]>
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 | { |
b033bb6d | 61 | /* See Bspec note for PSR2_CTL bit 31, Wa#828:skl,bxt,kbl */ |
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 MK |
67 | |
68 | /* WaEnableChickenDCPR:skl,bxt,kbl */ | |
69 | I915_WRITE(GEN8_CHICKEN_DCPR_1, | |
70 | I915_READ(GEN8_CHICKEN_DCPR_1) | MASK_WAKEMEM); | |
0f78dee6 MK |
71 | |
72 | /* WaFbcTurnOffFbcWatermark:skl,bxt,kbl */ | |
303d4ea5 MK |
73 | /* WaFbcWakeMemOn:skl,bxt,kbl */ |
74 | I915_WRITE(DISP_ARB_CTL, I915_READ(DISP_ARB_CTL) | | |
75 | DISP_FBC_WM_DIS | | |
76 | DISP_FBC_MEMORY_WAKE); | |
d1b4eefd MK |
77 | |
78 | /* WaFbcHighMemBwCorruptionAvoidance:skl,bxt,kbl */ | |
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 | */ | |
102 | if (IS_BXT_REVID(dev_priv, BXT_REVID_B0, REVID_FOREVER)) | |
103 | I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) | | |
104 | PWM1_GATING_DIS | PWM2_GATING_DIS); | |
a82abe43 ID |
105 | } |
106 | ||
148ac1f3 | 107 | static void i915_pineview_get_mem_freq(struct drm_i915_private *dev_priv) |
c921aba8 | 108 | { |
c921aba8 DV |
109 | u32 tmp; |
110 | ||
111 | tmp = I915_READ(CLKCFG); | |
112 | ||
113 | switch (tmp & CLKCFG_FSB_MASK) { | |
114 | case CLKCFG_FSB_533: | |
115 | dev_priv->fsb_freq = 533; /* 133*4 */ | |
116 | break; | |
117 | case CLKCFG_FSB_800: | |
118 | dev_priv->fsb_freq = 800; /* 200*4 */ | |
119 | break; | |
120 | case CLKCFG_FSB_667: | |
121 | dev_priv->fsb_freq = 667; /* 167*4 */ | |
122 | break; | |
123 | case CLKCFG_FSB_400: | |
124 | dev_priv->fsb_freq = 400; /* 100*4 */ | |
125 | break; | |
126 | } | |
127 | ||
128 | switch (tmp & CLKCFG_MEM_MASK) { | |
129 | case CLKCFG_MEM_533: | |
130 | dev_priv->mem_freq = 533; | |
131 | break; | |
132 | case CLKCFG_MEM_667: | |
133 | dev_priv->mem_freq = 667; | |
134 | break; | |
135 | case CLKCFG_MEM_800: | |
136 | dev_priv->mem_freq = 800; | |
137 | break; | |
138 | } | |
139 | ||
140 | /* detect pineview DDR3 setting */ | |
141 | tmp = I915_READ(CSHRDDR3CTL); | |
142 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; | |
143 | } | |
144 | ||
148ac1f3 | 145 | static void i915_ironlake_get_mem_freq(struct drm_i915_private *dev_priv) |
c921aba8 | 146 | { |
c921aba8 DV |
147 | u16 ddrpll, csipll; |
148 | ||
149 | ddrpll = I915_READ16(DDRMPLL1); | |
150 | csipll = I915_READ16(CSIPLL0); | |
151 | ||
152 | switch (ddrpll & 0xff) { | |
153 | case 0xc: | |
154 | dev_priv->mem_freq = 800; | |
155 | break; | |
156 | case 0x10: | |
157 | dev_priv->mem_freq = 1066; | |
158 | break; | |
159 | case 0x14: | |
160 | dev_priv->mem_freq = 1333; | |
161 | break; | |
162 | case 0x18: | |
163 | dev_priv->mem_freq = 1600; | |
164 | break; | |
165 | default: | |
166 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", | |
167 | ddrpll & 0xff); | |
168 | dev_priv->mem_freq = 0; | |
169 | break; | |
170 | } | |
171 | ||
20e4d407 | 172 | dev_priv->ips.r_t = dev_priv->mem_freq; |
c921aba8 DV |
173 | |
174 | switch (csipll & 0x3ff) { | |
175 | case 0x00c: | |
176 | dev_priv->fsb_freq = 3200; | |
177 | break; | |
178 | case 0x00e: | |
179 | dev_priv->fsb_freq = 3733; | |
180 | break; | |
181 | case 0x010: | |
182 | dev_priv->fsb_freq = 4266; | |
183 | break; | |
184 | case 0x012: | |
185 | dev_priv->fsb_freq = 4800; | |
186 | break; | |
187 | case 0x014: | |
188 | dev_priv->fsb_freq = 5333; | |
189 | break; | |
190 | case 0x016: | |
191 | dev_priv->fsb_freq = 5866; | |
192 | break; | |
193 | case 0x018: | |
194 | dev_priv->fsb_freq = 6400; | |
195 | break; | |
196 | default: | |
197 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", | |
198 | csipll & 0x3ff); | |
199 | dev_priv->fsb_freq = 0; | |
200 | break; | |
201 | } | |
202 | ||
203 | if (dev_priv->fsb_freq == 3200) { | |
20e4d407 | 204 | dev_priv->ips.c_m = 0; |
c921aba8 | 205 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
20e4d407 | 206 | dev_priv->ips.c_m = 1; |
c921aba8 | 207 | } else { |
20e4d407 | 208 | dev_priv->ips.c_m = 2; |
c921aba8 DV |
209 | } |
210 | } | |
211 | ||
b445e3b0 ED |
212 | static const struct cxsr_latency cxsr_latency_table[] = { |
213 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ | |
214 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ | |
215 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ | |
216 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ | |
217 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ | |
218 | ||
219 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ | |
220 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ | |
221 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ | |
222 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ | |
223 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ | |
224 | ||
225 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ | |
226 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ | |
227 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ | |
228 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ | |
229 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ | |
230 | ||
231 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ | |
232 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ | |
233 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ | |
234 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ | |
235 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ | |
236 | ||
237 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ | |
238 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ | |
239 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ | |
240 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ | |
241 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ | |
242 | ||
243 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ | |
244 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ | |
245 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ | |
246 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ | |
247 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ | |
248 | }; | |
249 | ||
44a655ca TU |
250 | static const struct cxsr_latency *intel_get_cxsr_latency(bool is_desktop, |
251 | bool is_ddr3, | |
b445e3b0 ED |
252 | int fsb, |
253 | int mem) | |
254 | { | |
255 | const struct cxsr_latency *latency; | |
256 | int i; | |
257 | ||
258 | if (fsb == 0 || mem == 0) | |
259 | return NULL; | |
260 | ||
261 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { | |
262 | latency = &cxsr_latency_table[i]; | |
263 | if (is_desktop == latency->is_desktop && | |
264 | is_ddr3 == latency->is_ddr3 && | |
265 | fsb == latency->fsb_freq && mem == latency->mem_freq) | |
266 | return latency; | |
267 | } | |
268 | ||
269 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
270 | ||
271 | return NULL; | |
272 | } | |
273 | ||
fc1ac8de VS |
274 | static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable) |
275 | { | |
276 | u32 val; | |
277 | ||
278 | mutex_lock(&dev_priv->rps.hw_lock); | |
279 | ||
280 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); | |
281 | if (enable) | |
282 | val &= ~FORCE_DDR_HIGH_FREQ; | |
283 | else | |
284 | val |= FORCE_DDR_HIGH_FREQ; | |
285 | val &= ~FORCE_DDR_LOW_FREQ; | |
286 | val |= FORCE_DDR_FREQ_REQ_ACK; | |
287 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); | |
288 | ||
289 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & | |
290 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) | |
291 | DRM_ERROR("timed out waiting for Punit DDR DVFS request\n"); | |
292 | ||
293 | mutex_unlock(&dev_priv->rps.hw_lock); | |
294 | } | |
295 | ||
cfb41411 VS |
296 | static void chv_set_memory_pm5(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_DSPFREQ); | |
303 | if (enable) | |
304 | val |= DSP_MAXFIFO_PM5_ENABLE; | |
305 | else | |
306 | val &= ~DSP_MAXFIFO_PM5_ENABLE; | |
307 | vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val); | |
308 | ||
309 | mutex_unlock(&dev_priv->rps.hw_lock); | |
310 | } | |
311 | ||
f4998963 VS |
312 | #define FW_WM(value, plane) \ |
313 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK) | |
314 | ||
5209b1f4 | 315 | void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable) |
b445e3b0 | 316 | { |
5209b1f4 | 317 | u32 val; |
b445e3b0 | 318 | |
920a14b2 | 319 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { |
5209b1f4 | 320 | I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0); |
a7a6c498 | 321 | POSTING_READ(FW_BLC_SELF_VLV); |
852eb00d | 322 | dev_priv->wm.vlv.cxsr = enable; |
9beb5fea | 323 | } else if (IS_G4X(dev_priv) || IS_CRESTLINE(dev_priv)) { |
5209b1f4 | 324 | I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0); |
a7a6c498 | 325 | POSTING_READ(FW_BLC_SELF); |
9b1e14f4 | 326 | } else if (IS_PINEVIEW(dev_priv)) { |
5209b1f4 ID |
327 | val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN; |
328 | val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0; | |
329 | I915_WRITE(DSPFW3, val); | |
a7a6c498 | 330 | POSTING_READ(DSPFW3); |
50a0bc90 | 331 | } else if (IS_I945G(dev_priv) || IS_I945GM(dev_priv)) { |
5209b1f4 ID |
332 | val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) : |
333 | _MASKED_BIT_DISABLE(FW_BLC_SELF_EN); | |
334 | I915_WRITE(FW_BLC_SELF, val); | |
a7a6c498 | 335 | POSTING_READ(FW_BLC_SELF); |
50a0bc90 | 336 | } else if (IS_I915GM(dev_priv)) { |
acb91359 VS |
337 | /* |
338 | * FIXME can't find a bit like this for 915G, and | |
339 | * and yet it does have the related watermark in | |
340 | * FW_BLC_SELF. What's going on? | |
341 | */ | |
5209b1f4 ID |
342 | val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) : |
343 | _MASKED_BIT_DISABLE(INSTPM_SELF_EN); | |
344 | I915_WRITE(INSTPM, val); | |
a7a6c498 | 345 | POSTING_READ(INSTPM); |
5209b1f4 ID |
346 | } else { |
347 | return; | |
348 | } | |
b445e3b0 | 349 | |
08c4d7fc | 350 | DRM_DEBUG_KMS("memory self-refresh is %s\n", enableddisabled(enable)); |
b445e3b0 ED |
351 | } |
352 | ||
fc1ac8de | 353 | |
b445e3b0 ED |
354 | /* |
355 | * Latency for FIFO fetches is dependent on several factors: | |
356 | * - memory configuration (speed, channels) | |
357 | * - chipset | |
358 | * - current MCH state | |
359 | * It can be fairly high in some situations, so here we assume a fairly | |
360 | * pessimal value. It's a tradeoff between extra memory fetches (if we | |
361 | * set this value too high, the FIFO will fetch frequently to stay full) | |
362 | * and power consumption (set it too low to save power and we might see | |
363 | * FIFO underruns and display "flicker"). | |
364 | * | |
365 | * A value of 5us seems to be a good balance; safe for very low end | |
366 | * platforms but not overly aggressive on lower latency configs. | |
367 | */ | |
5aef6003 | 368 | static const int pessimal_latency_ns = 5000; |
b445e3b0 | 369 | |
b5004720 VS |
370 | #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \ |
371 | ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8)) | |
372 | ||
ef0f5e93 | 373 | static int vlv_get_fifo_size(struct drm_i915_private *dev_priv, |
b5004720 VS |
374 | enum pipe pipe, int plane) |
375 | { | |
b5004720 VS |
376 | int sprite0_start, sprite1_start, size; |
377 | ||
378 | switch (pipe) { | |
379 | uint32_t dsparb, dsparb2, dsparb3; | |
380 | case PIPE_A: | |
381 | dsparb = I915_READ(DSPARB); | |
382 | dsparb2 = I915_READ(DSPARB2); | |
383 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0); | |
384 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4); | |
385 | break; | |
386 | case PIPE_B: | |
387 | dsparb = I915_READ(DSPARB); | |
388 | dsparb2 = I915_READ(DSPARB2); | |
389 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8); | |
390 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12); | |
391 | break; | |
392 | case PIPE_C: | |
393 | dsparb2 = I915_READ(DSPARB2); | |
394 | dsparb3 = I915_READ(DSPARB3); | |
395 | sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16); | |
396 | sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20); | |
397 | break; | |
398 | default: | |
399 | return 0; | |
400 | } | |
401 | ||
402 | switch (plane) { | |
403 | case 0: | |
404 | size = sprite0_start; | |
405 | break; | |
406 | case 1: | |
407 | size = sprite1_start - sprite0_start; | |
408 | break; | |
409 | case 2: | |
410 | size = 512 - 1 - sprite1_start; | |
411 | break; | |
412 | default: | |
413 | return 0; | |
414 | } | |
415 | ||
416 | DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n", | |
417 | pipe_name(pipe), plane == 0 ? "primary" : "sprite", | |
418 | plane == 0 ? plane_name(pipe) : sprite_name(pipe, plane - 1), | |
419 | size); | |
420 | ||
421 | return size; | |
422 | } | |
423 | ||
ef0f5e93 | 424 | static int i9xx_get_fifo_size(struct drm_i915_private *dev_priv, int plane) |
b445e3b0 | 425 | { |
b445e3b0 ED |
426 | uint32_t dsparb = I915_READ(DSPARB); |
427 | int size; | |
428 | ||
429 | size = dsparb & 0x7f; | |
430 | if (plane) | |
431 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; | |
432 | ||
433 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
434 | plane ? "B" : "A", size); | |
435 | ||
436 | return size; | |
437 | } | |
438 | ||
ef0f5e93 | 439 | static int i830_get_fifo_size(struct drm_i915_private *dev_priv, int plane) |
b445e3b0 | 440 | { |
b445e3b0 ED |
441 | uint32_t dsparb = I915_READ(DSPARB); |
442 | int size; | |
443 | ||
444 | size = dsparb & 0x1ff; | |
445 | if (plane) | |
446 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; | |
447 | size >>= 1; /* Convert to cachelines */ | |
448 | ||
449 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
450 | plane ? "B" : "A", size); | |
451 | ||
452 | return size; | |
453 | } | |
454 | ||
ef0f5e93 | 455 | static int i845_get_fifo_size(struct drm_i915_private *dev_priv, int plane) |
b445e3b0 | 456 | { |
b445e3b0 ED |
457 | uint32_t dsparb = I915_READ(DSPARB); |
458 | int size; | |
459 | ||
460 | size = dsparb & 0x7f; | |
461 | size >>= 2; /* Convert to cachelines */ | |
462 | ||
463 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
464 | plane ? "B" : "A", | |
465 | size); | |
466 | ||
467 | return size; | |
468 | } | |
469 | ||
b445e3b0 ED |
470 | /* Pineview has different values for various configs */ |
471 | static const struct intel_watermark_params pineview_display_wm = { | |
e0f0273e VS |
472 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
473 | .max_wm = PINEVIEW_MAX_WM, | |
474 | .default_wm = PINEVIEW_DFT_WM, | |
475 | .guard_size = PINEVIEW_GUARD_WM, | |
476 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
477 | }; |
478 | static const struct intel_watermark_params pineview_display_hplloff_wm = { | |
e0f0273e VS |
479 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
480 | .max_wm = PINEVIEW_MAX_WM, | |
481 | .default_wm = PINEVIEW_DFT_HPLLOFF_WM, | |
482 | .guard_size = PINEVIEW_GUARD_WM, | |
483 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
484 | }; |
485 | static const struct intel_watermark_params pineview_cursor_wm = { | |
e0f0273e VS |
486 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
487 | .max_wm = PINEVIEW_CURSOR_MAX_WM, | |
488 | .default_wm = PINEVIEW_CURSOR_DFT_WM, | |
489 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, | |
490 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
491 | }; |
492 | static const struct intel_watermark_params pineview_cursor_hplloff_wm = { | |
e0f0273e VS |
493 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
494 | .max_wm = PINEVIEW_CURSOR_MAX_WM, | |
495 | .default_wm = PINEVIEW_CURSOR_DFT_WM, | |
496 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, | |
497 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
498 | }; |
499 | static const struct intel_watermark_params g4x_wm_info = { | |
e0f0273e VS |
500 | .fifo_size = G4X_FIFO_SIZE, |
501 | .max_wm = G4X_MAX_WM, | |
502 | .default_wm = G4X_MAX_WM, | |
503 | .guard_size = 2, | |
504 | .cacheline_size = G4X_FIFO_LINE_SIZE, | |
b445e3b0 ED |
505 | }; |
506 | static const struct intel_watermark_params g4x_cursor_wm_info = { | |
e0f0273e VS |
507 | .fifo_size = I965_CURSOR_FIFO, |
508 | .max_wm = I965_CURSOR_MAX_WM, | |
509 | .default_wm = I965_CURSOR_DFT_WM, | |
510 | .guard_size = 2, | |
511 | .cacheline_size = G4X_FIFO_LINE_SIZE, | |
b445e3b0 | 512 | }; |
b445e3b0 | 513 | static const struct intel_watermark_params i965_cursor_wm_info = { |
e0f0273e VS |
514 | .fifo_size = I965_CURSOR_FIFO, |
515 | .max_wm = I965_CURSOR_MAX_WM, | |
516 | .default_wm = I965_CURSOR_DFT_WM, | |
517 | .guard_size = 2, | |
518 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 ED |
519 | }; |
520 | static const struct intel_watermark_params i945_wm_info = { | |
e0f0273e VS |
521 | .fifo_size = I945_FIFO_SIZE, |
522 | .max_wm = I915_MAX_WM, | |
523 | .default_wm = 1, | |
524 | .guard_size = 2, | |
525 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 ED |
526 | }; |
527 | static const struct intel_watermark_params i915_wm_info = { | |
e0f0273e VS |
528 | .fifo_size = I915_FIFO_SIZE, |
529 | .max_wm = I915_MAX_WM, | |
530 | .default_wm = 1, | |
531 | .guard_size = 2, | |
532 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 | 533 | }; |
9d539105 | 534 | static const struct intel_watermark_params i830_a_wm_info = { |
e0f0273e VS |
535 | .fifo_size = I855GM_FIFO_SIZE, |
536 | .max_wm = I915_MAX_WM, | |
537 | .default_wm = 1, | |
538 | .guard_size = 2, | |
539 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
b445e3b0 | 540 | }; |
9d539105 VS |
541 | static const struct intel_watermark_params i830_bc_wm_info = { |
542 | .fifo_size = I855GM_FIFO_SIZE, | |
543 | .max_wm = I915_MAX_WM/2, | |
544 | .default_wm = 1, | |
545 | .guard_size = 2, | |
546 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
547 | }; | |
feb56b93 | 548 | static const struct intel_watermark_params i845_wm_info = { |
e0f0273e VS |
549 | .fifo_size = I830_FIFO_SIZE, |
550 | .max_wm = I915_MAX_WM, | |
551 | .default_wm = 1, | |
552 | .guard_size = 2, | |
553 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
b445e3b0 ED |
554 | }; |
555 | ||
b445e3b0 ED |
556 | /** |
557 | * intel_calculate_wm - calculate watermark level | |
558 | * @clock_in_khz: pixel clock | |
559 | * @wm: chip FIFO params | |
ac484963 | 560 | * @cpp: bytes per pixel |
b445e3b0 ED |
561 | * @latency_ns: memory latency for the platform |
562 | * | |
563 | * Calculate the watermark level (the level at which the display plane will | |
564 | * start fetching from memory again). Each chip has a different display | |
565 | * FIFO size and allocation, so the caller needs to figure that out and pass | |
566 | * in the correct intel_watermark_params structure. | |
567 | * | |
568 | * As the pixel clock runs, the FIFO will be drained at a rate that depends | |
569 | * on the pixel size. When it reaches the watermark level, it'll start | |
570 | * fetching FIFO line sized based chunks from memory until the FIFO fills | |
571 | * past the watermark point. If the FIFO drains completely, a FIFO underrun | |
572 | * will occur, and a display engine hang could result. | |
573 | */ | |
574 | static unsigned long intel_calculate_wm(unsigned long clock_in_khz, | |
575 | const struct intel_watermark_params *wm, | |
ac484963 | 576 | int fifo_size, int cpp, |
b445e3b0 ED |
577 | unsigned long latency_ns) |
578 | { | |
579 | long entries_required, wm_size; | |
580 | ||
581 | /* | |
582 | * Note: we need to make sure we don't overflow for various clock & | |
583 | * latency values. | |
584 | * clocks go from a few thousand to several hundred thousand. | |
585 | * latency is usually a few thousand | |
586 | */ | |
ac484963 | 587 | entries_required = ((clock_in_khz / 1000) * cpp * latency_ns) / |
b445e3b0 ED |
588 | 1000; |
589 | entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size); | |
590 | ||
591 | DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required); | |
592 | ||
593 | wm_size = fifo_size - (entries_required + wm->guard_size); | |
594 | ||
595 | DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size); | |
596 | ||
597 | /* Don't promote wm_size to unsigned... */ | |
598 | if (wm_size > (long)wm->max_wm) | |
599 | wm_size = wm->max_wm; | |
600 | if (wm_size <= 0) | |
601 | wm_size = wm->default_wm; | |
d6feb196 VS |
602 | |
603 | /* | |
604 | * Bspec seems to indicate that the value shouldn't be lower than | |
605 | * 'burst size + 1'. Certainly 830 is quite unhappy with low values. | |
606 | * Lets go for 8 which is the burst size since certain platforms | |
607 | * already use a hardcoded 8 (which is what the spec says should be | |
608 | * done). | |
609 | */ | |
610 | if (wm_size <= 8) | |
611 | wm_size = 8; | |
612 | ||
b445e3b0 ED |
613 | return wm_size; |
614 | } | |
615 | ||
ffc7a76b | 616 | static struct intel_crtc *single_enabled_crtc(struct drm_i915_private *dev_priv) |
b445e3b0 | 617 | { |
efc2611e | 618 | struct intel_crtc *crtc, *enabled = NULL; |
b445e3b0 | 619 | |
ffc7a76b | 620 | for_each_intel_crtc(&dev_priv->drm, crtc) { |
efc2611e | 621 | if (intel_crtc_active(crtc)) { |
b445e3b0 ED |
622 | if (enabled) |
623 | return NULL; | |
624 | enabled = crtc; | |
625 | } | |
626 | } | |
627 | ||
628 | return enabled; | |
629 | } | |
630 | ||
432081bc | 631 | static void pineview_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 632 | { |
ffc7a76b | 633 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
efc2611e | 634 | struct intel_crtc *crtc; |
b445e3b0 ED |
635 | const struct cxsr_latency *latency; |
636 | u32 reg; | |
637 | unsigned long wm; | |
638 | ||
50a0bc90 TU |
639 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev_priv), |
640 | dev_priv->is_ddr3, | |
641 | dev_priv->fsb_freq, | |
642 | dev_priv->mem_freq); | |
b445e3b0 ED |
643 | if (!latency) { |
644 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
5209b1f4 | 645 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
646 | return; |
647 | } | |
648 | ||
ffc7a76b | 649 | crtc = single_enabled_crtc(dev_priv); |
b445e3b0 | 650 | if (crtc) { |
efc2611e VS |
651 | const struct drm_display_mode *adjusted_mode = |
652 | &crtc->config->base.adjusted_mode; | |
653 | const struct drm_framebuffer *fb = | |
654 | crtc->base.primary->state->fb; | |
655 | int cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
7c5f93b0 | 656 | int clock = adjusted_mode->crtc_clock; |
b445e3b0 ED |
657 | |
658 | /* Display SR */ | |
659 | wm = intel_calculate_wm(clock, &pineview_display_wm, | |
660 | pineview_display_wm.fifo_size, | |
ac484963 | 661 | cpp, latency->display_sr); |
b445e3b0 ED |
662 | reg = I915_READ(DSPFW1); |
663 | reg &= ~DSPFW_SR_MASK; | |
f4998963 | 664 | reg |= FW_WM(wm, SR); |
b445e3b0 ED |
665 | I915_WRITE(DSPFW1, reg); |
666 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); | |
667 | ||
668 | /* cursor SR */ | |
669 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, | |
670 | pineview_display_wm.fifo_size, | |
ac484963 | 671 | cpp, latency->cursor_sr); |
b445e3b0 ED |
672 | reg = I915_READ(DSPFW3); |
673 | reg &= ~DSPFW_CURSOR_SR_MASK; | |
f4998963 | 674 | reg |= FW_WM(wm, CURSOR_SR); |
b445e3b0 ED |
675 | I915_WRITE(DSPFW3, reg); |
676 | ||
677 | /* Display HPLL off SR */ | |
678 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, | |
679 | pineview_display_hplloff_wm.fifo_size, | |
ac484963 | 680 | cpp, latency->display_hpll_disable); |
b445e3b0 ED |
681 | reg = I915_READ(DSPFW3); |
682 | reg &= ~DSPFW_HPLL_SR_MASK; | |
f4998963 | 683 | reg |= FW_WM(wm, HPLL_SR); |
b445e3b0 ED |
684 | I915_WRITE(DSPFW3, reg); |
685 | ||
686 | /* cursor HPLL off SR */ | |
687 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, | |
688 | pineview_display_hplloff_wm.fifo_size, | |
ac484963 | 689 | cpp, latency->cursor_hpll_disable); |
b445e3b0 ED |
690 | reg = I915_READ(DSPFW3); |
691 | reg &= ~DSPFW_HPLL_CURSOR_MASK; | |
f4998963 | 692 | reg |= FW_WM(wm, HPLL_CURSOR); |
b445e3b0 ED |
693 | I915_WRITE(DSPFW3, reg); |
694 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); | |
695 | ||
5209b1f4 | 696 | intel_set_memory_cxsr(dev_priv, true); |
b445e3b0 | 697 | } else { |
5209b1f4 | 698 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
699 | } |
700 | } | |
701 | ||
f0ce2310 | 702 | static bool g4x_compute_wm0(struct drm_i915_private *dev_priv, |
b445e3b0 ED |
703 | int plane, |
704 | const struct intel_watermark_params *display, | |
705 | int display_latency_ns, | |
706 | const struct intel_watermark_params *cursor, | |
707 | int cursor_latency_ns, | |
708 | int *plane_wm, | |
709 | int *cursor_wm) | |
710 | { | |
efc2611e | 711 | struct intel_crtc *crtc; |
4fe8590a | 712 | const struct drm_display_mode *adjusted_mode; |
efc2611e | 713 | const struct drm_framebuffer *fb; |
ac484963 | 714 | int htotal, hdisplay, clock, cpp; |
b445e3b0 ED |
715 | int line_time_us, line_count; |
716 | int entries, tlb_miss; | |
717 | ||
b91eb5cc | 718 | crtc = intel_get_crtc_for_plane(dev_priv, plane); |
efc2611e | 719 | if (!intel_crtc_active(crtc)) { |
b445e3b0 ED |
720 | *cursor_wm = cursor->guard_size; |
721 | *plane_wm = display->guard_size; | |
722 | return false; | |
723 | } | |
724 | ||
efc2611e VS |
725 | adjusted_mode = &crtc->config->base.adjusted_mode; |
726 | fb = crtc->base.primary->state->fb; | |
241bfc38 | 727 | clock = adjusted_mode->crtc_clock; |
fec8cba3 | 728 | htotal = adjusted_mode->crtc_htotal; |
efc2611e VS |
729 | hdisplay = crtc->config->pipe_src_w; |
730 | cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
b445e3b0 ED |
731 | |
732 | /* Use the small buffer method to calculate plane watermark */ | |
ac484963 | 733 | entries = ((clock * cpp / 1000) * display_latency_ns) / 1000; |
b445e3b0 ED |
734 | tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8; |
735 | if (tlb_miss > 0) | |
736 | entries += tlb_miss; | |
737 | entries = DIV_ROUND_UP(entries, display->cacheline_size); | |
738 | *plane_wm = entries + display->guard_size; | |
739 | if (*plane_wm > (int)display->max_wm) | |
740 | *plane_wm = display->max_wm; | |
741 | ||
742 | /* Use the large buffer method to calculate cursor watermark */ | |
922044c9 | 743 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 | 744 | line_count = (cursor_latency_ns / line_time_us + 1000) / 1000; |
efc2611e | 745 | entries = line_count * crtc->base.cursor->state->crtc_w * cpp; |
b445e3b0 ED |
746 | tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8; |
747 | if (tlb_miss > 0) | |
748 | entries += tlb_miss; | |
749 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); | |
750 | *cursor_wm = entries + cursor->guard_size; | |
751 | if (*cursor_wm > (int)cursor->max_wm) | |
752 | *cursor_wm = (int)cursor->max_wm; | |
753 | ||
754 | return true; | |
755 | } | |
756 | ||
757 | /* | |
758 | * Check the wm result. | |
759 | * | |
760 | * If any calculated watermark values is larger than the maximum value that | |
761 | * can be programmed into the associated watermark register, that watermark | |
762 | * must be disabled. | |
763 | */ | |
f0ce2310 | 764 | static bool g4x_check_srwm(struct drm_i915_private *dev_priv, |
b445e3b0 ED |
765 | int display_wm, int cursor_wm, |
766 | const struct intel_watermark_params *display, | |
767 | const struct intel_watermark_params *cursor) | |
768 | { | |
