]>
Commit | Line | Data |
---|---|---|
85208be0 ED |
1 | /* |
2 | * Copyright © 2012 Intel Corporation | |
3 | * | |
4 | * Permission is hereby granted, free of charge, to any person obtaining a | |
5 | * copy of this software and associated documentation files (the "Software"), | |
6 | * to deal in the Software without restriction, including without limitation | |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
8 | * and/or sell copies of the Software, and to permit persons to whom the | |
9 | * Software is furnished to do so, subject to the following conditions: | |
10 | * | |
11 | * The above copyright notice and this permission notice (including the next | |
12 | * paragraph) shall be included in all copies or substantial portions of the | |
13 | * Software. | |
14 | * | |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
21 | * IN THE SOFTWARE. | |
22 | * | |
23 | * Authors: | |
24 | * Eugeni Dodonov <eugeni.dodonov@intel.com> | |
25 | * | |
26 | */ | |
27 | ||
2b4e57bd | 28 | #include <linux/cpufreq.h> |
9c2f7a9d | 29 | #include <drm/drm_plane_helper.h> |
85208be0 ED |
30 | #include "i915_drv.h" |
31 | #include "intel_drv.h" | |
eb48eb00 DV |
32 | #include "../../../platform/x86/intel_ips.h" |
33 | #include <linux/module.h> | |
c8fe32c1 | 34 | #include <drm/drm_atomic_helper.h> |
85208be0 | 35 | |
dc39fff7 | 36 | /** |
18afd443 JN |
37 | * DOC: RC6 |
38 | * | |
dc39fff7 BW |
39 | * RC6 is a special power stage which allows the GPU to enter an very |
40 | * low-voltage mode when idle, using down to 0V while at this stage. This | |
41 | * stage is entered automatically when the GPU is idle when RC6 support is | |
42 | * enabled, and as soon as new workload arises GPU wakes up automatically as well. | |
43 | * | |
44 | * There are different RC6 modes available in Intel GPU, which differentiate | |
45 | * among each other with the latency required to enter and leave RC6 and | |
46 | * voltage consumed by the GPU in different states. | |
47 | * | |
48 | * The combination of the following flags define which states GPU is allowed | |
49 | * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and | |
50 | * RC6pp is deepest RC6. Their support by hardware varies according to the | |
51 | * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one | |
52 | * which brings the most power savings; deeper states save more power, but | |
53 | * require higher latency to switch to and wake up. | |
54 | */ | |
55 | #define INTEL_RC6_ENABLE (1<<0) | |
56 | #define INTEL_RC6p_ENABLE (1<<1) | |
57 | #define INTEL_RC6pp_ENABLE (1<<2) | |
58 | ||
46f16e63 | 59 | static void gen9_init_clock_gating(struct drm_i915_private *dev_priv) |
a82abe43 | 60 | { |
82525c17 | 61 | /* See Bspec note for PSR2_CTL bit 31, Wa#828:skl,bxt,kbl,cfl */ |
dc00b6a0 DV |
62 | I915_WRITE(CHICKEN_PAR1_1, |
63 | I915_READ(CHICKEN_PAR1_1) | SKL_EDP_PSR_FIX_RDWRAP); | |
64 | ||
b033bb6d MK |
65 | I915_WRITE(GEN8_CONFIG0, |
66 | I915_READ(GEN8_CONFIG0) | GEN9_DEFAULT_FIXES); | |
590e8ff0 | 67 | |
82525c17 | 68 | /* WaEnableChickenDCPR:skl,bxt,kbl,glk,cfl */ |
590e8ff0 MK |
69 | I915_WRITE(GEN8_CHICKEN_DCPR_1, |
70 | I915_READ(GEN8_CHICKEN_DCPR_1) | MASK_WAKEMEM); | |
0f78dee6 | 71 | |
82525c17 RV |
72 | /* WaFbcTurnOffFbcWatermark:skl,bxt,kbl,cfl */ |
73 | /* WaFbcWakeMemOn:skl,bxt,kbl,glk,cfl */ | |
303d4ea5 MK |
74 | I915_WRITE(DISP_ARB_CTL, I915_READ(DISP_ARB_CTL) | |
75 | DISP_FBC_WM_DIS | | |
76 | DISP_FBC_MEMORY_WAKE); | |
d1b4eefd | 77 | |
82525c17 | 78 | /* WaFbcHighMemBwCorruptionAvoidance:skl,bxt,kbl,cfl */ |
d1b4eefd MK |
79 | I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) | |
80 | ILK_DPFC_DISABLE_DUMMY0); | |
b033bb6d MK |
81 | } |
82 | ||
46f16e63 | 83 | static void bxt_init_clock_gating(struct drm_i915_private *dev_priv) |
b033bb6d | 84 | { |
46f16e63 | 85 | gen9_init_clock_gating(dev_priv); |
b033bb6d | 86 | |
a7546159 NH |
87 | /* WaDisableSDEUnitClockGating:bxt */ |
88 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
89 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
90 | ||
32608ca2 ID |
91 | /* |
92 | * FIXME: | |
868434c5 | 93 | * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only. |
32608ca2 | 94 | */ |
32608ca2 | 95 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | |
868434c5 | 96 | GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ); |
d965e7ac ID |
97 | |
98 | /* | |
99 | * Wa: Backlight PWM may stop in the asserted state, causing backlight | |
100 | * to stay fully on. | |
101 | */ | |
8aeaf64c JN |
102 | I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) | |
103 | PWM1_GATING_DIS | PWM2_GATING_DIS); | |
a82abe43 ID |
104 | } |
105 | ||
9fb5026f ACO |
106 | static void glk_init_clock_gating(struct drm_i915_private *dev_priv) |
107 | { | |
108 | gen9_init_clock_gating(dev_priv); | |
109 | ||
110 | /* | |
111 | * WaDisablePWMClockGating:glk | |
112 | * Backlight PWM may stop in the asserted state, causing backlight | |
113 | * to stay fully on. | |
114 | */ | |
115 | I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) | | |
116 | PWM1_GATING_DIS | PWM2_GATING_DIS); | |
f4f4b59b ACO |
117 | |
118 | /* WaDDIIOTimeout:glk */ | |
119 | if (IS_GLK_REVID(dev_priv, 0, GLK_REVID_A1)) { | |
120 | u32 val = I915_READ(CHICKEN_MISC_2); | |
121 | val &= ~(GLK_CL0_PWR_DOWN | | |
122 | GLK_CL1_PWR_DOWN | | |
123 | GLK_CL2_PWR_DOWN); | |
124 | I915_WRITE(CHICKEN_MISC_2, val); | |
125 | } | |
126 | ||
9fb5026f ACO |
127 | } |
128 | ||
148ac1f3 | 129 | static void i915_pineview_get_mem_freq(struct drm_i915_private *dev_priv) |
c921aba8 | 130 | { |
c921aba8 DV |
131 | u32 tmp; |
132 | ||
133 | tmp = I915_READ(CLKCFG); | |
134 | ||
135 | switch (tmp & CLKCFG_FSB_MASK) { | |
136 | case CLKCFG_FSB_533: | |
137 | dev_priv->fsb_freq = 533; /* 133*4 */ | |
138 | break; | |
139 | case CLKCFG_FSB_800: | |
140 | dev_priv->fsb_freq = 800; /* 200*4 */ | |
141 | break; | |
142 | case CLKCFG_FSB_667: | |
143 | dev_priv->fsb_freq = 667; /* 167*4 */ | |
144 | break; | |
145 | case CLKCFG_FSB_400: | |
146 | dev_priv->fsb_freq = 400; /* 100*4 */ | |
147 | break; | |
148 | } | |
149 | ||
150 | switch (tmp & CLKCFG_MEM_MASK) { | |
151 | case CLKCFG_MEM_533: | |
152 | dev_priv->mem_freq = 533; | |
153 | break; | |
154 | case CLKCFG_MEM_667: | |
155 | dev_priv->mem_freq = 667; | |
156 | break; | |
157 | case CLKCFG_MEM_800: | |
158 | dev_priv->mem_freq = 800; | |
159 | break; | |
160 | } | |
161 | ||
162 | /* detect pineview DDR3 setting */ | |
163 | tmp = I915_READ(CSHRDDR3CTL); | |
164 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; | |
165 | } | |
166 | ||
148ac1f3 | 167 | static void i915_ironlake_get_mem_freq(struct drm_i915_private *dev_priv) |
c921aba8 | 168 | { |
c921aba8 DV |
169 | u16 ddrpll, csipll; |
170 | ||
171 | ddrpll = I915_READ16(DDRMPLL1); | |
172 | csipll = I915_READ16(CSIPLL0); | |
173 | ||
174 | switch (ddrpll & 0xff) { | |
175 | case 0xc: | |
176 | dev_priv->mem_freq = 800; | |
177 | break; | |
178 | case 0x10: | |
179 | dev_priv->mem_freq = 1066; | |
180 | break; | |
181 | case 0x14: | |
182 | dev_priv->mem_freq = 1333; | |
183 | break; | |
184 | case 0x18: | |
185 | dev_priv->mem_freq = 1600; | |
186 | break; | |
187 | default: | |
188 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", | |
189 | ddrpll & 0xff); | |
190 | dev_priv->mem_freq = 0; | |
191 | break; | |
192 | } | |
193 | ||
20e4d407 | 194 | dev_priv->ips.r_t = dev_priv->mem_freq; |
c921aba8 DV |
195 | |
196 | switch (csipll & 0x3ff) { | |
197 | case 0x00c: | |
198 | dev_priv->fsb_freq = 3200; | |
199 | break; | |
200 | case 0x00e: | |
201 | dev_priv->fsb_freq = 3733; | |
202 | break; | |
203 | case 0x010: | |
204 | dev_priv->fsb_freq = 4266; | |
205 | break; | |
206 | case 0x012: | |
207 | dev_priv->fsb_freq = 4800; | |
208 | break; | |
209 | case 0x014: | |
210 | dev_priv->fsb_freq = 5333; | |
211 | break; | |
212 | case 0x016: | |
213 | dev_priv->fsb_freq = 5866; | |
214 | break; | |
215 | case 0x018: | |
216 | dev_priv->fsb_freq = 6400; | |
217 | break; | |
218 | default: | |
219 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", | |
220 | csipll & 0x3ff); | |
221 | dev_priv->fsb_freq = 0; | |
222 | break; | |
223 | } | |
224 | ||
225 | if (dev_priv->fsb_freq == 3200) { | |
20e4d407 | 226 | dev_priv->ips.c_m = 0; |
c921aba8 | 227 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
20e4d407 | 228 | dev_priv->ips.c_m = 1; |
c921aba8 | 229 | } else { |
20e4d407 | 230 | dev_priv->ips.c_m = 2; |
c921aba8 DV |
231 | } |
232 | } | |
233 | ||
b445e3b0 ED |
234 | static const struct cxsr_latency cxsr_latency_table[] = { |
235 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ | |
236 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ | |
237 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ | |
238 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ | |
239 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ | |
240 | ||
241 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ | |
242 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ | |
243 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ | |
244 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ | |
245 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ | |
246 | ||
247 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ | |
248 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ | |
249 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ | |
250 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ | |
251 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ | |
252 | ||
253 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ | |
254 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ | |
255 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ | |
256 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ | |
257 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ | |
258 | ||
259 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ | |
260 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ | |
261 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ | |
262 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ | |
263 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ | |
264 | ||
265 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ | |
266 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ | |
267 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ | |
268 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ | |
269 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ | |
270 | }; | |
271 | ||
44a655ca TU |
272 | static const struct cxsr_latency *intel_get_cxsr_latency(bool is_desktop, |
273 | bool is_ddr3, | |
b445e3b0 ED |
274 | int fsb, |
275 | int mem) | |
276 | { | |
277 | const struct cxsr_latency *latency; | |
278 | int i; | |
279 | ||
280 | if (fsb == 0 || mem == 0) | |
281 | return NULL; | |
282 | ||
283 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { | |
284 | latency = &cxsr_latency_table[i]; | |
285 | if (is_desktop == latency->is_desktop && | |
286 | is_ddr3 == latency->is_ddr3 && | |
287 | fsb == latency->fsb_freq && mem == latency->mem_freq) | |
288 | return latency; | |
289 | } | |
290 | ||
291 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
292 | ||
293 | return NULL; | |
294 | } | |
295 | ||
fc1ac8de VS |
296 | static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable) |
297 | { | |
298 | u32 val; | |
299 | ||
300 | mutex_lock(&dev_priv->rps.hw_lock); | |
301 | ||
302 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); | |
303 | if (enable) | |
304 | val &= ~FORCE_DDR_HIGH_FREQ; | |
305 | else | |
306 | val |= FORCE_DDR_HIGH_FREQ; | |
307 | val &= ~FORCE_DDR_LOW_FREQ; | |
308 | val |= FORCE_DDR_FREQ_REQ_ACK; | |
309 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); | |
310 | ||
311 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & | |
312 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) | |
313 | DRM_ERROR("timed out waiting for Punit DDR DVFS request\n"); | |
314 | ||
315 | mutex_unlock(&dev_priv->rps.hw_lock); | |
316 | } | |
317 | ||
cfb41411 VS |
318 | static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable) |
319 | { | |
320 | u32 val; | |
321 | ||
322 | mutex_lock(&dev_priv->rps.hw_lock); | |
323 | ||
324 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); | |
325 | if (enable) | |
326 | val |= DSP_MAXFIFO_PM5_ENABLE; | |
327 | else | |
328 | val &= ~DSP_MAXFIFO_PM5_ENABLE; | |
329 | vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val); | |
330 | ||
331 | mutex_unlock(&dev_priv->rps.hw_lock); | |
332 | } | |
333 | ||
f4998963 VS |
334 | #define FW_WM(value, plane) \ |
335 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK) | |
336 | ||
11a85d6a | 337 | static bool _intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable) |
b445e3b0 | 338 | { |
11a85d6a | 339 | bool was_enabled; |
5209b1f4 | 340 | u32 val; |
b445e3b0 | 341 | |
920a14b2 | 342 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { |
11a85d6a | 343 | was_enabled = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN; |
5209b1f4 | 344 | I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0); |
a7a6c498 | 345 | POSTING_READ(FW_BLC_SELF_VLV); |
c0f86832 | 346 | } else if (IS_G4X(dev_priv) || IS_I965GM(dev_priv)) { |
11a85d6a | 347 | was_enabled = I915_READ(FW_BLC_SELF) & FW_BLC_SELF_EN; |
5209b1f4 | 348 | I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0); |
a7a6c498 | 349 | POSTING_READ(FW_BLC_SELF); |
9b1e14f4 | 350 | } else if (IS_PINEVIEW(dev_priv)) { |
11a85d6a VS |
351 | val = I915_READ(DSPFW3); |
352 | was_enabled = val & PINEVIEW_SELF_REFRESH_EN; | |
353 | if (enable) | |
354 | val |= PINEVIEW_SELF_REFRESH_EN; | |
355 | else | |
356 | val &= ~PINEVIEW_SELF_REFRESH_EN; | |
5209b1f4 | 357 | I915_WRITE(DSPFW3, val); |
a7a6c498 | 358 | POSTING_READ(DSPFW3); |
50a0bc90 | 359 | } else if (IS_I945G(dev_priv) || IS_I945GM(dev_priv)) { |
11a85d6a | 360 | was_enabled = I915_READ(FW_BLC_SELF) & FW_BLC_SELF_EN; |
5209b1f4 ID |
361 | val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) : |
362 | _MASKED_BIT_DISABLE(FW_BLC_SELF_EN); | |
363 | I915_WRITE(FW_BLC_SELF, val); | |
a7a6c498 | 364 | POSTING_READ(FW_BLC_SELF); |
50a0bc90 | 365 | } else if (IS_I915GM(dev_priv)) { |
acb91359 VS |
366 | /* |
367 | * FIXME can't find a bit like this for 915G, and | |
368 | * and yet it does have the related watermark in | |
369 | * FW_BLC_SELF. What's going on? | |
370 | */ | |
11a85d6a | 371 | was_enabled = I915_READ(INSTPM) & INSTPM_SELF_EN; |
5209b1f4 ID |
372 | val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) : |
373 | _MASKED_BIT_DISABLE(INSTPM_SELF_EN); | |
374 | I915_WRITE(INSTPM, val); | |
a7a6c498 | 375 | POSTING_READ(INSTPM); |
5209b1f4 | 376 | } else { |
11a85d6a | 377 | return false; |
5209b1f4 | 378 | } |
b445e3b0 | 379 | |
1489bba8 VS |
380 | trace_intel_memory_cxsr(dev_priv, was_enabled, enable); |
381 | ||
11a85d6a VS |
382 | DRM_DEBUG_KMS("memory self-refresh is %s (was %s)\n", |
383 | enableddisabled(enable), | |
384 | enableddisabled(was_enabled)); | |
385 | ||
386 | return was_enabled; | |
b445e3b0 ED |
387 | } |
388 | ||
62571fc3 VS |
389 | /** |
390 | * intel_set_memory_cxsr - Configure CxSR state | |
391 | * @dev_priv: i915 device | |
392 | * @enable: Allow vs. disallow CxSR | |
393 | * | |
394 | * Allow or disallow the system to enter a special CxSR | |
395 | * (C-state self refresh) state. What typically happens in CxSR mode | |
396 | * is that several display FIFOs may get combined into a single larger | |
397 | * FIFO for a particular plane (so called max FIFO mode) to allow the | |
398 | * system to defer memory fetches longer, and the memory will enter | |
399 | * self refresh. | |
400 | * | |
401 | * Note that enabling CxSR does not guarantee that the system enter | |
402 | * this special mode, nor does it guarantee that the system stays | |
403 | * in that mode once entered. So this just allows/disallows the system | |
404 | * to autonomously utilize the CxSR mode. Other factors such as core | |
405 | * C-states will affect when/if the system actually enters/exits the | |
406 | * CxSR mode. | |
407 | * | |
408 | * Note that on VLV/CHV this actually only controls the max FIFO mode, | |
409 | * and the system is free to enter/exit memory self refresh at any time | |
410 | * even when the use of CxSR has been disallowed. | |
411 | * | |
412 | * While the system is actually in the CxSR/max FIFO mode, some plane | |
413 | * control registers will not get latched on vblank. Thus in order to | |
414 | * guarantee the system will respond to changes in the plane registers | |
415 | * we must always disallow CxSR prior to making changes to those registers. | |
416 | * Unfortunately the system will re-evaluate the CxSR conditions at | |
417 | * frame start which happens after vblank start (which is when the plane | |
418 | * registers would get latched), so we can't proceed with the plane update | |
419 | * during the same frame where we disallowed CxSR. | |
420 | * | |
421 | * Certain platforms also have a deeper HPLL SR mode. Fortunately the | |
422 | * HPLL SR mode depends on CxSR itself, so we don't have to hand hold | |
423 | * the hardware w.r.t. HPLL SR when writing to plane registers. | |
424 | * Disallowing just CxSR is sufficient. | |
425 | */ | |
11a85d6a | 426 | bool intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable) |
3d90e649 | 427 | { |
11a85d6a VS |
428 | bool ret; |
429 | ||
3d90e649 | 430 | mutex_lock(&dev_priv->wm.wm_mutex); |
11a85d6a | 431 | ret = _intel_set_memory_cxsr(dev_priv, enable); |
04548cba VS |
432 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
433 | dev_priv->wm.vlv.cxsr = enable; | |
434 | else if (IS_G4X(dev_priv)) | |
435 | dev_priv->wm.g4x.cxsr = enable; | |
3d90e649 | 436 | mutex_unlock(&dev_priv->wm.wm_mutex); |
11a85d6a VS |
437 | |
438 | return ret; | |
3d90e649 | 439 | } |
fc1ac8de | 440 | |
b445e3b0 ED |
441 | /* |
442 | * Latency for FIFO fetches is dependent on several factors: | |
443 | * - memory configuration (speed, channels) | |
444 | * - chipset | |
445 | * - current MCH state | |
446 | * It can be fairly high in some situations, so here we assume a fairly | |
447 | * pessimal value. It's a tradeoff between extra memory fetches (if we | |
448 | * set this value too high, the FIFO will fetch frequently to stay full) | |
449 | * and power consumption (set it too low to save power and we might see | |
450 | * FIFO underruns and display "flicker"). | |
451 | * | |
452 | * A value of 5us seems to be a good balance; safe for very low end | |
453 | * platforms but not overly aggressive on lower latency configs. | |
454 | */ | |
5aef6003 | 455 | static const int pessimal_latency_ns = 5000; |
b445e3b0 | 456 | |
b5004720 VS |
457 | #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \ |
458 | ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8)) | |
459 | ||
814e7f0b | 460 | static void vlv_get_fifo_size(struct intel_crtc_state *crtc_state) |
b5004720 | 461 | { |
814e7f0b | 462 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); |
f07d43d2 | 463 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); |
814e7f0b | 464 | struct vlv_fifo_state *fifo_state = &crtc_state->wm.vlv.fifo_state; |
f07d43d2 VS |
465 | enum pipe pipe = crtc->pipe; |
466 | int sprite0_start, sprite1_start; | |
49845a23 | 467 | |
f07d43d2 | 468 | switch (pipe) { |
b5004720 VS |
469 | uint32_t dsparb, dsparb2, dsparb3; |
470 | case PIPE_A: | |
471 | dsparb = I915_READ(DSPARB); | |
472 | dsparb2 = I915_READ(DSPARB2); | |
473 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0); | |
474 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4); | |
475 | break; | |
476 | case PIPE_B: | |
477 | dsparb = I915_READ(DSPARB); | |
478 | dsparb2 = I915_READ(DSPARB2); | |
479 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8); | |
480 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12); | |
481 | break; | |
482 | case PIPE_C: | |
483 | dsparb2 = I915_READ(DSPARB2); | |
484 | dsparb3 = I915_READ(DSPARB3); | |
485 | sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16); | |
486 | sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20); | |
487 | break; | |
488 | default: | |
f07d43d2 VS |
489 | MISSING_CASE(pipe); |
490 | return; | |
b5004720 VS |
491 | } |
492 | ||
f07d43d2 VS |
493 | fifo_state->plane[PLANE_PRIMARY] = sprite0_start; |
494 | fifo_state->plane[PLANE_SPRITE0] = sprite1_start - sprite0_start; | |
495 | fifo_state->plane[PLANE_SPRITE1] = 511 - sprite1_start; | |
496 | fifo_state->plane[PLANE_CURSOR] = 63; | |
b5004720 VS |
497 | } |
498 | ||
ef0f5e93 | 499 | static int i9xx_get_fifo_size(struct drm_i915_private *dev_priv, int plane) |
b445e3b0 | 500 | { |
b445e3b0 ED |
501 | uint32_t dsparb = I915_READ(DSPARB); |
502 | int size; | |
503 | ||
504 | size = dsparb & 0x7f; | |
505 | if (plane) | |
506 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; | |
507 | ||
508 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
509 | plane ? "B" : "A", size); | |
510 | ||
511 | return size; | |
512 | } | |
513 | ||
ef0f5e93 | 514 | static int i830_get_fifo_size(struct drm_i915_private *dev_priv, int plane) |
b445e3b0 | 515 | { |
b445e3b0 ED |
516 | uint32_t dsparb = I915_READ(DSPARB); |
517 | int size; | |
518 | ||
519 | size = dsparb & 0x1ff; | |
520 | if (plane) | |
521 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; | |
522 | size >>= 1; /* Convert to cachelines */ | |
523 | ||
524 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
525 | plane ? "B" : "A", size); | |
526 | ||
527 | return size; | |
528 | } | |
529 | ||
ef0f5e93 | 530 | static int i845_get_fifo_size(struct drm_i915_private *dev_priv, int plane) |
b445e3b0 | 531 | { |
b445e3b0 ED |
532 | uint32_t dsparb = I915_READ(DSPARB); |
533 | int size; | |
534 | ||
535 | size = dsparb & 0x7f; | |
536 | size >>= 2; /* Convert to cachelines */ | |
537 | ||
538 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
539 | plane ? "B" : "A", | |
540 | size); | |
541 | ||
542 | return size; | |
543 | } | |
544 | ||
b445e3b0 ED |
545 | /* Pineview has different values for various configs */ |
546 | static const struct intel_watermark_params pineview_display_wm = { | |
e0f0273e VS |
547 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
548 | .max_wm = PINEVIEW_MAX_WM, | |
549 | .default_wm = PINEVIEW_DFT_WM, | |
550 | .guard_size = PINEVIEW_GUARD_WM, | |
551 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
552 | }; |
553 | static const struct intel_watermark_params pineview_display_hplloff_wm = { | |
e0f0273e VS |
554 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
555 | .max_wm = PINEVIEW_MAX_WM, | |
556 | .default_wm = PINEVIEW_DFT_HPLLOFF_WM, | |
557 | .guard_size = PINEVIEW_GUARD_WM, | |
558 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
559 | }; |
560 | static const struct intel_watermark_params pineview_cursor_wm = { | |
e0f0273e VS |
561 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
562 | .max_wm = PINEVIEW_CURSOR_MAX_WM, | |
563 | .default_wm = PINEVIEW_CURSOR_DFT_WM, | |
564 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, | |
565 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
566 | }; |
567 | static const struct intel_watermark_params pineview_cursor_hplloff_wm = { | |
e0f0273e VS |
568 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
569 | .max_wm = PINEVIEW_CURSOR_MAX_WM, | |
570 | .default_wm = PINEVIEW_CURSOR_DFT_WM, | |
571 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, | |
572 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 | 573 | }; |
b445e3b0 | 574 | static const struct intel_watermark_params i965_cursor_wm_info = { |
e0f0273e VS |
575 | .fifo_size = I965_CURSOR_FIFO, |
576 | .max_wm = I965_CURSOR_MAX_WM, | |
577 | .default_wm = I965_CURSOR_DFT_WM, | |
578 | .guard_size = 2, | |
579 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 ED |
580 | }; |
581 | static const struct intel_watermark_params i945_wm_info = { | |
e0f0273e VS |
582 | .fifo_size = I945_FIFO_SIZE, |
583 | .max_wm = I915_MAX_WM, | |
584 | .default_wm = 1, | |
585 | .guard_size = 2, | |
586 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 ED |
587 | }; |
588 | static const struct intel_watermark_params i915_wm_info = { | |
e0f0273e VS |
589 | .fifo_size = I915_FIFO_SIZE, |
590 | .max_wm = I915_MAX_WM, | |
591 | .default_wm = 1, | |
592 | .guard_size = 2, | |
593 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 | 594 | }; |
9d539105 | 595 | static const struct intel_watermark_params i830_a_wm_info = { |
e0f0273e VS |
596 | .fifo_size = I855GM_FIFO_SIZE, |
597 | .max_wm = I915_MAX_WM, | |
598 | .default_wm = 1, | |
599 | .guard_size = 2, | |
600 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
b445e3b0 | 601 | }; |
9d539105 VS |
602 | static const struct intel_watermark_params i830_bc_wm_info = { |
603 | .fifo_size = I855GM_FIFO_SIZE, | |
604 | .max_wm = I915_MAX_WM/2, | |
605 | .default_wm = 1, | |
606 | .guard_size = 2, | |
607 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
608 | }; | |
feb56b93 | 609 | static const struct intel_watermark_params i845_wm_info = { |
e0f0273e VS |
610 | .fifo_size = I830_FIFO_SIZE, |
611 | .max_wm = I915_MAX_WM, | |
612 | .default_wm = 1, | |
613 | .guard_size = 2, | |
614 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
b445e3b0 ED |
615 | }; |
616 | ||
baf69ca8 VS |
617 | /** |
618 | * intel_wm_method1 - Method 1 / "small buffer" watermark formula | |
619 | * @pixel_rate: Pipe pixel rate in kHz | |
620 | * @cpp: Plane bytes per pixel | |
621 | * @latency: Memory wakeup latency in 0.1us units | |
622 | * | |
623 | * Compute the watermark using the method 1 or "small buffer" | |
624 | * formula. The caller may additonally add extra cachelines | |
625 | * to account for TLB misses and clock crossings. | |
626 | * | |
627 | * This method is concerned with the short term drain rate | |
628 | * of the FIFO, ie. it does not account for blanking periods | |
629 | * which would effectively reduce the average drain rate across | |
630 | * a longer period. The name "small" refers to the fact the | |
631 | * FIFO is relatively small compared to the amount of data | |
632 | * fetched. | |
633 | * | |
634 | * The FIFO level vs. time graph might look something like: | |
635 | * | |
636 | * |\ |\ | |
637 | * | \ | \ | |
638 | * __---__---__ (- plane active, _ blanking) | |
639 | * -> time | |
640 | * | |
641 | * or perhaps like this: | |
642 | * | |
643 | * |\|\ |\|\ | |
644 | * __----__----__ (- plane active, _ blanking) | |
645 | * -> time | |
646 | * | |
647 | * Returns: | |
648 | * The watermark in bytes | |
649 | */ | |
650 | static unsigned int intel_wm_method1(unsigned int pixel_rate, | |
651 | unsigned int cpp, | |
652 | unsigned int latency) | |
653 | { | |
654 | uint64_t ret; | |
655 | ||
656 | ret = (uint64_t) pixel_rate * cpp * latency; | |
657 | ret = DIV_ROUND_UP_ULL(ret, 10000); | |
658 | ||
659 | return ret; | |
660 | } | |
661 | ||
662 | /** | |
663 | * intel_wm_method2 - Method 2 / "large buffer" watermark formula | |
664 | * @pixel_rate: Pipe pixel rate in kHz | |
665 | * @htotal: Pipe horizontal total | |
666 | * @width: Plane width in pixels | |
667 | * @cpp: Plane bytes per pixel | |
668 | * @latency: Memory wakeup latency in 0.1us units | |
669 | * | |
670 | * Compute the watermark using the method 2 or "large buffer" | |
671 | * formula. The caller may additonally add extra cachelines | |
672 | * to account for TLB misses and clock crossings. | |
673 | * | |
674 | * This method is concerned with the long term drain rate | |
675 | * of the FIFO, ie. it does account for blanking periods | |
676 | * which effectively reduce the average drain rate across | |
677 | * a longer period. The name "large" refers to the fact the | |
678 | * FIFO is relatively large compared to the amount of data | |
679 | * fetched. | |
680 | * | |
681 | * The FIFO level vs. time graph might look something like: | |
682 | * | |
683 | * |\___ |\___ | |
684 | * | \___ | \___ | |
685 | * | \ | \ | |
686 | * __ --__--__--__--__--__--__ (- plane active, _ blanking) | |
687 | * -> time | |
688 | * | |
689 | * Returns: | |
690 | * The watermark in bytes | |
691 | */ | |
692 | static unsigned int intel_wm_method2(unsigned int pixel_rate, | |
693 | unsigned int htotal, | |
694 | unsigned int width, | |
695 | unsigned int cpp, | |
696 | unsigned int latency) | |
697 | { | |
698 | unsigned int ret; | |
699 | ||
700 | /* | |
701 | * FIXME remove once all users are computing | |
702 | * watermarks in the correct place. | |
703 | */ | |
704 | if (WARN_ON_ONCE(htotal == 0)) | |
705 | htotal = 1; | |
706 | ||
707 | ret = (latency * pixel_rate) / (htotal * 10000); | |
708 | ret = (ret + 1) * width * cpp; | |
709 | ||
710 | return ret; | |
711 | } | |
712 | ||
b445e3b0 ED |
713 | /** |
714 | * intel_calculate_wm - calculate watermark level | |
baf69ca8 | 715 | * @pixel_rate: pixel clock |
b445e3b0 | 716 | * @wm: chip FIFO params |
ac484963 | 717 | * @cpp: bytes per pixel |
b445e3b0 ED |
718 | * @latency_ns: memory latency for the platform |
719 | * | |
720 | * Calculate the watermark level (the level at which the display plane will | |
721 | * start fetching from memory again). Each chip has a different display | |
722 | * FIFO size and allocation, so the caller needs to figure that out and pass | |
723 | * in the correct intel_watermark_params structure. | |
724 | * | |
725 | * As the pixel clock runs, the FIFO will be drained at a rate that depends | |
726 | * on the pixel size. When it reaches the watermark level, it'll start | |
727 | * fetching FIFO line sized based chunks from memory until the FIFO fills | |
728 | * past the watermark point. If the FIFO drains completely, a FIFO underrun | |
729 | * will occur, and a display engine hang could result. | |
730 | */ | |
baf69ca8 VS |
731 | static unsigned int intel_calculate_wm(int pixel_rate, |
732 | const struct intel_watermark_params *wm, | |
733 | int fifo_size, int cpp, | |
734 | unsigned int latency_ns) | |
b445e3b0 | 735 | { |
baf69ca8 | 736 | int entries, wm_size; |
b445e3b0 ED |
737 | |
738 | /* | |
739 | * Note: we need to make sure we don't overflow for various clock & | |
740 | * latency values. | |
741 | * clocks go from a few thousand to several hundred thousand. | |
742 | * latency is usually a few thousand | |
743 | */ | |
baf69ca8 VS |
744 | entries = intel_wm_method1(pixel_rate, cpp, |
745 | latency_ns / 100); | |
746 | entries = DIV_ROUND_UP(entries, wm->cacheline_size) + | |
747 | wm->guard_size; | |
748 | DRM_DEBUG_KMS("FIFO entries required for mode: %d\n", entries); | |
b445e3b0 | 749 | |
baf69ca8 VS |
750 | wm_size = fifo_size - entries; |
751 | DRM_DEBUG_KMS("FIFO watermark level: %d\n", wm_size); | |
b445e3b0 ED |
752 | |
753 | /* Don't promote wm_size to unsigned... */ | |
baf69ca8 | 754 | if (wm_size > wm->max_wm) |
b445e3b0 ED |
755 | wm_size = wm->max_wm; |
756 | if (wm_size <= 0) | |
757 | wm_size = wm->default_wm; | |
d6feb196 VS |
758 | |
759 | /* | |
760 | * Bspec seems to indicate that the value shouldn't be lower than | |
761 | * 'burst size + 1'. Certainly 830 is quite unhappy with low values. | |
762 | * Lets go for 8 which is the burst size since certain platforms | |
763 | * already use a hardcoded 8 (which is what the spec says should be | |
764 | * done). | |
765 | */ | |
766 | if (wm_size <= 8) | |
767 | wm_size = 8; | |
768 | ||
b445e3b0 ED |
769 | return wm_size; |
770 | } | |
771 | ||
04548cba VS |
772 | static bool is_disabling(int old, int new, int threshold) |
773 | { | |
774 | return old >= threshold && new < threshold; | |
775 | } | |
776 | ||
777 | static bool is_enabling(int old, int new, int threshold) | |
778 | { | |
779 | return old < threshold && new >= threshold; | |
780 | } | |
781 | ||
6d5019b6 VS |
782 | static int intel_wm_num_levels(struct drm_i915_private *dev_priv) |
783 | { | |
784 | return dev_priv->wm.max_level + 1; | |
785 | } | |
786 | ||
24304d81 VS |
787 | static bool intel_wm_plane_visible(const struct intel_crtc_state *crtc_state, |
788 | const struct intel_plane_state *plane_state) | |
789 | { | |
790 | struct intel_plane *plane = to_intel_plane(plane_state->base.plane); | |
791 | ||
792 | /* FIXME check the 'enable' instead */ | |
793 | if (!crtc_state->base.active) | |
794 | return false; | |
795 | ||
796 | /* | |
797 | * Treat cursor with fb as always visible since cursor updates | |
798 | * can happen faster than the vrefresh rate, and the current | |
799 | * watermark code doesn't handle that correctly. Cursor updates | |
800 | * which set/clear the fb or change the cursor size are going | |
801 | * to get throttled by intel_legacy_cursor_update() to work | |
802 | * around this problem with the watermark code. | |
803 | */ | |
804 | if (plane->id == PLANE_CURSOR) | |
805 | return plane_state->base.fb != NULL; | |
806 | else | |
807 | return plane_state->base.visible; | |
808 | } | |
809 | ||
ffc7a76b | 810 | static struct intel_crtc *single_enabled_crtc(struct drm_i915_private *dev_priv) |
b445e3b0 | 811 | { |
efc2611e | 812 | struct intel_crtc *crtc, *enabled = NULL; |
b445e3b0 | 813 | |
ffc7a76b | 814 | for_each_intel_crtc(&dev_priv->drm, crtc) { |
efc2611e | 815 | if (intel_crtc_active(crtc)) { |
b445e3b0 ED |
816 | if (enabled) |
817 | return NULL; | |
818 | enabled = crtc; | |
819 | } | |
820 | } | |
821 | ||
822 | return enabled; | |
823 | } | |
824 | ||
432081bc | 825 | static void pineview_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 826 | { |
ffc7a76b | 827 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
efc2611e | 828 | struct intel_crtc *crtc; |
b445e3b0 ED |
829 | const struct cxsr_latency *latency; |
830 | u32 reg; | |
baf69ca8 | 831 | unsigned int wm; |
b445e3b0 | 832 | |
50a0bc90 TU |
833 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev_priv), |
834 | dev_priv->is_ddr3, | |
835 | dev_priv->fsb_freq, | |
836 | dev_priv->mem_freq); | |
b445e3b0 ED |
837 | if (!latency) { |
838 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
5209b1f4 | 839 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
840 | return; |
841 | } | |
842 | ||
ffc7a76b | 843 | crtc = single_enabled_crtc(dev_priv); |
b445e3b0 | 844 | if (crtc) { |
efc2611e VS |
845 | const struct drm_display_mode *adjusted_mode = |
846 | &crtc->config->base.adjusted_mode; | |
847 | const struct drm_framebuffer *fb = | |
848 | crtc->base.primary->state->fb; | |
353c8598 | 849 | int cpp = fb->format->cpp[0]; |
7c5f93b0 | 850 | int clock = adjusted_mode->crtc_clock; |
b445e3b0 ED |
851 | |
852 | /* Display SR */ | |
853 | wm = intel_calculate_wm(clock, &pineview_display_wm, | |
854 | pineview_display_wm.fifo_size, | |
ac484963 | 855 | cpp, latency->display_sr); |
b445e3b0 ED |
856 | reg = I915_READ(DSPFW1); |
857 | reg &= ~DSPFW_SR_MASK; | |
f4998963 | 858 | reg |= FW_WM(wm, SR); |
b445e3b0 ED |
859 | I915_WRITE(DSPFW1, reg); |
860 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); | |
861 | ||
862 | /* cursor SR */ | |
863 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, | |
864 | pineview_display_wm.fifo_size, | |
99834b14 | 865 | 4, latency->cursor_sr); |
b445e3b0 ED |
866 | reg = I915_READ(DSPFW3); |
867 | reg &= ~DSPFW_CURSOR_SR_MASK; | |
f4998963 | 868 | reg |= FW_WM(wm, CURSOR_SR); |
b445e3b0 ED |
869 | I915_WRITE(DSPFW3, reg); |
870 | ||
871 | /* Display HPLL off SR */ | |
872 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, | |
873 | pineview_display_hplloff_wm.fifo_size, | |
ac484963 | 874 | cpp, latency->display_hpll_disable); |
b445e3b0 ED |
875 | reg = I915_READ(DSPFW3); |
876 | reg &= ~DSPFW_HPLL_SR_MASK; | |
f4998963 | 877 | reg |= FW_WM(wm, HPLL_SR); |
b445e3b0 ED |
878 | I915_WRITE(DSPFW3, reg); |
879 | ||
880 | /* cursor HPLL off SR */ | |
881 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, | |
882 | pineview_display_hplloff_wm.fifo_size, | |
99834b14 | 883 | 4, latency->cursor_hpll_disable); |
b445e3b0 ED |
884 | reg = I915_READ(DSPFW3); |
885 | reg &= ~DSPFW_HPLL_CURSOR_MASK; | |
f4998963 | 886 | reg |= FW_WM(wm, HPLL_CURSOR); |
b445e3b0 ED |
887 | I915_WRITE(DSPFW3, reg); |
888 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); | |
889 | ||
5209b1f4 | 890 | intel_set_memory_cxsr(dev_priv, true); |
b445e3b0 | 891 | } else { |
5209b1f4 | 892 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
893 | } |
894 | } | |
895 | ||
0f95ff85 VS |
896 | /* |
897 | * Documentation says: | |
898 | * "If the line size is small, the TLB fetches can get in the way of the | |
899 | * data fetches, causing some lag in the pixel data return which is not | |
900 | * accounted for in the above formulas. The following adjustment only | |
901 | * needs to be applied if eight whole lines fit in the buffer at once. | |
902 | * The WM is adjusted upwards by the difference between the FIFO size | |
903 | * and the size of 8 whole lines. This adjustment is always performed | |
904 | * in the actual pixel depth regardless of whether FBC is enabled or not." | |
905 | */ | |
906 | static int g4x_tlb_miss_wa(int fifo_size, int width, int cpp) | |
907 | { | |
908 | int tlb_miss = fifo_size * 64 - width * cpp * 8; | |
909 | ||
910 | return max(0, tlb_miss); | |
911 | } | |
912 | ||
04548cba VS |
913 | static void g4x_write_wm_values(struct drm_i915_private *dev_priv, |
914 | const struct g4x_wm_values *wm) | |
b445e3b0 | 915 | { |
e93329a5 VS |
916 | enum pipe pipe; |
917 | ||
918 | for_each_pipe(dev_priv, pipe) | |
919 | trace_g4x_wm(intel_get_crtc_for_pipe(dev_priv, pipe), wm); | |
920 | ||
04548cba VS |
921 | I915_WRITE(DSPFW1, |
922 | FW_WM(wm->sr.plane, SR) | | |
923 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_CURSOR], CURSORB) | | |
924 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_PRIMARY], PLANEB) | | |
925 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_PRIMARY], PLANEA)); | |
926 | I915_WRITE(DSPFW2, | |
927 | (wm->fbc_en ? DSPFW_FBC_SR_EN : 0) | | |
928 | FW_WM(wm->sr.fbc, FBC_SR) | | |
929 | FW_WM(wm->hpll.fbc, FBC_HPLL_SR) | | |
930 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE0], SPRITEB) | | |
931 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_CURSOR], CURSORA) | | |
932 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE0], SPRITEA)); | |
933 | I915_WRITE(DSPFW3, | |
934 | (wm->hpll_en ? DSPFW_HPLL_SR_EN : 0) | | |
935 | FW_WM(wm->sr.cursor, CURSOR_SR) | | |
936 | FW_WM(wm->hpll.cursor, HPLL_CURSOR) | | |
937 | FW_WM(wm->hpll.plane, HPLL_SR)); | |
b445e3b0 | 938 | |
04548cba | 939 | POSTING_READ(DSPFW1); |
b445e3b0 ED |
940 | } |
941 | ||
15665979 VS |
942 | #define FW_WM_VLV(value, plane) \ |
943 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV) | |
944 | ||
50f4caef | 945 | static void vlv_write_wm_values(struct drm_i915_private *dev_priv, |
0018fda1 VS |
946 | const struct vlv_wm_values *wm) |
947 | { | |
50f4caef VS |
948 | enum pipe pipe; |
949 | ||
950 | for_each_pipe(dev_priv, pipe) { | |
c137d660 VS |
951 | trace_vlv_wm(intel_get_crtc_for_pipe(dev_priv, pipe), wm); |
952 | ||
50f4caef VS |
953 | I915_WRITE(VLV_DDL(pipe), |
954 | (wm->ddl[pipe].plane[PLANE_CURSOR] << DDL_CURSOR_SHIFT) | | |
955 | (wm->ddl[pipe].plane[PLANE_SPRITE1] << DDL_SPRITE_SHIFT(1)) | | |
956 | (wm->ddl[pipe].plane[PLANE_SPRITE0] << DDL_SPRITE_SHIFT(0)) | | |
957 | (wm->ddl[pipe].plane[PLANE_PRIMARY] << DDL_PLANE_SHIFT)); | |
958 | } | |
0018fda1 | 959 | |
6fe6a7ff VS |
960 | /* |
961 | * Zero the (unused) WM1 watermarks, and also clear all the | |
962 | * high order bits so that there are no out of bounds values | |
963 | * present in the registers during the reprogramming. | |
964 | */ | |
965 | I915_WRITE(DSPHOWM, 0); | |
966 | I915_WRITE(DSPHOWM1, 0); | |
967 | I915_WRITE(DSPFW4, 0); | |
968 | I915_WRITE(DSPFW5, 0); | |
969 | I915_WRITE(DSPFW6, 0); | |
970 | ||
ae80152d | 971 | I915_WRITE(DSPFW1, |
15665979 | 972 | FW_WM(wm->sr.plane, SR) | |
1b31389c VS |
973 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_CURSOR], CURSORB) | |
974 | FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_PRIMARY], PLANEB) | | |
975 | FW_WM_VLV(wm->pipe[PIPE_A].plane[PLANE_PRIMARY], PLANEA)); | |
ae80152d | 976 | I915_WRITE(DSPFW2, |
1b31389c VS |
977 | FW_WM_VLV(wm->pipe[PIPE_A].plane[PLANE_SPRITE1], SPRITEB) | |
978 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_CURSOR], CURSORA) | | |
979 | FW_WM_VLV(wm->pipe[PIPE_A].plane[PLANE_SPRITE0], SPRITEA)); | |
ae80152d | 980 | I915_WRITE(DSPFW3, |
15665979 | 981 | FW_WM(wm->sr.cursor, CURSOR_SR)); |
ae80152d VS |
982 | |
983 | if (IS_CHERRYVIEW(dev_priv)) { | |
984 | I915_WRITE(DSPFW7_CHV, | |
1b31389c VS |
985 | FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE1], SPRITED) | |
986 | FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE0], SPRITEC)); | |
ae80152d | 987 | I915_WRITE(DSPFW8_CHV, |
1b31389c VS |
988 | FW_WM_VLV(wm->pipe[PIPE_C].plane[PLANE_SPRITE1], SPRITEF) | |
989 | FW_WM_VLV(wm->pipe[PIPE_C].plane[PLANE_SPRITE0], SPRITEE)); | |
ae80152d | 990 | I915_WRITE(DSPFW9_CHV, |
1b31389c VS |
991 | FW_WM_VLV(wm->pipe[PIPE_C].plane[PLANE_PRIMARY], PLANEC) | |
992 | FW_WM(wm->pipe[PIPE_C].plane[PLANE_CURSOR], CURSORC)); | |
ae80152d | 993 | I915_WRITE(DSPHOWM, |
15665979 | 994 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
1b31389c VS |
995 | FW_WM(wm->pipe[PIPE_C].plane[PLANE_SPRITE1] >> 8, SPRITEF_HI) | |
996 | FW_WM(wm->pipe[PIPE_C].plane[PLANE_SPRITE0] >> 8, SPRITEE_HI) | | |
997 | FW_WM(wm->pipe[PIPE_C].plane[PLANE_PRIMARY] >> 8, PLANEC_HI) | | |
998 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE1] >> 8, SPRITED_HI) | | |
999 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE0] >> 8, SPRITEC_HI) | | |
1000 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_PRIMARY] >> 8, PLANEB_HI) | | |
1001 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE1] >> 8, SPRITEB_HI) | | |
1002 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE0] >> 8, SPRITEA_HI) | | |
1003 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_PRIMARY] >> 8, PLANEA_HI)); | |
ae80152d VS |
1004 | } else { |
1005 | I915_WRITE(DSPFW7, | |
1b31389c VS |
1006 | FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE1], SPRITED) | |
1007 | FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE0], SPRITEC)); | |
ae80152d | 1008 | I915_WRITE(DSPHOWM, |
15665979 | 1009 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
1b31389c VS |
1010 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE1] >> 8, SPRITED_HI) | |
1011 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE0] >> 8, SPRITEC_HI) | | |
1012 | FW_WM(wm->pipe[PIPE_B].plane[PLANE_PRIMARY] >> 8, PLANEB_HI) | | |
1013 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE1] >> 8, SPRITEB_HI) | | |
1014 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE0] >> 8, SPRITEA_HI) | | |
1015 | FW_WM(wm->pipe[PIPE_A].plane[PLANE_PRIMARY] >> 8, PLANEA_HI)); | |
ae80152d VS |
1016 | } |
1017 | ||
1018 | POSTING_READ(DSPFW1); | |
0018fda1 VS |
1019 | } |
1020 | ||
15665979 VS |
1021 | #undef FW_WM_VLV |
1022 | ||
04548cba VS |
1023 | static void g4x_setup_wm_latency(struct drm_i915_private *dev_priv) |
1024 | { | |
1025 | /* all latencies in usec */ | |
1026 | dev_priv->wm.pri_latency[G4X_WM_LEVEL_NORMAL] = 5; | |
1027 | dev_priv->wm.pri_latency[G4X_WM_LEVEL_SR] = 12; | |
79d94306 | 1028 | dev_priv->wm.pri_latency[G4X_WM_LEVEL_HPLL] = 35; |
04548cba | 1029 | |
79d94306 | 1030 | dev_priv->wm.max_level = G4X_WM_LEVEL_HPLL; |
04548cba VS |
1031 | } |
1032 | ||
1033 | static int g4x_plane_fifo_size(enum plane_id plane_id, int level) | |
1034 | { | |
1035 | /* | |
1036 | * DSPCNTR[13] supposedly controls whether the | |
1037 | * primary plane can use the FIFO space otherwise | |
1038 | * reserved for the sprite plane. It's not 100% clear | |
1039 | * what the actual FIFO size is, but it looks like we | |
1040 | * can happily set both primary and sprite watermarks | |
1041 | * up to 127 cachelines. So that would seem to mean | |
1042 | * that either DSPCNTR[13] doesn't do anything, or that | |
1043 | * the total FIFO is >= 256 cachelines in size. Either | |
1044 | * way, we don't seem to have to worry about this | |
1045 | * repartitioning as the maximum watermark value the | |
1046 | * register can hold for each plane is lower than the | |
1047 | * minimum FIFO size. | |
1048 | */ | |
1049 | switch (plane_id) { | |
1050 | case PLANE_CURSOR: | |
1051 | return 63; | |
1052 | case PLANE_PRIMARY: | |
1053 | return level == G4X_WM_LEVEL_NORMAL ? 127 : 511; | |
1054 | case PLANE_SPRITE0: | |
1055 | return level == G4X_WM_LEVEL_NORMAL ? 127 : 0; | |
1056 | default: | |
1057 | MISSING_CASE(plane_id); | |
1058 | return 0; | |
1059 | } | |
1060 | } | |
1061 | ||
1062 | static int g4x_fbc_fifo_size(int level) | |
1063 | { | |
1064 | switch (level) { | |
1065 | case G4X_WM_LEVEL_SR: | |
1066 | return 7; | |
1067 | case G4X_WM_LEVEL_HPLL: | |
1068 | return 15; | |
1069 | default: | |
1070 | MISSING_CASE(level); | |
1071 | return 0; | |
1072 | } | |
1073 | } | |
1074 | ||
1075 | static uint16_t g4x_compute_wm(const struct intel_crtc_state *crtc_state, | |
1076 | const struct intel_plane_state *plane_state, | |
1077 | int level) | |
1078 | { | |
1079 | struct intel_plane *plane = to_intel_plane(plane_state->base.plane); | |
1080 | struct drm_i915_private *dev_priv = to_i915(plane->base.dev); | |
1081 | const struct drm_display_mode *adjusted_mode = | |
1082 | &crtc_state->base.adjusted_mode; | |
1083 | int clock, htotal, cpp, width, wm; | |
1084 | int latency = dev_priv->wm.pri_latency[level] * 10; | |
1085 | ||
1086 | if (latency == 0) | |
1087 | return USHRT_MAX; | |
1088 | ||
1089 | if (!intel_wm_plane_visible(crtc_state, plane_state)) | |
1090 | return 0; | |
1091 | ||
1092 | /* | |
1093 | * Not 100% sure which way ELK should go here as the | |
1094 | * spec only says CL/CTG should assume 32bpp and BW | |
1095 | * doesn't need to. But as these things followed the | |
1096 | * mobile vs. desktop lines on gen3 as well, let's | |
1097 | * assume ELK doesn't need this. | |
1098 | * | |
1099 | * The spec also fails to list such a restriction for | |
1100 | * the HPLL watermark, which seems a little strange. | |
1101 | * Let's use 32bpp for the HPLL watermark as well. | |
1102 | */ | |
1103 | if (IS_GM45(dev_priv) && plane->id == PLANE_PRIMARY && | |
1104 | level != G4X_WM_LEVEL_NORMAL) | |
1105 | cpp = 4; | |
1106 | else | |
1107 | cpp = plane_state->base.fb->format->cpp[0]; | |
1108 | ||
1109 | clock = adjusted_mode->crtc_clock; | |
1110 | htotal = adjusted_mode->crtc_htotal; | |
1111 | ||
1112 | if (plane->id == PLANE_CURSOR) | |
1113 | width = plane_state->base.crtc_w; | |
1114 | else | |
1115 | width = drm_rect_width(&plane_state->base.dst); | |
1116 | ||
1117 | if (plane->id == PLANE_CURSOR) { | |
1118 | wm = intel_wm_method2(clock, htotal, width, cpp, latency); | |
1119 | } else if (plane->id == PLANE_PRIMARY && | |
1120 | level == G4X_WM_LEVEL_NORMAL) { | |
1121 | wm = intel_wm_method1(clock, cpp, latency); | |
1122 | } else { | |
1123 | int small, large; | |
1124 | ||
1125 | small = intel_wm_method1(clock, cpp, latency); | |
1126 | large = intel_wm_method2(clock, htotal, width, cpp, latency); | |
1127 | ||
1128 | wm = min(small, large); | |
1129 | } | |
1130 | ||
1131 | wm += g4x_tlb_miss_wa(g4x_plane_fifo_size(plane->id, level), | |
1132 | width, cpp); | |
1133 | ||
1134 | wm = DIV_ROUND_UP(wm, 64) + 2; | |
1135 | ||
1136 | return min_t(int, wm, USHRT_MAX); | |
1137 | } | |
1138 | ||
1139 | static bool g4x_raw_plane_wm_set(struct intel_crtc_state *crtc_state, | |
1140 | int level, enum plane_id plane_id, u16 value) | |
1141 | { | |
1142 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
1143 | bool dirty = false; | |
1144 | ||
1145 | for (; level < intel_wm_num_levels(dev_priv); level++) { | |
1146 | struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level]; | |
1147 | ||
1148 | dirty |= raw->plane[plane_id] != value; | |
1149 | raw->plane[plane_id] = value; | |
1150 | } | |
1151 | ||
1152 | return dirty; | |
1153 | } | |
1154 | ||
1155 | static bool g4x_raw_fbc_wm_set(struct intel_crtc_state *crtc_state, | |
1156 | int level, u16 value) | |
1157 | { | |
1158 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
1159 | bool dirty = false; | |
1160 | ||
1161 | /* NORMAL level doesn't have an FBC watermark */ | |
1162 | level = max(level, G4X_WM_LEVEL_SR); | |
1163 | ||
1164 | for (; level < intel_wm_num_levels(dev_priv); level++) { | |
1165 | struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level]; | |
1166 | ||
1167 | dirty |= raw->fbc != value; | |
1168 | raw->fbc = value; | |
1169 | } | |
1170 | ||
1171 | return dirty; | |
1172 | } | |
1173 | ||
1174 | static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate, | |
1175 | const struct intel_plane_state *pstate, | |
1176 | uint32_t pri_val); | |
1177 | ||
1178 | static bool g4x_raw_plane_wm_compute(struct intel_crtc_state *crtc_state, | |
1179 | const struct intel_plane_state *plane_state) | |
1180 | { | |
1181 | struct intel_plane *plane = to_intel_plane(plane_state->base.plane); | |
1182 | int num_levels = intel_wm_num_levels(to_i915(plane->base.dev)); | |
1183 | enum plane_id plane_id = plane->id; | |
1184 | bool dirty = false; | |
1185 | int level; | |
1186 | ||
1187 | if (!intel_wm_plane_visible(crtc_state, plane_state)) { | |
1188 | dirty |= g4x_raw_plane_wm_set(crtc_state, 0, plane_id, 0); | |
1189 | if (plane_id == PLANE_PRIMARY) | |
1190 | dirty |= g4x_raw_fbc_wm_set(crtc_state, 0, 0); | |
1191 | goto out; | |
1192 | } | |
1193 | ||
1194 | for (level = 0; level < num_levels; level++) { | |
1195 | struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level]; | |
1196 | int wm, max_wm; | |
1197 | ||
1198 | wm = g4x_compute_wm(crtc_state, plane_state, level); | |
1199 | max_wm = g4x_plane_fifo_size(plane_id, level); | |
1200 | ||
1201 | if (wm > max_wm) | |
1202 | break; | |
1203 | ||
1204 | dirty |= raw->plane[plane_id] != wm; | |
1205 | raw->plane[plane_id] = wm; | |
1206 | ||
1207 | if (plane_id != PLANE_PRIMARY || | |
1208 | level == G4X_WM_LEVEL_NORMAL) | |
1209 | continue; | |
1210 | ||
1211 | wm = ilk_compute_fbc_wm(crtc_state, plane_state, | |
1212 | raw->plane[plane_id]); | |
1213 | max_wm = g4x_fbc_fifo_size(level); | |
1214 | ||
1215 | /* | |
1216 | * FBC wm is not mandatory as we | |
1217 | * can always just disable its use. | |
1218 | */ | |
1219 | if (wm > max_wm) | |
1220 | wm = USHRT_MAX; | |
1221 | ||
1222 | dirty |= raw->fbc != wm; | |
1223 | raw->fbc = wm; | |
1224 | } | |
1225 | ||
1226 | /* mark watermarks as invalid */ | |
1227 | dirty |= g4x_raw_plane_wm_set(crtc_state, level, plane_id, USHRT_MAX); | |
1228 | ||
1229 | if (plane_id == PLANE_PRIMARY) | |
1230 | dirty |= g4x_raw_fbc_wm_set(crtc_state, level, USHRT_MAX); | |
1231 | ||
1232 | out: | |
1233 | if (dirty) { | |
1234 | DRM_DEBUG_KMS("%s watermarks: normal=%d, SR=%d, HPLL=%d\n", | |
1235 | plane->base.name, | |
1236 | crtc_state->wm.g4x.raw[G4X_WM_LEVEL_NORMAL].plane[plane_id], | |
1237 | crtc_state->wm.g4x.raw[G4X_WM_LEVEL_SR].plane[plane_id], | |
1238 | crtc_state->wm.g4x.raw[G4X_WM_LEVEL_HPLL].plane[plane_id]); | |
1239 | ||
1240 | if (plane_id == PLANE_PRIMARY) | |
1241 | DRM_DEBUG_KMS("FBC watermarks: SR=%d, HPLL=%d\n", | |
1242 | crtc_state->wm.g4x.raw[G4X_WM_LEVEL_SR].fbc, | |
1243 | crtc_state->wm.g4x.raw[G4X_WM_LEVEL_HPLL].fbc); | |
1244 | } | |
1245 | ||
1246 | return dirty; | |
1247 | } | |
1248 | ||
1249 | static bool g4x_raw_plane_wm_is_valid(const struct intel_crtc_state *crtc_state, | |
1250 | enum plane_id plane_id, int level) | |
1251 | { | |
1252 | const struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level]; | |
1253 | ||
1254 | return raw->plane[plane_id] <= g4x_plane_fifo_size(plane_id, level); | |
1255 | } | |
1256 | ||
1257 | static bool g4x_raw_crtc_wm_is_valid(const struct intel_crtc_state *crtc_state, | |
1258 | int level) | |
1259 | { | |
1260 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
1261 | ||
1262 | if (level > dev_priv->wm.max_level) | |
1263 | return false; | |
1264 | ||
1265 | return g4x_raw_plane_wm_is_valid(crtc_state, PLANE_PRIMARY, level) && | |
1266 | g4x_raw_plane_wm_is_valid(crtc_state, PLANE_SPRITE0, level) && | |
1267 | g4x_raw_plane_wm_is_valid(crtc_state, PLANE_CURSOR, level); | |
1268 | } | |
1269 | ||
1270 | /* mark all levels starting from 'level' as invalid */ | |
1271 | static void g4x_invalidate_wms(struct intel_crtc *crtc, | |
1272 | struct g4x_wm_state *wm_state, int level) | |
1273 | { | |
1274 | if (level <= G4X_WM_LEVEL_NORMAL) { | |
1275 | enum plane_id plane_id; | |
1276 | ||
1277 | for_each_plane_id_on_crtc(crtc, plane_id) | |
1278 | wm_state->wm.plane[plane_id] = USHRT_MAX; | |
1279 | } | |
1280 | ||
1281 | if (level <= G4X_WM_LEVEL_SR) { | |
1282 | wm_state->cxsr = false; | |
1283 | wm_state->sr.cursor = USHRT_MAX; | |
1284 | wm_state->sr.plane = USHRT_MAX; | |
1285 | wm_state->sr.fbc = USHRT_MAX; | |
1286 | } | |
1287 | ||
1288 | if (level <= G4X_WM_LEVEL_HPLL) { | |
1289 | wm_state->hpll_en = false; | |
1290 | wm_state->hpll.cursor = USHRT_MAX; | |
1291 | wm_state->hpll.plane = USHRT_MAX; | |
1292 | wm_state->hpll.fbc = USHRT_MAX; | |
1293 | } | |
1294 | } | |
1295 | ||
1296 | static int g4x_compute_pipe_wm(struct intel_crtc_state *crtc_state) | |
1297 | { | |
1298 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); | |
1299 | struct intel_atomic_state *state = | |
1300 | to_intel_atomic_state(crtc_state->base.state); | |
1301 | struct g4x_wm_state *wm_state = &crtc_state->wm.g4x.optimal; | |
1302 | int num_active_planes = hweight32(crtc_state->active_planes & | |
1303 | ~BIT(PLANE_CURSOR)); | |
1304 | const struct g4x_pipe_wm *raw; | |
1305 | struct intel_plane_state *plane_state; | |
1306 | struct intel_plane *plane; | |
1307 | enum plane_id plane_id; | |
1308 | int i, level; | |
1309 | unsigned int dirty = 0; | |
1310 | ||
1311 | for_each_intel_plane_in_state(state, plane, plane_state, i) { | |
1312 | const struct intel_plane_state *old_plane_state = | |
1313 | to_intel_plane_state(plane->base.state); | |
1314 | ||
1315 | if (plane_state->base.crtc != &crtc->base && | |
1316 | old_plane_state->base.crtc != &crtc->base) | |
1317 | continue; | |
1318 | ||
1319 | if (g4x_raw_plane_wm_compute(crtc_state, plane_state)) | |
1320 | dirty |= BIT(plane->id); | |
1321 | } | |
1322 | ||
1323 | if (!dirty) | |
1324 | return 0; | |
1325 | ||
1326 | level = G4X_WM_LEVEL_NORMAL; | |
1327 | if (!g4x_raw_crtc_wm_is_valid(crtc_state, level)) | |
1328 | goto out; | |
1329 | ||
1330 | raw = &crtc_state->wm.g4x.raw[level]; | |
1331 | for_each_plane_id_on_crtc(crtc, plane_id) | |
1332 | wm_state->wm.plane[plane_id] = raw->plane[plane_id]; | |
1333 | ||
1334 | level = G4X_WM_LEVEL_SR; | |
1335 | ||
1336 | if (!g4x_raw_crtc_wm_is_valid(crtc_state, level)) | |
1337 | goto out; | |
1338 | ||
1339 | raw = &crtc_state->wm.g4x.raw[level]; | |
1340 | wm_state->sr.plane = raw->plane[PLANE_PRIMARY]; | |
1341 | wm_state->sr.cursor = raw->plane[PLANE_CURSOR]; | |
1342 | wm_state->sr.fbc = raw->fbc; | |
1343 | ||
1344 | wm_state->cxsr = num_active_planes == BIT(PLANE_PRIMARY); | |
1345 | ||
1346 | level = G4X_WM_LEVEL_HPLL; | |
1347 | ||
1348 | if (!g4x_raw_crtc_wm_is_valid(crtc_state, level)) | |
1349 | goto out; | |
1350 | ||
1351 | raw = &crtc_state->wm.g4x.raw[level]; | |
1352 | wm_state->hpll.plane = raw->plane[PLANE_PRIMARY]; | |
1353 | wm_state->hpll.cursor = raw->plane[PLANE_CURSOR]; | |
1354 | wm_state->hpll.fbc = raw->fbc; | |
1355 | ||
1356 | wm_state->hpll_en = wm_state->cxsr; | |
1357 | ||
1358 | level++; | |
1359 | ||
1360 | out: | |
1361 | if (level == G4X_WM_LEVEL_NORMAL) | |
1362 | return -EINVAL; | |
1363 | ||
1364 | /* invalidate the higher levels */ | |
1365 | g4x_invalidate_wms(crtc, wm_state, level); | |
1366 | ||
1367 | /* | |
1368 | * Determine if the FBC watermark(s) can be used. IF | |
1369 | * this isn't the case we prefer to disable the FBC | |
1370 | ( watermark(s) rather than disable the SR/HPLL | |
1371 | * level(s) entirely. | |
1372 | */ | |
1373 | wm_state->fbc_en = level > G4X_WM_LEVEL_NORMAL; | |
1374 | ||
1375 | if (level >= G4X_WM_LEVEL_SR && | |
1376 | wm_state->sr.fbc > g4x_fbc_fifo_size(G4X_WM_LEVEL_SR)) | |
1377 | wm_state->fbc_en = false; | |
1378 | else if (level >= G4X_WM_LEVEL_HPLL && | |
1379 | wm_state->hpll.fbc > g4x_fbc_fifo_size(G4X_WM_LEVEL_HPLL)) | |
1380 | wm_state->fbc_en = false; | |
1381 | ||
1382 | return 0; | |
1383 | } | |
1384 | ||
1385 | static int g4x_compute_intermediate_wm(struct drm_device *dev, | |
1386 | struct intel_crtc *crtc, | |
1387 | struct intel_crtc_state *crtc_state) | |
1388 | { | |
1389 | struct g4x_wm_state *intermediate = &crtc_state->wm.g4x.intermediate; | |
1390 | const struct g4x_wm_state *optimal = &crtc_state->wm.g4x.optimal; | |
1391 | const struct g4x_wm_state *active = &crtc->wm.active.g4x; | |
1392 | enum plane_id plane_id; | |
1393 | ||
1394 | intermediate->cxsr = optimal->cxsr && active->cxsr && | |
1395 | !crtc_state->disable_cxsr; | |
1396 | intermediate->hpll_en = optimal->hpll_en && active->hpll_en && | |
1397 | !crtc_state->disable_cxsr; | |
1398 | intermediate->fbc_en = optimal->fbc_en && active->fbc_en; | |
1399 | ||
1400 | for_each_plane_id_on_crtc(crtc, plane_id) { | |
1401 | intermediate->wm.plane[plane_id] = | |
1402 | max(optimal->wm.plane[plane_id], | |
1403 | active->wm.plane[plane_id]); | |
1404 | ||
1405 | WARN_ON(intermediate->wm.plane[plane_id] > | |
1406 | g4x_plane_fifo_size(plane_id, G4X_WM_LEVEL_NORMAL)); | |
1407 | } | |
1408 | ||
1409 | intermediate->sr.plane = max(optimal->sr.plane, | |
1410 | active->sr.plane); | |
1411 | intermediate->sr.cursor = max(optimal->sr.cursor, | |
1412 | active->sr.cursor); | |
1413 | intermediate->sr.fbc = max(optimal->sr.fbc, | |
1414 | active->sr.fbc); | |
1415 | ||
1416 | intermediate->hpll.plane = max(optimal->hpll.plane, | |
1417 | active->hpll.plane); | |
1418 | intermediate->hpll.cursor = max(optimal->hpll.cursor, | |
1419 | active->hpll.cursor); | |
1420 | intermediate->hpll.fbc = max(optimal->hpll.fbc, | |
1421 | active->hpll.fbc); | |
1422 | ||
1423 | WARN_ON((intermediate->sr.plane > | |
1424 | g4x_plane_fifo_size(PLANE_PRIMARY, G4X_WM_LEVEL_SR) || | |
1425 | intermediate->sr.cursor > | |
1426 | g4x_plane_fifo_size(PLANE_CURSOR, G4X_WM_LEVEL_SR)) && | |
1427 | intermediate->cxsr); | |
1428 | WARN_ON((intermediate->sr.plane > | |
1429 | g4x_plane_fifo_size(PLANE_PRIMARY, G4X_WM_LEVEL_HPLL) || | |
1430 | intermediate->sr.cursor > | |
1431 | g4x_plane_fifo_size(PLANE_CURSOR, G4X_WM_LEVEL_HPLL)) && | |
1432 | intermediate->hpll_en); | |
1433 | ||
1434 | WARN_ON(intermediate->sr.fbc > g4x_fbc_fifo_size(1) && | |
1435 | intermediate->fbc_en && intermediate->cxsr); | |
1436 | WARN_ON(intermediate->hpll.fbc > g4x_fbc_fifo_size(2) && | |
1437 | intermediate->fbc_en && intermediate->hpll_en); | |
1438 | ||
1439 | /* | |
1440 | * If our intermediate WM are identical to the final WM, then we can | |
1441 | * omit the post-vblank programming; only update if it's different. | |
1442 | */ | |
1443 | if (memcmp(intermediate, optimal, sizeof(*intermediate)) != 0) | |
1444 | crtc_state->wm.need_postvbl_update = true; | |
1445 | ||
1446 | return 0; | |
1447 | } | |
1448 | ||
1449 | static void g4x_merge_wm(struct drm_i915_private *dev_priv, | |
1450 | struct g4x_wm_values *wm) | |
1451 | { | |
1452 | struct intel_crtc *crtc; | |
1453 | int num_active_crtcs = 0; | |
1454 | ||
1455 | wm->cxsr = true; | |
1456 | wm->hpll_en = true; | |
1457 | wm->fbc_en = true; | |
1458 | ||
1459 | for_each_intel_crtc(&dev_priv->drm, crtc) { | |
1460 | const struct g4x_wm_state *wm_state = &crtc->wm.active.g4x; | |
1461 | ||
1462 | if (!crtc->active) | |
1463 | continue; | |
1464 | ||
1465 | if (!wm_state->cxsr) | |
1466 | wm->cxsr = false; | |
1467 | if (!wm_state->hpll_en) | |
1468 | wm->hpll_en = false; | |
1469 | if (!wm_state->fbc_en) | |
1470 | wm->fbc_en = false; | |
1471 | ||
1472 | num_active_crtcs++; | |
1473 | } | |
1474 | ||
1475 | if (num_active_crtcs != 1) { | |
1476 | wm->cxsr = false; | |
1477 | wm->hpll_en = false; | |
1478 | wm->fbc_en = false; | |
1479 | } | |
1480 | ||
1481 | for_each_intel_crtc(&dev_priv->drm, crtc) { | |
1482 | const struct g4x_wm_state *wm_state = &crtc->wm.active.g4x; | |
1483 | enum pipe pipe = crtc->pipe; | |
1484 | ||
1485 | wm->pipe[pipe] = wm_state->wm; | |
1486 | if (crtc->active && wm->cxsr) | |
1487 | wm->sr = wm_state->sr; | |
1488 | if (crtc->active && wm->hpll_en) | |
1489 | wm->hpll = wm_state->hpll; | |
1490 | } | |
1491 | } | |
1492 | ||
1493 | static void g4x_program_watermarks(struct drm_i915_private *dev_priv) | |
1494 | { | |
1495 | struct g4x_wm_values *old_wm = &dev_priv->wm.g4x; | |
1496 | struct g4x_wm_values new_wm = {}; | |
1497 | ||
1498 | g4x_merge_wm(dev_priv, &new_wm); | |
1499 | ||
1500 | if (memcmp(old_wm, &new_wm, sizeof(new_wm)) == 0) | |
1501 | return; | |
1502 | ||
1503 | if (is_disabling(old_wm->cxsr, new_wm.cxsr, true)) | |
1504 | _intel_set_memory_cxsr(dev_priv, false); | |
1505 | ||
1506 | g4x_write_wm_values(dev_priv, &new_wm); | |
1507 | ||
1508 | if (is_enabling(old_wm->cxsr, new_wm.cxsr, true)) | |
1509 | _intel_set_memory_cxsr(dev_priv, true); | |
1510 | ||
1511 | *old_wm = new_wm; | |
1512 | } | |
1513 | ||
1514 | static void g4x_initial_watermarks(struct intel_atomic_state *state, | |
1515 | struct intel_crtc_state *crtc_state) | |
1516 | { | |
1517 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
1518 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); | |
1519 | ||
1520 | mutex_lock(&dev_priv->wm.wm_mutex); | |
1521 | crtc->wm.active.g4x = crtc_state->wm.g4x.intermediate; | |
1522 | g4x_program_watermarks(dev_priv); | |
1523 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
1524 | } | |
1525 | ||
1526 | static void g4x_optimize_watermarks(struct intel_atomic_state *state, | |
1527 | struct intel_crtc_state *crtc_state) | |
1528 | { | |
1529 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
1530 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc_state->base.crtc); | |
1531 | ||
1532 | if (!crtc_state->wm.need_postvbl_update) | |
1533 | return; | |
1534 | ||
1535 | mutex_lock(&dev_priv->wm.wm_mutex); | |
1536 | intel_crtc->wm.active.g4x = crtc_state->wm.g4x.optimal; | |
1537 | g4x_program_watermarks(dev_priv); | |
1538 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
1539 | } | |
1540 | ||
262cd2e1 VS |
1541 | /* latency must be in 0.1us units. */ |
1542 | static unsigned int vlv_wm_method2(unsigned int pixel_rate, | |
baf69ca8 VS |
1543 | unsigned int htotal, |
1544 | unsigned int width, | |
ac484963 | 1545 | unsigned int cpp, |
262cd2e1 VS |
1546 | unsigned int latency) |
1547 | { | |
1548 | unsigned int ret; | |
1549 | ||
baf69ca8 VS |
1550 | ret = intel_wm_method2(pixel_rate, htotal, |
1551 | width, cpp, latency); | |
262cd2e1 VS |
1552 | ret = DIV_ROUND_UP(ret, 64); |
1553 | ||
1554 | return ret; | |
1555 | } | |
1556 | ||
bb726519 | 1557 | static void vlv_setup_wm_latency(struct drm_i915_private *dev_priv) |
262cd2e1 | 1558 | { |
262cd2e1 VS |
1559 | /* all latencies in usec */ |
1560 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3; | |
1561 | ||
58590c14 VS |
1562 | dev_priv->wm.max_level = VLV_WM_LEVEL_PM2; |
1563 | ||
262cd2e1 VS |
1564 | if (IS_CHERRYVIEW(dev_priv)) { |
1565 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12; | |
1566 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33; | |
58590c14 VS |
1567 | |
1568 | dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS; | |
262cd2e1 VS |
1569 | } |
1570 | } | |
1571 | ||
e339d67e VS |
1572 | static uint16_t vlv_compute_wm_level(const struct intel_crtc_state *crtc_state, |
1573 | const struct intel_plane_state *plane_state, | |
262cd2e1 VS |
1574 | int level) |
1575 | { | |
e339d67e | 1576 | struct intel_plane *plane = to_intel_plane(plane_state->base.plane); |
262cd2e1 | 1577 | struct drm_i915_private *dev_priv = to_i915(plane->base.dev); |
e339d67e VS |
1578 | const struct drm_display_mode *adjusted_mode = |
1579 | &crtc_state->base.adjusted_mode; | |
ac484963 | 1580 | int clock, htotal, cpp, width, wm; |
262cd2e1 VS |
1581 | |
1582 | if (dev_priv->wm.pri_latency[level] == 0) | |
1583 | return USHRT_MAX; | |
1584 | ||
a07102f1 | 1585 | if (!intel_wm_plane_visible(crtc_state, plane_state)) |
262cd2e1 VS |
1586 | return 0; |
1587 | ||
ef426c10 | 1588 | cpp = plane_state->base.fb->format->cpp[0]; |
e339d67e VS |
1589 | clock = adjusted_mode->crtc_clock; |
1590 | htotal = adjusted_mode->crtc_htotal; | |
1591 | width = crtc_state->pipe_src_w; | |
262cd2e1 | 1592 | |
709f3fc9 | 1593 | if (plane->id == PLANE_CURSOR) { |
262cd2e1 VS |
1594 | /* |
1595 | * FIXME the formula gives values that are | |
1596 | * too big for the cursor FIFO, and hence we | |
1597 | * would never be able to use cursors. For | |
1598 | * now just hardcode the watermark. | |
1599 | */ | |
1600 | wm = 63; | |
1601 | } else { | |
ac484963 | 1602 | wm = vlv_wm_method2(clock, htotal, width, cpp, |
262cd2e1 VS |
1603 | dev_priv->wm.pri_latency[level] * 10); |
1604 | } | |
1605 | ||
1606 | return min_t(int, wm, USHRT_MAX); | |
1607 | } | |
1608 | ||
1a10ae6b VS |
1609 | static bool vlv_need_sprite0_fifo_workaround(unsigned int active_planes) |
1610 | { | |
1611 | return (active_planes & (BIT(PLANE_SPRITE0) | | |
1612 | BIT(PLANE_SPRITE1))) == BIT(PLANE_SPRITE1); | |
1613 | } | |
1614 | ||
5012e604 | 1615 | static int vlv_compute_fifo(struct intel_crtc_state *crtc_state) |
54f1b6e1 | 1616 | { |
855c79f5 | 1617 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); |
114d7dc0 | 1618 | const struct g4x_pipe_wm *raw = |
5012e604 | 1619 | &crtc_state->wm.vlv.raw[VLV_WM_LEVEL_PM2]; |
814e7f0b | 1620 | struct vlv_fifo_state *fifo_state = &crtc_state->wm.vlv.fifo_state; |
5012e604 VS |
1621 | unsigned int active_planes = crtc_state->active_planes & ~BIT(PLANE_CURSOR); |
1622 | int num_active_planes = hweight32(active_planes); | |
1623 | const int fifo_size = 511; | |
54f1b6e1 | 1624 | int fifo_extra, fifo_left = fifo_size; |
1a10ae6b | 1625 | int sprite0_fifo_extra = 0; |
5012e604 VS |
1626 | unsigned int total_rate; |
1627 | enum plane_id plane_id; | |
54f1b6e1 | 1628 | |
1a10ae6b VS |
1629 | /* |
1630 | * When enabling sprite0 after sprite1 has already been enabled | |
1631 | * we tend to get an underrun unless sprite0 already has some | |
1632 | * FIFO space allcoated. Hence we always allocate at least one | |
1633 | * cacheline for sprite0 whenever sprite1 is enabled. | |
1634 | * | |
1635 | * All other plane enable sequences appear immune to this problem. | |
1636 | */ | |
1637 | if (vlv_need_sprite0_fifo_workaround(active_planes)) | |
1638 | sprite0_fifo_extra = 1; | |
1639 | ||
5012e604 VS |
1640 | total_rate = raw->plane[PLANE_PRIMARY] + |
1641 | raw->plane[PLANE_SPRITE0] + | |
1a10ae6b VS |
1642 | raw->plane[PLANE_SPRITE1] + |
1643 | sprite0_fifo_extra; | |
54f1b6e1 | 1644 | |
5012e604 VS |
1645 | if (total_rate > fifo_size) |
1646 | return -EINVAL; | |
54f1b6e1 | 1647 | |
5012e604 VS |
1648 | if (total_rate == 0) |
1649 | total_rate = 1; | |
54f1b6e1 | 1650 | |
5012e604 | 1651 | for_each_plane_id_on_crtc(crtc, plane_id) { |
54f1b6e1 VS |
1652 | unsigned int rate; |
1653 | ||
5012e604 VS |
1654 | if ((active_planes & BIT(plane_id)) == 0) { |
1655 | fifo_state->plane[plane_id] = 0; | |
54f1b6e1 VS |
1656 | continue; |
1657 | } | |
1658 | ||
5012e604 VS |
1659 | rate = raw->plane[plane_id]; |
1660 | fifo_state->plane[plane_id] = fifo_size * rate / total_rate; | |
1661 | fifo_left -= fifo_state->plane[plane_id]; | |
54f1b6e1 VS |
1662 | } |
1663 | ||
1a10ae6b VS |
1664 | fifo_state->plane[PLANE_SPRITE0] += sprite0_fifo_extra; |
1665 | fifo_left -= sprite0_fifo_extra; | |
1666 | ||
5012e604 VS |
1667 | fifo_state->plane[PLANE_CURSOR] = 63; |
1668 | ||
1669 | fifo_extra = DIV_ROUND_UP(fifo_left, num_active_planes ?: 1); | |
54f1b6e1 VS |
1670 | |
1671 | /* spread the remainder evenly */ | |
5012e604 | 1672 | for_each_plane_id_on_crtc(crtc, plane_id) { |
54f1b6e1 VS |
1673 | int plane_extra; |
1674 | ||
1675 | if (fifo_left == 0) | |
1676 | break; | |
1677 | ||
5012e604 | 1678 | if ((active_planes & BIT(plane_id)) == 0) |
54f1b6e1 VS |
1679 | continue; |
1680 | ||
1681 | plane_extra = min(fifo_extra, fifo_left); | |
5012e604 | 1682 | fifo_state->plane[plane_id] += plane_extra; |
54f1b6e1 VS |
1683 | fifo_left -= plane_extra; |
1684 | } | |
1685 | ||
5012e604 VS |
1686 | WARN_ON(active_planes != 0 && fifo_left != 0); |
1687 | ||
1688 | /* give it all to the first plane if none are active */ | |
1689 | if (active_planes == 0) { | |
1690 | WARN_ON(fifo_left != fifo_size); | |
1691 | fifo_state->plane[PLANE_PRIMARY] = fifo_left; | |
1692 | } | |
1693 | ||
1694 | return 0; | |
54f1b6e1 VS |
1695 | } |
1696 | ||
ff32c54e VS |
1697 | /* mark all levels starting from 'level' as invalid */ |
1698 | static void vlv_invalidate_wms(struct intel_crtc *crtc, | |
1699 | struct vlv_wm_state *wm_state, int level) | |
1700 | { | |
1701 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); | |
1702 | ||
6d5019b6 | 1703 | for (; level < intel_wm_num_levels(dev_priv); level++) { |
ff32c54e VS |
1704 | enum plane_id plane_id; |
1705 | ||
1706 | for_each_plane_id_on_crtc(crtc, plane_id) | |
1707 | wm_state->wm[level].plane[plane_id] = USHRT_MAX; | |
1708 | ||
1709 | wm_state->sr[level].cursor = USHRT_MAX; | |
1710 | wm_state->sr[level].plane = USHRT_MAX; | |
1711 | } | |
1712 | } | |
1713 | ||
26cca0e5 VS |
1714 | static u16 vlv_invert_wm_value(u16 wm, u16 fifo_size) |
1715 | { | |
1716 | if (wm > fifo_size) | |
1717 | return USHRT_MAX; | |
1718 | else | |
1719 | return fifo_size - wm; | |
1720 | } | |
1721 | ||
ff32c54e VS |
1722 | /* |
1723 | * Starting from 'level' set all higher | |
1724 | * levels to 'value' in the "raw" watermarks. | |
1725 | */ | |
236c48e6 | 1726 | static bool vlv_raw_plane_wm_set(struct intel_crtc_state *crtc_state, |
ff32c54e | 1727 | int level, enum plane_id plane_id, u16 value) |
262cd2e1 | 1728 | { |
ff32c54e | 1729 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); |
6d5019b6 | 1730 | int num_levels = intel_wm_num_levels(dev_priv); |
236c48e6 | 1731 | bool dirty = false; |
262cd2e1 | 1732 | |
ff32c54e | 1733 | for (; level < num_levels; level++) { |
114d7dc0 | 1734 | struct g4x_pipe_wm *raw = &crtc_state->wm.vlv.raw[level]; |
262cd2e1 | 1735 | |
236c48e6 | 1736 | dirty |= raw->plane[plane_id] != value; |
ff32c54e | 1737 | raw->plane[plane_id] = value; |
262cd2e1 | 1738 | } |
236c48e6 VS |
1739 | |
1740 | return dirty; | |
262cd2e1 VS |
1741 | } |
1742 | ||
77d14ee4 VS |
1743 | static bool vlv_raw_plane_wm_compute(struct intel_crtc_state *crtc_state, |
1744 | const struct intel_plane_state *plane_state) | |
262cd2e1 | 1745 | { |
ff32c54e VS |
1746 | struct intel_plane *plane = to_intel_plane(plane_state->base.plane); |
1747 | enum plane_id plane_id = plane->id; | |
6d5019b6 | 1748 | int num_levels = intel_wm_num_levels(to_i915(plane->base.dev)); |
262cd2e1 | 1749 | int level; |
236c48e6 | 1750 | bool dirty = false; |
262cd2e1 | 1751 | |
a07102f1 | 1752 | if (!intel_wm_plane_visible(crtc_state, plane_state)) { |
236c48e6 VS |
1753 | dirty |= vlv_raw_plane_wm_set(crtc_state, 0, plane_id, 0); |
1754 | goto out; | |
ff32c54e | 1755 | } |
262cd2e1 | 1756 | |
ff32c54e | 1757 | for (level = 0; level < num_levels; level++) { |
114d7dc0 | 1758 | struct g4x_pipe_wm *raw = &crtc_state->wm.vlv.raw[level]; |
ff32c54e VS |
1759 | int wm = vlv_compute_wm_level(crtc_state, plane_state, level); |
1760 | int max_wm = plane_id == PLANE_CURSOR ? 63 : 511; | |
262cd2e1 | 1761 | |
ff32c54e VS |
1762 | if (wm > max_wm) |
1763 | break; | |
262cd2e1 | 1764 | |
236c48e6 | 1765 | dirty |= raw->plane[plane_id] != wm; |
ff32c54e VS |
1766 | raw->plane[plane_id] = wm; |
1767 | } | |
262cd2e1 | 1768 | |
ff32c54e | 1769 | /* mark all higher levels as invalid */ |
236c48e6 | 1770 | dirty |= vlv_raw_plane_wm_set(crtc_state, level, plane_id, USHRT_MAX); |
262cd2e1 | 1771 | |
236c48e6 VS |
1772 | out: |
1773 | if (dirty) | |
57a6528a | 1774 | DRM_DEBUG_KMS("%s watermarks: PM2=%d, PM5=%d, DDR DVFS=%d\n", |
236c48e6 VS |
1775 | plane->base.name, |
1776 | crtc_state->wm.vlv.raw[VLV_WM_LEVEL_PM2].plane[plane_id], | |
1777 | crtc_state->wm.vlv.raw[VLV_WM_LEVEL_PM5].plane[plane_id], | |
1778 | crtc_state->wm.vlv.raw[VLV_WM_LEVEL_DDR_DVFS].plane[plane_id]); | |
1779 | ||
1780 | return dirty; | |
ff32c54e | 1781 | } |
262cd2e1 | 1782 | |
77d14ee4 VS |
1783 | static bool vlv_raw_plane_wm_is_valid(const struct intel_crtc_state *crtc_state, |
1784 | enum plane_id plane_id, int level) | |
ff32c54e | 1785 | { |
114d7dc0 | 1786 | const struct g4x_pipe_wm *raw = |
ff32c54e VS |
1787 | &crtc_state->wm.vlv.raw[level]; |
1788 | const struct vlv_fifo_state *fifo_state = | |
1789 | &crtc_state->wm.vlv.fifo_state; | |
262cd2e1 | 1790 | |
ff32c54e VS |
1791 | return raw->plane[plane_id] <= fifo_state->plane[plane_id]; |
1792 | } | |
262cd2e1 | 1793 | |
77d14ee4 | 1794 | static bool vlv_raw_crtc_wm_is_valid(const struct intel_crtc_state *crtc_state, int level) |
ff32c54e | 1795 | { |
77d14ee4 VS |
1796 | return vlv_raw_plane_wm_is_valid(crtc_state, PLANE_PRIMARY, level) && |
1797 | vlv_raw_plane_wm_is_valid(crtc_state, PLANE_SPRITE0, level) && | |
1798 | vlv_raw_plane_wm_is_valid(crtc_state, PLANE_SPRITE1, level) && | |
1799 | vlv_raw_plane_wm_is_valid(crtc_state, PLANE_CURSOR, level); | |
ff32c54e VS |
1800 | } |
1801 | ||
1802 | static int vlv_compute_pipe_wm(struct intel_crtc_state *crtc_state) | |
1803 | { | |
1804 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); | |
1805 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); | |
1806 | struct intel_atomic_state *state = | |
1807 | to_intel_atomic_state(crtc_state->base.state); | |
1808 | struct vlv_wm_state *wm_state = &crtc_state->wm.vlv.optimal; | |
1809 | const struct vlv_fifo_state *fifo_state = | |
1810 | &crtc_state->wm.vlv.fifo_state; | |
1811 | int num_active_planes = hweight32(crtc_state->active_planes & | |
1812 | ~BIT(PLANE_CURSOR)); | |
236c48e6 | 1813 | bool needs_modeset = drm_atomic_crtc_needs_modeset(&crtc_state->base); |
ff32c54e VS |
1814 | struct intel_plane_state *plane_state; |
1815 | struct intel_plane *plane; | |
1816 | enum plane_id plane_id; | |
1817 | int level, ret, i; | |
236c48e6 | 1818 | unsigned int dirty = 0; |
ff32c54e VS |
1819 | |
1820 | for_each_intel_plane_in_state(state, plane, plane_state, i) { | |
1821 | const struct intel_plane_state *old_plane_state = | |
1822 | to_intel_plane_state(plane->base.state); | |
1823 | ||
1824 | if (plane_state->base.crtc != &crtc->base && | |
1825 | old_plane_state->base.crtc != &crtc->base) | |
1826 | continue; | |
262cd2e1 | 1827 | |
77d14ee4 | 1828 | if (vlv_raw_plane_wm_compute(crtc_state, plane_state)) |
236c48e6 VS |
1829 | dirty |= BIT(plane->id); |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * DSPARB registers may have been reset due to the | |
1834 | * power well being turned off. Make sure we restore | |
1835 | * them to a consistent state even if no primary/sprite | |
1836 | * planes are initially active. | |
1837 | */ | |
1838 | if (needs_modeset) | |
1839 | crtc_state->fifo_changed = true; | |
1840 | ||
1841 | if (!dirty) | |
1842 | return 0; | |
1843 | ||
1844 | /* cursor changes don't warrant a FIFO recompute */ | |
1845 | if (dirty & ~BIT(PLANE_CURSOR)) { | |
1846 | const struct intel_crtc_state *old_crtc_state = | |
1847 | to_intel_crtc_state(crtc->base.state); | |
1848 | const struct vlv_fifo_state *old_fifo_state = | |
1849 | &old_crtc_state->wm.vlv.fifo_state; | |
1850 | ||
1851 | ret = vlv_compute_fifo(crtc_state); | |
1852 | if (ret) | |
1853 | return ret; | |
1854 | ||
1855 | if (needs_modeset || | |
1856 | memcmp(old_fifo_state, fifo_state, | |
1857 | sizeof(*fifo_state)) != 0) | |
1858 | crtc_state->fifo_changed = true; | |
5012e604 | 1859 | } |
262cd2e1 | 1860 | |
ff32c54e | 1861 | /* initially allow all levels */ |
6d5019b6 | 1862 | wm_state->num_levels = intel_wm_num_levels(dev_priv); |
ff32c54e VS |
1863 | /* |
1864 | * Note that enabling cxsr with no primary/sprite planes | |
1865 | * enabled can wedge the pipe. Hence we only allow cxsr | |
1866 | * with exactly one enabled primary/sprite plane. | |
1867 | */ | |
5eeb798b | 1868 | wm_state->cxsr = crtc->pipe != PIPE_C && num_active_planes == 1; |
ff32c54e | 1869 | |
5012e604 | 1870 | for (level = 0; level < wm_state->num_levels; level++) { |
114d7dc0 | 1871 | const struct g4x_pipe_wm *raw = &crtc_state->wm.vlv.raw[level]; |
ff32c54e | 1872 | const int sr_fifo_size = INTEL_INFO(dev_priv)->num_pipes * 512 - 1; |
5012e604 | 1873 | |
77d14ee4 | 1874 | if (!vlv_raw_crtc_wm_is_valid(crtc_state, level)) |
ff32c54e | 1875 | break; |
5012e604 | 1876 | |
ff32c54e VS |
1877 | for_each_plane_id_on_crtc(crtc, plane_id) { |
1878 | wm_state->wm[level].plane[plane_id] = | |
1879 | vlv_invert_wm_value(raw->plane[plane_id], | |
1880 | fifo_state->plane[plane_id]); | |
1881 | } | |
1882 | ||
1883 | wm_state->sr[level].plane = | |
1884 | vlv_invert_wm_value(max3(raw->plane[PLANE_PRIMARY], | |
5012e604 | 1885 | raw->plane[PLANE_SPRITE0], |
ff32c54e VS |
1886 | raw->plane[PLANE_SPRITE1]), |
1887 | sr_fifo_size); | |
262cd2e1 | 1888 | |
ff32c54e VS |
1889 | wm_state->sr[level].cursor = |
1890 | vlv_invert_wm_value(raw->plane[PLANE_CURSOR], | |
1891 | 63); | |
262cd2e1 VS |
1892 | } |
1893 | ||
ff32c54e VS |
1894 | if (level == 0) |
1895 | return -EINVAL; | |
1896 | ||
1897 | /* limit to only levels we can actually handle */ | |
1898 | wm_state->num_levels = level; | |
1899 | ||
1900 | /* invalidate the higher levels */ | |
1901 | vlv_invalidate_wms(crtc, wm_state, level); | |
1902 | ||
1903 | return 0; | |
262cd2e1 VS |
1904 | } |
1905 | ||
54f1b6e1 VS |
1906 | #define VLV_FIFO(plane, value) \ |
1907 | (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV) | |
1908 | ||
ff32c54e VS |
1909 | static void vlv_atomic_update_fifo(struct intel_atomic_state *state, |
1910 | struct intel_crtc_state *crtc_state) | |
54f1b6e1 | 1911 | { |
814e7f0b | 1912 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); |
f07d43d2 | 1913 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); |
814e7f0b VS |
1914 | const struct vlv_fifo_state *fifo_state = |
1915 | &crtc_state->wm.vlv.fifo_state; | |
f07d43d2 | 1916 | int sprite0_start, sprite1_start, fifo_size; |
54f1b6e1 | 1917 | |
236c48e6 VS |
1918 | if (!crtc_state->fifo_changed) |
1919 | return; | |
1920 | ||
f07d43d2 VS |
1921 | sprite0_start = fifo_state->plane[PLANE_PRIMARY]; |
1922 | sprite1_start = fifo_state->plane[PLANE_SPRITE0] + sprite0_start; | |
1923 | fifo_size = fifo_state->plane[PLANE_SPRITE1] + sprite1_start; | |
54f1b6e1 | 1924 | |
f07d43d2 VS |
1925 | WARN_ON(fifo_state->plane[PLANE_CURSOR] != 63); |
1926 | WARN_ON(fifo_size != 511); | |
54f1b6e1 | 1927 | |
c137d660 VS |
1928 | trace_vlv_fifo_size(crtc, sprite0_start, sprite1_start, fifo_size); |
1929 | ||
44e921d4 VS |
1930 | /* |
1931 | * uncore.lock serves a double purpose here. It allows us to | |
1932 | * use the less expensive I915_{READ,WRITE}_FW() functions, and | |
1933 | * it protects the DSPARB registers from getting clobbered by | |
1934 | * parallel updates from multiple pipes. | |
1935 | * | |
1936 | * intel_pipe_update_start() has already disabled interrupts | |
1937 | * for us, so a plain spin_lock() is sufficient here. | |
1938 | */ | |
1939 | spin_lock(&dev_priv->uncore.lock); | |
467a14d9 | 1940 | |
54f1b6e1 VS |
1941 | switch (crtc->pipe) { |
1942 | uint32_t dsparb, dsparb2, dsparb3; | |
1943 | case PIPE_A: | |
44e921d4 VS |
1944 | dsparb = I915_READ_FW(DSPARB); |
1945 | dsparb2 = I915_READ_FW(DSPARB2); | |
54f1b6e1 VS |
1946 | |
1947 | dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) | | |
1948 | VLV_FIFO(SPRITEB, 0xff)); | |
1949 | dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) | | |
1950 | VLV_FIFO(SPRITEB, sprite1_start)); | |
1951 | ||
1952 | dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) | | |
1953 | VLV_FIFO(SPRITEB_HI, 0x1)); | |
1954 | dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) | | |
1955 | VLV_FIFO(SPRITEB_HI, sprite1_start >> 8)); | |
1956 | ||
44e921d4 VS |
1957 | I915_WRITE_FW(DSPARB, dsparb); |
1958 | I915_WRITE_FW(DSPARB2, dsparb2); | |
54f1b6e1 VS |
1959 | break; |
1960 | case PIPE_B: | |
44e921d4 VS |
1961 | dsparb = I915_READ_FW(DSPARB); |
1962 | dsparb2 = I915_READ_FW(DSPARB2); | |
54f1b6e1 VS |
1963 | |
1964 | dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) | | |
1965 | VLV_FIFO(SPRITED, 0xff)); | |
1966 | dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) | | |
1967 | VLV_FIFO(SPRITED, sprite1_start)); | |
1968 | ||
1969 | dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) | | |
1970 | VLV_FIFO(SPRITED_HI, 0xff)); | |
1971 | dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) | | |
1972 | VLV_FIFO(SPRITED_HI, sprite1_start >> 8)); | |
1973 | ||
44e921d4 VS |
1974 | I915_WRITE_FW(DSPARB, dsparb); |
1975 | I915_WRITE_FW(DSPARB2, dsparb2); | |
54f1b6e1 VS |
1976 | break; |
1977 | case PIPE_C: | |
44e921d4 VS |
1978 | dsparb3 = I915_READ_FW(DSPARB3); |
1979 | dsparb2 = I915_READ_FW(DSPARB2); | |
54f1b6e1 VS |
1980 | |
1981 | dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) | | |
1982 | VLV_FIFO(SPRITEF, 0xff)); | |
1983 | dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) | | |
1984 | VLV_FIFO(SPRITEF, sprite1_start)); | |
1985 | ||
1986 | dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) | | |
1987 | VLV_FIFO(SPRITEF_HI, 0xff)); | |
1988 | dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) | | |
1989 | VLV_FIFO(SPRITEF_HI, sprite1_start >> 8)); | |
1990 | ||
44e921d4 VS |
1991 | I915_WRITE_FW(DSPARB3, dsparb3); |
1992 | I915_WRITE_FW(DSPARB2, dsparb2); | |
54f1b6e1 VS |
1993 | break; |
1994 | default: | |
1995 | break; | |
1996 | } | |
467a14d9 | 1997 | |
44e921d4 | 1998 | POSTING_READ_FW(DSPARB); |
467a14d9 | 1999 | |
44e921d4 | 2000 | spin_unlock(&dev_priv->uncore.lock); |
54f1b6e1 VS |
2001 | } |
2002 | ||
2003 | #undef VLV_FIFO | |
2004 | ||
4841da51 VS |
2005 | static int vlv_compute_intermediate_wm(struct drm_device *dev, |
2006 | struct intel_crtc *crtc, | |
2007 | struct intel_crtc_state *crtc_state) | |
2008 | { | |
2009 | struct vlv_wm_state *intermediate = &crtc_state->wm.vlv.intermediate; | |
2010 | const struct vlv_wm_state *optimal = &crtc_state->wm.vlv.optimal; | |
2011 | const struct vlv_wm_state *active = &crtc->wm.active.vlv; | |
2012 | int level; | |
2013 | ||
2014 | intermediate->num_levels = min(optimal->num_levels, active->num_levels); | |
5eeb798b VS |
2015 | intermediate->cxsr = optimal->cxsr && active->cxsr && |
2016 | !crtc_state->disable_cxsr; | |
4841da51 VS |
2017 | |
2018 | for (level = 0; level < intermediate->num_levels; level++) { | |
2019 | enum plane_id plane_id; | |
2020 | ||
2021 | for_each_plane_id_on_crtc(crtc, plane_id) { | |
2022 | intermediate->wm[level].plane[plane_id] = | |
2023 | min(optimal->wm[level].plane[plane_id], | |
2024 | active->wm[level].plane[plane_id]); | |
2025 | } | |
2026 | ||
2027 | intermediate->sr[level].plane = min(optimal->sr[level].plane, | |
2028 | active->sr[level].plane); | |
2029 | intermediate->sr[level].cursor = min(optimal->sr[level].cursor, | |
2030 | active->sr[level].cursor); | |
2031 | } | |
2032 | ||
2033 | vlv_invalidate_wms(crtc, intermediate, level); | |
2034 | ||
2035 | /* | |
2036 | * If our intermediate WM are identical to the final WM, then we can | |
2037 | * omit the post-vblank programming; only update if it's different. | |
2038 | */ | |
5eeb798b VS |
2039 | if (memcmp(intermediate, optimal, sizeof(*intermediate)) != 0) |
2040 | crtc_state->wm.need_postvbl_update = true; | |
4841da51 VS |
2041 | |
2042 | return 0; | |
2043 | } | |
2044 | ||
7c951c00 | 2045 | static void vlv_merge_wm(struct drm_i915_private *dev_priv, |
262cd2e1 VS |
2046 | struct vlv_wm_values *wm) |
2047 | { | |
2048 | struct intel_crtc *crtc; | |
2049 | int num_active_crtcs = 0; | |
2050 | ||
7c951c00 | 2051 | wm->level = dev_priv->wm.max_level; |
262cd2e1 VS |
2052 | wm->cxsr = true; |
2053 | ||
7c951c00 | 2054 | for_each_intel_crtc(&dev_priv->drm, crtc) { |
7eb4941f | 2055 | const struct vlv_wm_state *wm_state = &crtc->wm.active.vlv; |
262cd2e1 VS |
2056 | |
2057 | if (!crtc->active) | |
2058 | continue; | |
2059 | ||
2060 | if (!wm_state->cxsr) | |
2061 | wm->cxsr = false; | |
2062 | ||
2063 | num_active_crtcs++; | |
2064 | wm->level = min_t(int, wm->level, wm_state->num_levels - 1); | |
2065 | } | |
2066 | ||
2067 | if (num_active_crtcs != 1) | |
2068 | wm->cxsr = false; | |
2069 | ||
6f9c784b VS |
2070 | if (num_active_crtcs > 1) |
2071 | wm->level = VLV_WM_LEVEL_PM2; | |
2072 | ||
7c951c00 | 2073 | for_each_intel_crtc(&dev_priv->drm, crtc) { |
7eb4941f | 2074 | const struct vlv_wm_state *wm_state = &crtc->wm.active.vlv; |
262cd2e1 VS |
2075 | enum pipe pipe = crtc->pipe; |
2076 | ||
262cd2e1 | 2077 | wm->pipe[pipe] = wm_state->wm[wm->level]; |
ff32c54e | 2078 | if (crtc->active && wm->cxsr) |
262cd2e1 VS |
2079 | wm->sr = wm_state->sr[wm->level]; |
2080 | ||
1b31389c VS |
2081 | wm->ddl[pipe].plane[PLANE_PRIMARY] = DDL_PRECISION_HIGH | 2; |
2082 | wm->ddl[pipe].plane[PLANE_SPRITE0] = DDL_PRECISION_HIGH | 2; | |
2083 | wm->ddl[pipe].plane[PLANE_SPRITE1] = DDL_PRECISION_HIGH | 2; | |
2084 | wm->ddl[pipe].plane[PLANE_CURSOR] = DDL_PRECISION_HIGH | 2; | |
262cd2e1 VS |
2085 | } |
2086 | } | |
2087 | ||
ff32c54e | 2088 | static void vlv_program_watermarks(struct drm_i915_private *dev_priv) |
262cd2e1 | 2089 | { |
fa292a4b VS |
2090 | struct vlv_wm_values *old_wm = &dev_priv->wm.vlv; |
2091 | struct vlv_wm_values new_wm = {}; | |
262cd2e1 | 2092 | |
fa292a4b | 2093 | vlv_merge_wm(dev_priv, &new_wm); |
262cd2e1 | 2094 | |
ff32c54e | 2095 | if (memcmp(old_wm, &new_wm, sizeof(new_wm)) == 0) |
262cd2e1 VS |
2096 | return; |
2097 | ||
fa292a4b | 2098 | if (is_disabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_DDR_DVFS)) |
262cd2e1 VS |
2099 | chv_set_memory_dvfs(dev_priv, false); |
2100 | ||
fa292a4b | 2101 | if (is_disabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_PM5)) |
262cd2e1 VS |
2102 | chv_set_memory_pm5(dev_priv, false); |
2103 | ||
fa292a4b | 2104 | if (is_disabling(old_wm->cxsr, new_wm.cxsr, true)) |
3d90e649 | 2105 | _intel_set_memory_cxsr(dev_priv, false); |
262cd2e1 | 2106 | |
fa292a4b | 2107 | vlv_write_wm_values(dev_priv, &new_wm); |
262cd2e1 | 2108 | |
fa292a4b | 2109 | if (is_enabling(old_wm->cxsr, new_wm.cxsr, true)) |
3d90e649 | 2110 | _intel_set_memory_cxsr(dev_priv, true); |
262cd2e1 | 2111 | |
fa292a4b | 2112 | if (is_enabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_PM5)) |
262cd2e1 VS |
2113 | chv_set_memory_pm5(dev_priv, true); |
2114 | ||
fa292a4b | 2115 | if (is_enabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_DDR_DVFS)) |
262cd2e1 VS |
2116 | chv_set_memory_dvfs(dev_priv, true); |
2117 | ||
fa292a4b | 2118 | *old_wm = new_wm; |
3c2777fd VS |
2119 | } |
2120 | ||
ff32c54e VS |
2121 | static void vlv_initial_watermarks(struct intel_atomic_state *state, |
2122 | struct intel_crtc_state *crtc_state) | |
2123 | { | |
2124 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
2125 | struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc); | |
2126 | ||
2127 | mutex_lock(&dev_priv->wm.wm_mutex); | |
4841da51 VS |
2128 | crtc->wm.active.vlv = crtc_state->wm.vlv.intermediate; |
2129 | vlv_program_watermarks(dev_priv); | |
2130 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
2131 | } | |
2132 | ||
2133 | static void vlv_optimize_watermarks(struct intel_atomic_state *state, | |
2134 | struct intel_crtc_state *crtc_state) | |
2135 | { | |
2136 | struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev); | |
2137 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc_state->base.crtc); | |
2138 | ||
2139 | if (!crtc_state->wm.need_postvbl_update) | |
2140 | return; | |
2141 | ||
2142 | mutex_lock(&dev_priv->wm.wm_mutex); | |
2143 | intel_crtc->wm.active.vlv = crtc_state->wm.vlv.optimal; | |
ff32c54e VS |
2144 | vlv_program_watermarks(dev_priv); |
2145 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
2146 | } | |
2147 | ||
432081bc | 2148 | static void i965_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 2149 | { |
ffc7a76b | 2150 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
efc2611e | 2151 | struct intel_crtc *crtc; |
b445e3b0 ED |
2152 | int srwm = 1; |
2153 | int cursor_sr = 16; | |
9858425c | 2154 | bool cxsr_enabled; |
b445e3b0 ED |
2155 | |
2156 | /* Calc sr entries for one plane configs */ | |
ffc7a76b | 2157 | crtc = single_enabled_crtc(dev_priv); |
b445e3b0 ED |
2158 | if (crtc) { |
2159 | /* self-refresh has much higher latency */ | |
2160 | static const int sr_latency_ns = 12000; | |
efc2611e VS |
2161 | const struct drm_display_mode *adjusted_mode = |
2162 | &crtc->config->base.adjusted_mode; | |
2163 | const struct drm_framebuffer *fb = | |
2164 | crtc->base.primary->state->fb; | |
241bfc38 | 2165 | int clock = adjusted_mode->crtc_clock; |
fec8cba3 | 2166 | int htotal = adjusted_mode->crtc_htotal; |
efc2611e | 2167 | int hdisplay = crtc->config->pipe_src_w; |
353c8598 | 2168 | int cpp = fb->format->cpp[0]; |
b445e3b0 ED |
2169 | int entries; |
2170 | ||
baf69ca8 VS |
2171 | entries = intel_wm_method2(clock, htotal, |
2172 | hdisplay, cpp, sr_latency_ns / 100); | |
b445e3b0 ED |
2173 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); |
2174 | srwm = I965_FIFO_SIZE - entries; | |
2175 | if (srwm < 0) | |
2176 | srwm = 1; | |
2177 | srwm &= 0x1ff; | |
2178 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", | |
2179 | entries, srwm); | |
2180 | ||
baf69ca8 VS |
2181 | entries = intel_wm_method2(clock, htotal, |
2182 | crtc->base.cursor->state->crtc_w, 4, | |
2183 | sr_latency_ns / 100); | |
b445e3b0 | 2184 | entries = DIV_ROUND_UP(entries, |
baf69ca8 VS |
2185 | i965_cursor_wm_info.cacheline_size) + |
2186 | i965_cursor_wm_info.guard_size; | |
b445e3b0 | 2187 | |
baf69ca8 | 2188 | cursor_sr = i965_cursor_wm_info.fifo_size - entries; |
b445e3b0 ED |
2189 | if (cursor_sr > i965_cursor_wm_info.max_wm) |
2190 | cursor_sr = i965_cursor_wm_info.max_wm; | |
2191 | ||
2192 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " | |
2193 | "cursor %d\n", srwm, cursor_sr); | |
2194 | ||
9858425c | 2195 | cxsr_enabled = true; |
b445e3b0 | 2196 | } else { |
9858425c | 2197 | cxsr_enabled = false; |
b445e3b0 | 2198 | /* Turn off self refresh if both pipes are enabled */ |
5209b1f4 | 2199 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
2200 | } |
2201 | ||
2202 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", | |
2203 | srwm); | |
2204 | ||
2205 | /* 965 has limitations... */ | |
f4998963 VS |
2206 | I915_WRITE(DSPFW1, FW_WM(srwm, SR) | |
2207 | FW_WM(8, CURSORB) | | |
2208 | FW_WM(8, PLANEB) | | |
2209 | FW_WM(8, PLANEA)); | |
2210 | I915_WRITE(DSPFW2, FW_WM(8, CURSORA) | | |
2211 | FW_WM(8, PLANEC_OLD)); | |
b445e3b0 | 2212 | /* update cursor SR watermark */ |
f4998963 | 2213 | I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR)); |
9858425c ID |
2214 | |
2215 | if (cxsr_enabled) | |
2216 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
2217 | } |
2218 | ||
f4998963 VS |
2219 | #undef FW_WM |
2220 | ||
432081bc | 2221 | static void i9xx_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 2222 | { |
ffc7a76b | 2223 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
b445e3b0 ED |
2224 | const struct intel_watermark_params *wm_info; |
2225 | uint32_t fwater_lo; | |
2226 | uint32_t fwater_hi; | |
2227 | int cwm, srwm = 1; | |
2228 | int fifo_size; | |
2229 | int planea_wm, planeb_wm; | |
efc2611e | 2230 | struct intel_crtc *crtc, *enabled = NULL; |
b445e3b0 | 2231 | |
a9097be4 | 2232 | if (IS_I945GM(dev_priv)) |
b445e3b0 | 2233 | wm_info = &i945_wm_info; |
5db94019 | 2234 | else if (!IS_GEN2(dev_priv)) |
b445e3b0 ED |
2235 | wm_info = &i915_wm_info; |
2236 | else | |
9d539105 | 2237 | wm_info = &i830_a_wm_info; |
b445e3b0 | 2238 | |
ef0f5e93 | 2239 | fifo_size = dev_priv->display.get_fifo_size(dev_priv, 0); |
b91eb5cc | 2240 | crtc = intel_get_crtc_for_plane(dev_priv, 0); |
efc2611e VS |
2241 | if (intel_crtc_active(crtc)) { |
2242 | const struct drm_display_mode *adjusted_mode = | |
2243 | &crtc->config->base.adjusted_mode; | |
2244 | const struct drm_framebuffer *fb = | |
2245 | crtc->base.primary->state->fb; | |
2246 | int cpp; | |
2247 | ||
5db94019 | 2248 | if (IS_GEN2(dev_priv)) |
b9e0bda3 | 2249 | cpp = 4; |
efc2611e | 2250 | else |
353c8598 | 2251 | cpp = fb->format->cpp[0]; |
b9e0bda3 | 2252 | |
241bfc38 | 2253 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
b9e0bda3 | 2254 | wm_info, fifo_size, cpp, |
5aef6003 | 2255 | pessimal_latency_ns); |
b445e3b0 | 2256 | enabled = crtc; |
9d539105 | 2257 | } else { |
b445e3b0 | 2258 | planea_wm = fifo_size - wm_info->guard_size; |
9d539105 VS |
2259 | if (planea_wm > (long)wm_info->max_wm) |
2260 | planea_wm = wm_info->max_wm; | |
2261 | } | |
2262 | ||
5db94019 | 2263 | if (IS_GEN2(dev_priv)) |
9d539105 | 2264 | wm_info = &i830_bc_wm_info; |
b445e3b0 | 2265 | |
ef0f5e93 | 2266 | fifo_size = dev_priv->display.get_fifo_size(dev_priv, 1); |
b91eb5cc | 2267 | crtc = intel_get_crtc_for_plane(dev_priv, 1); |
efc2611e VS |
2268 | if (intel_crtc_active(crtc)) { |
2269 | const struct drm_display_mode *adjusted_mode = | |
2270 | &crtc->config->base.adjusted_mode; | |
2271 | const struct drm_framebuffer *fb = | |
2272 | crtc->base.primary->state->fb; | |
2273 | int cpp; | |
2274 | ||
5db94019 | 2275 | if (IS_GEN2(dev_priv)) |
b9e0bda3 | 2276 | cpp = 4; |
efc2611e | 2277 | else |
353c8598 | 2278 | cpp = fb->format->cpp[0]; |
b9e0bda3 | 2279 | |
241bfc38 | 2280 | planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
b9e0bda3 | 2281 | wm_info, fifo_size, cpp, |
5aef6003 | 2282 | pessimal_latency_ns); |
b445e3b0 ED |
2283 | if (enabled == NULL) |
2284 | enabled = crtc; | |
2285 | else | |
2286 | enabled = NULL; | |
9d539105 | 2287 | } else { |
b445e3b0 | 2288 | planeb_wm = fifo_size - wm_info->guard_size; |
9d539105 VS |
2289 | if (planeb_wm > (long)wm_info->max_wm) |
2290 | planeb_wm = wm_info->max_wm; | |
2291 | } | |
b445e3b0 ED |
2292 | |
2293 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); | |
2294 | ||
50a0bc90 | 2295 | if (IS_I915GM(dev_priv) && enabled) { |
2ff8fde1 | 2296 | struct drm_i915_gem_object *obj; |
2ab1bc9d | 2297 | |
efc2611e | 2298 | obj = intel_fb_obj(enabled->base.primary->state->fb); |
2ab1bc9d DV |
2299 | |
2300 | /* self-refresh seems busted with untiled */ | |
3e510a8e | 2301 | if (!i915_gem_object_is_tiled(obj)) |
2ab1bc9d DV |
2302 | enabled = NULL; |
2303 | } | |
2304 | ||
b445e3b0 ED |
2305 | /* |
2306 | * Overlay gets an aggressive default since video jitter is bad. | |
2307 | */ | |
2308 | cwm = 2; | |
2309 | ||
2310 | /* Play safe and disable self-refresh before adjusting watermarks. */ | |
5209b1f4 | 2311 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
2312 | |
2313 | /* Calc sr entries for one plane configs */ | |
03427fcb | 2314 | if (HAS_FW_BLC(dev_priv) && enabled) { |
b445e3b0 ED |
2315 | /* self-refresh has much higher latency */ |
2316 | static const int sr_latency_ns = 6000; | |
efc2611e VS |
2317 | const struct drm_display_mode *adjusted_mode = |
2318 | &enabled->config->base.adjusted_mode; | |
2319 | const struct drm_framebuffer *fb = | |
2320 | enabled->base.primary->state->fb; | |
241bfc38 | 2321 | int clock = adjusted_mode->crtc_clock; |
fec8cba3 | 2322 | int htotal = adjusted_mode->crtc_htotal; |
efc2611e VS |
2323 | int hdisplay = enabled->config->pipe_src_w; |
2324 | int cpp; | |
b445e3b0 ED |
2325 | int entries; |
2326 | ||
50a0bc90 | 2327 | if (IS_I915GM(dev_priv) || IS_I945GM(dev_priv)) |
2d1b5056 | 2328 | cpp = 4; |
efc2611e | 2329 | else |
353c8598 | 2330 | cpp = fb->format->cpp[0]; |
2d1b5056 | 2331 | |
baf69ca8 VS |
2332 | entries = intel_wm_method2(clock, htotal, hdisplay, cpp, |
2333 | sr_latency_ns / 100); | |
b445e3b0 ED |
2334 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); |
2335 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); | |
2336 | srwm = wm_info->fifo_size - entries; | |