769 | DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n", | |
770 | display_wm, cursor_wm); | |
771 | ||
772 | if (display_wm > display->max_wm) { | |
ae9400ca | 773 | DRM_DEBUG_KMS("display watermark is too large(%d/%u), disabling\n", |
b445e3b0 ED |
774 | display_wm, display->max_wm); |
775 | return false; | |
776 | } | |
777 | ||
778 | if (cursor_wm > cursor->max_wm) { | |
ae9400ca | 779 | DRM_DEBUG_KMS("cursor watermark is too large(%d/%u), disabling\n", |
b445e3b0 ED |
780 | cursor_wm, cursor->max_wm); |
781 | return false; | |
782 | } | |
783 | ||
784 | if (!(display_wm || cursor_wm)) { | |
785 | DRM_DEBUG_KMS("SR latency is 0, disabling\n"); | |
786 | return false; | |
787 | } | |
788 | ||
789 | return true; | |
790 | } | |
791 | ||
f0ce2310 | 792 | static bool g4x_compute_srwm(struct drm_i915_private *dev_priv, |
b445e3b0 ED |
793 | int plane, |
794 | int latency_ns, | |
795 | const struct intel_watermark_params *display, | |
796 | const struct intel_watermark_params *cursor, | |
797 | int *display_wm, int *cursor_wm) | |
798 | { | |
efc2611e | 799 | struct intel_crtc *crtc; |
4fe8590a | 800 | const struct drm_display_mode *adjusted_mode; |
efc2611e | 801 | const struct drm_framebuffer *fb; |
ac484963 | 802 | int hdisplay, htotal, cpp, clock; |
b445e3b0 ED |
803 | unsigned long line_time_us; |
804 | int line_count, line_size; | |
805 | int small, large; | |
806 | int entries; | |
807 | ||
808 | if (!latency_ns) { | |
809 | *display_wm = *cursor_wm = 0; | |
810 | return false; | |
811 | } | |
812 | ||
b91eb5cc | 813 | crtc = intel_get_crtc_for_plane(dev_priv, plane); |
efc2611e VS |
814 | adjusted_mode = &crtc->config->base.adjusted_mode; |
815 | fb = crtc->base.primary->state->fb; | |
241bfc38 | 816 | clock = adjusted_mode->crtc_clock; |
fec8cba3 | 817 | htotal = adjusted_mode->crtc_htotal; |
efc2611e VS |
818 | hdisplay = crtc->config->pipe_src_w; |
819 | cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
b445e3b0 | 820 | |
922044c9 | 821 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 | 822 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
ac484963 | 823 | line_size = hdisplay * cpp; |
b445e3b0 ED |
824 | |
825 | /* Use the minimum of the small and large buffer method for primary */ | |
ac484963 | 826 | small = ((clock * cpp / 1000) * latency_ns) / 1000; |
b445e3b0 ED |
827 | large = line_count * line_size; |
828 | ||
829 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); | |
830 | *display_wm = entries + display->guard_size; | |
831 | ||
832 | /* calculate the self-refresh watermark for display cursor */ | |
efc2611e | 833 | entries = line_count * cpp * crtc->base.cursor->state->crtc_w; |
b445e3b0 ED |
834 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
835 | *cursor_wm = entries + cursor->guard_size; | |
836 | ||
f0ce2310 | 837 | return g4x_check_srwm(dev_priv, |
b445e3b0 ED |
838 | *display_wm, *cursor_wm, |
839 | display, cursor); | |
840 | } | |
841 | ||
15665979 VS |
842 | #define FW_WM_VLV(value, plane) \ |
843 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV) | |
844 | ||
0018fda1 VS |
845 | static void vlv_write_wm_values(struct intel_crtc *crtc, |
846 | const struct vlv_wm_values *wm) | |
847 | { | |
848 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); | |
849 | enum pipe pipe = crtc->pipe; | |
850 | ||
851 | I915_WRITE(VLV_DDL(pipe), | |
852 | (wm->ddl[pipe].cursor << DDL_CURSOR_SHIFT) | | |
853 | (wm->ddl[pipe].sprite[1] << DDL_SPRITE_SHIFT(1)) | | |
854 | (wm->ddl[pipe].sprite[0] << DDL_SPRITE_SHIFT(0)) | | |
855 | (wm->ddl[pipe].primary << DDL_PLANE_SHIFT)); | |
856 | ||
ae80152d | 857 | I915_WRITE(DSPFW1, |
15665979 VS |
858 | FW_WM(wm->sr.plane, SR) | |
859 | FW_WM(wm->pipe[PIPE_B].cursor, CURSORB) | | |
860 | FW_WM_VLV(wm->pipe[PIPE_B].primary, PLANEB) | | |
861 | FW_WM_VLV(wm->pipe[PIPE_A].primary, PLANEA)); | |
ae80152d | 862 | I915_WRITE(DSPFW2, |
15665979 VS |
863 | FW_WM_VLV(wm->pipe[PIPE_A].sprite[1], SPRITEB) | |
864 | FW_WM(wm->pipe[PIPE_A].cursor, CURSORA) | | |
865 | FW_WM_VLV(wm->pipe[PIPE_A].sprite[0], SPRITEA)); | |
ae80152d | 866 | I915_WRITE(DSPFW3, |
15665979 | 867 | FW_WM(wm->sr.cursor, CURSOR_SR)); |
ae80152d VS |
868 | |
869 | if (IS_CHERRYVIEW(dev_priv)) { | |
870 | I915_WRITE(DSPFW7_CHV, | |
15665979 VS |
871 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) | |
872 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC)); | |
ae80152d | 873 | I915_WRITE(DSPFW8_CHV, |
15665979 VS |
874 | FW_WM_VLV(wm->pipe[PIPE_C].sprite[1], SPRITEF) | |
875 | FW_WM_VLV(wm->pipe[PIPE_C].sprite[0], SPRITEE)); | |
ae80152d | 876 | I915_WRITE(DSPFW9_CHV, |
15665979 VS |
877 | FW_WM_VLV(wm->pipe[PIPE_C].primary, PLANEC) | |
878 | FW_WM(wm->pipe[PIPE_C].cursor, CURSORC)); | |
ae80152d | 879 | I915_WRITE(DSPHOWM, |
15665979 VS |
880 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
881 | FW_WM(wm->pipe[PIPE_C].sprite[1] >> 8, SPRITEF_HI) | | |
882 | FW_WM(wm->pipe[PIPE_C].sprite[0] >> 8, SPRITEE_HI) | | |
883 | FW_WM(wm->pipe[PIPE_C].primary >> 8, PLANEC_HI) | | |
884 | FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) | | |
885 | FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) | | |
886 | FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) | | |
887 | FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) | | |
888 | FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) | | |
889 | FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI)); | |
ae80152d VS |
890 | } else { |
891 | I915_WRITE(DSPFW7, | |
15665979 VS |
892 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) | |
893 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC)); | |
ae80152d | 894 | I915_WRITE(DSPHOWM, |
15665979 VS |
895 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
896 | FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) | | |
897 | FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) | | |
898 | FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) | | |
899 | FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) | | |
900 | FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) | | |
901 | FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI)); | |
ae80152d VS |
902 | } |
903 | ||
2cb389b7 VS |
904 | /* zero (unused) WM1 watermarks */ |
905 | I915_WRITE(DSPFW4, 0); | |
906 | I915_WRITE(DSPFW5, 0); | |
907 | I915_WRITE(DSPFW6, 0); | |
908 | I915_WRITE(DSPHOWM1, 0); | |
909 | ||
ae80152d | 910 | POSTING_READ(DSPFW1); |
0018fda1 VS |
911 | } |
912 | ||
15665979 VS |
913 | #undef FW_WM_VLV |
914 | ||
6eb1a681 VS |
915 | enum vlv_wm_level { |
916 | VLV_WM_LEVEL_PM2, | |
917 | VLV_WM_LEVEL_PM5, | |
918 | VLV_WM_LEVEL_DDR_DVFS, | |
6eb1a681 VS |
919 | }; |
920 | ||
262cd2e1 VS |
921 | /* latency must be in 0.1us units. */ |
922 | static unsigned int vlv_wm_method2(unsigned int pixel_rate, | |
923 | unsigned int pipe_htotal, | |
924 | unsigned int horiz_pixels, | |
ac484963 | 925 | unsigned int cpp, |
262cd2e1 VS |
926 | unsigned int latency) |
927 | { | |
928 | unsigned int ret; | |
929 | ||
930 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); | |
ac484963 | 931 | ret = (ret + 1) * horiz_pixels * cpp; |
262cd2e1 VS |
932 | ret = DIV_ROUND_UP(ret, 64); |
933 | ||
934 | return ret; | |
935 | } | |
936 | ||
bb726519 | 937 | static void vlv_setup_wm_latency(struct drm_i915_private *dev_priv) |
262cd2e1 | 938 | { |
262cd2e1 VS |
939 | /* all latencies in usec */ |
940 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3; | |
941 | ||
58590c14 VS |
942 | dev_priv->wm.max_level = VLV_WM_LEVEL_PM2; |
943 | ||
262cd2e1 VS |
944 | if (IS_CHERRYVIEW(dev_priv)) { |
945 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12; | |
946 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33; | |
58590c14 VS |
947 | |
948 | dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS; | |
262cd2e1 VS |
949 | } |
950 | } | |
951 | ||
952 | static uint16_t vlv_compute_wm_level(struct intel_plane *plane, | |
953 | struct intel_crtc *crtc, | |
954 | const struct intel_plane_state *state, | |
955 | int level) | |
956 | { | |
957 | struct drm_i915_private *dev_priv = to_i915(plane->base.dev); | |
ac484963 | 958 | int clock, htotal, cpp, width, wm; |
262cd2e1 VS |
959 | |
960 | if (dev_priv->wm.pri_latency[level] == 0) | |
961 | return USHRT_MAX; | |
962 | ||
936e71e3 | 963 | if (!state->base.visible) |
262cd2e1 VS |
964 | return 0; |
965 | ||
ac484963 | 966 | cpp = drm_format_plane_cpp(state->base.fb->pixel_format, 0); |
262cd2e1 VS |
967 | clock = crtc->config->base.adjusted_mode.crtc_clock; |
968 | htotal = crtc->config->base.adjusted_mode.crtc_htotal; | |
969 | width = crtc->config->pipe_src_w; | |
970 | if (WARN_ON(htotal == 0)) | |
971 | htotal = 1; | |
972 | ||
973 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { | |
974 | /* | |
975 | * FIXME the formula gives values that are | |
976 | * too big for the cursor FIFO, and hence we | |
977 | * would never be able to use cursors. For | |
978 | * now just hardcode the watermark. | |
979 | */ | |
980 | wm = 63; | |
981 | } else { | |
ac484963 | 982 | wm = vlv_wm_method2(clock, htotal, width, cpp, |
262cd2e1 VS |
983 | dev_priv->wm.pri_latency[level] * 10); |
984 | } | |
985 | ||
986 | return min_t(int, wm, USHRT_MAX); | |
987 | } | |
988 | ||
54f1b6e1 VS |
989 | static void vlv_compute_fifo(struct intel_crtc *crtc) |
990 | { | |
991 | struct drm_device *dev = crtc->base.dev; | |
992 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
993 | struct intel_plane *plane; | |
994 | unsigned int total_rate = 0; | |
995 | const int fifo_size = 512 - 1; | |
996 | int fifo_extra, fifo_left = fifo_size; | |
997 | ||
998 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
999 | struct intel_plane_state *state = | |
1000 | to_intel_plane_state(plane->base.state); | |
1001 | ||
1002 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) | |
1003 | continue; | |
1004 | ||
936e71e3 | 1005 | if (state->base.visible) { |
54f1b6e1 VS |
1006 | wm_state->num_active_planes++; |
1007 | total_rate += drm_format_plane_cpp(state->base.fb->pixel_format, 0); | |
1008 | } | |
1009 | } | |
1010 | ||
1011 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1012 | struct intel_plane_state *state = | |
1013 | to_intel_plane_state(plane->base.state); | |
1014 | unsigned int rate; | |
1015 | ||
1016 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { | |
1017 | plane->wm.fifo_size = 63; | |
1018 | continue; | |
1019 | } | |
1020 | ||
936e71e3 | 1021 | if (!state->base.visible) { |
54f1b6e1 VS |
1022 | plane->wm.fifo_size = 0; |
1023 | continue; | |
1024 | } | |
1025 | ||
1026 | rate = drm_format_plane_cpp(state->base.fb->pixel_format, 0); | |
1027 | plane->wm.fifo_size = fifo_size * rate / total_rate; | |
1028 | fifo_left -= plane->wm.fifo_size; | |
1029 | } | |
1030 | ||
1031 | fifo_extra = DIV_ROUND_UP(fifo_left, wm_state->num_active_planes ?: 1); | |
1032 | ||
1033 | /* spread the remainder evenly */ | |
1034 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1035 | int plane_extra; | |
1036 | ||
1037 | if (fifo_left == 0) | |
1038 | break; | |
1039 | ||
1040 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) | |
1041 | continue; | |
1042 | ||
1043 | /* give it all to the first plane if none are active */ | |
1044 | if (plane->wm.fifo_size == 0 && | |
1045 | wm_state->num_active_planes) | |
1046 | continue; | |
1047 | ||
1048 | plane_extra = min(fifo_extra, fifo_left); | |
1049 | plane->wm.fifo_size += plane_extra; | |
1050 | fifo_left -= plane_extra; | |
1051 | } | |
1052 | ||
1053 | WARN_ON(fifo_left != 0); | |
1054 | } | |
1055 | ||
262cd2e1 VS |
1056 | static void vlv_invert_wms(struct intel_crtc *crtc) |
1057 | { | |
1058 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1059 | int level; | |
1060 | ||
1061 | for (level = 0; level < wm_state->num_levels; level++) { | |
1062 | struct drm_device *dev = crtc->base.dev; | |
b7f05d4a TU |
1063 | const int sr_fifo_size = |
1064 | INTEL_INFO(to_i915(dev))->num_pipes * 512 - 1; | |
262cd2e1 VS |
1065 | struct intel_plane *plane; |
1066 | ||
1067 | wm_state->sr[level].plane = sr_fifo_size - wm_state->sr[level].plane; | |
1068 | wm_state->sr[level].cursor = 63 - wm_state->sr[level].cursor; | |
1069 | ||
1070 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1071 | switch (plane->base.type) { | |
1072 | int sprite; | |
1073 | case DRM_PLANE_TYPE_CURSOR: | |
1074 | wm_state->wm[level].cursor = plane->wm.fifo_size - | |
1075 | wm_state->wm[level].cursor; | |
1076 | break; | |
1077 | case DRM_PLANE_TYPE_PRIMARY: | |
1078 | wm_state->wm[level].primary = plane->wm.fifo_size - | |
1079 | wm_state->wm[level].primary; | |
1080 | break; | |
1081 | case DRM_PLANE_TYPE_OVERLAY: | |
1082 | sprite = plane->plane; | |
1083 | wm_state->wm[level].sprite[sprite] = plane->wm.fifo_size - | |
1084 | wm_state->wm[level].sprite[sprite]; | |
1085 | break; | |
1086 | } | |
1087 | } | |
1088 | } | |
1089 | } | |
1090 | ||
26e1fe4f | 1091 | static void vlv_compute_wm(struct intel_crtc *crtc) |
262cd2e1 VS |
1092 | { |
1093 | struct drm_device *dev = crtc->base.dev; | |
b7f05d4a | 1094 | struct drm_i915_private *dev_priv = to_i915(dev); |
262cd2e1 VS |
1095 | struct vlv_wm_state *wm_state = &crtc->wm_state; |
1096 | struct intel_plane *plane; | |
b7f05d4a | 1097 | int sr_fifo_size = INTEL_INFO(dev_priv)->num_pipes * 512 - 1; |
262cd2e1 VS |
1098 | int level; |
1099 | ||
1100 | memset(wm_state, 0, sizeof(*wm_state)); | |
1101 | ||
852eb00d | 1102 | wm_state->cxsr = crtc->pipe != PIPE_C && crtc->wm.cxsr_allowed; |
b7f05d4a | 1103 | wm_state->num_levels = dev_priv->wm.max_level + 1; |
262cd2e1 VS |
1104 | |
1105 | wm_state->num_active_planes = 0; | |
262cd2e1 | 1106 | |
54f1b6e1 | 1107 | vlv_compute_fifo(crtc); |
262cd2e1 VS |
1108 | |
1109 | if (wm_state->num_active_planes != 1) | |
1110 | wm_state->cxsr = false; | |
1111 | ||
1112 | if (wm_state->cxsr) { | |
1113 | for (level = 0; level < wm_state->num_levels; level++) { | |
1114 | wm_state->sr[level].plane = sr_fifo_size; | |
1115 | wm_state->sr[level].cursor = 63; | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1120 | struct intel_plane_state *state = | |
1121 | to_intel_plane_state(plane->base.state); | |
1122 | ||
936e71e3 | 1123 | if (!state->base.visible) |
262cd2e1 VS |
1124 | continue; |
1125 | ||
1126 | /* normal watermarks */ | |
1127 | for (level = 0; level < wm_state->num_levels; level++) { | |
1128 | int wm = vlv_compute_wm_level(plane, crtc, state, level); | |
1129 | int max_wm = plane->base.type == DRM_PLANE_TYPE_CURSOR ? 63 : 511; | |
1130 | ||
1131 | /* hack */ | |
1132 | if (WARN_ON(level == 0 && wm > max_wm)) | |
1133 | wm = max_wm; | |
1134 | ||
1135 | if (wm > plane->wm.fifo_size) | |
1136 | break; | |
1137 | ||
1138 | switch (plane->base.type) { | |
1139 | int sprite; | |
1140 | case DRM_PLANE_TYPE_CURSOR: | |
1141 | wm_state->wm[level].cursor = wm; | |
1142 | break; | |
1143 | case DRM_PLANE_TYPE_PRIMARY: | |
1144 | wm_state->wm[level].primary = wm; | |
1145 | break; | |
1146 | case DRM_PLANE_TYPE_OVERLAY: | |
1147 | sprite = plane->plane; | |
1148 | wm_state->wm[level].sprite[sprite] = wm; | |
1149 | break; | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | wm_state->num_levels = level; | |
1154 | ||
1155 | if (!wm_state->cxsr) | |
1156 | continue; | |
1157 | ||
1158 | /* maxfifo watermarks */ | |
1159 | switch (plane->base.type) { | |
1160 | int sprite, level; | |
1161 | case DRM_PLANE_TYPE_CURSOR: | |
1162 | for (level = 0; level < wm_state->num_levels; level++) | |
1163 | wm_state->sr[level].cursor = | |
5a37ed0a | 1164 | wm_state->wm[level].cursor; |
262cd2e1 VS |
1165 | break; |
1166 | case DRM_PLANE_TYPE_PRIMARY: | |
1167 | for (level = 0; level < wm_state->num_levels; level++) | |
1168 | wm_state->sr[level].plane = | |
1169 | min(wm_state->sr[level].plane, | |
1170 | wm_state->wm[level].primary); | |
1171 | break; | |
1172 | case DRM_PLANE_TYPE_OVERLAY: | |
1173 | sprite = plane->plane; | |
1174 | for (level = 0; level < wm_state->num_levels; level++) | |
1175 | wm_state->sr[level].plane = | |
1176 | min(wm_state->sr[level].plane, | |
1177 | wm_state->wm[level].sprite[sprite]); | |
1178 | break; | |
1179 | } | |
1180 | } | |
1181 | ||
1182 | /* clear any (partially) filled invalid levels */ | |
b7f05d4a | 1183 | for (level = wm_state->num_levels; level < dev_priv->wm.max_level + 1; level++) { |
262cd2e1 VS |
1184 | memset(&wm_state->wm[level], 0, sizeof(wm_state->wm[level])); |
1185 | memset(&wm_state->sr[level], 0, sizeof(wm_state->sr[level])); | |
1186 | } | |
1187 | ||
1188 | vlv_invert_wms(crtc); | |
1189 | } | |
1190 | ||
54f1b6e1 VS |
1191 | #define VLV_FIFO(plane, value) \ |
1192 | (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV) | |
1193 | ||
1194 | static void vlv_pipe_set_fifo_size(struct intel_crtc *crtc) | |
1195 | { | |
1196 | struct drm_device *dev = crtc->base.dev; | |
1197 | struct drm_i915_private *dev_priv = to_i915(dev); | |
1198 | struct intel_plane *plane; | |
1199 | int sprite0_start = 0, sprite1_start = 0, fifo_size = 0; | |
1200 | ||
1201 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1202 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { | |
1203 | WARN_ON(plane->wm.fifo_size != 63); | |
1204 | continue; | |
1205 | } | |
1206 | ||
1207 | if (plane->base.type == DRM_PLANE_TYPE_PRIMARY) | |
1208 | sprite0_start = plane->wm.fifo_size; | |
1209 | else if (plane->plane == 0) | |
1210 | sprite1_start = sprite0_start + plane->wm.fifo_size; | |
1211 | else | |
1212 | fifo_size = sprite1_start + plane->wm.fifo_size; | |
1213 | } | |
1214 | ||
1215 | WARN_ON(fifo_size != 512 - 1); | |
1216 | ||
1217 | DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n", | |
1218 | pipe_name(crtc->pipe), sprite0_start, | |
1219 | sprite1_start, fifo_size); | |
1220 | ||
1221 | switch (crtc->pipe) { | |
1222 | uint32_t dsparb, dsparb2, dsparb3; | |
1223 | case PIPE_A: | |
1224 | dsparb = I915_READ(DSPARB); | |
1225 | dsparb2 = I915_READ(DSPARB2); | |
1226 | ||
1227 | dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) | | |
1228 | VLV_FIFO(SPRITEB, 0xff)); | |
1229 | dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) | | |
1230 | VLV_FIFO(SPRITEB, sprite1_start)); | |
1231 | ||
1232 | dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) | | |
1233 | VLV_FIFO(SPRITEB_HI, 0x1)); | |
1234 | dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) | | |
1235 | VLV_FIFO(SPRITEB_HI, sprite1_start >> 8)); | |
1236 | ||
1237 | I915_WRITE(DSPARB, dsparb); | |
1238 | I915_WRITE(DSPARB2, dsparb2); | |
1239 | break; | |
1240 | case PIPE_B: | |
1241 | dsparb = I915_READ(DSPARB); | |
1242 | dsparb2 = I915_READ(DSPARB2); | |
1243 | ||
1244 | dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) | | |
1245 | VLV_FIFO(SPRITED, 0xff)); | |
1246 | dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) | | |
1247 | VLV_FIFO(SPRITED, sprite1_start)); | |
1248 | ||
1249 | dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) | | |
1250 | VLV_FIFO(SPRITED_HI, 0xff)); | |
1251 | dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) | | |
1252 | VLV_FIFO(SPRITED_HI, sprite1_start >> 8)); | |
1253 | ||
1254 | I915_WRITE(DSPARB, dsparb); | |
1255 | I915_WRITE(DSPARB2, dsparb2); | |
1256 | break; | |
1257 | case PIPE_C: | |
1258 | dsparb3 = I915_READ(DSPARB3); | |
1259 | dsparb2 = I915_READ(DSPARB2); | |
1260 | ||
1261 | dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) | | |
1262 | VLV_FIFO(SPRITEF, 0xff)); | |
1263 | dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) | | |
1264 | VLV_FIFO(SPRITEF, sprite1_start)); | |
1265 | ||
1266 | dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) | | |
1267 | VLV_FIFO(SPRITEF_HI, 0xff)); | |
1268 | dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) | | |
1269 | VLV_FIFO(SPRITEF_HI, sprite1_start >> 8)); | |
1270 | ||
1271 | I915_WRITE(DSPARB3, dsparb3); | |
1272 | I915_WRITE(DSPARB2, dsparb2); | |
1273 | break; | |
1274 | default: | |
1275 | break; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | #undef VLV_FIFO | |
1280 | ||
262cd2e1 VS |
1281 | static void vlv_merge_wm(struct drm_device *dev, |
1282 | struct vlv_wm_values *wm) | |
1283 | { | |
1284 | struct intel_crtc *crtc; | |
1285 | int num_active_crtcs = 0; | |
1286 | ||
58590c14 | 1287 | wm->level = to_i915(dev)->wm.max_level; |
262cd2e1 VS |
1288 | wm->cxsr = true; |
1289 | ||
1290 | for_each_intel_crtc(dev, crtc) { | |
1291 | const struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1292 | ||
1293 | if (!crtc->active) | |
1294 | continue; | |
1295 | ||
1296 | if (!wm_state->cxsr) | |
1297 | wm->cxsr = false; | |
1298 | ||
1299 | num_active_crtcs++; | |
1300 | wm->level = min_t(int, wm->level, wm_state->num_levels - 1); | |
1301 | } | |
1302 | ||
1303 | if (num_active_crtcs != 1) | |
1304 | wm->cxsr = false; | |
1305 | ||
6f9c784b VS |
1306 | if (num_active_crtcs > 1) |
1307 | wm->level = VLV_WM_LEVEL_PM2; | |
1308 | ||
262cd2e1 VS |
1309 | for_each_intel_crtc(dev, crtc) { |
1310 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1311 | enum pipe pipe = crtc->pipe; | |
1312 | ||
1313 | if (!crtc->active) | |
1314 | continue; | |
1315 | ||
1316 | wm->pipe[pipe] = wm_state->wm[wm->level]; | |
1317 | if (wm->cxsr) | |
1318 | wm->sr = wm_state->sr[wm->level]; | |
1319 | ||
1320 | wm->ddl[pipe].primary = DDL_PRECISION_HIGH | 2; | |
1321 | wm->ddl[pipe].sprite[0] = DDL_PRECISION_HIGH | 2; | |
1322 | wm->ddl[pipe].sprite[1] = DDL_PRECISION_HIGH | 2; | |
1323 | wm->ddl[pipe].cursor = DDL_PRECISION_HIGH | 2; | |
1324 | } | |
1325 | } | |
1326 | ||
432081bc | 1327 | static void vlv_update_wm(struct intel_crtc *crtc) |
262cd2e1 | 1328 | { |
432081bc | 1329 | struct drm_device *dev = crtc->base.dev; |
fac5e23e | 1330 | struct drm_i915_private *dev_priv = to_i915(dev); |
432081bc | 1331 | enum pipe pipe = crtc->pipe; |
262cd2e1 VS |
1332 | struct vlv_wm_values wm = {}; |
1333 | ||
432081bc | 1334 | vlv_compute_wm(crtc); |
262cd2e1 VS |
1335 | vlv_merge_wm(dev, &wm); |
1336 | ||
54f1b6e1 VS |
1337 | if (memcmp(&dev_priv->wm.vlv, &wm, sizeof(wm)) == 0) { |
1338 | /* FIXME should be part of crtc atomic commit */ | |
432081bc | 1339 | vlv_pipe_set_fifo_size(crtc); |
262cd2e1 | 1340 | return; |
54f1b6e1 | 1341 | } |
262cd2e1 VS |
1342 | |
1343 | if (wm.level < VLV_WM_LEVEL_DDR_DVFS && | |
1344 | dev_priv->wm.vlv.level >= VLV_WM_LEVEL_DDR_DVFS) | |
1345 | chv_set_memory_dvfs(dev_priv, false); | |
1346 | ||
1347 | if (wm.level < VLV_WM_LEVEL_PM5 && | |
1348 | dev_priv->wm.vlv.level >= VLV_WM_LEVEL_PM5) | |
1349 | chv_set_memory_pm5(dev_priv, false); | |
1350 | ||
852eb00d | 1351 | if (!wm.cxsr && dev_priv->wm.vlv.cxsr) |
262cd2e1 | 1352 | intel_set_memory_cxsr(dev_priv, false); |
262cd2e1 | 1353 | |
54f1b6e1 | 1354 | /* FIXME should be part of crtc atomic commit */ |
432081bc | 1355 | vlv_pipe_set_fifo_size(crtc); |
54f1b6e1 | 1356 | |
432081bc | 1357 | vlv_write_wm_values(crtc, &wm); |
262cd2e1 VS |
1358 | |
1359 | DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, " | |
1360 | "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n", | |
1361 | pipe_name(pipe), wm.pipe[pipe].primary, wm.pipe[pipe].cursor, | |
1362 | wm.pipe[pipe].sprite[0], wm.pipe[pipe].sprite[1], | |
1363 | wm.sr.plane, wm.sr.cursor, wm.level, wm.cxsr); | |
1364 | ||
852eb00d | 1365 | if (wm.cxsr && !dev_priv->wm.vlv.cxsr) |
262cd2e1 | 1366 | intel_set_memory_cxsr(dev_priv, true); |
262cd2e1 VS |
1367 | |
1368 | if (wm.level >= VLV_WM_LEVEL_PM5 && | |
1369 | dev_priv->wm.vlv.level < VLV_WM_LEVEL_PM5) | |
1370 | chv_set_memory_pm5(dev_priv, true); | |
1371 | ||
1372 | if (wm.level >= VLV_WM_LEVEL_DDR_DVFS && | |
1373 | dev_priv->wm.vlv.level < VLV_WM_LEVEL_DDR_DVFS) | |
1374 | chv_set_memory_dvfs(dev_priv, true); | |
1375 | ||
1376 | dev_priv->wm.vlv = wm; | |
3c2777fd VS |
1377 | } |
1378 | ||
ae80152d VS |
1379 | #define single_plane_enabled(mask) is_power_of_2(mask) |
1380 | ||
432081bc | 1381 | static void g4x_update_wm(struct intel_crtc *crtc) |
b445e3b0 | 1382 | { |
b91eb5cc | 1383 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); |
b445e3b0 | 1384 | static const int sr_latency_ns = 12000; |
b445e3b0 ED |
1385 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; |
1386 | int plane_sr, cursor_sr; | |
1387 | unsigned int enabled = 0; | |
9858425c | 1388 | bool cxsr_enabled; |
b445e3b0 | 1389 | |
f0ce2310 | 1390 | if (g4x_compute_wm0(dev_priv, PIPE_A, |
5aef6003 CW |
1391 | &g4x_wm_info, pessimal_latency_ns, |
1392 | &g4x_cursor_wm_info, pessimal_latency_ns, | |
b445e3b0 | 1393 | &planea_wm, &cursora_wm)) |
51cea1f4 | 1394 | enabled |= 1 << PIPE_A; |
b445e3b0 | 1395 | |
f0ce2310 | 1396 | if (g4x_compute_wm0(dev_priv, PIPE_B, |
5aef6003 CW |
1397 | &g4x_wm_info, pessimal_latency_ns, |
1398 | &g4x_cursor_wm_info, pessimal_latency_ns, | |
b445e3b0 | 1399 | &planeb_wm, &cursorb_wm)) |
51cea1f4 | 1400 | enabled |= 1 << PIPE_B; |
b445e3b0 | 1401 | |
b445e3b0 | 1402 | if (single_plane_enabled(enabled) && |
f0ce2310 | 1403 | g4x_compute_srwm(dev_priv, ffs(enabled) - 1, |
b445e3b0 ED |
1404 | sr_latency_ns, |
1405 | &g4x_wm_info, | |
1406 | &g4x_cursor_wm_info, | |
52bd02d8 | 1407 | &plane_sr, &cursor_sr)) { |
9858425c | 1408 | cxsr_enabled = true; |
52bd02d8 | 1409 | } else { |
9858425c | 1410 | cxsr_enabled = false; |
5209b1f4 | 1411 | intel_set_memory_cxsr(dev_priv, false); |
52bd02d8 CW |
1412 | plane_sr = cursor_sr = 0; |
1413 | } | |
b445e3b0 | 1414 | |
a5043453 VS |
1415 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, " |
1416 | "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", | |
b445e3b0 ED |
1417 | planea_wm, cursora_wm, |
1418 | planeb_wm, cursorb_wm, | |
1419 | plane_sr, cursor_sr); | |
1420 | ||
1421 | I915_WRITE(DSPFW1, | |
f4998963 VS |
1422 | FW_WM(plane_sr, SR) | |
1423 | FW_WM(cursorb_wm, CURSORB) | | |
1424 | FW_WM(planeb_wm, PLANEB) | | |
1425 | FW_WM(planea_wm, PLANEA)); | |
b445e3b0 | 1426 | I915_WRITE(DSPFW2, |
8c919b28 | 1427 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
f4998963 | 1428 | FW_WM(cursora_wm, CURSORA)); |
b445e3b0 ED |
1429 | /* HPLL off in SR has some issues on G4x... disable it */ |
1430 | I915_WRITE(DSPFW3, | |
8c919b28 | 1431 | (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) | |
f4998963 | 1432 | FW_WM(cursor_sr, CURSOR_SR)); |
9858425c ID |
1433 | |
1434 | if (cxsr_enabled) | |
1435 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
1436 | } |
1437 | ||
432081bc | 1438 | static void i965_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 1439 | { |
ffc7a76b | 1440 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
efc2611e | 1441 | struct intel_crtc *crtc; |
b445e3b0 ED |
1442 | int srwm = 1; |
1443 | int cursor_sr = 16; | |
9858425c | 1444 | bool cxsr_enabled; |
b445e3b0 ED |
1445 | |
1446 | /* Calc sr entries for one plane configs */ | |
ffc7a76b | 1447 | crtc = single_enabled_crtc(dev_priv); |
b445e3b0 ED |
1448 | if (crtc) { |
1449 | /* self-refresh has much higher latency */ | |
1450 | static const int sr_latency_ns = 12000; | |
efc2611e VS |
1451 | const struct drm_display_mode *adjusted_mode = |
1452 | &crtc->config->base.adjusted_mode; | |