2337 | if (srwm < 0) | |
2338 | srwm = 1; | |
2339 | ||
50a0bc90 | 2340 | if (IS_I945G(dev_priv) || IS_I945GM(dev_priv)) |
b445e3b0 ED |
2341 | I915_WRITE(FW_BLC_SELF, |
2342 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); | |
acb91359 | 2343 | else |
b445e3b0 ED |
2344 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); |
2345 | } | |
2346 | ||
2347 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", | |
2348 | planea_wm, planeb_wm, cwm, srwm); | |
2349 | ||
2350 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); | |
2351 | fwater_hi = (cwm & 0x1f); | |
2352 | ||
2353 | /* Set request length to 8 cachelines per fetch */ | |
2354 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); | |
2355 | fwater_hi = fwater_hi | (1 << 8); | |
2356 | ||
2357 | I915_WRITE(FW_BLC, fwater_lo); | |
2358 | I915_WRITE(FW_BLC2, fwater_hi); | |
2359 | ||
5209b1f4 ID |
2360 | if (enabled) |
2361 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
2362 | } |
2363 | ||
432081bc | 2364 | static void i845_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 2365 | { |
ffc7a76b | 2366 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
efc2611e | 2367 | struct intel_crtc *crtc; |
241bfc38 | 2368 | const struct drm_display_mode *adjusted_mode; |
b445e3b0 ED |
2369 | uint32_t fwater_lo; |
2370 | int planea_wm; | |
2371 | ||
ffc7a76b | 2372 | crtc = single_enabled_crtc(dev_priv); |
b445e3b0 ED |
2373 | if (crtc == NULL) |
2374 | return; | |
2375 | ||
efc2611e | 2376 | adjusted_mode = &crtc->config->base.adjusted_mode; |
241bfc38 | 2377 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
feb56b93 | 2378 | &i845_wm_info, |
ef0f5e93 | 2379 | dev_priv->display.get_fifo_size(dev_priv, 0), |
5aef6003 | 2380 | 4, pessimal_latency_ns); |
b445e3b0 ED |
2381 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
2382 | fwater_lo |= (3<<8) | planea_wm; | |
2383 | ||
2384 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); | |
2385 | ||
2386 | I915_WRITE(FW_BLC, fwater_lo); | |
2387 | } | |
2388 | ||
37126462 | 2389 | /* latency must be in 0.1us units. */ |
baf69ca8 VS |
2390 | static unsigned int ilk_wm_method1(unsigned int pixel_rate, |
2391 | unsigned int cpp, | |
2392 | unsigned int latency) | |
801bcfff | 2393 | { |
baf69ca8 | 2394 | unsigned int ret; |
3312ba65 | 2395 | |
baf69ca8 VS |
2396 | ret = intel_wm_method1(pixel_rate, cpp, latency); |
2397 | ret = DIV_ROUND_UP(ret, 64) + 2; | |
801bcfff PZ |
2398 | |
2399 | return ret; | |
2400 | } | |
2401 | ||
37126462 | 2402 | /* latency must be in 0.1us units. */ |
baf69ca8 VS |
2403 | static unsigned int ilk_wm_method2(unsigned int pixel_rate, |
2404 | unsigned int htotal, | |
2405 | unsigned int width, | |
2406 | unsigned int cpp, | |
2407 | unsigned int latency) | |
801bcfff | 2408 | { |
baf69ca8 | 2409 | unsigned int ret; |
3312ba65 | 2410 | |
baf69ca8 VS |
2411 | ret = intel_wm_method2(pixel_rate, htotal, |
2412 | width, cpp, latency); | |
801bcfff | 2413 | ret = DIV_ROUND_UP(ret, 64) + 2; |
baf69ca8 | 2414 | |
801bcfff PZ |
2415 | return ret; |
2416 | } | |
2417 | ||
23297044 | 2418 | static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels, |
ac484963 | 2419 | uint8_t cpp) |
cca32e9a | 2420 | { |
15126882 MR |
2421 | /* |
2422 | * Neither of these should be possible since this function shouldn't be | |
2423 | * called if the CRTC is off or the plane is invisible. But let's be | |
2424 | * extra paranoid to avoid a potential divide-by-zero if we screw up | |
2425 | * elsewhere in the driver. | |
2426 | */ | |
ac484963 | 2427 | if (WARN_ON(!cpp)) |
15126882 MR |
2428 | return 0; |
2429 | if (WARN_ON(!horiz_pixels)) | |
2430 | return 0; | |
2431 | ||
ac484963 | 2432 | return DIV_ROUND_UP(pri_val * 64, horiz_pixels * cpp) + 2; |
cca32e9a PZ |
2433 | } |
2434 | ||
820c1980 | 2435 | struct ilk_wm_maximums { |
cca32e9a PZ |
2436 | uint16_t pri; |
2437 | uint16_t spr; | |
2438 | uint16_t cur; | |
2439 | uint16_t fbc; | |
2440 | }; | |
2441 | ||
37126462 VS |
2442 | /* |
2443 | * For both WM_PIPE and WM_LP. | |
2444 | * mem_value must be in 0.1us units. | |
2445 | */ | |
7221fc33 | 2446 | static uint32_t ilk_compute_pri_wm(const struct intel_crtc_state *cstate, |
43d59eda | 2447 | const struct intel_plane_state *pstate, |
cca32e9a PZ |
2448 | uint32_t mem_value, |
2449 | bool is_lp) | |
801bcfff | 2450 | { |
cca32e9a | 2451 | uint32_t method1, method2; |
8305494e | 2452 | int cpp; |
cca32e9a | 2453 | |
24304d81 | 2454 | if (!intel_wm_plane_visible(cstate, pstate)) |
801bcfff PZ |
2455 | return 0; |
2456 | ||
353c8598 | 2457 | cpp = pstate->base.fb->format->cpp[0]; |
8305494e | 2458 | |
a7d1b3f4 | 2459 | method1 = ilk_wm_method1(cstate->pixel_rate, cpp, mem_value); |
cca32e9a PZ |
2460 | |
2461 | if (!is_lp) | |
2462 | return method1; | |
2463 | ||
a7d1b3f4 | 2464 | method2 = ilk_wm_method2(cstate->pixel_rate, |
7221fc33 | 2465 | cstate->base.adjusted_mode.crtc_htotal, |
936e71e3 | 2466 | drm_rect_width(&pstate->base.dst), |
ac484963 | 2467 | cpp, mem_value); |
cca32e9a PZ |
2468 | |
2469 | return min(method1, method2); | |
801bcfff PZ |
2470 | } |
2471 | ||
37126462 VS |
2472 | /* |
2473 | * For both WM_PIPE and WM_LP. | |
2474 | * mem_value must be in 0.1us units. | |
2475 | */ | |
7221fc33 | 2476 | static uint32_t ilk_compute_spr_wm(const struct intel_crtc_state *cstate, |
43d59eda | 2477 | const struct intel_plane_state *pstate, |
801bcfff PZ |
2478 | uint32_t mem_value) |
2479 | { | |
2480 | uint32_t method1, method2; | |
8305494e | 2481 | int cpp; |
801bcfff | 2482 | |
24304d81 | 2483 | if (!intel_wm_plane_visible(cstate, pstate)) |
801bcfff PZ |
2484 | return 0; |
2485 | ||
353c8598 | 2486 | cpp = pstate->base.fb->format->cpp[0]; |
8305494e | 2487 | |
a7d1b3f4 VS |
2488 | method1 = ilk_wm_method1(cstate->pixel_rate, cpp, mem_value); |
2489 | method2 = ilk_wm_method2(cstate->pixel_rate, | |
7221fc33 | 2490 | cstate->base.adjusted_mode.crtc_htotal, |
936e71e3 | 2491 | drm_rect_width(&pstate->base.dst), |
ac484963 | 2492 | cpp, mem_value); |
801bcfff PZ |
2493 | return min(method1, method2); |
2494 | } | |
2495 | ||
37126462 VS |
2496 | /* |
2497 | * For both WM_PIPE and WM_LP. | |
2498 | * mem_value must be in 0.1us units. | |
2499 | */ | |
7221fc33 | 2500 | static uint32_t ilk_compute_cur_wm(const struct intel_crtc_state *cstate, |
43d59eda | 2501 | const struct intel_plane_state *pstate, |
801bcfff PZ |
2502 | uint32_t mem_value) |
2503 | { | |
a5509abd VS |
2504 | int cpp; |
2505 | ||
24304d81 | 2506 | if (!intel_wm_plane_visible(cstate, pstate)) |
801bcfff PZ |
2507 | return 0; |
2508 | ||
a5509abd VS |
2509 | cpp = pstate->base.fb->format->cpp[0]; |
2510 | ||
a7d1b3f4 | 2511 | return ilk_wm_method2(cstate->pixel_rate, |
7221fc33 | 2512 | cstate->base.adjusted_mode.crtc_htotal, |
a5509abd | 2513 | pstate->base.crtc_w, cpp, mem_value); |
801bcfff PZ |
2514 | } |
2515 | ||
cca32e9a | 2516 | /* Only for WM_LP. */ |
7221fc33 | 2517 | static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate, |
43d59eda | 2518 | const struct intel_plane_state *pstate, |
1fda9882 | 2519 | uint32_t pri_val) |
cca32e9a | 2520 | { |
8305494e | 2521 | int cpp; |
43d59eda | 2522 | |
24304d81 | 2523 | if (!intel_wm_plane_visible(cstate, pstate)) |
cca32e9a PZ |
2524 | return 0; |
2525 | ||
353c8598 | 2526 | cpp = pstate->base.fb->format->cpp[0]; |
8305494e | 2527 | |
936e71e3 | 2528 | return ilk_wm_fbc(pri_val, drm_rect_width(&pstate->base.dst), cpp); |
cca32e9a PZ |
2529 | } |
2530 | ||
175fded1 TU |
2531 | static unsigned int |
2532 | ilk_display_fifo_size(const struct drm_i915_private *dev_priv) | |
158ae64f | 2533 | { |
175fded1 | 2534 | if (INTEL_GEN(dev_priv) >= 8) |
416f4727 | 2535 | return 3072; |
175fded1 | 2536 | else if (INTEL_GEN(dev_priv) >= 7) |
158ae64f VS |
2537 | return 768; |
2538 | else | |
2539 | return 512; | |
2540 | } | |
2541 | ||
175fded1 TU |
2542 | static unsigned int |
2543 | ilk_plane_wm_reg_max(const struct drm_i915_private *dev_priv, | |
2544 | int level, bool is_sprite) | |
4e975081 | 2545 | { |
175fded1 | 2546 | if (INTEL_GEN(dev_priv) >= 8) |
4e975081 VS |
2547 | /* BDW primary/sprite plane watermarks */ |
2548 | return level == 0 ? 255 : 2047; | |
175fded1 | 2549 | else if (INTEL_GEN(dev_priv) >= 7) |
4e975081 VS |
2550 | /* IVB/HSW primary/sprite plane watermarks */ |
2551 | return level == 0 ? 127 : 1023; | |
2552 | else if (!is_sprite) | |
2553 | /* ILK/SNB primary plane watermarks */ | |
2554 | return level == 0 ? 127 : 511; | |
2555 | else | |
2556 | /* ILK/SNB sprite plane watermarks */ | |
2557 | return level == 0 ? 63 : 255; | |
2558 | } | |
2559 | ||
175fded1 TU |
2560 | static unsigned int |
2561 | ilk_cursor_wm_reg_max(const struct drm_i915_private *dev_priv, int level) | |
4e975081 | 2562 | { |
175fded1 | 2563 | if (INTEL_GEN(dev_priv) >= 7) |
4e975081 VS |
2564 | return level == 0 ? 63 : 255; |
2565 | else | |
2566 | return level == 0 ? 31 : 63; | |
2567 | } | |
2568 | ||
175fded1 | 2569 | static unsigned int ilk_fbc_wm_reg_max(const struct drm_i915_private *dev_priv) |
4e975081 | 2570 | { |
175fded1 | 2571 | if (INTEL_GEN(dev_priv) >= 8) |
4e975081 VS |
2572 | return 31; |
2573 | else | |
2574 | return 15; | |
2575 | } | |
2576 | ||
158ae64f VS |
2577 | /* Calculate the maximum primary/sprite plane watermark */ |
2578 | static unsigned int ilk_plane_wm_max(const struct drm_device *dev, | |
2579 | int level, | |
240264f4 | 2580 | const struct intel_wm_config *config, |
158ae64f VS |
2581 | enum intel_ddb_partitioning ddb_partitioning, |
2582 | bool is_sprite) | |
2583 | { | |
175fded1 TU |
2584 | struct drm_i915_private *dev_priv = to_i915(dev); |
2585 | unsigned int fifo_size = ilk_display_fifo_size(dev_priv); | |
158ae64f VS |
2586 | |
2587 | /* if sprites aren't enabled, sprites get nothing */ | |
240264f4 | 2588 | if (is_sprite && !config->sprites_enabled) |
158ae64f VS |
2589 | return 0; |
2590 | ||
2591 | /* HSW allows LP1+ watermarks even with multiple pipes */ | |
240264f4 | 2592 | if (level == 0 || config->num_pipes_active > 1) { |
175fded1 | 2593 | fifo_size /= INTEL_INFO(dev_priv)->num_pipes; |
158ae64f VS |
2594 | |
2595 | /* | |
2596 | * For some reason the non self refresh | |
2597 | * FIFO size is only half of the self | |
2598 | * refresh FIFO size on ILK/SNB. | |
2599 | */ | |
175fded1 | 2600 | if (INTEL_GEN(dev_priv) <= 6) |
158ae64f VS |
2601 | fifo_size /= 2; |
2602 | } | |
2603 | ||
240264f4 | 2604 | if (config->sprites_enabled) { |
158ae64f VS |
2605 | /* level 0 is always calculated with 1:1 split */ |
2606 | if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) { | |
2607 | if (is_sprite) | |
2608 | fifo_size *= 5; | |
2609 | fifo_size /= 6; | |
2610 | } else { | |
2611 | fifo_size /= 2; | |
2612 | } | |
2613 | } | |
2614 | ||
2615 | /* clamp to max that the registers can hold */ | |
175fded1 | 2616 | return min(fifo_size, ilk_plane_wm_reg_max(dev_priv, level, is_sprite)); |
158ae64f VS |
2617 | } |
2618 | ||
2619 | /* Calculate the maximum cursor plane watermark */ | |
2620 | static unsigned int ilk_cursor_wm_max(const struct drm_device *dev, | |
240264f4 VS |
2621 | int level, |
2622 | const struct intel_wm_config *config) | |
158ae64f VS |
2623 | { |
2624 | /* HSW LP1+ watermarks w/ multiple pipes */ | |
240264f4 | 2625 | if (level > 0 && config->num_pipes_active > 1) |
158ae64f VS |
2626 | return 64; |
2627 | ||
2628 | /* otherwise just report max that registers can hold */ | |
175fded1 | 2629 | return ilk_cursor_wm_reg_max(to_i915(dev), level); |
158ae64f VS |
2630 | } |
2631 | ||
d34ff9c6 | 2632 | static void ilk_compute_wm_maximums(const struct drm_device *dev, |
34982fe1 VS |
2633 | int level, |
2634 | const struct intel_wm_config *config, | |
2635 | enum intel_ddb_partitioning ddb_partitioning, | |
820c1980 | 2636 | struct ilk_wm_maximums *max) |
158ae64f | 2637 | { |
240264f4 VS |
2638 | max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false); |
2639 | max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true); | |
2640 | max->cur = ilk_cursor_wm_max(dev, level, config); | |
175fded1 | 2641 | max->fbc = ilk_fbc_wm_reg_max(to_i915(dev)); |
158ae64f VS |
2642 | } |
2643 | ||
175fded1 | 2644 | static void ilk_compute_wm_reg_maximums(const struct drm_i915_private *dev_priv, |
a3cb4048 VS |
2645 | int level, |
2646 | struct ilk_wm_maximums *max) | |
2647 | { | |
175fded1 TU |
2648 | max->pri = ilk_plane_wm_reg_max(dev_priv, level, false); |
2649 | max->spr = ilk_plane_wm_reg_max(dev_priv, level, true); | |
2650 | max->cur = ilk_cursor_wm_reg_max(dev_priv, level); | |
2651 | max->fbc = ilk_fbc_wm_reg_max(dev_priv); | |
a3cb4048 VS |
2652 | } |
2653 | ||
d9395655 | 2654 | static bool ilk_validate_wm_level(int level, |
820c1980 | 2655 | const struct ilk_wm_maximums *max, |
d9395655 | 2656 | struct intel_wm_level *result) |
a9786a11 VS |
2657 | { |
2658 | bool ret; | |
2659 | ||
2660 | /* already determined to be invalid? */ | |
2661 | if (!result->enable) | |
2662 | return false; | |
2663 | ||
2664 | result->enable = result->pri_val <= max->pri && | |
2665 | result->spr_val <= max->spr && | |
2666 | result->cur_val <= max->cur; | |
2667 | ||
2668 | ret = result->enable; | |
2669 | ||
2670 | /* | |
2671 | * HACK until we can pre-compute everything, | |
2672 | * and thus fail gracefully if LP0 watermarks | |
2673 | * are exceeded... | |
2674 | */ | |
2675 | if (level == 0 && !result->enable) { | |
2676 | if (result->pri_val > max->pri) | |
2677 | DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n", | |
2678 | level, result->pri_val, max->pri); | |
2679 | if (result->spr_val > max->spr) | |
2680 | DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n", | |
2681 | level, result->spr_val, max->spr); | |
2682 | if (result->cur_val > max->cur) | |
2683 | DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n", | |
2684 | level, result->cur_val, max->cur); | |
2685 | ||
2686 | result->pri_val = min_t(uint32_t, result->pri_val, max->pri); | |
2687 | result->spr_val = min_t(uint32_t, result->spr_val, max->spr); | |
2688 | result->cur_val = min_t(uint32_t, result->cur_val, max->cur); | |
2689 | result->enable = true; | |
2690 | } | |
2691 | ||
a9786a11 VS |
2692 | return ret; |
2693 | } | |
2694 | ||
d34ff9c6 | 2695 | static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv, |
43d59eda | 2696 | const struct intel_crtc *intel_crtc, |
6f5ddd17 | 2697 | int level, |
7221fc33 | 2698 | struct intel_crtc_state *cstate, |
0434a5fb ML |
2699 | const struct intel_plane_state *pristate, |
2700 | const struct intel_plane_state *sprstate, | |
2701 | const struct intel_plane_state *curstate, | |
1fd527cc | 2702 | struct intel_wm_level *result) |
6f5ddd17 VS |
2703 | { |
2704 | uint16_t pri_latency = dev_priv->wm.pri_latency[level]; | |
2705 | uint16_t spr_latency = dev_priv->wm.spr_latency[level]; | |
2706 | uint16_t cur_latency = dev_priv->wm.cur_latency[level]; | |
2707 | ||
2708 | /* WM1+ latency values stored in 0.5us units */ | |
2709 | if (level > 0) { | |
2710 | pri_latency *= 5; | |
2711 | spr_latency *= 5; | |
2712 | cur_latency *= 5; | |
2713 | } | |
2714 | ||
e3bddded ML |
2715 | if (pristate) { |
2716 | result->pri_val = ilk_compute_pri_wm(cstate, pristate, | |
2717 | pri_latency, level); | |
2718 | result->fbc_val = ilk_compute_fbc_wm(cstate, pristate, result->pri_val); | |
2719 | } | |
2720 | ||
2721 | if (sprstate) | |
2722 | result->spr_val = ilk_compute_spr_wm(cstate, sprstate, spr_latency); | |
2723 | ||
2724 | if (curstate) | |
2725 | result->cur_val = ilk_compute_cur_wm(cstate, curstate, cur_latency); | |
2726 | ||
6f5ddd17 VS |
2727 | result->enable = true; |
2728 | } | |
2729 | ||
801bcfff | 2730 | static uint32_t |
532f7a7f | 2731 | hsw_compute_linetime_wm(const struct intel_crtc_state *cstate) |
1f8eeabf | 2732 | { |
532f7a7f VS |
2733 | const struct intel_atomic_state *intel_state = |
2734 | to_intel_atomic_state(cstate->base.state); | |
ee91a159 MR |
2735 | const struct drm_display_mode *adjusted_mode = |
2736 | &cstate->base.adjusted_mode; | |
85a02deb | 2737 | u32 linetime, ips_linetime; |
1f8eeabf | 2738 | |
ee91a159 MR |
2739 | if (!cstate->base.active) |
2740 | return 0; | |
2741 | if (WARN_ON(adjusted_mode->crtc_clock == 0)) | |
2742 | return 0; | |
bb0f4aab | 2743 | if (WARN_ON(intel_state->cdclk.logical.cdclk == 0)) |
801bcfff | 2744 | return 0; |
1011d8c4 | 2745 | |
1f8eeabf ED |
2746 | /* The WM are computed with base on how long it takes to fill a single |
2747 | * row at the given clock rate, multiplied by 8. | |
2748 | * */ | |
124abe07 VS |
2749 | linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8, |
2750 | adjusted_mode->crtc_clock); | |
2751 | ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8, | |
bb0f4aab | 2752 | intel_state->cdclk.logical.cdclk); |
1f8eeabf | 2753 | |
801bcfff PZ |
2754 | return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) | |
2755 | PIPE_WM_LINETIME_TIME(linetime); | |
1f8eeabf ED |
2756 | } |
2757 | ||
bb726519 VS |
2758 | static void intel_read_wm_latency(struct drm_i915_private *dev_priv, |
2759 | uint16_t wm[8]) | |
12b134df | 2760 | { |
5db94019 | 2761 | if (IS_GEN9(dev_priv)) { |
2af30a5c | 2762 | uint32_t val; |
4f947386 | 2763 | int ret, i; |
5db94019 | 2764 | int level, max_level = ilk_wm_max_level(dev_priv); |
2af30a5c PB |
2765 | |
2766 | /* read the first set of memory latencies[0:3] */ | |
2767 | val = 0; /* data0 to be programmed to 0 for first set */ | |
2768 | mutex_lock(&dev_priv->rps.hw_lock); | |
2769 | ret = sandybridge_pcode_read(dev_priv, | |
2770 | GEN9_PCODE_READ_MEM_LATENCY, | |
2771 | &val); | |
2772 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2773 | ||
2774 | if (ret) { | |
2775 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); | |
2776 | return; | |
2777 | } | |
2778 | ||
2779 | wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK; | |
2780 | wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & | |
2781 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2782 | wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & | |
2783 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2784 | wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & | |
2785 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2786 | ||
2787 | /* read the second set of memory latencies[4:7] */ | |
2788 | val = 1; /* data0 to be programmed to 1 for second set */ | |
2789 | mutex_lock(&dev_priv->rps.hw_lock); | |
2790 | ret = sandybridge_pcode_read(dev_priv, | |
2791 | GEN9_PCODE_READ_MEM_LATENCY, | |
2792 | &val); | |
2793 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2794 | if (ret) { | |
2795 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); | |
2796 | return; | |
2797 | } | |
2798 | ||
2799 | wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK; | |
2800 | wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & | |
2801 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2802 | wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & | |
2803 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2804 | wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & | |
2805 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2806 | ||
0727e40a PZ |
2807 | /* |
2808 | * If a level n (n > 1) has a 0us latency, all levels m (m >= n) | |
2809 | * need to be disabled. We make sure to sanitize the values out | |
2810 | * of the punit to satisfy this requirement. | |
2811 | */ | |
2812 | for (level = 1; level <= max_level; level++) { | |
2813 | if (wm[level] == 0) { | |
2814 | for (i = level + 1; i <= max_level; i++) | |
2815 | wm[i] = 0; | |
2816 | break; | |
2817 | } | |
2818 | } | |
2819 | ||
367294be | 2820 | /* |
9fb5026f | 2821 | * WaWmMemoryReadLatency:skl,glk |
6f97235b | 2822 | * |
367294be | 2823 | * punit doesn't take into account the read latency so we need |
0727e40a PZ |
2824 | * to add 2us to the various latency levels we retrieve from the |
2825 | * punit when level 0 response data us 0us. | |
367294be | 2826 | */ |
0727e40a PZ |
2827 | if (wm[0] == 0) { |
2828 | wm[0] += 2; | |
2829 | for (level = 1; level <= max_level; level++) { | |
2830 | if (wm[level] == 0) | |
2831 | break; | |
367294be | 2832 | wm[level] += 2; |
4f947386 | 2833 | } |
0727e40a PZ |
2834 | } |
2835 | ||
8652744b | 2836 | } else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) { |
12b134df VS |
2837 | uint64_t sskpd = I915_READ64(MCH_SSKPD); |
2838 | ||
2839 | wm[0] = (sskpd >> 56) & 0xFF; | |
2840 | if (wm[0] == 0) | |
2841 | wm[0] = sskpd & 0xF; | |
e5d5019e VS |
2842 | wm[1] = (sskpd >> 4) & 0xFF; |
2843 | wm[2] = (sskpd >> 12) & 0xFF; | |
2844 | wm[3] = (sskpd >> 20) & 0x1FF; | |
2845 | wm[4] = (sskpd >> 32) & 0x1FF; | |
bb726519 | 2846 | } else if (INTEL_GEN(dev_priv) >= 6) { |
63cf9a13 VS |
2847 | uint32_t sskpd = I915_READ(MCH_SSKPD); |
2848 | ||
2849 | wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK; | |
2850 | wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK; | |
2851 | wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK; | |
2852 | wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK; | |
bb726519 | 2853 | } else if (INTEL_GEN(dev_priv) >= 5) { |
3a88d0ac VS |
2854 | uint32_t mltr = I915_READ(MLTR_ILK); |
2855 | ||
2856 | /* ILK primary LP0 latency is 700 ns */ | |
2857 | wm[0] = 7; | |
2858 | wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK; | |
2859 | wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK; | |
12b134df VS |
2860 | } |
2861 | } | |
2862 | ||
5db94019 TU |
2863 | static void intel_fixup_spr_wm_latency(struct drm_i915_private *dev_priv, |
2864 | uint16_t wm[5]) | |
53615a5e VS |
2865 | { |
2866 | /* ILK sprite LP0 latency is 1300 ns */ | |
5db94019 | 2867 | if (IS_GEN5(dev_priv)) |
53615a5e VS |
2868 | wm[0] = 13; |
2869 | } | |
2870 | ||
fd6b8f43 TU |
2871 | static void intel_fixup_cur_wm_latency(struct drm_i915_private *dev_priv, |
2872 | uint16_t wm[5]) | |
53615a5e VS |
2873 | { |
2874 | /* ILK cursor LP0 latency is 1300 ns */ | |
fd6b8f43 | 2875 | if (IS_GEN5(dev_priv)) |
53615a5e VS |
2876 | wm[0] = 13; |
2877 | ||
2878 | /* WaDoubleCursorLP3Latency:ivb */ | |
fd6b8f43 | 2879 | if (IS_IVYBRIDGE(dev_priv)) |
53615a5e VS |
2880 | wm[3] *= 2; |
2881 | } | |
2882 | ||
5db94019 | 2883 | int ilk_wm_max_level(const struct drm_i915_private *dev_priv) |
26ec971e | 2884 | { |
26ec971e | 2885 | /* how many WM levels are we expecting */ |
8652744b | 2886 | if (INTEL_GEN(dev_priv) >= 9) |
2af30a5c | 2887 | return 7; |
8652744b | 2888 | else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
ad0d6dc4 | 2889 | return 4; |
8652744b | 2890 | else if (INTEL_GEN(dev_priv) >= 6) |
ad0d6dc4 | 2891 | return 3; |
26ec971e | 2892 | else |
ad0d6dc4 VS |
2893 | return 2; |
2894 | } | |
7526ed79 | 2895 | |
5db94019 | 2896 | static void intel_print_wm_latency(struct drm_i915_private *dev_priv, |
ad0d6dc4 | 2897 | const char *name, |
2af30a5c | 2898 | const uint16_t wm[8]) |
ad0d6dc4 | 2899 | { |
5db94019 | 2900 | int level, max_level = ilk_wm_max_level(dev_priv); |
26ec971e VS |
2901 | |
2902 | for (level = 0; level <= max_level; level++) { | |
2903 | unsigned int latency = wm[level]; | |
2904 | ||
2905 | if (latency == 0) { | |
2906 | DRM_ERROR("%s WM%d latency not provided\n", | |
2907 | name, level); | |
2908 | continue; | |
2909 | } | |
2910 | ||
2af30a5c PB |
2911 | /* |
2912 | * - latencies are in us on gen9. | |
2913 | * - before then, WM1+ latency values are in 0.5us units | |
2914 | */ | |
5db94019 | 2915 | if (IS_GEN9(dev_priv)) |
2af30a5c PB |
2916 | latency *= 10; |
2917 | else if (level > 0) | |
26ec971e VS |
2918 | latency *= 5; |
2919 | ||
2920 | DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n", | |
2921 | name, level, wm[level], | |
2922 | latency / 10, latency % 10); | |
2923 | } | |
2924 | } | |
2925 | ||
e95a2f75 VS |
2926 | static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv, |
2927 | uint16_t wm[5], uint16_t min) | |
2928 | { | |
5db94019 | 2929 | int level, max_level = ilk_wm_max_level(dev_priv); |
e95a2f75 VS |
2930 | |
2931 | if (wm[0] >= min) | |
2932 | return false; | |
2933 | ||
2934 | wm[0] = max(wm[0], min); | |
2935 | for (level = 1; level <= max_level; level++) | |
2936 | wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5)); | |
2937 | ||
2938 | return true; | |
2939 | } | |
2940 | ||
bb726519 | 2941 | static void snb_wm_latency_quirk(struct drm_i915_private *dev_priv) |
e95a2f75 | 2942 | { |
e95a2f75 VS |
2943 | bool changed; |
2944 | ||
2945 | /* | |
2946 | * The BIOS provided WM memory latency values are often | |
2947 | * inadequate for high resolution displays. Adjust them. | |
2948 | */ | |
2949 | changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) | | |
2950 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) | | |
2951 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12); | |
2952 | ||
2953 | if (!changed) | |
2954 | return; | |
2955 | ||
2956 | DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n"); | |
5db94019 TU |
2957 | intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency); |
2958 | intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency); | |
2959 | intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency); | |
e95a2f75 VS |
2960 | } |
2961 | ||
bb726519 | 2962 | static void ilk_setup_wm_latency(struct drm_i915_private *dev_priv) |
53615a5e | 2963 | { |
bb726519 | 2964 | intel_read_wm_latency(dev_priv, dev_priv->wm.pri_latency); |
53615a5e VS |
2965 | |
2966 | memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency, | |
2967 | sizeof(dev_priv->wm.pri_latency)); | |
2968 | memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency, | |
2969 | sizeof(dev_priv->wm.pri_latency)); | |
2970 | ||
5db94019 | 2971 | intel_fixup_spr_wm_latency(dev_priv, dev_priv->wm.spr_latency); |
fd6b8f43 | 2972 | intel_fixup_cur_wm_latency(dev_priv, dev_priv->wm.cur_latency); |
26ec971e | 2973 | |
5db94019 TU |
2974 | intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency); |
2975 | intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency); | |
2976 | intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency); | |
e95a2f75 | 2977 | |
5db94019 | 2978 | if (IS_GEN6(dev_priv)) |
bb726519 | 2979 | snb_wm_latency_quirk(dev_priv); |
53615a5e VS |
2980 | } |
2981 | ||
bb726519 | 2982 | static void skl_setup_wm_latency(struct drm_i915_private *dev_priv) |
2af30a5c | 2983 | { |
bb726519 | 2984 | intel_read_wm_latency(dev_priv, dev_priv->wm.skl_latency); |
5db94019 | 2985 | intel_print_wm_latency(dev_priv, "Gen9 Plane", dev_priv->wm.skl_latency); |
2af30a5c PB |
2986 | } |
2987 | ||
ed4a6a7c MR |
2988 | static bool ilk_validate_pipe_wm(struct drm_device *dev, |
2989 | struct intel_pipe_wm *pipe_wm) | |
2990 | { | |
2991 | /* LP0 watermark maximums depend on this pipe alone */ | |
2992 | const struct intel_wm_config config = { | |
2993 | .num_pipes_active = 1, | |
2994 | .sprites_enabled = pipe_wm->sprites_enabled, | |
2995 | .sprites_scaled = pipe_wm->sprites_scaled, | |
2996 | }; | |
2997 | struct ilk_wm_maximums max; | |
2998 | ||
2999 | /* LP0 watermarks always use 1/2 DDB partitioning */ | |
3000 | ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max); | |
3001 | ||
3002 | /* At least LP0 must be valid */ | |
3003 | if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) { | |
3004 | DRM_DEBUG_KMS("LP0 watermark invalid\n"); | |
3005 | return false; | |
3006 | } | |
3007 | ||
3008 | return true; | |
3009 | } | |
3010 | ||
0b2ae6d7 | 3011 | /* Compute new watermarks for the pipe */ |
e3bddded | 3012 | static int ilk_compute_pipe_wm(struct intel_crtc_state *cstate) |
0b2ae6d7 | 3013 | { |
e3bddded ML |
3014 | struct drm_atomic_state *state = cstate->base.state; |
3015 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); | |
86c8bbbe | 3016 | struct intel_pipe_wm *pipe_wm; |
e3bddded | 3017 | struct drm_device *dev = state->dev; |
fac5e23e | 3018 | const struct drm_i915_private *dev_priv = to_i915(dev); |
0434a5fb ML |
3019 | struct drm_plane *plane; |
3020 | const struct drm_plane_state *plane_state; | |
3021 | const struct intel_plane_state *pristate = NULL; | |
3022 | const struct intel_plane_state *sprstate = NULL; | |
3023 | const struct intel_plane_state *curstate = NULL; | |
5db94019 | 3024 | int level, max_level = ilk_wm_max_level(dev_priv), usable_level; |
820c1980 | 3025 | struct ilk_wm_maximums max; |
0b2ae6d7 | 3026 | |
e8f1f02e | 3027 | pipe_wm = &cstate->wm.ilk.optimal; |
86c8bbbe | 3028 | |
0434a5fb ML |
3029 | drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, &cstate->base) { |
3030 | const struct intel_plane_state *ps = to_intel_plane_state(plane_state); | |
e3bddded | 3031 | |
0434a5fb | 3032 | if (plane->type == DRM_PLANE_TYPE_PRIMARY) |
e3bddded | 3033 | pristate = ps; |
0434a5fb | 3034 | else if (plane->type == DRM_PLANE_TYPE_OVERLAY) |
e3bddded | 3035 | sprstate = ps; |
0434a5fb | 3036 | else if (plane->type == DRM_PLANE_TYPE_CURSOR) |
e3bddded | 3037 | curstate = ps; |
43d59eda MR |
3038 | } |
3039 | ||
ed4a6a7c | 3040 | pipe_wm->pipe_enabled = cstate->base.active; |
e3bddded | 3041 | if (sprstate) { |
936e71e3 VS |
3042 | pipe_wm->sprites_enabled = sprstate->base.visible; |
3043 | pipe_wm->sprites_scaled = sprstate->base.visible && | |
3044 | (drm_rect_width(&sprstate->base.dst) != drm_rect_width(&sprstate->base.src) >> 16 || | |
3045 | drm_rect_height(&sprstate->base.dst) != drm_rect_height(&sprstate->base.src) >> 16); | |
e3bddded ML |
3046 | } |
3047 | ||
d81f04c5 ML |
3048 | usable_level = max_level; |
3049 | ||
7b39a0b7 | 3050 | /* ILK/SNB: LP2+ watermarks only w/o sprites */ |
175fded1 | 3051 | if (INTEL_GEN(dev_priv) <= 6 && pipe_wm->sprites_enabled) |
d81f04c5 | 3052 | usable_level = 1; |
7b39a0b7 VS |
3053 | |
3054 | /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */ | |
ed4a6a7c | 3055 | if (pipe_wm->sprites_scaled) |
d81f04c5 | 3056 | usable_level = 0; |
7b39a0b7 | 3057 | |
71f0a626 | 3058 | memset(&pipe_wm->wm, 0, sizeof(pipe_wm->wm)); |
0434a5fb ML |
3059 | ilk_compute_wm_level(dev_priv, intel_crtc, 0, cstate, |
3060 | pristate, sprstate, curstate, &pipe_wm->wm[0]); | |
0b2ae6d7 | 3061 | |
8652744b | 3062 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
532f7a7f | 3063 | pipe_wm->linetime = hsw_compute_linetime_wm(cstate); |
0b2ae6d7 | 3064 | |
ed4a6a7c | 3065 | if (!ilk_validate_pipe_wm(dev, pipe_wm)) |
1a426d61 | 3066 | return -EINVAL; |
a3cb4048 | 3067 | |
175fded1 | 3068 | ilk_compute_wm_reg_maximums(dev_priv, 1, &max); |
a3cb4048 | 3069 | |
0434a5fb ML |
3070 | for (level = 1; level <= usable_level; level++) { |
3071 | struct intel_wm_level *wm = &pipe_wm->wm[level]; | |
a3cb4048 | 3072 | |
86c8bbbe | 3073 | ilk_compute_wm_level(dev_priv, intel_crtc, level, cstate, |
d81f04c5 | 3074 | pristate, sprstate, curstate, wm); |
a3cb4048 VS |
3075 | |
3076 | /* | |
3077 | * Disable any watermark level that exceeds the | |
3078 | * register maximums since such watermarks are | |
3079 | * always invalid. | |
3080 | */ | |
0434a5fb ML |
3081 | if (!ilk_validate_wm_level(level, &max, wm)) { |
3082 | memset(wm, 0, sizeof(*wm)); | |
3083 | break; | |
3084 | } | |
a3cb4048 VS |
3085 | } |
3086 | ||
86c8bbbe | 3087 | return 0; |
0b2ae6d7 VS |
3088 | } |
3089 | ||
ed4a6a7c MR |
3090 | /* |
3091 | * Build a set of 'intermediate' watermark values that satisfy both the old | |
3092 | * state and the new state. These can be programmed to the hardware | |
3093 | * immediately. | |
3094 | */ | |
3095 | static int ilk_compute_intermediate_wm(struct drm_device *dev, | |
3096 | struct intel_crtc *intel_crtc, | |
3097 | struct intel_crtc_state *newstate) | |
3098 | { | |
e8f1f02e | 3099 | struct intel_pipe_wm *a = &newstate->wm.ilk.intermediate; |
ed4a6a7c | 3100 | struct intel_pipe_wm *b = &intel_crtc->wm.active.ilk; |
5db94019 | 3101 | int level, max_level = ilk_wm_max_level(to_i915(dev)); |
ed4a6a7c MR |
3102 | |
3103 | /* | |
3104 | * Start with the final, target watermarks, then combine with the | |
3105 | * currently active watermarks to get values that are safe both before | |
3106 | * and after the vblank. | |
3107 | */ | |
e8f1f02e | 3108 | *a = newstate->wm.ilk.optimal; |
ed4a6a7c MR |
3109 | a->pipe_enabled |= b->pipe_enabled; |
3110 | a->sprites_enabled |= b->sprites_enabled; | |
3111 | a->sprites_scaled |= b->sprites_scaled; | |
3112 | ||
3113 | for (level = 0; level <= max_level; level++) { | |
3114 | struct intel_wm_level *a_wm = &a->wm[level]; | |
3115 | const struct intel_wm_level *b_wm = &b->wm[level]; | |
3116 | ||
3117 | a_wm->enable &= b_wm->enable; | |
3118 | a_wm->pri_val = max(a_wm->pri_val, b_wm->pri_val); | |
3119 | a_wm->spr_val = max(a_wm->spr_val, b_wm->spr_val); | |
3120 | a_wm->cur_val = max(a_wm->cur_val, b_wm->cur_val); | |
3121 | a_wm->fbc_val = max(a_wm->fbc_val, b_wm->fbc_val); | |
3122 | } | |
3123 | ||
3124 | /* | |
3125 | * We need to make sure that these merged watermark values are | |
3126 | * actually a valid configuration themselves. If they're not, | |
3127 | * there's no safe way to transition from the old state to | |
3128 | * the new state, so we need to fail the atomic transaction. | |
3129 | */ | |
3130 | if (!ilk_validate_pipe_wm(dev, a)) | |
3131 | return -EINVAL; | |
3132 | ||
3133 | /* | |
3134 | * If our intermediate WM are identical to the final WM, then we can | |
3135 | * omit the post-vblank programming; only update if it's different. | |
3136 | */ | |
5eeb798b VS |
3137 | if (memcmp(a, &newstate->wm.ilk.optimal, sizeof(*a)) != 0) |
3138 | newstate->wm.need_postvbl_update = true; | |
ed4a6a7c MR |
3139 | |
3140 | return 0; | |
3141 | } | |
3142 | ||
0b2ae6d7 VS |
3143 | /* |
3144 | * Merge the watermarks from all active pipes for a specific level. | |
3145 | */ | |
3146 | static void ilk_merge_wm_level(struct drm_device *dev, | |
3147 | int level, | |
3148 | struct intel_wm_level *ret_wm) | |
3149 | { | |
3150 | const struct intel_crtc *intel_crtc; | |
3151 | ||
d52fea5b VS |
3152 | ret_wm->enable = true; |
3153 | ||
d3fcc808 | 3154 | for_each_intel_crtc(dev, intel_crtc) { |
ed4a6a7c | 3155 | const struct intel_pipe_wm *active = &intel_crtc->wm.active.ilk; |
fe392efd VS |
3156 | const struct intel_wm_level *wm = &active->wm[level]; |
3157 | ||
3158 | if (!active->pipe_enabled) | |
3159 | continue; | |
0b2ae6d7 | 3160 | |
d52fea5b VS |
3161 | /* |
3162 | * The watermark values may have been used in the past, | |
3163 | * so we must maintain them in the registers for some | |
3164 | * time even if the level is now disabled. | |
3165 | */ | |
0b2ae6d7 | 3166 | if (!wm->enable) |
d52fea5b | 3167 | ret_wm->enable = false; |
0b2ae6d7 VS |
3168 | |
3169 | ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val); | |
3170 | ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val); | |
3171 | ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val); | |
3172 | ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val); | |
3173 | } | |
0b2ae6d7 VS |
3174 | } |
3175 | ||
3176 | /* | |
3177 | * Merge all low power watermarks for all active pipes. | |
3178 | */ | |
3179 | static void ilk_wm_merge(struct drm_device *dev, | |
0ba22e26 | 3180 | const struct intel_wm_config *config, |
820c1980 | 3181 | const struct ilk_wm_maximums *max, |
0b2ae6d7 VS |
3182 | struct intel_pipe_wm *merged) |
3183 | { | |
fac5e23e | 3184 | struct drm_i915_private *dev_priv = to_i915(dev); |
5db94019 | 3185 | int level, max_level = ilk_wm_max_level(dev_priv); |
d52fea5b | 3186 | int last_enabled_level = max_level; |
0b2ae6d7 | 3187 | |
0ba22e26 | 3188 | /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */ |
fd6b8f43 | 3189 | if ((INTEL_GEN(dev_priv) <= 6 || IS_IVYBRIDGE(dev_priv)) && |
0ba22e26 | 3190 | config->num_pipes_active > 1) |
1204d5ba | 3191 | last_enabled_level = 0; |
0ba22e26 | 3192 | |
6c8b6c28 | 3193 | /* ILK: FBC WM must be disabled always */ |
175fded1 | 3194 | merged->fbc_wm_enabled = INTEL_GEN(dev_priv) >= 6; |
0b2ae6d7 VS |
3195 | |
3196 | /* merge each WM1+ level */ | |
3197 | for (level = 1; level <= max_level; level++) { | |
3198 | struct intel_wm_level *wm = &merged->wm[level]; | |
3199 | ||
3200 | ilk_merge_wm_level(dev, level, wm); | |
3201 | ||
d52fea5b VS |
3202 | if (level > last_enabled_level) |
3203 | wm->enable = false; | |
3204 | else if (!ilk_validate_wm_level(level, max, wm)) | |
3205 | /* make sure all following levels get disabled */ | |
3206 | last_enabled_level = level - 1; | |
0b2ae6d7 VS |
3207 | |
3208 | /* | |
3209 | * The spec says it is preferred to disable | |
3210 | * FBC WMs instead of disabling a WM level. | |
3211 | */ | |
3212 | if (wm->fbc_val > max->fbc) { | |
d52fea5b VS |
3213 | if (wm->enable) |
3214 | merged->fbc_wm_enabled = false; | |
0b2ae6d7 VS |
3215 | wm->fbc_val = 0; |
3216 | } | |
3217 | } | |
6c8b6c28 VS |
3218 | |
3219 | /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */ | |
3220 | /* | |
3221 | * FIXME this is racy. FBC might get enabled later. | |
3222 | * What we should check here is whether FBC can be | |
3223 | * enabled sometime later. | |
3224 | */ | |
5db94019 | 3225 | if (IS_GEN5(dev_priv) && !merged->fbc_wm_enabled && |
0e631adc | 3226 | intel_fbc_is_active(dev_priv)) { |
6c8b6c28 VS |
3227 | for (level = 2; level <= max_level; level++) { |
3228 | struct intel_wm_level *wm = &merged->wm[level]; | |
3229 | ||
3230 | wm->enable = false; | |
3231 | } | |
3232 | } | |
0b2ae6d7 VS |
3233 | } |
3234 | ||
b380ca3c VS |
3235 | static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm) |
3236 | { | |
3237 | /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */ | |
3238 | return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable); | |
3239 | } | |
3240 | ||
a68d68ee VS |
3241 | /* The value we need to program into the WM_LPx latency field */ |
3242 | static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level) | |
3243 | { | |
fac5e23e | 3244 | struct drm_i915_private *dev_priv = to_i915(dev); |
a68d68ee | 3245 | |
8652744b | 3246 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
a68d68ee VS |
3247 | return 2 * level; |
3248 | else | |
3249 | return dev_priv->wm.pri_latency[level]; | |
3250 | } | |
3251 | ||
820c1980 | 3252 | static void ilk_compute_wm_results(struct drm_device *dev, |
0362c781 | 3253 | const struct intel_pipe_wm *merged, |
609cedef | 3254 | enum intel_ddb_partitioning partitioning, |
820c1980 | 3255 | struct ilk_wm_values *results) |
801bcfff | 3256 | { |
175fded1 | 3257 | struct drm_i915_private *dev_priv = to_i915(dev); |
0b2ae6d7 VS |
3258 | struct intel_crtc *intel_crtc; |
3259 | int level, wm_lp; | |
cca32e9a | 3260 | |
0362c781 | 3261 | results->enable_fbc_wm = merged->fbc_wm_enabled; |
609cedef | 3262 | results->partitioning = partitioning; |
cca32e9a | 3263 | |
0b2ae6d7 | 3264 | /* LP1+ register values */ |
cca32e9a | 3265 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
1fd527cc | 3266 | const struct intel_wm_level *r; |
801bcfff | 3267 | |
b380ca3c | 3268 | level = ilk_wm_lp_to_level(wm_lp, merged); |
0b2ae6d7 | 3269 | |
0362c781 | 3270 | r = &merged->wm[level]; |
cca32e9a | 3271 | |
d52fea5b VS |
3272 | /* |
3273 | * Maintain the watermark values even if the level is | |
3274 | * disabled. Doing otherwise could cause underruns. | |
3275 | */ | |
3276 | results->wm_lp[wm_lp - 1] = | |
a68d68ee | 3277 | (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) | |
416f4727 VS |
3278 | (r->pri_val << WM1_LP_SR_SHIFT) | |
3279 | r->cur_val; | |
3280 | ||
d52fea5b VS |
3281 | if (r->enable) |
3282 | results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN; | |
3283 | ||
175fded1 | 3284 | if (INTEL_GEN(dev_priv) >= 8) |
416f4727 VS |
3285 | results->wm_lp[wm_lp - 1] |= |
3286 | r->fbc_val << WM1_LP_FBC_SHIFT_BDW; | |
3287 | else | |
3288 | results->wm_lp[wm_lp - 1] |= | |
3289 | r->fbc_val << WM1_LP_FBC_SHIFT; | |
3290 | ||
d52fea5b VS |
3291 | /* |
3292 | * Always set WM1S_LP_EN when spr_val != 0, even if the | |
3293 | * level is disabled. Doing otherwise could cause underruns. | |
3294 | */ | |
175fded1 | 3295 | if (INTEL_GEN(dev_priv) <= 6 && r->spr_val) { |
6cef2b8a VS |
3296 | WARN_ON(wm_lp != 1); |
3297 | results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val; | |
3298 | } else | |
3299 | results->wm_lp_spr[wm_lp - 1] = r->spr_val; | |
cca32e9a | 3300 | } |
801bcfff | 3301 | |
0b2ae6d7 | 3302 | /* LP0 register values */ |
d3fcc808 | 3303 | for_each_intel_crtc(dev, intel_crtc) { |
0b2ae6d7 | 3304 | enum pipe pipe = intel_crtc->pipe; |
ed4a6a7c MR |
3305 | const struct intel_wm_level *r = |
3306 | &intel_crtc->wm.active.ilk.wm[0]; | |
0b2ae6d7 VS |
3307 | |
3308 | if (WARN_ON(!r->enable)) | |
3309 | continue; | |
3310 | ||
ed4a6a7c | 3311 | results->wm_linetime[pipe] = intel_crtc->wm.active.ilk.linetime; |
1011d8c4 | 3312 | |
0b2ae6d7 VS |
3313 | results->wm_pipe[pipe] = |
3314 | (r->pri_val << WM0_PIPE_PLANE_SHIFT) | | |
3315 | (r->spr_val << WM0_PIPE_SPRITE_SHIFT) | | |
3316 | r->cur_val; | |
801bcfff PZ |
3317 | } |
3318 | } | |
3319 | ||
861f3389 PZ |
3320 | /* Find the result with the highest level enabled. Check for enable_fbc_wm in |
3321 | * case both are at the same level. Prefer r1 in case they're the same. */ | |
820c1980 | 3322 | static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev, |
198a1e9b VS |
3323 | struct intel_pipe_wm *r1, |
3324 | struct intel_pipe_wm *r2) | |
861f3389 | 3325 | { |
5db94019 | 3326 | int level, max_level = ilk_wm_max_level(to_i915(dev)); |
198a1e9b | 3327 | int level1 = 0, level2 = 0; |
861f3389 | 3328 | |
198a1e9b VS |
3329 | for (level = 1; level <= max_level; level++) { |
3330 | if (r1->wm[level].enable) | |
3331 | level1 = level; | |
3332 | if (r2->wm[level].enable) | |
3333 | level2 = level; | |
861f3389 PZ |
3334 | } |
3335 | ||
198a1e9b VS |
3336 | if (level1 == level2) { |
3337 | if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled) | |
861f3389 PZ |
3338 | return r2; |
3339 | else | |
3340 | return r1; | |
198a1e9b | 3341 | } else if (level1 > level2) { |
861f3389 PZ |
3342 | return r1; |
3343 | } else { | |
3344 | return r2; | |
3345 | } | |
3346 | } | |
3347 | ||
49a687c4 VS |
3348 | /* dirty bits used to track which watermarks need changes */ |
3349 | #define WM_DIRTY_PIPE(pipe) (1 << (pipe)) | |
3350 | #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe))) | |
3351 | #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp))) | |
3352 | #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3)) | |
3353 | #define WM_DIRTY_FBC (1 << 24) | |
3354 | #define WM_DIRTY_DDB (1 << 25) | |
3355 | ||
055e393f | 3356 | static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv, |
820c1980 ID |
3357 | const struct ilk_wm_values *old, |
3358 | const struct ilk_wm_values *new) | |
49a687c4 VS |
3359 | { |
3360 | unsigned int dirty = 0; | |
3361 | enum pipe pipe; | |
3362 | int wm_lp; | |
3363 | ||
055e393f | 3364 | for_each_pipe(dev_priv, pipe) { |
49a687c4 VS |
3365 | if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) { |
3366 | dirty |= WM_DIRTY_LINETIME(pipe); | |
3367 | /* Must disable LP1+ watermarks too */ | |
3368 | dirty |= WM_DIRTY_LP_ALL; | |
3369 | } | |
3370 | ||
3371 | if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) { | |
3372 | dirty |= WM_DIRTY_PIPE(pipe); | |
3373 | /* Must disable LP1+ watermarks too */ | |
3374 | dirty |= WM_DIRTY_LP_ALL; | |
3375 | } | |
3376 | } | |
3377 | ||
3378 | if (old->enable_fbc_wm != new->enable_fbc_wm) { | |
3379 | dirty |= WM_DIRTY_FBC; | |
3380 | /* Must disable LP1+ watermarks too */ | |
3381 | dirty |= WM_DIRTY_LP_ALL; | |
3382 | } | |
3383 | ||
3384 | if (old->partitioning != new->partitioning) { | |
3385 | dirty |= WM_DIRTY_DDB; | |
3386 | /* Must disable LP1+ watermarks too */ | |
3387 | dirty |= WM_DIRTY_LP_ALL; | |
3388 | } | |
3389 | ||
3390 | /* LP1+ watermarks already deemed dirty, no need to continue */ | |
3391 | if (dirty & WM_DIRTY_LP_ALL) | |
3392 | return dirty; | |
3393 | ||
3394 | /* Find the lowest numbered LP1+ watermark in need of an update... */ | |
3395 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { | |
3396 | if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] || | |
3397 | old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1]) | |
3398 | break; | |
3399 | } | |
3400 | ||
3401 | /* ...and mark it and all higher numbered LP1+ watermarks as dirty */ | |
3402 | for (; wm_lp <= 3; wm_lp++) | |
3403 | dirty |= WM_DIRTY_LP(wm_lp); | |
3404 | ||
3405 | return dirty; | |
3406 | } | |
3407 | ||
8553c18e VS |
3408 | static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv, |
3409 | unsigned int dirty) | |
801bcfff | 3410 | { |
820c1980 | 3411 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
8553c18e | 3412 | bool changed = false; |
801bcfff | 3413 | |
facd619b VS |
3414 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) { |
3415 | previous->wm_lp[2] &= ~WM1_LP_SR_EN; | |
3416 | I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]); | |
8553c18e | 3417 | changed = true; |
facd619b VS |
3418 | } |
3419 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) { | |
3420 | previous->wm_lp[1] &= ~WM1_LP_SR_EN; | |
3421 | I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]); | |
8553c18e | 3422 | changed = true; |
facd619b VS |
3423 | } |
3424 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) { | |
3425 | previous->wm_lp[0] &= ~WM1_LP_SR_EN; | |
3426 | I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]); | |
8553c18e | 3427 | changed = true; |
facd619b | 3428 | } |
801bcfff | 3429 | |
facd619b VS |
3430 | /* |
3431 | * Don't touch WM1S_LP_EN here. | |
3432 | * Doing so could cause underruns. | |
3433 | */ | |
6cef2b8a | 3434 | |
8553c18e VS |
3435 | return changed; |
3436 | } | |
3437 | ||
3438 | /* | |
3439 | * The spec says we shouldn't write when we don't need, because every write | |
3440 | * causes WMs to be re-evaluated, expending some power. | |
3441 | */ | |
820c1980 ID |
3442 | static void ilk_write_wm_values(struct drm_i915_private *dev_priv, |
3443 | struct ilk_wm_values *results) | |
8553c18e | 3444 | { |
820c1980 | 3445 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
8553c18e VS |
3446 | unsigned int dirty; |
3447 | uint32_t val; | |
3448 | ||
055e393f | 3449 | dirty = ilk_compute_wm_dirty(dev_priv, previous, results); |
8553c18e VS |
3450 | if (!dirty) |
3451 | return; | |
3452 | ||
3453 | _ilk_disable_lp_wm(dev_priv, dirty); | |
3454 | ||
49a687c4 | 3455 | if (dirty & WM_DIRTY_PIPE(PIPE_A)) |
801bcfff | 3456 | I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]); |
49a687c4 | 3457 | if (dirty & WM_DIRTY_PIPE(PIPE_B)) |
801bcfff | 3458 | I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]); |
49a687c4 | 3459 | if (dirty & WM_DIRTY_PIPE(PIPE_C)) |
801bcfff PZ |
3460 | I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]); |
3461 | ||
49a687c4 | 3462 | if (dirty & WM_DIRTY_LINETIME(PIPE_A)) |
801bcfff | 3463 | I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]); |
49a687c4 | 3464 | if (dirty & WM_DIRTY_LINETIME(PIPE_B)) |
801bcfff | 3465 | I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]); |
49a687c4 | 3466 | if (dirty & WM_DIRTY_LINETIME(PIPE_C)) |
801bcfff PZ |
3467 | I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]); |
3468 | ||
49a687c4 | 3469 | if (dirty & WM_DIRTY_DDB) { |
8652744b | 3470 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) { |
ac9545fd VS |
3471 | val = I915_READ(WM_MISC); |
3472 | if (results->partitioning == INTEL_DDB_PART_1_2) | |
3473 | val &= ~WM_MISC_DATA_PARTITION_5_6; | |
3474 | else | |
3475 | val |= WM_MISC_DATA_PARTITION_5_6; | |
3476 | I915_WRITE(WM_MISC, val); | |
3477 | } else { | |
3478 | val = I915_READ(DISP_ARB_CTL2); | |
3479 | if (results->partitioning == INTEL_DDB_PART_1_2) | |
3480 | val &= ~DISP_DATA_PARTITION_5_6; | |
3481 | else | |
3482 | val |= DISP_DATA_PARTITION_5_6; | |
3483 | I915_WRITE(DISP_ARB_CTL2, val); | |
3484 | } | |
1011d8c4 PZ |
3485 | } |
3486 | ||
49a687c4 | 3487 | if (dirty & WM_DIRTY_FBC) { |
cca32e9a PZ |
3488 | val = I915_READ(DISP_ARB_CTL); |
3489 | if (results->enable_fbc_wm) | |
3490 | val &= ~DISP_FBC_WM_DIS; | |
3491 | else | |
3492 | val |= DISP_FBC_WM_DIS; | |
3493 | I915_WRITE(DISP_ARB_CTL, val); | |
3494 | } | |
3495 | ||
954911eb ID |
3496 | if (dirty & WM_DIRTY_LP(1) && |
3497 | previous->wm_lp_spr[0] != results->wm_lp_spr[0]) | |
3498 | I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]); | |
3499 | ||
175fded1 | 3500 | if (INTEL_GEN(dev_priv) >= 7) { |
6cef2b8a VS |
3501 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1]) |
3502 | I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]); | |
3503 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2]) | |
3504 | I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]); | |
3505 | } | |
801bcfff | 3506 | |
facd619b | 3507 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0]) |
801bcfff | 3508 | I915_WRITE(WM1_LP_ILK, results->wm_lp[0]); |
facd619b | 3509 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1]) |
801bcfff | 3510 | I915_WRITE(WM2_LP_ILK, results->wm_lp[1]); |
facd619b | 3511 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2]) |
801bcfff | 3512 | I915_WRITE(WM3_LP_ILK, results->wm_lp[2]); |
609cedef VS |
3513 | |
3514 | dev_priv->wm.hw = *results; | |
801bcfff PZ |
3515 | } |
3516 | ||
ed4a6a7c | 3517 | bool ilk_disable_lp_wm(struct drm_device *dev) |
8553c18e | 3518 | { |
fac5e23e | 3519 | struct drm_i915_private *dev_priv = to_i915(dev); |
8553c18e VS |
3520 | |
3521 | return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL); | |
3522 | } | |
3523 | ||
656d1b89 | 3524 | #define SKL_SAGV_BLOCK_TIME 30 /* µs */ |
b9cec075 | 3525 | |
ee3d532f PZ |
3526 | /* |
3527 | * FIXME: We still don't have the proper code detect if we need to apply the WA, | |
3528 | * so assume we'll always need it in order to avoid underruns. | |
3529 | */ | |
3530 | static bool skl_needs_memory_bw_wa(struct intel_atomic_state *state) | |
3531 | { | |
3532 | struct drm_i915_private *dev_priv = to_i915(state->base.dev); | |
3533 | ||
b976dc53 | 3534 | if (IS_GEN9_BC(dev_priv) || IS_BROXTON(dev_priv)) |
ee3d532f PZ |
3535 | return true; |
3536 | ||
3537 | return false; | |
3538 | } | |
3539 | ||
56feca91 PZ |
3540 | static bool |
3541 | intel_has_sagv(struct drm_i915_private *dev_priv) | |
3542 | { | |
82525c17 | 3543 | if (IS_KABYLAKE(dev_priv) || IS_COFFEELAKE(dev_priv)) |
6e3100ec PZ |
3544 | return true; |
3545 | ||
3546 | if (IS_SKYLAKE(dev_priv) && | |
3547 | dev_priv->sagv_status != I915_SAGV_NOT_CONTROLLED) | |
3548 | return true; | |
3549 | ||
3550 | return false; | |
56feca91 PZ |
3551 | } |
3552 | ||
656d1b89 L |
3553 | /* |
3554 | * SAGV dynamically adjusts the system agent voltage and clock frequencies | |
3555 | * depending on power and performance requirements. The display engine access | |
3556 | * to system memory is blocked during the adjustment time. Because of the | |
3557 | * blocking time, having this enabled can cause full system hangs and/or pipe | |
3558 | * underruns if we don't meet all of the following requirements: | |
3559 | * | |
3560 | * - <= 1 pipe enabled | |
3561 | * - All planes can enable watermarks for latencies >= SAGV engine block time | |
3562 | * - We're not using an interlaced display configuration | |
3563 | */ | |
3564 | int | |
16dcdc4e | 3565 | intel_enable_sagv(struct drm_i915_private *dev_priv) |
656d1b89 L |
3566 | { |
3567 | int ret; | |
3568 | ||
56feca91 PZ |
3569 | if (!intel_has_sagv(dev_priv)) |
3570 | return 0; | |
3571 | ||
3572 | if (dev_priv->sagv_status == I915_SAGV_ENABLED) | |
656d1b89 L |
3573 | return 0; |
3574 | ||
3575 | DRM_DEBUG_KMS("Enabling the SAGV\n"); | |
3576 | mutex_lock(&dev_priv->rps.hw_lock); | |
3577 | ||
3578 | ret = sandybridge_pcode_write(dev_priv, GEN9_PCODE_SAGV_CONTROL, | |
3579 | GEN9_SAGV_ENABLE); | |
3580 | ||
3581 | /* We don't need to wait for the SAGV when enabling */ | |
3582 | mutex_unlock(&dev_priv->rps.hw_lock); | |
3583 | ||
3584 | /* | |
3585 | * Some skl systems, pre-release machines in particular, | |
3586 | * don't actually have an SAGV. | |
3587 | */ | |
6e3100ec | 3588 | if (IS_SKYLAKE(dev_priv) && ret == -ENXIO) { |
656d1b89 | 3589 | DRM_DEBUG_DRIVER("No SAGV found on system, ignoring\n"); |
16dcdc4e | 3590 | dev_priv->sagv_status = I915_SAGV_NOT_CONTROLLED; |
656d1b89 L |
3591 | return 0; |
3592 | } else if (ret < 0) { | |
3593 | DRM_ERROR("Failed to enable the SAGV\n"); | |
3594 | return ret; | |
3595 | } | |
3596 | ||
16dcdc4e | 3597 | dev_priv->sagv_status = I915_SAGV_ENABLED; |
656d1b89 L |
3598 | return 0; |
3599 | } | |
3600 | ||
656d1b89 | 3601 | int |
16dcdc4e | 3602 | intel_disable_sagv(struct drm_i915_private *dev_priv) |
656d1b89 | 3603 | { |
b3b8e999 | 3604 | int ret; |
656d1b89 | 3605 | |
56feca91 PZ |
3606 | if (!intel_has_sagv(dev_priv)) |
3607 | return 0; | |
3608 | ||
3609 | if (dev_priv->sagv_status == I915_SAGV_DISABLED) | |
656d1b89 L |
3610 | return 0; |
3611 | ||
3612 | DRM_DEBUG_KMS("Disabling the SAGV\n"); | |
3613 | mutex_lock(&dev_priv->rps.hw_lock); | |
3614 | ||
3615 | /* bspec says to keep retrying for at least 1 ms */ | |
b3b8e999 ID |
3616 | ret = skl_pcode_request(dev_priv, GEN9_PCODE_SAGV_CONTROL, |
3617 | GEN9_SAGV_DISABLE, | |
3618 | GEN9_SAGV_IS_DISABLED, GEN9_SAGV_IS_DISABLED, | |
3619 | 1); | |
656d1b89 L |
3620 | mutex_unlock(&dev_priv->rps.hw_lock); |
3621 | ||
656d1b89 L |
3622 | /* |
3623 | * Some skl systems, pre-release machines in particular, | |
3624 | * don't actually have an SAGV. | |
3625 | */ | |
b3b8e999 | 3626 | if (IS_SKYLAKE(dev_priv) && ret == -ENXIO) { |
656d1b89 | 3627 | DRM_DEBUG_DRIVER("No SAGV found on system, ignoring\n"); |
16dcdc4e | 3628 | dev_priv->sagv_status = I915_SAGV_NOT_CONTROLLED; |
656d1b89 | 3629 | return 0; |
b3b8e999 ID |
3630 | } else if (ret < 0) { |
3631 | DRM_ERROR("Failed to disable the SAGV (%d)\n", ret); | |
3632 | return ret; | |
656d1b89 L |
3633 | } |
3634 | ||
16dcdc4e | 3635 | dev_priv->sagv_status = I915_SAGV_DISABLED; |
656d1b89 L |
3636 | return 0; |
3637 | } | |
3638 | ||
16dcdc4e | 3639 | bool intel_can_enable_sagv(struct drm_atomic_state *state) |
656d1b89 L |
3640 | { |
3641 | struct drm_device *dev = state->dev; | |
3642 | struct drm_i915_private *dev_priv = to_i915(dev); | |
3643 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
ee3d532f PZ |
3644 | struct intel_crtc *crtc; |
3645 | struct intel_plane *plane; | |
d8c0fafc | 3646 | struct intel_crtc_state *cstate; |
656d1b89 | 3647 | enum pipe pipe; |
d8c0fafc | 3648 | int level, latency; |
656d1b89 | 3649 | |
56feca91 PZ |
3650 | if (!intel_has_sagv(dev_priv)) |
3651 | return false; | |
3652 | ||
656d1b89 L |
3653 | /* |
3654 | * SKL workaround: bspec recommends we disable the SAGV when we have | |
3655 | * more then one pipe enabled | |
3656 | * | |
3657 | * If there are no active CRTCs, no additional checks need be performed | |
3658 | */ | |
3659 | if (hweight32(intel_state->active_crtcs) == 0) | |
3660 | return true; | |
3661 | else if (hweight32(intel_state->active_crtcs) > 1) | |
3662 | return false; | |
3663 | ||
3664 | /* Since we're now guaranteed to only have one active CRTC... */ | |
3665 | pipe = ffs(intel_state->active_crtcs) - 1; | |
98187836 | 3666 | crtc = intel_get_crtc_for_pipe(dev_priv, pipe); |
d8c0fafc | 3667 | cstate = to_intel_crtc_state(crtc->base.state); |
656d1b89 | 3668 | |
c89cadd5 | 3669 | if (crtc->base.state->adjusted_mode.flags & DRM_MODE_FLAG_INTERLACE) |
656d1b89 L |
3670 | return false; |
3671 | ||
ee3d532f | 3672 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
d5cdfdf5 VS |
3673 | struct skl_plane_wm *wm = |
3674 | &cstate->wm.skl.optimal.planes[plane->id]; | |
ee3d532f | 3675 | |
656d1b89 | 3676 | /* Skip this plane if it's not enabled */ |
d8c0fafc | 3677 | if (!wm->wm[0].plane_en) |
656d1b89 L |
3678 | continue; |
3679 | ||
3680 | /* Find the highest enabled wm level for this plane */ | |
5db94019 | 3681 | for (level = ilk_wm_max_level(dev_priv); |
d8c0fafc | 3682 | !wm->wm[level].plane_en; --level) |
656d1b89 L |
3683 | { } |
3684 | ||
ee3d532f PZ |
3685 | latency = dev_priv->wm.skl_latency[level]; |
3686 | ||
3687 | if (skl_needs_memory_bw_wa(intel_state) && | |
bae781b2 | 3688 | plane->base.state->fb->modifier == |
ee3d532f PZ |
3689 | I915_FORMAT_MOD_X_TILED) |
3690 | latency += 15; | |
3691 | ||
656d1b89 L |
3692 | /* |
3693 | * If any of the planes on this pipe don't enable wm levels | |
3694 | * that incur memory latencies higher then 30µs we can't enable | |
3695 | * the SAGV | |
3696 | */ | |
ee3d532f | 3697 | if (latency < SKL_SAGV_BLOCK_TIME) |
656d1b89 L |
3698 | return false; |
3699 | } | |
3700 | ||
3701 | return true; | |
3702 | } | |
3703 | ||
b9cec075 DL |
3704 | static void |
3705 | skl_ddb_get_pipe_allocation_limits(struct drm_device *dev, | |
024c9045 | 3706 | const struct intel_crtc_state *cstate, |
c107acfe MR |
3707 | struct skl_ddb_entry *alloc, /* out */ |
3708 | int *num_active /* out */) | |
b9cec075 | 3709 | { |
c107acfe MR |
3710 | struct drm_atomic_state *state = cstate->base.state; |
3711 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
3712 | struct drm_i915_private *dev_priv = to_i915(dev); | |
024c9045 | 3713 | struct drm_crtc *for_crtc = cstate->base.crtc; |
b9cec075 DL |
3714 | unsigned int pipe_size, ddb_size; |
3715 | int nth_active_pipe; | |
c107acfe | 3716 | |
a6d3460e | 3717 | if (WARN_ON(!state) || !cstate->base.active) { |
b9cec075 DL |
3718 | alloc->start = 0; |
3719 | alloc->end = 0; | |
a6d3460e | 3720 | *num_active = hweight32(dev_priv->active_crtcs); |
b9cec075 DL |
3721 | return; |
3722 | } | |
3723 | ||
a6d3460e MR |
3724 | if (intel_state->active_pipe_changes) |
3725 | *num_active = hweight32(intel_state->active_crtcs); | |
3726 | else | |
3727 | *num_active = hweight32(dev_priv->active_crtcs); | |
3728 | ||
6f3fff60 D |
3729 | ddb_size = INTEL_INFO(dev_priv)->ddb_size; |
3730 | WARN_ON(ddb_size == 0); | |
b9cec075 DL |
3731 | |
3732 | ddb_size -= 4; /* 4 blocks for bypass path allocation */ | |
3733 | ||
c107acfe | 3734 | /* |
a6d3460e MR |
3735 | * If the state doesn't change the active CRTC's, then there's |
3736 | * no need to recalculate; the existing pipe allocation limits | |
3737 | * should remain unchanged. Note that we're safe from racing | |
3738 | * commits since any racing commit that changes the active CRTC | |
3739 | * list would need to grab _all_ crtc locks, including the one | |
3740 | * we currently hold. | |
c107acfe | 3741 | */ |
a6d3460e | 3742 | if (!intel_state->active_pipe_changes) { |
512b5527 ML |
3743 | /* |
3744 | * alloc may be cleared by clear_intel_crtc_state, | |
3745 | * copy from old state to be sure | |
3746 | */ | |
3747 | *alloc = to_intel_crtc_state(for_crtc->state)->wm.skl.ddb; | |
a6d3460e | 3748 | return; |
c107acfe | 3749 | } |
a6d3460e MR |
3750 | |
3751 | nth_active_pipe = hweight32(intel_state->active_crtcs & | |
3752 | (drm_crtc_mask(for_crtc) - 1)); | |
3753 | pipe_size = ddb_size / hweight32(intel_state->active_crtcs); | |
3754 | alloc->start = nth_active_pipe * ddb_size / *num_active; | |
3755 | alloc->end = alloc->start + pipe_size; | |
b9cec075 DL |
3756 | } |
3757 | ||
c107acfe | 3758 | static unsigned int skl_cursor_allocation(int num_active) |
b9cec075 | 3759 | { |
c107acfe | 3760 | if (num_active == 1) |
b9cec075 DL |
3761 | return 32; |
3762 | ||
3763 | return 8; | |
3764 | } | |
3765 | ||
a269c583 DL |
3766 | static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg) |
3767 | { | |
3768 | entry->start = reg & 0x3ff; | |
3769 | entry->end = (reg >> 16) & 0x3ff; | |
16160e3d DL |
3770 | if (entry->end) |
3771 | entry->end += 1; | |
a269c583 DL |
3772 | } |
3773 | ||
08db6652 DL |
3774 | void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv, |
3775 | struct skl_ddb_allocation *ddb /* out */) | |
a269c583 | 3776 | { |
d5cdfdf5 | 3777 | struct intel_crtc *crtc; |
a269c583 | 3778 | |
b10f1b20 ML |
3779 | memset(ddb, 0, sizeof(*ddb)); |
3780 | ||
d5cdfdf5 | 3781 | for_each_intel_crtc(&dev_priv->drm, crtc) { |
4d800030 | 3782 | enum intel_display_power_domain power_domain; |
d5cdfdf5 VS |
3783 | enum plane_id plane_id; |
3784 | enum pipe pipe = crtc->pipe; | |
4d800030 ID |
3785 | |
3786 | power_domain = POWER_DOMAIN_PIPE(pipe); | |
3787 | if (!intel_display_power_get_if_enabled(dev_priv, power_domain)) | |
b10f1b20 ML |
3788 | continue; |
3789 | ||
d5cdfdf5 VS |
3790 | for_each_plane_id_on_crtc(crtc, plane_id) { |
3791 | u32 val; | |
3792 | ||
3793 | if (plane_id != PLANE_CURSOR) | |
3794 | val = I915_READ(PLANE_BUF_CFG(pipe, plane_id)); | |
3795 | else | |
3796 | val = I915_READ(CUR_BUF_CFG(pipe)); | |
a269c583 | 3797 | |
d5cdfdf5 VS |
3798 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane_id], val); |
3799 | } | |
4d800030 ID |
3800 | |
3801 | intel_display_power_put(dev_priv, power_domain); | |
a269c583 DL |
3802 | } |
3803 | } | |
3804 | ||
9c2f7a9d KM |
3805 | /* |
3806 | * Determines the downscale amount of a plane for the purposes of watermark calculations. | |
3807 | * The bspec defines downscale amount as: | |
3808 | * | |
3809 | * """ | |
3810 | * Horizontal down scale amount = maximum[1, Horizontal source size / | |
3811 | * Horizontal destination size] | |
3812 | * Vertical down scale amount = maximum[1, Vertical source size / | |
3813 | * Vertical destination size] | |
3814 | * Total down scale amount = Horizontal down scale amount * | |
3815 | * Vertical down scale amount | |
3816 | * """ | |
3817 | * | |
3818 | * Return value is provided in 16.16 fixed point form to retain fractional part. | |
3819 | * Caller should take care of dividing & rounding off the value. | |
3820 | */ | |
7084b50b | 3821 | static uint_fixed_16_16_t |
93aa2a1c VS |
3822 | skl_plane_downscale_amount(const struct intel_crtc_state *cstate, |
3823 | const struct intel_plane_state *pstate) | |
9c2f7a9d | 3824 | { |
93aa2a1c | 3825 | struct intel_plane *plane = to_intel_plane(pstate->base.plane); |
9c2f7a9d | 3826 | uint32_t src_w, src_h, dst_w, dst_h; |
7084b50b KM |
3827 | uint_fixed_16_16_t fp_w_ratio, fp_h_ratio; |
3828 | uint_fixed_16_16_t downscale_h, downscale_w; | |
9c2f7a9d | 3829 | |
93aa2a1c | 3830 | if (WARN_ON(!intel_wm_plane_visible(cstate, pstate))) |
7084b50b | 3831 | return u32_to_fixed_16_16(0); |
9c2f7a9d KM |
3832 | |
3833 | /* n.b., src is 16.16 fixed point, dst is whole integer */ | |
93aa2a1c | 3834 | if (plane->id == PLANE_CURSOR) { |
fce5adf5 VS |
3835 | /* |
3836 | * Cursors only support 0/180 degree rotation, | |
3837 | * hence no need to account for rotation here. | |
3838 | */ | |
7084b50b KM |
3839 | src_w = pstate->base.src_w >> 16; |
3840 | src_h = pstate->base.src_h >> 16; | |
93aa2a1c VS |
3841 | dst_w = pstate->base.crtc_w; |
3842 | dst_h = pstate->base.crtc_h; | |
3843 | } else { | |
fce5adf5 VS |
3844 | /* |
3845 | * Src coordinates are already rotated by 270 degrees for | |
3846 | * the 90/270 degree plane rotation cases (to match the | |
3847 | * GTT mapping), hence no need to account for rotation here. | |
3848 | */ | |
7084b50b KM |
3849 | src_w = drm_rect_width(&pstate->base.src) >> 16; |
3850 | src_h = drm_rect_height(&pstate->base.src) >> 16; | |
93aa2a1c VS |
3851 | dst_w = drm_rect_width(&pstate->base.dst); |
3852 | dst_h = drm_rect_height(&pstate->base.dst); | |
3853 | } | |
3854 | ||
7084b50b KM |
3855 | fp_w_ratio = fixed_16_16_div(src_w, dst_w); |
3856 | fp_h_ratio = fixed_16_16_div(src_h, dst_h); | |
3857 | downscale_w = max_fixed_16_16(fp_w_ratio, u32_to_fixed_16_16(1)); | |
3858 | downscale_h = max_fixed_16_16(fp_h_ratio, u32_to_fixed_16_16(1)); | |
9c2f7a9d | 3859 | |
7084b50b | 3860 | return mul_fixed16(downscale_w, downscale_h); |
9c2f7a9d KM |
3861 | } |
3862 | ||
73b0ca8e MK |
3863 | static uint_fixed_16_16_t |
3864 | skl_pipe_downscale_amount(const struct intel_crtc_state *crtc_state) | |
3865 | { | |
3866 | uint_fixed_16_16_t pipe_downscale = u32_to_fixed_16_16(1); | |
3867 | ||
3868 | if (!crtc_state->base.enable) | |
3869 | return pipe_downscale; | |
3870 | ||
3871 | if (crtc_state->pch_pfit.enabled) { | |
3872 | uint32_t src_w, src_h, dst_w, dst_h; | |
3873 | uint32_t pfit_size = crtc_state->pch_pfit.size; | |
3874 | uint_fixed_16_16_t fp_w_ratio, fp_h_ratio; | |
3875 | uint_fixed_16_16_t downscale_h, downscale_w; | |
3876 | ||
3877 | src_w = crtc_state->pipe_src_w; | |
3878 | src_h = crtc_state->pipe_src_h; | |
3879 | dst_w = pfit_size >> 16; | |
3880 | dst_h = pfit_size & 0xffff; | |
3881 | ||
3882 | if (!dst_w || !dst_h) | |
3883 | return pipe_downscale; | |
3884 | ||
3885 | fp_w_ratio = fixed_16_16_div(src_w, dst_w); | |
3886 | fp_h_ratio = fixed_16_16_div(src_h, dst_h); | |
3887 | downscale_w = max_fixed_16_16(fp_w_ratio, u32_to_fixed_16_16(1)); | |
3888 | downscale_h = max_fixed_16_16(fp_h_ratio, u32_to_fixed_16_16(1)); | |
3889 | ||
3890 | pipe_downscale = mul_fixed16(downscale_w, downscale_h); | |
3891 | } | |
3892 | ||
3893 | return pipe_downscale; | |
3894 | } | |
3895 | ||
3896 | int skl_check_pipe_max_pixel_rate(struct intel_crtc *intel_crtc, | |
3897 | struct intel_crtc_state *cstate) | |
3898 | { | |
3899 | struct drm_crtc_state *crtc_state = &cstate->base; | |
3900 | struct drm_atomic_state *state = crtc_state->state; | |
3901 | struct drm_plane *plane; | |
3902 | const struct drm_plane_state *pstate; | |
3903 | struct intel_plane_state *intel_pstate; | |
789f35d7 | 3904 | int crtc_clock, dotclk; |
73b0ca8e MK |
3905 | uint32_t pipe_max_pixel_rate; |
3906 | uint_fixed_16_16_t pipe_downscale; | |
3907 | uint_fixed_16_16_t max_downscale = u32_to_fixed_16_16(1); | |
3908 | ||
3909 | if (!cstate->base.enable) | |
3910 | return 0; | |
3911 | ||
3912 | drm_atomic_crtc_state_for_each_plane_state(plane, pstate, crtc_state) { | |
3913 | uint_fixed_16_16_t plane_downscale; | |
3914 | uint_fixed_16_16_t fp_9_div_8 = fixed_16_16_div(9, 8); | |
3915 | int bpp; | |
3916 | ||
3917 | if (!intel_wm_plane_visible(cstate, | |
3918 | to_intel_plane_state(pstate))) | |
3919 | continue; | |
3920 | ||
3921 | if (WARN_ON(!pstate->fb)) | |
3922 | return -EINVAL; | |
3923 | ||
3924 | intel_pstate = to_intel_plane_state(pstate); | |
3925 | plane_downscale = skl_plane_downscale_amount(cstate, | |
3926 | intel_pstate); | |
3927 | bpp = pstate->fb->format->cpp[0] * 8; | |
3928 | if (bpp == 64) | |
3929 | plane_downscale = mul_fixed16(plane_downscale, | |
3930 | fp_9_div_8); | |
3931 | ||
3932 | max_downscale = max_fixed_16_16(plane_downscale, max_downscale); | |
3933 | } | |
3934 | pipe_downscale = skl_pipe_downscale_amount(cstate); | |
3935 | ||
3936 | pipe_downscale = mul_fixed16(pipe_downscale, max_downscale); | |
3937 | ||
3938 | crtc_clock = crtc_state->adjusted_mode.crtc_clock; | |
789f35d7 ML |
3939 | dotclk = to_intel_atomic_state(state)->cdclk.logical.cdclk; |
3940 | ||
3941 | if (IS_GEMINILAKE(to_i915(intel_crtc->base.dev))) | |
3942 | dotclk *= 2; | |
3943 | ||
3944 | pipe_max_pixel_rate = div_round_up_u32_fixed16(dotclk, pipe_downscale); | |
73b0ca8e MK |
3945 | |
3946 | if (pipe_max_pixel_rate < crtc_clock) { | |
789f35d7 | 3947 | DRM_DEBUG_KMS("Max supported pixel clock with scaling exceeded\n"); |
73b0ca8e MK |
3948 | return -EINVAL; |
3949 | } | |
3950 | ||
3951 | return 0; | |
3952 | } | |
3953 | ||
b9cec075 | 3954 | static unsigned int |
024c9045 MR |
3955 | skl_plane_relative_data_rate(const struct intel_crtc_state *cstate, |
3956 | const struct drm_plane_state *pstate, | |
3957 | int y) | |
b9cec075 | 3958 | { |
93aa2a1c | 3959 | struct intel_plane *plane = to_intel_plane(pstate->plane); |
a280f7dd | 3960 | struct intel_plane_state *intel_pstate = to_intel_plane_state(pstate); |
7084b50b | 3961 | uint32_t data_rate; |
a280f7dd | 3962 | uint32_t width = 0, height = 0; |
8305494e VS |
3963 | struct drm_framebuffer *fb; |
3964 | u32 format; | |
7084b50b | 3965 | uint_fixed_16_16_t down_scale_amount; |
a1de91e5 | 3966 | |
936e71e3 | 3967 | if (!intel_pstate->base.visible) |
a1de91e5 | 3968 | return 0; |
8305494e VS |
3969 | |
3970 | fb = pstate->fb; | |
438b74a5 | 3971 | format = fb->format->format; |
8305494e | 3972 | |
93aa2a1c | 3973 | if (plane->id == PLANE_CURSOR) |
a1de91e5 MR |
3974 | return 0; |
3975 | if (y && format != DRM_FORMAT_NV12) | |
3976 | return 0; | |
a280f7dd | 3977 | |
fce5adf5 VS |
3978 | /* |
3979 | * Src coordinates are already rotated by 270 degrees for | |
3980 | * the 90/270 degree plane rotation cases (to match the | |
3981 | * GTT mapping), hence no need to account for rotation here. | |
3982 | */ | |
936e71e3 VS |
3983 | width = drm_rect_width(&intel_pstate->base.src) >> 16; |
3984 | height = drm_rect_height(&intel_pstate->base.src) >> 16; | |
a280f7dd | 3985 | |
2cd601c6 | 3986 | /* for planar format */ |
a1de91e5 | 3987 | if (format == DRM_FORMAT_NV12) { |
2cd601c6 | 3988 | if (y) /* y-plane data rate */ |
8d19d7d9 | 3989 | data_rate = width * height * |
353c8598 | 3990 | fb->format->cpp[0]; |
2cd601c6 | 3991 | else /* uv-plane data rate */ |
8d19d7d9 | 3992 | data_rate = (width / 2) * (height / 2) * |
353c8598 | 3993 | fb->format->cpp[1]; |
8d19d7d9 KM |
3994 | } else { |
3995 | /* for packed formats */ | |
353c8598 | 3996 | data_rate = width * height * fb->format->cpp[0]; |
2cd601c6 CK |
3997 | } |
3998 | ||
93aa2a1c | 3999 | down_scale_amount = skl_plane_downscale_amount(cstate, intel_pstate); |
8d19d7d9 | 4000 | |
7084b50b | 4001 | return mul_round_up_u32_fixed16(data_rate, down_scale_amount); |
b9cec075 DL |
4002 | } |
4003 | ||
4004 | /* | |
4005 | * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching | |
4006 | * a 8192x4096@32bpp framebuffer: | |
4007 | * 3 * 4096 * 8192 * 4 < 2^32 | |
4008 | */ | |
4009 | static unsigned int | |
1e6ee542 ML |
4010 | skl_get_total_relative_data_rate(struct intel_crtc_state *intel_cstate, |
4011 | unsigned *plane_data_rate, | |
4012 | unsigned *plane_y_data_rate) | |
b9cec075 | 4013 | { |
9c74d826 MR |
4014 | struct drm_crtc_state *cstate = &intel_cstate->base; |
4015 | struct drm_atomic_state *state = cstate->state; | |
c8fe32c1 | 4016 | struct drm_plane *plane; |
c8fe32c1 | 4017 | const struct drm_plane_state *pstate; |
d5cdfdf5 | 4018 | unsigned int total_data_rate = 0; |
a6d3460e MR |
4019 | |
4020 | if (WARN_ON(!state)) | |
4021 | return 0; | |
b9cec075 | 4022 | |
a1de91e5 | 4023 | /* Calculate and cache data rate for each plane */ |
c8fe32c1 | 4024 | drm_atomic_crtc_state_for_each_plane_state(plane, pstate, cstate) { |
d5cdfdf5 VS |
4025 | enum plane_id plane_id = to_intel_plane(plane)->id; |
4026 | unsigned int rate; | |
a6d3460e | 4027 | |
a6d3460e MR |
4028 | /* packed/uv */ |
4029 | rate = skl_plane_relative_data_rate(intel_cstate, | |
4030 | pstate, 0); | |
d5cdfdf5 | 4031 | plane_data_rate[plane_id] = rate; |
1e6ee542 ML |
4032 | |
4033 | total_data_rate += rate; | |
a6d3460e MR |
4034 | |
4035 | /* y-plane */ | |
4036 | rate = skl_plane_relative_data_rate(intel_cstate, | |
4037 | pstate, 1); | |
d5cdfdf5 | 4038 | plane_y_data_rate[plane_id] = rate; |
024c9045 | 4039 | |
1e6ee542 | 4040 | total_data_rate += rate; |
b9cec075 DL |
4041 | } |
4042 | ||
4043 | return total_data_rate; | |
4044 | } | |
4045 | ||
cbcfd14b KM |
4046 | static uint16_t |
4047 | skl_ddb_min_alloc(const struct drm_plane_state *pstate, | |
4048 | const int y) | |
4049 | { | |
4050 | struct drm_framebuffer *fb = pstate->fb; | |
4051 | struct intel_plane_state *intel_pstate = to_intel_plane_state(pstate); | |
4052 | uint32_t src_w, src_h; | |
4053 | uint32_t min_scanlines = 8; | |
4054 | uint8_t plane_bpp; | |
4055 | ||
4056 | if (WARN_ON(!fb)) | |
4057 | return 0; | |
4058 | ||
4059 | /* For packed formats, no y-plane, return 0 */ | |
438b74a5 | 4060 | if (y && fb->format->format != DRM_FORMAT_NV12) |
cbcfd14b KM |
4061 | return 0; |
4062 | ||
4063 | /* For Non Y-tile return 8-blocks */ | |
bae781b2 VS |
4064 | if (fb->modifier != I915_FORMAT_MOD_Y_TILED && |
4065 | fb->modifier != I915_FORMAT_MOD_Yf_TILED) | |
cbcfd14b KM |
4066 | return 8; |
4067 | ||
fce5adf5 VS |
4068 | /* |
4069 | * Src coordinates are already rotated by 270 degrees for | |
4070 | * the 90/270 degree plane rotation cases (to match the | |
4071 | * GTT mapping), hence no need to account for rotation here. | |
4072 | */ | |
936e71e3 VS |
4073 | src_w = drm_rect_width(&intel_pstate->base.src) >> 16; |
4074 | src_h = drm_rect_height(&intel_pstate->base.src) >> 16; | |
cbcfd14b | 4075 | |
cbcfd14b | 4076 | /* Halve UV plane width and height for NV12 */ |
438b74a5 | 4077 | if (fb->format->format == DRM_FORMAT_NV12 && !y) { |
cbcfd14b KM |
4078 | src_w /= 2; |
4079 | src_h /= 2; | |
4080 | } | |
4081 | ||
438b74a5 | 4082 | if (fb->format->format == DRM_FORMAT_NV12 && !y) |
353c8598 | 4083 | plane_bpp = fb->format->cpp[1]; |
cbcfd14b | 4084 | else |
353c8598 | 4085 | plane_bpp = fb->format->cpp[0]; |
cbcfd14b | 4086 | |
bd2ef25d | 4087 | if (drm_rotation_90_or_270(pstate->rotation)) { |
cbcfd14b KM |
4088 | switch (plane_bpp) { |
4089 | case 1: | |
4090 | min_scanlines = 32; | |
4091 | break; | |
4092 | case 2: | |
4093 | min_scanlines = 16; | |
4094 | break; | |
4095 | case 4: | |
4096 | min_scanlines = 8; | |
4097 | break; | |
4098 | case 8: | |
4099 | min_scanlines = 4; | |
4100 | break; | |
4101 | default: | |
4102 | WARN(1, "Unsupported pixel depth %u for rotation", | |
4103 | plane_bpp); | |
4104 | min_scanlines = 32; | |
4105 | } | |
4106 | } | |
4107 | ||
4108 | return DIV_ROUND_UP((4 * src_w * plane_bpp), 512) * min_scanlines/4 + 3; | |
4109 | } | |
4110 | ||
49845a7a ML |
4111 | static void |
4112 | skl_ddb_calc_min(const struct intel_crtc_state *cstate, int num_active, | |
4113 | uint16_t *minimum, uint16_t *y_minimum) | |
4114 | { | |
4115 | const struct drm_plane_state *pstate; | |
4116 | struct drm_plane *plane; | |
4117 | ||
4118 | drm_atomic_crtc_state_for_each_plane_state(plane, pstate, &cstate->base) { | |
d5cdfdf5 | 4119 | enum plane_id plane_id = to_intel_plane(plane)->id; |
49845a7a | 4120 | |
d5cdfdf5 | 4121 | if (plane_id == PLANE_CURSOR) |
49845a7a ML |
4122 | continue; |
4123 | ||
4124 | if (!pstate->visible) | |
4125 | continue; | |
4126 | ||
d5cdfdf5 VS |
4127 | minimum[plane_id] = skl_ddb_min_alloc(pstate, 0); |
4128 | y_minimum[plane_id] = skl_ddb_min_alloc(pstate, 1); | |
49845a7a ML |
4129 | } |
4130 | ||
4131 | minimum[PLANE_CURSOR] = skl_cursor_allocation(num_active); | |
4132 | } | |
4133 | ||
c107acfe | 4134 | static int |
024c9045 | 4135 | skl_allocate_pipe_ddb(struct intel_crtc_state *cstate, |
b9cec075 DL |
4136 | struct skl_ddb_allocation *ddb /* out */) |
4137 | { | |
c107acfe | 4138 | struct drm_atomic_state *state = cstate->base.state; |
024c9045 | 4139 | struct drm_crtc *crtc = cstate->base.crtc; |
b9cec075 DL |
4140 | struct drm_device *dev = crtc->dev; |
4141 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
4142 | enum pipe pipe = intel_crtc->pipe; | |
ce0ba283 | 4143 | struct skl_ddb_entry *alloc = &cstate->wm.skl.ddb; |
49845a7a | 4144 | uint16_t alloc_size, start; |
fefdd810 ML |
4145 | uint16_t minimum[I915_MAX_PLANES] = {}; |
4146 | uint16_t y_minimum[I915_MAX_PLANES] = {}; | |
b9cec075 | 4147 | unsigned int total_data_rate; |
d5cdfdf5 | 4148 | enum plane_id plane_id; |
c107acfe | 4149 | int num_active; |
1e6ee542 ML |
4150 | unsigned plane_data_rate[I915_MAX_PLANES] = {}; |
4151 | unsigned plane_y_data_rate[I915_MAX_PLANES] = {}; | |
5ba6faaf | 4152 | uint16_t total_min_blocks = 0; |
b9cec075 | 4153 | |
5a920b85 PZ |
4154 | /* Clear the partitioning for disabled planes. */ |
4155 | memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe])); | |
4156 | memset(ddb->y_plane[pipe], 0, sizeof(ddb->y_plane[pipe])); | |
4157 | ||
a6d3460e MR |
4158 | if (WARN_ON(!state)) |
4159 | return 0; | |
4160 | ||
c107acfe | 4161 | if (!cstate->base.active) { |
ce0ba283 | 4162 | alloc->start = alloc->end = 0; |
c107acfe MR |
4163 | return 0; |
4164 | } | |
4165 | ||
a6d3460e | 4166 | skl_ddb_get_pipe_allocation_limits(dev, cstate, alloc, &num_active); |
34bb56af | 4167 | alloc_size = skl_ddb_entry_size(alloc); |
336031ea | 4168 | if (alloc_size == 0) |
c107acfe | 4169 | return 0; |
b9cec075 | 4170 | |
49845a7a | 4171 | skl_ddb_calc_min(cstate, num_active, minimum, y_minimum); |
a6d3460e | 4172 | |
49845a7a ML |
4173 | /* |
4174 | * 1. Allocate the mininum required blocks for each active plane | |
4175 | * and allocate the cursor, it doesn't require extra allocation | |
4176 | * proportional to the data rate. | |
4177 | */ | |
80958155 | 4178 | |
d5cdfdf5 | 4179 | for_each_plane_id_on_crtc(intel_crtc, plane_id) { |
5ba6faaf KM |
4180 | total_min_blocks += minimum[plane_id]; |
4181 | total_min_blocks += y_minimum[plane_id]; | |
80958155 DL |
4182 | } |
4183 | ||
5ba6faaf KM |
4184 | if (total_min_blocks > alloc_size) { |
4185 | DRM_DEBUG_KMS("Requested display configuration exceeds system DDB limitations"); | |
4186 | DRM_DEBUG_KMS("minimum required %d/%d\n", total_min_blocks, | |
4187 | alloc_size); | |
4188 | return -EINVAL; | |
4189 | } | |
4190 | ||
9a30a261 RV |
4191 | alloc_size -= total_min_blocks; |
4192 | ddb->plane[pipe][PLANE_CURSOR].start = alloc->end - minimum[PLANE_CURSOR]; | |
49845a7a ML |
4193 | ddb->plane[pipe][PLANE_CURSOR].end = alloc->end; |
4194 | ||
b9cec075 | 4195 | /* |
80958155 DL |
4196 | * 2. Distribute the remaining space in proportion to the amount of |
4197 | * data each plane needs to fetch from memory. | |
b9cec075 DL |
4198 | * |
4199 | * FIXME: we may not allocate every single block here. | |
4200 | */ | |
1e6ee542 ML |
4201 | total_data_rate = skl_get_total_relative_data_rate(cstate, |
4202 | plane_data_rate, | |
4203 | plane_y_data_rate); | |
a1de91e5 | 4204 | if (total_data_rate == 0) |
c107acfe | 4205 | return 0; |
b9cec075 | 4206 | |
34bb56af | 4207 | start = alloc->start; |
d5cdfdf5 | 4208 | for_each_plane_id_on_crtc(intel_crtc, plane_id) { |
2cd601c6 | 4209 | unsigned int data_rate, y_data_rate; |
9a30a261 | 4210 | uint16_t plane_blocks, y_plane_blocks = 0; |
b9cec075 | 4211 | |
d5cdfdf5 | 4212 | if (plane_id == PLANE_CURSOR) |
49845a7a ML |
4213 | continue; |
4214 | ||
d5cdfdf5 | 4215 | data_rate = plane_data_rate[plane_id]; |
b9cec075 DL |
4216 | |
4217 | /* | |
2cd601c6 | 4218 | * allocation for (packed formats) or (uv-plane part of planar format): |
b9cec075 DL |
4219 | * promote the expression to 64 bits to avoid overflowing, the |
4220 | * result is < available as data_rate / total_data_rate < 1 | |
4221 | */ | |
9a30a261 RV |
4222 | plane_blocks = minimum[plane_id]; |
4223 | plane_blocks += div_u64((uint64_t)alloc_size * data_rate, | |
4224 | total_data_rate); | |
b9cec075 | 4225 | |
c107acfe MR |
4226 | /* Leave disabled planes at (0,0) */ |
4227 | if (data_rate) { | |
d5cdfdf5 VS |
4228 | ddb->plane[pipe][plane_id].start = start; |
4229 | ddb->plane[pipe][plane_id].end = start + plane_blocks; | |
c107acfe | 4230 | } |
b9cec075 | 4231 | |
9a30a261 RV |
4232 | start += plane_blocks; |
4233 | ||
2cd601c6 CK |
4234 | /* |
4235 | * allocation for y_plane part of planar format: | |
4236 | */ | |
d5cdfdf5 | 4237 | y_data_rate = plane_y_data_rate[plane_id]; |
a1de91e5 | 4238 | |
9a30a261 RV |
4239 | y_plane_blocks = y_minimum[plane_id]; |
4240 | y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate, | |
4241 | total_data_rate); | |
4242 | ||
c107acfe | 4243 | if (y_data_rate) { |
d5cdfdf5 VS |
4244 | ddb->y_plane[pipe][plane_id].start = start; |
4245 | ddb->y_plane[pipe][plane_id].end = start + y_plane_blocks; | |
c107acfe | 4246 | } |
9a30a261 RV |
4247 | |
4248 | start += y_plane_blocks; | |
b9cec075 DL |
4249 | } |
4250 | ||
c107acfe | 4251 | return 0; |
b9cec075 DL |
4252 | } |
4253 | ||
2d41c0b5 PB |
4254 | /* |
4255 | * The max latency should be 257 (max the punit can code is 255 and we add 2us | |
ac484963 | 4256 | * for the read latency) and cpp should always be <= 8, so that |
2d41c0b5 PB |
4257 | * should allow pixel_rate up to ~2 GHz which seems sufficient since max |
4258 | * 2xcdclk is 1350 MHz and the pixel rate should never exceed that. | |
4259 | */ | |
b95320bd MK |
4260 | static uint_fixed_16_16_t skl_wm_method1(uint32_t pixel_rate, uint8_t cpp, |
4261 | uint32_t latency) | |
2d41c0b5 | 4262 | { |
b95320bd MK |
4263 | uint32_t wm_intermediate_val; |
4264 | uint_fixed_16_16_t ret; | |
2d41c0b5 PB |
4265 | |
4266 | if (latency == 0) | |
b95320bd | 4267 | return FP_16_16_MAX; |
2d41c0b5 | 4268 | |
b95320bd | 4269 | wm_intermediate_val = latency * pixel_rate * cpp; |
afbc95cd | 4270 | ret = fixed_16_16_div_u64(wm_intermediate_val, 1000 * 512); |
2d41c0b5 PB |
4271 | return ret; |
4272 | } | |
4273 | ||
b95320bd MK |
4274 | static uint_fixed_16_16_t skl_wm_method2(uint32_t pixel_rate, |
4275 | uint32_t pipe_htotal, | |
4276 | uint32_t latency, | |
4277 | uint_fixed_16_16_t plane_blocks_per_line) | |
2d41c0b5 | 4278 | { |
d4c2aa60 | 4279 | uint32_t wm_intermediate_val; |
b95320bd | 4280 | uint_fixed_16_16_t ret; |
2d41c0b5 PB |
4281 | |
4282 | if (latency == 0) | |
b95320bd | 4283 | return FP_16_16_MAX; |
2d41c0b5 | 4284 | |
2d41c0b5 | 4285 | wm_intermediate_val = latency * pixel_rate; |
b95320bd MK |
4286 | wm_intermediate_val = DIV_ROUND_UP(wm_intermediate_val, |
4287 | pipe_htotal * 1000); | |
4288 | ret = mul_u32_fixed_16_16(wm_intermediate_val, plane_blocks_per_line); | |
2d41c0b5 PB |
4289 | return ret; |
4290 | } | |
4291 | ||
d555cb58 KM |
4292 | static uint_fixed_16_16_t |
4293 | intel_get_linetime_us(struct intel_crtc_state *cstate) | |
4294 | { | |
4295 | uint32_t pixel_rate; | |
4296 | uint32_t crtc_htotal; | |
4297 | uint_fixed_16_16_t linetime_us; | |
4298 | ||
4299 | if (!cstate->base.active) | |
4300 | return u32_to_fixed_16_16(0); | |
4301 | ||
4302 | pixel_rate = cstate->pixel_rate; | |
4303 | ||
4304 | if (WARN_ON(pixel_rate == 0)) | |
4305 | return u32_to_fixed_16_16(0); | |
4306 | ||
4307 | crtc_htotal = cstate->base.adjusted_mode.crtc_htotal; | |
4308 | linetime_us = fixed_16_16_div_u64(crtc_htotal * 1000, pixel_rate); | |
4309 | ||
4310 | return linetime_us; | |
4311 | } | |
4312 | ||
eb2fdcdf KM |
4313 | static uint32_t |
4314 | skl_adjusted_plane_pixel_rate(const struct intel_crtc_state *cstate, | |
4315 | const struct intel_plane_state *pstate) | |
9c2f7a9d KM |
4316 | { |
4317 | uint64_t adjusted_pixel_rate; | |
7084b50b | 4318 | uint_fixed_16_16_t downscale_amount; |
9c2f7a9d KM |
4319 | |
4320 | /* Shouldn't reach here on disabled planes... */ | |
93aa2a1c | 4321 | if (WARN_ON(!intel_wm_plane_visible(cstate, pstate))) |
9c2f7a9d KM |
4322 | return 0; |
4323 | ||
4324 | /* | |
4325 | * Adjusted plane pixel rate is just the pipe's adjusted pixel rate | |
4326 | * with additional adjustments for plane-specific scaling. | |
4327 | */ | |
a7d1b3f4 | 4328 | adjusted_pixel_rate = cstate->pixel_rate; |
93aa2a1c | 4329 | downscale_amount = skl_plane_downscale_amount(cstate, pstate); |
9c2f7a9d | 4330 | |
7084b50b KM |
4331 | return mul_round_up_u32_fixed16(adjusted_pixel_rate, |
4332 | downscale_amount); | |
9c2f7a9d KM |
4333 | } |
4334 | ||
55994c2c MR |
4335 | static int skl_compute_plane_wm(const struct drm_i915_private *dev_priv, |
4336 | struct intel_crtc_state *cstate, | |
eb2fdcdf | 4337 | const struct intel_plane_state *intel_pstate, |
9a30a261 | 4338 | uint16_t ddb_allocation, |
55994c2c MR |
4339 | int level, |
4340 | uint16_t *out_blocks, /* out */ | |
9a30a261 RV |
4341 | uint8_t *out_lines, /* out */ |
4342 | bool *enabled /* out */) | |
2d41c0b5 | 4343 | { |
93aa2a1c | 4344 | struct intel_plane *plane = to_intel_plane(intel_pstate->base.plane); |
eb2fdcdf KM |
4345 | const struct drm_plane_state *pstate = &intel_pstate->base; |
4346 | const struct drm_framebuffer *fb = pstate->fb; | |
d4c2aa60 | 4347 | uint32_t latency = dev_priv->wm.skl_latency[level]; |
b95320bd MK |
4348 | uint_fixed_16_16_t method1, method2; |
4349 | uint_fixed_16_16_t plane_blocks_per_line; | |
4350 | uint_fixed_16_16_t selected_result; | |
4351 | uint32_t interm_pbpl; | |
4352 | uint32_t plane_bytes_per_line; | |
d4c2aa60 | 4353 | uint32_t res_blocks, res_lines; |
ac484963 | 4354 | uint8_t cpp; |
a280f7dd | 4355 | uint32_t width = 0, height = 0; |
9c2f7a9d | 4356 | uint32_t plane_pixel_rate; |
b95320bd MK |
4357 | uint_fixed_16_16_t y_tile_minimum; |
4358 | uint32_t y_min_scanlines; | |
ee3d532f PZ |
4359 | struct intel_atomic_state *state = |
4360 | to_intel_atomic_state(cstate->base.state); | |
4361 | bool apply_memory_bw_wa = skl_needs_memory_bw_wa(state); | |
ef8a4fb4 | 4362 | bool y_tiled, x_tiled; |
2d41c0b5 | 4363 | |
93aa2a1c | 4364 | if (latency == 0 || |
9a30a261 RV |
4365 | !intel_wm_plane_visible(cstate, intel_pstate)) { |
4366 | *enabled = false; | |
55994c2c | 4367 | return 0; |
9a30a261 | 4368 | } |
2d41c0b5 | 4369 | |
ef8a4fb4 MK |
4370 | y_tiled = fb->modifier == I915_FORMAT_MOD_Y_TILED || |
4371 | fb->modifier == I915_FORMAT_MOD_Yf_TILED; | |
4372 | x_tiled = fb->modifier == I915_FORMAT_MOD_X_TILED; | |
4373 | ||
82525c17 RV |
4374 | /* Display WA #1141: kbl,cfl */ |
4375 | if ((IS_KABYLAKE(dev_priv) || IS_COFFEELAKE(dev_priv)) && | |
4376 | dev_priv->ipc_enabled) | |
4b7b2331 MK |
4377 | latency += 4; |
4378 | ||
ef8a4fb4 | 4379 | if (apply_memory_bw_wa && x_tiled) |
ee3d532f PZ |
4380 | latency += 15; |
4381 | ||
93aa2a1c VS |
4382 | if (plane->id == PLANE_CURSOR) { |
4383 | width = intel_pstate->base.crtc_w; | |
4384 | height = intel_pstate->base.crtc_h; | |
4385 | } else { | |
fce5adf5 VS |
4386 | /* |
4387 | * Src coordinates are already rotated by 270 degrees for | |
4388 | * the 90/270 degree plane rotation cases (to match the | |
4389 | * GTT mapping), hence no need to account for rotation here. | |
4390 | */ | |
93aa2a1c VS |
4391 | width = drm_rect_width(&intel_pstate->base.src) >> 16; |
4392 | height = drm_rect_height(&intel_pstate->base.src) >> 16; | |
4393 | } | |
a280f7dd | 4394 | |
353c8598 | 4395 | cpp = fb->format->cpp[0]; |
9c2f7a9d KM |
4396 | plane_pixel_rate = skl_adjusted_plane_pixel_rate(cstate, intel_pstate); |
4397 | ||
61d0a04d | 4398 | if (drm_rotation_90_or_270(pstate->rotation)) { |
438b74a5 | 4399 | int cpp = (fb->format->format == DRM_FORMAT_NV12) ? |
353c8598 VS |
4400 | fb->format->cpp[1] : |
4401 | fb->format->cpp[0]; | |
1186fa85 PZ |
4402 | |
4403 | switch (cpp) { | |
4404 | case 1: | |
4405 | y_min_scanlines = 16; | |
4406 | break; | |
4407 | case 2: | |
4408 | y_min_scanlines = 8; | |
4409 | break; | |
1186fa85 PZ |
4410 | case 4: |
4411 | y_min_scanlines = 4; | |
4412 | break; | |
86a462bc PZ |
4413 | default: |
4414 | MISSING_CASE(cpp); | |
4415 | return -EINVAL; | |
1186fa85 PZ |
4416 | } |
4417 | } else { | |
4418 | y_min_scanlines = 4; | |
4419 | } | |
4420 | ||
2ef32dee PZ |
4421 | if (apply_memory_bw_wa) |
4422 | y_min_scanlines *= 2; | |
4423 | ||
7a1a8aed | 4424 | plane_bytes_per_line = width * cpp; |
ef8a4fb4 | 4425 | if (y_tiled) { |
b95320bd MK |
4426 | interm_pbpl = DIV_ROUND_UP(plane_bytes_per_line * |
4427 | y_min_scanlines, 512); | |
afbc95cd KM |
4428 | plane_blocks_per_line = fixed_16_16_div(interm_pbpl, |
4429 | y_min_scanlines); | |
ef8a4fb4 | 4430 | } else if (x_tiled) { |
b95320bd MK |
4431 | interm_pbpl = DIV_ROUND_UP(plane_bytes_per_line, 512); |
4432 | plane_blocks_per_line = u32_to_fixed_16_16(interm_pbpl); | |
ef8a4fb4 | 4433 | } else { |
b95320bd MK |
4434 | interm_pbpl = DIV_ROUND_UP(plane_bytes_per_line, 512) + 1; |
4435 | plane_blocks_per_line = u32_to_fixed_16_16(interm_pbpl); | |
7a1a8aed PZ |
4436 | } |
4437 | ||
9c2f7a9d KM |
4438 | method1 = skl_wm_method1(plane_pixel_rate, cpp, latency); |
4439 | method2 = skl_wm_method2(plane_pixel_rate, | |
024c9045 | 4440 | cstate->base.adjusted_mode.crtc_htotal, |
1186fa85 | 4441 | latency, |
7a1a8aed | 4442 | plane_blocks_per_line); |
2d41c0b5 | 4443 | |
b95320bd MK |
4444 | y_tile_minimum = mul_u32_fixed_16_16(y_min_scanlines, |
4445 | plane_blocks_per_line); | |
75676ed4 | 4446 | |
ef8a4fb4 | 4447 | if (y_tiled) { |
b95320bd | 4448 | selected_result = max_fixed_16_16(method2, y_tile_minimum); |
0fda6568 | 4449 | } else { |
d555cb58 KM |
4450 | uint32_t linetime_us; |
4451 | ||
4452 | linetime_us = fixed_16_16_to_u32_round_up( | |
4453 | intel_get_linetime_us(cstate)); | |
f1db3eaf PZ |
4454 | if ((cpp * cstate->base.adjusted_mode.crtc_htotal / 512 < 1) && |
4455 | (plane_bytes_per_line / 512 < 1)) | |
4456 | selected_result = method2; | |
e9ba4032 ML |
4457 | else if (ddb_allocation >= |
4458 | fixed_16_16_to_u32_round_up(plane_blocks_per_line)) | |
9a30a261 | 4459 | selected_result = min_fixed_16_16(method1, method2); |
d555cb58 KM |
4460 | else if (latency >= linetime_us) |
4461 | selected_result = min_fixed_16_16(method1, method2); | |
0fda6568 TU |
4462 | else |
4463 | selected_result = method1; | |
4464 | } | |
2d41c0b5 | 4465 | |
b95320bd | 4466 | res_blocks = fixed_16_16_to_u32_round_up(selected_result) + 1; |
d273ecce KM |
4467 | res_lines = div_round_up_fixed16(selected_result, |
4468 | plane_blocks_per_line); | |
e6d66171 | 4469 | |
0fda6568 | 4470 | if (level >= 1 && level <= 7) { |
ef8a4fb4 | 4471 | if (y_tiled) { |
b95320bd | 4472 | res_blocks += fixed_16_16_to_u32_round_up(y_tile_minimum); |
1186fa85 | 4473 | res_lines += y_min_scanlines; |
75676ed4 | 4474 | } else { |
0fda6568 | 4475 | res_blocks++; |
75676ed4 | 4476 | } |
0fda6568 | 4477 | } |
e6d66171 | 4478 | |
9a30a261 RV |
4479 | if (res_blocks >= ddb_allocation || res_lines > 31) { |
4480 | *enabled = false; | |
d5cdfdf5 | 4481 | |
9a30a261 RV |
4482 | /* |
4483 | * If there are no valid level 0 watermarks, then we can't | |
4484 | * support this display configuration. | |
4485 | */ | |
4486 | if (level) { | |
4487 | return 0; | |
4488 | } else { | |
4489 | struct drm_plane *plane = pstate->plane; | |
4490 | ||
4491 | DRM_DEBUG_KMS("Requested display configuration exceeds system watermark limitations\n"); | |
4492 | DRM_DEBUG_KMS("[PLANE:%d:%s] blocks required = %u/%u, lines required = %u/31\n", | |
4493 | plane->base.id, plane->name, | |
4494 | res_blocks, ddb_allocation, res_lines); | |
4495 | return -EINVAL; | |
4496 | } | |
55994c2c | 4497 | } |
e6d66171 DL |
4498 | |
4499 | *out_blocks = res_blocks; | |
4500 | *out_lines = res_lines; | |
9a30a261 | 4501 | *enabled = true; |
2d41c0b5 | 4502 | |
55994c2c | 4503 | return 0; |
2d41c0b5 PB |
4504 | } |
4505 | ||
f4a96752 | 4506 | static int |
d2f5e36d | 4507 | skl_compute_wm_levels(const struct drm_i915_private *dev_priv, |
9a30a261 | 4508 | struct skl_ddb_allocation *ddb, |
d2f5e36d KM |
4509 | struct intel_crtc_state *cstate, |
4510 | const struct intel_plane_state *intel_pstate, | |
4511 | struct skl_plane_wm *wm) | |
2d41c0b5 | 4512 | { |
9a30a261 RV |
4513 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); |
4514 | struct drm_plane *plane = intel_pstate->base.plane; | |
4515 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
4516 | uint16_t ddb_blocks; | |
4517 | enum pipe pipe = intel_crtc->pipe; | |
d2f5e36d | 4518 | int level, max_level = ilk_wm_max_level(dev_priv); |
55994c2c | 4519 | int ret; |
a62163e9 | 4520 | |
7b75119c KM |
4521 | if (WARN_ON(!intel_pstate->base.fb)) |
4522 | return -EINVAL; | |
f4a96752 | 4523 | |
9a30a261 RV |
4524 | ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][intel_plane->id]); |
4525 | ||
d2f5e36d KM |
4526 | for (level = 0; level <= max_level; level++) { |
4527 | struct skl_wm_level *result = &wm->wm[level]; | |
4528 | ||
4529 | ret = skl_compute_plane_wm(dev_priv, | |
4530 | cstate, | |
4531 | intel_pstate, | |
9a30a261 | 4532 | ddb_blocks, |
d2f5e36d KM |
4533 | level, |
4534 | &result->plane_res_b, | |
9a30a261 RV |
4535 | &result->plane_res_l, |
4536 | &result->plane_en); | |
d2f5e36d KM |
4537 | if (ret) |
4538 | return ret; | |
4539 | } | |
f4a96752 MR |
4540 | |
4541 | return 0; | |
2d41c0b5 PB |
4542 | } |
4543 | ||
407b50f3 | 4544 | static uint32_t |
024c9045 | 4545 | skl_compute_linetime_wm(struct intel_crtc_state *cstate) |
407b50f3 | 4546 | { |
a3a8986c MK |
4547 | struct drm_atomic_state *state = cstate->base.state; |
4548 | struct drm_i915_private *dev_priv = to_i915(state->dev); | |
d555cb58 | 4549 | uint_fixed_16_16_t linetime_us; |
a3a8986c | 4550 | uint32_t linetime_wm; |
30d1b5fe | 4551 | |
d555cb58 | 4552 | linetime_us = intel_get_linetime_us(cstate); |
407b50f3 | 4553 | |
d555cb58 | 4554 | if (is_fixed16_zero(linetime_us)) |
661abfc0 | 4555 | return 0; |
407b50f3 | 4556 | |
d555cb58 KM |
4557 | linetime_wm = fixed_16_16_to_u32_round_up(mul_u32_fixed_16_16(8, |
4558 | linetime_us)); | |
a3a8986c MK |
4559 | |
4560 | /* Display WA #1135: bxt. */ | |
4561 | if (IS_BROXTON(dev_priv) && dev_priv->ipc_enabled) | |
4562 | linetime_wm = DIV_ROUND_UP(linetime_wm, 2); | |
4563 | ||
4564 | return linetime_wm; | |
407b50f3 DL |
4565 | } |
4566 | ||
024c9045 | 4567 | static void skl_compute_transition_wm(struct intel_crtc_state *cstate, |
9414f563 | 4568 | struct skl_wm_level *trans_wm /* out */) |
407b50f3 | 4569 | { |
024c9045 | 4570 | if (!cstate->base.active) |
407b50f3 | 4571 | return; |
9414f563 DL |
4572 | |
4573 | /* Until we know more, just disable transition WMs */ | |
a62163e9 | 4574 | trans_wm->plane_en = false; |
407b50f3 DL |
4575 | } |
4576 | ||
55994c2c MR |
4577 | static int skl_build_pipe_wm(struct intel_crtc_state *cstate, |
4578 | struct skl_ddb_allocation *ddb, | |
4579 | struct skl_pipe_wm *pipe_wm) | |
2d41c0b5 | 4580 | { |
024c9045 | 4581 | struct drm_device *dev = cstate->base.crtc->dev; |
eb2fdcdf | 4582 | struct drm_crtc_state *crtc_state = &cstate->base; |
fac5e23e | 4583 | const struct drm_i915_private *dev_priv = to_i915(dev); |
eb2fdcdf KM |
4584 | struct drm_plane *plane; |
4585 | const struct drm_plane_state *pstate; | |
a62163e9 | 4586 | struct skl_plane_wm *wm; |
55994c2c | 4587 | int ret; |
2d41c0b5 | 4588 | |
a62163e9 L |
4589 | /* |
4590 | * We'll only calculate watermarks for planes that are actually | |
4591 | * enabled, so make sure all other planes are set as disabled. | |
4592 | */ | |
4593 | memset(pipe_wm->planes, 0, sizeof(pipe_wm->planes)); | |
4594 | ||
eb2fdcdf KM |
4595 | drm_atomic_crtc_state_for_each_plane_state(plane, pstate, crtc_state) { |
4596 | const struct intel_plane_state *intel_pstate = | |
4597 | to_intel_plane_state(pstate); | |
4598 | enum plane_id plane_id = to_intel_plane(plane)->id; | |
4599 | ||
4600 | wm = &pipe_wm->planes[plane_id]; | |
a62163e9 | 4601 | |
9a30a261 RV |
4602 | ret = skl_compute_wm_levels(dev_priv, ddb, cstate, |
4603 | intel_pstate, wm); | |
d2f5e36d KM |
4604 | if (ret) |
4605 | return ret; | |
a62163e9 | 4606 | skl_compute_transition_wm(cstate, &wm->trans_wm); |
2d41c0b5 | 4607 | } |
024c9045 | 4608 | pipe_wm->linetime = skl_compute_linetime_wm(cstate); |
2d41c0b5 | 4609 | |
55994c2c | 4610 | return 0; |
2d41c0b5 PB |
4611 | } |
4612 | ||
f0f59a00 VS |
4613 | static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, |
4614 | i915_reg_t reg, | |
16160e3d DL |
4615 | const struct skl_ddb_entry *entry) |
4616 | { | |
4617 | if (entry->end) | |
4618 | I915_WRITE(reg, (entry->end - 1) << 16 | entry->start); | |
4619 | else | |
4620 | I915_WRITE(reg, 0); | |
4621 | } | |
4622 | ||
d8c0fafc | 4623 | static void skl_write_wm_level(struct drm_i915_private *dev_priv, |
4624 | i915_reg_t reg, | |
4625 | const struct skl_wm_level *level) | |
4626 | { | |
4627 | uint32_t val = 0; | |
4628 | ||
4629 | if (level->plane_en) { | |
4630 | val |= PLANE_WM_EN; | |
4631 | val |= level->plane_res_b; | |
4632 | val |= level->plane_res_l << PLANE_WM_LINES_SHIFT; | |
4633 | } | |
4634 | ||
4635 | I915_WRITE(reg, val); | |
4636 | } | |
4637 | ||
d9348dec VS |
4638 | static void skl_write_plane_wm(struct intel_crtc *intel_crtc, |
4639 | const struct skl_plane_wm *wm, | |
4640 | const struct skl_ddb_allocation *ddb, | |
d5cdfdf5 | 4641 | enum plane_id plane_id) |
62e0fb88 L |
4642 | { |
4643 | struct drm_crtc *crtc = &intel_crtc->base; | |
4644 | struct drm_device *dev = crtc->dev; | |
4645 | struct drm_i915_private *dev_priv = to_i915(dev); | |
5db94019 | 4646 | int level, max_level = ilk_wm_max_level(dev_priv); |
62e0fb88 L |
4647 | enum pipe pipe = intel_crtc->pipe; |
4648 | ||
4649 | for (level = 0; level <= max_level; level++) { | |
d5cdfdf5 | 4650 | skl_write_wm_level(dev_priv, PLANE_WM(pipe, plane_id, level), |
d8c0fafc | 4651 | &wm->wm[level]); |
62e0fb88 | 4652 | } |
d5cdfdf5 | 4653 | skl_write_wm_level(dev_priv, PLANE_WM_TRANS(pipe, plane_id), |
d8c0fafc | 4654 | &wm->trans_wm); |
27082493 | 4655 | |