1453 | const struct drm_framebuffer *fb = | |
1454 | crtc->base.primary->state->fb; | |
241bfc38 | 1455 | int clock = adjusted_mode->crtc_clock; |
fec8cba3 | 1456 | int htotal = adjusted_mode->crtc_htotal; |
efc2611e VS |
1457 | int hdisplay = crtc->config->pipe_src_w; |
1458 | int cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
b445e3b0 ED |
1459 | unsigned long line_time_us; |
1460 | int entries; | |
1461 | ||
922044c9 | 1462 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 ED |
1463 | |
1464 | /* Use ns/us then divide to preserve precision */ | |
1465 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
ac484963 | 1466 | cpp * hdisplay; |
b445e3b0 ED |
1467 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); |
1468 | srwm = I965_FIFO_SIZE - entries; | |
1469 | if (srwm < 0) | |
1470 | srwm = 1; | |
1471 | srwm &= 0x1ff; | |
1472 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", | |
1473 | entries, srwm); | |
1474 | ||
1475 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
efc2611e | 1476 | cpp * crtc->base.cursor->state->crtc_w; |
b445e3b0 ED |
1477 | entries = DIV_ROUND_UP(entries, |
1478 | i965_cursor_wm_info.cacheline_size); | |
1479 | cursor_sr = i965_cursor_wm_info.fifo_size - | |
1480 | (entries + i965_cursor_wm_info.guard_size); | |
1481 | ||
1482 | if (cursor_sr > i965_cursor_wm_info.max_wm) | |
1483 | cursor_sr = i965_cursor_wm_info.max_wm; | |
1484 | ||
1485 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " | |
1486 | "cursor %d\n", srwm, cursor_sr); | |
1487 | ||
9858425c | 1488 | cxsr_enabled = true; |
b445e3b0 | 1489 | } else { |
9858425c | 1490 | cxsr_enabled = false; |
b445e3b0 | 1491 | /* Turn off self refresh if both pipes are enabled */ |
5209b1f4 | 1492 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
1493 | } |
1494 | ||
1495 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", | |
1496 | srwm); | |
1497 | ||
1498 | /* 965 has limitations... */ | |
f4998963 VS |
1499 | I915_WRITE(DSPFW1, FW_WM(srwm, SR) | |
1500 | FW_WM(8, CURSORB) | | |
1501 | FW_WM(8, PLANEB) | | |
1502 | FW_WM(8, PLANEA)); | |
1503 | I915_WRITE(DSPFW2, FW_WM(8, CURSORA) | | |
1504 | FW_WM(8, PLANEC_OLD)); | |
b445e3b0 | 1505 | /* update cursor SR watermark */ |
f4998963 | 1506 | I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR)); |
9858425c ID |
1507 | |
1508 | if (cxsr_enabled) | |
1509 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
1510 | } |
1511 | ||
f4998963 VS |
1512 | #undef FW_WM |
1513 | ||
432081bc | 1514 | static void i9xx_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 1515 | { |
ffc7a76b | 1516 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
b445e3b0 ED |
1517 | const struct intel_watermark_params *wm_info; |
1518 | uint32_t fwater_lo; | |
1519 | uint32_t fwater_hi; | |
1520 | int cwm, srwm = 1; | |
1521 | int fifo_size; | |
1522 | int planea_wm, planeb_wm; | |
efc2611e | 1523 | struct intel_crtc *crtc, *enabled = NULL; |
b445e3b0 | 1524 | |
a9097be4 | 1525 | if (IS_I945GM(dev_priv)) |
b445e3b0 | 1526 | wm_info = &i945_wm_info; |
5db94019 | 1527 | else if (!IS_GEN2(dev_priv)) |
b445e3b0 ED |
1528 | wm_info = &i915_wm_info; |
1529 | else | |
9d539105 | 1530 | wm_info = &i830_a_wm_info; |
b445e3b0 | 1531 | |
ef0f5e93 | 1532 | fifo_size = dev_priv->display.get_fifo_size(dev_priv, 0); |
b91eb5cc | 1533 | crtc = intel_get_crtc_for_plane(dev_priv, 0); |
efc2611e VS |
1534 | if (intel_crtc_active(crtc)) { |
1535 | const struct drm_display_mode *adjusted_mode = | |
1536 | &crtc->config->base.adjusted_mode; | |
1537 | const struct drm_framebuffer *fb = | |
1538 | crtc->base.primary->state->fb; | |
1539 | int cpp; | |
1540 | ||
5db94019 | 1541 | if (IS_GEN2(dev_priv)) |
b9e0bda3 | 1542 | cpp = 4; |
efc2611e VS |
1543 | else |
1544 | cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
b9e0bda3 | 1545 | |
241bfc38 | 1546 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
b9e0bda3 | 1547 | wm_info, fifo_size, cpp, |
5aef6003 | 1548 | pessimal_latency_ns); |
b445e3b0 | 1549 | enabled = crtc; |
9d539105 | 1550 | } else { |
b445e3b0 | 1551 | planea_wm = fifo_size - wm_info->guard_size; |
9d539105 VS |
1552 | if (planea_wm > (long)wm_info->max_wm) |
1553 | planea_wm = wm_info->max_wm; | |
1554 | } | |
1555 | ||
5db94019 | 1556 | if (IS_GEN2(dev_priv)) |
9d539105 | 1557 | wm_info = &i830_bc_wm_info; |
b445e3b0 | 1558 | |
ef0f5e93 | 1559 | fifo_size = dev_priv->display.get_fifo_size(dev_priv, 1); |
b91eb5cc | 1560 | crtc = intel_get_crtc_for_plane(dev_priv, 1); |
efc2611e VS |
1561 | if (intel_crtc_active(crtc)) { |
1562 | const struct drm_display_mode *adjusted_mode = | |
1563 | &crtc->config->base.adjusted_mode; | |
1564 | const struct drm_framebuffer *fb = | |
1565 | crtc->base.primary->state->fb; | |
1566 | int cpp; | |
1567 | ||
5db94019 | 1568 | if (IS_GEN2(dev_priv)) |
b9e0bda3 | 1569 | cpp = 4; |
efc2611e VS |
1570 | else |
1571 | cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
b9e0bda3 | 1572 | |
241bfc38 | 1573 | planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
b9e0bda3 | 1574 | wm_info, fifo_size, cpp, |
5aef6003 | 1575 | pessimal_latency_ns); |
b445e3b0 ED |
1576 | if (enabled == NULL) |
1577 | enabled = crtc; | |
1578 | else | |
1579 | enabled = NULL; | |
9d539105 | 1580 | } else { |
b445e3b0 | 1581 | planeb_wm = fifo_size - wm_info->guard_size; |
9d539105 VS |
1582 | if (planeb_wm > (long)wm_info->max_wm) |
1583 | planeb_wm = wm_info->max_wm; | |
1584 | } | |
b445e3b0 ED |
1585 | |
1586 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); | |
1587 | ||
50a0bc90 | 1588 | if (IS_I915GM(dev_priv) && enabled) { |
2ff8fde1 | 1589 | struct drm_i915_gem_object *obj; |
2ab1bc9d | 1590 | |
efc2611e | 1591 | obj = intel_fb_obj(enabled->base.primary->state->fb); |
2ab1bc9d DV |
1592 | |
1593 | /* self-refresh seems busted with untiled */ | |
3e510a8e | 1594 | if (!i915_gem_object_is_tiled(obj)) |
2ab1bc9d DV |
1595 | enabled = NULL; |
1596 | } | |
1597 | ||
b445e3b0 ED |
1598 | /* |
1599 | * Overlay gets an aggressive default since video jitter is bad. | |
1600 | */ | |
1601 | cwm = 2; | |
1602 | ||
1603 | /* Play safe and disable self-refresh before adjusting watermarks. */ | |
5209b1f4 | 1604 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
1605 | |
1606 | /* Calc sr entries for one plane configs */ | |
03427fcb | 1607 | if (HAS_FW_BLC(dev_priv) && enabled) { |
b445e3b0 ED |
1608 | /* self-refresh has much higher latency */ |
1609 | static const int sr_latency_ns = 6000; | |
efc2611e VS |
1610 | const struct drm_display_mode *adjusted_mode = |
1611 | &enabled->config->base.adjusted_mode; | |
1612 | const struct drm_framebuffer *fb = | |
1613 | enabled->base.primary->state->fb; | |
241bfc38 | 1614 | int clock = adjusted_mode->crtc_clock; |
fec8cba3 | 1615 | int htotal = adjusted_mode->crtc_htotal; |
efc2611e VS |
1616 | int hdisplay = enabled->config->pipe_src_w; |
1617 | int cpp; | |
b445e3b0 ED |
1618 | unsigned long line_time_us; |
1619 | int entries; | |
1620 | ||
50a0bc90 | 1621 | if (IS_I915GM(dev_priv) || IS_I945GM(dev_priv)) |
2d1b5056 | 1622 | cpp = 4; |
efc2611e VS |
1623 | else |
1624 | cpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
2d1b5056 | 1625 | |
922044c9 | 1626 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 ED |
1627 | |
1628 | /* Use ns/us then divide to preserve precision */ | |
1629 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
ac484963 | 1630 | cpp * hdisplay; |
b445e3b0 ED |
1631 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); |
1632 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); | |
1633 | srwm = wm_info->fifo_size - entries; | |
1634 | if (srwm < 0) | |
1635 | srwm = 1; | |
1636 | ||
50a0bc90 | 1637 | if (IS_I945G(dev_priv) || IS_I945GM(dev_priv)) |
b445e3b0 ED |
1638 | I915_WRITE(FW_BLC_SELF, |
1639 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); | |
acb91359 | 1640 | else |
b445e3b0 ED |
1641 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); |
1642 | } | |
1643 | ||
1644 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", | |
1645 | planea_wm, planeb_wm, cwm, srwm); | |
1646 | ||
1647 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); | |
1648 | fwater_hi = (cwm & 0x1f); | |
1649 | ||
1650 | /* Set request length to 8 cachelines per fetch */ | |
1651 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); | |
1652 | fwater_hi = fwater_hi | (1 << 8); | |
1653 | ||
1654 | I915_WRITE(FW_BLC, fwater_lo); | |
1655 | I915_WRITE(FW_BLC2, fwater_hi); | |
1656 | ||
5209b1f4 ID |
1657 | if (enabled) |
1658 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
1659 | } |
1660 | ||
432081bc | 1661 | static void i845_update_wm(struct intel_crtc *unused_crtc) |
b445e3b0 | 1662 | { |
ffc7a76b | 1663 | struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev); |
efc2611e | 1664 | struct intel_crtc *crtc; |
241bfc38 | 1665 | const struct drm_display_mode *adjusted_mode; |
b445e3b0 ED |
1666 | uint32_t fwater_lo; |
1667 | int planea_wm; | |
1668 | ||
ffc7a76b | 1669 | crtc = single_enabled_crtc(dev_priv); |
b445e3b0 ED |
1670 | if (crtc == NULL) |
1671 | return; | |
1672 | ||
efc2611e | 1673 | adjusted_mode = &crtc->config->base.adjusted_mode; |
241bfc38 | 1674 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
feb56b93 | 1675 | &i845_wm_info, |
ef0f5e93 | 1676 | dev_priv->display.get_fifo_size(dev_priv, 0), |
5aef6003 | 1677 | 4, pessimal_latency_ns); |
b445e3b0 ED |
1678 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
1679 | fwater_lo |= (3<<8) | planea_wm; | |
1680 | ||
1681 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); | |
1682 | ||
1683 | I915_WRITE(FW_BLC, fwater_lo); | |
1684 | } | |
1685 | ||
8cfb3407 | 1686 | uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state *pipe_config) |
801bcfff | 1687 | { |
fd4daa9c | 1688 | uint32_t pixel_rate; |
801bcfff | 1689 | |
8cfb3407 | 1690 | pixel_rate = pipe_config->base.adjusted_mode.crtc_clock; |
801bcfff PZ |
1691 | |
1692 | /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to | |
1693 | * adjust the pixel_rate here. */ | |
1694 | ||
8cfb3407 | 1695 | if (pipe_config->pch_pfit.enabled) { |
801bcfff | 1696 | uint64_t pipe_w, pipe_h, pfit_w, pfit_h; |
8cfb3407 VS |
1697 | uint32_t pfit_size = pipe_config->pch_pfit.size; |
1698 | ||
1699 | pipe_w = pipe_config->pipe_src_w; | |
1700 | pipe_h = pipe_config->pipe_src_h; | |
801bcfff | 1701 | |
801bcfff PZ |
1702 | pfit_w = (pfit_size >> 16) & 0xFFFF; |
1703 | pfit_h = pfit_size & 0xFFFF; | |
1704 | if (pipe_w < pfit_w) | |
1705 | pipe_w = pfit_w; | |
1706 | if (pipe_h < pfit_h) | |
1707 | pipe_h = pfit_h; | |
1708 | ||
15126882 MR |
1709 | if (WARN_ON(!pfit_w || !pfit_h)) |
1710 | return pixel_rate; | |
1711 | ||
801bcfff PZ |
1712 | pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h, |
1713 | pfit_w * pfit_h); | |
1714 | } | |
1715 | ||
1716 | return pixel_rate; | |
1717 | } | |
1718 | ||
37126462 | 1719 | /* latency must be in 0.1us units. */ |
ac484963 | 1720 | static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t cpp, uint32_t latency) |
801bcfff PZ |
1721 | { |
1722 | uint64_t ret; | |
1723 | ||
3312ba65 VS |
1724 | if (WARN(latency == 0, "Latency value missing\n")) |
1725 | return UINT_MAX; | |
1726 | ||
ac484963 | 1727 | ret = (uint64_t) pixel_rate * cpp * latency; |
801bcfff PZ |
1728 | ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2; |
1729 | ||
1730 | return ret; | |
1731 | } | |
1732 | ||
37126462 | 1733 | /* latency must be in 0.1us units. */ |
23297044 | 1734 | static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, |
ac484963 | 1735 | uint32_t horiz_pixels, uint8_t cpp, |
801bcfff PZ |
1736 | uint32_t latency) |
1737 | { | |
1738 | uint32_t ret; | |
1739 | ||
3312ba65 VS |
1740 | if (WARN(latency == 0, "Latency value missing\n")) |
1741 | return UINT_MAX; | |
15126882 MR |
1742 | if (WARN_ON(!pipe_htotal)) |
1743 | return UINT_MAX; | |
3312ba65 | 1744 | |
801bcfff | 1745 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); |
ac484963 | 1746 | ret = (ret + 1) * horiz_pixels * cpp; |
801bcfff PZ |
1747 | ret = DIV_ROUND_UP(ret, 64) + 2; |
1748 | return ret; | |
1749 | } | |
1750 | ||
23297044 | 1751 | static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels, |
ac484963 | 1752 | uint8_t cpp) |
cca32e9a | 1753 | { |
15126882 MR |
1754 | /* |
1755 | * Neither of these should be possible since this function shouldn't be | |
1756 | * called if the CRTC is off or the plane is invisible. But let's be | |
1757 | * extra paranoid to avoid a potential divide-by-zero if we screw up | |
1758 | * elsewhere in the driver. | |
1759 | */ | |
ac484963 | 1760 | if (WARN_ON(!cpp)) |
15126882 MR |
1761 | return 0; |
1762 | if (WARN_ON(!horiz_pixels)) | |
1763 | return 0; | |
1764 | ||
ac484963 | 1765 | return DIV_ROUND_UP(pri_val * 64, horiz_pixels * cpp) + 2; |
cca32e9a PZ |
1766 | } |
1767 | ||
820c1980 | 1768 | struct ilk_wm_maximums { |
cca32e9a PZ |
1769 | uint16_t pri; |
1770 | uint16_t spr; | |
1771 | uint16_t cur; | |
1772 | uint16_t fbc; | |
1773 | }; | |
1774 | ||
37126462 VS |
1775 | /* |
1776 | * For both WM_PIPE and WM_LP. | |
1777 | * mem_value must be in 0.1us units. | |
1778 | */ | |
7221fc33 | 1779 | static uint32_t ilk_compute_pri_wm(const struct intel_crtc_state *cstate, |
43d59eda | 1780 | const struct intel_plane_state *pstate, |
cca32e9a PZ |
1781 | uint32_t mem_value, |
1782 | bool is_lp) | |
801bcfff | 1783 | { |
ac484963 VS |
1784 | int cpp = pstate->base.fb ? |
1785 | drm_format_plane_cpp(pstate->base.fb->pixel_format, 0) : 0; | |
cca32e9a PZ |
1786 | uint32_t method1, method2; |
1787 | ||
936e71e3 | 1788 | if (!cstate->base.active || !pstate->base.visible) |
801bcfff PZ |
1789 | return 0; |
1790 | ||
ac484963 | 1791 | method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), cpp, mem_value); |
cca32e9a PZ |
1792 | |
1793 | if (!is_lp) | |
1794 | return method1; | |
1795 | ||
7221fc33 MR |
1796 | method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate), |
1797 | cstate->base.adjusted_mode.crtc_htotal, | |
936e71e3 | 1798 | drm_rect_width(&pstate->base.dst), |
ac484963 | 1799 | cpp, mem_value); |
cca32e9a PZ |
1800 | |
1801 | return min(method1, method2); | |
801bcfff PZ |
1802 | } |
1803 | ||
37126462 VS |
1804 | /* |
1805 | * For both WM_PIPE and WM_LP. | |
1806 | * mem_value must be in 0.1us units. | |
1807 | */ | |
7221fc33 | 1808 | static uint32_t ilk_compute_spr_wm(const struct intel_crtc_state *cstate, |
43d59eda | 1809 | const struct intel_plane_state *pstate, |
801bcfff PZ |
1810 | uint32_t mem_value) |
1811 | { | |
ac484963 VS |
1812 | int cpp = pstate->base.fb ? |
1813 | drm_format_plane_cpp(pstate->base.fb->pixel_format, 0) : 0; | |
801bcfff PZ |
1814 | uint32_t method1, method2; |
1815 | ||
936e71e3 | 1816 | if (!cstate->base.active || !pstate->base.visible) |
801bcfff PZ |
1817 | return 0; |
1818 | ||
ac484963 | 1819 | method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), cpp, mem_value); |
7221fc33 MR |
1820 | method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate), |
1821 | cstate->base.adjusted_mode.crtc_htotal, | |
936e71e3 | 1822 | drm_rect_width(&pstate->base.dst), |
ac484963 | 1823 | cpp, mem_value); |
801bcfff PZ |
1824 | return min(method1, method2); |
1825 | } | |
1826 | ||
37126462 VS |
1827 | /* |
1828 | * For both WM_PIPE and WM_LP. | |
1829 | * mem_value must be in 0.1us units. | |
1830 | */ | |
7221fc33 | 1831 | static uint32_t ilk_compute_cur_wm(const struct intel_crtc_state *cstate, |
43d59eda | 1832 | const struct intel_plane_state *pstate, |
801bcfff PZ |
1833 | uint32_t mem_value) |
1834 | { | |
b2435692 MR |
1835 | /* |
1836 | * We treat the cursor plane as always-on for the purposes of watermark | |
1837 | * calculation. Until we have two-stage watermark programming merged, | |
1838 | * this is necessary to avoid flickering. | |
1839 | */ | |
1840 | int cpp = 4; | |
936e71e3 | 1841 | int width = pstate->base.visible ? pstate->base.crtc_w : 64; |
43d59eda | 1842 | |
b2435692 | 1843 | if (!cstate->base.active) |
801bcfff PZ |
1844 | return 0; |
1845 | ||
7221fc33 MR |
1846 | return ilk_wm_method2(ilk_pipe_pixel_rate(cstate), |
1847 | cstate->base.adjusted_mode.crtc_htotal, | |
b2435692 | 1848 | width, cpp, mem_value); |
801bcfff PZ |
1849 | } |
1850 | ||
cca32e9a | 1851 | /* Only for WM_LP. */ |
7221fc33 | 1852 | static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate, |
43d59eda | 1853 | const struct intel_plane_state *pstate, |
1fda9882 | 1854 | uint32_t pri_val) |
cca32e9a | 1855 | { |
ac484963 VS |
1856 | int cpp = pstate->base.fb ? |
1857 | drm_format_plane_cpp(pstate->base.fb->pixel_format, 0) : 0; | |
43d59eda | 1858 | |
936e71e3 | 1859 | if (!cstate->base.active || !pstate->base.visible) |
cca32e9a PZ |
1860 | return 0; |
1861 | ||
936e71e3 | 1862 | return ilk_wm_fbc(pri_val, drm_rect_width(&pstate->base.dst), cpp); |
cca32e9a PZ |
1863 | } |
1864 | ||
175fded1 TU |
1865 | static unsigned int |
1866 | ilk_display_fifo_size(const struct drm_i915_private *dev_priv) | |
158ae64f | 1867 | { |
175fded1 | 1868 | if (INTEL_GEN(dev_priv) >= 8) |
416f4727 | 1869 | return 3072; |
175fded1 | 1870 | else if (INTEL_GEN(dev_priv) >= 7) |
158ae64f VS |
1871 | return 768; |
1872 | else | |
1873 | return 512; | |
1874 | } | |
1875 | ||
175fded1 TU |
1876 | static unsigned int |
1877 | ilk_plane_wm_reg_max(const struct drm_i915_private *dev_priv, | |
1878 | int level, bool is_sprite) | |
4e975081 | 1879 | { |
175fded1 | 1880 | if (INTEL_GEN(dev_priv) >= 8) |
4e975081 VS |
1881 | /* BDW primary/sprite plane watermarks */ |
1882 | return level == 0 ? 255 : 2047; | |
175fded1 | 1883 | else if (INTEL_GEN(dev_priv) >= 7) |
4e975081 VS |
1884 | /* IVB/HSW primary/sprite plane watermarks */ |
1885 | return level == 0 ? 127 : 1023; | |
1886 | else if (!is_sprite) | |
1887 | /* ILK/SNB primary plane watermarks */ | |
1888 | return level == 0 ? 127 : 511; | |
1889 | else | |
1890 | /* ILK/SNB sprite plane watermarks */ | |
1891 | return level == 0 ? 63 : 255; | |
1892 | } | |
1893 | ||
175fded1 TU |
1894 | static unsigned int |
1895 | ilk_cursor_wm_reg_max(const struct drm_i915_private *dev_priv, int level) | |
4e975081 | 1896 | { |
175fded1 | 1897 | if (INTEL_GEN(dev_priv) >= 7) |
4e975081 VS |
1898 | return level == 0 ? 63 : 255; |
1899 | else | |
1900 | return level == 0 ? 31 : 63; | |
1901 | } | |
1902 | ||
175fded1 | 1903 | static unsigned int ilk_fbc_wm_reg_max(const struct drm_i915_private *dev_priv) |
4e975081 | 1904 | { |
175fded1 | 1905 | if (INTEL_GEN(dev_priv) >= 8) |
4e975081 VS |
1906 | return 31; |
1907 | else | |
1908 | return 15; | |
1909 | } | |
1910 | ||
158ae64f VS |
1911 | /* Calculate the maximum primary/sprite plane watermark */ |
1912 | static unsigned int ilk_plane_wm_max(const struct drm_device *dev, | |
1913 | int level, | |
240264f4 | 1914 | const struct intel_wm_config *config, |
158ae64f VS |
1915 | enum intel_ddb_partitioning ddb_partitioning, |
1916 | bool is_sprite) | |
1917 | { | |
175fded1 TU |
1918 | struct drm_i915_private *dev_priv = to_i915(dev); |
1919 | unsigned int fifo_size = ilk_display_fifo_size(dev_priv); | |
158ae64f VS |
1920 | |
1921 | /* if sprites aren't enabled, sprites get nothing */ | |
240264f4 | 1922 | if (is_sprite && !config->sprites_enabled) |
158ae64f VS |
1923 | return 0; |
1924 | ||
1925 | /* HSW allows LP1+ watermarks even with multiple pipes */ | |
240264f4 | 1926 | if (level == 0 || config->num_pipes_active > 1) { |
175fded1 | 1927 | fifo_size /= INTEL_INFO(dev_priv)->num_pipes; |
158ae64f VS |
1928 | |
1929 | /* | |
1930 | * For some reason the non self refresh | |
1931 | * FIFO size is only half of the self | |
1932 | * refresh FIFO size on ILK/SNB. | |
1933 | */ | |
175fded1 | 1934 | if (INTEL_GEN(dev_priv) <= 6) |
158ae64f VS |
1935 | fifo_size /= 2; |
1936 | } | |
1937 | ||
240264f4 | 1938 | if (config->sprites_enabled) { |
158ae64f VS |
1939 | /* level 0 is always calculated with 1:1 split */ |
1940 | if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) { | |
1941 | if (is_sprite) | |
1942 | fifo_size *= 5; | |
1943 | fifo_size /= 6; | |
1944 | } else { | |
1945 | fifo_size /= 2; | |
1946 | } | |
1947 | } | |
1948 | ||
1949 | /* clamp to max that the registers can hold */ | |
175fded1 | 1950 | return min(fifo_size, ilk_plane_wm_reg_max(dev_priv, level, is_sprite)); |
158ae64f VS |
1951 | } |
1952 | ||
1953 | /* Calculate the maximum cursor plane watermark */ | |
1954 | static unsigned int ilk_cursor_wm_max(const struct drm_device *dev, | |
240264f4 VS |
1955 | int level, |
1956 | const struct intel_wm_config *config) | |
158ae64f VS |
1957 | { |
1958 | /* HSW LP1+ watermarks w/ multiple pipes */ | |
240264f4 | 1959 | if (level > 0 && config->num_pipes_active > 1) |
158ae64f VS |
1960 | return 64; |
1961 | ||
1962 | /* otherwise just report max that registers can hold */ | |
175fded1 | 1963 | return ilk_cursor_wm_reg_max(to_i915(dev), level); |
158ae64f VS |
1964 | } |
1965 | ||
d34ff9c6 | 1966 | static void ilk_compute_wm_maximums(const struct drm_device *dev, |
34982fe1 VS |
1967 | int level, |
1968 | const struct intel_wm_config *config, | |
1969 | enum intel_ddb_partitioning ddb_partitioning, | |
820c1980 | 1970 | struct ilk_wm_maximums *max) |
158ae64f | 1971 | { |
240264f4 VS |
1972 | max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false); |
1973 | max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true); | |
1974 | max->cur = ilk_cursor_wm_max(dev, level, config); | |
175fded1 | 1975 | max->fbc = ilk_fbc_wm_reg_max(to_i915(dev)); |
158ae64f VS |
1976 | } |
1977 | ||
175fded1 | 1978 | static void ilk_compute_wm_reg_maximums(const struct drm_i915_private *dev_priv, |
a3cb4048 VS |
1979 | int level, |
1980 | struct ilk_wm_maximums *max) | |
1981 | { | |
175fded1 TU |
1982 | max->pri = ilk_plane_wm_reg_max(dev_priv, level, false); |
1983 | max->spr = ilk_plane_wm_reg_max(dev_priv, level, true); | |
1984 | max->cur = ilk_cursor_wm_reg_max(dev_priv, level); | |
1985 | max->fbc = ilk_fbc_wm_reg_max(dev_priv); | |
a3cb4048 VS |
1986 | } |
1987 | ||
d9395655 | 1988 | static bool ilk_validate_wm_level(int level, |
820c1980 | 1989 | const struct ilk_wm_maximums *max, |
d9395655 | 1990 | struct intel_wm_level *result) |
a9786a11 VS |
1991 | { |
1992 | bool ret; | |
1993 | ||
1994 | /* already determined to be invalid? */ | |
1995 | if (!result->enable) | |
1996 | return false; | |
1997 | ||
1998 | result->enable = result->pri_val <= max->pri && | |
1999 | result->spr_val <= max->spr && | |
2000 | result->cur_val <= max->cur; | |
2001 | ||
2002 | ret = result->enable; | |
2003 | ||
2004 | /* | |
2005 | * HACK until we can pre-compute everything, | |
2006 | * and thus fail gracefully if LP0 watermarks | |
2007 | * are exceeded... | |
2008 | */ | |
2009 | if (level == 0 && !result->enable) { | |
2010 | if (result->pri_val > max->pri) | |
2011 | DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n", | |
2012 | level, result->pri_val, max->pri); | |
2013 | if (result->spr_val > max->spr) | |
2014 | DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n", | |
2015 | level, result->spr_val, max->spr); | |
2016 | if (result->cur_val > max->cur) | |
2017 | DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n", | |
2018 | level, result->cur_val, max->cur); | |
2019 | ||
2020 | result->pri_val = min_t(uint32_t, result->pri_val, max->pri); | |
2021 | result->spr_val = min_t(uint32_t, result->spr_val, max->spr); | |
2022 | result->cur_val = min_t(uint32_t, result->cur_val, max->cur); | |
2023 | result->enable = true; | |
2024 | } | |
2025 | ||
a9786a11 VS |
2026 | return ret; |
2027 | } | |
2028 | ||
d34ff9c6 | 2029 | static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv, |
43d59eda | 2030 | const struct intel_crtc *intel_crtc, |
6f5ddd17 | 2031 | int level, |
7221fc33 | 2032 | struct intel_crtc_state *cstate, |
86c8bbbe MR |
2033 | struct intel_plane_state *pristate, |
2034 | struct intel_plane_state *sprstate, | |
2035 | struct intel_plane_state *curstate, | |
1fd527cc | 2036 | struct intel_wm_level *result) |
6f5ddd17 VS |
2037 | { |
2038 | uint16_t pri_latency = dev_priv->wm.pri_latency[level]; | |
2039 | uint16_t spr_latency = dev_priv->wm.spr_latency[level]; | |
2040 | uint16_t cur_latency = dev_priv->wm.cur_latency[level]; | |
2041 | ||
2042 | /* WM1+ latency values stored in 0.5us units */ | |
2043 | if (level > 0) { | |
2044 | pri_latency *= 5; | |
2045 | spr_latency *= 5; | |
2046 | cur_latency *= 5; | |
2047 | } | |
2048 | ||
e3bddded ML |
2049 | if (pristate) { |
2050 | result->pri_val = ilk_compute_pri_wm(cstate, pristate, | |
2051 | pri_latency, level); | |
2052 | result->fbc_val = ilk_compute_fbc_wm(cstate, pristate, result->pri_val); | |
2053 | } | |
2054 | ||
2055 | if (sprstate) | |
2056 | result->spr_val = ilk_compute_spr_wm(cstate, sprstate, spr_latency); | |
2057 | ||
2058 | if (curstate) | |
2059 | result->cur_val = ilk_compute_cur_wm(cstate, curstate, cur_latency); | |
2060 | ||
6f5ddd17 VS |
2061 | result->enable = true; |
2062 | } | |
2063 | ||
801bcfff | 2064 | static uint32_t |
532f7a7f | 2065 | hsw_compute_linetime_wm(const struct intel_crtc_state *cstate) |
1f8eeabf | 2066 | { |
532f7a7f VS |
2067 | const struct intel_atomic_state *intel_state = |
2068 | to_intel_atomic_state(cstate->base.state); | |
ee91a159 MR |
2069 | const struct drm_display_mode *adjusted_mode = |
2070 | &cstate->base.adjusted_mode; | |
85a02deb | 2071 | u32 linetime, ips_linetime; |
1f8eeabf | 2072 | |
ee91a159 MR |
2073 | if (!cstate->base.active) |
2074 | return 0; | |
2075 | if (WARN_ON(adjusted_mode->crtc_clock == 0)) | |
2076 | return 0; | |
532f7a7f | 2077 | if (WARN_ON(intel_state->cdclk == 0)) |
801bcfff | 2078 | return 0; |
1011d8c4 | 2079 | |
1f8eeabf ED |
2080 | /* The WM are computed with base on how long it takes to fill a single |
2081 | * row at the given clock rate, multiplied by 8. | |
2082 | * */ | |
124abe07 VS |
2083 | linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8, |
2084 | adjusted_mode->crtc_clock); | |
2085 | ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8, | |
532f7a7f | 2086 | intel_state->cdclk); |
1f8eeabf | 2087 | |
801bcfff PZ |
2088 | return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) | |
2089 | PIPE_WM_LINETIME_TIME(linetime); | |
1f8eeabf ED |
2090 | } |
2091 | ||
bb726519 VS |
2092 | static void intel_read_wm_latency(struct drm_i915_private *dev_priv, |
2093 | uint16_t wm[8]) | |
12b134df | 2094 | { |
5db94019 | 2095 | if (IS_GEN9(dev_priv)) { |
2af30a5c | 2096 | uint32_t val; |
4f947386 | 2097 | int ret, i; |
5db94019 | 2098 | int level, max_level = ilk_wm_max_level(dev_priv); |
2af30a5c PB |
2099 | |
2100 | /* read the first set of memory latencies[0:3] */ | |
2101 | val = 0; /* data0 to be programmed to 0 for first set */ | |
2102 | mutex_lock(&dev_priv->rps.hw_lock); | |
2103 | ret = sandybridge_pcode_read(dev_priv, | |
2104 | GEN9_PCODE_READ_MEM_LATENCY, | |
2105 | &val); | |
2106 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2107 | ||
2108 | if (ret) { | |
2109 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); | |