d5cdfdf5 VS |
4656 | skl_ddb_entry_write(dev_priv, PLANE_BUF_CFG(pipe, plane_id), |
4657 | &ddb->plane[pipe][plane_id]); | |
4658 | skl_ddb_entry_write(dev_priv, PLANE_NV12_BUF_CFG(pipe, plane_id), | |
4659 | &ddb->y_plane[pipe][plane_id]); | |
62e0fb88 L |
4660 | } |
4661 | ||
d9348dec VS |
4662 | static void skl_write_cursor_wm(struct intel_crtc *intel_crtc, |
4663 | const struct skl_plane_wm *wm, | |
4664 | const struct skl_ddb_allocation *ddb) | |
62e0fb88 L |
4665 | { |
4666 | struct drm_crtc *crtc = &intel_crtc->base; | |
4667 | struct drm_device *dev = crtc->dev; | |
4668 | struct drm_i915_private *dev_priv = to_i915(dev); | |
5db94019 | 4669 | int level, max_level = ilk_wm_max_level(dev_priv); |
62e0fb88 L |
4670 | enum pipe pipe = intel_crtc->pipe; |
4671 | ||
4672 | for (level = 0; level <= max_level; level++) { | |
d8c0fafc | 4673 | skl_write_wm_level(dev_priv, CUR_WM(pipe, level), |
4674 | &wm->wm[level]); | |
62e0fb88 | 4675 | } |
d8c0fafc | 4676 | skl_write_wm_level(dev_priv, CUR_WM_TRANS(pipe), &wm->trans_wm); |
5d374d96 | 4677 | |
27082493 | 4678 | skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe), |
d8c0fafc | 4679 | &ddb->plane[pipe][PLANE_CURSOR]); |
2d41c0b5 PB |
4680 | } |
4681 | ||
45ece230 | 4682 | bool skl_wm_level_equals(const struct skl_wm_level *l1, |
4683 | const struct skl_wm_level *l2) | |
4684 | { | |
4685 | if (l1->plane_en != l2->plane_en) | |
4686 | return false; | |
4687 | ||
4688 | /* If both planes aren't enabled, the rest shouldn't matter */ | |
4689 | if (!l1->plane_en) | |
4690 | return true; | |
4691 | ||
4692 | return (l1->plane_res_l == l2->plane_res_l && | |
4693 | l1->plane_res_b == l2->plane_res_b); | |
4694 | } | |
4695 | ||
27082493 L |
4696 | static inline bool skl_ddb_entries_overlap(const struct skl_ddb_entry *a, |
4697 | const struct skl_ddb_entry *b) | |
0e8fb7ba | 4698 | { |
27082493 | 4699 | return a->start < b->end && b->start < a->end; |
0e8fb7ba DL |
4700 | } |
4701 | ||
5eff503b ML |
4702 | bool skl_ddb_allocation_overlaps(const struct skl_ddb_entry **entries, |
4703 | const struct skl_ddb_entry *ddb, | |
4704 | int ignore) | |
0e8fb7ba | 4705 | { |
ce0ba283 | 4706 | int i; |
0e8fb7ba | 4707 | |
5eff503b ML |
4708 | for (i = 0; i < I915_MAX_PIPES; i++) |
4709 | if (i != ignore && entries[i] && | |
4710 | skl_ddb_entries_overlap(ddb, entries[i])) | |
27082493 | 4711 | return true; |
0e8fb7ba | 4712 | |
27082493 | 4713 | return false; |
0e8fb7ba DL |
4714 | } |
4715 | ||
55994c2c | 4716 | static int skl_update_pipe_wm(struct drm_crtc_state *cstate, |
03af79e0 | 4717 | const struct skl_pipe_wm *old_pipe_wm, |
55994c2c | 4718 | struct skl_pipe_wm *pipe_wm, /* out */ |
03af79e0 | 4719 | struct skl_ddb_allocation *ddb, /* out */ |
55994c2c | 4720 | bool *changed /* out */) |
2d41c0b5 | 4721 | { |
f4a96752 | 4722 | struct intel_crtc_state *intel_cstate = to_intel_crtc_state(cstate); |
55994c2c | 4723 | int ret; |
2d41c0b5 | 4724 | |
55994c2c MR |
4725 | ret = skl_build_pipe_wm(intel_cstate, ddb, pipe_wm); |
4726 | if (ret) | |
4727 | return ret; | |
2d41c0b5 | 4728 | |
03af79e0 | 4729 | if (!memcmp(old_pipe_wm, pipe_wm, sizeof(*pipe_wm))) |
55994c2c MR |
4730 | *changed = false; |
4731 | else | |
4732 | *changed = true; | |
2d41c0b5 | 4733 | |
55994c2c | 4734 | return 0; |
2d41c0b5 PB |
4735 | } |
4736 | ||
9b613022 MR |
4737 | static uint32_t |
4738 | pipes_modified(struct drm_atomic_state *state) | |
4739 | { | |
4740 | struct drm_crtc *crtc; | |
4741 | struct drm_crtc_state *cstate; | |
4742 | uint32_t i, ret = 0; | |
4743 | ||
6ebdb5a0 | 4744 | for_each_new_crtc_in_state(state, crtc, cstate, i) |
9b613022 MR |
4745 | ret |= drm_crtc_mask(crtc); |
4746 | ||
4747 | return ret; | |
4748 | } | |
4749 | ||
bb7791bd | 4750 | static int |
9a30a261 RV |
4751 | skl_ddb_add_affected_planes(struct intel_crtc_state *cstate) |
4752 | { | |
4753 | struct drm_atomic_state *state = cstate->base.state; | |
4754 | struct drm_device *dev = state->dev; | |
4755 | struct drm_crtc *crtc = cstate->base.crtc; | |
4756 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
4757 | struct drm_i915_private *dev_priv = to_i915(dev); | |
4758 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
4759 | struct skl_ddb_allocation *new_ddb = &intel_state->wm_results.ddb; | |
4760 | struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb; | |
4761 | struct drm_plane_state *plane_state; | |
4762 | struct drm_plane *plane; | |
4763 | enum pipe pipe = intel_crtc->pipe; | |
4764 | ||
4765 | WARN_ON(!drm_atomic_get_existing_crtc_state(state, crtc)); | |
4766 | ||
4767 | drm_for_each_plane_mask(plane, dev, cstate->base.plane_mask) { | |
4768 | enum plane_id plane_id = to_intel_plane(plane)->id; | |
4769 | ||
4770 | if (skl_ddb_entry_equal(&cur_ddb->plane[pipe][plane_id], | |
4771 | &new_ddb->plane[pipe][plane_id]) && | |
4772 | skl_ddb_entry_equal(&cur_ddb->y_plane[pipe][plane_id], | |
4773 | &new_ddb->y_plane[pipe][plane_id])) | |
4774 | continue; | |
4775 | ||
4776 | plane_state = drm_atomic_get_plane_state(state, plane); | |
4777 | if (IS_ERR(plane_state)) | |
4778 | return PTR_ERR(plane_state); | |
4779 | } | |
4780 | ||
4781 | return 0; | |
4782 | } | |
4783 | ||
4784 | static int | |
4785 | skl_compute_ddb(struct drm_atomic_state *state) | |
98d39494 MR |
4786 | { |
4787 | struct drm_device *dev = state->dev; | |
4788 | struct drm_i915_private *dev_priv = to_i915(dev); | |
4789 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
4790 | struct intel_crtc *intel_crtc; | |
734fa01f | 4791 | struct skl_ddb_allocation *ddb = &intel_state->wm_results.ddb; |
9b613022 | 4792 | uint32_t realloc_pipes = pipes_modified(state); |
98d39494 MR |
4793 | int ret; |
4794 | ||
4795 | /* | |
4796 | * If this is our first atomic update following hardware readout, | |
4797 | * we can't trust the DDB that the BIOS programmed for us. Let's | |
4798 | * pretend that all pipes switched active status so that we'll | |
4799 | * ensure a full DDB recompute. | |
4800 | */ | |
1b54a880 MR |
4801 | if (dev_priv->wm.distrust_bios_wm) { |
4802 | ret = drm_modeset_lock(&dev->mode_config.connection_mutex, | |
4803 | state->acquire_ctx); | |
4804 | if (ret) | |
4805 | return ret; | |
4806 | ||
98d39494 MR |
4807 | intel_state->active_pipe_changes = ~0; |
4808 | ||
1b54a880 MR |
4809 | /* |
4810 | * We usually only initialize intel_state->active_crtcs if we | |
4811 | * we're doing a modeset; make sure this field is always | |
4812 | * initialized during the sanitization process that happens | |
4813 | * on the first commit too. | |
4814 | */ | |
4815 | if (!intel_state->modeset) | |
4816 | intel_state->active_crtcs = dev_priv->active_crtcs; | |
4817 | } | |
4818 | ||
98d39494 MR |
4819 | /* |
4820 | * If the modeset changes which CRTC's are active, we need to | |
4821 | * recompute the DDB allocation for *all* active pipes, even | |
4822 | * those that weren't otherwise being modified in any way by this | |
4823 | * atomic commit. Due to the shrinking of the per-pipe allocations | |
4824 | * when new active CRTC's are added, it's possible for a pipe that | |
4825 | * we were already using and aren't changing at all here to suddenly | |
4826 | * become invalid if its DDB needs exceeds its new allocation. | |
4827 | * | |
4828 | * Note that if we wind up doing a full DDB recompute, we can't let | |
4829 | * any other display updates race with this transaction, so we need | |
4830 | * to grab the lock on *all* CRTC's. | |
4831 | */ | |
734fa01f | 4832 | if (intel_state->active_pipe_changes) { |
98d39494 | 4833 | realloc_pipes = ~0; |
734fa01f MR |
4834 | intel_state->wm_results.dirty_pipes = ~0; |
4835 | } | |
98d39494 | 4836 | |
5a920b85 PZ |
4837 | /* |
4838 | * We're not recomputing for the pipes not included in the commit, so | |
4839 | * make sure we start with the current state. | |
4840 | */ | |
4841 | memcpy(ddb, &dev_priv->wm.skl_hw.ddb, sizeof(*ddb)); | |
4842 | ||
98d39494 MR |
4843 | for_each_intel_crtc_mask(dev, intel_crtc, realloc_pipes) { |
4844 | struct intel_crtc_state *cstate; | |
4845 | ||
4846 | cstate = intel_atomic_get_crtc_state(state, intel_crtc); | |
4847 | if (IS_ERR(cstate)) | |
4848 | return PTR_ERR(cstate); | |
9a30a261 RV |
4849 | |
4850 | ret = skl_allocate_pipe_ddb(cstate, ddb); | |
4851 | if (ret) | |
4852 | return ret; | |
4853 | ||
4854 | ret = skl_ddb_add_affected_planes(cstate); | |
4855 | if (ret) | |
4856 | return ret; | |
98d39494 MR |
4857 | } |
4858 | ||
4859 | return 0; | |
4860 | } | |
4861 | ||
2722efb9 MR |
4862 | static void |
4863 | skl_copy_wm_for_pipe(struct skl_wm_values *dst, | |
4864 | struct skl_wm_values *src, | |
4865 | enum pipe pipe) | |
4866 | { | |
2722efb9 MR |
4867 | memcpy(dst->ddb.y_plane[pipe], src->ddb.y_plane[pipe], |
4868 | sizeof(dst->ddb.y_plane[pipe])); | |
4869 | memcpy(dst->ddb.plane[pipe], src->ddb.plane[pipe], | |
4870 | sizeof(dst->ddb.plane[pipe])); | |
4871 | } | |
4872 | ||
413fc530 | 4873 | static void |
4874 | skl_print_wm_changes(const struct drm_atomic_state *state) | |
4875 | { | |
4876 | const struct drm_device *dev = state->dev; | |
4877 | const struct drm_i915_private *dev_priv = to_i915(dev); | |
4878 | const struct intel_atomic_state *intel_state = | |
4879 | to_intel_atomic_state(state); | |
4880 | const struct drm_crtc *crtc; | |
4881 | const struct drm_crtc_state *cstate; | |
413fc530 | 4882 | const struct intel_plane *intel_plane; |
413fc530 | 4883 | const struct skl_ddb_allocation *old_ddb = &dev_priv->wm.skl_hw.ddb; |
4884 | const struct skl_ddb_allocation *new_ddb = &intel_state->wm_results.ddb; | |
7570498e | 4885 | int i; |
413fc530 | 4886 | |
6ebdb5a0 | 4887 | for_each_new_crtc_in_state(state, crtc, cstate, i) { |
7570498e ML |
4888 | const struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
4889 | enum pipe pipe = intel_crtc->pipe; | |
413fc530 | 4890 | |
7570498e | 4891 | for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) { |
d5cdfdf5 | 4892 | enum plane_id plane_id = intel_plane->id; |
413fc530 | 4893 | const struct skl_ddb_entry *old, *new; |
4894 | ||
d5cdfdf5 VS |
4895 | old = &old_ddb->plane[pipe][plane_id]; |
4896 | new = &new_ddb->plane[pipe][plane_id]; | |
413fc530 | 4897 | |
413fc530 | 4898 | if (skl_ddb_entry_equal(old, new)) |
4899 | continue; | |
4900 | ||
7570498e ML |
4901 | DRM_DEBUG_ATOMIC("[PLANE:%d:%s] ddb (%d - %d) -> (%d - %d)\n", |
4902 | intel_plane->base.base.id, | |
4903 | intel_plane->base.name, | |
4904 | old->start, old->end, | |
4905 | new->start, new->end); | |
413fc530 | 4906 | } |
4907 | } | |
4908 | } | |
4909 | ||
98d39494 MR |
4910 | static int |
4911 | skl_compute_wm(struct drm_atomic_state *state) | |
4912 | { | |
4913 | struct drm_crtc *crtc; | |
4914 | struct drm_crtc_state *cstate; | |
734fa01f MR |
4915 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); |
4916 | struct skl_wm_values *results = &intel_state->wm_results; | |
367d73d2 | 4917 | struct drm_device *dev = state->dev; |
734fa01f | 4918 | struct skl_pipe_wm *pipe_wm; |
98d39494 | 4919 | bool changed = false; |
734fa01f | 4920 | int ret, i; |
98d39494 | 4921 | |
367d73d2 ML |
4922 | /* |
4923 | * When we distrust bios wm we always need to recompute to set the | |
4924 | * expected DDB allocations for each CRTC. | |
4925 | */ | |
4926 | if (to_i915(dev)->wm.distrust_bios_wm) | |
4927 | changed = true; | |
4928 | ||
98d39494 MR |
4929 | /* |
4930 | * If this transaction isn't actually touching any CRTC's, don't | |
4931 | * bother with watermark calculation. Note that if we pass this | |
4932 | * test, we're guaranteed to hold at least one CRTC state mutex, | |
4933 | * which means we can safely use values like dev_priv->active_crtcs | |
4934 | * since any racing commits that want to update them would need to | |
4935 | * hold _all_ CRTC state mutexes. | |
4936 | */ | |
6ebdb5a0 | 4937 | for_each_new_crtc_in_state(state, crtc, cstate, i) |
98d39494 | 4938 | changed = true; |
367d73d2 | 4939 | |
98d39494 MR |
4940 | if (!changed) |
4941 | return 0; | |
4942 | ||
734fa01f MR |
4943 | /* Clear all dirty flags */ |
4944 | results->dirty_pipes = 0; | |
4945 | ||
9a30a261 | 4946 | ret = skl_compute_ddb(state); |
98d39494 MR |
4947 | if (ret) |
4948 | return ret; | |
4949 | ||
734fa01f MR |
4950 | /* |
4951 | * Calculate WM's for all pipes that are part of this transaction. | |
4952 | * Note that the DDB allocation above may have added more CRTC's that | |
4953 | * weren't otherwise being modified (and set bits in dirty_pipes) if | |
4954 | * pipe allocations had to change. | |
4955 | * | |
4956 | * FIXME: Now that we're doing this in the atomic check phase, we | |
4957 | * should allow skl_update_pipe_wm() to return failure in cases where | |
4958 | * no suitable watermark values can be found. | |
4959 | */ | |
6ebdb5a0 | 4960 | for_each_new_crtc_in_state(state, crtc, cstate, i) { |
734fa01f MR |
4961 | struct intel_crtc_state *intel_cstate = |
4962 | to_intel_crtc_state(cstate); | |
03af79e0 ML |
4963 | const struct skl_pipe_wm *old_pipe_wm = |
4964 | &to_intel_crtc_state(crtc->state)->wm.skl.optimal; | |
734fa01f MR |
4965 | |
4966 | pipe_wm = &intel_cstate->wm.skl.optimal; | |
03af79e0 ML |
4967 | ret = skl_update_pipe_wm(cstate, old_pipe_wm, pipe_wm, |
4968 | &results->ddb, &changed); | |
734fa01f MR |
4969 | if (ret) |
4970 | return ret; | |
4971 | ||
4972 | if (changed) | |
4973 | results->dirty_pipes |= drm_crtc_mask(crtc); | |
4974 | ||
4975 | if ((results->dirty_pipes & drm_crtc_mask(crtc)) == 0) | |
4976 | /* This pipe's WM's did not change */ | |
4977 | continue; | |
4978 | ||
4979 | intel_cstate->update_wm_pre = true; | |
734fa01f MR |
4980 | } |
4981 | ||
413fc530 | 4982 | skl_print_wm_changes(state); |
4983 | ||
98d39494 MR |
4984 | return 0; |
4985 | } | |
4986 | ||
ccf010fb ML |
4987 | static void skl_atomic_update_crtc_wm(struct intel_atomic_state *state, |
4988 | struct intel_crtc_state *cstate) | |
4989 | { | |
4990 | struct intel_crtc *crtc = to_intel_crtc(cstate->base.crtc); | |
4991 | struct drm_i915_private *dev_priv = to_i915(state->base.dev); | |
4992 | struct skl_pipe_wm *pipe_wm = &cstate->wm.skl.optimal; | |
e62929b3 | 4993 | const struct skl_ddb_allocation *ddb = &state->wm_results.ddb; |
ccf010fb | 4994 | enum pipe pipe = crtc->pipe; |
d5cdfdf5 | 4995 | enum plane_id plane_id; |
e62929b3 ML |
4996 | |
4997 | if (!(state->wm_results.dirty_pipes & drm_crtc_mask(&crtc->base))) | |
4998 | return; | |
ccf010fb ML |
4999 | |
5000 | I915_WRITE(PIPE_WM_LINETIME(pipe), pipe_wm->linetime); | |
e62929b3 | 5001 | |
d5cdfdf5 VS |
5002 | for_each_plane_id_on_crtc(crtc, plane_id) { |
5003 | if (plane_id != PLANE_CURSOR) | |
5004 | skl_write_plane_wm(crtc, &pipe_wm->planes[plane_id], | |
5005 | ddb, plane_id); | |
5006 | else | |
5007 | skl_write_cursor_wm(crtc, &pipe_wm->planes[plane_id], | |
5008 | ddb); | |
5009 | } | |
ccf010fb ML |
5010 | } |
5011 | ||
e62929b3 ML |
5012 | static void skl_initial_wm(struct intel_atomic_state *state, |
5013 | struct intel_crtc_state *cstate) | |
2d41c0b5 | 5014 | { |
e62929b3 | 5015 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); |
432081bc | 5016 | struct drm_device *dev = intel_crtc->base.dev; |
fac5e23e | 5017 | struct drm_i915_private *dev_priv = to_i915(dev); |
e62929b3 | 5018 | struct skl_wm_values *results = &state->wm_results; |
2722efb9 | 5019 | struct skl_wm_values *hw_vals = &dev_priv->wm.skl_hw; |
27082493 | 5020 | enum pipe pipe = intel_crtc->pipe; |
adda50b8 | 5021 | |
432081bc | 5022 | if ((results->dirty_pipes & drm_crtc_mask(&intel_crtc->base)) == 0) |
2d41c0b5 PB |
5023 | return; |
5024 | ||
734fa01f | 5025 | mutex_lock(&dev_priv->wm.wm_mutex); |
2d41c0b5 | 5026 | |
e62929b3 ML |
5027 | if (cstate->base.active_changed) |
5028 | skl_atomic_update_crtc_wm(state, cstate); | |
27082493 L |
5029 | |
5030 | skl_copy_wm_for_pipe(hw_vals, results, pipe); | |
734fa01f MR |
5031 | |
5032 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
2d41c0b5 PB |
5033 | } |
5034 | ||
d890565c VS |
5035 | static void ilk_compute_wm_config(struct drm_device *dev, |
5036 | struct intel_wm_config *config) | |
5037 | { | |
5038 | struct intel_crtc *crtc; | |
5039 | ||
5040 | /* Compute the currently _active_ config */ | |
5041 | for_each_intel_crtc(dev, crtc) { | |
5042 | const struct intel_pipe_wm *wm = &crtc->wm.active.ilk; | |
5043 | ||
5044 | if (!wm->pipe_enabled) | |
5045 | continue; | |
5046 | ||
5047 | config->sprites_enabled |= wm->sprites_enabled; | |
5048 | config->sprites_scaled |= wm->sprites_scaled; | |
5049 | config->num_pipes_active++; | |
5050 | } | |
5051 | } | |
5052 | ||
ed4a6a7c | 5053 | static void ilk_program_watermarks(struct drm_i915_private *dev_priv) |
801bcfff | 5054 | { |
91c8a326 | 5055 | struct drm_device *dev = &dev_priv->drm; |
b9d5c839 | 5056 | struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm; |
820c1980 | 5057 | struct ilk_wm_maximums max; |
d890565c | 5058 | struct intel_wm_config config = {}; |
820c1980 | 5059 | struct ilk_wm_values results = {}; |
77c122bc | 5060 | enum intel_ddb_partitioning partitioning; |
261a27d1 | 5061 | |
d890565c VS |
5062 | ilk_compute_wm_config(dev, &config); |
5063 | ||
5064 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max); | |
5065 | ilk_wm_merge(dev, &config, &max, &lp_wm_1_2); | |
a485bfb8 VS |
5066 | |
5067 | /* 5/6 split only in single pipe config on IVB+ */ | |
175fded1 | 5068 | if (INTEL_GEN(dev_priv) >= 7 && |
d890565c VS |
5069 | config.num_pipes_active == 1 && config.sprites_enabled) { |
5070 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max); | |
5071 | ilk_wm_merge(dev, &config, &max, &lp_wm_5_6); | |
0362c781 | 5072 | |
820c1980 | 5073 | best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6); |
861f3389 | 5074 | } else { |
198a1e9b | 5075 | best_lp_wm = &lp_wm_1_2; |
861f3389 PZ |
5076 | } |
5077 | ||
198a1e9b | 5078 | partitioning = (best_lp_wm == &lp_wm_1_2) ? |
77c122bc | 5079 | INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6; |
801bcfff | 5080 | |
820c1980 | 5081 | ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results); |
609cedef | 5082 | |
820c1980 | 5083 | ilk_write_wm_values(dev_priv, &results); |
1011d8c4 PZ |
5084 | } |
5085 | ||
ccf010fb ML |
5086 | static void ilk_initial_watermarks(struct intel_atomic_state *state, |
5087 | struct intel_crtc_state *cstate) | |
b9d5c839 | 5088 | { |
ed4a6a7c MR |
5089 | struct drm_i915_private *dev_priv = to_i915(cstate->base.crtc->dev); |
5090 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); | |
b9d5c839 | 5091 | |
ed4a6a7c | 5092 | mutex_lock(&dev_priv->wm.wm_mutex); |
e8f1f02e | 5093 | intel_crtc->wm.active.ilk = cstate->wm.ilk.intermediate; |
ed4a6a7c MR |
5094 | ilk_program_watermarks(dev_priv); |
5095 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
5096 | } | |
bf220452 | 5097 | |
ccf010fb ML |
5098 | static void ilk_optimize_watermarks(struct intel_atomic_state *state, |
5099 | struct intel_crtc_state *cstate) | |
ed4a6a7c MR |
5100 | { |
5101 | struct drm_i915_private *dev_priv = to_i915(cstate->base.crtc->dev); | |
5102 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); | |
bf220452 | 5103 | |
ed4a6a7c MR |
5104 | mutex_lock(&dev_priv->wm.wm_mutex); |
5105 | if (cstate->wm.need_postvbl_update) { | |
e8f1f02e | 5106 | intel_crtc->wm.active.ilk = cstate->wm.ilk.optimal; |
ed4a6a7c MR |
5107 | ilk_program_watermarks(dev_priv); |
5108 | } | |
5109 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
b9d5c839 VS |
5110 | } |
5111 | ||
d8c0fafc | 5112 | static inline void skl_wm_level_from_reg_val(uint32_t val, |
5113 | struct skl_wm_level *level) | |
3078999f | 5114 | { |
d8c0fafc | 5115 | level->plane_en = val & PLANE_WM_EN; |
5116 | level->plane_res_b = val & PLANE_WM_BLOCKS_MASK; | |
5117 | level->plane_res_l = (val >> PLANE_WM_LINES_SHIFT) & | |
5118 | PLANE_WM_LINES_MASK; | |
3078999f PB |
5119 | } |
5120 | ||
bf9d99ad | 5121 | void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc, |
5122 | struct skl_pipe_wm *out) | |
3078999f | 5123 | { |
d5cdfdf5 | 5124 | struct drm_i915_private *dev_priv = to_i915(crtc->dev); |
3078999f | 5125 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3078999f | 5126 | enum pipe pipe = intel_crtc->pipe; |
d5cdfdf5 VS |
5127 | int level, max_level; |
5128 | enum plane_id plane_id; | |
d8c0fafc | 5129 | uint32_t val; |
3078999f | 5130 | |
5db94019 | 5131 | max_level = ilk_wm_max_level(dev_priv); |
3078999f | 5132 | |
d5cdfdf5 VS |
5133 | for_each_plane_id_on_crtc(intel_crtc, plane_id) { |
5134 | struct skl_plane_wm *wm = &out->planes[plane_id]; | |
3078999f | 5135 | |
d8c0fafc | 5136 | for (level = 0; level <= max_level; level++) { |
d5cdfdf5 VS |
5137 | if (plane_id != PLANE_CURSOR) |
5138 | val = I915_READ(PLANE_WM(pipe, plane_id, level)); | |
d8c0fafc | 5139 | else |
5140 | val = I915_READ(CUR_WM(pipe, level)); | |
3078999f | 5141 | |
d8c0fafc | 5142 | skl_wm_level_from_reg_val(val, &wm->wm[level]); |
3078999f | 5143 | } |
3078999f | 5144 | |
d5cdfdf5 VS |
5145 | if (plane_id != PLANE_CURSOR) |
5146 | val = I915_READ(PLANE_WM_TRANS(pipe, plane_id)); | |
d8c0fafc | 5147 | else |
5148 | val = I915_READ(CUR_WM_TRANS(pipe)); | |
5149 | ||
5150 | skl_wm_level_from_reg_val(val, &wm->trans_wm); | |
3078999f PB |
5151 | } |
5152 | ||
d8c0fafc | 5153 | if (!intel_crtc->active) |
5154 | return; | |
4e0963c7 | 5155 | |
bf9d99ad | 5156 | out->linetime = I915_READ(PIPE_WM_LINETIME(pipe)); |
3078999f PB |
5157 | } |
5158 | ||
5159 | void skl_wm_get_hw_state(struct drm_device *dev) | |
5160 | { | |
fac5e23e | 5161 | struct drm_i915_private *dev_priv = to_i915(dev); |
bf9d99ad | 5162 | struct skl_wm_values *hw = &dev_priv->wm.skl_hw; |
a269c583 | 5163 | struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb; |
3078999f | 5164 | struct drm_crtc *crtc; |
bf9d99ad | 5165 | struct intel_crtc *intel_crtc; |
5166 | struct intel_crtc_state *cstate; | |
3078999f | 5167 | |
a269c583 | 5168 | skl_ddb_get_hw_state(dev_priv, ddb); |
bf9d99ad | 5169 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { |
5170 | intel_crtc = to_intel_crtc(crtc); | |
5171 | cstate = to_intel_crtc_state(crtc->state); | |
5172 | ||
5173 | skl_pipe_wm_get_hw_state(crtc, &cstate->wm.skl.optimal); | |
5174 | ||
03af79e0 | 5175 | if (intel_crtc->active) |
bf9d99ad | 5176 | hw->dirty_pipes |= drm_crtc_mask(crtc); |
bf9d99ad | 5177 | } |
a1de91e5 | 5178 | |
279e99d7 MR |
5179 | if (dev_priv->active_crtcs) { |
5180 | /* Fully recompute DDB on first atomic commit */ | |
5181 | dev_priv->wm.distrust_bios_wm = true; | |
5182 | } else { | |
5183 | /* Easy/common case; just sanitize DDB now if everything off */ | |
5184 | memset(ddb, 0, sizeof(*ddb)); | |
5185 | } | |
3078999f PB |
5186 | } |
5187 | ||
243e6a44 VS |
5188 | static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
5189 | { | |
5190 | struct drm_device *dev = crtc->dev; | |
fac5e23e | 5191 | struct drm_i915_private *dev_priv = to_i915(dev); |
820c1980 | 5192 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
243e6a44 | 5193 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
4e0963c7 | 5194 | struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state); |
e8f1f02e | 5195 | struct intel_pipe_wm *active = &cstate->wm.ilk.optimal; |
243e6a44 | 5196 | enum pipe pipe = intel_crtc->pipe; |
f0f59a00 | 5197 | static const i915_reg_t wm0_pipe_reg[] = { |
243e6a44 VS |
5198 | [PIPE_A] = WM0_PIPEA_ILK, |
5199 | [PIPE_B] = WM0_PIPEB_ILK, | |
5200 | [PIPE_C] = WM0_PIPEC_IVB, | |
5201 | }; | |
5202 | ||
5203 | hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]); | |
8652744b | 5204 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
ce0e0713 | 5205 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
243e6a44 | 5206 | |
15606534 VS |
5207 | memset(active, 0, sizeof(*active)); |
5208 | ||
3ef00284 | 5209 | active->pipe_enabled = intel_crtc->active; |
2a44b76b VS |
5210 | |
5211 | if (active->pipe_enabled) { | |
243e6a44 VS |
5212 | u32 tmp = hw->wm_pipe[pipe]; |
5213 | ||
5214 | /* | |
5215 | * For active pipes LP0 watermark is marked as | |
5216 | * enabled, and LP1+ watermaks as disabled since | |
5217 | * we can't really reverse compute them in case | |
5218 | * multiple pipes are active. | |
5219 | */ | |
5220 | active->wm[0].enable = true; | |
5221 | active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT; | |
5222 | active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT; | |
5223 | active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK; | |
5224 | active->linetime = hw->wm_linetime[pipe]; | |
5225 | } else { | |
5db94019 | 5226 | int level, max_level = ilk_wm_max_level(dev_priv); |
243e6a44 VS |
5227 | |
5228 | /* | |
5229 | * For inactive pipes, all watermark levels | |
5230 | * should be marked as enabled but zeroed, | |
5231 | * which is what we'd compute them to. | |
5232 | */ | |
5233 | for (level = 0; level <= max_level; level++) | |
5234 | active->wm[level].enable = true; | |
5235 | } | |
4e0963c7 MR |
5236 | |
5237 | intel_crtc->wm.active.ilk = *active; | |
243e6a44 VS |
5238 | } |
5239 | ||
6eb1a681 VS |
5240 | #define _FW_WM(value, plane) \ |
5241 | (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT) | |
5242 | #define _FW_WM_VLV(value, plane) \ | |
5243 | (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT) | |
5244 | ||
04548cba VS |
5245 | static void g4x_read_wm_values(struct drm_i915_private *dev_priv, |
5246 | struct g4x_wm_values *wm) | |
5247 | { | |
5248 | uint32_t tmp; | |
5249 | ||
5250 | tmp = I915_READ(DSPFW1); | |
5251 | wm->sr.plane = _FW_WM(tmp, SR); | |
5252 | wm->pipe[PIPE_B].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORB); | |
5253 | wm->pipe[PIPE_B].plane[PLANE_PRIMARY] = _FW_WM(tmp, PLANEB); | |
5254 | wm->pipe[PIPE_A].plane[PLANE_PRIMARY] = _FW_WM(tmp, PLANEA); | |
5255 | ||
5256 | tmp = I915_READ(DSPFW2); | |
5257 | wm->fbc_en = tmp & DSPFW_FBC_SR_EN; | |
5258 | wm->sr.fbc = _FW_WM(tmp, FBC_SR); | |
5259 | wm->hpll.fbc = _FW_WM(tmp, FBC_HPLL_SR); | |
5260 | wm->pipe[PIPE_B].plane[PLANE_SPRITE0] = _FW_WM(tmp, SPRITEB); | |
5261 | wm->pipe[PIPE_A].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORA); | |
5262 | wm->pipe[PIPE_A].plane[PLANE_SPRITE0] = _FW_WM(tmp, SPRITEA); | |
5263 | ||
5264 | tmp = I915_READ(DSPFW3); | |
5265 | wm->hpll_en = tmp & DSPFW_HPLL_SR_EN; | |
5266 | wm->sr.cursor = _FW_WM(tmp, CURSOR_SR); | |
5267 | wm->hpll.cursor = _FW_WM(tmp, HPLL_CURSOR); | |
5268 | wm->hpll.plane = _FW_WM(tmp, HPLL_SR); | |
5269 | } | |
5270 | ||
6eb1a681 VS |
5271 | static void vlv_read_wm_values(struct drm_i915_private *dev_priv, |
5272 | struct vlv_wm_values *wm) | |
5273 | { | |
5274 | enum pipe pipe; | |
5275 | uint32_t tmp; | |
5276 | ||
5277 | for_each_pipe(dev_priv, pipe) { | |
5278 | tmp = I915_READ(VLV_DDL(pipe)); | |
5279 | ||
1b31389c | 5280 | wm->ddl[pipe].plane[PLANE_PRIMARY] = |
6eb1a681 | 5281 | (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
1b31389c | 5282 | wm->ddl[pipe].plane[PLANE_CURSOR] = |
6eb1a681 | 5283 | (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
1b31389c | 5284 | wm->ddl[pipe].plane[PLANE_SPRITE0] = |
6eb1a681 | 5285 | (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
1b31389c | 5286 | wm->ddl[pipe].plane[PLANE_SPRITE1] = |
6eb1a681 VS |
5287 | (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); |
5288 | } | |
5289 | ||
5290 | tmp = I915_READ(DSPFW1); | |
5291 | wm->sr.plane = _FW_WM(tmp, SR); | |
1b31389c VS |
5292 | wm->pipe[PIPE_B].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORB); |
5293 | wm->pipe[PIPE_B].plane[PLANE_PRIMARY] = _FW_WM_VLV(tmp, PLANEB); | |
5294 | wm->pipe[PIPE_A].plane[PLANE_PRIMARY] = _FW_WM_VLV(tmp, PLANEA); | |
6eb1a681 VS |
5295 | |
5296 | tmp = I915_READ(DSPFW2); | |
1b31389c VS |
5297 | wm->pipe[PIPE_A].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITEB); |
5298 | wm->pipe[PIPE_A].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORA); | |
5299 | wm->pipe[PIPE_A].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEA); | |
6eb1a681 VS |
5300 | |
5301 | tmp = I915_READ(DSPFW3); | |
5302 | wm->sr.cursor = _FW_WM(tmp, CURSOR_SR); | |
5303 | ||
5304 | if (IS_CHERRYVIEW(dev_priv)) { | |
5305 | tmp = I915_READ(DSPFW7_CHV); | |
1b31389c VS |
5306 | wm->pipe[PIPE_B].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITED); |
5307 | wm->pipe[PIPE_B].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEC); | |
6eb1a681 VS |
5308 | |
5309 | tmp = I915_READ(DSPFW8_CHV); | |
1b31389c VS |
5310 | wm->pipe[PIPE_C].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITEF); |
5311 | wm->pipe[PIPE_C].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEE); | |
6eb1a681 VS |
5312 | |
5313 | tmp = I915_READ(DSPFW9_CHV); | |
1b31389c VS |
5314 | wm->pipe[PIPE_C].plane[PLANE_PRIMARY] = _FW_WM_VLV(tmp, PLANEC); |
5315 | wm->pipe[PIPE_C].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORC); | |
6eb1a681 VS |
5316 | |
5317 | tmp = I915_READ(DSPHOWM); | |
5318 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; | |
1b31389c VS |
5319 | wm->pipe[PIPE_C].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITEF_HI) << 8; |
5320 | wm->pipe[PIPE_C].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEE_HI) << 8; | |
5321 | wm->pipe[PIPE_C].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEC_HI) << 8; | |
5322 | wm->pipe[PIPE_B].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITED_HI) << 8; | |
5323 | wm->pipe[PIPE_B].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEC_HI) << 8; | |
5324 | wm->pipe[PIPE_B].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEB_HI) << 8; | |
5325 | wm->pipe[PIPE_A].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITEB_HI) << 8; | |
5326 | wm->pipe[PIPE_A].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEA_HI) << 8; | |
5327 | wm->pipe[PIPE_A].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEA_HI) << 8; | |
6eb1a681 VS |
5328 | } else { |
5329 | tmp = I915_READ(DSPFW7); | |
1b31389c VS |
5330 | wm->pipe[PIPE_B].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITED); |
5331 | wm->pipe[PIPE_B].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEC); | |
6eb1a681 VS |
5332 | |
5333 | tmp = I915_READ(DSPHOWM); | |
5334 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; | |
1b31389c VS |
5335 | wm->pipe[PIPE_B].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITED_HI) << 8; |
5336 | wm->pipe[PIPE_B].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEC_HI) << 8; | |
5337 | wm->pipe[PIPE_B].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEB_HI) << 8; | |
5338 | wm->pipe[PIPE_A].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITEB_HI) << 8; | |
5339 | wm->pipe[PIPE_A].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEA_HI) << 8; | |
5340 | wm->pipe[PIPE_A].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEA_HI) << 8; | |
6eb1a681 VS |
5341 | } |
5342 | } | |
5343 | ||
5344 | #undef _FW_WM | |
5345 | #undef _FW_WM_VLV | |
5346 | ||
04548cba VS |
5347 | void g4x_wm_get_hw_state(struct drm_device *dev) |
5348 | { | |
5349 | struct drm_i915_private *dev_priv = to_i915(dev); | |
5350 | struct g4x_wm_values *wm = &dev_priv->wm.g4x; | |
5351 | struct intel_crtc *crtc; | |
5352 | ||
5353 | g4x_read_wm_values(dev_priv, wm); | |
5354 | ||
5355 | wm->cxsr = I915_READ(FW_BLC_SELF) & FW_BLC_SELF_EN; | |
5356 | ||
5357 | for_each_intel_crtc(dev, crtc) { | |
5358 | struct intel_crtc_state *crtc_state = | |
5359 | to_intel_crtc_state(crtc->base.state); | |
5360 | struct g4x_wm_state *active = &crtc->wm.active.g4x; | |
5361 | struct g4x_pipe_wm *raw; | |
5362 | enum pipe pipe = crtc->pipe; | |
5363 | enum plane_id plane_id; | |
5364 | int level, max_level; | |
5365 | ||
5366 | active->cxsr = wm->cxsr; | |
5367 | active->hpll_en = wm->hpll_en; | |
5368 | active->fbc_en = wm->fbc_en; | |
5369 | ||
5370 | active->sr = wm->sr; | |
5371 | active->hpll = wm->hpll; | |
5372 | ||
5373 | for_each_plane_id_on_crtc(crtc, plane_id) { | |
5374 | active->wm.plane[plane_id] = | |
5375 | wm->pipe[pipe].plane[plane_id]; | |
5376 | } | |
5377 | ||
5378 | if (wm->cxsr && wm->hpll_en) | |
5379 | max_level = G4X_WM_LEVEL_HPLL; | |
5380 | else if (wm->cxsr) | |
5381 | max_level = G4X_WM_LEVEL_SR; | |
5382 | else | |
5383 | max_level = G4X_WM_LEVEL_NORMAL; | |
5384 | ||
5385 | level = G4X_WM_LEVEL_NORMAL; | |
5386 | raw = &crtc_state->wm.g4x.raw[level]; | |
5387 | for_each_plane_id_on_crtc(crtc, plane_id) | |
5388 | raw->plane[plane_id] = active->wm.plane[plane_id]; | |
5389 | ||
5390 | if (++level > max_level) | |
5391 | goto out; | |
5392 | ||
5393 | raw = &crtc_state->wm.g4x.raw[level]; | |
5394 | raw->plane[PLANE_PRIMARY] = active->sr.plane; | |
5395 | raw->plane[PLANE_CURSOR] = active->sr.cursor; | |
5396 | raw->plane[PLANE_SPRITE0] = 0; | |
5397 | raw->fbc = active->sr.fbc; | |
5398 | ||
5399 | if (++level > max_level) | |
5400 | goto out; | |
5401 | ||
5402 | raw = &crtc_state->wm.g4x.raw[level]; | |
5403 | raw->plane[PLANE_PRIMARY] = active->hpll.plane; | |
5404 | raw->plane[PLANE_CURSOR] = active->hpll.cursor; | |
5405 | raw->plane[PLANE_SPRITE0] = 0; | |
5406 | raw->fbc = active->hpll.fbc; | |
5407 | ||
5408 | out: | |
5409 | for_each_plane_id_on_crtc(crtc, plane_id) | |
5410 | g4x_raw_plane_wm_set(crtc_state, level, | |
5411 | plane_id, USHRT_MAX); | |
5412 | g4x_raw_fbc_wm_set(crtc_state, level, USHRT_MAX); | |
5413 | ||
5414 | crtc_state->wm.g4x.optimal = *active; | |
5415 | crtc_state->wm.g4x.intermediate = *active; | |
5416 | ||
5417 | DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite=%d\n", | |
5418 | pipe_name(pipe), | |
5419 | wm->pipe[pipe].plane[PLANE_PRIMARY], | |
5420 | wm->pipe[pipe].plane[PLANE_CURSOR], | |
5421 | wm->pipe[pipe].plane[PLANE_SPRITE0]); | |
5422 | } | |
5423 | ||
5424 | DRM_DEBUG_KMS("Initial SR watermarks: plane=%d, cursor=%d fbc=%d\n", | |
5425 | wm->sr.plane, wm->sr.cursor, wm->sr.fbc); | |
5426 | DRM_DEBUG_KMS("Initial HPLL watermarks: plane=%d, SR cursor=%d fbc=%d\n", | |
5427 | wm->hpll.plane, wm->hpll.cursor, wm->hpll.fbc); | |
5428 | DRM_DEBUG_KMS("Initial SR=%s HPLL=%s FBC=%s\n", | |
5429 | yesno(wm->cxsr), yesno(wm->hpll_en), yesno(wm->fbc_en)); | |
5430 | } | |
5431 | ||
5432 | void g4x_wm_sanitize(struct drm_i915_private *dev_priv) | |
5433 | { | |
5434 | struct intel_plane *plane; | |
5435 | struct intel_crtc *crtc; | |
5436 | ||
5437 | mutex_lock(&dev_priv->wm.wm_mutex); | |
5438 | ||
5439 | for_each_intel_plane(&dev_priv->drm, plane) { | |
5440 | struct intel_crtc *crtc = | |
5441 | intel_get_crtc_for_pipe(dev_priv, plane->pipe); | |
5442 | struct intel_crtc_state *crtc_state = | |
5443 | to_intel_crtc_state(crtc->base.state); | |
5444 | struct intel_plane_state *plane_state = | |
5445 | to_intel_plane_state(plane->base.state); | |
5446 | struct g4x_wm_state *wm_state = &crtc_state->wm.g4x.optimal; | |
5447 | enum plane_id plane_id = plane->id; | |
5448 | int level; | |
5449 | ||
5450 | if (plane_state->base.visible) | |
5451 | continue; | |
5452 | ||
5453 | for (level = 0; level < 3; level++) { | |
5454 | struct g4x_pipe_wm *raw = | |
5455 | &crtc_state->wm.g4x.raw[level]; | |
5456 | ||
5457 | raw->plane[plane_id] = 0; | |
5458 | wm_state->wm.plane[plane_id] = 0; | |
5459 | } | |
5460 | ||
5461 | if (plane_id == PLANE_PRIMARY) { | |
5462 | for (level = 0; level < 3; level++) { | |
5463 | struct g4x_pipe_wm *raw = | |
5464 | &crtc_state->wm.g4x.raw[level]; | |
5465 | raw->fbc = 0; | |
5466 | } | |
5467 | ||
5468 | wm_state->sr.fbc = 0; | |
5469 | wm_state->hpll.fbc = 0; | |
5470 | wm_state->fbc_en = false; | |
5471 | } | |
5472 | } | |
5473 | ||
5474 | for_each_intel_crtc(&dev_priv->drm, crtc) { | |
5475 | struct intel_crtc_state *crtc_state = | |
5476 | to_intel_crtc_state(crtc->base.state); | |
5477 | ||
5478 | crtc_state->wm.g4x.intermediate = | |
5479 | crtc_state->wm.g4x.optimal; | |
5480 | crtc->wm.active.g4x = crtc_state->wm.g4x.optimal; | |
5481 | } | |
5482 | ||
5483 | g4x_program_watermarks(dev_priv); | |
5484 | ||
5485 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
5486 | } | |
5487 | ||
6eb1a681 VS |
5488 | void vlv_wm_get_hw_state(struct drm_device *dev) |
5489 | { | |
5490 | struct drm_i915_private *dev_priv = to_i915(dev); | |
5491 | struct vlv_wm_values *wm = &dev_priv->wm.vlv; | |
f07d43d2 | 5492 | struct intel_crtc *crtc; |
6eb1a681 VS |
5493 | u32 val; |
5494 | ||
5495 | vlv_read_wm_values(dev_priv, wm); | |
5496 | ||
6eb1a681 VS |
5497 | wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN; |
5498 | wm->level = VLV_WM_LEVEL_PM2; | |
5499 | ||
5500 | if (IS_CHERRYVIEW(dev_priv)) { | |
5501 | mutex_lock(&dev_priv->rps.hw_lock); | |
5502 | ||
5503 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); | |
5504 | if (val & DSP_MAXFIFO_PM5_ENABLE) | |
5505 | wm->level = VLV_WM_LEVEL_PM5; | |
5506 | ||
58590c14 VS |
5507 | /* |
5508 | * If DDR DVFS is disabled in the BIOS, Punit | |
5509 | * will never ack the request. So if that happens | |
5510 | * assume we don't have to enable/disable DDR DVFS | |
5511 | * dynamically. To test that just set the REQ_ACK | |
5512 | * bit to poke the Punit, but don't change the | |
5513 | * HIGH/LOW bits so that we don't actually change | |
5514 | * the current state. | |
5515 | */ | |
6eb1a681 | 5516 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); |
58590c14 VS |
5517 | val |= FORCE_DDR_FREQ_REQ_ACK; |
5518 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); | |
5519 | ||
5520 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & | |
5521 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) { | |
5522 | DRM_DEBUG_KMS("Punit not acking DDR DVFS request, " | |
5523 | "assuming DDR DVFS is disabled\n"); | |
5524 | dev_priv->wm.max_level = VLV_WM_LEVEL_PM5; | |
5525 | } else { | |
5526 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); | |
5527 | if ((val & FORCE_DDR_HIGH_FREQ) == 0) | |
5528 | wm->level = VLV_WM_LEVEL_DDR_DVFS; | |
5529 | } | |
6eb1a681 VS |
5530 | |
5531 | mutex_unlock(&dev_priv->rps.hw_lock); | |
5532 | } | |
5533 | ||
ff32c54e VS |
5534 | for_each_intel_crtc(dev, crtc) { |
5535 | struct intel_crtc_state *crtc_state = | |
5536 | to_intel_crtc_state(crtc->base.state); | |
5537 | struct vlv_wm_state *active = &crtc->wm.active.vlv; | |
5538 | const struct vlv_fifo_state *fifo_state = | |
5539 | &crtc_state->wm.vlv.fifo_state; | |
5540 | enum pipe pipe = crtc->pipe; | |
5541 | enum plane_id plane_id; | |
5542 | int level; | |
5543 | ||
5544 | vlv_get_fifo_size(crtc_state); | |
5545 | ||
5546 | active->num_levels = wm->level + 1; | |
5547 | active->cxsr = wm->cxsr; | |
5548 | ||
ff32c54e | 5549 | for (level = 0; level < active->num_levels; level++) { |
114d7dc0 | 5550 | struct g4x_pipe_wm *raw = |
ff32c54e VS |
5551 | &crtc_state->wm.vlv.raw[level]; |
5552 | ||
5553 | active->sr[level].plane = wm->sr.plane; | |
5554 | active->sr[level].cursor = wm->sr.cursor; | |
5555 | ||
5556 | for_each_plane_id_on_crtc(crtc, plane_id) { | |
5557 | active->wm[level].plane[plane_id] = | |
5558 | wm->pipe[pipe].plane[plane_id]; | |
5559 | ||
5560 | raw->plane[plane_id] = | |
5561 | vlv_invert_wm_value(active->wm[level].plane[plane_id], | |
5562 | fifo_state->plane[plane_id]); | |
5563 | } | |
5564 | } | |
5565 | ||
5566 | for_each_plane_id_on_crtc(crtc, plane_id) | |
5567 | vlv_raw_plane_wm_set(crtc_state, level, | |
5568 | plane_id, USHRT_MAX); | |
5569 | vlv_invalidate_wms(crtc, active, level); | |
5570 | ||
5571 | crtc_state->wm.vlv.optimal = *active; | |
4841da51 | 5572 | crtc_state->wm.vlv.intermediate = *active; |
ff32c54e | 5573 | |
6eb1a681 | 5574 | DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n", |
1b31389c VS |
5575 | pipe_name(pipe), |
5576 | wm->pipe[pipe].plane[PLANE_PRIMARY], | |
5577 | wm->pipe[pipe].plane[PLANE_CURSOR], | |
5578 | wm->pipe[pipe].plane[PLANE_SPRITE0], | |
5579 | wm->pipe[pipe].plane[PLANE_SPRITE1]); | |
ff32c54e | 5580 | } |
6eb1a681 VS |
5581 | |
5582 | DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n", | |
5583 | wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr); | |
5584 | } | |
5585 | ||
602ae835 VS |
5586 | void vlv_wm_sanitize(struct drm_i915_private *dev_priv) |
5587 | { | |
5588 | struct intel_plane *plane; | |
5589 | struct intel_crtc *crtc; | |
5590 | ||
5591 | mutex_lock(&dev_priv->wm.wm_mutex); | |
5592 | ||
5593 | for_each_intel_plane(&dev_priv->drm, plane) { | |
5594 | struct intel_crtc *crtc = | |
5595 | intel_get_crtc_for_pipe(dev_priv, plane->pipe); | |
5596 | struct intel_crtc_state *crtc_state = | |
5597 | to_intel_crtc_state(crtc->base.state); | |
5598 | struct intel_plane_state *plane_state = | |
5599 | to_intel_plane_state(plane->base.state); | |
5600 | struct vlv_wm_state *wm_state = &crtc_state->wm.vlv.optimal; | |
5601 | const struct vlv_fifo_state *fifo_state = | |
5602 | &crtc_state->wm.vlv.fifo_state; | |
5603 | enum plane_id plane_id = plane->id; | |
5604 | int level; | |
5605 | ||
5606 | if (plane_state->base.visible) | |
5607 | continue; | |
5608 | ||
5609 | for (level = 0; level < wm_state->num_levels; level++) { | |
114d7dc0 | 5610 | struct g4x_pipe_wm *raw = |
602ae835 VS |
5611 | &crtc_state->wm.vlv.raw[level]; |
5612 | ||
5613 | raw->plane[plane_id] = 0; | |
5614 | ||
5615 | wm_state->wm[level].plane[plane_id] = | |
5616 | vlv_invert_wm_value(raw->plane[plane_id], | |
5617 | fifo_state->plane[plane_id]); | |
5618 | } | |
5619 | } | |
5620 | ||
5621 | for_each_intel_crtc(&dev_priv->drm, crtc) { | |
5622 | struct intel_crtc_state *crtc_state = | |
5623 | to_intel_crtc_state(crtc->base.state); | |
5624 | ||
5625 | crtc_state->wm.vlv.intermediate = | |
5626 | crtc_state->wm.vlv.optimal; | |
5627 | crtc->wm.active.vlv = crtc_state->wm.vlv.optimal; | |
5628 | } | |
5629 | ||
5630 | vlv_program_watermarks(dev_priv); | |
5631 | ||
5632 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
5633 | } | |
5634 | ||
243e6a44 VS |
5635 | void ilk_wm_get_hw_state(struct drm_device *dev) |
5636 | { | |
fac5e23e | 5637 | struct drm_i915_private *dev_priv = to_i915(dev); |
820c1980 | 5638 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
243e6a44 VS |
5639 | struct drm_crtc *crtc; |
5640 | ||
70e1e0ec | 5641 | for_each_crtc(dev, crtc) |
243e6a44 VS |
5642 | ilk_pipe_wm_get_hw_state(crtc); |
5643 | ||
5644 | hw->wm_lp[0] = I915_READ(WM1_LP_ILK); | |
5645 | hw->wm_lp[1] = I915_READ(WM2_LP_ILK); | |
5646 | hw->wm_lp[2] = I915_READ(WM3_LP_ILK); | |
5647 | ||
5648 | hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK); | |
175fded1 | 5649 | if (INTEL_GEN(dev_priv) >= 7) { |
cfa7698b VS |
5650 | hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB); |
5651 | hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB); | |
5652 | } | |
243e6a44 | 5653 | |
8652744b | 5654 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
ac9545fd VS |
5655 | hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ? |
5656 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; | |
fd6b8f43 | 5657 | else if (IS_IVYBRIDGE(dev_priv)) |
ac9545fd VS |
5658 | hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ? |
5659 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; | |
243e6a44 VS |
5660 | |
5661 | hw->enable_fbc_wm = | |
5662 | !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS); | |
5663 | } | |
5664 | ||
b445e3b0 ED |
5665 | /** |
5666 | * intel_update_watermarks - update FIFO watermark values based on current modes | |
5667 | * | |
5668 | * Calculate watermark values for the various WM regs based on current mode | |
5669 | * and plane configuration. | |
5670 | * | |
5671 | * There are several cases to deal with here: | |
5672 | * - normal (i.e. non-self-refresh) | |
5673 | * - self-refresh (SR) mode | |
5674 | * - lines are large relative to FIFO size (buffer can hold up to 2) | |
5675 | * - lines are small relative to FIFO size (buffer can hold more than 2 | |
5676 | * lines), so need to account for TLB latency | |
5677 | * | |
5678 | * The normal calculation is: | |
5679 | * watermark = dotclock * bytes per pixel * latency | |
5680 | * where latency is platform & configuration dependent (we assume pessimal | |
5681 | * values here). | |
5682 | * | |
5683 | * The SR calculation is: | |
5684 | * watermark = (trunc(latency/line time)+1) * surface width * | |
5685 | * bytes per pixel | |
5686 | * where | |
5687 | * line time = htotal / dotclock | |
5688 | * surface width = hdisplay for normal plane and 64 for cursor | |
5689 | * and latency is assumed to be high, as above. | |
5690 | * | |
5691 | * The final value programmed to the register should always be rounded up, | |
5692 | * and include an extra 2 entries to account for clock crossings. | |
5693 | * | |
5694 | * We don't use the sprite, so we can ignore that. And on Crestline we have | |
5695 | * to set the non-SR watermarks to 8. | |
5696 | */ | |
432081bc | 5697 | void intel_update_watermarks(struct intel_crtc *crtc) |
b445e3b0 | 5698 | { |
432081bc | 5699 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); |
b445e3b0 ED |
5700 | |
5701 | if (dev_priv->display.update_wm) | |
46ba614c | 5702 | dev_priv->display.update_wm(crtc); |
b445e3b0 ED |
5703 | } |
5704 | ||
e2828914 | 5705 | /* |
9270388e | 5706 | * Lock protecting IPS related data structures |
9270388e DV |
5707 | */ |
5708 | DEFINE_SPINLOCK(mchdev_lock); | |
5709 | ||
5710 | /* Global for IPS driver to get at the current i915 device. Protected by | |
5711 | * mchdev_lock. */ | |
5712 | static struct drm_i915_private *i915_mch_dev; | |
5713 | ||
91d14251 | 5714 | bool ironlake_set_drps(struct drm_i915_private *dev_priv, u8 val) |
2b4e57bd | 5715 | { |
2b4e57bd ED |
5716 | u16 rgvswctl; |
5717 | ||
67520415 | 5718 | lockdep_assert_held(&mchdev_lock); |
9270388e | 5719 | |
2b4e57bd ED |
5720 | rgvswctl = I915_READ16(MEMSWCTL); |
5721 | if (rgvswctl & MEMCTL_CMD_STS) { | |
5722 | DRM_DEBUG("gpu busy, RCS change rejected\n"); | |
5723 | return false; /* still busy with another command */ | |
5724 | } | |
5725 | ||
5726 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | | |
5727 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; | |
5728 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
5729 | POSTING_READ16(MEMSWCTL); | |
5730 | ||
5731 | rgvswctl |= MEMCTL_CMD_STS; | |
5732 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
5733 | ||
5734 | return true; | |
5735 | } | |
5736 | ||
91d14251 | 5737 | static void ironlake_enable_drps(struct drm_i915_private *dev_priv) |
2b4e57bd | 5738 | { |
84f1b20f | 5739 | u32 rgvmodectl; |
2b4e57bd ED |
5740 | u8 fmax, fmin, fstart, vstart; |
5741 | ||
9270388e DV |
5742 | spin_lock_irq(&mchdev_lock); |
5743 | ||
84f1b20f TU |
5744 | rgvmodectl = I915_READ(MEMMODECTL); |
5745 | ||
2b4e57bd ED |
5746 | /* Enable temp reporting */ |
5747 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); | |
5748 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); | |
5749 | ||
5750 | /* 100ms RC evaluation intervals */ | |
5751 | I915_WRITE(RCUPEI, 100000); | |
5752 | I915_WRITE(RCDNEI, 100000); | |
5753 | ||
5754 | /* Set max/min thresholds to 90ms and 80ms respectively */ | |
5755 | I915_WRITE(RCBMAXAVG, 90000); | |
5756 | I915_WRITE(RCBMINAVG, 80000); | |
5757 | ||
5758 | I915_WRITE(MEMIHYST, 1); | |
5759 | ||
5760 | /* Set up min, max, and cur for interrupt handling */ | |
5761 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; | |
5762 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); | |
5763 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> | |
5764 | MEMMODE_FSTART_SHIFT; | |
5765 | ||
616847e7 | 5766 | vstart = (I915_READ(PXVFREQ(fstart)) & PXVFREQ_PX_MASK) >> |
2b4e57bd ED |
5767 | PXVFREQ_PX_SHIFT; |
5768 | ||
20e4d407 DV |
5769 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
5770 | dev_priv->ips.fstart = fstart; | |
2b4e57bd | 5771 | |
20e4d407 DV |
5772 | dev_priv->ips.max_delay = fstart; |
5773 | dev_priv->ips.min_delay = fmin; | |
5774 | dev_priv->ips.cur_delay = fstart; | |
2b4e57bd ED |
5775 | |
5776 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", | |
5777 | fmax, fmin, fstart); | |
5778 | ||
5779 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); | |
5780 | ||
5781 | /* | |
5782 | * Interrupts will be enabled in ironlake_irq_postinstall | |
5783 | */ | |
5784 | ||
5785 | I915_WRITE(VIDSTART, vstart); | |
5786 | POSTING_READ(VIDSTART); | |
5787 | ||
5788 | rgvmodectl |= MEMMODE_SWMODE_EN; | |
5789 | I915_WRITE(MEMMODECTL, rgvmodectl); | |
5790 | ||
9270388e | 5791 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
2b4e57bd | 5792 | DRM_ERROR("stuck trying to change perf mode\n"); |
dd92d8de | 5793 | mdelay(1); |
2b4e57bd | 5794 | |
91d14251 | 5795 | ironlake_set_drps(dev_priv, fstart); |
2b4e57bd | 5796 | |
7d81c3e0 VS |
5797 | dev_priv->ips.last_count1 = I915_READ(DMIEC) + |
5798 | I915_READ(DDREC) + I915_READ(CSIEC); | |
20e4d407 | 5799 | dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies); |
7d81c3e0 | 5800 | dev_priv->ips.last_count2 = I915_READ(GFXEC); |
5ed0bdf2 | 5801 | dev_priv->ips.last_time2 = ktime_get_raw_ns(); |
9270388e DV |
5802 | |
5803 | spin_unlock_irq(&mchdev_lock); | |
2b4e57bd ED |
5804 | } |
5805 | ||
91d14251 | 5806 | static void ironlake_disable_drps(struct drm_i915_private *dev_priv) |
2b4e57bd | 5807 | { |
9270388e DV |
5808 | u16 rgvswctl; |
5809 | ||
5810 | spin_lock_irq(&mchdev_lock); | |
5811 | ||
5812 | rgvswctl = I915_READ16(MEMSWCTL); | |
2b4e57bd ED |
5813 | |
5814 | /* Ack interrupts, disable EFC interrupt */ | |
5815 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); | |
5816 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); | |
5817 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); | |
5818 | I915_WRITE(DEIIR, DE_PCU_EVENT); | |
5819 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); | |
5820 | ||
5821 | /* Go back to the starting frequency */ | |
91d14251 | 5822 | ironlake_set_drps(dev_priv, dev_priv->ips.fstart); |
dd92d8de | 5823 | mdelay(1); |
2b4e57bd ED |
5824 | rgvswctl |= MEMCTL_CMD_STS; |
5825 | I915_WRITE(MEMSWCTL, rgvswctl); | |
dd92d8de | 5826 | mdelay(1); |
2b4e57bd | 5827 | |
9270388e | 5828 | spin_unlock_irq(&mchdev_lock); |
2b4e57bd ED |
5829 | } |
5830 | ||
acbe9475 DV |
5831 | /* There's a funny hw issue where the hw returns all 0 when reading from |
5832 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value | |
5833 | * ourselves, instead of doing a rmw cycle (which might result in us clearing | |
5834 | * all limits and the gpu stuck at whatever frequency it is at atm). | |
5835 | */ | |
74ef1173 | 5836 | static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val) |
2b4e57bd | 5837 | { |
7b9e0ae6 | 5838 | u32 limits; |
2b4e57bd | 5839 | |
20b46e59 DV |
5840 | /* Only set the down limit when we've reached the lowest level to avoid |
5841 | * getting more interrupts, otherwise leave this clear. This prevents a | |
5842 | * race in the hw when coming out of rc6: There's a tiny window where | |
5843 | * the hw runs at the minimal clock before selecting the desired | |
5844 | * frequency, if the down threshold expires in that window we will not | |
5845 | * receive a down interrupt. */ | |
2d1fe073 | 5846 | if (IS_GEN9(dev_priv)) { |
74ef1173 AG |
5847 | limits = (dev_priv->rps.max_freq_softlimit) << 23; |
5848 | if (val <= dev_priv->rps.min_freq_softlimit) | |
5849 | limits |= (dev_priv->rps.min_freq_softlimit) << 14; | |
5850 | } else { | |
5851 | limits = dev_priv->rps.max_freq_softlimit << 24; | |
5852 | if (val <= dev_priv->rps.min_freq_softlimit) | |
5853 | limits |= dev_priv->rps.min_freq_softlimit << 16; | |
5854 | } | |
20b46e59 DV |
5855 | |
5856 | return limits; | |
5857 | } | |
5858 | ||
dd75fdc8 CW |
5859 | static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val) |
5860 | { | |
5861 | int new_power; | |
8a586437 AG |
5862 | u32 threshold_up = 0, threshold_down = 0; /* in % */ |
5863 | u32 ei_up = 0, ei_down = 0; | |
dd75fdc8 CW |
5864 | |
5865 | new_power = dev_priv->rps.power; | |
5866 | switch (dev_priv->rps.power) { | |
5867 | case LOW_POWER: | |
a72b5623 CW |
5868 | if (val > dev_priv->rps.efficient_freq + 1 && |
5869 | val > dev_priv->rps.cur_freq) | |
dd75fdc8 CW |
5870 | new_power = BETWEEN; |
5871 | break; | |
5872 | ||
5873 | case BETWEEN: | |
a72b5623 CW |
5874 | if (val <= dev_priv->rps.efficient_freq && |
5875 | val < dev_priv->rps.cur_freq) | |
dd75fdc8 | 5876 | new_power = LOW_POWER; |
a72b5623 CW |
5877 | else if (val >= dev_priv->rps.rp0_freq && |
5878 | val > dev_priv->rps.cur_freq) | |
dd75fdc8 CW |
5879 | new_power = HIGH_POWER; |
5880 | break; | |
5881 | ||
5882 | case HIGH_POWER: | |
a72b5623 CW |
5883 | if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && |
5884 | val < dev_priv->rps.cur_freq) | |
dd75fdc8 CW |
5885 | new_power = BETWEEN; |
5886 | break; | |
5887 | } | |
5888 | /* Max/min bins are special */ | |
aed242ff | 5889 | if (val <= dev_priv->rps.min_freq_softlimit) |
dd75fdc8 | 5890 | new_power = LOW_POWER; |
aed242ff | 5891 | if (val >= dev_priv->rps.max_freq_softlimit) |
dd75fdc8 CW |
5892 | new_power = HIGH_POWER; |
5893 | if (new_power == dev_priv->rps.power) | |
5894 | return; | |
5895 | ||
5896 | /* Note the units here are not exactly 1us, but 1280ns. */ | |
5897 | switch (new_power) { | |
5898 | case LOW_POWER: | |
5899 | /* Upclock if more than 95% busy over 16ms */ | |
8a586437 AG |
5900 | ei_up = 16000; |
5901 | threshold_up = 95; | |
dd75fdc8 CW |
5902 | |
5903 | /* Downclock if less than 85% busy over 32ms */ | |
8a586437 AG |
5904 | ei_down = 32000; |
5905 | threshold_down = 85; | |
dd75fdc8 CW |
5906 | break; |
5907 | ||
5908 | case BETWEEN: | |
5909 | /* Upclock if more than 90% busy over 13ms */ | |
8a586437 AG |
5910 | ei_up = 13000; |
5911 | threshold_up = 90; | |
dd75fdc8 CW |
5912 | |
5913 | /* Downclock if less than 75% busy over 32ms */ | |
8a586437 AG |
5914 | ei_down = 32000; |
5915 | threshold_down = 75; | |
dd75fdc8 CW |
5916 | break; |
5917 | ||
5918 | case HIGH_POWER: | |
5919 | /* Upclock if more than 85% busy over 10ms */ | |
8a586437 AG |
5920 | ei_up = 10000; |
5921 | threshold_up = 85; | |
dd75fdc8 CW |
5922 | |
5923 | /* Downclock if less than 60% busy over 32ms */ | |
8a586437 AG |
5924 | ei_down = 32000; |
5925 | threshold_down = 60; | |
dd75fdc8 CW |
5926 | break; |
5927 | } | |
5928 | ||
6067a27d MK |
5929 | /* When byt can survive without system hang with dynamic |
5930 | * sw freq adjustments, this restriction can be lifted. | |
5931 | */ | |
5932 | if (IS_VALLEYVIEW(dev_priv)) | |
5933 | goto skip_hw_write; | |
5934 | ||
8a586437 | 5935 | I915_WRITE(GEN6_RP_UP_EI, |
a72b5623 | 5936 | GT_INTERVAL_FROM_US(dev_priv, ei_up)); |
8a586437 | 5937 | I915_WRITE(GEN6_RP_UP_THRESHOLD, |
a72b5623 CW |
5938 | GT_INTERVAL_FROM_US(dev_priv, |
5939 | ei_up * threshold_up / 100)); | |
8a586437 AG |
5940 | |
5941 | I915_WRITE(GEN6_RP_DOWN_EI, | |
a72b5623 | 5942 | GT_INTERVAL_FROM_US(dev_priv, ei_down)); |
8a586437 | 5943 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, |
a72b5623 CW |
5944 | GT_INTERVAL_FROM_US(dev_priv, |
5945 | ei_down * threshold_down / 100)); | |
5946 | ||
5947 | I915_WRITE(GEN6_RP_CONTROL, | |
5948 | GEN6_RP_MEDIA_TURBO | | |
5949 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
5950 | GEN6_RP_MEDIA_IS_GFX | | |
5951 | GEN6_RP_ENABLE | | |
5952 | GEN6_RP_UP_BUSY_AVG | | |
5953 | GEN6_RP_DOWN_IDLE_AVG); | |
8a586437 | 5954 | |
6067a27d | 5955 | skip_hw_write: |
dd75fdc8 | 5956 | dev_priv->rps.power = new_power; |
8fb55197 CW |
5957 | dev_priv->rps.up_threshold = threshold_up; |
5958 | dev_priv->rps.down_threshold = threshold_down; | |
dd75fdc8 CW |
5959 | dev_priv->rps.last_adj = 0; |
5960 | } | |
5961 | ||
2876ce73 CW |
5962 | static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val) |
5963 | { | |
5964 | u32 mask = 0; | |
5965 | ||
e0e8c7cb | 5966 | /* We use UP_EI_EXPIRED interupts for both up/down in manual mode */ |
2876ce73 | 5967 | if (val > dev_priv->rps.min_freq_softlimit) |
e0e8c7cb | 5968 | mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT; |
2876ce73 | 5969 | if (val < dev_priv->rps.max_freq_softlimit) |
6f4b12f8 | 5970 | mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD; |
2876ce73 | 5971 | |
7b3c29f6 CW |
5972 | mask &= dev_priv->pm_rps_events; |
5973 | ||
59d02a1f | 5974 | return gen6_sanitize_rps_pm_mask(dev_priv, ~mask); |
2876ce73 CW |
5975 | } |
5976 | ||
b8a5ff8d JM |
5977 | /* gen6_set_rps is called to update the frequency request, but should also be |
5978 | * called when the range (min_delay and max_delay) is modified so that we can | |
5979 | * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */ | |
9fcee2f7 | 5980 | static int gen6_set_rps(struct drm_i915_private *dev_priv, u8 val) |
20b46e59 | 5981 | { |
eb64cad1 CW |
5982 | /* min/max delay may still have been modified so be sure to |
5983 | * write the limits value. | |
5984 | */ | |
5985 | if (val != dev_priv->rps.cur_freq) { | |
5986 | gen6_set_rps_thresholds(dev_priv, val); | |
b8a5ff8d | 5987 | |
dc97997a | 5988 | if (IS_GEN9(dev_priv)) |
5704195c AG |
5989 | I915_WRITE(GEN6_RPNSWREQ, |
5990 | GEN9_FREQUENCY(val)); | |
dc97997a | 5991 | else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
eb64cad1 CW |
5992 | I915_WRITE(GEN6_RPNSWREQ, |
5993 | HSW_FREQUENCY(val)); | |
5994 | else | |
5995 | I915_WRITE(GEN6_RPNSWREQ, | |
5996 | GEN6_FREQUENCY(val) | | |
5997 | GEN6_OFFSET(0) | | |
5998 | GEN6_AGGRESSIVE_TURBO); | |
b8a5ff8d | 5999 | } |
7b9e0ae6 | 6000 | |
7b9e0ae6 CW |
6001 | /* Make sure we continue to get interrupts |
6002 | * until we hit the minimum or maximum frequencies. | |
6003 | */ | |
74ef1173 | 6004 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val)); |
2876ce73 | 6005 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
7b9e0ae6 | 6006 | |
b39fb297 | 6007 | dev_priv->rps.cur_freq = val; |
0f94592e | 6008 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); |
9fcee2f7 CW |
6009 | |
6010 | return 0; | |
2b4e57bd ED |
6011 | } |
6012 | ||
9fcee2f7 | 6013 | static int valleyview_set_rps(struct drm_i915_private *dev_priv, u8 val) |
ffe02b40 | 6014 | { |
9fcee2f7 CW |
6015 | int err; |
6016 | ||
dc97997a | 6017 | if (WARN_ONCE(IS_CHERRYVIEW(dev_priv) && (val & 1), |
ffe02b40 VS |
6018 | "Odd GPU freq value\n")) |
6019 | val &= ~1; | |
6020 | ||
cd25dd5b D |
6021 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
6022 | ||
8fb55197 | 6023 | if (val != dev_priv->rps.cur_freq) { |
9fcee2f7 CW |
6024 | err = vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val); |
6025 | if (err) | |
6026 | return err; | |
6027 | ||
db4c5e0b | 6028 | gen6_set_rps_thresholds(dev_priv, val); |
8fb55197 | 6029 | } |
ffe02b40 | 6030 | |
ffe02b40 VS |
6031 | dev_priv->rps.cur_freq = val; |
6032 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); | |
9fcee2f7 CW |
6033 | |
6034 | return 0; | |
ffe02b40 VS |
6035 | } |
6036 | ||
a7f6e231 | 6037 | /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down |
76c3552f D |
6038 | * |
6039 | * * If Gfx is Idle, then | |
a7f6e231 D |
6040 | * 1. Forcewake Media well. |
6041 | * 2. Request idle freq. | |
6042 | * 3. Release Forcewake of Media well. | |
76c3552f D |
6043 | */ |
6044 | static void vlv_set_rps_idle(struct drm_i915_private *dev_priv) | |
6045 | { | |
aed242ff | 6046 | u32 val = dev_priv->rps.idle_freq; |
9fcee2f7 | 6047 | int err; |
5549d25f | 6048 | |
aed242ff | 6049 | if (dev_priv->rps.cur_freq <= val) |
76c3552f D |
6050 | return; |
6051 | ||
c9efef7b CW |
6052 | /* The punit delays the write of the frequency and voltage until it |
6053 | * determines the GPU is awake. During normal usage we don't want to | |
6054 | * waste power changing the frequency if the GPU is sleeping (rc6). | |
6055 | * However, the GPU and driver is now idle and we do not want to delay | |
6056 | * switching to minimum voltage (reducing power whilst idle) as we do | |
6057 | * not expect to be woken in the near future and so must flush the | |
6058 | * change by waking the device. | |
6059 | * | |
6060 | * We choose to take the media powerwell (either would do to trick the | |
6061 | * punit into committing the voltage change) as that takes a lot less | |
6062 | * power than the render powerwell. | |
6063 | */ | |
a7f6e231 | 6064 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA); |
9fcee2f7 | 6065 | err = valleyview_set_rps(dev_priv, val); |
a7f6e231 | 6066 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA); |
9fcee2f7 CW |
6067 | |
6068 | if (err) | |
6069 | DRM_ERROR("Failed to set RPS for idle\n"); | |
76c3552f D |
6070 | } |
6071 | ||
43cf3bf0 CW |
6072 | void gen6_rps_busy(struct drm_i915_private *dev_priv) |
6073 | { | |
6074 | mutex_lock(&dev_priv->rps.hw_lock); | |
6075 | if (dev_priv->rps.enabled) { | |
bd64818d CW |
6076 | u8 freq; |
6077 | ||
e0e8c7cb | 6078 | if (dev_priv->pm_rps_events & GEN6_PM_RP_UP_EI_EXPIRED) |
43cf3bf0 CW |
6079 | gen6_rps_reset_ei(dev_priv); |
6080 | I915_WRITE(GEN6_PMINTRMSK, | |
6081 | gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq)); | |
2b83c4c4 | 6082 | |
c33d247d CW |
6083 | gen6_enable_rps_interrupts(dev_priv); |
6084 | ||
bd64818d CW |
6085 | /* Use the user's desired frequency as a guide, but for better |
6086 | * performance, jump directly to RPe as our starting frequency. | |
6087 | */ | |
6088 | freq = max(dev_priv->rps.cur_freq, | |
6089 | dev_priv->rps.efficient_freq); | |
6090 | ||
9fcee2f7 | 6091 | if (intel_set_rps(dev_priv, |
bd64818d | 6092 | clamp(freq, |
9fcee2f7 CW |
6093 | dev_priv->rps.min_freq_softlimit, |
6094 | dev_priv->rps.max_freq_softlimit))) | |
6095 | DRM_DEBUG_DRIVER("Failed to set idle frequency\n"); | |
43cf3bf0 CW |
6096 | } |
6097 | mutex_unlock(&dev_priv->rps.hw_lock); | |
6098 | } | |
6099 | ||
b29c19b6 CW |
6100 | void gen6_rps_idle(struct drm_i915_private *dev_priv) |
6101 | { | |
c33d247d CW |
6102 | /* Flush our bottom-half so that it does not race with us |
6103 | * setting the idle frequency and so that it is bounded by | |
6104 | * our rpm wakeref. And then disable the interrupts to stop any | |
6105 | * futher RPS reclocking whilst we are asleep. | |
6106 | */ | |
6107 | gen6_disable_rps_interrupts(dev_priv); | |
6108 | ||
b29c19b6 | 6109 | mutex_lock(&dev_priv->rps.hw_lock); |
c0951f0c | 6110 | if (dev_priv->rps.enabled) { |
dc97997a | 6111 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
76c3552f | 6112 | vlv_set_rps_idle(dev_priv); |
7526ed79 | 6113 | else |
dc97997a | 6114 | gen6_set_rps(dev_priv, dev_priv->rps.idle_freq); |
c0951f0c | 6115 | dev_priv->rps.last_adj = 0; |
12c100bf VS |
6116 | I915_WRITE(GEN6_PMINTRMSK, |
6117 | gen6_sanitize_rps_pm_mask(dev_priv, ~0)); | |
c0951f0c | 6118 | } |
8d3afd7d | 6119 | mutex_unlock(&dev_priv->rps.hw_lock); |
1854d5ca | 6120 | |
8d3afd7d | 6121 | spin_lock(&dev_priv->rps.client_lock); |
1854d5ca CW |
6122 | while (!list_empty(&dev_priv->rps.clients)) |
6123 | list_del_init(dev_priv->rps.clients.next); | |
8d3afd7d | 6124 | spin_unlock(&dev_priv->rps.client_lock); |
b29c19b6 CW |
6125 | } |
6126 | ||
1854d5ca | 6127 | void gen6_rps_boost(struct drm_i915_private *dev_priv, |
e61b9958 CW |
6128 | struct intel_rps_client *rps, |
6129 | unsigned long submitted) | |
b29c19b6 | 6130 | { |
8d3afd7d CW |
6131 | /* This is intentionally racy! We peek at the state here, then |
6132 | * validate inside the RPS worker. | |
6133 | */ | |
67d97da3 | 6134 | if (!(dev_priv->gt.awake && |
8d3afd7d | 6135 | dev_priv->rps.enabled && |
29ecd78d | 6136 | dev_priv->rps.cur_freq < dev_priv->rps.boost_freq)) |
8d3afd7d | 6137 | return; |
43cf3bf0 | 6138 | |
e61b9958 CW |
6139 | /* Force a RPS boost (and don't count it against the client) if |
6140 | * the GPU is severely congested. | |
6141 | */ | |
d0bc54f2 | 6142 | if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES)) |
e61b9958 CW |
6143 | rps = NULL; |
6144 | ||
8d3afd7d CW |
6145 | spin_lock(&dev_priv->rps.client_lock); |
6146 | if (rps == NULL || list_empty(&rps->link)) { | |
6147 | spin_lock_irq(&dev_priv->irq_lock); | |
6148 | if (dev_priv->rps.interrupts_enabled) { | |
6149 | dev_priv->rps.client_boost = true; | |
c33d247d | 6150 | schedule_work(&dev_priv->rps.work); |
8d3afd7d CW |
6151 | } |
6152 | spin_unlock_irq(&dev_priv->irq_lock); | |
1854d5ca | 6153 | |
2e1b8730 CW |
6154 | if (rps != NULL) { |
6155 | list_add(&rps->link, &dev_priv->rps.clients); | |
6156 | rps->boosts++; | |
1854d5ca CW |
6157 | } else |
6158 | dev_priv->rps.boosts++; | |
c0951f0c | 6159 | } |
8d3afd7d | 6160 | spin_unlock(&dev_priv->rps.client_lock); |
b29c19b6 CW |
6161 | } |
6162 | ||
9fcee2f7 | 6163 | int intel_set_rps(struct drm_i915_private *dev_priv, u8 val) |
0a073b84 | 6164 | { |
9fcee2f7 CW |
6165 | int err; |
6166 | ||
cfd1c488 CW |
6167 | lockdep_assert_held(&dev_priv->rps.hw_lock); |
6168 | GEM_BUG_ON(val > dev_priv->rps.max_freq); | |
6169 | GEM_BUG_ON(val < dev_priv->rps.min_freq); | |
6170 | ||
76e4e4b5 CW |
6171 | if (!dev_priv->rps.enabled) { |
6172 | dev_priv->rps.cur_freq = val; | |
6173 | return 0; | |
6174 | } | |
6175 | ||
dc97997a | 6176 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
9fcee2f7 | 6177 | err = valleyview_set_rps(dev_priv, val); |
ffe02b40 | 6178 | else |
9fcee2f7 CW |
6179 | err = gen6_set_rps(dev_priv, val); |
6180 | ||
6181 | return err; | |
0a073b84 JB |
6182 | } |
6183 | ||
dc97997a | 6184 | static void gen9_disable_rc6(struct drm_i915_private *dev_priv) |
20e49366 | 6185 | { |
20e49366 | 6186 | I915_WRITE(GEN6_RC_CONTROL, 0); |
38c23527 | 6187 | I915_WRITE(GEN9_PG_ENABLE, 0); |
20e49366 ZW |
6188 | } |
6189 | ||
dc97997a | 6190 | static void gen9_disable_rps(struct drm_i915_private *dev_priv) |
2030d684 | 6191 | { |
2030d684 AG |
6192 | I915_WRITE(GEN6_RP_CONTROL, 0); |
6193 | } | |
6194 | ||
dc97997a | 6195 | static void gen6_disable_rps(struct drm_i915_private *dev_priv) |
d20d4f0c | 6196 | { |
d20d4f0c | 6197 | I915_WRITE(GEN6_RC_CONTROL, 0); |
44fc7d5c | 6198 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
2030d684 | 6199 | I915_WRITE(GEN6_RP_CONTROL, 0); |
44fc7d5c DV |
6200 | } |
6201 | ||
dc97997a | 6202 | static void cherryview_disable_rps(struct drm_i915_private *dev_priv) |
38807746 | 6203 | { |
38807746 D |
6204 | I915_WRITE(GEN6_RC_CONTROL, 0); |
6205 | } | |
6206 | ||
dc97997a | 6207 | static void valleyview_disable_rps(struct drm_i915_private *dev_priv) |
44fc7d5c | 6208 | { |
98a2e5f9 D |
6209 | /* we're doing forcewake before Disabling RC6, |
6210 | * This what the BIOS expects when going into suspend */ | |
59bad947 | 6211 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
98a2e5f9 | 6212 | |
44fc7d5c | 6213 | I915_WRITE(GEN6_RC_CONTROL, 0); |
d20d4f0c | 6214 | |
59bad947 | 6215 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
d20d4f0c JB |
6216 | } |
6217 | ||
dc97997a | 6218 | static void intel_print_rc6_info(struct drm_i915_private *dev_priv, u32 mode) |
dc39fff7 | 6219 | { |
dc97997a | 6220 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { |
91ca689a ID |
6221 | if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1))) |
6222 | mode = GEN6_RC_CTL_RC6_ENABLE; | |
6223 | else | |
6224 | mode = 0; | |
6225 | } | |
dc97997a | 6226 | if (HAS_RC6p(dev_priv)) |
b99d49cc ID |
6227 | DRM_DEBUG_DRIVER("Enabling RC6 states: " |
6228 | "RC6 %s RC6p %s RC6pp %s\n", | |
6229 | onoff(mode & GEN6_RC_CTL_RC6_ENABLE), | |
6230 | onoff(mode & GEN6_RC_CTL_RC6p_ENABLE), | |
6231 | onoff(mode & GEN6_RC_CTL_RC6pp_ENABLE)); | |
58abf1da RV |
6232 | |
6233 | else | |
b99d49cc ID |
6234 | DRM_DEBUG_DRIVER("Enabling RC6 states: RC6 %s\n", |
6235 | onoff(mode & GEN6_RC_CTL_RC6_ENABLE)); | |
dc39fff7 BW |
6236 | } |
6237 | ||
dc97997a | 6238 | static bool bxt_check_bios_rc6_setup(struct drm_i915_private *dev_priv) |
274008e8 | 6239 | { |
72e96d64 | 6240 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
274008e8 SAK |
6241 | bool enable_rc6 = true; |
6242 | unsigned long rc6_ctx_base; | |
fc619841 ID |
6243 | u32 rc_ctl; |
6244 | int rc_sw_target; | |
6245 | ||
6246 | rc_ctl = I915_READ(GEN6_RC_CONTROL); | |
6247 | rc_sw_target = (I915_READ(GEN6_RC_STATE) & RC_SW_TARGET_STATE_MASK) >> | |
6248 | RC_SW_TARGET_STATE_SHIFT; | |
6249 | DRM_DEBUG_DRIVER("BIOS enabled RC states: " | |
6250 | "HW_CTRL %s HW_RC6 %s SW_TARGET_STATE %x\n", | |
6251 | onoff(rc_ctl & GEN6_RC_CTL_HW_ENABLE), | |
6252 | onoff(rc_ctl & GEN6_RC_CTL_RC6_ENABLE), | |
6253 | rc_sw_target); | |
274008e8 SAK |
6254 | |
6255 | if (!(I915_READ(RC6_LOCATION) & RC6_CTX_IN_DRAM)) { | |
b99d49cc | 6256 | DRM_DEBUG_DRIVER("RC6 Base location not set properly.\n"); |
274008e8 SAK |
6257 | enable_rc6 = false; |
6258 | } | |
6259 | ||
6260 | /* | |
6261 | * The exact context size is not known for BXT, so assume a page size | |
6262 | * for this check. | |
6263 | */ | |
6264 | rc6_ctx_base = I915_READ(RC6_CTX_BASE) & RC6_CTX_BASE_MASK; | |
72e96d64 JL |
6265 | if (!((rc6_ctx_base >= ggtt->stolen_reserved_base) && |
6266 | (rc6_ctx_base + PAGE_SIZE <= ggtt->stolen_reserved_base + | |
6267 | ggtt->stolen_reserved_size))) { | |
b99d49cc | 6268 | DRM_DEBUG_DRIVER("RC6 Base address not as expected.\n"); |
274008e8 SAK |
6269 | enable_rc6 = false; |
6270 | } | |
6271 | ||
6272 | if (!(((I915_READ(PWRCTX_MAXCNT_RCSUNIT) & IDLE_TIME_MASK) > 1) && | |
6273 | ((I915_READ(PWRCTX_MAXCNT_VCSUNIT0) & IDLE_TIME_MASK) > 1) && | |
6274 | ((I915_READ(PWRCTX_MAXCNT_BCSUNIT) & IDLE_TIME_MASK) > 1) && | |
6275 | ((I915_READ(PWRCTX_MAXCNT_VECSUNIT) & IDLE_TIME_MASK) > 1))) { | |
b99d49cc | 6276 | DRM_DEBUG_DRIVER("Engine Idle wait time not set properly.\n"); |
274008e8 SAK |
6277 | enable_rc6 = false; |
6278 | } | |
6279 | ||
fc619841 ID |
6280 | if (!I915_READ(GEN8_PUSHBUS_CONTROL) || |
6281 | !I915_READ(GEN8_PUSHBUS_ENABLE) || | |
6282 | !I915_READ(GEN8_PUSHBUS_SHIFT)) { | |
6283 | DRM_DEBUG_DRIVER("Pushbus not setup properly.\n"); | |
6284 | enable_rc6 = false; | |
6285 | } | |
6286 | ||
6287 | if (!I915_READ(GEN6_GFXPAUSE)) { | |
6288 | DRM_DEBUG_DRIVER("GFX pause not setup properly.\n"); | |
6289 | enable_rc6 = false; | |
6290 | } | |
6291 | ||
6292 | if (!I915_READ(GEN8_MISC_CTRL0)) { | |
6293 | DRM_DEBUG_DRIVER("GPM control not setup properly.\n"); | |
274008e8 SAK |
6294 | enable_rc6 = false; |
6295 | } | |
6296 | ||
6297 | return enable_rc6; | |
6298 | } | |
6299 | ||
dc97997a | 6300 | int sanitize_rc6_option(struct drm_i915_private *dev_priv, int enable_rc6) |
2b4e57bd | 6301 | { |
e7d66d89 | 6302 | /* No RC6 before Ironlake and code is gone for ilk. */ |
dc97997a | 6303 | if (INTEL_INFO(dev_priv)->gen < 6) |
e6069ca8 ID |
6304 | return 0; |
6305 | ||
274008e8 SAK |
6306 | if (!enable_rc6) |
6307 | return 0; | |
6308 | ||
cc3f90f0 | 6309 | if (IS_GEN9_LP(dev_priv) && !bxt_check_bios_rc6_setup(dev_priv)) { |
274008e8 SAK |
6310 | DRM_INFO("RC6 disabled by BIOS\n"); |
6311 | return 0; | |
6312 | } | |
6313 | ||
456470eb | 6314 | /* Respect the kernel parameter if it is set */ |
e6069ca8 ID |
6315 | if (enable_rc6 >= 0) { |
6316 | int mask; | |
6317 | ||
dc97997a | 6318 | if (HAS_RC6p(dev_priv)) |
e6069ca8 ID |
6319 | mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE | |
6320 | INTEL_RC6pp_ENABLE; | |
6321 | else | |
6322 | mask = INTEL_RC6_ENABLE; | |
6323 | ||
6324 | if ((enable_rc6 & mask) != enable_rc6) | |
b99d49cc ID |
6325 | DRM_DEBUG_DRIVER("Adjusting RC6 mask to %d " |
6326 | "(requested %d, valid %d)\n", | |
6327 | enable_rc6 & mask, enable_rc6, mask); | |
e6069ca8 ID |
6328 | |
6329 | return enable_rc6 & mask; | |
6330 | } | |
2b4e57bd | 6331 | |
dc97997a | 6332 | if (IS_IVYBRIDGE(dev_priv)) |
cca84a1f | 6333 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); |
8bade1ad BW |
6334 | |
6335 | return INTEL_RC6_ENABLE; | |
2b4e57bd ED |
6336 | } |
6337 | ||
dc97997a | 6338 | static void gen6_init_rps_frequencies(struct drm_i915_private *dev_priv) |
3280e8b0 BW |
6339 | { |
6340 | /* All of these values are in units of 50MHz */ | |
773ea9a8 | 6341 | |
93ee2920 | 6342 | /* static values from HW: RP0 > RP1 > RPn (min_freq) */ |
cc3f90f0 | 6343 | if (IS_GEN9_LP(dev_priv)) { |
773ea9a8 | 6344 | u32 rp_state_cap = I915_READ(BXT_RP_STATE_CAP); |
35040562 BP |
6345 | dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff; |
6346 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; | |
6347 | dev_priv->rps.min_freq = (rp_state_cap >> 0) & 0xff; | |
6348 | } else { | |
773ea9a8 | 6349 | u32 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); |
35040562 BP |
6350 | dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff; |
6351 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; | |
6352 | dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff; | |
6353 | } | |
3280e8b0 | 6354 | /* hw_max = RP0 until we check for overclocking */ |
773ea9a8 | 6355 | dev_priv->rps.max_freq = dev_priv->rps.rp0_freq; |
3280e8b0 | 6356 | |
93ee2920 | 6357 | dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq; |
dc97997a | 6358 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv) || |
b976dc53 | 6359 | IS_GEN9_BC(dev_priv)) { |
773ea9a8 CW |
6360 | u32 ddcc_status = 0; |
6361 | ||
6362 | if (sandybridge_pcode_read(dev_priv, | |
6363 | HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL, | |
6364 | &ddcc_status) == 0) | |
93ee2920 | 6365 | dev_priv->rps.efficient_freq = |
46efa4ab TR |
6366 | clamp_t(u8, |
6367 | ((ddcc_status >> 8) & 0xff), | |
6368 | dev_priv->rps.min_freq, | |
6369 | dev_priv->rps.max_freq); | |
93ee2920 TR |
6370 | } |
6371 | ||
b976dc53 | 6372 | if (IS_GEN9_BC(dev_priv)) { |
c5e0688c | 6373 | /* Store the frequency values in 16.66 MHZ units, which is |
773ea9a8 CW |
6374 | * the natural hardware unit for SKL |
6375 | */ | |
c5e0688c AG |
6376 | dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER; |
6377 | dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER; | |
6378 | dev_priv->rps.min_freq *= GEN9_FREQ_SCALER; | |
6379 | dev_priv->rps.max_freq *= GEN9_FREQ_SCALER; | |
6380 | dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER; | |
6381 | } | |
3280e8b0 BW |
6382 | } |
6383 | ||
3a45b05c | 6384 | static void reset_rps(struct drm_i915_private *dev_priv, |
9fcee2f7 | 6385 | int (*set)(struct drm_i915_private *, u8)) |
3a45b05c CW |
6386 | { |
6387 | u8 freq = dev_priv->rps.cur_freq; | |
6388 | ||
6389 | /* force a reset */ | |
6390 | dev_priv->rps.power = -1; | |
6391 | dev_priv->rps.cur_freq = -1; | |
6392 | ||
9fcee2f7 CW |
6393 | if (set(dev_priv, freq)) |
6394 | DRM_ERROR("Failed to reset RPS to initial values\n"); | |
3a45b05c CW |
6395 | } |
6396 | ||
b6fef0ef | 6397 | /* See the Gen9_GT_PM_Programming_Guide doc for the below */ |
dc97997a | 6398 | static void gen9_enable_rps(struct drm_i915_private *dev_priv) |
b6fef0ef | 6399 | { |
b6fef0ef JB |
6400 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
6401 | ||
0beb059a AG |
6402 | /* Program defaults and thresholds for RPS*/ |
6403 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
6404 | GEN9_FREQUENCY(dev_priv->rps.rp1_freq)); | |
6405 | ||
6406 | /* 1 second timeout*/ | |
6407 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, | |
6408 | GT_INTERVAL_FROM_US(dev_priv, 1000000)); | |
6409 | ||
b6fef0ef | 6410 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa); |
b6fef0ef | 6411 | |
0beb059a AG |
6412 | /* Leaning on the below call to gen6_set_rps to program/setup the |
6413 | * Up/Down EI & threshold registers, as well as the RP_CONTROL, | |
6414 | * RP_INTERRUPT_LIMITS & RPNSWREQ registers */ | |
3a45b05c | 6415 | reset_rps(dev_priv, gen6_set_rps); |
b6fef0ef JB |
6416 | |
6417 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); | |
6418 | } | |
6419 | ||
dc97997a | 6420 | static void gen9_enable_rc6(struct drm_i915_private *dev_priv) |
20e49366 | 6421 | { |
e2f80391 | 6422 | struct intel_engine_cs *engine; |
3b3f1650 | 6423 | enum intel_engine_id id; |
20e49366 | 6424 | uint32_t rc6_mask = 0; |
20e49366 ZW |
6425 | |
6426 | /* 1a: Software RC state - RC0 */ | |
6427 | I915_WRITE(GEN6_RC_STATE, 0); | |
6428 | ||
6429 | /* 1b: Get forcewake during program sequence. Although the driver | |
6430 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 6431 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
20e49366 ZW |
6432 | |
6433 | /* 2a: Disable RC states. */ | |
6434 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
6435 | ||
6436 | /* 2b: Program RC6 thresholds.*/ | |
63a4dec2 SAK |
6437 | |
6438 | /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */ | |
dc97997a | 6439 | if (IS_SKYLAKE(dev_priv)) |
63a4dec2 SAK |
6440 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16); |
6441 | else | |
6442 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16); | |
20e49366 ZW |
6443 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
6444 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
3b3f1650 | 6445 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 6446 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
97c322e7 | 6447 | |
1a3d1898 | 6448 | if (HAS_GUC(dev_priv)) |
97c322e7 SAK |
6449 | I915_WRITE(GUC_MAX_IDLE_COUNT, 0xA); |
6450 | ||
20e49366 | 6451 | I915_WRITE(GEN6_RC_SLEEP, 0); |
20e49366 | 6452 | |
38c23527 ZW |
6453 | /* 2c: Program Coarse Power Gating Policies. */ |
6454 | I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25); | |
6455 | I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25); | |
6456 | ||
20e49366 | 6457 | /* 3a: Enable RC6 */ |
dc97997a | 6458 | if (intel_enable_rc6() & INTEL_RC6_ENABLE) |
20e49366 | 6459 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
87ad3212 | 6460 | DRM_INFO("RC6 %s\n", onoff(rc6_mask & GEN6_RC_CTL_RC6_ENABLE)); |
1c044f9b CW |
6461 | I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */ |
6462 | I915_WRITE(GEN6_RC_CONTROL, | |
6463 | GEN6_RC_CTL_HW_ENABLE | GEN6_RC_CTL_EI_MODE(1) | rc6_mask); | |
20e49366 | 6464 | |
cb07bae0 SK |
6465 | /* |
6466 | * 3b: Enable Coarse Power Gating only when RC6 is enabled. | |
f2d2fe95 | 6467 | * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6. |
cb07bae0 | 6468 | */ |
dc97997a | 6469 | if (NEEDS_WaRsDisableCoarsePowerGating(dev_priv)) |
f2d2fe95 SAK |
6470 | I915_WRITE(GEN9_PG_ENABLE, 0); |
6471 | else | |
6472 | I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? | |
6473 | (GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE) : 0); | |
38c23527 | 6474 | |
59bad947 | 6475 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
20e49366 ZW |
6476 | } |
6477 | ||
dc97997a | 6478 | static void gen8_enable_rps(struct drm_i915_private *dev_priv) |
6edee7f3 | 6479 | { |
e2f80391 | 6480 | struct intel_engine_cs *engine; |
3b3f1650 | 6481 | enum intel_engine_id id; |
93ee2920 | 6482 | uint32_t rc6_mask = 0; |
6edee7f3 BW |
6483 | |
6484 | /* 1a: Software RC state - RC0 */ | |
6485 | I915_WRITE(GEN6_RC_STATE, 0); | |
6486 | ||
6487 | /* 1c & 1d: Get forcewake during program sequence. Although the driver | |
6488 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 6489 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
6edee7f3 BW |
6490 | |
6491 | /* 2a: Disable RC states. */ | |
6492 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
6493 | ||
6edee7f3 BW |
6494 | /* 2b: Program RC6 thresholds.*/ |
6495 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); | |
6496 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
6497 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
3b3f1650 | 6498 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 6499 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
6edee7f3 | 6500 | I915_WRITE(GEN6_RC_SLEEP, 0); |
dc97997a | 6501 | if (IS_BROADWELL(dev_priv)) |
0d68b25e TR |
6502 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */ |
6503 | else | |
6504 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */ | |
6edee7f3 BW |
6505 | |
6506 | /* 3: Enable RC6 */ | |
dc97997a | 6507 | if (intel_enable_rc6() & INTEL_RC6_ENABLE) |
6edee7f3 | 6508 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
dc97997a CW |
6509 | intel_print_rc6_info(dev_priv, rc6_mask); |
6510 | if (IS_BROADWELL(dev_priv)) | |
0d68b25e TR |
6511 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
6512 | GEN7_RC_CTL_TO_MODE | | |
6513 | rc6_mask); | |
6514 | else | |
6515 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | | |
6516 | GEN6_RC_CTL_EI_MODE(1) | | |
6517 | rc6_mask); | |
6edee7f3 BW |
6518 | |
6519 | /* 4 Program defaults and thresholds for RPS*/ | |
f9bdc585 BW |
6520 | I915_WRITE(GEN6_RPNSWREQ, |
6521 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); | |
6522 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
6523 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); | |
7526ed79 DV |
6524 | /* NB: Docs say 1s, and 1000000 - which aren't equivalent */ |
6525 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */ | |
6526 | ||
6527 | /* Docs recommend 900MHz, and 300 MHz respectively */ | |
6528 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, | |
6529 | dev_priv->rps.max_freq_softlimit << 24 | | |
6530 | dev_priv->rps.min_freq_softlimit << 16); | |
6531 | ||
6532 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */ | |
6533 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/ | |
6534 | I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */ | |
6535 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */ | |
6536 | ||
6537 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
6edee7f3 BW |
6538 | |
6539 | /* 5: Enable RPS */ | |
7526ed79 DV |
6540 | I915_WRITE(GEN6_RP_CONTROL, |
6541 | GEN6_RP_MEDIA_TURBO | | |
6542 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
6543 | GEN6_RP_MEDIA_IS_GFX | | |
6544 | GEN6_RP_ENABLE | | |
6545 | GEN6_RP_UP_BUSY_AVG | | |
6546 | GEN6_RP_DOWN_IDLE_AVG); | |
6547 | ||
6548 | /* 6: Ring frequency + overclocking (our driver does this later */ | |
6549 | ||
3a45b05c | 6550 | reset_rps(dev_priv, gen6_set_rps); |
7526ed79 | 6551 | |
59bad947 | 6552 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
6edee7f3 BW |
6553 | } |
6554 | ||
dc97997a | 6555 | static void gen6_enable_rps(struct drm_i915_private *dev_priv) |
2b4e57bd | 6556 | { |
e2f80391 | 6557 | struct intel_engine_cs *engine; |
3b3f1650 | 6558 | enum intel_engine_id id; |
99ac9612 | 6559 | u32 rc6vids, rc6_mask = 0; |
2b4e57bd | 6560 | u32 gtfifodbg; |
2b4e57bd | 6561 | int rc6_mode; |
b4ac5afc | 6562 | int ret; |
2b4e57bd | 6563 | |
4fc688ce | 6564 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 6565 | |
2b4e57bd ED |
6566 | /* Here begins a magic sequence of register writes to enable |
6567 | * auto-downclocking. | |
6568 | * | |
6569 | * Perhaps there might be some value in exposing these to | |
6570 | * userspace... | |
6571 | */ | |
6572 | I915_WRITE(GEN6_RC_STATE, 0); | |
2b4e57bd ED |
6573 | |
6574 | /* Clear the DBG now so we don't confuse earlier errors */ | |
297b32ec VS |
6575 | gtfifodbg = I915_READ(GTFIFODBG); |
6576 | if (gtfifodbg) { | |
2b4e57bd ED |
6577 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); |
6578 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
6579 | } | |
6580 | ||
59bad947 | 6581 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
2b4e57bd ED |
6582 | |
6583 | /* disable the counters and set deterministic thresholds */ | |
6584 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
6585 | ||
6586 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); | |
6587 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); | |
6588 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); | |
6589 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
6590 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
6591 | ||
3b3f1650 | 6592 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 6593 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
2b4e57bd ED |
6594 | |
6595 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
6596 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); | |
dc97997a | 6597 | if (IS_IVYBRIDGE(dev_priv)) |
351aa566 SM |
6598 | I915_WRITE(GEN6_RC6_THRESHOLD, 125000); |
6599 | else | |
6600 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); | |
0920a487 | 6601 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
2b4e57bd ED |
6602 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
6603 | ||
5a7dc92a | 6604 | /* Check if we are enabling RC6 */ |
dc97997a | 6605 | rc6_mode = intel_enable_rc6(); |
2b4e57bd ED |
6606 | if (rc6_mode & INTEL_RC6_ENABLE) |
6607 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; | |
6608 | ||
5a7dc92a | 6609 | /* We don't use those on Haswell */ |
dc97997a | 6610 | if (!IS_HASWELL(dev_priv)) { |
5a7dc92a ED |
6611 | if (rc6_mode & INTEL_RC6p_ENABLE) |
6612 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; | |
2b4e57bd | 6613 | |
5a7dc92a ED |
6614 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
6615 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; | |
6616 | } | |
2b4e57bd | 6617 | |
dc97997a | 6618 | intel_print_rc6_info(dev_priv, rc6_mask); |
2b4e57bd ED |
6619 | |
6620 | I915_WRITE(GEN6_RC_CONTROL, | |
6621 | rc6_mask | | |
6622 | GEN6_RC_CTL_EI_MODE(1) | | |
6623 | GEN6_RC_CTL_HW_ENABLE); | |
6624 | ||
dd75fdc8 CW |
6625 | /* Power down if completely idle for over 50ms */ |
6626 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000); | |
2b4e57bd | 6627 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
2b4e57bd | 6628 | |
3a45b05c | 6629 | reset_rps(dev_priv, gen6_set_rps); |
2b4e57bd | 6630 | |
31643d54 BW |
6631 | rc6vids = 0; |
6632 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); | |
dc97997a | 6633 | if (IS_GEN6(dev_priv) && ret) { |
31643d54 | 6634 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); |
dc97997a | 6635 | } else if (IS_GEN6(dev_priv) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { |
31643d54 BW |
6636 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", |
6637 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); | |
6638 | rc6vids &= 0xffff00; | |
6639 | rc6vids |= GEN6_ENCODE_RC6_VID(450); | |
6640 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); | |
6641 | if (ret) | |
6642 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); | |
6643 | } | |
6644 | ||
59bad947 | 6645 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
2b4e57bd ED |
6646 | } |
6647 | ||
fb7404e8 | 6648 | static void gen6_update_ring_freq(struct drm_i915_private *dev_priv) |
2b4e57bd ED |
6649 | { |
6650 | int min_freq = 15; | |
3ebecd07 CW |
6651 | unsigned int gpu_freq; |
6652 | unsigned int max_ia_freq, min_ring_freq; | |
4c8c7743 | 6653 | unsigned int max_gpu_freq, min_gpu_freq; |
2b4e57bd | 6654 | int scaling_factor = 180; |
eda79642 | 6655 | struct cpufreq_policy *policy; |
2b4e57bd | 6656 | |
4fc688ce | 6657 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 6658 | |
eda79642 BW |
6659 | policy = cpufreq_cpu_get(0); |
6660 | if (policy) { | |
6661 | max_ia_freq = policy->cpuinfo.max_freq; | |
6662 | cpufreq_cpu_put(policy); | |
6663 | } else { | |
6664 | /* | |
6665 | * Default to measured freq if none found, PCU will ensure we | |
6666 | * don't go over | |
6667 | */ | |
2b4e57bd | 6668 | max_ia_freq = tsc_khz; |
eda79642 | 6669 | } |
2b4e57bd ED |
6670 | |
6671 | /* Convert from kHz to MHz */ | |
6672 | max_ia_freq /= 1000; | |
6673 | ||
153b4b95 | 6674 | min_ring_freq = I915_READ(DCLK) & 0xf; |
f6aca45c BW |
6675 | /* convert DDR frequency from units of 266.6MHz to bandwidth */ |
6676 | min_ring_freq = mult_frac(min_ring_freq, 8, 3); | |
3ebecd07 | 6677 | |
b976dc53 | 6678 | if (IS_GEN9_BC(dev_priv)) { |
4c8c7743 AG |
6679 | /* Convert GT frequency to 50 HZ units */ |
6680 | min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER; | |
6681 | max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER; | |
6682 | } else { | |
6683 | min_gpu_freq = dev_priv->rps.min_freq; | |
6684 | max_gpu_freq = dev_priv->rps.max_freq; | |
6685 | } | |
6686 | ||
2b4e57bd ED |
6687 | /* |
6688 | * For each potential GPU frequency, load a ring frequency we'd like | |
6689 | * to use for memory access. We do this by specifying the IA frequency | |
6690 | * the PCU should use as a reference to determine the ring frequency. | |
6691 | */ | |
4c8c7743 AG |
6692 | for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) { |
6693 | int diff = max_gpu_freq - gpu_freq; | |
3ebecd07 CW |
6694 | unsigned int ia_freq = 0, ring_freq = 0; |
6695 | ||
b976dc53 | 6696 | if (IS_GEN9_BC(dev_priv)) { |
4c8c7743 AG |
6697 | /* |
6698 | * ring_freq = 2 * GT. ring_freq is in 100MHz units | |
6699 | * No floor required for ring frequency on SKL. | |
6700 | */ | |
6701 | ring_freq = gpu_freq; | |
dc97997a | 6702 | } else if (INTEL_INFO(dev_priv)->gen >= 8) { |
46c764d4 BW |
6703 | /* max(2 * GT, DDR). NB: GT is 50MHz units */ |
6704 | ring_freq = max(min_ring_freq, gpu_freq); | |
dc97997a | 6705 | } else if (IS_HASWELL(dev_priv)) { |
f6aca45c | 6706 | ring_freq = mult_frac(gpu_freq, 5, 4); |
3ebecd07 CW |
6707 | ring_freq = max(min_ring_freq, ring_freq); |
6708 | /* leave ia_freq as the default, chosen by cpufreq */ | |
6709 | } else { | |
6710 | /* On older processors, there is no separate ring | |
6711 | * clock domain, so in order to boost the bandwidth | |
6712 | * of the ring, we need to upclock the CPU (ia_freq). | |
6713 | * | |
6714 | * For GPU frequencies less than 750MHz, | |
6715 | * just use the lowest ring freq. | |
6716 | */ | |
6717 | if (gpu_freq < min_freq) | |
6718 | ia_freq = 800; | |
6719 | else | |
6720 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); | |
6721 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); | |
6722 | } | |
2b4e57bd | 6723 | |
42c0526c BW |
6724 | sandybridge_pcode_write(dev_priv, |
6725 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, | |
3ebecd07 CW |
6726 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
6727 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | | |
6728 | gpu_freq); | |
2b4e57bd | 6729 | } |
2b4e57bd ED |
6730 | } |
6731 | ||
03af2045 | 6732 | static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv) |
2b6b3a09 D |
6733 | { |
6734 | u32 val, rp0; | |
6735 | ||
5b5929cb | 6736 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); |
2b6b3a09 | 6737 | |
43b67998 | 6738 | switch (INTEL_INFO(dev_priv)->sseu.eu_total) { |
5b5929cb JN |
6739 | case 8: |
6740 | /* (2 * 4) config */ | |
6741 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT); | |
6742 | break; | |
6743 | case 12: | |
6744 | /* (2 * 6) config */ | |
6745 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT); | |
6746 | break; | |
6747 | case 16: | |
6748 | /* (2 * 8) config */ | |
6749 | default: | |
6750 | /* Setting (2 * 8) Min RP0 for any other combination */ | |
6751 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT); | |
6752 | break; | |
095acd5f | 6753 | } |
5b5929cb JN |
6754 | |
6755 | rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK); | |
6756 | ||
2b6b3a09 D |
6757 | return rp0; |
6758 | } | |
6759 | ||
6760 | static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
6761 | { | |
6762 | u32 val, rpe; | |
6763 | ||
6764 | val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG); | |
6765 | rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK; | |
6766 | ||
6767 | return rpe; | |
6768 | } | |
6769 | ||
7707df4a D |
6770 | static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv) |
6771 | { | |
6772 | u32 val, rp1; | |
6773 | ||
5b5929cb JN |
6774 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); |
6775 | rp1 = (val & FB_GFX_FREQ_FUSE_MASK); | |
6776 | ||
7707df4a D |
6777 | return rp1; |
6778 | } | |
6779 | ||
96676fe3 D |
6780 | static u32 cherryview_rps_min_freq(struct drm_i915_private *dev_priv) |
6781 | { | |
6782 | u32 val, rpn; | |
6783 | ||
6784 | val = vlv_punit_read(dev_priv, FB_GFX_FMIN_AT_VMIN_FUSE); | |
6785 | rpn = ((val >> FB_GFX_FMIN_AT_VMIN_FUSE_SHIFT) & | |
6786 | FB_GFX_FREQ_FUSE_MASK); | |
6787 | ||
6788 | return rpn; | |
6789 | } | |
6790 | ||
f8f2b001 D |
6791 | static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv) |
6792 | { | |
6793 | u32 val, rp1; | |
6794 | ||
6795 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); | |
6796 | ||
6797 | rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT; | |
6798 | ||
6799 | return rp1; | |
6800 | } | |
6801 | ||
03af2045 | 6802 | static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv) |
0a073b84 JB |
6803 | { |
6804 | u32 val, rp0; | |
6805 | ||
64936258 | 6806 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
0a073b84 JB |
6807 | |
6808 | rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT; | |
6809 | /* Clamp to max */ | |
6810 | rp0 = min_t(u32, rp0, 0xea); | |
6811 | ||
6812 | return rp0; | |
6813 | } | |
6814 | ||
6815 | static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
6816 | { | |
6817 | u32 val, rpe; | |
6818 | ||
64936258 | 6819 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO); |
0a073b84 | 6820 | rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT; |
64936258 | 6821 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI); |
0a073b84 JB |
6822 | rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5; |
6823 | ||
6824 | return rpe; | |
6825 | } | |
6826 | ||
03af2045 | 6827 | static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv) |
0a073b84 | 6828 | { |
36146035 ID |
6829 | u32 val; |
6830 | ||
6831 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff; | |
6832 | /* | |
6833 | * According to the BYT Punit GPU turbo HAS 1.1.6.3 the minimum value | |
6834 | * for the minimum frequency in GPLL mode is 0xc1. Contrary to this on | |
6835 | * a BYT-M B0 the above register contains 0xbf. Moreover when setting | |
6836 | * a frequency Punit will not allow values below 0xc0. Clamp it 0xc0 | |
6837 | * to make sure it matches what Punit accepts. | |
6838 | */ | |
6839 | return max_t(u32, val, 0xc0); | |
0a073b84 JB |
6840 | } |
6841 | ||
ae48434c ID |
6842 | /* Check that the pctx buffer wasn't move under us. */ |
6843 | static void valleyview_check_pctx(struct drm_i915_private *dev_priv) | |
6844 | { | |
6845 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; | |
6846 | ||
6847 | WARN_ON(pctx_addr != dev_priv->mm.stolen_base + | |
6848 | dev_priv->vlv_pctx->stolen->start); | |
6849 | } | |
6850 | ||
38807746 D |
6851 | |
6852 | /* Check that the pcbr address is not empty. */ | |
6853 | static void cherryview_check_pctx(struct drm_i915_private *dev_priv) | |
6854 | { | |
6855 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; | |
6856 | ||
6857 | WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0); | |
6858 | } | |
6859 | ||
dc97997a | 6860 | static void cherryview_setup_pctx(struct drm_i915_private *dev_priv) |
38807746 | 6861 | { |
62106b4f | 6862 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
72e96d64 | 6863 | unsigned long pctx_paddr, paddr; |
38807746 D |
6864 | u32 pcbr; |
6865 | int pctx_size = 32*1024; | |
6866 | ||
38807746 D |
6867 | pcbr = I915_READ(VLV_PCBR); |
6868 | if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) { | |
ce611ef8 | 6869 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
38807746 | 6870 | paddr = (dev_priv->mm.stolen_base + |
62106b4f | 6871 | (ggtt->stolen_size - pctx_size)); |
38807746 D |
6872 | |
6873 | pctx_paddr = (paddr & (~4095)); | |
6874 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
6875 | } | |
ce611ef8 VS |
6876 | |
6877 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); | |
38807746 D |
6878 | } |
6879 | ||
dc97997a | 6880 | static void valleyview_setup_pctx(struct drm_i915_private *dev_priv) |
c9cddffc | 6881 | { |
c9cddffc JB |
6882 | struct drm_i915_gem_object *pctx; |
6883 | unsigned long pctx_paddr; | |
6884 | u32 pcbr; | |
6885 | int pctx_size = 24*1024; | |
6886 | ||
6887 | pcbr = I915_READ(VLV_PCBR); | |
6888 | if (pcbr) { | |
6889 | /* BIOS set it up already, grab the pre-alloc'd space */ | |
6890 | int pcbr_offset; | |
6891 | ||
6892 | pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base; | |
187685cb | 6893 | pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv, |
c9cddffc | 6894 | pcbr_offset, |
190d6cd5 | 6895 | I915_GTT_OFFSET_NONE, |
c9cddffc JB |
6896 | pctx_size); |
6897 | goto out; | |
6898 | } | |
6899 | ||
ce611ef8 VS |
6900 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
6901 | ||
c9cddffc JB |
6902 | /* |
6903 | * From the Gunit register HAS: | |
6904 | * The Gfx driver is expected to program this register and ensure | |
6905 | * proper allocation within Gfx stolen memory. For example, this | |