2110 | return; | |
2111 | } | |
2112 | ||
2113 | wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK; | |
2114 | wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & | |
2115 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2116 | wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & | |
2117 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2118 | wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & | |
2119 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2120 | ||
2121 | /* read the second set of memory latencies[4:7] */ | |
2122 | val = 1; /* data0 to be programmed to 1 for second set */ | |
2123 | mutex_lock(&dev_priv->rps.hw_lock); | |
2124 | ret = sandybridge_pcode_read(dev_priv, | |
2125 | GEN9_PCODE_READ_MEM_LATENCY, | |
2126 | &val); | |
2127 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2128 | if (ret) { | |
2129 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); | |
2130 | return; | |
2131 | } | |
2132 | ||
2133 | wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK; | |
2134 | wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & | |
2135 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2136 | wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & | |
2137 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2138 | wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & | |
2139 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2140 | ||
0727e40a PZ |
2141 | /* |
2142 | * If a level n (n > 1) has a 0us latency, all levels m (m >= n) | |
2143 | * need to be disabled. We make sure to sanitize the values out | |
2144 | * of the punit to satisfy this requirement. | |
2145 | */ | |
2146 | for (level = 1; level <= max_level; level++) { | |
2147 | if (wm[level] == 0) { | |
2148 | for (i = level + 1; i <= max_level; i++) | |
2149 | wm[i] = 0; | |
2150 | break; | |
2151 | } | |
2152 | } | |
2153 | ||
367294be | 2154 | /* |
6f97235b DL |
2155 | * WaWmMemoryReadLatency:skl |
2156 | * | |
367294be | 2157 | * punit doesn't take into account the read latency so we need |
0727e40a PZ |
2158 | * to add 2us to the various latency levels we retrieve from the |
2159 | * punit when level 0 response data us 0us. | |
367294be | 2160 | */ |
0727e40a PZ |
2161 | if (wm[0] == 0) { |
2162 | wm[0] += 2; | |
2163 | for (level = 1; level <= max_level; level++) { | |
2164 | if (wm[level] == 0) | |
2165 | break; | |
367294be | 2166 | wm[level] += 2; |
4f947386 | 2167 | } |
0727e40a PZ |
2168 | } |
2169 | ||
8652744b | 2170 | } else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) { |
12b134df VS |
2171 | uint64_t sskpd = I915_READ64(MCH_SSKPD); |
2172 | ||
2173 | wm[0] = (sskpd >> 56) & 0xFF; | |
2174 | if (wm[0] == 0) | |
2175 | wm[0] = sskpd & 0xF; | |
e5d5019e VS |
2176 | wm[1] = (sskpd >> 4) & 0xFF; |
2177 | wm[2] = (sskpd >> 12) & 0xFF; | |
2178 | wm[3] = (sskpd >> 20) & 0x1FF; | |
2179 | wm[4] = (sskpd >> 32) & 0x1FF; | |
bb726519 | 2180 | } else if (INTEL_GEN(dev_priv) >= 6) { |
63cf9a13 VS |
2181 | uint32_t sskpd = I915_READ(MCH_SSKPD); |
2182 | ||
2183 | wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK; | |
2184 | wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK; | |
2185 | wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK; | |
2186 | wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK; | |
bb726519 | 2187 | } else if (INTEL_GEN(dev_priv) >= 5) { |
3a88d0ac VS |
2188 | uint32_t mltr = I915_READ(MLTR_ILK); |
2189 | ||
2190 | /* ILK primary LP0 latency is 700 ns */ | |
2191 | wm[0] = 7; | |
2192 | wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK; | |
2193 | wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK; | |
12b134df VS |
2194 | } |
2195 | } | |
2196 | ||
5db94019 TU |
2197 | static void intel_fixup_spr_wm_latency(struct drm_i915_private *dev_priv, |
2198 | uint16_t wm[5]) | |
53615a5e VS |
2199 | { |
2200 | /* ILK sprite LP0 latency is 1300 ns */ | |
5db94019 | 2201 | if (IS_GEN5(dev_priv)) |
53615a5e VS |
2202 | wm[0] = 13; |
2203 | } | |
2204 | ||
fd6b8f43 TU |
2205 | static void intel_fixup_cur_wm_latency(struct drm_i915_private *dev_priv, |
2206 | uint16_t wm[5]) | |
53615a5e VS |
2207 | { |
2208 | /* ILK cursor LP0 latency is 1300 ns */ | |
fd6b8f43 | 2209 | if (IS_GEN5(dev_priv)) |
53615a5e VS |
2210 | wm[0] = 13; |
2211 | ||
2212 | /* WaDoubleCursorLP3Latency:ivb */ | |
fd6b8f43 | 2213 | if (IS_IVYBRIDGE(dev_priv)) |
53615a5e VS |
2214 | wm[3] *= 2; |
2215 | } | |
2216 | ||
5db94019 | 2217 | int ilk_wm_max_level(const struct drm_i915_private *dev_priv) |
26ec971e | 2218 | { |
26ec971e | 2219 | /* how many WM levels are we expecting */ |
8652744b | 2220 | if (INTEL_GEN(dev_priv) >= 9) |
2af30a5c | 2221 | return 7; |
8652744b | 2222 | else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
ad0d6dc4 | 2223 | return 4; |
8652744b | 2224 | else if (INTEL_GEN(dev_priv) >= 6) |
ad0d6dc4 | 2225 | return 3; |
26ec971e | 2226 | else |
ad0d6dc4 VS |
2227 | return 2; |
2228 | } | |
7526ed79 | 2229 | |
5db94019 | 2230 | static void intel_print_wm_latency(struct drm_i915_private *dev_priv, |
ad0d6dc4 | 2231 | const char *name, |
2af30a5c | 2232 | const uint16_t wm[8]) |
ad0d6dc4 | 2233 | { |
5db94019 | 2234 | int level, max_level = ilk_wm_max_level(dev_priv); |
26ec971e VS |
2235 | |
2236 | for (level = 0; level <= max_level; level++) { | |
2237 | unsigned int latency = wm[level]; | |
2238 | ||
2239 | if (latency == 0) { | |
2240 | DRM_ERROR("%s WM%d latency not provided\n", | |
2241 | name, level); | |
2242 | continue; | |
2243 | } | |
2244 | ||
2af30a5c PB |
2245 | /* |
2246 | * - latencies are in us on gen9. | |
2247 | * - before then, WM1+ latency values are in 0.5us units | |
2248 | */ | |
5db94019 | 2249 | if (IS_GEN9(dev_priv)) |
2af30a5c PB |
2250 | latency *= 10; |
2251 | else if (level > 0) | |
26ec971e VS |
2252 | latency *= 5; |
2253 | ||
2254 | DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n", | |
2255 | name, level, wm[level], | |
2256 | latency / 10, latency % 10); | |
2257 | } | |
2258 | } | |
2259 | ||
e95a2f75 VS |
2260 | static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv, |
2261 | uint16_t wm[5], uint16_t min) | |
2262 | { | |
5db94019 | 2263 | int level, max_level = ilk_wm_max_level(dev_priv); |
e95a2f75 VS |
2264 | |
2265 | if (wm[0] >= min) | |
2266 | return false; | |
2267 | ||
2268 | wm[0] = max(wm[0], min); | |
2269 | for (level = 1; level <= max_level; level++) | |
2270 | wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5)); | |
2271 | ||
2272 | return true; | |
2273 | } | |
2274 | ||
bb726519 | 2275 | static void snb_wm_latency_quirk(struct drm_i915_private *dev_priv) |
e95a2f75 | 2276 | { |
e95a2f75 VS |
2277 | bool changed; |
2278 | ||
2279 | /* | |
2280 | * The BIOS provided WM memory latency values are often | |
2281 | * inadequate for high resolution displays. Adjust them. | |
2282 | */ | |
2283 | changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) | | |
2284 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) | | |
2285 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12); | |
2286 | ||
2287 | if (!changed) | |
2288 | return; | |
2289 | ||
2290 | DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n"); | |
5db94019 TU |
2291 | intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency); |
2292 | intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency); | |
2293 | intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency); | |
e95a2f75 VS |
2294 | } |
2295 | ||
bb726519 | 2296 | static void ilk_setup_wm_latency(struct drm_i915_private *dev_priv) |
53615a5e | 2297 | { |
bb726519 | 2298 | intel_read_wm_latency(dev_priv, dev_priv->wm.pri_latency); |
53615a5e VS |
2299 | |
2300 | memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency, | |
2301 | sizeof(dev_priv->wm.pri_latency)); | |
2302 | memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency, | |
2303 | sizeof(dev_priv->wm.pri_latency)); | |
2304 | ||
5db94019 | 2305 | intel_fixup_spr_wm_latency(dev_priv, dev_priv->wm.spr_latency); |
fd6b8f43 | 2306 | intel_fixup_cur_wm_latency(dev_priv, dev_priv->wm.cur_latency); |
26ec971e | 2307 | |
5db94019 TU |
2308 | intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency); |
2309 | intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency); | |
2310 | intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency); | |
e95a2f75 | 2311 | |
5db94019 | 2312 | if (IS_GEN6(dev_priv)) |
bb726519 | 2313 | snb_wm_latency_quirk(dev_priv); |
53615a5e VS |
2314 | } |
2315 | ||
bb726519 | 2316 | static void skl_setup_wm_latency(struct drm_i915_private *dev_priv) |
2af30a5c | 2317 | { |
bb726519 | 2318 | intel_read_wm_latency(dev_priv, dev_priv->wm.skl_latency); |
5db94019 | 2319 | intel_print_wm_latency(dev_priv, "Gen9 Plane", dev_priv->wm.skl_latency); |
2af30a5c PB |
2320 | } |
2321 | ||
ed4a6a7c MR |
2322 | static bool ilk_validate_pipe_wm(struct drm_device *dev, |
2323 | struct intel_pipe_wm *pipe_wm) | |
2324 | { | |
2325 | /* LP0 watermark maximums depend on this pipe alone */ | |
2326 | const struct intel_wm_config config = { | |
2327 | .num_pipes_active = 1, | |
2328 | .sprites_enabled = pipe_wm->sprites_enabled, | |
2329 | .sprites_scaled = pipe_wm->sprites_scaled, | |
2330 | }; | |
2331 | struct ilk_wm_maximums max; | |
2332 | ||
2333 | /* LP0 watermarks always use 1/2 DDB partitioning */ | |
2334 | ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max); | |
2335 | ||
2336 | /* At least LP0 must be valid */ | |
2337 | if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) { | |
2338 | DRM_DEBUG_KMS("LP0 watermark invalid\n"); | |
2339 | return false; | |
2340 | } | |
2341 | ||
2342 | return true; | |
2343 | } | |
2344 | ||
0b2ae6d7 | 2345 | /* Compute new watermarks for the pipe */ |
e3bddded | 2346 | static int ilk_compute_pipe_wm(struct intel_crtc_state *cstate) |
0b2ae6d7 | 2347 | { |
e3bddded ML |
2348 | struct drm_atomic_state *state = cstate->base.state; |
2349 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); | |
86c8bbbe | 2350 | struct intel_pipe_wm *pipe_wm; |
e3bddded | 2351 | struct drm_device *dev = state->dev; |
fac5e23e | 2352 | const struct drm_i915_private *dev_priv = to_i915(dev); |
43d59eda | 2353 | struct intel_plane *intel_plane; |
86c8bbbe | 2354 | struct intel_plane_state *pristate = NULL; |
43d59eda | 2355 | struct intel_plane_state *sprstate = NULL; |
86c8bbbe | 2356 | struct intel_plane_state *curstate = NULL; |
5db94019 | 2357 | int level, max_level = ilk_wm_max_level(dev_priv), usable_level; |
820c1980 | 2358 | struct ilk_wm_maximums max; |
0b2ae6d7 | 2359 | |
e8f1f02e | 2360 | pipe_wm = &cstate->wm.ilk.optimal; |
86c8bbbe | 2361 | |
43d59eda | 2362 | for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) { |
e3bddded ML |
2363 | struct intel_plane_state *ps; |
2364 | ||
2365 | ps = intel_atomic_get_existing_plane_state(state, | |
2366 | intel_plane); | |
2367 | if (!ps) | |
2368 | continue; | |
86c8bbbe MR |
2369 | |
2370 | if (intel_plane->base.type == DRM_PLANE_TYPE_PRIMARY) | |
e3bddded | 2371 | pristate = ps; |
86c8bbbe | 2372 | else if (intel_plane->base.type == DRM_PLANE_TYPE_OVERLAY) |
e3bddded | 2373 | sprstate = ps; |
86c8bbbe | 2374 | else if (intel_plane->base.type == DRM_PLANE_TYPE_CURSOR) |
e3bddded | 2375 | curstate = ps; |
43d59eda MR |
2376 | } |
2377 | ||
ed4a6a7c | 2378 | pipe_wm->pipe_enabled = cstate->base.active; |
e3bddded | 2379 | if (sprstate) { |
936e71e3 VS |
2380 | pipe_wm->sprites_enabled = sprstate->base.visible; |
2381 | pipe_wm->sprites_scaled = sprstate->base.visible && | |
2382 | (drm_rect_width(&sprstate->base.dst) != drm_rect_width(&sprstate->base.src) >> 16 || | |
2383 | drm_rect_height(&sprstate->base.dst) != drm_rect_height(&sprstate->base.src) >> 16); | |
e3bddded ML |
2384 | } |
2385 | ||
d81f04c5 ML |
2386 | usable_level = max_level; |
2387 | ||
7b39a0b7 | 2388 | /* ILK/SNB: LP2+ watermarks only w/o sprites */ |
175fded1 | 2389 | if (INTEL_GEN(dev_priv) <= 6 && pipe_wm->sprites_enabled) |
d81f04c5 | 2390 | usable_level = 1; |
7b39a0b7 VS |
2391 | |
2392 | /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */ | |
ed4a6a7c | 2393 | if (pipe_wm->sprites_scaled) |
d81f04c5 | 2394 | usable_level = 0; |
7b39a0b7 | 2395 | |
86c8bbbe | 2396 | ilk_compute_wm_level(dev_priv, intel_crtc, 0, cstate, |
71f0a626 ML |
2397 | pristate, sprstate, curstate, &pipe_wm->raw_wm[0]); |
2398 | ||
2399 | memset(&pipe_wm->wm, 0, sizeof(pipe_wm->wm)); | |
2400 | pipe_wm->wm[0] = pipe_wm->raw_wm[0]; | |
0b2ae6d7 | 2401 | |
8652744b | 2402 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
532f7a7f | 2403 | pipe_wm->linetime = hsw_compute_linetime_wm(cstate); |
0b2ae6d7 | 2404 | |
ed4a6a7c | 2405 | if (!ilk_validate_pipe_wm(dev, pipe_wm)) |
1a426d61 | 2406 | return -EINVAL; |
a3cb4048 | 2407 | |
175fded1 | 2408 | ilk_compute_wm_reg_maximums(dev_priv, 1, &max); |
a3cb4048 VS |
2409 | |
2410 | for (level = 1; level <= max_level; level++) { | |
71f0a626 | 2411 | struct intel_wm_level *wm = &pipe_wm->raw_wm[level]; |
a3cb4048 | 2412 | |
86c8bbbe | 2413 | ilk_compute_wm_level(dev_priv, intel_crtc, level, cstate, |
d81f04c5 | 2414 | pristate, sprstate, curstate, wm); |
a3cb4048 VS |
2415 | |
2416 | /* | |
2417 | * Disable any watermark level that exceeds the | |
2418 | * register maximums since such watermarks are | |
2419 | * always invalid. | |
2420 | */ | |
71f0a626 ML |
2421 | if (level > usable_level) |
2422 | continue; | |
2423 | ||
2424 | if (ilk_validate_wm_level(level, &max, wm)) | |
2425 | pipe_wm->wm[level] = *wm; | |
2426 | else | |
d81f04c5 | 2427 | usable_level = level; |
a3cb4048 VS |
2428 | } |
2429 | ||
86c8bbbe | 2430 | return 0; |
0b2ae6d7 VS |
2431 | } |
2432 | ||
ed4a6a7c MR |
2433 | /* |
2434 | * Build a set of 'intermediate' watermark values that satisfy both the old | |
2435 | * state and the new state. These can be programmed to the hardware | |
2436 | * immediately. | |
2437 | */ | |
2438 | static int ilk_compute_intermediate_wm(struct drm_device *dev, | |
2439 | struct intel_crtc *intel_crtc, | |
2440 | struct intel_crtc_state *newstate) | |
2441 | { | |
e8f1f02e | 2442 | struct intel_pipe_wm *a = &newstate->wm.ilk.intermediate; |
ed4a6a7c | 2443 | struct intel_pipe_wm *b = &intel_crtc->wm.active.ilk; |
5db94019 | 2444 | int level, max_level = ilk_wm_max_level(to_i915(dev)); |
ed4a6a7c MR |
2445 | |
2446 | /* | |
2447 | * Start with the final, target watermarks, then combine with the | |
2448 | * currently active watermarks to get values that are safe both before | |
2449 | * and after the vblank. | |
2450 | */ | |
e8f1f02e | 2451 | *a = newstate->wm.ilk.optimal; |
ed4a6a7c MR |
2452 | a->pipe_enabled |= b->pipe_enabled; |
2453 | a->sprites_enabled |= b->sprites_enabled; | |
2454 | a->sprites_scaled |= b->sprites_scaled; | |
2455 | ||
2456 | for (level = 0; level <= max_level; level++) { | |
2457 | struct intel_wm_level *a_wm = &a->wm[level]; | |
2458 | const struct intel_wm_level *b_wm = &b->wm[level]; | |
2459 | ||
2460 | a_wm->enable &= b_wm->enable; | |
2461 | a_wm->pri_val = max(a_wm->pri_val, b_wm->pri_val); | |
2462 | a_wm->spr_val = max(a_wm->spr_val, b_wm->spr_val); | |
2463 | a_wm->cur_val = max(a_wm->cur_val, b_wm->cur_val); | |
2464 | a_wm->fbc_val = max(a_wm->fbc_val, b_wm->fbc_val); | |
2465 | } | |
2466 | ||
2467 | /* | |
2468 | * We need to make sure that these merged watermark values are | |
2469 | * actually a valid configuration themselves. If they're not, | |
2470 | * there's no safe way to transition from the old state to | |
2471 | * the new state, so we need to fail the atomic transaction. | |
2472 | */ | |
2473 | if (!ilk_validate_pipe_wm(dev, a)) | |
2474 | return -EINVAL; | |
2475 | ||
2476 | /* | |
2477 | * If our intermediate WM are identical to the final WM, then we can | |
2478 | * omit the post-vblank programming; only update if it's different. | |
2479 | */ | |
e8f1f02e | 2480 | if (memcmp(a, &newstate->wm.ilk.optimal, sizeof(*a)) == 0) |
ed4a6a7c MR |
2481 | newstate->wm.need_postvbl_update = false; |
2482 | ||
2483 | return 0; | |
2484 | } | |
2485 | ||
0b2ae6d7 VS |
2486 | /* |
2487 | * Merge the watermarks from all active pipes for a specific level. | |
2488 | */ | |
2489 | static void ilk_merge_wm_level(struct drm_device *dev, | |
2490 | int level, | |
2491 | struct intel_wm_level *ret_wm) | |
2492 | { | |
2493 | const struct intel_crtc *intel_crtc; | |
2494 | ||
d52fea5b VS |
2495 | ret_wm->enable = true; |
2496 | ||
d3fcc808 | 2497 | for_each_intel_crtc(dev, intel_crtc) { |
ed4a6a7c | 2498 | const struct intel_pipe_wm *active = &intel_crtc->wm.active.ilk; |
fe392efd VS |
2499 | const struct intel_wm_level *wm = &active->wm[level]; |
2500 | ||
2501 | if (!active->pipe_enabled) | |
2502 | continue; | |
0b2ae6d7 | 2503 | |
d52fea5b VS |
2504 | /* |
2505 | * The watermark values may have been used in the past, | |
2506 | * so we must maintain them in the registers for some | |
2507 | * time even if the level is now disabled. | |
2508 | */ | |
0b2ae6d7 | 2509 | if (!wm->enable) |
d52fea5b | 2510 | ret_wm->enable = false; |
0b2ae6d7 VS |
2511 | |
2512 | ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val); | |
2513 | ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val); | |
2514 | ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val); | |
2515 | ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val); | |
2516 | } | |
0b2ae6d7 VS |
2517 | } |
2518 | ||
2519 | /* | |
2520 | * Merge all low power watermarks for all active pipes. | |
2521 | */ | |
2522 | static void ilk_wm_merge(struct drm_device *dev, | |
0ba22e26 | 2523 | const struct intel_wm_config *config, |
820c1980 | 2524 | const struct ilk_wm_maximums *max, |
0b2ae6d7 VS |
2525 | struct intel_pipe_wm *merged) |
2526 | { | |
fac5e23e | 2527 | struct drm_i915_private *dev_priv = to_i915(dev); |
5db94019 | 2528 | int level, max_level = ilk_wm_max_level(dev_priv); |
d52fea5b | 2529 | int last_enabled_level = max_level; |
0b2ae6d7 | 2530 | |
0ba22e26 | 2531 | /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */ |
fd6b8f43 | 2532 | if ((INTEL_GEN(dev_priv) <= 6 || IS_IVYBRIDGE(dev_priv)) && |
0ba22e26 | 2533 | config->num_pipes_active > 1) |
1204d5ba | 2534 | last_enabled_level = 0; |
0ba22e26 | 2535 | |
6c8b6c28 | 2536 | /* ILK: FBC WM must be disabled always */ |
175fded1 | 2537 | merged->fbc_wm_enabled = INTEL_GEN(dev_priv) >= 6; |
0b2ae6d7 VS |
2538 | |
2539 | /* merge each WM1+ level */ | |
2540 | for (level = 1; level <= max_level; level++) { | |
2541 | struct intel_wm_level *wm = &merged->wm[level]; | |
2542 | ||
2543 | ilk_merge_wm_level(dev, level, wm); | |
2544 | ||
d52fea5b VS |
2545 | if (level > last_enabled_level) |
2546 | wm->enable = false; | |
2547 | else if (!ilk_validate_wm_level(level, max, wm)) | |
2548 | /* make sure all following levels get disabled */ | |
2549 | last_enabled_level = level - 1; | |
0b2ae6d7 VS |
2550 | |
2551 | /* | |
2552 | * The spec says it is preferred to disable | |
2553 | * FBC WMs instead of disabling a WM level. | |
2554 | */ | |
2555 | if (wm->fbc_val > max->fbc) { | |
d52fea5b VS |
2556 | if (wm->enable) |
2557 | merged->fbc_wm_enabled = false; | |
0b2ae6d7 VS |
2558 | wm->fbc_val = 0; |
2559 | } | |
2560 | } | |
6c8b6c28 VS |
2561 | |
2562 | /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */ | |
2563 | /* | |
2564 | * FIXME this is racy. FBC might get enabled later. | |
2565 | * What we should check here is whether FBC can be | |
2566 | * enabled sometime later. | |
2567 | */ | |
5db94019 | 2568 | if (IS_GEN5(dev_priv) && !merged->fbc_wm_enabled && |
0e631adc | 2569 | intel_fbc_is_active(dev_priv)) { |
6c8b6c28 VS |
2570 | for (level = 2; level <= max_level; level++) { |
2571 | struct intel_wm_level *wm = &merged->wm[level]; | |
2572 | ||
2573 | wm->enable = false; | |
2574 | } | |
2575 | } | |
0b2ae6d7 VS |
2576 | } |
2577 | ||
b380ca3c VS |
2578 | static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm) |
2579 | { | |
2580 | /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */ | |
2581 | return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable); | |
2582 | } | |
2583 | ||
a68d68ee VS |
2584 | /* The value we need to program into the WM_LPx latency field */ |
2585 | static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level) | |
2586 | { | |
fac5e23e | 2587 | struct drm_i915_private *dev_priv = to_i915(dev); |
a68d68ee | 2588 | |
8652744b | 2589 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
a68d68ee VS |
2590 | return 2 * level; |
2591 | else | |
2592 | return dev_priv->wm.pri_latency[level]; | |
2593 | } | |
2594 | ||
820c1980 | 2595 | static void ilk_compute_wm_results(struct drm_device *dev, |
0362c781 | 2596 | const struct intel_pipe_wm *merged, |
609cedef | 2597 | enum intel_ddb_partitioning partitioning, |
820c1980 | 2598 | struct ilk_wm_values *results) |
801bcfff | 2599 | { |
175fded1 | 2600 | struct drm_i915_private *dev_priv = to_i915(dev); |
0b2ae6d7 VS |
2601 | struct intel_crtc *intel_crtc; |
2602 | int level, wm_lp; | |
cca32e9a | 2603 | |
0362c781 | 2604 | results->enable_fbc_wm = merged->fbc_wm_enabled; |
609cedef | 2605 | results->partitioning = partitioning; |
cca32e9a | 2606 | |
0b2ae6d7 | 2607 | /* LP1+ register values */ |
cca32e9a | 2608 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
1fd527cc | 2609 | const struct intel_wm_level *r; |
801bcfff | 2610 | |
b380ca3c | 2611 | level = ilk_wm_lp_to_level(wm_lp, merged); |
0b2ae6d7 | 2612 | |
0362c781 | 2613 | r = &merged->wm[level]; |
cca32e9a | 2614 | |
d52fea5b VS |
2615 | /* |
2616 | * Maintain the watermark values even if the level is | |
2617 | * disabled. Doing otherwise could cause underruns. | |
2618 | */ | |
2619 | results->wm_lp[wm_lp - 1] = | |
a68d68ee | 2620 | (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) | |
416f4727 VS |
2621 | (r->pri_val << WM1_LP_SR_SHIFT) | |
2622 | r->cur_val; | |
2623 | ||
d52fea5b VS |
2624 | if (r->enable) |
2625 | results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN; | |
2626 | ||
175fded1 | 2627 | if (INTEL_GEN(dev_priv) >= 8) |
416f4727 VS |
2628 | results->wm_lp[wm_lp - 1] |= |
2629 | r->fbc_val << WM1_LP_FBC_SHIFT_BDW; | |
2630 | else | |
2631 | results->wm_lp[wm_lp - 1] |= | |
2632 | r->fbc_val << WM1_LP_FBC_SHIFT; | |
2633 | ||
d52fea5b VS |
2634 | /* |
2635 | * Always set WM1S_LP_EN when spr_val != 0, even if the | |
2636 | * level is disabled. Doing otherwise could cause underruns. | |
2637 | */ | |
175fded1 | 2638 | if (INTEL_GEN(dev_priv) <= 6 && r->spr_val) { |
6cef2b8a VS |
2639 | WARN_ON(wm_lp != 1); |
2640 | results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val; | |
2641 | } else | |
2642 | results->wm_lp_spr[wm_lp - 1] = r->spr_val; | |
cca32e9a | 2643 | } |
801bcfff | 2644 | |
0b2ae6d7 | 2645 | /* LP0 register values */ |
d3fcc808 | 2646 | for_each_intel_crtc(dev, intel_crtc) { |
0b2ae6d7 | 2647 | enum pipe pipe = intel_crtc->pipe; |
ed4a6a7c MR |
2648 | const struct intel_wm_level *r = |
2649 | &intel_crtc->wm.active.ilk.wm[0]; | |
0b2ae6d7 VS |
2650 | |
2651 | if (WARN_ON(!r->enable)) | |
2652 | continue; | |
2653 | ||
ed4a6a7c | 2654 | results->wm_linetime[pipe] = intel_crtc->wm.active.ilk.linetime; |
1011d8c4 | 2655 | |
0b2ae6d7 VS |
2656 | results->wm_pipe[pipe] = |
2657 | (r->pri_val << WM0_PIPE_PLANE_SHIFT) | | |
2658 | (r->spr_val << WM0_PIPE_SPRITE_SHIFT) | | |
2659 | r->cur_val; | |
801bcfff PZ |
2660 | } |
2661 | } | |
2662 | ||
861f3389 PZ |
2663 | /* Find the result with the highest level enabled. Check for enable_fbc_wm in |
2664 | * case both are at the same level. Prefer r1 in case they're the same. */ | |
820c1980 | 2665 | static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev, |
198a1e9b VS |
2666 | struct intel_pipe_wm *r1, |
2667 | struct intel_pipe_wm *r2) | |
861f3389 | 2668 | { |
5db94019 | 2669 | int level, max_level = ilk_wm_max_level(to_i915(dev)); |
198a1e9b | 2670 | int level1 = 0, level2 = 0; |
861f3389 | 2671 | |
198a1e9b VS |
2672 | for (level = 1; level <= max_level; level++) { |
2673 | if (r1->wm[level].enable) | |
2674 | level1 = level; | |
2675 | if (r2->wm[level].enable) | |
2676 | level2 = level; | |
861f3389 PZ |
2677 | } |
2678 | ||
198a1e9b VS |
2679 | if (level1 == level2) { |
2680 | if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled) | |
861f3389 PZ |
2681 | return r2; |
2682 | else | |
2683 | return r1; | |
198a1e9b | 2684 | } else if (level1 > level2) { |
861f3389 PZ |
2685 | return r1; |
2686 | } else { | |
2687 | return r2; | |
2688 | } | |
2689 | } | |
2690 | ||
49a687c4 VS |
2691 | /* dirty bits used to track which watermarks need changes */ |
2692 | #define WM_DIRTY_PIPE(pipe) (1 << (pipe)) | |
2693 | #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe))) | |
2694 | #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp))) | |
2695 | #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3)) | |
2696 | #define WM_DIRTY_FBC (1 << 24) | |
2697 | #define WM_DIRTY_DDB (1 << 25) | |
2698 | ||
055e393f | 2699 | static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv, |
820c1980 ID |
2700 | const struct ilk_wm_values *old, |
2701 | const struct ilk_wm_values *new) | |
49a687c4 VS |
2702 | { |
2703 | unsigned int dirty = 0; | |
2704 | enum pipe pipe; | |
2705 | int wm_lp; | |
2706 | ||
055e393f | 2707 | for_each_pipe(dev_priv, pipe) { |
49a687c4 VS |
2708 | if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) { |
2709 | dirty |= WM_DIRTY_LINETIME(pipe); | |
2710 | /* Must disable LP1+ watermarks too */ | |
2711 | dirty |= WM_DIRTY_LP_ALL; | |
2712 | } | |
2713 | ||
2714 | if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) { | |
2715 | dirty |= WM_DIRTY_PIPE(pipe); | |
2716 | /* Must disable LP1+ watermarks too */ | |
2717 | dirty |= WM_DIRTY_LP_ALL; | |
2718 | } | |
2719 | } | |
2720 | ||
2721 | if (old->enable_fbc_wm != new->enable_fbc_wm) { | |
2722 | dirty |= WM_DIRTY_FBC; | |
2723 | /* Must disable LP1+ watermarks too */ | |
2724 | dirty |= WM_DIRTY_LP_ALL; | |
2725 | } | |
2726 | ||
2727 | if (old->partitioning != new->partitioning) { | |
2728 | dirty |= WM_DIRTY_DDB; | |
2729 | /* Must disable LP1+ watermarks too */ | |
2730 | dirty |= WM_DIRTY_LP_ALL; | |
2731 | } | |
2732 | ||
2733 | /* LP1+ watermarks already deemed dirty, no need to continue */ | |
2734 | if (dirty & WM_DIRTY_LP_ALL) | |
2735 | return dirty; | |
2736 | ||
2737 | /* Find the lowest numbered LP1+ watermark in need of an update... */ | |
2738 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { | |
2739 | if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] || | |
2740 | old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1]) | |
2741 | break; | |
2742 | } | |
2743 | ||
2744 | /* ...and mark it and all higher numbered LP1+ watermarks as dirty */ | |
2745 | for (; wm_lp <= 3; wm_lp++) | |
2746 | dirty |= WM_DIRTY_LP(wm_lp); | |
2747 | ||
2748 | return dirty; | |
2749 | } | |
2750 | ||
8553c18e VS |
2751 | static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv, |
2752 | unsigned int dirty) | |
801bcfff | 2753 | { |
820c1980 | 2754 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