6906 | * register should be programmed such than the PCBR range does not | |
6907 | * overlap with other ranges, such as the frame buffer, protected | |
6908 | * memory, or any other relevant ranges. | |
6909 | */ | |
187685cb | 6910 | pctx = i915_gem_object_create_stolen(dev_priv, pctx_size); |
c9cddffc JB |
6911 | if (!pctx) { |
6912 | DRM_DEBUG("not enough stolen space for PCTX, disabling\n"); | |
ee504898 | 6913 | goto out; |
c9cddffc JB |
6914 | } |
6915 | ||
6916 | pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start; | |
6917 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
6918 | ||
6919 | out: | |
ce611ef8 | 6920 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
c9cddffc JB |
6921 | dev_priv->vlv_pctx = pctx; |
6922 | } | |
6923 | ||
dc97997a | 6924 | static void valleyview_cleanup_pctx(struct drm_i915_private *dev_priv) |
ae48434c | 6925 | { |
ae48434c ID |
6926 | if (WARN_ON(!dev_priv->vlv_pctx)) |
6927 | return; | |
6928 | ||
f0cd5182 | 6929 | i915_gem_object_put(dev_priv->vlv_pctx); |
ae48434c ID |
6930 | dev_priv->vlv_pctx = NULL; |
6931 | } | |
6932 | ||
c30fec65 VS |
6933 | static void vlv_init_gpll_ref_freq(struct drm_i915_private *dev_priv) |
6934 | { | |
6935 | dev_priv->rps.gpll_ref_freq = | |
6936 | vlv_get_cck_clock(dev_priv, "GPLL ref", | |
6937 | CCK_GPLL_CLOCK_CONTROL, | |
6938 | dev_priv->czclk_freq); | |
6939 | ||
6940 | DRM_DEBUG_DRIVER("GPLL reference freq: %d kHz\n", | |
6941 | dev_priv->rps.gpll_ref_freq); | |
6942 | } | |
6943 | ||
dc97997a | 6944 | static void valleyview_init_gt_powersave(struct drm_i915_private *dev_priv) |
4e80519e | 6945 | { |
2bb25c17 | 6946 | u32 val; |
4e80519e | 6947 | |
dc97997a | 6948 | valleyview_setup_pctx(dev_priv); |
4e80519e | 6949 | |
c30fec65 VS |
6950 | vlv_init_gpll_ref_freq(dev_priv); |
6951 | ||
2bb25c17 VS |
6952 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
6953 | switch ((val >> 6) & 3) { | |
6954 | case 0: | |
6955 | case 1: | |
6956 | dev_priv->mem_freq = 800; | |
6957 | break; | |
6958 | case 2: | |
6959 | dev_priv->mem_freq = 1066; | |
6960 | break; | |
6961 | case 3: | |
6962 | dev_priv->mem_freq = 1333; | |
6963 | break; | |
6964 | } | |
80b83b62 | 6965 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
2bb25c17 | 6966 | |
4e80519e ID |
6967 | dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv); |
6968 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; | |
6969 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 6970 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
4e80519e ID |
6971 | dev_priv->rps.max_freq); |
6972 | ||
6973 | dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv); | |
6974 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 6975 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
4e80519e ID |
6976 | dev_priv->rps.efficient_freq); |
6977 | ||
f8f2b001 D |
6978 | dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv); |
6979 | DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 6980 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
f8f2b001 D |
6981 | dev_priv->rps.rp1_freq); |
6982 | ||
4e80519e ID |
6983 | dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv); |
6984 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 6985 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
4e80519e | 6986 | dev_priv->rps.min_freq); |
4e80519e ID |
6987 | } |
6988 | ||
dc97997a | 6989 | static void cherryview_init_gt_powersave(struct drm_i915_private *dev_priv) |
38807746 | 6990 | { |
2bb25c17 | 6991 | u32 val; |
2b6b3a09 | 6992 | |
dc97997a | 6993 | cherryview_setup_pctx(dev_priv); |
2b6b3a09 | 6994 | |
c30fec65 VS |
6995 | vlv_init_gpll_ref_freq(dev_priv); |
6996 | ||
a580516d | 6997 | mutex_lock(&dev_priv->sb_lock); |
c6e8f39d | 6998 | val = vlv_cck_read(dev_priv, CCK_FUSE_REG); |
a580516d | 6999 | mutex_unlock(&dev_priv->sb_lock); |
c6e8f39d | 7000 | |
2bb25c17 | 7001 | switch ((val >> 2) & 0x7) { |
2bb25c17 | 7002 | case 3: |
2bb25c17 VS |
7003 | dev_priv->mem_freq = 2000; |
7004 | break; | |
bfa7df01 | 7005 | default: |
2bb25c17 VS |
7006 | dev_priv->mem_freq = 1600; |
7007 | break; | |
7008 | } | |
80b83b62 | 7009 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
2bb25c17 | 7010 | |
2b6b3a09 D |
7011 | dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv); |
7012 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; | |
7013 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 7014 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
2b6b3a09 D |
7015 | dev_priv->rps.max_freq); |
7016 | ||
7017 | dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv); | |
7018 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 7019 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
2b6b3a09 D |
7020 | dev_priv->rps.efficient_freq); |
7021 | ||
7707df4a D |
7022 | dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv); |
7023 | DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 7024 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
7707df4a D |
7025 | dev_priv->rps.rp1_freq); |
7026 | ||
96676fe3 | 7027 | dev_priv->rps.min_freq = cherryview_rps_min_freq(dev_priv); |
2b6b3a09 | 7028 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
7c59a9c1 | 7029 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
2b6b3a09 D |
7030 | dev_priv->rps.min_freq); |
7031 | ||
1c14762d VS |
7032 | WARN_ONCE((dev_priv->rps.max_freq | |
7033 | dev_priv->rps.efficient_freq | | |
7034 | dev_priv->rps.rp1_freq | | |
7035 | dev_priv->rps.min_freq) & 1, | |
7036 | "Odd GPU freq values\n"); | |
38807746 D |
7037 | } |
7038 | ||
dc97997a | 7039 | static void valleyview_cleanup_gt_powersave(struct drm_i915_private *dev_priv) |
4e80519e | 7040 | { |
dc97997a | 7041 | valleyview_cleanup_pctx(dev_priv); |
4e80519e ID |
7042 | } |
7043 | ||
dc97997a | 7044 | static void cherryview_enable_rps(struct drm_i915_private *dev_priv) |
38807746 | 7045 | { |
e2f80391 | 7046 | struct intel_engine_cs *engine; |
3b3f1650 | 7047 | enum intel_engine_id id; |
2b6b3a09 | 7048 | u32 gtfifodbg, val, rc6_mode = 0, pcbr; |
38807746 D |
7049 | |
7050 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
7051 | ||
297b32ec VS |
7052 | gtfifodbg = I915_READ(GTFIFODBG) & ~(GT_FIFO_SBDEDICATE_FREE_ENTRY_CHV | |
7053 | GT_FIFO_FREE_ENTRIES_CHV); | |
38807746 D |
7054 | if (gtfifodbg) { |
7055 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", | |
7056 | gtfifodbg); | |
7057 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
7058 | } | |
7059 | ||
7060 | cherryview_check_pctx(dev_priv); | |
7061 | ||
7062 | /* 1a & 1b: Get forcewake during program sequence. Although the driver | |
7063 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 7064 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
38807746 | 7065 | |
160614a2 VS |
7066 | /* Disable RC states. */ |
7067 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
7068 | ||
38807746 D |
7069 | /* 2a: Program RC6 thresholds.*/ |
7070 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); | |
7071 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
7072 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
7073 | ||
3b3f1650 | 7074 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 7075 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
38807746 D |
7076 | I915_WRITE(GEN6_RC_SLEEP, 0); |
7077 | ||
f4f71c7d D |
7078 | /* TO threshold set to 500 us ( 0x186 * 1.28 us) */ |
7079 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x186); | |
38807746 D |
7080 | |
7081 | /* allows RC6 residency counter to work */ | |
7082 | I915_WRITE(VLV_COUNTER_CONTROL, | |
7083 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH | | |
7084 | VLV_MEDIA_RC6_COUNT_EN | | |
7085 | VLV_RENDER_RC6_COUNT_EN)); | |
7086 | ||
7087 | /* For now we assume BIOS is allocating and populating the PCBR */ | |
7088 | pcbr = I915_READ(VLV_PCBR); | |
7089 | ||
38807746 | 7090 | /* 3: Enable RC6 */ |
dc97997a CW |
7091 | if ((intel_enable_rc6() & INTEL_RC6_ENABLE) && |
7092 | (pcbr >> VLV_PCBR_ADDR_SHIFT)) | |
af5a75a3 | 7093 | rc6_mode = GEN7_RC_CTL_TO_MODE; |
38807746 D |
7094 | |
7095 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); | |
7096 | ||
2b6b3a09 | 7097 | /* 4 Program defaults and thresholds for RPS*/ |
3cbdb48f | 7098 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
2b6b3a09 D |
7099 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
7100 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
7101 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
7102 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
7103 | ||
7104 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
7105 | ||
7106 | /* 5: Enable RPS */ | |
7107 | I915_WRITE(GEN6_RP_CONTROL, | |
7108 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
eb973a5e | 7109 | GEN6_RP_MEDIA_IS_GFX | |
2b6b3a09 D |
7110 | GEN6_RP_ENABLE | |
7111 | GEN6_RP_UP_BUSY_AVG | | |
7112 | GEN6_RP_DOWN_IDLE_AVG); | |
7113 | ||
3ef62342 D |
7114 | /* Setting Fixed Bias */ |
7115 | val = VLV_OVERRIDE_EN | | |
7116 | VLV_SOC_TDP_EN | | |
7117 | CHV_BIAS_CPU_50_SOC_50; | |
7118 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); | |
7119 | ||
2b6b3a09 D |
7120 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
7121 | ||
8d40c3ae VS |
7122 | /* RPS code assumes GPLL is used */ |
7123 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); | |
7124 | ||
742f491d | 7125 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE)); |
2b6b3a09 D |
7126 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
7127 | ||
3a45b05c | 7128 | reset_rps(dev_priv, valleyview_set_rps); |
2b6b3a09 | 7129 | |
59bad947 | 7130 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
38807746 D |
7131 | } |
7132 | ||
dc97997a | 7133 | static void valleyview_enable_rps(struct drm_i915_private *dev_priv) |
0a073b84 | 7134 | { |
e2f80391 | 7135 | struct intel_engine_cs *engine; |
3b3f1650 | 7136 | enum intel_engine_id id; |
2a5913a8 | 7137 | u32 gtfifodbg, val, rc6_mode = 0; |
0a073b84 JB |
7138 | |
7139 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
7140 | ||
ae48434c ID |
7141 | valleyview_check_pctx(dev_priv); |
7142 | ||
297b32ec VS |
7143 | gtfifodbg = I915_READ(GTFIFODBG); |
7144 | if (gtfifodbg) { | |
f7d85c1e JB |
7145 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
7146 | gtfifodbg); | |
0a073b84 JB |
7147 | I915_WRITE(GTFIFODBG, gtfifodbg); |
7148 | } | |
7149 | ||
c8d9a590 | 7150 | /* If VLV, Forcewake all wells, else re-direct to regular path */ |
59bad947 | 7151 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
0a073b84 | 7152 | |
160614a2 VS |
7153 | /* Disable RC states. */ |
7154 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
7155 | ||
cad725fe | 7156 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
0a073b84 JB |
7157 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
7158 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
7159 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
7160 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
7161 | ||
7162 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
7163 | ||
7164 | I915_WRITE(GEN6_RP_CONTROL, | |
7165 | GEN6_RP_MEDIA_TURBO | | |
7166 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
7167 | GEN6_RP_MEDIA_IS_GFX | | |
7168 | GEN6_RP_ENABLE | | |
7169 | GEN6_RP_UP_BUSY_AVG | | |
7170 | GEN6_RP_DOWN_IDLE_CONT); | |
7171 | ||
7172 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000); | |
7173 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
7174 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
7175 | ||
3b3f1650 | 7176 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 7177 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
0a073b84 | 7178 | |
2f0aa304 | 7179 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x557); |
0a073b84 JB |
7180 | |
7181 | /* allows RC6 residency counter to work */ | |
49798eb2 | 7182 | I915_WRITE(VLV_COUNTER_CONTROL, |
6b7f6aa7 MK |
7183 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH | |
7184 | VLV_MEDIA_RC0_COUNT_EN | | |
31685c25 | 7185 | VLV_RENDER_RC0_COUNT_EN | |
49798eb2 JB |
7186 | VLV_MEDIA_RC6_COUNT_EN | |
7187 | VLV_RENDER_RC6_COUNT_EN)); | |
31685c25 | 7188 | |
dc97997a | 7189 | if (intel_enable_rc6() & INTEL_RC6_ENABLE) |
6b88f295 | 7190 | rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL; |
dc39fff7 | 7191 | |
dc97997a | 7192 | intel_print_rc6_info(dev_priv, rc6_mode); |
dc39fff7 | 7193 | |
a2b23fe0 | 7194 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
0a073b84 | 7195 | |
3ef62342 D |
7196 | /* Setting Fixed Bias */ |
7197 | val = VLV_OVERRIDE_EN | | |
7198 | VLV_SOC_TDP_EN | | |
7199 | VLV_BIAS_CPU_125_SOC_875; | |
7200 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); | |
7201 | ||
64936258 | 7202 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
0a073b84 | 7203 | |
8d40c3ae VS |
7204 | /* RPS code assumes GPLL is used */ |
7205 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); | |
7206 | ||
742f491d | 7207 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE)); |
0a073b84 JB |
7208 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
7209 | ||
3a45b05c | 7210 | reset_rps(dev_priv, valleyview_set_rps); |
0a073b84 | 7211 | |
59bad947 | 7212 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
0a073b84 JB |
7213 | } |
7214 | ||
dde18883 ED |
7215 | static unsigned long intel_pxfreq(u32 vidfreq) |
7216 | { | |
7217 | unsigned long freq; | |
7218 | int div = (vidfreq & 0x3f0000) >> 16; | |
7219 | int post = (vidfreq & 0x3000) >> 12; | |
7220 | int pre = (vidfreq & 0x7); | |
7221 | ||
7222 | if (!pre) | |
7223 | return 0; | |
7224 | ||
7225 | freq = ((div * 133333) / ((1<<post) * pre)); | |
7226 | ||
7227 | return freq; | |
7228 | } | |
7229 | ||
eb48eb00 DV |
7230 | static const struct cparams { |
7231 | u16 i; | |
7232 | u16 t; | |
7233 | u16 m; | |
7234 | u16 c; | |
7235 | } cparams[] = { | |
7236 | { 1, 1333, 301, 28664 }, | |
7237 | { 1, 1066, 294, 24460 }, | |
7238 | { 1, 800, 294, 25192 }, | |
7239 | { 0, 1333, 276, 27605 }, | |
7240 | { 0, 1066, 276, 27605 }, | |
7241 | { 0, 800, 231, 23784 }, | |
7242 | }; | |
7243 | ||
f531dcb2 | 7244 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
7245 | { |
7246 | u64 total_count, diff, ret; | |
7247 | u32 count1, count2, count3, m = 0, c = 0; | |
7248 | unsigned long now = jiffies_to_msecs(jiffies), diff1; | |
7249 | int i; | |
7250 | ||
67520415 | 7251 | lockdep_assert_held(&mchdev_lock); |
02d71956 | 7252 | |
20e4d407 | 7253 | diff1 = now - dev_priv->ips.last_time1; |
eb48eb00 DV |
7254 | |
7255 | /* Prevent division-by-zero if we are asking too fast. | |
7256 | * Also, we don't get interesting results if we are polling | |
7257 | * faster than once in 10ms, so just return the saved value | |
7258 | * in such cases. | |
7259 | */ | |
7260 | if (diff1 <= 10) | |
20e4d407 | 7261 | return dev_priv->ips.chipset_power; |
eb48eb00 DV |
7262 | |
7263 | count1 = I915_READ(DMIEC); | |
7264 | count2 = I915_READ(DDREC); | |
7265 | count3 = I915_READ(CSIEC); | |
7266 | ||
7267 | total_count = count1 + count2 + count3; | |
7268 | ||
7269 | /* FIXME: handle per-counter overflow */ | |
20e4d407 DV |
7270 | if (total_count < dev_priv->ips.last_count1) { |
7271 | diff = ~0UL - dev_priv->ips.last_count1; | |
eb48eb00 DV |
7272 | diff += total_count; |
7273 | } else { | |
20e4d407 | 7274 | diff = total_count - dev_priv->ips.last_count1; |
eb48eb00 DV |
7275 | } |
7276 | ||
7277 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { | |
20e4d407 DV |
7278 | if (cparams[i].i == dev_priv->ips.c_m && |
7279 | cparams[i].t == dev_priv->ips.r_t) { | |
eb48eb00 DV |
7280 | m = cparams[i].m; |
7281 | c = cparams[i].c; | |
7282 | break; | |
7283 | } | |
7284 | } | |
7285 | ||
7286 | diff = div_u64(diff, diff1); | |
7287 | ret = ((m * diff) + c); | |
7288 | ret = div_u64(ret, 10); | |
7289 | ||
20e4d407 DV |
7290 | dev_priv->ips.last_count1 = total_count; |
7291 | dev_priv->ips.last_time1 = now; | |
eb48eb00 | 7292 | |
20e4d407 | 7293 | dev_priv->ips.chipset_power = ret; |
eb48eb00 DV |
7294 | |
7295 | return ret; | |
7296 | } | |
7297 | ||
f531dcb2 CW |
7298 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
7299 | { | |
7300 | unsigned long val; | |
7301 | ||
dc97997a | 7302 | if (INTEL_INFO(dev_priv)->gen != 5) |
f531dcb2 CW |
7303 | return 0; |
7304 | ||
7305 | spin_lock_irq(&mchdev_lock); | |
7306 | ||
7307 | val = __i915_chipset_val(dev_priv); | |
7308 | ||
7309 | spin_unlock_irq(&mchdev_lock); | |
7310 | ||
7311 | return val; | |
7312 | } | |
7313 | ||
eb48eb00 DV |
7314 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
7315 | { | |
7316 | unsigned long m, x, b; | |
7317 | u32 tsfs; | |
7318 | ||
7319 | tsfs = I915_READ(TSFS); | |
7320 | ||
7321 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); | |
7322 | x = I915_READ8(TR1); | |
7323 | ||
7324 | b = tsfs & TSFS_INTR_MASK; | |
7325 | ||
7326 | return ((m * x) / 127) - b; | |
7327 | } | |
7328 | ||
d972d6ee MK |
7329 | static int _pxvid_to_vd(u8 pxvid) |
7330 | { | |
7331 | if (pxvid == 0) | |
7332 | return 0; | |
7333 | ||
7334 | if (pxvid >= 8 && pxvid < 31) | |
7335 | pxvid = 31; | |
7336 | ||
7337 | return (pxvid + 2) * 125; | |
7338 | } | |
7339 | ||
7340 | static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) | |
eb48eb00 | 7341 | { |
d972d6ee MK |
7342 | const int vd = _pxvid_to_vd(pxvid); |
7343 | const int vm = vd - 1125; | |
7344 | ||
dc97997a | 7345 | if (INTEL_INFO(dev_priv)->is_mobile) |
d972d6ee MK |
7346 | return vm > 0 ? vm : 0; |
7347 | ||
7348 | return vd; | |
eb48eb00 DV |
7349 | } |
7350 | ||
02d71956 | 7351 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 | 7352 | { |
5ed0bdf2 | 7353 | u64 now, diff, diffms; |
eb48eb00 DV |
7354 | u32 count; |
7355 | ||
67520415 | 7356 | lockdep_assert_held(&mchdev_lock); |
eb48eb00 | 7357 | |
5ed0bdf2 TG |
7358 | now = ktime_get_raw_ns(); |
7359 | diffms = now - dev_priv->ips.last_time2; | |
7360 | do_div(diffms, NSEC_PER_MSEC); | |
eb48eb00 DV |
7361 | |
7362 | /* Don't divide by 0 */ | |
eb48eb00 DV |
7363 | if (!diffms) |
7364 | return; | |
7365 | ||
7366 | count = I915_READ(GFXEC); | |
7367 | ||
20e4d407 DV |
7368 | if (count < dev_priv->ips.last_count2) { |
7369 | diff = ~0UL - dev_priv->ips.last_count2; | |
eb48eb00 DV |
7370 | diff += count; |
7371 | } else { | |
20e4d407 | 7372 | diff = count - dev_priv->ips.last_count2; |
eb48eb00 DV |
7373 | } |
7374 | ||
20e4d407 DV |
7375 | dev_priv->ips.last_count2 = count; |
7376 | dev_priv->ips.last_time2 = now; | |
eb48eb00 DV |
7377 | |
7378 | /* More magic constants... */ | |
7379 | diff = diff * 1181; | |
7380 | diff = div_u64(diff, diffms * 10); | |
20e4d407 | 7381 | dev_priv->ips.gfx_power = diff; |
eb48eb00 DV |
7382 | } |
7383 | ||
02d71956 DV |
7384 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
7385 | { | |
dc97997a | 7386 | if (INTEL_INFO(dev_priv)->gen != 5) |
02d71956 DV |
7387 | return; |
7388 | ||
9270388e | 7389 | spin_lock_irq(&mchdev_lock); |
02d71956 DV |
7390 | |
7391 | __i915_update_gfx_val(dev_priv); | |
7392 | ||
9270388e | 7393 | spin_unlock_irq(&mchdev_lock); |
02d71956 DV |
7394 | } |
7395 | ||
f531dcb2 | 7396 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
7397 | { |
7398 | unsigned long t, corr, state1, corr2, state2; | |
7399 | u32 pxvid, ext_v; | |
7400 | ||
67520415 | 7401 | lockdep_assert_held(&mchdev_lock); |
02d71956 | 7402 | |
616847e7 | 7403 | pxvid = I915_READ(PXVFREQ(dev_priv->rps.cur_freq)); |
eb48eb00 DV |
7404 | pxvid = (pxvid >> 24) & 0x7f; |
7405 | ext_v = pvid_to_extvid(dev_priv, pxvid); | |
7406 | ||
7407 | state1 = ext_v; | |
7408 | ||
7409 | t = i915_mch_val(dev_priv); | |
7410 | ||
7411 | /* Revel in the empirically derived constants */ | |
7412 | ||
7413 | /* Correction factor in 1/100000 units */ | |
7414 | if (t > 80) | |
7415 | corr = ((t * 2349) + 135940); | |
7416 | else if (t >= 50) | |
7417 | corr = ((t * 964) + 29317); | |
7418 | else /* < 50 */ | |
7419 | corr = ((t * 301) + 1004); | |
7420 | ||
7421 | corr = corr * ((150142 * state1) / 10000 - 78642); | |
7422 | corr /= 100000; | |
20e4d407 | 7423 | corr2 = (corr * dev_priv->ips.corr); |
eb48eb00 DV |
7424 | |
7425 | state2 = (corr2 * state1) / 10000; | |
7426 | state2 /= 100; /* convert to mW */ | |
7427 | ||
02d71956 | 7428 | __i915_update_gfx_val(dev_priv); |
eb48eb00 | 7429 | |
20e4d407 | 7430 | return dev_priv->ips.gfx_power + state2; |
eb48eb00 DV |
7431 | } |
7432 | ||
f531dcb2 CW |
7433 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
7434 | { | |
7435 | unsigned long val; | |
7436 | ||
dc97997a | 7437 | if (INTEL_INFO(dev_priv)->gen != 5) |
f531dcb2 CW |
7438 | return 0; |
7439 | ||
7440 | spin_lock_irq(&mchdev_lock); | |
7441 | ||
7442 | val = __i915_gfx_val(dev_priv); | |
7443 | ||
7444 | spin_unlock_irq(&mchdev_lock); | |
7445 | ||
7446 | return val; | |
7447 | } | |
7448 | ||
eb48eb00 DV |
7449 | /** |
7450 | * i915_read_mch_val - return value for IPS use | |
7451 | * | |
7452 | * Calculate and return a value for the IPS driver to use when deciding whether | |
7453 | * we have thermal and power headroom to increase CPU or GPU power budget. | |
7454 | */ | |
7455 | unsigned long i915_read_mch_val(void) | |
7456 | { | |
7457 | struct drm_i915_private *dev_priv; | |
7458 | unsigned long chipset_val, graphics_val, ret = 0; | |
7459 | ||
9270388e | 7460 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
7461 | if (!i915_mch_dev) |
7462 | goto out_unlock; | |
7463 | dev_priv = i915_mch_dev; | |
7464 | ||
f531dcb2 CW |
7465 | chipset_val = __i915_chipset_val(dev_priv); |
7466 | graphics_val = __i915_gfx_val(dev_priv); | |
eb48eb00 DV |
7467 | |
7468 | ret = chipset_val + graphics_val; | |
7469 | ||
7470 | out_unlock: | |
9270388e | 7471 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
7472 | |
7473 | return ret; | |
7474 | } | |
7475 | EXPORT_SYMBOL_GPL(i915_read_mch_val); | |
7476 | ||
7477 | /** | |
7478 | * i915_gpu_raise - raise GPU frequency limit | |
7479 | * | |
7480 | * Raise the limit; IPS indicates we have thermal headroom. | |
7481 | */ | |
7482 | bool i915_gpu_raise(void) | |
7483 | { | |
7484 | struct drm_i915_private *dev_priv; | |
7485 | bool ret = true; | |
7486 | ||
9270388e | 7487 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
7488 | if (!i915_mch_dev) { |
7489 | ret = false; | |
7490 | goto out_unlock; | |
7491 | } | |
7492 | dev_priv = i915_mch_dev; | |
7493 | ||
20e4d407 DV |
7494 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
7495 | dev_priv->ips.max_delay--; | |
eb48eb00 DV |
7496 | |
7497 | out_unlock: | |
9270388e | 7498 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
7499 | |
7500 | return ret; | |
7501 | } | |
7502 | EXPORT_SYMBOL_GPL(i915_gpu_raise); | |
7503 | ||
7504 | /** | |
7505 | * i915_gpu_lower - lower GPU frequency limit | |
7506 | * | |
7507 | * IPS indicates we're close to a thermal limit, so throttle back the GPU | |
7508 | * frequency maximum. | |
7509 | */ | |
7510 | bool i915_gpu_lower(void) | |
7511 | { | |
7512 | struct drm_i915_private *dev_priv; | |
7513 | bool ret = true; | |
7514 | ||
9270388e | 7515 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
7516 | if (!i915_mch_dev) { |
7517 | ret = false; | |
7518 | goto out_unlock; | |
7519 | } | |
7520 | dev_priv = i915_mch_dev; | |
7521 | ||
20e4d407 DV |
7522 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
7523 | dev_priv->ips.max_delay++; | |
eb48eb00 DV |
7524 | |
7525 | out_unlock: | |
9270388e | 7526 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
7527 | |
7528 | return ret; | |
7529 | } | |
7530 | EXPORT_SYMBOL_GPL(i915_gpu_lower); | |
7531 | ||
7532 | /** | |
7533 | * i915_gpu_busy - indicate GPU business to IPS | |
7534 | * | |
7535 | * Tell the IPS driver whether or not the GPU is busy. | |
7536 | */ | |
7537 | bool i915_gpu_busy(void) | |
7538 | { | |
eb48eb00 DV |
7539 | bool ret = false; |
7540 | ||
9270388e | 7541 | spin_lock_irq(&mchdev_lock); |
dcff85c8 CW |
7542 | if (i915_mch_dev) |
7543 | ret = i915_mch_dev->gt.awake; | |
9270388e | 7544 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
7545 | |
7546 | return ret; | |
7547 | } | |
7548 | EXPORT_SYMBOL_GPL(i915_gpu_busy); | |
7549 | ||
7550 | /** | |
7551 | * i915_gpu_turbo_disable - disable graphics turbo | |
7552 | * | |
7553 | * Disable graphics turbo by resetting the max frequency and setting the | |
7554 | * current frequency to the default. | |
7555 | */ | |
7556 | bool i915_gpu_turbo_disable(void) | |
7557 | { | |
7558 | struct drm_i915_private *dev_priv; | |
7559 | bool ret = true; | |
7560 | ||
9270388e | 7561 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
7562 | if (!i915_mch_dev) { |
7563 | ret = false; | |
7564 | goto out_unlock; | |
7565 | } | |
7566 | dev_priv = i915_mch_dev; | |
7567 | ||
20e4d407 | 7568 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
eb48eb00 | 7569 | |
91d14251 | 7570 | if (!ironlake_set_drps(dev_priv, dev_priv->ips.fstart)) |
eb48eb00 DV |
7571 | ret = false; |
7572 | ||
7573 | out_unlock: | |
9270388e | 7574 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
7575 | |
7576 | return ret; | |
7577 | } | |
7578 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); | |
7579 | ||
7580 | /** | |
7581 | * Tells the intel_ips driver that the i915 driver is now loaded, if | |
7582 | * IPS got loaded first. | |
7583 | * | |
7584 | * This awkward dance is so that neither module has to depend on the | |
7585 | * other in order for IPS to do the appropriate communication of | |
7586 | * GPU turbo limits to i915. | |
7587 | */ | |
7588 | static void | |
7589 | ips_ping_for_i915_load(void) | |
7590 | { | |
7591 | void (*link)(void); | |
7592 | ||
7593 | link = symbol_get(ips_link_to_i915_driver); | |
7594 | if (link) { | |
7595 | link(); | |
7596 | symbol_put(ips_link_to_i915_driver); | |
7597 | } | |
7598 | } | |
7599 | ||
7600 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) | |
7601 | { | |
02d71956 DV |
7602 | /* We only register the i915 ips part with intel-ips once everything is |
7603 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ | |
9270388e | 7604 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 7605 | i915_mch_dev = dev_priv; |
9270388e | 7606 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
7607 | |
7608 | ips_ping_for_i915_load(); | |
7609 | } | |
7610 | ||
7611 | void intel_gpu_ips_teardown(void) | |
7612 | { | |
9270388e | 7613 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 7614 | i915_mch_dev = NULL; |
9270388e | 7615 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 | 7616 | } |
76c3552f | 7617 | |
dc97997a | 7618 | static void intel_init_emon(struct drm_i915_private *dev_priv) |
dde18883 | 7619 | { |
dde18883 ED |
7620 | u32 lcfuse; |
7621 | u8 pxw[16]; | |
7622 | int i; | |
7623 | ||
7624 | /* Disable to program */ | |
7625 | I915_WRITE(ECR, 0); | |
7626 | POSTING_READ(ECR); | |
7627 | ||
7628 | /* Program energy weights for various events */ | |
7629 | I915_WRITE(SDEW, 0x15040d00); | |
7630 | I915_WRITE(CSIEW0, 0x007f0000); | |
7631 | I915_WRITE(CSIEW1, 0x1e220004); | |
7632 | I915_WRITE(CSIEW2, 0x04000004); | |
7633 | ||
7634 | for (i = 0; i < 5; i++) | |
616847e7 | 7635 | I915_WRITE(PEW(i), 0); |
dde18883 | 7636 | for (i = 0; i < 3; i++) |
616847e7 | 7637 | I915_WRITE(DEW(i), 0); |
dde18883 ED |
7638 | |
7639 | /* Program P-state weights to account for frequency power adjustment */ | |
7640 | for (i = 0; i < 16; i++) { | |
616847e7 | 7641 | u32 pxvidfreq = I915_READ(PXVFREQ(i)); |
dde18883 ED |
7642 | unsigned long freq = intel_pxfreq(pxvidfreq); |
7643 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> | |
7644 | PXVFREQ_PX_SHIFT; | |
7645 | unsigned long val; | |
7646 | ||
7647 | val = vid * vid; | |
7648 | val *= (freq / 1000); | |
7649 | val *= 255; | |
7650 | val /= (127*127*900); | |
7651 | if (val > 0xff) | |
7652 | DRM_ERROR("bad pxval: %ld\n", val); | |
7653 | pxw[i] = val; | |
7654 | } | |
7655 | /* Render standby states get 0 weight */ | |
7656 | pxw[14] = 0; | |
7657 | pxw[15] = 0; | |
7658 | ||
7659 | for (i = 0; i < 4; i++) { | |
7660 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | | |
7661 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); | |
616847e7 | 7662 | I915_WRITE(PXW(i), val); |
dde18883 ED |
7663 | } |
7664 | ||
7665 | /* Adjust magic regs to magic values (more experimental results) */ | |
7666 | I915_WRITE(OGW0, 0); | |
7667 | I915_WRITE(OGW1, 0); | |
7668 | I915_WRITE(EG0, 0x00007f00); | |
7669 | I915_WRITE(EG1, 0x0000000e); | |
7670 | I915_WRITE(EG2, 0x000e0000); | |
7671 | I915_WRITE(EG3, 0x68000300); | |
7672 | I915_WRITE(EG4, 0x42000000); | |
7673 | I915_WRITE(EG5, 0x00140031); | |
7674 | I915_WRITE(EG6, 0); | |
7675 | I915_WRITE(EG7, 0); | |
7676 | ||
7677 | for (i = 0; i < 8; i++) | |
616847e7 | 7678 | I915_WRITE(PXWL(i), 0); |
dde18883 ED |
7679 | |
7680 | /* Enable PMON + select events */ | |
7681 | I915_WRITE(ECR, 0x80000019); | |
7682 | ||
7683 | lcfuse = I915_READ(LCFUSE02); | |
7684 | ||
20e4d407 | 7685 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
dde18883 ED |
7686 | } |
7687 | ||
dc97997a | 7688 | void intel_init_gt_powersave(struct drm_i915_private *dev_priv) |
ae48434c | 7689 | { |
b268c699 ID |
7690 | /* |
7691 | * RPM depends on RC6 to save restore the GT HW context, so make RC6 a | |
7692 | * requirement. | |
7693 | */ | |
7694 | if (!i915.enable_rc6) { | |
7695 | DRM_INFO("RC6 disabled, disabling runtime PM support\n"); | |
7696 | intel_runtime_pm_get(dev_priv); | |
7697 | } | |
e6069ca8 | 7698 | |
b5163dbb | 7699 | mutex_lock(&dev_priv->drm.struct_mutex); |
773ea9a8 CW |
7700 | mutex_lock(&dev_priv->rps.hw_lock); |
7701 | ||
7702 | /* Initialize RPS limits (for userspace) */ | |
dc97997a CW |
7703 | if (IS_CHERRYVIEW(dev_priv)) |
7704 | cherryview_init_gt_powersave(dev_priv); | |
7705 | else if (IS_VALLEYVIEW(dev_priv)) | |
7706 | valleyview_init_gt_powersave(dev_priv); | |
2a13ae79 | 7707 | else if (INTEL_GEN(dev_priv) >= 6) |
773ea9a8 CW |
7708 | gen6_init_rps_frequencies(dev_priv); |
7709 | ||
7710 | /* Derive initial user preferences/limits from the hardware limits */ | |
7711 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; | |
7712 | dev_priv->rps.cur_freq = dev_priv->rps.idle_freq; | |
7713 | ||
7714 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; | |
7715 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; | |
7716 | ||
7717 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) | |
7718 | dev_priv->rps.min_freq_softlimit = | |
7719 | max_t(int, | |
7720 | dev_priv->rps.efficient_freq, | |
7721 | intel_freq_opcode(dev_priv, 450)); | |
7722 | ||
99ac9612 CW |
7723 | /* After setting max-softlimit, find the overclock max freq */ |
7724 | if (IS_GEN6(dev_priv) || | |
7725 | IS_IVYBRIDGE(dev_priv) || IS_HASWELL(dev_priv)) { | |
7726 | u32 params = 0; | |
7727 | ||
7728 | sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, ¶ms); | |
7729 | if (params & BIT(31)) { /* OC supported */ | |
7730 | DRM_DEBUG_DRIVER("Overclocking supported, max: %dMHz, overclock: %dMHz\n", | |
7731 | (dev_priv->rps.max_freq & 0xff) * 50, | |
7732 | (params & 0xff) * 50); | |
7733 | dev_priv->rps.max_freq = params & 0xff; | |
7734 | } | |
7735 | } | |
7736 | ||
29ecd78d CW |
7737 | /* Finally allow us to boost to max by default */ |
7738 | dev_priv->rps.boost_freq = dev_priv->rps.max_freq; | |
7739 | ||
773ea9a8 | 7740 | mutex_unlock(&dev_priv->rps.hw_lock); |
b5163dbb | 7741 | mutex_unlock(&dev_priv->drm.struct_mutex); |
54b4f68f CW |
7742 | |
7743 | intel_autoenable_gt_powersave(dev_priv); | |
ae48434c ID |
7744 | } |
7745 | ||
dc97997a | 7746 | void intel_cleanup_gt_powersave(struct drm_i915_private *dev_priv) |
ae48434c | 7747 | { |
8dac1e1f | 7748 | if (IS_VALLEYVIEW(dev_priv)) |
dc97997a | 7749 | valleyview_cleanup_gt_powersave(dev_priv); |
b268c699 ID |
7750 | |
7751 | if (!i915.enable_rc6) | |
7752 | intel_runtime_pm_put(dev_priv); | |
ae48434c ID |
7753 | } |
7754 | ||
54b4f68f CW |
7755 | /** |
7756 | * intel_suspend_gt_powersave - suspend PM work and helper threads | |
7757 | * @dev_priv: i915 device | |
7758 | * | |
7759 | * We don't want to disable RC6 or other features here, we just want | |
7760 | * to make sure any work we've queued has finished and won't bother | |
7761 | * us while we're suspended. | |
7762 | */ | |
7763 | void intel_suspend_gt_powersave(struct drm_i915_private *dev_priv) | |
7764 | { | |
7765 | if (INTEL_GEN(dev_priv) < 6) | |
7766 | return; | |
7767 | ||
7768 | if (cancel_delayed_work_sync(&dev_priv->rps.autoenable_work)) | |
7769 | intel_runtime_pm_put(dev_priv); | |
7770 | ||
7771 | /* gen6_rps_idle() will be called later to disable interrupts */ | |
7772 | } | |
7773 | ||
b7137e0c CW |
7774 | void intel_sanitize_gt_powersave(struct drm_i915_private *dev_priv) |
7775 | { | |
7776 | dev_priv->rps.enabled = true; /* force disabling */ | |
7777 | intel_disable_gt_powersave(dev_priv); | |
54b4f68f CW |
7778 | |
7779 | gen6_reset_rps_interrupts(dev_priv); | |
156c7ca0 JB |
7780 | } |
7781 | ||
dc97997a | 7782 | void intel_disable_gt_powersave(struct drm_i915_private *dev_priv) |
8090c6b9 | 7783 | { |
b7137e0c CW |
7784 | if (!READ_ONCE(dev_priv->rps.enabled)) |
7785 | return; | |
e494837a | 7786 | |
b7137e0c | 7787 | mutex_lock(&dev_priv->rps.hw_lock); |
e534770a | 7788 | |
b7137e0c CW |
7789 | if (INTEL_GEN(dev_priv) >= 9) { |
7790 | gen9_disable_rc6(dev_priv); | |
7791 | gen9_disable_rps(dev_priv); | |
7792 | } else if (IS_CHERRYVIEW(dev_priv)) { | |
7793 | cherryview_disable_rps(dev_priv); | |
7794 | } else if (IS_VALLEYVIEW(dev_priv)) { | |
7795 | valleyview_disable_rps(dev_priv); | |
7796 | } else if (INTEL_GEN(dev_priv) >= 6) { | |
7797 | gen6_disable_rps(dev_priv); | |
7798 | } else if (IS_IRONLAKE_M(dev_priv)) { | |
7799 | ironlake_disable_drps(dev_priv); | |
930ebb46 | 7800 | } |
b7137e0c CW |
7801 | |
7802 | dev_priv->rps.enabled = false; | |
7803 | mutex_unlock(&dev_priv->rps.hw_lock); | |
8090c6b9 DV |
7804 | } |
7805 | ||
b7137e0c | 7806 | void intel_enable_gt_powersave(struct drm_i915_private *dev_priv) |
1a01ab3b | 7807 | { |
54b4f68f CW |
7808 | /* We shouldn't be disabling as we submit, so this should be less |
7809 | * racy than it appears! | |
7810 | */ | |
b7137e0c CW |
7811 | if (READ_ONCE(dev_priv->rps.enabled)) |
7812 | return; | |
1a01ab3b | 7813 | |
b7137e0c CW |
7814 | /* Powersaving is controlled by the host when inside a VM */ |
7815 | if (intel_vgpu_active(dev_priv)) | |
7816 | return; | |
0a073b84 | 7817 | |
b7137e0c | 7818 | mutex_lock(&dev_priv->rps.hw_lock); |
dc97997a CW |
7819 | |
7820 | if (IS_CHERRYVIEW(dev_priv)) { | |
7821 | cherryview_enable_rps(dev_priv); | |
7822 | } else if (IS_VALLEYVIEW(dev_priv)) { | |
7823 | valleyview_enable_rps(dev_priv); | |
b7137e0c | 7824 | } else if (INTEL_GEN(dev_priv) >= 9) { |
dc97997a CW |
7825 | gen9_enable_rc6(dev_priv); |
7826 | gen9_enable_rps(dev_priv); | |
b976dc53 | 7827 | if (IS_GEN9_BC(dev_priv)) |
fb7404e8 | 7828 | gen6_update_ring_freq(dev_priv); |
dc97997a CW |
7829 | } else if (IS_BROADWELL(dev_priv)) { |
7830 | gen8_enable_rps(dev_priv); | |
fb7404e8 | 7831 | gen6_update_ring_freq(dev_priv); |
b7137e0c | 7832 | } else if (INTEL_GEN(dev_priv) >= 6) { |
dc97997a | 7833 | gen6_enable_rps(dev_priv); |
fb7404e8 | 7834 | gen6_update_ring_freq(dev_priv); |
b7137e0c CW |
7835 | } else if (IS_IRONLAKE_M(dev_priv)) { |
7836 | ironlake_enable_drps(dev_priv); | |
7837 | intel_init_emon(dev_priv); | |
0a073b84 | 7838 | } |
aed242ff CW |
7839 | |
7840 | WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq); | |
7841 | WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq); | |
7842 | ||
7843 | WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq); | |
7844 | WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq); | |
7845 | ||
54b4f68f | 7846 | dev_priv->rps.enabled = true; |
b7137e0c CW |
7847 | mutex_unlock(&dev_priv->rps.hw_lock); |
7848 | } | |
3cc134e3 | 7849 | |
54b4f68f CW |
7850 | static void __intel_autoenable_gt_powersave(struct work_struct *work) |
7851 | { | |
7852 | struct drm_i915_private *dev_priv = | |
7853 | container_of(work, typeof(*dev_priv), rps.autoenable_work.work); | |
7854 | struct intel_engine_cs *rcs; | |
7855 | struct drm_i915_gem_request *req; | |
7856 | ||
7857 | if (READ_ONCE(dev_priv->rps.enabled)) | |
7858 | goto out; | |
7859 | ||
3b3f1650 | 7860 | rcs = dev_priv->engine[RCS]; |
e8a9c58f | 7861 | if (rcs->last_retired_context) |
54b4f68f CW |
7862 | goto out; |
7863 | ||
7864 | if (!rcs->init_context) | |
7865 | goto out; | |
7866 | ||
7867 | mutex_lock(&dev_priv->drm.struct_mutex); | |
7868 | ||
7869 | req = i915_gem_request_alloc(rcs, dev_priv->kernel_context); | |
7870 | if (IS_ERR(req)) | |
7871 | goto unlock; | |
7872 | ||
7873 | if (!i915.enable_execlists && i915_switch_context(req) == 0) | |
7874 | rcs->init_context(req); | |
7875 | ||
7876 | /* Mark the device busy, calling intel_enable_gt_powersave() */ | |
e642c85b | 7877 | i915_add_request(req); |
54b4f68f CW |
7878 | |
7879 | unlock: | |
7880 | mutex_unlock(&dev_priv->drm.struct_mutex); | |
7881 | out: | |
7882 | intel_runtime_pm_put(dev_priv); | |
7883 | } | |
7884 | ||
7885 | void intel_autoenable_gt_powersave(struct drm_i915_private *dev_priv) | |
7886 | { | |
7887 | if (READ_ONCE(dev_priv->rps.enabled)) | |
7888 | return; | |
7889 | ||
7890 | if (IS_IRONLAKE_M(dev_priv)) { | |
7891 | ironlake_enable_drps(dev_priv); | |
54b4f68f | 7892 | intel_init_emon(dev_priv); |
54b4f68f CW |
7893 | } else if (INTEL_INFO(dev_priv)->gen >= 6) { |
7894 | /* | |
7895 | * PCU communication is slow and this doesn't need to be | |
7896 | * done at any specific time, so do this out of our fast path | |
7897 | * to make resume and init faster. | |
7898 | * | |
7899 | * We depend on the HW RC6 power context save/restore | |
7900 | * mechanism when entering D3 through runtime PM suspend. So | |
7901 | * disable RPM until RPS/RC6 is properly setup. We can only | |
7902 | * get here via the driver load/system resume/runtime resume | |
7903 | * paths, so the _noresume version is enough (and in case of | |
7904 | * runtime resume it's necessary). | |
7905 | */ | |
7906 | if (queue_delayed_work(dev_priv->wq, | |
7907 | &dev_priv->rps.autoenable_work, | |
7908 | round_jiffies_up_relative(HZ))) | |
7909 | intel_runtime_pm_get_noresume(dev_priv); | |
7910 | } | |
7911 | } | |
7912 | ||
46f16e63 | 7913 | static void ibx_init_clock_gating(struct drm_i915_private *dev_priv) |
3107bd48 | 7914 | { |
3107bd48 DV |
7915 | /* |
7916 | * On Ibex Peak and Cougar Point, we need to disable clock | |
7917 | * gating for the panel power sequencer or it will fail to | |
7918 | * start up when no ports are active. | |
7919 | */ | |
7920 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); | |
7921 | } | |
7922 | ||
46f16e63 | 7923 | static void g4x_disable_trickle_feed(struct drm_i915_private *dev_priv) |
0e088b8f | 7924 | { |
b12ce1d8 | 7925 | enum pipe pipe; |
0e088b8f | 7926 | |
055e393f | 7927 | for_each_pipe(dev_priv, pipe) { |
0e088b8f VS |
7928 | I915_WRITE(DSPCNTR(pipe), |
7929 | I915_READ(DSPCNTR(pipe)) | | |
7930 | DISPPLANE_TRICKLE_FEED_DISABLE); | |
b12ce1d8 VS |
7931 | |
7932 | I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe))); | |
7933 | POSTING_READ(DSPSURF(pipe)); | |
0e088b8f VS |
7934 | } |
7935 | } | |
7936 | ||
46f16e63 | 7937 | static void ilk_init_lp_watermarks(struct drm_i915_private *dev_priv) |
017636cc | 7938 | { |
017636cc VS |
7939 | I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN); |
7940 | I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN); | |
7941 | I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN); | |
7942 | ||
7943 | /* | |
7944 | * Don't touch WM1S_LP_EN here. | |
7945 | * Doing so could cause underruns. | |
7946 | */ | |
7947 | } | |
7948 | ||
46f16e63 | 7949 | static void ironlake_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7950 | { |
231e54f6 | 7951 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 7952 | |
f1e8fa56 DL |
7953 | /* |
7954 | * Required for FBC | |
7955 | * WaFbcDisableDpfcClockGating:ilk | |
7956 | */ | |
4d47e4f5 DL |
7957 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
7958 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | | |
7959 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; | |
6f1d69b0 ED |
7960 | |
7961 | I915_WRITE(PCH_3DCGDIS0, | |
7962 | MARIUNIT_CLOCK_GATE_DISABLE | | |
7963 | SVSMUNIT_CLOCK_GATE_DISABLE); | |
7964 | I915_WRITE(PCH_3DCGDIS1, | |
7965 | VFMUNIT_CLOCK_GATE_DISABLE); | |
7966 | ||
6f1d69b0 ED |
7967 | /* |
7968 | * According to the spec the following bits should be set in | |
7969 | * order to enable memory self-refresh | |
7970 | * The bit 22/21 of 0x42004 | |
7971 | * The bit 5 of 0x42020 | |
7972 | * The bit 15 of 0x45000 | |
7973 | */ | |
7974 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
7975 | (I915_READ(ILK_DISPLAY_CHICKEN2) | | |
7976 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); | |
4d47e4f5 | 7977 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
6f1d69b0 ED |
7978 | I915_WRITE(DISP_ARB_CTL, |
7979 | (I915_READ(DISP_ARB_CTL) | | |
7980 | DISP_FBC_WM_DIS)); | |
017636cc | 7981 | |
46f16e63 | 7982 | ilk_init_lp_watermarks(dev_priv); |
6f1d69b0 ED |
7983 | |
7984 | /* | |
7985 | * Based on the document from hardware guys the following bits | |
7986 | * should be set unconditionally in order to enable FBC. | |
7987 | * The bit 22 of 0x42000 | |
7988 | * The bit 22 of 0x42004 | |
7989 | * The bit 7,8,9 of 0x42020. | |
7990 | */ | |
50a0bc90 | 7991 | if (IS_IRONLAKE_M(dev_priv)) { |
4bb35334 | 7992 | /* WaFbcAsynchFlipDisableFbcQueue:ilk */ |
6f1d69b0 ED |
7993 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
7994 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
7995 | ILK_FBCQ_DIS); | |
7996 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
7997 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
7998 | ILK_DPARB_GATE); | |
6f1d69b0 ED |
7999 | } |
8000 | ||
4d47e4f5 DL |
8001 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
8002 | ||
6f1d69b0 ED |
8003 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
8004 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
8005 | ILK_ELPIN_409_SELECT); | |
8006 | I915_WRITE(_3D_CHICKEN2, | |
8007 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | | |
8008 | _3D_CHICKEN2_WM_READ_PIPELINED); | |
4358a374 | 8009 | |
ecdb4eb7 | 8010 | /* WaDisableRenderCachePipelinedFlush:ilk */ |
4358a374 DV |
8011 | I915_WRITE(CACHE_MODE_0, |
8012 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
3107bd48 | 8013 | |
4e04632e AG |
8014 | /* WaDisable_RenderCache_OperationalFlush:ilk */ |
8015 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
8016 | ||
46f16e63 | 8017 | g4x_disable_trickle_feed(dev_priv); |
bdad2b2f | 8018 | |
46f16e63 | 8019 | ibx_init_clock_gating(dev_priv); |
3107bd48 DV |
8020 | } |
8021 | ||
46f16e63 | 8022 | static void cpt_init_clock_gating(struct drm_i915_private *dev_priv) |
3107bd48 | 8023 | { |
3107bd48 | 8024 | int pipe; |
3f704fa2 | 8025 | uint32_t val; |
3107bd48 DV |
8026 | |
8027 | /* | |
8028 | * On Ibex Peak and Cougar Point, we need to disable clock | |
8029 | * gating for the panel power sequencer or it will fail to | |
8030 | * start up when no ports are active. | |
8031 | */ | |
cd664078 JB |
8032 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE | |
8033 | PCH_DPLUNIT_CLOCK_GATE_DISABLE | | |
8034 | PCH_CPUNIT_CLOCK_GATE_DISABLE); | |
3107bd48 DV |
8035 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | |
8036 | DPLS_EDP_PPS_FIX_DIS); | |
335c07b7 TI |
8037 | /* The below fixes the weird display corruption, a few pixels shifted |
8038 | * downward, on (only) LVDS of some HP laptops with IVY. | |
8039 | */ | |
055e393f | 8040 | for_each_pipe(dev_priv, pipe) { |
dc4bd2d1 PZ |
8041 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
8042 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; | |
8043 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; | |
41aa3448 | 8044 | if (dev_priv->vbt.fdi_rx_polarity_inverted) |
3f704fa2 | 8045 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
dc4bd2d1 PZ |
8046 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
8047 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; | |
8048 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; | |
3f704fa2 PZ |
8049 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
8050 | } | |
3107bd48 | 8051 | /* WADP0ClockGatingDisable */ |
055e393f | 8052 | for_each_pipe(dev_priv, pipe) { |
3107bd48 DV |
8053 | I915_WRITE(TRANS_CHICKEN1(pipe), |
8054 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); | |
8055 | } | |
6f1d69b0 ED |
8056 | } |
8057 | ||
46f16e63 | 8058 | static void gen6_check_mch_setup(struct drm_i915_private *dev_priv) |
1d7aaa0c | 8059 | { |
1d7aaa0c DV |
8060 | uint32_t tmp; |
8061 | ||
8062 | tmp = I915_READ(MCH_SSKPD); | |
df662a28 DV |
8063 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) |
8064 | DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n", | |
8065 | tmp); | |
1d7aaa0c DV |
8066 | } |
8067 | ||
46f16e63 | 8068 | static void gen6_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8069 | { |
231e54f6 | 8070 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 8071 | |
231e54f6 | 8072 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6f1d69b0 ED |
8073 | |
8074 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
8075 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
8076 | ILK_ELPIN_409_SELECT); | |
8077 | ||
ecdb4eb7 | 8078 | /* WaDisableHiZPlanesWhenMSAAEnabled:snb */ |
4283908e DV |
8079 | I915_WRITE(_3D_CHICKEN, |
8080 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); | |
8081 | ||
4e04632e AG |
8082 | /* WaDisable_RenderCache_OperationalFlush:snb */ |
8083 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
8084 | ||
8d85d272 VS |
8085 | /* |
8086 | * BSpec recoomends 8x4 when MSAA is used, | |
8087 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
8088 | * |
8089 | * Note that PS/WM thread counts depend on the WIZ hashing | |
8090 | * disable bit, which we don't touch here, but it's good | |
8091 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
8d85d272 VS |
8092 | */ |
8093 | I915_WRITE(GEN6_GT_MODE, | |
98533251 | 8094 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
8d85d272 | 8095 | |
46f16e63 | 8096 | ilk_init_lp_watermarks(dev_priv); |
6f1d69b0 | 8097 | |
6f1d69b0 | 8098 | I915_WRITE(CACHE_MODE_0, |
50743298 | 8099 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
6f1d69b0 ED |
8100 | |
8101 | I915_WRITE(GEN6_UCGCTL1, | |
8102 | I915_READ(GEN6_UCGCTL1) | | |
8103 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | | |
8104 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
8105 | ||
8106 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock | |
8107 | * gating disable must be set. Failure to set it results in | |
8108 | * flickering pixels due to Z write ordering failures after | |
8109 | * some amount of runtime in the Mesa "fire" demo, and Unigine | |
8110 | * Sanctuary and Tropics, and apparently anything else with | |
8111 | * alpha test or pixel discard. | |
8112 | * | |
8113 | * According to the spec, bit 11 (RCCUNIT) must also be set, | |
8114 | * but we didn't debug actual testcases to find it out. | |
0f846f81 | 8115 | * |
ef59318c VS |
8116 | * WaDisableRCCUnitClockGating:snb |
8117 | * WaDisableRCPBUnitClockGating:snb | |
6f1d69b0 ED |
8118 | */ |
8119 | I915_WRITE(GEN6_UCGCTL2, | |
8120 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | | |
8121 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); | |
8122 | ||
5eb146dd | 8123 | /* WaStripsFansDisableFastClipPerformanceFix:snb */ |
743b57d8 VS |
8124 | I915_WRITE(_3D_CHICKEN3, |
8125 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL)); | |
6f1d69b0 | 8126 | |
e927ecde VS |
8127 | /* |
8128 | * Bspec says: | |
8129 | * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and | |
8130 | * 3DSTATE_SF number of SF output attributes is more than 16." | |
8131 | */ | |
8132 | I915_WRITE(_3D_CHICKEN3, | |
8133 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH)); | |
8134 | ||
6f1d69b0 ED |
8135 | /* |
8136 | * According to the spec the following bits should be | |
8137 | * set in order to enable memory self-refresh and fbc: | |
8138 | * The bit21 and bit22 of 0x42000 | |
8139 | * The bit21 and bit22 of 0x42004 | |
8140 | * The bit5 and bit7 of 0x42020 | |
8141 | * The bit14 of 0x70180 | |
8142 | * The bit14 of 0x71180 | |
4bb35334 DL |
8143 | * |
8144 | * WaFbcAsynchFlipDisableFbcQueue:snb | |
6f1d69b0 ED |
8145 | */ |
8146 | I915_WRITE(ILK_DISPLAY_CHICKEN1, | |
8147 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
8148 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); | |
8149 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
8150 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
8151 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); | |
231e54f6 DL |
8152 | I915_WRITE(ILK_DSPCLK_GATE_D, |
8153 | I915_READ(ILK_DSPCLK_GATE_D) | | |
8154 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | | |
8155 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); | |
6f1d69b0 | 8156 | |
46f16e63 | 8157 | g4x_disable_trickle_feed(dev_priv); |
f8f2ac9a | 8158 | |
46f16e63 | 8159 | cpt_init_clock_gating(dev_priv); |
1d7aaa0c | 8160 | |
46f16e63 | 8161 | gen6_check_mch_setup(dev_priv); |
6f1d69b0 ED |
8162 | } |
8163 | ||
8164 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) | |
8165 | { | |
8166 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); | |
8167 | ||
3aad9059 | 8168 | /* |
46680e0a | 8169 | * WaVSThreadDispatchOverride:ivb,vlv |
3aad9059 VS |
8170 | * |
8171 | * This actually overrides the dispatch | |
8172 | * mode for all thread types. | |
8173 | */ | |
6f1d69b0 ED |
8174 | reg &= ~GEN7_FF_SCHED_MASK; |
8175 | reg |= GEN7_FF_TS_SCHED_HW; | |
8176 | reg |= GEN7_FF_VS_SCHED_HW; | |
8177 | reg |= GEN7_FF_DS_SCHED_HW; | |
8178 | ||
8179 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); | |
8180 | } | |
8181 | ||
46f16e63 | 8182 | static void lpt_init_clock_gating(struct drm_i915_private *dev_priv) |
17a303ec | 8183 | { |
17a303ec PZ |
8184 | /* |
8185 | * TODO: this bit should only be enabled when really needed, then | |
8186 | * disabled when not needed anymore in order to save power. | |
8187 | */ | |
4f8036a2 | 8188 | if (HAS_PCH_LPT_LP(dev_priv)) |
17a303ec PZ |
8189 | I915_WRITE(SOUTH_DSPCLK_GATE_D, |
8190 | I915_READ(SOUTH_DSPCLK_GATE_D) | | |
8191 | PCH_LP_PARTITION_LEVEL_DISABLE); | |
0a790cdb PZ |
8192 | |
8193 | /* WADPOClockGatingDisable:hsw */ | |
36c0d0cf VS |
8194 | I915_WRITE(TRANS_CHICKEN1(PIPE_A), |
8195 | I915_READ(TRANS_CHICKEN1(PIPE_A)) | | |
0a790cdb | 8196 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
17a303ec PZ |
8197 | } |
8198 | ||
712bf364 | 8199 | static void lpt_suspend_hw(struct drm_i915_private *dev_priv) |
7d708ee4 | 8200 | { |
4f8036a2 | 8201 | if (HAS_PCH_LPT_LP(dev_priv)) { |
7d708ee4 ID |
8202 | uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D); |
8203 | ||
8204 | val &= ~PCH_LP_PARTITION_LEVEL_DISABLE; | |
8205 | I915_WRITE(SOUTH_DSPCLK_GATE_D, val); | |
8206 | } | |
8207 | } | |
8208 | ||
450174fe ID |
8209 | static void gen8_set_l3sqc_credits(struct drm_i915_private *dev_priv, |
8210 | int general_prio_credits, | |
8211 | int high_prio_credits) | |
8212 | { | |
8213 | u32 misccpctl; | |
8214 | ||
8215 | /* WaTempDisableDOPClkGating:bdw */ | |
8216 | misccpctl = I915_READ(GEN7_MISCCPCTL); | |
8217 | I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE); | |
8218 | ||
8219 | I915_WRITE(GEN8_L3SQCREG1, | |
8220 | L3_GENERAL_PRIO_CREDITS(general_prio_credits) | | |
8221 | L3_HIGH_PRIO_CREDITS(high_prio_credits)); | |
8222 | ||
8223 | /* | |
8224 | * Wait at least 100 clocks before re-enabling clock gating. | |
8225 | * See the definition of L3SQCREG1 in BSpec. | |
8226 | */ | |
8227 | POSTING_READ(GEN8_L3SQCREG1); | |
8228 | udelay(1); | |
8229 | I915_WRITE(GEN7_MISCCPCTL, misccpctl); | |
8230 | } | |
8231 | ||
46f16e63 | 8232 | static void kabylake_init_clock_gating(struct drm_i915_private *dev_priv) |
9498dba7 | 8233 | { |
46f16e63 | 8234 | gen9_init_clock_gating(dev_priv); |
9498dba7 MK |
8235 | |
8236 | /* WaDisableSDEUnitClockGating:kbl */ | |
8237 | if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0)) | |
8238 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
8239 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
8aeb7f62 MK |
8240 | |
8241 | /* WaDisableGamClockGating:kbl */ | |
8242 | if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0)) | |
8243 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | | |
8244 | GEN6_GAMUNIT_CLOCK_GATE_DISABLE); | |
031cd8c8 | 8245 | |
82525c17 | 8246 | /* WaFbcNukeOnHostModify:kbl,cfl */ |
031cd8c8 MK |
8247 | I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) | |
8248 | ILK_DPFC_NUKE_ON_ANY_MODIFICATION); | |
9498dba7 MK |
8249 | } |
8250 | ||
46f16e63 | 8251 | static void skylake_init_clock_gating(struct drm_i915_private *dev_priv) |
dc00b6a0 | 8252 | { |
46f16e63 | 8253 | gen9_init_clock_gating(dev_priv); |
44fff99f MK |
8254 | |
8255 | /* WAC6entrylatency:skl */ | |
8256 | I915_WRITE(FBC_LLC_READ_CTRL, I915_READ(FBC_LLC_READ_CTRL) | | |
8257 | FBC_LLC_FULLY_OPEN); | |
031cd8c8 MK |
8258 | |
8259 | /* WaFbcNukeOnHostModify:skl */ | |
8260 | I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) | | |
8261 | ILK_DPFC_NUKE_ON_ANY_MODIFICATION); | |
dc00b6a0 DV |
8262 | } |
8263 | ||
46f16e63 | 8264 | static void broadwell_init_clock_gating(struct drm_i915_private *dev_priv) |
1020a5c2 | 8265 | { |
07d27e20 | 8266 | enum pipe pipe; |
1020a5c2 | 8267 | |
46f16e63 | 8268 | ilk_init_lp_watermarks(dev_priv); |
50ed5fbd | 8269 | |
ab57fff1 | 8270 | /* WaSwitchSolVfFArbitrationPriority:bdw */ |
50ed5fbd | 8271 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
fe4ab3ce | 8272 | |
ab57fff1 | 8273 | /* WaPsrDPAMaskVBlankInSRD:bdw */ |
fe4ab3ce BW |
8274 | I915_WRITE(CHICKEN_PAR1_1, |
8275 | I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD); | |
8276 | ||
ab57fff1 | 8277 | /* WaPsrDPRSUnmaskVBlankInSRD:bdw */ |
055e393f | 8278 | for_each_pipe(dev_priv, pipe) { |
07d27e20 | 8279 | I915_WRITE(CHICKEN_PIPESL_1(pipe), |
c7c65622 | 8280 | I915_READ(CHICKEN_PIPESL_1(pipe)) | |
8f670bb1 | 8281 | BDW_DPRS_MASK_VBLANK_SRD); |
fe4ab3ce | 8282 | } |
63801f21 | 8283 | |
ab57fff1 BW |
8284 | /* WaVSRefCountFullforceMissDisable:bdw */ |
8285 | /* WaDSRefCountFullforceMissDisable:bdw */ | |
8286 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
8287 | I915_READ(GEN7_FF_THREAD_MODE) & | |
8288 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); | |
36075a4c | 8289 | |
295e8bb7 VS |
8290 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
8291 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); | |
4f1ca9e9 VS |
8292 | |
8293 | /* WaDisableSDEUnitClockGating:bdw */ | |
8294 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
8295 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
5d708680 | 8296 | |
450174fe ID |
8297 | /* WaProgramL3SqcReg1Default:bdw */ |
8298 | gen8_set_l3sqc_credits(dev_priv, 30, 2); | |
4d487cff | 8299 | |
6d50b065 VS |
8300 | /* |
8301 | * WaGttCachingOffByDefault:bdw | |
8302 | * GTT cache may not work with big pages, so if those | |
8303 | * are ever enabled GTT cache may need to be disabled. | |
8304 | */ | |
8305 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); | |
8306 | ||
17e0adf0 MK |
8307 | /* WaKVMNotificationOnConfigChange:bdw */ |
8308 | I915_WRITE(CHICKEN_PAR2_1, I915_READ(CHICKEN_PAR2_1) | |
8309 | | KVM_CONFIG_CHANGE_NOTIFICATION_SELECT); | |
8310 | ||
46f16e63 | 8311 | lpt_init_clock_gating(dev_priv); |
9cc19733 RB |
8312 | |
8313 | /* WaDisableDopClockGating:bdw | |
8314 | * | |
8315 | * Also see the CHICKEN2 write in bdw_init_workarounds() to disable DOP | |
8316 | * clock gating. | |
8317 | */ | |
8318 | I915_WRITE(GEN6_UCGCTL1, | |
8319 | I915_READ(GEN6_UCGCTL1) | GEN6_EU_TCUNIT_CLOCK_GATE_DISABLE); | |
1020a5c2 BW |
8320 | } |
8321 | ||
46f16e63 | 8322 | static void haswell_init_clock_gating(struct drm_i915_private *dev_priv) |
cad2a2d7 | 8323 | { |
46f16e63 | 8324 | ilk_init_lp_watermarks(dev_priv); |
cad2a2d7 | 8325 | |
f3fc4884 FJ |
8326 | /* L3 caching of data atomics doesn't work -- disable it. */ |
8327 | I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE); | |
8328 | I915_WRITE(HSW_ROW_CHICKEN3, | |
8329 | _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE)); | |
8330 | ||
ecdb4eb7 | 8331 | /* This is required by WaCatErrorRejectionIssue:hsw */ |
cad2a2d7 ED |
8332 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
8333 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
8334 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
8335 | ||
e36ea7ff VS |
8336 | /* WaVSRefCountFullforceMissDisable:hsw */ |
8337 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
8338 | I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME); | |
cad2a2d7 | 8339 | |
4e04632e AG |
8340 | /* WaDisable_RenderCache_OperationalFlush:hsw */ |
8341 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
8342 | ||
fe27c606 CW |
8343 | /* enable HiZ Raw Stall Optimization */ |
8344 | I915_WRITE(CACHE_MODE_0_GEN7, | |
8345 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); | |
8346 | ||
ecdb4eb7 | 8347 | /* WaDisable4x2SubspanOptimization:hsw */ |
cad2a2d7 ED |
8348 | I915_WRITE(CACHE_MODE_1, |
8349 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
1544d9d5 | 8350 | |
a12c4967 VS |
8351 | /* |
8352 | * BSpec recommends 8x4 when MSAA is used, | |
8353 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
8354 | * |
8355 | * Note that PS/WM thread counts depend on the WIZ hashing | |
8356 | * disable bit, which we don't touch here, but it's good | |
8357 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
a12c4967 VS |
8358 | */ |
8359 | I915_WRITE(GEN7_GT_MODE, | |
98533251 | 8360 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
a12c4967 | 8361 | |
94411593 KG |
8362 | /* WaSampleCChickenBitEnable:hsw */ |
8363 | I915_WRITE(HALF_SLICE_CHICKEN3, | |
8364 | _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE)); | |
8365 | ||
ecdb4eb7 | 8366 | /* WaSwitchSolVfFArbitrationPriority:hsw */ |
e3dff585 BW |
8367 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
8368 | ||
90a88643 PZ |
8369 | /* WaRsPkgCStateDisplayPMReq:hsw */ |
8370 | I915_WRITE(CHICKEN_PAR1_1, | |
8371 | I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES); | |
1544d9d5 | 8372 | |
46f16e63 | 8373 | lpt_init_clock_gating(dev_priv); |
cad2a2d7 ED |
8374 | } |
8375 | ||
46f16e63 | 8376 | static void ivybridge_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8377 | { |
20848223 | 8378 | uint32_t snpcr; |
6f1d69b0 | 8379 | |
46f16e63 | 8380 | ilk_init_lp_watermarks(dev_priv); |
6f1d69b0 | 8381 | |
231e54f6 | 8382 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
6f1d69b0 | 8383 | |
ecdb4eb7 | 8384 | /* WaDisableEarlyCull:ivb */ |
87f8020e JB |
8385 | I915_WRITE(_3D_CHICKEN3, |
8386 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
8387 | ||
ecdb4eb7 | 8388 | /* WaDisableBackToBackFlipFix:ivb */ |
6f1d69b0 ED |
8389 | I915_WRITE(IVB_CHICKEN3, |
8390 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
8391 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
8392 | ||
ecdb4eb7 | 8393 | /* WaDisablePSDDualDispatchEnable:ivb */ |
50a0bc90 | 8394 | if (IS_IVB_GT1(dev_priv)) |
12f3382b JB |
8395 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
8396 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 8397 | |
4e04632e AG |
8398 | /* WaDisable_RenderCache_OperationalFlush:ivb */ |
8399 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
8400 | ||
ecdb4eb7 | 8401 | /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */ |
6f1d69b0 ED |
8402 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
8403 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); | |
8404 | ||
ecdb4eb7 | 8405 | /* WaApplyL3ControlAndL3ChickenMode:ivb */ |
6f1d69b0 ED |
8406 | I915_WRITE(GEN7_L3CNTLREG1, |
8407 | GEN7_WA_FOR_GEN7_L3_CONTROL); | |
8408 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, | |
8ab43976 | 8409 | GEN7_WA_L3_CHICKEN_MODE); |
50a0bc90 | 8410 | if (IS_IVB_GT1(dev_priv)) |
8ab43976 JB |
8411 | I915_WRITE(GEN7_ROW_CHICKEN2, |
8412 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
412236c2 VS |
8413 | else { |
8414 | /* must write both registers */ | |
8415 | I915_WRITE(GEN7_ROW_CHICKEN2, | |
8416 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
8ab43976 JB |
8417 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, |
8418 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
412236c2 | 8419 | } |
6f1d69b0 | 8420 | |
ecdb4eb7 | 8421 | /* WaForceL3Serialization:ivb */ |
61939d97 JB |
8422 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
8423 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
8424 | ||
1b80a19a | 8425 | /* |
0f846f81 | 8426 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
ecdb4eb7 | 8427 | * This implements the WaDisableRCZUnitClockGating:ivb workaround. |
0f846f81 JB |
8428 | */ |
8429 | I915_WRITE(GEN6_UCGCTL2, | |
28acf3b2 | 8430 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
0f846f81 | 8431 | |
ecdb4eb7 | 8432 | /* This is required by WaCatErrorRejectionIssue:ivb */ |
6f1d69b0 ED |
8433 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
8434 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
8435 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
8436 | ||
46f16e63 | 8437 | g4x_disable_trickle_feed(dev_priv); |
6f1d69b0 ED |
8438 | |
8439 | gen7_setup_fixed_func_scheduler(dev_priv); | |
97e1930f | 8440 | |
22721343 CW |
8441 | if (0) { /* causes HiZ corruption on ivb:gt1 */ |
8442 | /* enable HiZ Raw Stall Optimization */ | |
8443 | I915_WRITE(CACHE_MODE_0_GEN7, | |
8444 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); | |
8445 | } | |
116f2b6d | 8446 | |
ecdb4eb7 | 8447 | /* WaDisable4x2SubspanOptimization:ivb */ |
97e1930f DV |
8448 | I915_WRITE(CACHE_MODE_1, |
8449 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
20848223 | 8450 | |
a607c1a4 VS |
8451 | /* |
8452 | * BSpec recommends 8x4 when MSAA is used, | |
8453 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
8454 | * |
8455 | * Note that PS/WM thread counts depend on the WIZ hashing | |
8456 | * disable bit, which we don't touch here, but it's good | |
8457 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
a607c1a4 VS |
8458 | */ |
8459 | I915_WRITE(GEN7_GT_MODE, | |
98533251 | 8460 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
a607c1a4 | 8461 | |
20848223 BW |
8462 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); |
8463 | snpcr &= ~GEN6_MBC_SNPCR_MASK; | |
8464 | snpcr |= GEN6_MBC_SNPCR_MED; | |
8465 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); | |
3107bd48 | 8466 | |
6e266956 | 8467 | if (!HAS_PCH_NOP(dev_priv)) |
46f16e63 | 8468 | cpt_init_clock_gating(dev_priv); |
1d7aaa0c | 8469 | |
46f16e63 | 8470 | gen6_check_mch_setup(dev_priv); |
6f1d69b0 ED |
8471 | } |
8472 | ||
46f16e63 | 8473 | static void valleyview_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8474 | { |
ecdb4eb7 | 8475 | /* WaDisableEarlyCull:vlv */ |
87f8020e JB |
8476 | I915_WRITE(_3D_CHICKEN3, |
8477 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
8478 | ||
ecdb4eb7 | 8479 | /* WaDisableBackToBackFlipFix:vlv */ |
6f1d69b0 ED |
8480 | I915_WRITE(IVB_CHICKEN3, |
8481 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
8482 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
8483 | ||
fad7d36e | 8484 | /* WaPsdDispatchEnable:vlv */ |
ecdb4eb7 | 8485 | /* WaDisablePSDDualDispatchEnable:vlv */ |
12f3382b | 8486 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
d3bc0303 JB |
8487 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
8488 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 8489 | |
4e04632e AG |
8490 | /* WaDisable_RenderCache_OperationalFlush:vlv */ |
8491 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
8492 | ||
ecdb4eb7 | 8493 | /* WaForceL3Serialization:vlv */ |
61939d97 JB |
8494 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
8495 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
8496 | ||
ecdb4eb7 | 8497 | /* WaDisableDopClockGating:vlv */ |
8ab43976 JB |
8498 | I915_WRITE(GEN7_ROW_CHICKEN2, |
8499 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
8500 | ||
ecdb4eb7 | 8501 | /* This is required by WaCatErrorRejectionIssue:vlv */ |
6f1d69b0 ED |
8502 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
8503 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
8504 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
8505 | ||
46680e0a VS |
8506 | gen7_setup_fixed_func_scheduler(dev_priv); |
8507 | ||
3c0edaeb | 8508 | /* |
0f846f81 | 8509 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
ecdb4eb7 | 8510 | * This implements the WaDisableRCZUnitClockGating:vlv workaround. |
0f846f81 JB |
8511 | */ |
8512 | I915_WRITE(GEN6_UCGCTL2, | |
3c0edaeb | 8513 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
0f846f81 | 8514 | |
c98f5062 AG |
8515 | /* WaDisableL3Bank2xClockGate:vlv |
8516 | * Disabling L3 clock gating- MMIO 940c[25] = 1 | |
8517 | * Set bit 25, to disable L3_BANK_2x_CLK_GATING */ | |
8518 | I915_WRITE(GEN7_UCGCTL4, | |
8519 | I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE); | |
e3f33d46 | 8520 | |
afd58e79 VS |
8521 | /* |
8522 | * BSpec says this must be set, even though | |
8523 | * WaDisable4x2SubspanOptimization isn't listed for VLV. | |
8524 | */ | |
6b26c86d DV |
8525 | I915_WRITE(CACHE_MODE_1, |
8526 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
7983117f | 8527 | |
da2518f9 VS |
8528 | /* |
8529 | * BSpec recommends 8x4 when MSAA is used, | |
8530 | * however in practice 16x4 seems fastest. | |
8531 | * | |
8532 | * Note that PS/WM thread counts depend on the WIZ hashing | |
8533 | * disable bit, which we don't touch here, but it's good | |
8534 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
8535 | */ | |
8536 | I915_WRITE(GEN7_GT_MODE, | |
8537 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); | |
8538 | ||
031994ee VS |
8539 | /* |
8540 | * WaIncreaseL3CreditsForVLVB0:vlv | |
8541 | * This is the hardware default actually. | |
8542 | */ | |
8543 | I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE); | |
8544 | ||
2d809570 | 8545 | /* |
ecdb4eb7 | 8546 | * WaDisableVLVClockGating_VBIIssue:vlv |
2d809570 JB |
8547 | * Disable clock gating on th GCFG unit to prevent a delay |
8548 | * in the reporting of vblank events. | |
8549 | */ | |
7a0d1eed | 8550 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS); |
6f1d69b0 ED |
8551 | } |
8552 | ||
46f16e63 | 8553 | static void cherryview_init_clock_gating(struct drm_i915_private *dev_priv) |
a4565da8 | 8554 | { |
232ce337 VS |
8555 | /* WaVSRefCountFullforceMissDisable:chv */ |
8556 | /* WaDSRefCountFullforceMissDisable:chv */ | |
8557 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
8558 | I915_READ(GEN7_FF_THREAD_MODE) & | |
8559 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); | |
acea6f95 VS |
8560 | |
8561 | /* WaDisableSemaphoreAndSyncFlipWait:chv */ | |
8562 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, | |
8563 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); | |
0846697c VS |
8564 | |
8565 | /* WaDisableCSUnitClockGating:chv */ | |
8566 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | | |
8567 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
c631780f VS |
8568 | |
8569 | /* WaDisableSDEUnitClockGating:chv */ | |
8570 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
8571 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
6d50b065 | 8572 | |
450174fe ID |
8573 | /* |
8574 | * WaProgramL3SqcReg1Default:chv | |
8575 | * See gfxspecs/Related Documents/Performance Guide/ | |
8576 | * LSQC Setting Recommendations. | |
8577 | */ | |
8578 | gen8_set_l3sqc_credits(dev_priv, 38, 2); | |
8579 | ||
6d50b065 VS |
8580 | /* |
8581 | * GTT cache may not work with big pages, so if those | |
8582 | * are ever enabled GTT cache may need to be disabled. | |
8583 | */ | |
8584 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); | |
a4565da8 VS |
8585 | } |
8586 | ||
46f16e63 | 8587 | static void g4x_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8588 | { |
6f1d69b0 ED |
8589 | uint32_t dspclk_gate; |
8590 | ||
8591 | I915_WRITE(RENCLK_GATE_D1, 0); | |
8592 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | | |
8593 | GS_UNIT_CLOCK_GATE_DISABLE | | |
8594 | CL_UNIT_CLOCK_GATE_DISABLE); | |
8595 | I915_WRITE(RAMCLK_GATE_D, 0); | |
8596 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | | |
8597 | OVRUNIT_CLOCK_GATE_DISABLE | | |
8598 | OVCUNIT_CLOCK_GATE_DISABLE; | |
50a0bc90 | 8599 | if (IS_GM45(dev_priv)) |
6f1d69b0 ED |
8600 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; |
8601 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); | |
4358a374 DV |
8602 | |
8603 | /* WaDisableRenderCachePipelinedFlush */ | |
8604 | I915_WRITE(CACHE_MODE_0, | |
8605 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
de1aa629 | 8606 | |
4e04632e AG |
8607 | /* WaDisable_RenderCache_OperationalFlush:g4x */ |
8608 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
8609 | ||
46f16e63 | 8610 | g4x_disable_trickle_feed(dev_priv); |
6f1d69b0 ED |
8611 | } |
8612 | ||
46f16e63 | 8613 | static void crestline_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8614 | { |
6f1d69b0 ED |
8615 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); |
8616 | I915_WRITE(RENCLK_GATE_D2, 0); | |
8617 | I915_WRITE(DSPCLK_GATE_D, 0); | |
8618 | I915_WRITE(RAMCLK_GATE_D, 0); | |
8619 | I915_WRITE16(DEUC, 0); | |
20f94967 VS |
8620 | I915_WRITE(MI_ARB_STATE, |
8621 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
4e04632e AG |
8622 | |
8623 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ | |
8624 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6f1d69b0 ED |
8625 | } |
8626 | ||
46f16e63 | 8627 | static void broadwater_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8628 | { |
6f1d69b0 ED |
8629 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | |
8630 | I965_RCC_CLOCK_GATE_DISABLE | | |
8631 | I965_RCPB_CLOCK_GATE_DISABLE | | |
8632 | I965_ISC_CLOCK_GATE_DISABLE | | |
8633 | I965_FBC_CLOCK_GATE_DISABLE); | |
8634 | I915_WRITE(RENCLK_GATE_D2, 0); | |
20f94967 VS |
8635 | I915_WRITE(MI_ARB_STATE, |
8636 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
4e04632e AG |
8637 | |
8638 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ | |
8639 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6f1d69b0 ED |
8640 | } |
8641 | ||
46f16e63 | 8642 | static void gen3_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8643 | { |
6f1d69b0 ED |
8644 | u32 dstate = I915_READ(D_STATE); |
8645 | ||
8646 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | | |
8647 | DSTATE_DOT_CLOCK_GATING; | |
8648 | I915_WRITE(D_STATE, dstate); | |
13a86b85 | 8649 | |
9b1e14f4 | 8650 | if (IS_PINEVIEW(dev_priv)) |
13a86b85 | 8651 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); |
974a3b0f DV |
8652 | |
8653 | /* IIR "flip pending" means done if this bit is set */ | |
8654 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); | |
12fabbcb VS |
8655 | |
8656 | /* interrupts should cause a wake up from C3 */ | |
3299254f | 8657 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN)); |
dbb42748 VS |
8658 | |
8659 | /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ | |
8660 | I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); | |
1038392b VS |
8661 | |
8662 | I915_WRITE(MI_ARB_STATE, | |
8663 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
8664 | } |
8665 | ||
46f16e63 | 8666 | static void i85x_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8667 | { |
6f1d69b0 | 8668 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); |
54e472ae VS |
8669 | |
8670 | /* interrupts should cause a wake up from C3 */ | |
8671 | I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) | | |
8672 | _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE)); | |
1038392b VS |
8673 | |
8674 | I915_WRITE(MEM_MODE, | |
8675 | _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
8676 | } |
8677 | ||
46f16e63 | 8678 | static void i830_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8679 | { |
1038392b VS |
8680 | I915_WRITE(MEM_MODE, |
8681 | _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) | | |
8682 | _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
8683 | } |
8684 | ||
46f16e63 | 8685 | void intel_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 8686 | { |
46f16e63 | 8687 | dev_priv->display.init_clock_gating(dev_priv); |
6f1d69b0 ED |
8688 | } |
8689 | ||
712bf364 | 8690 | void intel_suspend_hw(struct drm_i915_private *dev_priv) |
7d708ee4 | 8691 | { |
712bf364 VS |
8692 | if (HAS_PCH_LPT(dev_priv)) |
8693 | lpt_suspend_hw(dev_priv); | |
7d708ee4 ID |
8694 | } |
8695 | ||
46f16e63 | 8696 | static void nop_init_clock_gating(struct drm_i915_private *dev_priv) |
bb400da9 ID |
8697 | { |
8698 | DRM_DEBUG_KMS("No clock gating settings or workarounds applied.\n"); | |
8699 | } | |
8700 | ||
8701 | /** | |
8702 | * intel_init_clock_gating_hooks - setup the clock gating hooks | |
8703 | * @dev_priv: device private | |
8704 | * | |
8705 | * Setup the hooks that configure which clocks of a given platform can be | |
8706 | * gated and also apply various GT and display specific workarounds for these | |
8707 | * platforms. Note that some GT specific workarounds are applied separately | |
8708 | * when GPU contexts or batchbuffers start their execution. | |
8709 | */ | |
8710 | void intel_init_clock_gating_hooks(struct drm_i915_private *dev_priv) | |
8711 | { | |
8712 | if (IS_SKYLAKE(dev_priv)) | |
dc00b6a0 | 8713 | dev_priv->display.init_clock_gating = skylake_init_clock_gating; |
82525c17 | 8714 | else if (IS_KABYLAKE(dev_priv) || IS_COFFEELAKE(dev_priv)) |
9498dba7 | 8715 | dev_priv->display.init_clock_gating = kabylake_init_clock_gating; |
9fb5026f | 8716 | else if (IS_BROXTON(dev_priv)) |
bb400da9 | 8717 | dev_priv->display.init_clock_gating = bxt_init_clock_gating; |
9fb5026f ACO |
8718 | else if (IS_GEMINILAKE(dev_priv)) |
8719 | dev_priv->display.init_clock_gating = glk_init_clock_gating; | |
bb400da9 ID |
8720 | else if (IS_BROADWELL(dev_priv)) |
8721 | dev_priv->display.init_clock_gating = broadwell_init_clock_gating; | |
8722 | else if (IS_CHERRYVIEW(dev_priv)) | |
8723 | dev_priv->display.init_clock_gating = cherryview_init_clock_gating; | |
8724 | else if (IS_HASWELL(dev_priv)) | |
8725 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; | |
8726 | else if (IS_IVYBRIDGE(dev_priv)) | |
8727 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; | |
8728 | else if (IS_VALLEYVIEW(dev_priv)) | |
8729 | dev_priv->display.init_clock_gating = valleyview_init_clock_gating; | |
8730 | else if (IS_GEN6(dev_priv)) | |
8731 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; | |
8732 | else if (IS_GEN5(dev_priv)) | |
8733 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; | |
8734 | else if (IS_G4X(dev_priv)) | |
8735 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; | |
c0f86832 | 8736 | else if (IS_I965GM(dev_priv)) |
bb400da9 | 8737 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; |
c0f86832 | 8738 | else if (IS_I965G(dev_priv)) |
bb400da9 ID |
8739 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; |
8740 | else if (IS_GEN3(dev_priv)) | |
8741 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; | |
8742 | else if (IS_I85X(dev_priv) || IS_I865G(dev_priv)) | |
8743 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; | |
8744 | else if (IS_GEN2(dev_priv)) | |
8745 | dev_priv->display.init_clock_gating = i830_init_clock_gating; | |
8746 | else { | |
8747 | MISSING_CASE(INTEL_DEVID(dev_priv)); | |
8748 | dev_priv->display.init_clock_gating = nop_init_clock_gating; | |
8749 | } | |
8750 | } | |
8751 | ||
1fa61106 | 8752 | /* Set up chip specific power management-related functions */ |
62d75df7 | 8753 | void intel_init_pm(struct drm_i915_private *dev_priv) |
1fa61106 | 8754 | { |
7ff0ebcc | 8755 | intel_fbc_init(dev_priv); |
1fa61106 | 8756 | |
c921aba8 | 8757 | /* For cxsr */ |
9b1e14f4 | 8758 | if (IS_PINEVIEW(dev_priv)) |
148ac1f3 | 8759 | i915_pineview_get_mem_freq(dev_priv); |
5db94019 | 8760 | else if (IS_GEN5(dev_priv)) |
148ac1f3 | 8761 | i915_ironlake_get_mem_freq(dev_priv); |
c921aba8 | 8762 | |
1fa61106 | 8763 | /* For FIFO watermark updates */ |
62d75df7 | 8764 | if (INTEL_GEN(dev_priv) >= 9) { |
bb726519 | 8765 | skl_setup_wm_latency(dev_priv); |
e62929b3 | 8766 | dev_priv->display.initial_watermarks = skl_initial_wm; |
ccf010fb | 8767 | dev_priv->display.atomic_update_watermarks = skl_atomic_update_crtc_wm; |
98d39494 | 8768 | dev_priv->display.compute_global_watermarks = skl_compute_wm; |
6e266956 | 8769 | } else if (HAS_PCH_SPLIT(dev_priv)) { |
bb726519 | 8770 | ilk_setup_wm_latency(dev_priv); |
53615a5e | 8771 | |
5db94019 | 8772 | if ((IS_GEN5(dev_priv) && dev_priv->wm.pri_latency[1] && |
bd602544 | 8773 | dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) || |
5db94019 | 8774 | (!IS_GEN5(dev_priv) && dev_priv->wm.pri_latency[0] && |
bd602544 | 8775 | dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) { |
86c8bbbe | 8776 | dev_priv->display.compute_pipe_wm = ilk_compute_pipe_wm; |
ed4a6a7c MR |
8777 | dev_priv->display.compute_intermediate_wm = |
8778 | ilk_compute_intermediate_wm; | |
8779 | dev_priv->display.initial_watermarks = | |
8780 | ilk_initial_watermarks; | |
8781 | dev_priv->display.optimize_watermarks = | |
8782 | ilk_optimize_watermarks; | |
bd602544 VS |
8783 | } else { |
8784 | DRM_DEBUG_KMS("Failed to read display plane latency. " | |
8785 | "Disable CxSR\n"); | |
8786 | } | |
6b6b3eef | 8787 | } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { |
bb726519 | 8788 | vlv_setup_wm_latency(dev_priv); |
ff32c54e | 8789 | dev_priv->display.compute_pipe_wm = vlv_compute_pipe_wm; |
4841da51 | 8790 | dev_priv->display.compute_intermediate_wm = vlv_compute_intermediate_wm; |
ff32c54e | 8791 | dev_priv->display.initial_watermarks = vlv_initial_watermarks; |
4841da51 | 8792 | dev_priv->display.optimize_watermarks = vlv_optimize_watermarks; |
ff32c54e | 8793 | dev_priv->display.atomic_update_watermarks = vlv_atomic_update_fifo; |
04548cba VS |
8794 | } else if (IS_G4X(dev_priv)) { |
8795 | g4x_setup_wm_latency(dev_priv); | |
8796 | dev_priv->display.compute_pipe_wm = g4x_compute_pipe_wm; | |
8797 | dev_priv->display.compute_intermediate_wm = g4x_compute_intermediate_wm; | |
8798 | dev_priv->display.initial_watermarks = g4x_initial_watermarks; | |
8799 | dev_priv->display.optimize_watermarks = g4x_optimize_watermarks; | |
9b1e14f4 | 8800 | } else if (IS_PINEVIEW(dev_priv)) { |
50a0bc90 | 8801 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev_priv), |
1fa61106 ED |
8802 | dev_priv->is_ddr3, |
8803 | dev_priv->fsb_freq, | |
8804 | dev_priv->mem_freq)) { | |
8805 | DRM_INFO("failed to find known CxSR latency " | |
8806 | "(found ddr%s fsb freq %d, mem freq %d), " | |
8807 | "disabling CxSR\n", | |
8808 | (dev_priv->is_ddr3 == 1) ? "3" : "2", | |
8809 | dev_priv->fsb_freq, dev_priv->mem_freq); | |
8810 | /* Disable CxSR and never update its watermark again */ | |
5209b1f4 | 8811 | intel_set_memory_cxsr(dev_priv, false); |
1fa61106 ED |
8812 | dev_priv->display.update_wm = NULL; |
8813 | } else | |
8814 | dev_priv->display.update_wm = pineview_update_wm; | |
5db94019 | 8815 | } else if (IS_GEN4(dev_priv)) { |
1fa61106 | 8816 | dev_priv->display.update_wm = i965_update_wm; |
5db94019 | 8817 | } else if (IS_GEN3(dev_priv)) { |
1fa61106 ED |
8818 | dev_priv->display.update_wm = i9xx_update_wm; |
8819 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; | |
5db94019 | 8820 | } else if (IS_GEN2(dev_priv)) { |
62d75df7 | 8821 | if (INTEL_INFO(dev_priv)->num_pipes == 1) { |
feb56b93 | 8822 | dev_priv->display.update_wm = i845_update_wm; |
1fa61106 | 8823 | dev_priv->display.get_fifo_size = i845_get_fifo_size; |
feb56b93 DV |
8824 | } else { |
8825 | dev_priv->display.update_wm = i9xx_update_wm; | |
1fa61106 | 8826 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
feb56b93 | 8827 | } |
feb56b93 DV |
8828 | } else { |
8829 | DRM_ERROR("unexpected fall-through in intel_init_pm\n"); | |
1fa61106 ED |
8830 | } |
8831 | } | |
8832 | ||
87660502 L |
8833 | static inline int gen6_check_mailbox_status(struct drm_i915_private *dev_priv) |
8834 | { | |
8835 | uint32_t flags = | |
8836 | I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_ERROR_MASK; | |
8837 | ||
8838 | switch (flags) { | |
8839 | case GEN6_PCODE_SUCCESS: | |
8840 | return 0; | |
8841 | case GEN6_PCODE_UNIMPLEMENTED_CMD: | |
8842 | case GEN6_PCODE_ILLEGAL_CMD: | |
8843 | return -ENXIO; | |
8844 | case GEN6_PCODE_MIN_FREQ_TABLE_GT_RATIO_OUT_OF_RANGE: | |
7850d1c3 | 8845 | case GEN7_PCODE_MIN_FREQ_TABLE_GT_RATIO_OUT_OF_RANGE: |
87660502 L |
8846 | return -EOVERFLOW; |
8847 | case GEN6_PCODE_TIMEOUT: | |
8848 | return -ETIMEDOUT; | |
8849 | default: | |
f0d66153 | 8850 | MISSING_CASE(flags); |
87660502 L |
8851 | return 0; |
8852 | } | |
8853 | } | |
8854 | ||
8855 | static inline int gen7_check_mailbox_status(struct drm_i915_private *dev_priv) | |
8856 | { | |
8857 | uint32_t flags = | |
8858 | I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_ERROR_MASK; | |
8859 | ||
8860 | switch (flags) { | |
8861 | case GEN6_PCODE_SUCCESS: | |
8862 | return 0; | |
8863 | case GEN6_PCODE_ILLEGAL_CMD: | |
8864 | return -ENXIO; | |
8865 | case GEN7_PCODE_TIMEOUT: | |
8866 | return -ETIMEDOUT; | |
8867 | case GEN7_PCODE_ILLEGAL_DATA: | |
8868 | return -EINVAL; | |
8869 | case GEN7_PCODE_MIN_FREQ_TABLE_GT_RATIO_OUT_OF_RANGE: | |
8870 | return -EOVERFLOW; | |
8871 | default: | |
8872 | MISSING_CASE(flags); | |
8873 | return 0; | |
8874 | } | |
8875 | } | |
8876 | ||
151a49d0 | 8877 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val) |
42c0526c | 8878 | { |
87660502 L |
8879 | int status; |
8880 | ||
4fc688ce | 8881 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c | 8882 | |
3f5582dd CW |
8883 | /* GEN6_PCODE_* are outside of the forcewake domain, we can |
8884 | * use te fw I915_READ variants to reduce the amount of work | |
8885 | * required when reading/writing. | |
8886 | */ | |
8887 | ||
8888 | if (I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
42c0526c BW |
8889 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); |
8890 | return -EAGAIN; | |
8891 | } | |
8892 | ||
3f5582dd CW |
8893 | I915_WRITE_FW(GEN6_PCODE_DATA, *val); |
8894 | I915_WRITE_FW(GEN6_PCODE_DATA1, 0); | |
8895 | I915_WRITE_FW(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); | |
42c0526c | 8896 | |
e09a3036 CW |
8897 | if (__intel_wait_for_register_fw(dev_priv, |
8898 | GEN6_PCODE_MAILBOX, GEN6_PCODE_READY, 0, | |
8899 | 500, 0, NULL)) { | |
42c0526c BW |
8900 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); |
8901 | return -ETIMEDOUT; | |
8902 | } | |
8903 | ||
3f5582dd CW |
8904 | *val = I915_READ_FW(GEN6_PCODE_DATA); |
8905 | I915_WRITE_FW(GEN6_PCODE_DATA, 0); | |
42c0526c | 8906 | |
87660502 L |
8907 | if (INTEL_GEN(dev_priv) > 6) |
8908 | status = gen7_check_mailbox_status(dev_priv); | |
8909 | else | |
8910 | status = gen6_check_mailbox_status(dev_priv); | |
8911 | ||
8912 | if (status) { | |
8913 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed: %d\n", | |
8914 | status); | |
8915 | return status; | |
8916 | } | |
8917 | ||
42c0526c BW |
8918 | return 0; |
8919 | } | |
8920 | ||
3f5582dd | 8921 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, |
87660502 | 8922 | u32 mbox, u32 val) |
42c0526c | 8923 | { |
87660502 L |
8924 | int status; |
8925 | ||
4fc688ce | 8926 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c | 8927 | |
3f5582dd CW |
8928 | /* GEN6_PCODE_* are outside of the forcewake domain, we can |
8929 | * use te fw I915_READ variants to reduce the amount of work | |
8930 | * required when reading/writing. | |
8931 | */ | |
8932 | ||
8933 | if (I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
42c0526c BW |
8934 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); |
8935 | return -EAGAIN; | |
8936 | } | |
8937 | ||
3f5582dd | 8938 | I915_WRITE_FW(GEN6_PCODE_DATA, val); |
8bf41b72 | 8939 | I915_WRITE_FW(GEN6_PCODE_DATA1, 0); |
3f5582dd | 8940 | I915_WRITE_FW(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
42c0526c | 8941 | |
e09a3036 CW |
8942 | if (__intel_wait_for_register_fw(dev_priv, |
8943 | GEN6_PCODE_MAILBOX, GEN6_PCODE_READY, 0, | |
8944 | 500, 0, NULL)) { | |
42c0526c BW |
8945 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); |
8946 | return -ETIMEDOUT; | |
8947 | } | |
8948 | ||
3f5582dd | 8949 | I915_WRITE_FW(GEN6_PCODE_DATA, 0); |
42c0526c | 8950 | |
87660502 L |
8951 | if (INTEL_GEN(dev_priv) > 6) |
8952 | status = gen7_check_mailbox_status(dev_priv); | |
8953 | else | |
8954 | status = gen6_check_mailbox_status(dev_priv); | |
8955 | ||
8956 | if (status) { | |
8957 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed: %d\n", | |
8958 | status); | |
8959 | return status; | |
8960 | } | |
8961 | ||
42c0526c BW |
8962 | return 0; |
8963 | } | |
a0e4e199 | 8964 | |
a0b8a1fe ID |
8965 | static bool skl_pcode_try_request(struct drm_i915_private *dev_priv, u32 mbox, |
8966 | u32 request, u32 reply_mask, u32 reply, | |
8967 | u32 *status) | |
8968 | { | |
8969 | u32 val = request; | |
8970 | ||
8971 | *status = sandybridge_pcode_read(dev_priv, mbox, &val); | |
8972 | ||
8973 | return *status || ((val & reply_mask) == reply); | |
8974 | } | |
8975 | ||
8976 | /** | |
8977 | * skl_pcode_request - send PCODE request until acknowledgment | |
8978 | * @dev_priv: device private | |
8979 | * @mbox: PCODE mailbox ID the request is targeted for | |
8980 | * @request: request ID | |
8981 | * @reply_mask: mask used to check for request acknowledgment | |
8982 | * @reply: value used to check for request acknowledgment | |
8983 | * @timeout_base_ms: timeout for polling with preemption enabled | |
8984 | * | |
8985 | * Keep resending the @request to @mbox until PCODE acknowledges it, PCODE | |
0129936d | 8986 | * reports an error or an overall timeout of @timeout_base_ms+50 ms expires. |
a0b8a1fe ID |
8987 | * The request is acknowledged once the PCODE reply dword equals @reply after |
8988 | * applying @reply_mask. Polling is first attempted with preemption enabled | |
0129936d | 8989 | * for @timeout_base_ms and if this times out for another 50 ms with |
a0b8a1fe ID |
8990 | * preemption disabled. |
8991 | * | |
8992 | * Returns 0 on success, %-ETIMEDOUT in case of a timeout, <0 in case of some | |
8993 | * other error as reported by PCODE. | |
8994 | */ | |
8995 | int skl_pcode_request(struct drm_i915_private *dev_priv, u32 mbox, u32 request, | |
8996 | u32 reply_mask, u32 reply, int timeout_base_ms) | |
8997 | { | |
8998 | u32 status; | |
8999 | int ret; | |
9000 | ||
9001 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
9002 | ||
9003 | #define COND skl_pcode_try_request(dev_priv, mbox, request, reply_mask, reply, \ | |
9004 | &status) | |
9005 | ||
9006 | /* | |
9007 | * Prime the PCODE by doing a request first. Normally it guarantees | |
9008 | * that a subsequent request, at most @timeout_base_ms later, succeeds. | |
9009 | * _wait_for() doesn't guarantee when its passed condition is evaluated | |
9010 | * first, so send the first request explicitly. | |
9011 | */ | |
9012 | if (COND) { | |
9013 | ret = 0; | |
9014 | goto out; | |
9015 | } | |
9016 | ret = _wait_for(COND, timeout_base_ms * 1000, 10); | |
9017 | if (!ret) | |
9018 | goto out; | |
9019 | ||
9020 | /* | |
9021 | * The above can time out if the number of requests was low (2 in the | |
9022 | * worst case) _and_ PCODE was busy for some reason even after a | |
9023 | * (queued) request and @timeout_base_ms delay. As a workaround retry | |
9024 | * the poll with preemption disabled to maximize the number of | |
0129936d | 9025 | * requests. Increase the timeout from @timeout_base_ms to 50ms to |
a0b8a1fe | 9026 | * account for interrupts that could reduce the number of these |
0129936d ID |
9027 | * requests, and for any quirks of the PCODE firmware that delays |
9028 | * the request completion. | |
a0b8a1fe ID |
9029 | */ |
9030 | DRM_DEBUG_KMS("PCODE timeout, retrying with preemption disabled\n"); | |
9031 | WARN_ON_ONCE(timeout_base_ms > 3); | |
9032 | preempt_disable(); | |
0129936d | 9033 | ret = wait_for_atomic(COND, 50); |
a0b8a1fe ID |
9034 | preempt_enable(); |
9035 | ||
9036 | out: | |
9037 | return ret ? ret : status; | |
9038 | #undef COND | |
9039 | } | |
9040 | ||
dd06f88c VS |
9041 | static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val) |
9042 | { | |
c30fec65 VS |
9043 | /* |
9044 | * N = val - 0xb7 | |
9045 | * Slow = Fast = GPLL ref * N | |
9046 | */ | |
9047 | return DIV_ROUND_CLOSEST(dev_priv->rps.gpll_ref_freq * (val - 0xb7), 1000); | |
855ba3be JB |
9048 | } |
9049 | ||
b55dd647 | 9050 | static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val) |
855ba3be | 9051 | { |
c30fec65 | 9052 | return DIV_ROUND_CLOSEST(1000 * val, dev_priv->rps.gpll_ref_freq) + 0xb7; |
855ba3be JB |
9053 | } |
9054 | ||
b55dd647 | 9055 | static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 9056 | { |
c30fec65 VS |
9057 | /* |
9058 | * N = val / 2 | |
9059 | * CU (slow) = CU2x (fast) / 2 = GPLL ref * N / 2 | |
9060 | */ | |
9061 | return DIV_ROUND_CLOSEST(dev_priv->rps.gpll_ref_freq * val, 2 * 2 * 1000); | |
22b1b2f8 D |
9062 | } |
9063 | ||
b55dd647 | 9064 | static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 9065 | { |
1c14762d | 9066 | /* CHV needs even values */ |
c30fec65 | 9067 | return DIV_ROUND_CLOSEST(2 * 1000 * val, dev_priv->rps.gpll_ref_freq) * 2; |
22b1b2f8 D |
9068 | } |
9069 | ||
616bc820 | 9070 | int intel_gpu_freq(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 9071 | { |
2d1fe073 | 9072 | if (IS_GEN9(dev_priv)) |
500a3d2e MK |
9073 | return DIV_ROUND_CLOSEST(val * GT_FREQUENCY_MULTIPLIER, |
9074 | GEN9_FREQ_SCALER); | |
2d1fe073 | 9075 | else if (IS_CHERRYVIEW(dev_priv)) |
616bc820 | 9076 | return chv_gpu_freq(dev_priv, val); |
2d1fe073 | 9077 | else if (IS_VALLEYVIEW(dev_priv)) |
616bc820 VS |
9078 | return byt_gpu_freq(dev_priv, val); |
9079 | else | |
9080 | return val * GT_FREQUENCY_MULTIPLIER; | |
22b1b2f8 D |
9081 | } |
9082 | ||
616bc820 VS |
9083 | int intel_freq_opcode(struct drm_i915_private *dev_priv, int val) |
9084 | { | |
2d1fe073 | 9085 | if (IS_GEN9(dev_priv)) |
500a3d2e MK |
9086 | return DIV_ROUND_CLOSEST(val * GEN9_FREQ_SCALER, |
9087 | GT_FREQUENCY_MULTIPLIER); | |
2d1fe073 | 9088 | else if (IS_CHERRYVIEW(dev_priv)) |
616bc820 | 9089 | return chv_freq_opcode(dev_priv, val); |
2d1fe073 | 9090 | else if (IS_VALLEYVIEW(dev_priv)) |
616bc820 VS |
9091 | return byt_freq_opcode(dev_priv, val); |
9092 | else | |
500a3d2e | 9093 | return DIV_ROUND_CLOSEST(val, GT_FREQUENCY_MULTIPLIER); |
616bc820 | 9094 | } |
22b1b2f8 | 9095 | |
6ad790c0 CW |
9096 | struct request_boost { |
9097 | struct work_struct work; | |
eed29a5b | 9098 | struct drm_i915_gem_request *req; |
6ad790c0 CW |
9099 | }; |
9100 | ||
9101 | static void __intel_rps_boost_work(struct work_struct *work) | |
9102 | { | |
9103 | struct request_boost *boost = container_of(work, struct request_boost, work); | |
e61b9958 | 9104 | struct drm_i915_gem_request *req = boost->req; |
6ad790c0 | 9105 | |
f69a02c9 | 9106 | if (!i915_gem_request_completed(req)) |
c033666a | 9107 | gen6_rps_boost(req->i915, NULL, req->emitted_jiffies); |
6ad790c0 | 9108 | |
e8a261ea | 9109 | i915_gem_request_put(req); |
6ad790c0 CW |
9110 | kfree(boost); |
9111 | } | |
9112 | ||
91d14251 | 9113 | void intel_queue_rps_boost_for_request(struct drm_i915_gem_request *req) |
6ad790c0 CW |
9114 | { |
9115 | struct request_boost *boost; | |
9116 | ||
91d14251 | 9117 | if (req == NULL || INTEL_GEN(req->i915) < 6) |
6ad790c0 CW |
9118 | return; |
9119 | ||
f69a02c9 | 9120 | if (i915_gem_request_completed(req)) |
e61b9958 CW |
9121 | return; |
9122 | ||
6ad790c0 CW |
9123 | boost = kmalloc(sizeof(*boost), GFP_ATOMIC); |
9124 | if (boost == NULL) | |
9125 | return; | |
9126 | ||
e8a261ea | 9127 | boost->req = i915_gem_request_get(req); |
6ad790c0 CW |
9128 | |
9129 | INIT_WORK(&boost->work, __intel_rps_boost_work); | |
91d14251 | 9130 | queue_work(req->i915->wq, &boost->work); |
6ad790c0 CW |
9131 | } |
9132 | ||
192aa181 | 9133 | void intel_pm_setup(struct drm_i915_private *dev_priv) |
907b28c5 | 9134 | { |
f742a552 | 9135 | mutex_init(&dev_priv->rps.hw_lock); |
8d3afd7d | 9136 | spin_lock_init(&dev_priv->rps.client_lock); |
f742a552 | 9137 | |
54b4f68f CW |
9138 | INIT_DELAYED_WORK(&dev_priv->rps.autoenable_work, |
9139 | __intel_autoenable_gt_powersave); | |
1854d5ca | 9140 | INIT_LIST_HEAD(&dev_priv->rps.clients); |
5d584b2e | 9141 | |
33688d95 | 9142 | dev_priv->pm.suspended = false; |
1f814dac | 9143 | atomic_set(&dev_priv->pm.wakeref_count, 0); |
907b28c5 | 9144 | } |
135bafa5 | 9145 | |
47c21d9a MK |
9146 | static u64 vlv_residency_raw(struct drm_i915_private *dev_priv, |
9147 | const i915_reg_t reg) | |
9148 | { | |
facbecad | 9149 | u32 lower, upper, tmp; |
71cc2b18 | 9150 | int loop = 2; |
47c21d9a MK |
9151 | |
9152 | /* The register accessed do not need forcewake. We borrow | |
9153 | * uncore lock to prevent concurrent access to range reg. | |
9154 | */ | |
9155 | spin_lock_irq(&dev_priv->uncore.lock); | |
47c21d9a MK |
9156 | |
9157 | /* vlv and chv residency counters are 40 bits in width. | |
9158 | * With a control bit, we can choose between upper or lower | |
9159 | * 32bit window into this counter. | |
facbecad CW |
9160 | * |
9161 | * Although we always use the counter in high-range mode elsewhere, | |
9162 | * userspace may attempt to read the value before rc6 is initialised, | |
9163 | * before we have set the default VLV_COUNTER_CONTROL value. So always | |
9164 | * set the high bit to be safe. | |
47c21d9a | 9165 | */ |
facbecad CW |
9166 | I915_WRITE_FW(VLV_COUNTER_CONTROL, |
9167 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH)); | |
47c21d9a MK |
9168 | upper = I915_READ_FW(reg); |
9169 | do { | |
9170 | tmp = upper; | |
9171 | ||
9172 | I915_WRITE_FW(VLV_COUNTER_CONTROL, | |
9173 | _MASKED_BIT_DISABLE(VLV_COUNT_RANGE_HIGH)); | |
9174 | lower = I915_READ_FW(reg); | |
9175 | ||
9176 | I915_WRITE_FW(VLV_COUNTER_CONTROL, | |
9177 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH)); | |
9178 | upper = I915_READ_FW(reg); | |
71cc2b18 | 9179 | } while (upper != tmp && --loop); |
47c21d9a | 9180 | |
facbecad CW |
9181 | /* Everywhere else we always use VLV_COUNTER_CONTROL with the |
9182 | * VLV_COUNT_RANGE_HIGH bit set - so it is safe to leave it set | |
9183 | * now. | |
9184 | */ | |
9185 | ||
47c21d9a MK |
9186 | spin_unlock_irq(&dev_priv->uncore.lock); |
9187 | ||
9188 | return lower | (u64)upper << 8; | |
9189 | } | |
9190 | ||
c5a0ad11 MK |
9191 | u64 intel_rc6_residency_us(struct drm_i915_private *dev_priv, |
9192 | const i915_reg_t reg) | |
135bafa5 | 9193 | { |
47c21d9a | 9194 | u64 time_hw, units, div; |
135bafa5 MK |
9195 | |
9196 | if (!intel_enable_rc6()) | |
9197 | return 0; | |
9198 | ||
9199 | intel_runtime_pm_get(dev_priv); | |
9200 | ||
9201 | /* On VLV and CHV, residency time is in CZ units rather than 1.28us */ | |
9202 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { | |
c5a0ad11 | 9203 | units = 1000; |
135bafa5 MK |
9204 | div = dev_priv->czclk_freq; |
9205 | ||
47c21d9a | 9206 | time_hw = vlv_residency_raw(dev_priv, reg); |
135bafa5 | 9207 | } else if (IS_GEN9_LP(dev_priv)) { |
c5a0ad11 | 9208 | units = 1000; |
135bafa5 | 9209 | div = 1200; /* 833.33ns */ |
135bafa5 | 9210 | |
47c21d9a MK |
9211 | time_hw = I915_READ(reg); |
9212 | } else { | |
9213 | units = 128000; /* 1.28us */ | |
9214 | div = 100000; | |
9215 | ||
9216 | time_hw = I915_READ(reg); | |
9217 | } | |
135bafa5 MK |
9218 | |
9219 | intel_runtime_pm_put(dev_priv); | |
47c21d9a | 9220 | return DIV_ROUND_UP_ULL(time_hw * units, div); |
135bafa5 | 9221 | } |