8553c18e | 2755 | bool changed = false; |
801bcfff | 2756 | |
facd619b VS |
2757 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) { |
2758 | previous->wm_lp[2] &= ~WM1_LP_SR_EN; | |
2759 | I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]); | |
8553c18e | 2760 | changed = true; |
facd619b VS |
2761 | } |
2762 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) { | |
2763 | previous->wm_lp[1] &= ~WM1_LP_SR_EN; | |
2764 | I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]); | |
8553c18e | 2765 | changed = true; |
facd619b VS |
2766 | } |
2767 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) { | |
2768 | previous->wm_lp[0] &= ~WM1_LP_SR_EN; | |
2769 | I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]); | |
8553c18e | 2770 | changed = true; |
facd619b | 2771 | } |
801bcfff | 2772 | |
facd619b VS |
2773 | /* |
2774 | * Don't touch WM1S_LP_EN here. | |
2775 | * Doing so could cause underruns. | |
2776 | */ | |
6cef2b8a | 2777 | |
8553c18e VS |
2778 | return changed; |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * The spec says we shouldn't write when we don't need, because every write | |
2783 | * causes WMs to be re-evaluated, expending some power. | |
2784 | */ | |
820c1980 ID |
2785 | static void ilk_write_wm_values(struct drm_i915_private *dev_priv, |
2786 | struct ilk_wm_values *results) | |
8553c18e | 2787 | { |
820c1980 | 2788 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
8553c18e VS |
2789 | unsigned int dirty; |
2790 | uint32_t val; | |
2791 | ||
055e393f | 2792 | dirty = ilk_compute_wm_dirty(dev_priv, previous, results); |
8553c18e VS |
2793 | if (!dirty) |
2794 | return; | |
2795 | ||
2796 | _ilk_disable_lp_wm(dev_priv, dirty); | |
2797 | ||
49a687c4 | 2798 | if (dirty & WM_DIRTY_PIPE(PIPE_A)) |
801bcfff | 2799 | I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]); |
49a687c4 | 2800 | if (dirty & WM_DIRTY_PIPE(PIPE_B)) |
801bcfff | 2801 | I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]); |
49a687c4 | 2802 | if (dirty & WM_DIRTY_PIPE(PIPE_C)) |
801bcfff PZ |
2803 | I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]); |
2804 | ||
49a687c4 | 2805 | if (dirty & WM_DIRTY_LINETIME(PIPE_A)) |
801bcfff | 2806 | I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]); |
49a687c4 | 2807 | if (dirty & WM_DIRTY_LINETIME(PIPE_B)) |
801bcfff | 2808 | I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]); |
49a687c4 | 2809 | if (dirty & WM_DIRTY_LINETIME(PIPE_C)) |
801bcfff PZ |
2810 | I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]); |
2811 | ||
49a687c4 | 2812 | if (dirty & WM_DIRTY_DDB) { |
8652744b | 2813 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) { |
ac9545fd VS |
2814 | val = I915_READ(WM_MISC); |
2815 | if (results->partitioning == INTEL_DDB_PART_1_2) | |
2816 | val &= ~WM_MISC_DATA_PARTITION_5_6; | |
2817 | else | |
2818 | val |= WM_MISC_DATA_PARTITION_5_6; | |
2819 | I915_WRITE(WM_MISC, val); | |
2820 | } else { | |
2821 | val = I915_READ(DISP_ARB_CTL2); | |
2822 | if (results->partitioning == INTEL_DDB_PART_1_2) | |
2823 | val &= ~DISP_DATA_PARTITION_5_6; | |
2824 | else | |
2825 | val |= DISP_DATA_PARTITION_5_6; | |
2826 | I915_WRITE(DISP_ARB_CTL2, val); | |
2827 | } | |
1011d8c4 PZ |
2828 | } |
2829 | ||
49a687c4 | 2830 | if (dirty & WM_DIRTY_FBC) { |
cca32e9a PZ |
2831 | val = I915_READ(DISP_ARB_CTL); |
2832 | if (results->enable_fbc_wm) | |
2833 | val &= ~DISP_FBC_WM_DIS; | |
2834 | else | |
2835 | val |= DISP_FBC_WM_DIS; | |
2836 | I915_WRITE(DISP_ARB_CTL, val); | |
2837 | } | |
2838 | ||
954911eb ID |
2839 | if (dirty & WM_DIRTY_LP(1) && |
2840 | previous->wm_lp_spr[0] != results->wm_lp_spr[0]) | |
2841 | I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]); | |
2842 | ||
175fded1 | 2843 | if (INTEL_GEN(dev_priv) >= 7) { |
6cef2b8a VS |
2844 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1]) |
2845 | I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]); | |
2846 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2]) | |
2847 | I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]); | |
2848 | } | |
801bcfff | 2849 | |
facd619b | 2850 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0]) |
801bcfff | 2851 | I915_WRITE(WM1_LP_ILK, results->wm_lp[0]); |
facd619b | 2852 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1]) |
801bcfff | 2853 | I915_WRITE(WM2_LP_ILK, results->wm_lp[1]); |
facd619b | 2854 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2]) |
801bcfff | 2855 | I915_WRITE(WM3_LP_ILK, results->wm_lp[2]); |
609cedef VS |
2856 | |
2857 | dev_priv->wm.hw = *results; | |
801bcfff PZ |
2858 | } |
2859 | ||
ed4a6a7c | 2860 | bool ilk_disable_lp_wm(struct drm_device *dev) |
8553c18e | 2861 | { |
fac5e23e | 2862 | struct drm_i915_private *dev_priv = to_i915(dev); |
8553c18e VS |
2863 | |
2864 | return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL); | |
2865 | } | |
2866 | ||
656d1b89 | 2867 | #define SKL_SAGV_BLOCK_TIME 30 /* µs */ |
b9cec075 | 2868 | |
024c9045 MR |
2869 | /* |
2870 | * Return the index of a plane in the SKL DDB and wm result arrays. Primary | |
2871 | * plane is always in slot 0, cursor is always in slot I915_MAX_PLANES-1, and | |
2872 | * other universal planes are in indices 1..n. Note that this may leave unused | |
2873 | * indices between the top "sprite" plane and the cursor. | |
2874 | */ | |
2875 | static int | |
2876 | skl_wm_plane_id(const struct intel_plane *plane) | |
2877 | { | |
2878 | switch (plane->base.type) { | |
2879 | case DRM_PLANE_TYPE_PRIMARY: | |
2880 | return 0; | |
2881 | case DRM_PLANE_TYPE_CURSOR: | |
2882 | return PLANE_CURSOR; | |
2883 | case DRM_PLANE_TYPE_OVERLAY: | |
2884 | return plane->plane + 1; | |
2885 | default: | |
2886 | MISSING_CASE(plane->base.type); | |
2887 | return plane->plane; | |
2888 | } | |
2889 | } | |
2890 | ||
ee3d532f PZ |
2891 | /* |
2892 | * FIXME: We still don't have the proper code detect if we need to apply the WA, | |
2893 | * so assume we'll always need it in order to avoid underruns. | |
2894 | */ | |
2895 | static bool skl_needs_memory_bw_wa(struct intel_atomic_state *state) | |
2896 | { | |
2897 | struct drm_i915_private *dev_priv = to_i915(state->base.dev); | |
2898 | ||
2899 | if (IS_SKYLAKE(dev_priv) || IS_BROXTON(dev_priv) || | |
2900 | IS_KABYLAKE(dev_priv)) | |
2901 | return true; | |
2902 | ||
2903 | return false; | |
2904 | } | |
2905 | ||
56feca91 PZ |
2906 | static bool |
2907 | intel_has_sagv(struct drm_i915_private *dev_priv) | |
2908 | { | |
6e3100ec PZ |
2909 | if (IS_KABYLAKE(dev_priv)) |
2910 | return true; | |
2911 | ||
2912 | if (IS_SKYLAKE(dev_priv) && | |
2913 | dev_priv->sagv_status != I915_SAGV_NOT_CONTROLLED) | |
2914 | return true; | |
2915 | ||
2916 | return false; | |
56feca91 PZ |
2917 | } |
2918 | ||
656d1b89 L |
2919 | /* |
2920 | * SAGV dynamically adjusts the system agent voltage and clock frequencies | |
2921 | * depending on power and performance requirements. The display engine access | |
2922 | * to system memory is blocked during the adjustment time. Because of the | |
2923 | * blocking time, having this enabled can cause full system hangs and/or pipe | |
2924 | * underruns if we don't meet all of the following requirements: | |
2925 | * | |
2926 | * - <= 1 pipe enabled | |
2927 | * - All planes can enable watermarks for latencies >= SAGV engine block time | |
2928 | * - We're not using an interlaced display configuration | |
2929 | */ | |
2930 | int | |
16dcdc4e | 2931 | intel_enable_sagv(struct drm_i915_private *dev_priv) |
656d1b89 L |
2932 | { |
2933 | int ret; | |
2934 | ||
56feca91 PZ |
2935 | if (!intel_has_sagv(dev_priv)) |
2936 | return 0; | |
2937 | ||
2938 | if (dev_priv->sagv_status == I915_SAGV_ENABLED) | |
656d1b89 L |
2939 | return 0; |
2940 | ||
2941 | DRM_DEBUG_KMS("Enabling the SAGV\n"); | |
2942 | mutex_lock(&dev_priv->rps.hw_lock); | |
2943 | ||
2944 | ret = sandybridge_pcode_write(dev_priv, GEN9_PCODE_SAGV_CONTROL, | |
2945 | GEN9_SAGV_ENABLE); | |
2946 | ||
2947 | /* We don't need to wait for the SAGV when enabling */ | |
2948 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2949 | ||
2950 | /* | |
2951 | * Some skl systems, pre-release machines in particular, | |
2952 | * don't actually have an SAGV. | |
2953 | */ | |
6e3100ec | 2954 | if (IS_SKYLAKE(dev_priv) && ret == -ENXIO) { |
656d1b89 | 2955 | DRM_DEBUG_DRIVER("No SAGV found on system, ignoring\n"); |
16dcdc4e | 2956 | dev_priv->sagv_status = I915_SAGV_NOT_CONTROLLED; |
656d1b89 L |
2957 | return 0; |
2958 | } else if (ret < 0) { | |
2959 | DRM_ERROR("Failed to enable the SAGV\n"); | |
2960 | return ret; | |
2961 | } | |
2962 | ||
16dcdc4e | 2963 | dev_priv->sagv_status = I915_SAGV_ENABLED; |
656d1b89 L |
2964 | return 0; |
2965 | } | |
2966 | ||
656d1b89 | 2967 | int |
16dcdc4e | 2968 | intel_disable_sagv(struct drm_i915_private *dev_priv) |
656d1b89 | 2969 | { |
dccf82ad | 2970 | int ret; |
656d1b89 | 2971 | |
56feca91 PZ |
2972 | if (!intel_has_sagv(dev_priv)) |
2973 | return 0; | |
2974 | ||
2975 | if (dev_priv->sagv_status == I915_SAGV_DISABLED) | |
656d1b89 L |
2976 | return 0; |
2977 | ||
2978 | DRM_DEBUG_KMS("Disabling the SAGV\n"); | |
2979 | mutex_lock(&dev_priv->rps.hw_lock); | |
2980 | ||
2981 | /* bspec says to keep retrying for at least 1 ms */ | |
dccf82ad ID |
2982 | ret = skl_pcode_request(dev_priv, GEN9_PCODE_SAGV_CONTROL, |
2983 | GEN9_SAGV_DISABLE, | |
2984 | GEN9_SAGV_IS_DISABLED, GEN9_SAGV_IS_DISABLED, | |
2985 | 1); | |
656d1b89 L |
2986 | mutex_unlock(&dev_priv->rps.hw_lock); |
2987 | ||
656d1b89 L |
2988 | /* |
2989 | * Some skl systems, pre-release machines in particular, | |
2990 | * don't actually have an SAGV. | |
2991 | */ | |
dccf82ad | 2992 | if (IS_SKYLAKE(dev_priv) && ret == -ENXIO) { |
656d1b89 | 2993 | DRM_DEBUG_DRIVER("No SAGV found on system, ignoring\n"); |
16dcdc4e | 2994 | dev_priv->sagv_status = I915_SAGV_NOT_CONTROLLED; |
656d1b89 | 2995 | return 0; |
dccf82ad ID |
2996 | } else if (ret < 0) { |
2997 | DRM_ERROR("Failed to disable the SAGV (%d)\n", ret); | |
2998 | return ret; | |
656d1b89 L |
2999 | } |
3000 | ||
16dcdc4e | 3001 | dev_priv->sagv_status = I915_SAGV_DISABLED; |
656d1b89 L |
3002 | return 0; |
3003 | } | |
3004 | ||
16dcdc4e | 3005 | bool intel_can_enable_sagv(struct drm_atomic_state *state) |
656d1b89 L |
3006 | { |
3007 | struct drm_device *dev = state->dev; | |
3008 | struct drm_i915_private *dev_priv = to_i915(dev); | |
3009 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
ee3d532f PZ |
3010 | struct intel_crtc *crtc; |
3011 | struct intel_plane *plane; | |
d8c0fafc | 3012 | struct intel_crtc_state *cstate; |
3013 | struct skl_plane_wm *wm; | |
656d1b89 | 3014 | enum pipe pipe; |
d8c0fafc | 3015 | int level, latency; |
656d1b89 | 3016 | |
56feca91 PZ |
3017 | if (!intel_has_sagv(dev_priv)) |
3018 | return false; | |
3019 | ||
656d1b89 L |
3020 | /* |
3021 | * SKL workaround: bspec recommends we disable the SAGV when we have | |
3022 | * more then one pipe enabled | |
3023 | * | |
3024 | * If there are no active CRTCs, no additional checks need be performed | |
3025 | */ | |
3026 | if (hweight32(intel_state->active_crtcs) == 0) | |
3027 | return true; | |
3028 | else if (hweight32(intel_state->active_crtcs) > 1) | |
3029 | return false; | |
3030 | ||
3031 | /* Since we're now guaranteed to only have one active CRTC... */ | |
3032 | pipe = ffs(intel_state->active_crtcs) - 1; | |
98187836 | 3033 | crtc = intel_get_crtc_for_pipe(dev_priv, pipe); |
d8c0fafc | 3034 | cstate = to_intel_crtc_state(crtc->base.state); |
656d1b89 | 3035 | |
c89cadd5 | 3036 | if (crtc->base.state->adjusted_mode.flags & DRM_MODE_FLAG_INTERLACE) |
656d1b89 L |
3037 | return false; |
3038 | ||
ee3d532f | 3039 | for_each_intel_plane_on_crtc(dev, crtc, plane) { |
d8c0fafc | 3040 | wm = &cstate->wm.skl.optimal.planes[skl_wm_plane_id(plane)]; |
ee3d532f | 3041 | |
656d1b89 | 3042 | /* Skip this plane if it's not enabled */ |
d8c0fafc | 3043 | if (!wm->wm[0].plane_en) |
656d1b89 L |
3044 | continue; |
3045 | ||
3046 | /* Find the highest enabled wm level for this plane */ | |
5db94019 | 3047 | for (level = ilk_wm_max_level(dev_priv); |
d8c0fafc | 3048 | !wm->wm[level].plane_en; --level) |
656d1b89 L |
3049 | { } |
3050 | ||
ee3d532f PZ |
3051 | latency = dev_priv->wm.skl_latency[level]; |
3052 | ||
3053 | if (skl_needs_memory_bw_wa(intel_state) && | |
bae781b2 | 3054 | plane->base.state->fb->modifier == |
ee3d532f PZ |
3055 | I915_FORMAT_MOD_X_TILED) |
3056 | latency += 15; | |
3057 | ||
656d1b89 L |
3058 | /* |
3059 | * If any of the planes on this pipe don't enable wm levels | |
3060 | * that incur memory latencies higher then 30µs we can't enable | |
3061 | * the SAGV | |
3062 | */ | |
ee3d532f | 3063 | if (latency < SKL_SAGV_BLOCK_TIME) |
656d1b89 L |
3064 | return false; |
3065 | } | |
3066 | ||
3067 | return true; | |
3068 | } | |
3069 | ||
b9cec075 DL |
3070 | static void |
3071 | skl_ddb_get_pipe_allocation_limits(struct drm_device *dev, | |
024c9045 | 3072 | const struct intel_crtc_state *cstate, |
c107acfe MR |
3073 | struct skl_ddb_entry *alloc, /* out */ |
3074 | int *num_active /* out */) | |
b9cec075 | 3075 | { |
c107acfe MR |
3076 | struct drm_atomic_state *state = cstate->base.state; |
3077 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
3078 | struct drm_i915_private *dev_priv = to_i915(dev); | |
024c9045 | 3079 | struct drm_crtc *for_crtc = cstate->base.crtc; |
b9cec075 DL |
3080 | unsigned int pipe_size, ddb_size; |
3081 | int nth_active_pipe; | |
c107acfe | 3082 | |
a6d3460e | 3083 | if (WARN_ON(!state) || !cstate->base.active) { |
b9cec075 DL |
3084 | alloc->start = 0; |
3085 | alloc->end = 0; | |
a6d3460e | 3086 | *num_active = hweight32(dev_priv->active_crtcs); |
b9cec075 DL |
3087 | return; |
3088 | } | |
3089 | ||
a6d3460e MR |
3090 | if (intel_state->active_pipe_changes) |
3091 | *num_active = hweight32(intel_state->active_crtcs); | |
3092 | else | |
3093 | *num_active = hweight32(dev_priv->active_crtcs); | |
3094 | ||
6f3fff60 D |
3095 | ddb_size = INTEL_INFO(dev_priv)->ddb_size; |
3096 | WARN_ON(ddb_size == 0); | |
b9cec075 DL |
3097 | |
3098 | ddb_size -= 4; /* 4 blocks for bypass path allocation */ | |
3099 | ||
c107acfe | 3100 | /* |
a6d3460e MR |
3101 | * If the state doesn't change the active CRTC's, then there's |
3102 | * no need to recalculate; the existing pipe allocation limits | |
3103 | * should remain unchanged. Note that we're safe from racing | |
3104 | * commits since any racing commit that changes the active CRTC | |
3105 | * list would need to grab _all_ crtc locks, including the one | |
3106 | * we currently hold. | |
c107acfe | 3107 | */ |
a6d3460e | 3108 | if (!intel_state->active_pipe_changes) { |
512b5527 ML |
3109 | /* |
3110 | * alloc may be cleared by clear_intel_crtc_state, | |
3111 | * copy from old state to be sure | |
3112 | */ | |
3113 | *alloc = to_intel_crtc_state(for_crtc->state)->wm.skl.ddb; | |
a6d3460e | 3114 | return; |
c107acfe | 3115 | } |
a6d3460e MR |
3116 | |
3117 | nth_active_pipe = hweight32(intel_state->active_crtcs & | |
3118 | (drm_crtc_mask(for_crtc) - 1)); | |
3119 | pipe_size = ddb_size / hweight32(intel_state->active_crtcs); | |
3120 | alloc->start = nth_active_pipe * ddb_size / *num_active; | |
3121 | alloc->end = alloc->start + pipe_size; | |
b9cec075 DL |
3122 | } |
3123 | ||
c107acfe | 3124 | static unsigned int skl_cursor_allocation(int num_active) |
b9cec075 | 3125 | { |
c107acfe | 3126 | if (num_active == 1) |
b9cec075 DL |
3127 | return 32; |
3128 | ||
3129 | return 8; | |
3130 | } | |
3131 | ||
a269c583 DL |
3132 | static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg) |
3133 | { | |
3134 | entry->start = reg & 0x3ff; | |
3135 | entry->end = (reg >> 16) & 0x3ff; | |
16160e3d DL |
3136 | if (entry->end) |
3137 | entry->end += 1; | |
a269c583 DL |
3138 | } |
3139 | ||
08db6652 DL |
3140 | void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv, |
3141 | struct skl_ddb_allocation *ddb /* out */) | |
a269c583 | 3142 | { |
a269c583 DL |
3143 | enum pipe pipe; |
3144 | int plane; | |
3145 | u32 val; | |
3146 | ||
b10f1b20 ML |
3147 | memset(ddb, 0, sizeof(*ddb)); |
3148 | ||
a269c583 | 3149 | for_each_pipe(dev_priv, pipe) { |
4d800030 ID |
3150 | enum intel_display_power_domain power_domain; |
3151 | ||
3152 | power_domain = POWER_DOMAIN_PIPE(pipe); | |
3153 | if (!intel_display_power_get_if_enabled(dev_priv, power_domain)) | |
b10f1b20 ML |
3154 | continue; |
3155 | ||
8b364b41 | 3156 | for_each_universal_plane(dev_priv, pipe, plane) { |
a269c583 DL |
3157 | val = I915_READ(PLANE_BUF_CFG(pipe, plane)); |
3158 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane], | |
3159 | val); | |
3160 | } | |
3161 | ||
3162 | val = I915_READ(CUR_BUF_CFG(pipe)); | |
4969d33e MR |
3163 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][PLANE_CURSOR], |
3164 | val); | |
4d800030 ID |
3165 | |
3166 | intel_display_power_put(dev_priv, power_domain); | |
a269c583 DL |
3167 | } |
3168 | } | |
3169 | ||
9c2f7a9d KM |
3170 | /* |
3171 | * Determines the downscale amount of a plane for the purposes of watermark calculations. | |
3172 | * The bspec defines downscale amount as: | |
3173 | * | |
3174 | * """ | |
3175 | * Horizontal down scale amount = maximum[1, Horizontal source size / | |
3176 | * Horizontal destination size] | |
3177 | * Vertical down scale amount = maximum[1, Vertical source size / | |
3178 | * Vertical destination size] | |
3179 | * Total down scale amount = Horizontal down scale amount * | |
3180 | * Vertical down scale amount | |
3181 | * """ | |
3182 | * | |
3183 | * Return value is provided in 16.16 fixed point form to retain fractional part. | |
3184 | * Caller should take care of dividing & rounding off the value. | |
3185 | */ | |
3186 | static uint32_t | |
3187 | skl_plane_downscale_amount(const struct intel_plane_state *pstate) | |
3188 | { | |
3189 | uint32_t downscale_h, downscale_w; | |
3190 | uint32_t src_w, src_h, dst_w, dst_h; | |
3191 | ||
936e71e3 | 3192 | if (WARN_ON(!pstate->base.visible)) |
9c2f7a9d KM |
3193 | return DRM_PLANE_HELPER_NO_SCALING; |
3194 | ||
3195 | /* n.b., src is 16.16 fixed point, dst is whole integer */ | |
936e71e3 VS |
3196 | src_w = drm_rect_width(&pstate->base.src); |
3197 | src_h = drm_rect_height(&pstate->base.src); | |
3198 | dst_w = drm_rect_width(&pstate->base.dst); | |
3199 | dst_h = drm_rect_height(&pstate->base.dst); | |
bd2ef25d | 3200 | if (drm_rotation_90_or_270(pstate->base.rotation)) |
9c2f7a9d KM |
3201 | swap(dst_w, dst_h); |
3202 | ||
3203 | downscale_h = max(src_h / dst_h, (uint32_t)DRM_PLANE_HELPER_NO_SCALING); | |
3204 | downscale_w = max(src_w / dst_w, (uint32_t)DRM_PLANE_HELPER_NO_SCALING); | |
3205 | ||
3206 | /* Provide result in 16.16 fixed point */ | |
3207 | return (uint64_t)downscale_w * downscale_h >> 16; | |
3208 | } | |
3209 | ||
b9cec075 | 3210 | static unsigned int |
024c9045 MR |
3211 | skl_plane_relative_data_rate(const struct intel_crtc_state *cstate, |
3212 | const struct drm_plane_state *pstate, | |
3213 | int y) | |
b9cec075 | 3214 | { |
a280f7dd | 3215 | struct intel_plane_state *intel_pstate = to_intel_plane_state(pstate); |
024c9045 | 3216 | struct drm_framebuffer *fb = pstate->fb; |
8d19d7d9 | 3217 | uint32_t down_scale_amount, data_rate; |
a280f7dd | 3218 | uint32_t width = 0, height = 0; |
a1de91e5 MR |
3219 | unsigned format = fb ? fb->pixel_format : DRM_FORMAT_XRGB8888; |
3220 | ||
936e71e3 | 3221 | if (!intel_pstate->base.visible) |
a1de91e5 MR |
3222 | return 0; |
3223 | if (pstate->plane->type == DRM_PLANE_TYPE_CURSOR) | |
3224 | return 0; | |
3225 | if (y && format != DRM_FORMAT_NV12) | |
3226 | return 0; | |
a280f7dd | 3227 | |
936e71e3 VS |
3228 | width = drm_rect_width(&intel_pstate->base.src) >> 16; |
3229 | height = drm_rect_height(&intel_pstate->base.src) >> 16; | |
a280f7dd | 3230 | |
bd2ef25d | 3231 | if (drm_rotation_90_or_270(pstate->rotation)) |
a280f7dd | 3232 | swap(width, height); |
2cd601c6 CK |
3233 | |
3234 | /* for planar format */ | |
a1de91e5 | 3235 | if (format == DRM_FORMAT_NV12) { |
2cd601c6 | 3236 | if (y) /* y-plane data rate */ |
8d19d7d9 | 3237 | data_rate = width * height * |
a1de91e5 | 3238 | drm_format_plane_cpp(format, 0); |
2cd601c6 | 3239 | else /* uv-plane data rate */ |
8d19d7d9 | 3240 | data_rate = (width / 2) * (height / 2) * |
a1de91e5 | 3241 | drm_format_plane_cpp(format, 1); |
8d19d7d9 KM |
3242 | } else { |
3243 | /* for packed formats */ | |
3244 | data_rate = width * height * drm_format_plane_cpp(format, 0); | |
2cd601c6 CK |
3245 | } |
3246 | ||
8d19d7d9 KM |
3247 | down_scale_amount = skl_plane_downscale_amount(intel_pstate); |
3248 | ||
3249 | return (uint64_t)data_rate * down_scale_amount >> 16; | |
b9cec075 DL |
3250 | } |
3251 | ||
3252 | /* | |
3253 | * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching | |
3254 | * a 8192x4096@32bpp framebuffer: | |
3255 | * 3 * 4096 * 8192 * 4 < 2^32 | |
3256 | */ | |
3257 | static unsigned int | |
1e6ee542 ML |
3258 | skl_get_total_relative_data_rate(struct intel_crtc_state *intel_cstate, |
3259 | unsigned *plane_data_rate, | |
3260 | unsigned *plane_y_data_rate) | |
b9cec075 | 3261 | { |
9c74d826 MR |
3262 | struct drm_crtc_state *cstate = &intel_cstate->base; |
3263 | struct drm_atomic_state *state = cstate->state; | |
c8fe32c1 | 3264 | struct drm_plane *plane; |
024c9045 | 3265 | const struct intel_plane *intel_plane; |
c8fe32c1 | 3266 | const struct drm_plane_state *pstate; |
a1de91e5 | 3267 | unsigned int rate, total_data_rate = 0; |
9c74d826 | 3268 | int id; |
a6d3460e MR |
3269 | |
3270 | if (WARN_ON(!state)) | |
3271 | return 0; | |
b9cec075 | 3272 | |
a1de91e5 | 3273 | /* Calculate and cache data rate for each plane */ |
c8fe32c1 | 3274 | drm_atomic_crtc_state_for_each_plane_state(plane, pstate, cstate) { |
a6d3460e MR |
3275 | id = skl_wm_plane_id(to_intel_plane(plane)); |
3276 | intel_plane = to_intel_plane(plane); | |
3277 | ||
a6d3460e MR |
3278 | /* packed/uv */ |
3279 | rate = skl_plane_relative_data_rate(intel_cstate, | |
3280 | pstate, 0); | |
1e6ee542 ML |
3281 | plane_data_rate[id] = rate; |
3282 | ||
3283 | total_data_rate += rate; | |
a6d3460e MR |
3284 | |
3285 | /* y-plane */ | |
3286 | rate = skl_plane_relative_data_rate(intel_cstate, | |
3287 | pstate, 1); | |
1e6ee542 | 3288 | plane_y_data_rate[id] = rate; |
024c9045 | 3289 | |
1e6ee542 | 3290 | total_data_rate += rate; |
b9cec075 DL |
3291 | } |
3292 | ||
3293 | return total_data_rate; | |
3294 | } | |
3295 | ||
cbcfd14b KM |
3296 | static uint16_t |
3297 | skl_ddb_min_alloc(const struct drm_plane_state *pstate, | |
3298 | const int y) | |
3299 | { | |
3300 | struct drm_framebuffer *fb = pstate->fb; | |
3301 | struct intel_plane_state *intel_pstate = to_intel_plane_state(pstate); | |
3302 | uint32_t src_w, src_h; | |
3303 | uint32_t min_scanlines = 8; | |
3304 | uint8_t plane_bpp; | |
3305 | ||
3306 | if (WARN_ON(!fb)) | |
3307 | return 0; | |
3308 | ||
3309 | /* For packed formats, no y-plane, return 0 */ | |
3310 | if (y && fb->pixel_format != DRM_FORMAT_NV12) | |
3311 | return 0; | |
3312 | ||
3313 | /* For Non Y-tile return 8-blocks */ | |
bae781b2 VS |
3314 | if (fb->modifier != I915_FORMAT_MOD_Y_TILED && |
3315 | fb->modifier != I915_FORMAT_MOD_Yf_TILED) | |
cbcfd14b KM |
3316 | return 8; |
3317 | ||
936e71e3 VS |
3318 | src_w = drm_rect_width(&intel_pstate->base.src) >> 16; |
3319 | src_h = drm_rect_height(&intel_pstate->base.src) >> 16; | |
cbcfd14b | 3320 | |
bd2ef25d | 3321 | if (drm_rotation_90_or_270(pstate->rotation)) |
cbcfd14b KM |
3322 | swap(src_w, src_h); |
3323 | ||
3324 | /* Halve UV plane width and height for NV12 */ | |
3325 | if (fb->pixel_format == DRM_FORMAT_NV12 && !y) { | |
3326 | src_w /= 2; | |
3327 | src_h /= 2; | |
3328 | } | |
3329 | ||
3330 | if (fb->pixel_format == DRM_FORMAT_NV12 && !y) | |
3331 | plane_bpp = drm_format_plane_cpp(fb->pixel_format, 1); | |
3332 | else | |
3333 | plane_bpp = drm_format_plane_cpp(fb->pixel_format, 0); | |
3334 | ||
bd2ef25d | 3335 | if (drm_rotation_90_or_270(pstate->rotation)) { |
cbcfd14b KM |
3336 | switch (plane_bpp) { |
3337 | case 1: | |
3338 | min_scanlines = 32; | |
3339 | break; | |
3340 | case 2: | |
3341 | min_scanlines = 16; | |
3342 | break; | |
3343 | case 4: | |
3344 | min_scanlines = 8; | |
3345 | break; | |
3346 | case 8: | |
3347 | min_scanlines = 4; | |
3348 | break; | |
3349 | default: | |
3350 | WARN(1, "Unsupported pixel depth %u for rotation", | |
3351 | plane_bpp); | |
3352 | min_scanlines = 32; | |
3353 | } | |
3354 | } | |
3355 | ||
3356 | return DIV_ROUND_UP((4 * src_w * plane_bpp), 512) * min_scanlines/4 + 3; | |
3357 | } | |
3358 | ||
49845a7a ML |
3359 | static void |
3360 | skl_ddb_calc_min(const struct intel_crtc_state *cstate, int num_active, | |
3361 | uint16_t *minimum, uint16_t *y_minimum) | |
3362 | { | |
3363 | const struct drm_plane_state *pstate; | |
3364 | struct drm_plane *plane; | |
3365 | ||
3366 | drm_atomic_crtc_state_for_each_plane_state(plane, pstate, &cstate->base) { | |
3367 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
3368 | int id = skl_wm_plane_id(intel_plane); | |
3369 | ||
3370 | if (id == PLANE_CURSOR) | |
3371 | continue; | |
3372 | ||
3373 | if (!pstate->visible) | |
3374 | continue; | |
3375 | ||
3376 | minimum[id] = skl_ddb_min_alloc(pstate, 0); | |
3377 | y_minimum[id] = skl_ddb_min_alloc(pstate, 1); | |
3378 | } | |
3379 | ||
3380 | minimum[PLANE_CURSOR] = skl_cursor_allocation(num_active); | |
3381 | } | |
3382 | ||
c107acfe | 3383 | static int |
024c9045 | 3384 | skl_allocate_pipe_ddb(struct intel_crtc_state *cstate, |
b9cec075 DL |
3385 | struct skl_ddb_allocation *ddb /* out */) |
3386 | { | |
c107acfe | 3387 | struct drm_atomic_state *state = cstate->base.state; |
024c9045 | 3388 | struct drm_crtc *crtc = cstate->base.crtc; |
b9cec075 DL |
3389 | struct drm_device *dev = crtc->dev; |
3390 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3391 | enum pipe pipe = intel_crtc->pipe; | |
ce0ba283 | 3392 | struct skl_ddb_entry *alloc = &cstate->wm.skl.ddb; |
49845a7a | 3393 | uint16_t alloc_size, start; |
fefdd810 ML |
3394 | uint16_t minimum[I915_MAX_PLANES] = {}; |
3395 | uint16_t y_minimum[I915_MAX_PLANES] = {}; | |
b9cec075 | 3396 | unsigned int total_data_rate; |
c107acfe MR |
3397 | int num_active; |
3398 | int id, i; | |
1e6ee542 ML |
3399 | unsigned plane_data_rate[I915_MAX_PLANES] = {}; |
3400 | unsigned plane_y_data_rate[I915_MAX_PLANES] = {}; | |
b9cec075 | 3401 | |
5a920b85 PZ |
3402 | /* Clear the partitioning for disabled planes. */ |
3403 | memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe])); | |
3404 | memset(ddb->y_plane[pipe], 0, sizeof(ddb->y_plane[pipe])); | |
3405 | ||
a6d3460e MR |
3406 | if (WARN_ON(!state)) |
3407 | return 0; | |
3408 | ||
c107acfe | 3409 | if (!cstate->base.active) { |
ce0ba283 | 3410 | alloc->start = alloc->end = 0; |
c107acfe MR |
3411 | return 0; |
3412 | } | |
3413 | ||
a6d3460e | 3414 | skl_ddb_get_pipe_allocation_limits(dev, cstate, alloc, &num_active); |
34bb56af | 3415 | alloc_size = skl_ddb_entry_size(alloc); |
b9cec075 DL |
3416 | if (alloc_size == 0) { |
3417 | memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe])); | |
c107acfe | 3418 | return 0; |
b9cec075 DL |
3419 | } |
3420 | ||
49845a7a | 3421 | skl_ddb_calc_min(cstate, num_active, minimum, y_minimum); |
a6d3460e | 3422 | |
49845a7a ML |
3423 | /* |
3424 | * 1. Allocate the mininum required blocks for each active plane | |
3425 | * and allocate the cursor, it doesn't require extra allocation | |
3426 | * proportional to the data rate. | |
3427 | */ | |
80958155 | 3428 | |
49845a7a | 3429 | for (i = 0; i < I915_MAX_PLANES; i++) { |
c107acfe MR |
3430 | alloc_size -= minimum[i]; |
3431 | alloc_size -= y_minimum[i]; | |
80958155 DL |
3432 | } |
3433 | ||
49845a7a ML |
3434 | ddb->plane[pipe][PLANE_CURSOR].start = alloc->end - minimum[PLANE_CURSOR]; |
3435 | ddb->plane[pipe][PLANE_CURSOR].end = alloc->end; | |
3436 | ||
b9cec075 | 3437 | /* |
80958155 DL |
3438 | * 2. Distribute the remaining space in proportion to the amount of |
3439 | * data each plane needs to fetch from memory. | |
b9cec075 DL |
3440 | * |
3441 | * FIXME: we may not allocate every single block here. | |
3442 | */ | |
1e6ee542 ML |
3443 | total_data_rate = skl_get_total_relative_data_rate(cstate, |
3444 | plane_data_rate, | |
3445 | plane_y_data_rate); | |
a1de91e5 | 3446 | if (total_data_rate == 0) |
c107acfe | 3447 | return 0; |
b9cec075 | 3448 | |
34bb56af | 3449 | start = alloc->start; |
1e6ee542 | 3450 | for (id = 0; id < I915_MAX_PLANES; id++) { |
2cd601c6 CK |
3451 | unsigned int data_rate, y_data_rate; |
3452 | uint16_t plane_blocks, y_plane_blocks = 0; | |
b9cec075 | 3453 | |
49845a7a ML |
3454 | if (id == PLANE_CURSOR) |
3455 | continue; | |
3456 | ||
1e6ee542 | 3457 | data_rate = plane_data_rate[id]; |
b9cec075 DL |
3458 | |
3459 | /* | |
2cd601c6 | 3460 | * allocation for (packed formats) or (uv-plane part of planar format): |
b9cec075 DL |
3461 | * promote the expression to 64 bits to avoid overflowing, the |
3462 | * result is < available as data_rate / total_data_rate < 1 | |
3463 | */ | |
024c9045 | 3464 | plane_blocks = minimum[id]; |
80958155 DL |
3465 | plane_blocks += div_u64((uint64_t)alloc_size * data_rate, |
3466 | total_data_rate); | |
b9cec075 | 3467 | |
c107acfe MR |
3468 | /* Leave disabled planes at (0,0) */ |
3469 | if (data_rate) { | |
3470 | ddb->plane[pipe][id].start = start; | |
3471 | ddb->plane[pipe][id].end = start + plane_blocks; | |
3472 | } | |
b9cec075 DL |
3473 | |
3474 | start += plane_blocks; | |
2cd601c6 CK |
3475 | |
3476 | /* | |
3477 | * allocation for y_plane part of planar format: | |
3478 | */ | |
1e6ee542 | 3479 | y_data_rate = plane_y_data_rate[id]; |
a1de91e5 MR |
3480 | |
3481 | y_plane_blocks = y_minimum[id]; | |
3482 | y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate, | |
3483 | total_data_rate); | |
2cd601c6 | 3484 | |
c107acfe MR |
3485 | if (y_data_rate) { |
3486 | ddb->y_plane[pipe][id].start = start; | |
3487 | ddb->y_plane[pipe][id].end = start + y_plane_blocks; | |
3488 | } | |
a1de91e5 MR |
3489 | |
3490 | start += y_plane_blocks; | |
b9cec075 DL |
3491 | } |
3492 | ||
c107acfe | 3493 | return 0; |
b9cec075 DL |
3494 | } |
3495 | ||
2d41c0b5 PB |
3496 | /* |
3497 | * The max latency should be 257 (max the punit can code is 255 and we add 2us | |
ac484963 | 3498 | * for the read latency) and cpp should always be <= 8, so that |
2d41c0b5 PB |
3499 | * should allow pixel_rate up to ~2 GHz which seems sufficient since max |
3500 | * 2xcdclk is 1350 MHz and the pixel rate should never exceed that. | |
3501 | */ | |
ac484963 | 3502 | static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t cpp, uint32_t latency) |
2d41c0b5 PB |
3503 | { |
3504 | uint32_t wm_intermediate_val, ret; | |
3505 | ||
3506 | if (latency == 0) | |
3507 | return UINT_MAX; | |
3508 | ||
ac484963 | 3509 | wm_intermediate_val = latency * pixel_rate * cpp / 512; |
2d41c0b5 PB |
3510 | ret = DIV_ROUND_UP(wm_intermediate_val, 1000); |
3511 | ||
3512 | return ret; | |
3513 | } | |
3514 | ||
3515 | static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, | |
7a1a8aed | 3516 | uint32_t latency, uint32_t plane_blocks_per_line) |
2d41c0b5 | 3517 | { |
d4c2aa60 | 3518 | uint32_t ret; |
d4c2aa60 | 3519 | uint32_t wm_intermediate_val; |
2d41c0b5 PB |
3520 | |
3521 | if (latency == 0) | |
3522 | return UINT_MAX; | |
3523 | ||
2d41c0b5 PB |
3524 | wm_intermediate_val = latency * pixel_rate; |
3525 | ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) * | |
d4c2aa60 | 3526 | plane_blocks_per_line; |
2d41c0b5 PB |
3527 | |
3528 | return ret; | |
3529 | } | |
3530 | ||
9c2f7a9d KM |
3531 | static uint32_t skl_adjusted_plane_pixel_rate(const struct intel_crtc_state *cstate, |
3532 | struct intel_plane_state *pstate) | |
3533 | { | |
3534 | uint64_t adjusted_pixel_rate; | |
3535 | uint64_t downscale_amount; | |
3536 | uint64_t pixel_rate; | |
3537 | ||
3538 | /* Shouldn't reach here on disabled planes... */ | |
936e71e3 | 3539 | if (WARN_ON(!pstate->base.visible)) |
9c2f7a9d KM |
3540 | return 0; |
3541 | ||
3542 | /* | |
3543 | * Adjusted plane pixel rate is just the pipe's adjusted pixel rate | |
3544 | * with additional adjustments for plane-specific scaling. | |
3545 | */ | |
cfd7e3a2 | 3546 | adjusted_pixel_rate = ilk_pipe_pixel_rate(cstate); |
9c2f7a9d KM |
3547 | downscale_amount = skl_plane_downscale_amount(pstate); |
3548 | ||
3549 | pixel_rate = adjusted_pixel_rate * downscale_amount >> 16; | |
3550 | WARN_ON(pixel_rate != clamp_t(uint32_t, pixel_rate, 0, ~0)); | |
3551 | ||
3552 | return pixel_rate; | |
3553 | } | |
3554 | ||
55994c2c MR |
3555 | static int skl_compute_plane_wm(const struct drm_i915_private *dev_priv, |
3556 | struct intel_crtc_state *cstate, | |
3557 | struct intel_plane_state *intel_pstate, | |
3558 | uint16_t ddb_allocation, | |
3559 | int level, | |
3560 | uint16_t *out_blocks, /* out */ | |
3561 | uint8_t *out_lines, /* out */ | |
3562 | bool *enabled /* out */) | |
2d41c0b5 | 3563 | { |
33815fa5 MR |
3564 | struct drm_plane_state *pstate = &intel_pstate->base; |
3565 | struct drm_framebuffer *fb = pstate->fb; | |
d4c2aa60 TU |
3566 | uint32_t latency = dev_priv->wm.skl_latency[level]; |
3567 | uint32_t method1, method2; | |
3568 | uint32_t plane_bytes_per_line, plane_blocks_per_line; | |
3569 | uint32_t res_blocks, res_lines; | |
3570 | uint32_t selected_result; | |
ac484963 | 3571 | uint8_t cpp; |
a280f7dd | 3572 | uint32_t width = 0, height = 0; |
9c2f7a9d | 3573 | uint32_t plane_pixel_rate; |
75676ed4 | 3574 | uint32_t y_tile_minimum, y_min_scanlines; |
ee3d532f PZ |
3575 | struct intel_atomic_state *state = |
3576 | to_intel_atomic_state(cstate->base.state); | |
3577 | bool apply_memory_bw_wa = skl_needs_memory_bw_wa(state); | |
2d41c0b5 | 3578 | |
936e71e3 | 3579 | if (latency == 0 || !cstate->base.active || !intel_pstate->base.visible) { |
55994c2c MR |
3580 | *enabled = false; |
3581 | return 0; | |
3582 | } | |
2d41c0b5 | 3583 | |
bae781b2 | 3584 | if (apply_memory_bw_wa && fb->modifier == I915_FORMAT_MOD_X_TILED) |
ee3d532f PZ |
3585 | latency += 15; |
3586 | ||
936e71e3 VS |
3587 | width = drm_rect_width(&intel_pstate->base.src) >> 16; |
3588 | height = drm_rect_height(&intel_pstate->base.src) >> 16; | |
a280f7dd | 3589 | |
bd2ef25d | 3590 | if (drm_rotation_90_or_270(pstate->rotation)) |
a280f7dd KM |
3591 | swap(width, height); |
3592 | ||
ac484963 | 3593 | cpp = drm_format_plane_cpp(fb->pixel_format, 0); |
9c2f7a9d KM |
3594 | plane_pixel_rate = skl_adjusted_plane_pixel_rate(cstate, intel_pstate); |
3595 | ||
61d0a04d | 3596 | if (drm_rotation_90_or_270(pstate->rotation)) { |
1186fa85 PZ |
3597 | int cpp = (fb->pixel_format == DRM_FORMAT_NV12) ? |
3598 | drm_format_plane_cpp(fb->pixel_format, 1) : | |
3599 | drm_format_plane_cpp(fb->pixel_format, 0); | |
3600 | ||
3601 | switch (cpp) { | |
3602 | case 1: | |
3603 | y_min_scanlines = 16; | |
3604 | break; | |
3605 | case 2: | |
3606 | y_min_scanlines = 8; | |
3607 | break; | |
1186fa85 PZ |
3608 | case 4: |
3609 | y_min_scanlines = 4; | |
3610 | break; | |
86a462bc PZ |
3611 | default: |
3612 | MISSING_CASE(cpp); | |
3613 | return -EINVAL; | |
1186fa85 PZ |
3614 | } |
3615 | } else { | |
3616 | y_min_scanlines = 4; | |
3617 | } | |
3618 | ||
2ef32dee PZ |
3619 | if (apply_memory_bw_wa) |
3620 | y_min_scanlines *= 2; | |
3621 | ||
7a1a8aed | 3622 | plane_bytes_per_line = width * cpp; |
bae781b2 VS |
3623 | if (fb->modifier == I915_FORMAT_MOD_Y_TILED || |
3624 | fb->modifier == I915_FORMAT_MOD_Yf_TILED) { | |
7a1a8aed PZ |
3625 | plane_blocks_per_line = |
3626 | DIV_ROUND_UP(plane_bytes_per_line * y_min_scanlines, 512); | |
3627 | plane_blocks_per_line /= y_min_scanlines; | |
bae781b2 | 3628 | } else if (fb->modifier == DRM_FORMAT_MOD_NONE) { |
7a1a8aed PZ |
3629 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512) |
3630 | + 1; | |
3631 | } else { | |
3632 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); | |
3633 | } | |
3634 | ||
9c2f7a9d KM |
3635 | method1 = skl_wm_method1(plane_pixel_rate, cpp, latency); |
3636 | method2 = skl_wm_method2(plane_pixel_rate, | |
024c9045 | 3637 | cstate->base.adjusted_mode.crtc_htotal, |
1186fa85 | 3638 | latency, |
7a1a8aed | 3639 | plane_blocks_per_line); |
2d41c0b5 | 3640 | |
75676ed4 PZ |
3641 | y_tile_minimum = plane_blocks_per_line * y_min_scanlines; |
3642 | ||
bae781b2 VS |
3643 | if (fb->modifier == I915_FORMAT_MOD_Y_TILED || |
3644 | fb->modifier == I915_FORMAT_MOD_Yf_TILED) { | |
0fda6568 TU |
3645 | selected_result = max(method2, y_tile_minimum); |
3646 | } else { | |
f1db3eaf PZ |
3647 | if ((cpp * cstate->base.adjusted_mode.crtc_htotal / 512 < 1) && |
3648 | (plane_bytes_per_line / 512 < 1)) | |
3649 | selected_result = method2; | |
3650 | else if ((ddb_allocation / plane_blocks_per_line) >= 1) | |
0fda6568 TU |
3651 | selected_result = min(method1, method2); |
3652 | else | |
3653 | selected_result = method1; | |
3654 | } | |
2d41c0b5 | 3655 | |
d4c2aa60 TU |
3656 | res_blocks = selected_result + 1; |
3657 | res_lines = DIV_ROUND_UP(selected_result, plane_blocks_per_line); | |
e6d66171 | 3658 | |
0fda6568 | 3659 | if (level >= 1 && level <= 7) { |
bae781b2 VS |
3660 | if (fb->modifier == I915_FORMAT_MOD_Y_TILED || |
3661 | fb->modifier == I915_FORMAT_MOD_Yf_TILED) { | |
75676ed4 | 3662 | res_blocks += y_tile_minimum; |
1186fa85 | 3663 | res_lines += y_min_scanlines; |
75676ed4 | 3664 | } else { |
0fda6568 | 3665 | res_blocks++; |
75676ed4 | 3666 | } |
0fda6568 | 3667 | } |
e6d66171 | 3668 | |
55994c2c MR |
3669 | if (res_blocks >= ddb_allocation || res_lines > 31) { |
3670 | *enabled = false; | |
6b6bada7 MR |
3671 | |
3672 | /* | |
3673 | * If there are no valid level 0 watermarks, then we can't | |
3674 | * support this display configuration. | |
3675 | */ | |
3676 | if (level) { | |
3677 | return 0; | |
3678 | } else { | |
3679 | DRM_DEBUG_KMS("Requested display configuration exceeds system watermark limitations\n"); | |
3680 | DRM_DEBUG_KMS("Plane %d.%d: blocks required = %u/%u, lines required = %u/31\n", | |
3681 | to_intel_crtc(cstate->base.crtc)->pipe, | |
3682 | skl_wm_plane_id(to_intel_plane(pstate->plane)), | |
3683 | res_blocks, ddb_allocation, res_lines); | |
3684 | ||
3685 | return -EINVAL; | |
3686 | } | |
55994c2c | 3687 | } |
e6d66171 DL |
3688 | |
3689 | *out_blocks = res_blocks; | |
3690 | *out_lines = res_lines; | |
55994c2c | 3691 | *enabled = true; |
2d41c0b5 | 3692 | |
55994c2c | 3693 | return 0; |
2d41c0b5 PB |
3694 | } |
3695 | ||
f4a96752 MR |
3696 | static int |
3697 | skl_compute_wm_level(const struct drm_i915_private *dev_priv, | |
3698 | struct skl_ddb_allocation *ddb, | |
3699 | struct intel_crtc_state *cstate, | |
a62163e9 | 3700 | struct intel_plane *intel_plane, |
f4a96752 MR |
3701 | int level, |
3702 | struct skl_wm_level *result) | |
2d41c0b5 | 3703 | { |
f4a96752 | 3704 | struct drm_atomic_state *state = cstate->base.state; |
024c9045 | 3705 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); |
a62163e9 L |
3706 | struct drm_plane *plane = &intel_plane->base; |
3707 | struct intel_plane_state *intel_pstate = NULL; | |
2d41c0b5 | 3708 | uint16_t ddb_blocks; |
024c9045 | 3709 | enum pipe pipe = intel_crtc->pipe; |
55994c2c | 3710 | int ret; |
a62163e9 L |
3711 | int i = skl_wm_plane_id(intel_plane); |
3712 | ||
3713 | if (state) | |
3714 | intel_pstate = | |
3715 | intel_atomic_get_existing_plane_state(state, | |
3716 | intel_plane); | |
024c9045 | 3717 | |
f4a96752 | 3718 | /* |
a62163e9 L |
3719 | * Note: If we start supporting multiple pending atomic commits against |
3720 | * the same planes/CRTC's in the future, plane->state will no longer be | |
3721 | * the correct pre-state to use for the calculations here and we'll | |
3722 | * need to change where we get the 'unchanged' plane data from. | |
3723 | * | |
3724 | * For now this is fine because we only allow one queued commit against | |
3725 | * a CRTC. Even if the plane isn't modified by this transaction and we | |
3726 | * don't have a plane lock, we still have the CRTC's lock, so we know | |
3727 | * that no other transactions are racing with us to update it. | |
f4a96752 | 3728 | */ |
a62163e9 L |
3729 | if (!intel_pstate) |
3730 | intel_pstate = to_intel_plane_state(plane->state); | |
f4a96752 | 3731 | |
a62163e9 | 3732 | WARN_ON(!intel_pstate->base.fb); |
f4a96752 | 3733 | |
a62163e9 | 3734 | ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]); |
2d41c0b5 | 3735 | |
a62163e9 L |
3736 | ret = skl_compute_plane_wm(dev_priv, |
3737 | cstate, | |
3738 | intel_pstate, | |
3739 | ddb_blocks, | |
3740 | level, | |
3741 | &result->plane_res_b, | |
3742 | &result->plane_res_l, | |
3743 | &result->plane_en); | |
3744 | if (ret) | |
3745 | return ret; | |
f4a96752 MR |
3746 | |
3747 | return 0; | |
2d41c0b5 PB |
3748 | } |
3749 | ||
407b50f3 | 3750 | static uint32_t |
024c9045 | 3751 | skl_compute_linetime_wm(struct intel_crtc_state *cstate) |
407b50f3 | 3752 | { |
30d1b5fe PZ |
3753 | uint32_t pixel_rate; |
3754 | ||
024c9045 | 3755 | if (!cstate->base.active) |
407b50f3 DL |
3756 | return 0; |
3757 | ||
30d1b5fe PZ |
3758 | pixel_rate = ilk_pipe_pixel_rate(cstate); |
3759 | ||
3760 | if (WARN_ON(pixel_rate == 0)) | |
661abfc0 | 3761 | return 0; |
407b50f3 | 3762 | |
024c9045 | 3763 | return DIV_ROUND_UP(8 * cstate->base.adjusted_mode.crtc_htotal * 1000, |
30d1b5fe | 3764 | pixel_rate); |
407b50f3 DL |
3765 | } |
3766 | ||
024c9045 | 3767 | static void skl_compute_transition_wm(struct intel_crtc_state *cstate, |
9414f563 | 3768 | struct skl_wm_level *trans_wm /* out */) |
407b50f3 | 3769 | { |
024c9045 | 3770 | if (!cstate->base.active) |
407b50f3 | 3771 | return; |
9414f563 DL |
3772 | |
3773 | /* Until we know more, just disable transition WMs */ | |
a62163e9 | 3774 | trans_wm->plane_en = false; |
407b50f3 DL |
3775 | } |
3776 | ||
55994c2c MR |
3777 | static int skl_build_pipe_wm(struct intel_crtc_state *cstate, |
3778 | struct skl_ddb_allocation *ddb, | |
3779 | struct skl_pipe_wm *pipe_wm) | |
2d41c0b5 | 3780 | { |
024c9045 | 3781 | struct drm_device *dev = cstate->base.crtc->dev; |
fac5e23e | 3782 | const struct drm_i915_private *dev_priv = to_i915(dev); |
a62163e9 L |
3783 | struct intel_plane *intel_plane; |
3784 | struct skl_plane_wm *wm; | |
5db94019 | 3785 | int level, max_level = ilk_wm_max_level(dev_priv); |
55994c2c | 3786 | int ret; |
2d41c0b5 | 3787 | |
a62163e9 L |
3788 | /* |
3789 | * We'll only calculate watermarks for planes that are actually | |
3790 | * enabled, so make sure all other planes are set as disabled. | |
3791 | */ | |
3792 | memset(pipe_wm->planes, 0, sizeof(pipe_wm->planes)); | |
3793 | ||
3794 | for_each_intel_plane_mask(&dev_priv->drm, | |
3795 | intel_plane, | |
3796 | cstate->base.plane_mask) { | |
3797 | wm = &pipe_wm->planes[skl_wm_plane_id(intel_plane)]; | |
3798 | ||
3799 | for (level = 0; level <= max_level; level++) { | |
3800 | ret = skl_compute_wm_level(dev_priv, ddb, cstate, | |
3801 | intel_plane, level, | |
3802 | &wm->wm[level]); | |
3803 | if (ret) | |
3804 | return ret; | |
3805 | } | |
3806 | skl_compute_transition_wm(cstate, &wm->trans_wm); | |
2d41c0b5 | 3807 | } |
024c9045 | 3808 | pipe_wm->linetime = skl_compute_linetime_wm(cstate); |
2d41c0b5 | 3809 | |
55994c2c | 3810 | return 0; |
2d41c0b5 PB |
3811 | } |
3812 | ||
f0f59a00 VS |
3813 | static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, |
3814 | i915_reg_t reg, | |
16160e3d DL |
3815 | const struct skl_ddb_entry *entry) |
3816 | { | |
3817 | if (entry->end) | |
3818 | I915_WRITE(reg, (entry->end - 1) << 16 | entry->start); | |
3819 | else | |
3820 | I915_WRITE(reg, 0); | |
3821 | } | |
3822 | ||
d8c0fafc | 3823 | static void skl_write_wm_level(struct drm_i915_private *dev_priv, |
3824 | i915_reg_t reg, | |
3825 | const struct skl_wm_level *level) | |
3826 | { | |
3827 | uint32_t val = 0; | |
3828 | ||
3829 | if (level->plane_en) { | |
3830 | val |= PLANE_WM_EN; | |
3831 | val |= level->plane_res_b; | |
3832 | val |= level->plane_res_l << PLANE_WM_LINES_SHIFT; | |
3833 | } | |
3834 | ||
3835 | I915_WRITE(reg, val); | |
3836 | } | |
3837 | ||
99101574 VS |
3838 | static void skl_write_plane_wm(struct intel_crtc *intel_crtc, |
3839 | const struct skl_plane_wm *wm, | |
3840 | const struct skl_ddb_allocation *ddb, | |
3841 | int plane) | |
62e0fb88 L |
3842 | { |
3843 | struct drm_crtc *crtc = &intel_crtc->base; | |
3844 | struct drm_device *dev = crtc->dev; | |
3845 | struct drm_i915_private *dev_priv = to_i915(dev); | |
5db94019 | 3846 | int level, max_level = ilk_wm_max_level(dev_priv); |
62e0fb88 L |
3847 | enum pipe pipe = intel_crtc->pipe; |
3848 | ||
3849 | for (level = 0; level <= max_level; level++) { | |
d8c0fafc | 3850 | skl_write_wm_level(dev_priv, PLANE_WM(pipe, plane, level), |
3851 | &wm->wm[level]); | |
62e0fb88 | 3852 | } |
d8c0fafc | 3853 | skl_write_wm_level(dev_priv, PLANE_WM_TRANS(pipe, plane), |
3854 | &wm->trans_wm); | |
27082493 L |
3855 | |
3856 | skl_ddb_entry_write(dev_priv, PLANE_BUF_CFG(pipe, plane), | |
d8c0fafc | 3857 | &ddb->plane[pipe][plane]); |
27082493 | 3858 | skl_ddb_entry_write(dev_priv, PLANE_NV12_BUF_CFG(pipe, plane), |
d8c0fafc | 3859 | &ddb->y_plane[pipe][plane]); |
62e0fb88 L |
3860 | } |
3861 | ||
99101574 VS |
3862 | static void skl_write_cursor_wm(struct intel_crtc *intel_crtc, |
3863 | const struct skl_plane_wm *wm, | |
3864 | const struct skl_ddb_allocation *ddb) | |
62e0fb88 L |
3865 | { |
3866 | struct drm_crtc *crtc = &intel_crtc->base; | |
3867 | struct drm_device *dev = crtc->dev; | |
3868 | struct drm_i915_private *dev_priv = to_i915(dev); | |
5db94019 | 3869 | int level, max_level = ilk_wm_max_level(dev_priv); |
62e0fb88 L |
3870 | enum pipe pipe = intel_crtc->pipe; |
3871 | ||
3872 | for (level = 0; level <= max_level; level++) { | |
d8c0fafc | 3873 | skl_write_wm_level(dev_priv, CUR_WM(pipe, level), |
3874 | &wm->wm[level]); | |
62e0fb88 | 3875 | } |
d8c0fafc | 3876 | skl_write_wm_level(dev_priv, CUR_WM_TRANS(pipe), &wm->trans_wm); |
5d374d96 | 3877 | |
27082493 | 3878 | skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe), |
d8c0fafc | 3879 | &ddb->plane[pipe][PLANE_CURSOR]); |
2d41c0b5 PB |
3880 | } |
3881 | ||
45ece230 | 3882 | bool skl_wm_level_equals(const struct skl_wm_level *l1, |
3883 | const struct skl_wm_level *l2) | |
3884 | { | |
3885 | if (l1->plane_en != l2->plane_en) | |
3886 | return false; | |
3887 | ||
3888 | /* If both planes aren't enabled, the rest shouldn't matter */ | |
3889 | if (!l1->plane_en) | |
3890 | return true; | |
3891 | ||
3892 | return (l1->plane_res_l == l2->plane_res_l && | |
3893 | l1->plane_res_b == l2->plane_res_b); | |
3894 | } | |
3895 | ||
27082493 L |
3896 | static inline bool skl_ddb_entries_overlap(const struct skl_ddb_entry *a, |
3897 | const struct skl_ddb_entry *b) | |
0e8fb7ba | 3898 | { |
27082493 | 3899 | return a->start < b->end && b->start < a->end; |
0e8fb7ba DL |
3900 | } |
3901 | ||
5eff503b ML |
3902 | bool skl_ddb_allocation_overlaps(const struct skl_ddb_entry **entries, |
3903 | const struct skl_ddb_entry *ddb, | |
3904 | int ignore) | |
0e8fb7ba | 3905 | { |
ce0ba283 | 3906 | int i; |
0e8fb7ba | 3907 | |
5eff503b ML |
3908 | for (i = 0; i < I915_MAX_PIPES; i++) |
3909 | if (i != ignore && entries[i] && | |
3910 | skl_ddb_entries_overlap(ddb, entries[i])) | |
27082493 | 3911 | return true; |
0e8fb7ba | 3912 | |
27082493 | 3913 | return false; |
0e8fb7ba DL |
3914 | } |
3915 | ||
55994c2c | 3916 | static int skl_update_pipe_wm(struct drm_crtc_state *cstate, |
03af79e0 | 3917 | const struct skl_pipe_wm *old_pipe_wm, |
55994c2c | 3918 | struct skl_pipe_wm *pipe_wm, /* out */ |
03af79e0 | 3919 | struct skl_ddb_allocation *ddb, /* out */ |
55994c2c | 3920 | bool *changed /* out */) |
2d41c0b5 | 3921 | { |
f4a96752 | 3922 | struct intel_crtc_state *intel_cstate = to_intel_crtc_state(cstate); |
55994c2c | 3923 | int ret; |
2d41c0b5 | 3924 | |
55994c2c MR |
3925 | ret = skl_build_pipe_wm(intel_cstate, ddb, pipe_wm); |
3926 | if (ret) | |
3927 | return ret; | |
2d41c0b5 | 3928 | |
03af79e0 | 3929 | if (!memcmp(old_pipe_wm, pipe_wm, sizeof(*pipe_wm))) |
55994c2c MR |
3930 | *changed = false; |
3931 | else | |
3932 | *changed = true; | |
2d41c0b5 | 3933 | |
55994c2c | 3934 | return 0; |
2d41c0b5 PB |
3935 | } |
3936 | ||
9b613022 MR |
3937 | static uint32_t |
3938 | pipes_modified(struct drm_atomic_state *state) | |
3939 | { | |
3940 | struct drm_crtc *crtc; | |
3941 | struct drm_crtc_state *cstate; | |
3942 | uint32_t i, ret = 0; | |
3943 | ||
3944 | for_each_crtc_in_state(state, crtc, cstate, i) | |
3945 | ret |= drm_crtc_mask(crtc); | |
3946 | ||
3947 | return ret; | |
3948 | } | |
3949 | ||
bb7791bd | 3950 | static int |
7f60e200 PZ |
3951 | skl_ddb_add_affected_planes(struct intel_crtc_state *cstate) |
3952 | { | |
3953 | struct drm_atomic_state *state = cstate->base.state; | |
3954 | struct drm_device *dev = state->dev; | |
3955 | struct drm_crtc *crtc = cstate->base.crtc; | |
3956 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3957 | struct drm_i915_private *dev_priv = to_i915(dev); | |
3958 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
3959 | struct skl_ddb_allocation *new_ddb = &intel_state->wm_results.ddb; | |
3960 | struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb; | |
3961 | struct drm_plane_state *plane_state; | |
3962 | struct drm_plane *plane; | |
3963 | enum pipe pipe = intel_crtc->pipe; | |
3964 | int id; | |
3965 | ||
3966 | WARN_ON(!drm_atomic_get_existing_crtc_state(state, crtc)); | |
3967 | ||
220b0965 | 3968 | drm_for_each_plane_mask(plane, dev, cstate->base.plane_mask) { |
7f60e200 PZ |
3969 | id = skl_wm_plane_id(to_intel_plane(plane)); |
3970 | ||
3971 | if (skl_ddb_entry_equal(&cur_ddb->plane[pipe][id], | |
3972 | &new_ddb->plane[pipe][id]) && | |
3973 | skl_ddb_entry_equal(&cur_ddb->y_plane[pipe][id], | |
3974 | &new_ddb->y_plane[pipe][id])) | |
3975 | continue; | |
3976 | ||
3977 | plane_state = drm_atomic_get_plane_state(state, plane); | |
3978 | if (IS_ERR(plane_state)) | |
3979 | return PTR_ERR(plane_state); | |
3980 | } | |
3981 | ||
3982 | return 0; | |
3983 | } | |
3984 | ||
98d39494 MR |
3985 | static int |
3986 | skl_compute_ddb(struct drm_atomic_state *state) | |
3987 | { | |
3988 | struct drm_device *dev = state->dev; | |
3989 | struct drm_i915_private *dev_priv = to_i915(dev); | |
3990 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); | |
3991 | struct intel_crtc *intel_crtc; | |
734fa01f | 3992 | struct skl_ddb_allocation *ddb = &intel_state->wm_results.ddb; |
9b613022 | 3993 | uint32_t realloc_pipes = pipes_modified(state); |
98d39494 MR |
3994 | int ret; |
3995 | ||
3996 | /* | |
3997 | * If this is our first atomic update following hardware readout, | |
3998 | * we can't trust the DDB that the BIOS programmed for us. Let's | |
3999 | * pretend that all pipes switched active status so that we'll | |
4000 | * ensure a full DDB recompute. | |
4001 | */ | |
1b54a880 MR |
4002 | if (dev_priv->wm.distrust_bios_wm) { |
4003 | ret = drm_modeset_lock(&dev->mode_config.connection_mutex, | |
4004 | state->acquire_ctx); | |
4005 | if (ret) | |
4006 | return ret; | |
4007 | ||
98d39494 MR |
4008 | intel_state->active_pipe_changes = ~0; |
4009 | ||
1b54a880 MR |
4010 | /* |
4011 | * We usually only initialize intel_state->active_crtcs if we | |
4012 | * we're doing a modeset; make sure this field is always | |
4013 | * initialized during the sanitization process that happens | |
4014 | * on the first commit too. | |
4015 | */ | |
4016 | if (!intel_state->modeset) | |
4017 | intel_state->active_crtcs = dev_priv->active_crtcs; | |
4018 | } | |
4019 | ||
98d39494 MR |
4020 | /* |
4021 | * If the modeset changes which CRTC's are active, we need to | |
4022 | * recompute the DDB allocation for *all* active pipes, even | |
4023 | * those that weren't otherwise being modified in any way by this | |
4024 | * atomic commit. Due to the shrinking of the per-pipe allocations | |
4025 | * when new active CRTC's are added, it's possible for a pipe that | |
4026 | * we were already using and aren't changing at all here to suddenly | |
4027 | * become invalid if its DDB needs exceeds its new allocation. | |
4028 | * | |
4029 | * Note that if we wind up doing a full DDB recompute, we can't let | |
4030 | * any other display updates race with this transaction, so we need | |
4031 | * to grab the lock on *all* CRTC's. | |
4032 | */ | |
734fa01f | 4033 | if (intel_state->active_pipe_changes) { |
98d39494 | 4034 | realloc_pipes = ~0; |
734fa01f MR |
4035 | intel_state->wm_results.dirty_pipes = ~0; |
4036 | } | |
98d39494 | 4037 | |
5a920b85 PZ |
4038 | /* |
4039 | * We're not recomputing for the pipes not included in the commit, so | |
4040 | * make sure we start with the current state. | |
4041 | */ | |
4042 | memcpy(ddb, &dev_priv->wm.skl_hw.ddb, sizeof(*ddb)); | |
4043 | ||
98d39494 MR |
4044 | for_each_intel_crtc_mask(dev, intel_crtc, realloc_pipes) { |
4045 | struct intel_crtc_state *cstate; | |
4046 | ||
4047 | cstate = intel_atomic_get_crtc_state(state, intel_crtc); | |
4048 | if (IS_ERR(cstate)) | |
4049 | return PTR_ERR(cstate); | |
4050 | ||
734fa01f | 4051 | ret = skl_allocate_pipe_ddb(cstate, ddb); |
98d39494 MR |
4052 | if (ret) |
4053 | return ret; | |
05a76d3d | 4054 | |
7f60e200 | 4055 | ret = skl_ddb_add_affected_planes(cstate); |
05a76d3d L |
4056 | if (ret) |
4057 | return ret; | |
98d39494 MR |
4058 | } |
4059 | ||
4060 | return 0; | |
4061 | } | |
4062 | ||
2722efb9 MR |
4063 | static void |
4064 | skl_copy_wm_for_pipe(struct skl_wm_values *dst, | |
4065 | struct skl_wm_values *src, | |
4066 | enum pipe pipe) | |
4067 | { | |
2722efb9 MR |
4068 | memcpy(dst->ddb.y_plane[pipe], src->ddb.y_plane[pipe], |
4069 | sizeof(dst->ddb.y_plane[pipe])); | |
4070 | memcpy(dst->ddb.plane[pipe], src->ddb.plane[pipe], | |
4071 | sizeof(dst->ddb.plane[pipe])); | |
4072 | } | |
4073 | ||
413fc530 | 4074 | static void |
4075 | skl_print_wm_changes(const struct drm_atomic_state *state) | |
4076 | { | |
4077 | const struct drm_device *dev = state->dev; | |
4078 | const struct drm_i915_private *dev_priv = to_i915(dev); | |
4079 | const struct intel_atomic_state *intel_state = | |
4080 | to_intel_atomic_state(state); | |
4081 | const struct drm_crtc *crtc; | |
4082 | const struct drm_crtc_state *cstate; | |
413fc530 | 4083 | const struct intel_plane *intel_plane; |
413fc530 | 4084 | const struct skl_ddb_allocation *old_ddb = &dev_priv->wm.skl_hw.ddb; |
4085 | const struct skl_ddb_allocation *new_ddb = &intel_state->wm_results.ddb; | |
413fc530 | 4086 | int id; |
7570498e | 4087 | int i; |
413fc530 | 4088 | |
4089 | for_each_crtc_in_state(state, crtc, cstate, i) { | |
7570498e ML |
4090 | const struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
4091 | enum pipe pipe = intel_crtc->pipe; | |
413fc530 | 4092 | |
7570498e | 4093 | for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) { |
413fc530 | 4094 | const struct skl_ddb_entry *old, *new; |
4095 | ||
413fc530 | 4096 | id = skl_wm_plane_id(intel_plane); |
4097 | old = &old_ddb->plane[pipe][id]; | |
4098 | new = &new_ddb->plane[pipe][id]; | |
4099 | ||
413fc530 | 4100 | if (skl_ddb_entry_equal(old, new)) |
4101 | continue; | |
4102 | ||
7570498e ML |
4103 | DRM_DEBUG_ATOMIC("[PLANE:%d:%s] ddb (%d - %d) -> (%d - %d)\n", |
4104 | intel_plane->base.base.id, | |
4105 | intel_plane->base.name, | |
4106 | old->start, old->end, | |
4107 | new->start, new->end); | |
413fc530 | 4108 | } |
4109 | } | |
4110 | } | |
4111 | ||
98d39494 MR |
4112 | static int |
4113 | skl_compute_wm(struct drm_atomic_state *state) | |
4114 | { | |
4115 | struct drm_crtc *crtc; | |
4116 | struct drm_crtc_state *cstate; | |
734fa01f MR |
4117 | struct intel_atomic_state *intel_state = to_intel_atomic_state(state); |
4118 | struct skl_wm_values *results = &intel_state->wm_results; | |
4119 | struct skl_pipe_wm *pipe_wm; | |
98d39494 | 4120 | bool changed = false; |
734fa01f | 4121 | int ret, i; |
98d39494 MR |
4122 | |
4123 | /* | |
4124 | * If this transaction isn't actually touching any CRTC's, don't | |
4125 | * bother with watermark calculation. Note that if we pass this | |
4126 | * test, we're guaranteed to hold at least one CRTC state mutex, | |
4127 | * which means we can safely use values like dev_priv->active_crtcs | |
4128 | * since any racing commits that want to update them would need to | |
4129 | * hold _all_ CRTC state mutexes. | |
4130 | */ | |
4131 | for_each_crtc_in_state(state, crtc, cstate, i) | |
4132 | changed = true; | |
4133 | if (!changed) | |
4134 | return 0; | |
4135 | ||
734fa01f MR |
4136 | /* Clear all dirty flags */ |
4137 | results->dirty_pipes = 0; | |
4138 | ||
98d39494 MR |
4139 | ret = skl_compute_ddb(state); |
4140 | if (ret) | |
4141 | return ret; | |
4142 | ||
734fa01f MR |
4143 | /* |
4144 | * Calculate WM's for all pipes that are part of this transaction. | |
4145 | * Note that the DDB allocation above may have added more CRTC's that | |
4146 | * weren't otherwise being modified (and set bits in dirty_pipes) if | |
4147 | * pipe allocations had to change. | |
4148 | * | |
4149 | * FIXME: Now that we're doing this in the atomic check phase, we | |
4150 | * should allow skl_update_pipe_wm() to return failure in cases where | |
4151 | * no suitable watermark values can be found. | |
4152 | */ | |
4153 | for_each_crtc_in_state(state, crtc, cstate, i) { | |
734fa01f MR |
4154 | struct intel_crtc_state *intel_cstate = |
4155 | to_intel_crtc_state(cstate); | |
03af79e0 ML |
4156 | const struct skl_pipe_wm *old_pipe_wm = |
4157 | &to_intel_crtc_state(crtc->state)->wm.skl.optimal; | |
734fa01f MR |
4158 | |
4159 | pipe_wm = &intel_cstate->wm.skl.optimal; | |
03af79e0 ML |
4160 | ret = skl_update_pipe_wm(cstate, old_pipe_wm, pipe_wm, |
4161 | &results->ddb, &changed); | |
734fa01f MR |
4162 | if (ret) |
4163 | return ret; | |
4164 | ||
4165 | if (changed) | |
4166 | results->dirty_pipes |= drm_crtc_mask(crtc); | |
4167 | ||
4168 | if ((results->dirty_pipes & drm_crtc_mask(crtc)) == 0) | |
4169 | /* This pipe's WM's did not change */ | |
4170 | continue; | |
4171 | ||
4172 | intel_cstate->update_wm_pre = true; | |
734fa01f MR |
4173 | } |
4174 | ||
413fc530 | 4175 | skl_print_wm_changes(state); |
4176 | ||
98d39494 MR |
4177 | return 0; |
4178 | } | |
4179 | ||
ccf010fb ML |
4180 | static void skl_atomic_update_crtc_wm(struct intel_atomic_state *state, |
4181 | struct intel_crtc_state *cstate) | |
4182 | { | |
4183 | struct intel_crtc *crtc = to_intel_crtc(cstate->base.crtc); | |
4184 | struct drm_i915_private *dev_priv = to_i915(state->base.dev); | |
4185 | struct skl_pipe_wm *pipe_wm = &cstate->wm.skl.optimal; | |
e62929b3 | 4186 | const struct skl_ddb_allocation *ddb = &state->wm_results.ddb; |
ccf010fb | 4187 | enum pipe pipe = crtc->pipe; |
e62929b3 ML |
4188 | int plane; |
4189 | ||
4190 | if (!(state->wm_results.dirty_pipes & drm_crtc_mask(&crtc->base))) | |
4191 | return; | |
ccf010fb ML |
4192 | |
4193 | I915_WRITE(PIPE_WM_LINETIME(pipe), pipe_wm->linetime); | |
e62929b3 ML |
4194 | |
4195 | for_each_universal_plane(dev_priv, pipe, plane) | |
4196 | skl_write_plane_wm(crtc, &pipe_wm->planes[plane], ddb, plane); | |
4197 | ||
4198 | skl_write_cursor_wm(crtc, &pipe_wm->planes[PLANE_CURSOR], ddb); | |
ccf010fb ML |
4199 | } |
4200 | ||
e62929b3 ML |
4201 | static void skl_initial_wm(struct intel_atomic_state *state, |
4202 | struct intel_crtc_state *cstate) | |
2d41c0b5 | 4203 | { |
e62929b3 | 4204 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); |
432081bc | 4205 | struct drm_device *dev = intel_crtc->base.dev; |
fac5e23e | 4206 | struct drm_i915_private *dev_priv = to_i915(dev); |
e62929b3 | 4207 | struct skl_wm_values *results = &state->wm_results; |
2722efb9 | 4208 | struct skl_wm_values *hw_vals = &dev_priv->wm.skl_hw; |
27082493 | 4209 | enum pipe pipe = intel_crtc->pipe; |
adda50b8 | 4210 | |
432081bc | 4211 | if ((results->dirty_pipes & drm_crtc_mask(&intel_crtc->base)) == 0) |
2d41c0b5 PB |
4212 | return; |
4213 | ||
734fa01f | 4214 | mutex_lock(&dev_priv->wm.wm_mutex); |
2d41c0b5 | 4215 | |
e62929b3 ML |
4216 | if (cstate->base.active_changed) |
4217 | skl_atomic_update_crtc_wm(state, cstate); | |
27082493 L |
4218 | |
4219 | skl_copy_wm_for_pipe(hw_vals, results, pipe); | |
734fa01f MR |
4220 | |
4221 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
2d41c0b5 PB |
4222 | } |
4223 | ||
d890565c VS |
4224 | static void ilk_compute_wm_config(struct drm_device *dev, |
4225 | struct intel_wm_config *config) | |
4226 | { | |
4227 | struct intel_crtc *crtc; | |
4228 | ||
4229 | /* Compute the currently _active_ config */ | |
4230 | for_each_intel_crtc(dev, crtc) { | |
4231 | const struct intel_pipe_wm *wm = &crtc->wm.active.ilk; | |
4232 | ||
4233 | if (!wm->pipe_enabled) | |
4234 | continue; | |
4235 | ||
4236 | config->sprites_enabled |= wm->sprites_enabled; | |
4237 | config->sprites_scaled |= wm->sprites_scaled; | |
4238 | config->num_pipes_active++; | |
4239 | } | |
4240 | } | |
4241 | ||
ed4a6a7c | 4242 | static void ilk_program_watermarks(struct drm_i915_private *dev_priv) |
801bcfff | 4243 | { |
91c8a326 | 4244 | struct drm_device *dev = &dev_priv->drm; |
b9d5c839 | 4245 | struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm; |
820c1980 | 4246 | struct ilk_wm_maximums max; |
d890565c | 4247 | struct intel_wm_config config = {}; |
820c1980 | 4248 | struct ilk_wm_values results = {}; |
77c122bc | 4249 | enum intel_ddb_partitioning partitioning; |
261a27d1 | 4250 | |
d890565c VS |
4251 | ilk_compute_wm_config(dev, &config); |
4252 | ||
4253 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max); | |
4254 | ilk_wm_merge(dev, &config, &max, &lp_wm_1_2); | |
a485bfb8 VS |
4255 | |
4256 | /* 5/6 split only in single pipe config on IVB+ */ | |
175fded1 | 4257 | if (INTEL_GEN(dev_priv) >= 7 && |
d890565c VS |
4258 | config.num_pipes_active == 1 && config.sprites_enabled) { |
4259 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max); | |
4260 | ilk_wm_merge(dev, &config, &max, &lp_wm_5_6); | |
0362c781 | 4261 | |
820c1980 | 4262 | best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6); |
861f3389 | 4263 | } else { |
198a1e9b | 4264 | best_lp_wm = &lp_wm_1_2; |
861f3389 PZ |
4265 | } |
4266 | ||
198a1e9b | 4267 | partitioning = (best_lp_wm == &lp_wm_1_2) ? |
77c122bc | 4268 | INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6; |
801bcfff | 4269 | |
820c1980 | 4270 | ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results); |
609cedef | 4271 | |
820c1980 | 4272 | ilk_write_wm_values(dev_priv, &results); |
1011d8c4 PZ |
4273 | } |
4274 | ||
ccf010fb ML |
4275 | static void ilk_initial_watermarks(struct intel_atomic_state *state, |
4276 | struct intel_crtc_state *cstate) | |
b9d5c839 | 4277 | { |
ed4a6a7c MR |
4278 | struct drm_i915_private *dev_priv = to_i915(cstate->base.crtc->dev); |
4279 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); | |
b9d5c839 | 4280 | |
ed4a6a7c | 4281 | mutex_lock(&dev_priv->wm.wm_mutex); |
e8f1f02e | 4282 | intel_crtc->wm.active.ilk = cstate->wm.ilk.intermediate; |
ed4a6a7c MR |
4283 | ilk_program_watermarks(dev_priv); |
4284 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
4285 | } | |
bf220452 | 4286 | |
ccf010fb ML |
4287 | static void ilk_optimize_watermarks(struct intel_atomic_state *state, |
4288 | struct intel_crtc_state *cstate) | |
ed4a6a7c MR |
4289 | { |
4290 | struct drm_i915_private *dev_priv = to_i915(cstate->base.crtc->dev); | |
4291 | struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc); | |
bf220452 | 4292 | |
ed4a6a7c MR |
4293 | mutex_lock(&dev_priv->wm.wm_mutex); |
4294 | if (cstate->wm.need_postvbl_update) { | |
e8f1f02e | 4295 | intel_crtc->wm.active.ilk = cstate->wm.ilk.optimal; |
ed4a6a7c MR |
4296 | ilk_program_watermarks(dev_priv); |
4297 | } | |
4298 | mutex_unlock(&dev_priv->wm.wm_mutex); | |
b9d5c839 VS |
4299 | } |
4300 | ||
d8c0fafc | 4301 | static inline void skl_wm_level_from_reg_val(uint32_t val, |
4302 | struct skl_wm_level *level) | |
3078999f | 4303 | { |
d8c0fafc | 4304 | level->plane_en = val & PLANE_WM_EN; |
4305 | level->plane_res_b = val & PLANE_WM_BLOCKS_MASK; | |
4306 | level->plane_res_l = (val >> PLANE_WM_LINES_SHIFT) & | |
4307 | PLANE_WM_LINES_MASK; | |
3078999f PB |
4308 | } |
4309 | ||
bf9d99ad | 4310 | void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc, |
4311 | struct skl_pipe_wm *out) | |
3078999f PB |
4312 | { |
4313 | struct drm_device *dev = crtc->dev; | |
fac5e23e | 4314 | struct drm_i915_private *dev_priv = to_i915(dev); |
3078999f | 4315 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
d8c0fafc | 4316 | struct intel_plane *intel_plane; |
d8c0fafc | 4317 | struct skl_plane_wm *wm; |
3078999f | 4318 | enum pipe pipe = intel_crtc->pipe; |
d8c0fafc | 4319 | int level, id, max_level; |
4320 | uint32_t val; | |
3078999f | 4321 | |
5db94019 | 4322 | max_level = ilk_wm_max_level(dev_priv); |
3078999f | 4323 | |
d8c0fafc | 4324 | for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) { |
4325 | id = skl_wm_plane_id(intel_plane); | |
bf9d99ad | 4326 | wm = &out->planes[id]; |
3078999f | 4327 | |
d8c0fafc | 4328 | for (level = 0; level <= max_level; level++) { |
4329 | if (id != PLANE_CURSOR) | |
4330 | val = I915_READ(PLANE_WM(pipe, id, level)); | |
4331 | else | |
4332 | val = I915_READ(CUR_WM(pipe, level)); | |
3078999f | 4333 | |
d8c0fafc | 4334 | skl_wm_level_from_reg_val(val, &wm->wm[level]); |
3078999f | 4335 | } |
3078999f | 4336 | |
d8c0fafc | 4337 | if (id != PLANE_CURSOR) |
4338 | val = I915_READ(PLANE_WM_TRANS(pipe, id)); | |
4339 | else | |
4340 | val = I915_READ(CUR_WM_TRANS(pipe)); | |
4341 | ||
4342 | skl_wm_level_from_reg_val(val, &wm->trans_wm); | |
3078999f PB |
4343 | } |
4344 | ||
d8c0fafc | 4345 | if (!intel_crtc->active) |
4346 | return; | |
4e0963c7 | 4347 | |
bf9d99ad | 4348 | out->linetime = I915_READ(PIPE_WM_LINETIME(pipe)); |
3078999f PB |
4349 | } |
4350 | ||
4351 | void skl_wm_get_hw_state(struct drm_device *dev) | |
4352 | { | |
fac5e23e | 4353 | struct drm_i915_private *dev_priv = to_i915(dev); |
bf9d99ad | 4354 | struct skl_wm_values *hw = &dev_priv->wm.skl_hw; |
a269c583 | 4355 | struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb; |
3078999f | 4356 | struct drm_crtc *crtc; |
bf9d99ad | 4357 | struct intel_crtc *intel_crtc; |
4358 | struct intel_crtc_state *cstate; | |
3078999f | 4359 | |
a269c583 | 4360 | skl_ddb_get_hw_state(dev_priv, ddb); |
bf9d99ad | 4361 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { |
4362 | intel_crtc = to_intel_crtc(crtc); | |
4363 | cstate = to_intel_crtc_state(crtc->state); | |
4364 | ||
4365 | skl_pipe_wm_get_hw_state(crtc, &cstate->wm.skl.optimal); | |
4366 | ||
03af79e0 | 4367 | if (intel_crtc->active) |
bf9d99ad | 4368 | hw->dirty_pipes |= drm_crtc_mask(crtc); |
bf9d99ad | 4369 | } |
a1de91e5 | 4370 | |
279e99d7 MR |
4371 | if (dev_priv->active_crtcs) { |
4372 | /* Fully recompute DDB on first atomic commit */ | |
4373 | dev_priv->wm.distrust_bios_wm = true; | |
4374 | } else { | |
4375 | /* Easy/common case; just sanitize DDB now if everything off */ | |
4376 | memset(ddb, 0, sizeof(*ddb)); | |
4377 | } | |
3078999f PB |
4378 | } |
4379 | ||
243e6a44 VS |
4380 | static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
4381 | { | |
4382 | struct drm_device *dev = crtc->dev; | |
fac5e23e | 4383 | struct drm_i915_private *dev_priv = to_i915(dev); |
820c1980 | 4384 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
243e6a44 | 4385 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
4e0963c7 | 4386 | struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state); |
e8f1f02e | 4387 | struct intel_pipe_wm *active = &cstate->wm.ilk.optimal; |
243e6a44 | 4388 | enum pipe pipe = intel_crtc->pipe; |
f0f59a00 | 4389 | static const i915_reg_t wm0_pipe_reg[] = { |
243e6a44 VS |
4390 | [PIPE_A] = WM0_PIPEA_ILK, |
4391 | [PIPE_B] = WM0_PIPEB_ILK, | |
4392 | [PIPE_C] = WM0_PIPEC_IVB, | |
4393 | }; | |
4394 | ||
4395 | hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]); | |
8652744b | 4396 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
ce0e0713 | 4397 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
243e6a44 | 4398 | |
15606534 VS |
4399 | memset(active, 0, sizeof(*active)); |
4400 | ||
3ef00284 | 4401 | active->pipe_enabled = intel_crtc->active; |
2a44b76b VS |
4402 | |
4403 | if (active->pipe_enabled) { | |
243e6a44 VS |
4404 | u32 tmp = hw->wm_pipe[pipe]; |
4405 | ||
4406 | /* | |
4407 | * For active pipes LP0 watermark is marked as | |
4408 | * enabled, and LP1+ watermaks as disabled since | |
4409 | * we can't really reverse compute them in case | |
4410 | * multiple pipes are active. | |
4411 | */ | |
4412 | active->wm[0].enable = true; | |
4413 | active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT; | |
4414 | active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT; | |
4415 | active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK; | |
4416 | active->linetime = hw->wm_linetime[pipe]; | |
4417 | } else { | |
5db94019 | 4418 | int level, max_level = ilk_wm_max_level(dev_priv); |
243e6a44 VS |
4419 | |
4420 | /* | |
4421 | * For inactive pipes, all watermark levels | |
4422 | * should be marked as enabled but zeroed, | |
4423 | * which is what we'd compute them to. | |
4424 | */ | |
4425 | for (level = 0; level <= max_level; level++) | |
4426 | active->wm[level].enable = true; | |
4427 | } | |
4e0963c7 MR |
4428 | |
4429 | intel_crtc->wm.active.ilk = *active; | |
243e6a44 VS |
4430 | } |
4431 | ||
6eb1a681 VS |
4432 | #define _FW_WM(value, plane) \ |
4433 | (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT) | |
4434 | #define _FW_WM_VLV(value, plane) \ | |
4435 | (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT) | |
4436 | ||
4437 | static void vlv_read_wm_values(struct drm_i915_private *dev_priv, | |
4438 | struct vlv_wm_values *wm) | |
4439 | { | |
4440 | enum pipe pipe; | |
4441 | uint32_t tmp; | |
4442 | ||
4443 | for_each_pipe(dev_priv, pipe) { | |
4444 | tmp = I915_READ(VLV_DDL(pipe)); | |
4445 | ||
4446 | wm->ddl[pipe].primary = | |
4447 | (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
4448 | wm->ddl[pipe].cursor = | |
4449 | (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
4450 | wm->ddl[pipe].sprite[0] = | |
4451 | (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
4452 | wm->ddl[pipe].sprite[1] = | |
4453 | (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
4454 | } | |
4455 | ||
4456 | tmp = I915_READ(DSPFW1); | |
4457 | wm->sr.plane = _FW_WM(tmp, SR); | |
4458 | wm->pipe[PIPE_B].cursor = _FW_WM(tmp, CURSORB); | |
4459 | wm->pipe[PIPE_B].primary = _FW_WM_VLV(tmp, PLANEB); | |
4460 | wm->pipe[PIPE_A].primary = _FW_WM_VLV(tmp, PLANEA); | |
4461 | ||
4462 | tmp = I915_READ(DSPFW2); | |
4463 | wm->pipe[PIPE_A].sprite[1] = _FW_WM_VLV(tmp, SPRITEB); | |
4464 | wm->pipe[PIPE_A].cursor = _FW_WM(tmp, CURSORA); | |
4465 | wm->pipe[PIPE_A].sprite[0] = _FW_WM_VLV(tmp, SPRITEA); | |
4466 | ||
4467 | tmp = I915_READ(DSPFW3); | |
4468 | wm->sr.cursor = _FW_WM(tmp, CURSOR_SR); | |
4469 | ||
4470 | if (IS_CHERRYVIEW(dev_priv)) { | |
4471 | tmp = I915_READ(DSPFW7_CHV); | |
4472 | wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED); | |
4473 | wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC); | |
4474 | ||
4475 | tmp = I915_READ(DSPFW8_CHV); | |
4476 | wm->pipe[PIPE_C].sprite[1] = _FW_WM_VLV(tmp, SPRITEF); | |
4477 | wm->pipe[PIPE_C].sprite[0] = _FW_WM_VLV(tmp, SPRITEE); | |
4478 | ||
4479 | tmp = I915_READ(DSPFW9_CHV); | |
4480 | wm->pipe[PIPE_C].primary = _FW_WM_VLV(tmp, PLANEC); | |
4481 | wm->pipe[PIPE_C].cursor = _FW_WM(tmp, CURSORC); | |
4482 | ||
4483 | tmp = I915_READ(DSPHOWM); | |
4484 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; | |
4485 | wm->pipe[PIPE_C].sprite[1] |= _FW_WM(tmp, SPRITEF_HI) << 8; | |
4486 | wm->pipe[PIPE_C].sprite[0] |= _FW_WM(tmp, SPRITEE_HI) << 8; | |
4487 | wm->pipe[PIPE_C].primary |= _FW_WM(tmp, PLANEC_HI) << 8; | |
4488 | wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8; | |
4489 | wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8; | |
4490 | wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8; | |
4491 | wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8; | |
4492 | wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8; | |
4493 | wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8; | |
4494 | } else { | |
4495 | tmp = I915_READ(DSPFW7); | |
4496 | wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED); | |
4497 | wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC); | |
4498 | ||
4499 | tmp = I915_READ(DSPHOWM); | |
4500 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; | |
4501 | wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8; | |
4502 | wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8; | |
4503 | wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8; | |
4504 | wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8; | |
4505 | wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8; | |
4506 | wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8; | |
4507 | } | |
4508 | } | |
4509 | ||
4510 | #undef _FW_WM | |
4511 | #undef _FW_WM_VLV | |
4512 | ||
4513 | void vlv_wm_get_hw_state(struct drm_device *dev) | |
4514 | { | |
4515 | struct drm_i915_private *dev_priv = to_i915(dev); | |
4516 | struct vlv_wm_values *wm = &dev_priv->wm.vlv; | |
4517 | struct intel_plane *plane; | |
4518 | enum pipe pipe; | |
4519 | u32 val; | |
4520 | ||
4521 | vlv_read_wm_values(dev_priv, wm); | |
4522 | ||
4523 | for_each_intel_plane(dev, plane) { | |
4524 | switch (plane->base.type) { | |
4525 | int sprite; | |
4526 | case DRM_PLANE_TYPE_CURSOR: | |
4527 | plane->wm.fifo_size = 63; | |
4528 | break; | |
4529 | case DRM_PLANE_TYPE_PRIMARY: | |
ef0f5e93 | 4530 | plane->wm.fifo_size = vlv_get_fifo_size(dev_priv, plane->pipe, 0); |
6eb1a681 VS |
4531 | break; |
4532 | case DRM_PLANE_TYPE_OVERLAY: | |
4533 | sprite = plane->plane; | |
ef0f5e93 | 4534 | plane->wm.fifo_size = vlv_get_fifo_size(dev_priv, plane->pipe, sprite + 1); |
6eb1a681 VS |
4535 | break; |
4536 | } | |
4537 | } | |
4538 | ||
4539 | wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN; | |
4540 | wm->level = VLV_WM_LEVEL_PM2; | |
4541 | ||
4542 | if (IS_CHERRYVIEW(dev_priv)) { | |
4543 | mutex_lock(&dev_priv->rps.hw_lock); | |
4544 | ||
4545 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); | |
4546 | if (val & DSP_MAXFIFO_PM5_ENABLE) | |
4547 | wm->level = VLV_WM_LEVEL_PM5; | |
4548 | ||
58590c14 VS |
4549 | /* |
4550 | * If DDR DVFS is disabled in the BIOS, Punit | |
4551 | * will never ack the request. So if that happens | |
4552 | * assume we don't have to enable/disable DDR DVFS | |
4553 | * dynamically. To test that just set the REQ_ACK | |
4554 | * bit to poke the Punit, but don't change the | |
4555 | * HIGH/LOW bits so that we don't actually change | |
4556 | * the current state. | |
4557 | */ | |
6eb1a681 | 4558 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); |
58590c14 VS |
4559 | val |= FORCE_DDR_FREQ_REQ_ACK; |
4560 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); | |
4561 | ||
4562 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & | |
4563 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) { | |
4564 | DRM_DEBUG_KMS("Punit not acking DDR DVFS request, " | |
4565 | "assuming DDR DVFS is disabled\n"); | |
4566 | dev_priv->wm.max_level = VLV_WM_LEVEL_PM5; | |
4567 | } else { | |
4568 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); | |
4569 | if ((val & FORCE_DDR_HIGH_FREQ) == 0) | |
4570 | wm->level = VLV_WM_LEVEL_DDR_DVFS; | |
4571 | } | |
6eb1a681 VS |
4572 | |
4573 | mutex_unlock(&dev_priv->rps.hw_lock); | |
4574 | } | |
4575 | ||
4576 | for_each_pipe(dev_priv, pipe) | |
4577 | DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n", | |
4578 | pipe_name(pipe), wm->pipe[pipe].primary, wm->pipe[pipe].cursor, | |
4579 | wm->pipe[pipe].sprite[0], wm->pipe[pipe].sprite[1]); | |
4580 | ||
4581 | DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n", | |
4582 | wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr); | |
4583 | } | |
4584 | ||
243e6a44 VS |
4585 | void ilk_wm_get_hw_state(struct drm_device *dev) |
4586 | { | |
fac5e23e | 4587 | struct drm_i915_private *dev_priv = to_i915(dev); |
820c1980 | 4588 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
243e6a44 VS |
4589 | struct drm_crtc *crtc; |
4590 | ||
70e1e0ec | 4591 | for_each_crtc(dev, crtc) |
243e6a44 VS |
4592 | ilk_pipe_wm_get_hw_state(crtc); |
4593 | ||
4594 | hw->wm_lp[0] = I915_READ(WM1_LP_ILK); | |
4595 | hw->wm_lp[1] = I915_READ(WM2_LP_ILK); | |
4596 | hw->wm_lp[2] = I915_READ(WM3_LP_ILK); | |
4597 | ||
4598 | hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK); | |
175fded1 | 4599 | if (INTEL_GEN(dev_priv) >= 7) { |
cfa7698b VS |
4600 | hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB); |
4601 | hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB); | |
4602 | } | |
243e6a44 | 4603 | |
8652744b | 4604 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
ac9545fd VS |
4605 | hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ? |
4606 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; | |
fd6b8f43 | 4607 | else if (IS_IVYBRIDGE(dev_priv)) |
ac9545fd VS |
4608 | hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ? |
4609 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; | |
243e6a44 VS |
4610 | |
4611 | hw->enable_fbc_wm = | |
4612 | !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS); | |
4613 | } | |
4614 | ||
b445e3b0 ED |
4615 | /** |
4616 | * intel_update_watermarks - update FIFO watermark values based on current modes | |
4617 | * | |
4618 | * Calculate watermark values for the various WM regs based on current mode | |
4619 | * and plane configuration. | |
4620 | * | |
4621 | * There are several cases to deal with here: | |
4622 | * - normal (i.e. non-self-refresh) | |
4623 | * - self-refresh (SR) mode | |
4624 | * - lines are large relative to FIFO size (buffer can hold up to 2) | |
4625 | * - lines are small relative to FIFO size (buffer can hold more than 2 | |
4626 | * lines), so need to account for TLB latency | |
4627 | * | |
4628 | * The normal calculation is: | |
4629 | * watermark = dotclock * bytes per pixel * latency | |
4630 | * where latency is platform & configuration dependent (we assume pessimal | |
4631 | * values here). | |
4632 | * | |
4633 | * The SR calculation is: | |
4634 | * watermark = (trunc(latency/line time)+1) * surface width * | |
4635 | * bytes per pixel | |
4636 | * where | |
4637 | * line time = htotal / dotclock | |
4638 | * surface width = hdisplay for normal plane and 64 for cursor | |
4639 | * and latency is assumed to be high, as above. | |
4640 | * | |
4641 | * The final value programmed to the register should always be rounded up, | |
4642 | * and include an extra 2 entries to account for clock crossings. | |
4643 | * | |
4644 | * We don't use the sprite, so we can ignore that. And on Crestline we have | |
4645 | * to set the non-SR watermarks to 8. | |
4646 | */ | |
432081bc | 4647 | void intel_update_watermarks(struct intel_crtc *crtc) |
b445e3b0 | 4648 | { |
432081bc | 4649 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); |
b445e3b0 ED |
4650 | |
4651 | if (dev_priv->display.update_wm) | |
46ba614c | 4652 | dev_priv->display.update_wm(crtc); |
b445e3b0 ED |
4653 | } |
4654 | ||
e2828914 | 4655 | /* |
9270388e | 4656 | * Lock protecting IPS related data structures |
9270388e DV |
4657 | */ |
4658 | DEFINE_SPINLOCK(mchdev_lock); | |
4659 | ||
4660 | /* Global for IPS driver to get at the current i915 device. Protected by | |
4661 | * mchdev_lock. */ | |
4662 | static struct drm_i915_private *i915_mch_dev; | |
4663 | ||
91d14251 | 4664 | bool ironlake_set_drps(struct drm_i915_private *dev_priv, u8 val) |
2b4e57bd | 4665 | { |
2b4e57bd ED |
4666 | u16 rgvswctl; |
4667 | ||
9270388e DV |
4668 | assert_spin_locked(&mchdev_lock); |
4669 | ||
2b4e57bd ED |
4670 | rgvswctl = I915_READ16(MEMSWCTL); |
4671 | if (rgvswctl & MEMCTL_CMD_STS) { | |
4672 | DRM_DEBUG("gpu busy, RCS change rejected\n"); | |
4673 | return false; /* still busy with another command */ | |
4674 | } | |
4675 | ||
4676 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | | |
4677 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; | |
4678 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
4679 | POSTING_READ16(MEMSWCTL); | |
4680 | ||
4681 | rgvswctl |= MEMCTL_CMD_STS; | |
4682 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
4683 | ||
4684 | return true; | |
4685 | } | |
4686 | ||
91d14251 | 4687 | static void ironlake_enable_drps(struct drm_i915_private *dev_priv) |
2b4e57bd | 4688 | { |
84f1b20f | 4689 | u32 rgvmodectl; |
2b4e57bd ED |
4690 | u8 fmax, fmin, fstart, vstart; |
4691 | ||
9270388e DV |
4692 | spin_lock_irq(&mchdev_lock); |
4693 | ||
84f1b20f TU |
4694 | rgvmodectl = I915_READ(MEMMODECTL); |
4695 | ||
2b4e57bd ED |
4696 | /* Enable temp reporting */ |
4697 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); | |
4698 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); | |
4699 | ||
4700 | /* 100ms RC evaluation intervals */ | |
4701 | I915_WRITE(RCUPEI, 100000); | |
4702 | I915_WRITE(RCDNEI, 100000); | |
4703 | ||
4704 | /* Set max/min thresholds to 90ms and 80ms respectively */ | |
4705 | I915_WRITE(RCBMAXAVG, 90000); | |
4706 | I915_WRITE(RCBMINAVG, 80000); | |
4707 | ||
4708 | I915_WRITE(MEMIHYST, 1); | |
4709 | ||
4710 | /* Set up min, max, and cur for interrupt handling */ | |
4711 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; | |
4712 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); | |
4713 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> | |
4714 | MEMMODE_FSTART_SHIFT; | |
4715 | ||
616847e7 | 4716 | vstart = (I915_READ(PXVFREQ(fstart)) & PXVFREQ_PX_MASK) >> |
2b4e57bd ED |
4717 | PXVFREQ_PX_SHIFT; |
4718 | ||
20e4d407 DV |
4719 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
4720 | dev_priv->ips.fstart = fstart; | |
2b4e57bd | 4721 | |
20e4d407 DV |
4722 | dev_priv->ips.max_delay = fstart; |
4723 | dev_priv->ips.min_delay = fmin; | |
4724 | dev_priv->ips.cur_delay = fstart; | |
2b4e57bd ED |
4725 | |
4726 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", | |
4727 | fmax, fmin, fstart); | |
4728 | ||
4729 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); | |
4730 | ||
4731 | /* | |
4732 | * Interrupts will be enabled in ironlake_irq_postinstall | |
4733 | */ | |
4734 | ||
4735 | I915_WRITE(VIDSTART, vstart); | |
4736 | POSTING_READ(VIDSTART); | |
4737 | ||
4738 | rgvmodectl |= MEMMODE_SWMODE_EN; | |
4739 | I915_WRITE(MEMMODECTL, rgvmodectl); | |
4740 | ||
9270388e | 4741 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
2b4e57bd | 4742 | DRM_ERROR("stuck trying to change perf mode\n"); |
dd92d8de | 4743 | mdelay(1); |
2b4e57bd | 4744 | |
91d14251 | 4745 | ironlake_set_drps(dev_priv, fstart); |
2b4e57bd | 4746 | |
7d81c3e0 VS |
4747 | dev_priv->ips.last_count1 = I915_READ(DMIEC) + |
4748 | I915_READ(DDREC) + I915_READ(CSIEC); | |
20e4d407 | 4749 | dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies); |
7d81c3e0 | 4750 | dev_priv->ips.last_count2 = I915_READ(GFXEC); |
5ed0bdf2 | 4751 | dev_priv->ips.last_time2 = ktime_get_raw_ns(); |
9270388e DV |
4752 | |
4753 | spin_unlock_irq(&mchdev_lock); | |
2b4e57bd ED |
4754 | } |
4755 | ||
91d14251 | 4756 | static void ironlake_disable_drps(struct drm_i915_private *dev_priv) |
2b4e57bd | 4757 | { |
9270388e DV |
4758 | u16 rgvswctl; |
4759 | ||
4760 | spin_lock_irq(&mchdev_lock); | |
4761 | ||
4762 | rgvswctl = I915_READ16(MEMSWCTL); | |
2b4e57bd ED |
4763 | |
4764 | /* Ack interrupts, disable EFC interrupt */ | |
4765 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); | |
4766 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); | |
4767 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); | |
4768 | I915_WRITE(DEIIR, DE_PCU_EVENT); | |
4769 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); | |
4770 | ||
4771 | /* Go back to the starting frequency */ | |
91d14251 | 4772 | ironlake_set_drps(dev_priv, dev_priv->ips.fstart); |
dd92d8de | 4773 | mdelay(1); |
2b4e57bd ED |
4774 | rgvswctl |= MEMCTL_CMD_STS; |
4775 | I915_WRITE(MEMSWCTL, rgvswctl); | |
dd92d8de | 4776 | mdelay(1); |
2b4e57bd | 4777 | |
9270388e | 4778 | spin_unlock_irq(&mchdev_lock); |
2b4e57bd ED |
4779 | } |
4780 | ||
acbe9475 DV |
4781 | /* There's a funny hw issue where the hw returns all 0 when reading from |
4782 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value | |
4783 | * ourselves, instead of doing a rmw cycle (which might result in us clearing | |
4784 | * all limits and the gpu stuck at whatever frequency it is at atm). | |
4785 | */ | |
74ef1173 | 4786 | static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val) |
2b4e57bd | 4787 | { |
7b9e0ae6 | 4788 | u32 limits; |
2b4e57bd | 4789 | |
20b46e59 DV |
4790 | /* Only set the down limit when we've reached the lowest level to avoid |
4791 | * getting more interrupts, otherwise leave this clear. This prevents a | |
4792 | * race in the hw when coming out of rc6: There's a tiny window where | |
4793 | * the hw runs at the minimal clock before selecting the desired | |
4794 | * frequency, if the down threshold expires in that window we will not | |
4795 | * receive a down interrupt. */ | |
2d1fe073 | 4796 | if (IS_GEN9(dev_priv)) { |
74ef1173 AG |
4797 | limits = (dev_priv->rps.max_freq_softlimit) << 23; |
4798 | if (val <= dev_priv->rps.min_freq_softlimit) | |
4799 | limits |= (dev_priv->rps.min_freq_softlimit) << 14; | |
4800 | } else { | |
4801 | limits = dev_priv->rps.max_freq_softlimit << 24; | |
4802 | if (val <= dev_priv->rps.min_freq_softlimit) | |
4803 | limits |= dev_priv->rps.min_freq_softlimit << 16; | |
4804 | } | |
20b46e59 DV |
4805 | |
4806 | return limits; | |
4807 | } | |
4808 | ||
dd75fdc8 CW |
4809 | static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val) |
4810 | { | |
4811 | int new_power; | |
8a586437 AG |
4812 | u32 threshold_up = 0, threshold_down = 0; /* in % */ |
4813 | u32 ei_up = 0, ei_down = 0; | |
dd75fdc8 CW |
4814 | |
4815 | new_power = dev_priv->rps.power; | |
4816 | switch (dev_priv->rps.power) { | |
4817 | case LOW_POWER: | |
a72b5623 CW |
4818 | if (val > dev_priv->rps.efficient_freq + 1 && |
4819 | val > dev_priv->rps.cur_freq) | |
dd75fdc8 CW |
4820 | new_power = BETWEEN; |
4821 | break; | |
4822 | ||
4823 | case BETWEEN: | |
a72b5623 CW |
4824 | if (val <= dev_priv->rps.efficient_freq && |
4825 | val < dev_priv->rps.cur_freq) | |
dd75fdc8 | 4826 | new_power = LOW_POWER; |
a72b5623 CW |
4827 | else if (val >= dev_priv->rps.rp0_freq && |
4828 | val > dev_priv->rps.cur_freq) | |
dd75fdc8 CW |
4829 | new_power = HIGH_POWER; |
4830 | break; | |
4831 | ||
4832 | case HIGH_POWER: | |
a72b5623 CW |
4833 | if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && |
4834 | val < dev_priv->rps.cur_freq) | |
dd75fdc8 CW |
4835 | new_power = BETWEEN; |
4836 | break; | |
4837 | } | |
4838 | /* Max/min bins are special */ | |
aed242ff | 4839 | if (val <= dev_priv->rps.min_freq_softlimit) |
dd75fdc8 | 4840 | new_power = LOW_POWER; |
aed242ff | 4841 | if (val >= dev_priv->rps.max_freq_softlimit) |
dd75fdc8 CW |
4842 | new_power = HIGH_POWER; |
4843 | if (new_power == dev_priv->rps.power) | |
4844 | return; | |
4845 | ||
4846 | /* Note the units here are not exactly 1us, but 1280ns. */ | |
4847 | switch (new_power) { | |
4848 | case LOW_POWER: | |
4849 | /* Upclock if more than 95% busy over 16ms */ | |
8a586437 AG |
4850 | ei_up = 16000; |
4851 | threshold_up = 95; | |
dd75fdc8 CW |
4852 | |
4853 | /* Downclock if less than 85% busy over 32ms */ | |
8a586437 AG |
4854 | ei_down = 32000; |
4855 | threshold_down = 85; | |
dd75fdc8 CW |
4856 | break; |
4857 | ||
4858 | case BETWEEN: | |
4859 | /* Upclock if more than 90% busy over 13ms */ | |
8a586437 AG |
4860 | ei_up = 13000; |
4861 | threshold_up = 90; | |
dd75fdc8 CW |
4862 | |
4863 | /* Downclock if less than 75% busy over 32ms */ | |
8a586437 AG |
4864 | ei_down = 32000; |
4865 | threshold_down = 75; | |
dd75fdc8 CW |
4866 | break; |
4867 | ||
4868 | case HIGH_POWER: | |
4869 | /* Upclock if more than 85% busy over 10ms */ | |
8a586437 AG |
4870 | ei_up = 10000; |
4871 | threshold_up = 85; | |
dd75fdc8 CW |
4872 | |
4873 | /* Downclock if less than 60% busy over 32ms */ | |
8a586437 AG |
4874 | ei_down = 32000; |
4875 | threshold_down = 60; | |
dd75fdc8 CW |
4876 | break; |
4877 | } | |
4878 | ||
1a6ffd54 MK |
4879 | /* When byt can survive without system hang with dynamic |
4880 | * sw freq adjustments, this restriction can be lifted. | |
4881 | */ | |
4882 | if (IS_VALLEYVIEW(dev_priv)) | |
4883 | goto skip_hw_write; | |
4884 | ||
8a586437 | 4885 | I915_WRITE(GEN6_RP_UP_EI, |
a72b5623 | 4886 | GT_INTERVAL_FROM_US(dev_priv, ei_up)); |
8a586437 | 4887 | I915_WRITE(GEN6_RP_UP_THRESHOLD, |
a72b5623 CW |
4888 | GT_INTERVAL_FROM_US(dev_priv, |
4889 | ei_up * threshold_up / 100)); | |
8a586437 AG |
4890 | |
4891 | I915_WRITE(GEN6_RP_DOWN_EI, | |
a72b5623 | 4892 | GT_INTERVAL_FROM_US(dev_priv, ei_down)); |
8a586437 | 4893 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, |
a72b5623 CW |
4894 | GT_INTERVAL_FROM_US(dev_priv, |
4895 | ei_down * threshold_down / 100)); | |
4896 | ||
4897 | I915_WRITE(GEN6_RP_CONTROL, | |
4898 | GEN6_RP_MEDIA_TURBO | | |
4899 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
4900 | GEN6_RP_MEDIA_IS_GFX | | |
4901 | GEN6_RP_ENABLE | | |
4902 | GEN6_RP_UP_BUSY_AVG | | |
4903 | GEN6_RP_DOWN_IDLE_AVG); | |
8a586437 | 4904 | |
1a6ffd54 | 4905 | skip_hw_write: |
dd75fdc8 | 4906 | dev_priv->rps.power = new_power; |
8fb55197 CW |
4907 | dev_priv->rps.up_threshold = threshold_up; |
4908 | dev_priv->rps.down_threshold = threshold_down; | |
dd75fdc8 CW |
4909 | dev_priv->rps.last_adj = 0; |
4910 | } | |
4911 | ||
2876ce73 CW |
4912 | static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val) |
4913 | { | |
4914 | u32 mask = 0; | |
4915 | ||
4916 | if (val > dev_priv->rps.min_freq_softlimit) | |
6f4b12f8 | 4917 | mask |= GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT; |
2876ce73 | 4918 | if (val < dev_priv->rps.max_freq_softlimit) |
6f4b12f8 | 4919 | mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD; |
2876ce73 | 4920 | |
7b3c29f6 CW |
4921 | mask &= dev_priv->pm_rps_events; |
4922 | ||
59d02a1f | 4923 | return gen6_sanitize_rps_pm_mask(dev_priv, ~mask); |
2876ce73 CW |
4924 | } |
4925 | ||
b8a5ff8d JM |
4926 | /* gen6_set_rps is called to update the frequency request, but should also be |
4927 | * called when the range (min_delay and max_delay) is modified so that we can | |
4928 | * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */ | |
dc97997a | 4929 | static void gen6_set_rps(struct drm_i915_private *dev_priv, u8 val) |
20b46e59 | 4930 | { |
23eafea6 | 4931 | /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */ |
dc97997a | 4932 | if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) |
23eafea6 SAK |
4933 | return; |
4934 | ||
4fc688ce | 4935 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
aed242ff CW |
4936 | WARN_ON(val > dev_priv->rps.max_freq); |
4937 | WARN_ON(val < dev_priv->rps.min_freq); | |
004777cb | 4938 | |
eb64cad1 CW |
4939 | /* min/max delay may still have been modified so be sure to |
4940 | * write the limits value. | |
4941 | */ | |
4942 | if (val != dev_priv->rps.cur_freq) { | |
4943 | gen6_set_rps_thresholds(dev_priv, val); | |
b8a5ff8d | 4944 | |
dc97997a | 4945 | if (IS_GEN9(dev_priv)) |
5704195c AG |
4946 | I915_WRITE(GEN6_RPNSWREQ, |
4947 | GEN9_FREQUENCY(val)); | |
dc97997a | 4948 | else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
eb64cad1 CW |
4949 | I915_WRITE(GEN6_RPNSWREQ, |
4950 | HSW_FREQUENCY(val)); | |
4951 | else | |
4952 | I915_WRITE(GEN6_RPNSWREQ, | |
4953 | GEN6_FREQUENCY(val) | | |
4954 | GEN6_OFFSET(0) | | |
4955 | GEN6_AGGRESSIVE_TURBO); | |
b8a5ff8d | 4956 | } |
7b9e0ae6 | 4957 | |
7b9e0ae6 CW |
4958 | /* Make sure we continue to get interrupts |
4959 | * until we hit the minimum or maximum frequencies. | |
4960 | */ | |
74ef1173 | 4961 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val)); |
2876ce73 | 4962 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
7b9e0ae6 | 4963 | |
d5570a72 BW |
4964 | POSTING_READ(GEN6_RPNSWREQ); |
4965 | ||
b39fb297 | 4966 | dev_priv->rps.cur_freq = val; |
0f94592e | 4967 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); |
2b4e57bd ED |
4968 | } |
4969 | ||
dc97997a | 4970 | static void valleyview_set_rps(struct drm_i915_private *dev_priv, u8 val) |
ffe02b40 | 4971 | { |
ffe02b40 | 4972 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
aed242ff CW |
4973 | WARN_ON(val > dev_priv->rps.max_freq); |
4974 | WARN_ON(val < dev_priv->rps.min_freq); | |
ffe02b40 | 4975 | |
dc97997a | 4976 | if (WARN_ONCE(IS_CHERRYVIEW(dev_priv) && (val & 1), |
ffe02b40 VS |
4977 | "Odd GPU freq value\n")) |
4978 | val &= ~1; | |
4979 | ||
cd25dd5b D |
4980 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
4981 | ||
8fb55197 | 4982 | if (val != dev_priv->rps.cur_freq) { |
ffe02b40 | 4983 | vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val); |
8fb55197 CW |
4984 | if (!IS_CHERRYVIEW(dev_priv)) |
4985 | gen6_set_rps_thresholds(dev_priv, val); | |
4986 | } | |
ffe02b40 | 4987 | |
ffe02b40 VS |
4988 | dev_priv->rps.cur_freq = val; |
4989 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); | |
4990 | } | |
4991 | ||
a7f6e231 | 4992 | /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down |
76c3552f D |
4993 | * |
4994 | * * If Gfx is Idle, then | |
a7f6e231 D |
4995 | * 1. Forcewake Media well. |
4996 | * 2. Request idle freq. | |
4997 | * 3. Release Forcewake of Media well. | |
76c3552f D |
4998 | */ |
4999 | static void vlv_set_rps_idle(struct drm_i915_private *dev_priv) | |
5000 | { | |
aed242ff | 5001 | u32 val = dev_priv->rps.idle_freq; |
5549d25f | 5002 | |
aed242ff | 5003 | if (dev_priv->rps.cur_freq <= val) |
76c3552f D |
5004 | return; |
5005 | ||
a7f6e231 D |
5006 | /* Wake up the media well, as that takes a lot less |
5007 | * power than the Render well. */ | |
5008 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA); | |
dc97997a | 5009 | valleyview_set_rps(dev_priv, val); |
a7f6e231 | 5010 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA); |
76c3552f D |
5011 | } |
5012 | ||
43cf3bf0 CW |
5013 | void gen6_rps_busy(struct drm_i915_private *dev_priv) |
5014 | { | |
5015 | mutex_lock(&dev_priv->rps.hw_lock); | |
5016 | if (dev_priv->rps.enabled) { | |
5017 | if (dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED)) | |
5018 | gen6_rps_reset_ei(dev_priv); | |
5019 | I915_WRITE(GEN6_PMINTRMSK, | |
5020 | gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq)); | |
2b83c4c4 | 5021 | |
c33d247d CW |
5022 | gen6_enable_rps_interrupts(dev_priv); |
5023 | ||
2b83c4c4 MW |
5024 | /* Ensure we start at the user's desired frequency */ |
5025 | intel_set_rps(dev_priv, | |
5026 | clamp(dev_priv->rps.cur_freq, | |
5027 | dev_priv->rps.min_freq_softlimit, | |
5028 | dev_priv->rps.max_freq_softlimit)); | |
43cf3bf0 CW |
5029 | } |
5030 | mutex_unlock(&dev_priv->rps.hw_lock); | |
5031 | } | |
5032 | ||
b29c19b6 CW |
5033 | void gen6_rps_idle(struct drm_i915_private *dev_priv) |
5034 | { | |
c33d247d CW |
5035 | /* Flush our bottom-half so that it does not race with us |
5036 | * setting the idle frequency and so that it is bounded by | |
5037 | * our rpm wakeref. And then disable the interrupts to stop any | |
5038 | * futher RPS reclocking whilst we are asleep. | |
5039 | */ | |
5040 | gen6_disable_rps_interrupts(dev_priv); | |
5041 | ||
b29c19b6 | 5042 | mutex_lock(&dev_priv->rps.hw_lock); |
c0951f0c | 5043 | if (dev_priv->rps.enabled) { |
dc97997a | 5044 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
76c3552f | 5045 | vlv_set_rps_idle(dev_priv); |
7526ed79 | 5046 | else |
dc97997a | 5047 | gen6_set_rps(dev_priv, dev_priv->rps.idle_freq); |
c0951f0c | 5048 | dev_priv->rps.last_adj = 0; |
12c100bf VS |
5049 | I915_WRITE(GEN6_PMINTRMSK, |
5050 | gen6_sanitize_rps_pm_mask(dev_priv, ~0)); | |
c0951f0c | 5051 | } |
8d3afd7d | 5052 | mutex_unlock(&dev_priv->rps.hw_lock); |
1854d5ca | 5053 | |
8d3afd7d | 5054 | spin_lock(&dev_priv->rps.client_lock); |
1854d5ca CW |
5055 | while (!list_empty(&dev_priv->rps.clients)) |
5056 | list_del_init(dev_priv->rps.clients.next); | |
8d3afd7d | 5057 | spin_unlock(&dev_priv->rps.client_lock); |
b29c19b6 CW |
5058 | } |
5059 | ||
1854d5ca | 5060 | void gen6_rps_boost(struct drm_i915_private *dev_priv, |
e61b9958 CW |
5061 | struct intel_rps_client *rps, |
5062 | unsigned long submitted) | |
b29c19b6 | 5063 | { |
8d3afd7d CW |
5064 | /* This is intentionally racy! We peek at the state here, then |
5065 | * validate inside the RPS worker. | |
5066 | */ | |
67d97da3 | 5067 | if (!(dev_priv->gt.awake && |
8d3afd7d | 5068 | dev_priv->rps.enabled && |
29ecd78d | 5069 | dev_priv->rps.cur_freq < dev_priv->rps.boost_freq)) |
8d3afd7d | 5070 | return; |
43cf3bf0 | 5071 | |
e61b9958 CW |
5072 | /* Force a RPS boost (and don't count it against the client) if |
5073 | * the GPU is severely congested. | |
5074 | */ | |
d0bc54f2 | 5075 | if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES)) |
e61b9958 CW |
5076 | rps = NULL; |
5077 | ||
8d3afd7d CW |
5078 | spin_lock(&dev_priv->rps.client_lock); |
5079 | if (rps == NULL || list_empty(&rps->link)) { | |
5080 | spin_lock_irq(&dev_priv->irq_lock); | |
5081 | if (dev_priv->rps.interrupts_enabled) { | |
5082 | dev_priv->rps.client_boost = true; | |
c33d247d | 5083 | schedule_work(&dev_priv->rps.work); |
8d3afd7d CW |
5084 | } |
5085 | spin_unlock_irq(&dev_priv->irq_lock); | |
1854d5ca | 5086 | |
2e1b8730 CW |
5087 | if (rps != NULL) { |
5088 | list_add(&rps->link, &dev_priv->rps.clients); | |
5089 | rps->boosts++; | |
1854d5ca CW |
5090 | } else |
5091 | dev_priv->rps.boosts++; | |
c0951f0c | 5092 | } |
8d3afd7d | 5093 | spin_unlock(&dev_priv->rps.client_lock); |
b29c19b6 CW |
5094 | } |
5095 | ||
dc97997a | 5096 | void intel_set_rps(struct drm_i915_private *dev_priv, u8 val) |
0a073b84 | 5097 | { |
dc97997a CW |
5098 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
5099 | valleyview_set_rps(dev_priv, val); | |
ffe02b40 | 5100 | else |
dc97997a | 5101 | gen6_set_rps(dev_priv, val); |
0a073b84 JB |
5102 | } |
5103 | ||
dc97997a | 5104 | static void gen9_disable_rc6(struct drm_i915_private *dev_priv) |
20e49366 | 5105 | { |
20e49366 | 5106 | I915_WRITE(GEN6_RC_CONTROL, 0); |
38c23527 | 5107 | I915_WRITE(GEN9_PG_ENABLE, 0); |
20e49366 ZW |
5108 | } |
5109 | ||
dc97997a | 5110 | static void gen9_disable_rps(struct drm_i915_private *dev_priv) |
2030d684 | 5111 | { |
2030d684 AG |
5112 | I915_WRITE(GEN6_RP_CONTROL, 0); |
5113 | } | |
5114 | ||
dc97997a | 5115 | static void gen6_disable_rps(struct drm_i915_private *dev_priv) |
d20d4f0c | 5116 | { |
d20d4f0c | 5117 | I915_WRITE(GEN6_RC_CONTROL, 0); |
44fc7d5c | 5118 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
2030d684 | 5119 | I915_WRITE(GEN6_RP_CONTROL, 0); |
44fc7d5c DV |
5120 | } |
5121 | ||
dc97997a | 5122 | static void cherryview_disable_rps(struct drm_i915_private *dev_priv) |
38807746 | 5123 | { |
38807746 D |
5124 | I915_WRITE(GEN6_RC_CONTROL, 0); |
5125 | } | |
5126 | ||
dc97997a | 5127 | static void valleyview_disable_rps(struct drm_i915_private *dev_priv) |
44fc7d5c | 5128 | { |
98a2e5f9 D |
5129 | /* we're doing forcewake before Disabling RC6, |
5130 | * This what the BIOS expects when going into suspend */ | |
59bad947 | 5131 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
98a2e5f9 | 5132 | |
44fc7d5c | 5133 | I915_WRITE(GEN6_RC_CONTROL, 0); |
d20d4f0c | 5134 | |
59bad947 | 5135 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
d20d4f0c JB |
5136 | } |
5137 | ||
dc97997a | 5138 | static void intel_print_rc6_info(struct drm_i915_private *dev_priv, u32 mode) |
dc39fff7 | 5139 | { |
dc97997a | 5140 | if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { |
91ca689a ID |
5141 | if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1))) |
5142 | mode = GEN6_RC_CTL_RC6_ENABLE; | |
5143 | else | |
5144 | mode = 0; | |
5145 | } | |
dc97997a | 5146 | if (HAS_RC6p(dev_priv)) |
b99d49cc ID |
5147 | DRM_DEBUG_DRIVER("Enabling RC6 states: " |
5148 | "RC6 %s RC6p %s RC6pp %s\n", | |
5149 | onoff(mode & GEN6_RC_CTL_RC6_ENABLE), | |
5150 | onoff(mode & GEN6_RC_CTL_RC6p_ENABLE), | |
5151 | onoff(mode & GEN6_RC_CTL_RC6pp_ENABLE)); | |
58abf1da RV |
5152 | |
5153 | else | |
b99d49cc ID |
5154 | DRM_DEBUG_DRIVER("Enabling RC6 states: RC6 %s\n", |
5155 | onoff(mode & GEN6_RC_CTL_RC6_ENABLE)); | |
dc39fff7 BW |
5156 | } |
5157 | ||
dc97997a | 5158 | static bool bxt_check_bios_rc6_setup(struct drm_i915_private *dev_priv) |
274008e8 | 5159 | { |
72e96d64 | 5160 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
274008e8 SAK |
5161 | bool enable_rc6 = true; |
5162 | unsigned long rc6_ctx_base; | |
fc619841 ID |
5163 | u32 rc_ctl; |
5164 | int rc_sw_target; | |
5165 | ||
5166 | rc_ctl = I915_READ(GEN6_RC_CONTROL); | |
5167 | rc_sw_target = (I915_READ(GEN6_RC_STATE) & RC_SW_TARGET_STATE_MASK) >> | |
5168 | RC_SW_TARGET_STATE_SHIFT; | |
5169 | DRM_DEBUG_DRIVER("BIOS enabled RC states: " | |
5170 | "HW_CTRL %s HW_RC6 %s SW_TARGET_STATE %x\n", | |
5171 | onoff(rc_ctl & GEN6_RC_CTL_HW_ENABLE), | |
5172 | onoff(rc_ctl & GEN6_RC_CTL_RC6_ENABLE), | |
5173 | rc_sw_target); | |
274008e8 SAK |
5174 | |
5175 | if (!(I915_READ(RC6_LOCATION) & RC6_CTX_IN_DRAM)) { | |
b99d49cc | 5176 | DRM_DEBUG_DRIVER("RC6 Base location not set properly.\n"); |
274008e8 SAK |
5177 | enable_rc6 = false; |
5178 | } | |
5179 | ||
5180 | /* | |
5181 | * The exact context size is not known for BXT, so assume a page size | |
5182 | * for this check. | |
5183 | */ | |
5184 | rc6_ctx_base = I915_READ(RC6_CTX_BASE) & RC6_CTX_BASE_MASK; | |
72e96d64 JL |
5185 | if (!((rc6_ctx_base >= ggtt->stolen_reserved_base) && |
5186 | (rc6_ctx_base + PAGE_SIZE <= ggtt->stolen_reserved_base + | |
5187 | ggtt->stolen_reserved_size))) { | |
b99d49cc | 5188 | DRM_DEBUG_DRIVER("RC6 Base address not as expected.\n"); |
274008e8 SAK |
5189 | enable_rc6 = false; |
5190 | } | |
5191 | ||
5192 | if (!(((I915_READ(PWRCTX_MAXCNT_RCSUNIT) & IDLE_TIME_MASK) > 1) && | |
5193 | ((I915_READ(PWRCTX_MAXCNT_VCSUNIT0) & IDLE_TIME_MASK) > 1) && | |
5194 | ((I915_READ(PWRCTX_MAXCNT_BCSUNIT) & IDLE_TIME_MASK) > 1) && | |
5195 | ((I915_READ(PWRCTX_MAXCNT_VECSUNIT) & IDLE_TIME_MASK) > 1))) { | |
b99d49cc | 5196 | DRM_DEBUG_DRIVER("Engine Idle wait time not set properly.\n"); |
274008e8 SAK |
5197 | enable_rc6 = false; |
5198 | } | |
5199 | ||
fc619841 ID |
5200 | if (!I915_READ(GEN8_PUSHBUS_CONTROL) || |
5201 | !I915_READ(GEN8_PUSHBUS_ENABLE) || | |
5202 | !I915_READ(GEN8_PUSHBUS_SHIFT)) { | |
5203 | DRM_DEBUG_DRIVER("Pushbus not setup properly.\n"); | |
5204 | enable_rc6 = false; | |
5205 | } | |
5206 | ||
5207 | if (!I915_READ(GEN6_GFXPAUSE)) { | |
5208 | DRM_DEBUG_DRIVER("GFX pause not setup properly.\n"); | |
5209 | enable_rc6 = false; | |
5210 | } | |
5211 | ||
5212 | if (!I915_READ(GEN8_MISC_CTRL0)) { | |
5213 | DRM_DEBUG_DRIVER("GPM control not setup properly.\n"); | |
274008e8 SAK |
5214 | enable_rc6 = false; |
5215 | } | |
5216 | ||
5217 | return enable_rc6; | |
5218 | } | |
5219 | ||
dc97997a | 5220 | int sanitize_rc6_option(struct drm_i915_private *dev_priv, int enable_rc6) |
2b4e57bd | 5221 | { |
e7d66d89 | 5222 | /* No RC6 before Ironlake and code is gone for ilk. */ |
dc97997a | 5223 | if (INTEL_INFO(dev_priv)->gen < 6) |
e6069ca8 ID |
5224 | return 0; |
5225 | ||
274008e8 SAK |
5226 | if (!enable_rc6) |
5227 | return 0; | |
5228 | ||
dc97997a | 5229 | if (IS_BROXTON(dev_priv) && !bxt_check_bios_rc6_setup(dev_priv)) { |
274008e8 SAK |
5230 | DRM_INFO("RC6 disabled by BIOS\n"); |
5231 | return 0; | |
5232 | } | |
5233 | ||
456470eb | 5234 | /* Respect the kernel parameter if it is set */ |
e6069ca8 ID |
5235 | if (enable_rc6 >= 0) { |
5236 | int mask; | |
5237 | ||
dc97997a | 5238 | if (HAS_RC6p(dev_priv)) |
e6069ca8 ID |
5239 | mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE | |
5240 | INTEL_RC6pp_ENABLE; | |
5241 | else | |
5242 | mask = INTEL_RC6_ENABLE; | |
5243 | ||
5244 | if ((enable_rc6 & mask) != enable_rc6) | |
b99d49cc ID |
5245 | DRM_DEBUG_DRIVER("Adjusting RC6 mask to %d " |
5246 | "(requested %d, valid %d)\n", | |
5247 | enable_rc6 & mask, enable_rc6, mask); | |
e6069ca8 ID |
5248 | |
5249 | return enable_rc6 & mask; | |
5250 | } | |
2b4e57bd | 5251 | |
dc97997a | 5252 | if (IS_IVYBRIDGE(dev_priv)) |
cca84a1f | 5253 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); |
8bade1ad BW |
5254 | |
5255 | return INTEL_RC6_ENABLE; | |
2b4e57bd ED |
5256 | } |
5257 | ||
dc97997a | 5258 | static void gen6_init_rps_frequencies(struct drm_i915_private *dev_priv) |
3280e8b0 BW |
5259 | { |
5260 | /* All of these values are in units of 50MHz */ | |
773ea9a8 | 5261 | |
93ee2920 | 5262 | /* static values from HW: RP0 > RP1 > RPn (min_freq) */ |
dc97997a | 5263 | if (IS_BROXTON(dev_priv)) { |
773ea9a8 | 5264 | u32 rp_state_cap = I915_READ(BXT_RP_STATE_CAP); |
35040562 BP |
5265 | dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff; |
5266 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; | |
5267 | dev_priv->rps.min_freq = (rp_state_cap >> 0) & 0xff; | |
5268 | } else { | |
773ea9a8 | 5269 | u32 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); |
35040562 BP |
5270 | dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff; |
5271 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; | |
5272 | dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff; | |
5273 | } | |
3280e8b0 | 5274 | /* hw_max = RP0 until we check for overclocking */ |
773ea9a8 | 5275 | dev_priv->rps.max_freq = dev_priv->rps.rp0_freq; |
3280e8b0 | 5276 | |
93ee2920 | 5277 | dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq; |
dc97997a CW |
5278 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv) || |
5279 | IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) { | |
773ea9a8 CW |
5280 | u32 ddcc_status = 0; |
5281 | ||
5282 | if (sandybridge_pcode_read(dev_priv, | |
5283 | HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL, | |
5284 | &ddcc_status) == 0) | |
93ee2920 | 5285 | dev_priv->rps.efficient_freq = |
46efa4ab TR |
5286 | clamp_t(u8, |
5287 | ((ddcc_status >> 8) & 0xff), | |
5288 | dev_priv->rps.min_freq, | |
5289 | dev_priv->rps.max_freq); | |
93ee2920 TR |
5290 | } |
5291 | ||
dc97997a | 5292 | if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) { |
c5e0688c | 5293 | /* Store the frequency values in 16.66 MHZ units, which is |
773ea9a8 CW |
5294 | * the natural hardware unit for SKL |
5295 | */ | |
c5e0688c AG |
5296 | dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER; |
5297 | dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER; | |
5298 | dev_priv->rps.min_freq *= GEN9_FREQ_SCALER; | |
5299 | dev_priv->rps.max_freq *= GEN9_FREQ_SCALER; | |
5300 | dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER; | |
5301 | } | |
3280e8b0 BW |
5302 | } |
5303 | ||
3a45b05c CW |
5304 | static void reset_rps(struct drm_i915_private *dev_priv, |
5305 | void (*set)(struct drm_i915_private *, u8)) | |
5306 | { | |
5307 | u8 freq = dev_priv->rps.cur_freq; | |
5308 | ||
5309 | /* force a reset */ | |
5310 | dev_priv->rps.power = -1; | |
5311 | dev_priv->rps.cur_freq = -1; | |
5312 | ||
5313 | set(dev_priv, freq); | |
5314 | } | |
5315 | ||
b6fef0ef | 5316 | /* See the Gen9_GT_PM_Programming_Guide doc for the below */ |
dc97997a | 5317 | static void gen9_enable_rps(struct drm_i915_private *dev_priv) |
b6fef0ef | 5318 | { |
b6fef0ef JB |
5319 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
5320 | ||
23eafea6 | 5321 | /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */ |
dc97997a | 5322 | if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) { |
2030d684 AG |
5323 | /* |
5324 | * BIOS could leave the Hw Turbo enabled, so need to explicitly | |
5325 | * clear out the Control register just to avoid inconsitency | |
5326 | * with debugfs interface, which will show Turbo as enabled | |
5327 | * only and that is not expected by the User after adding the | |
5328 | * WaGsvDisableTurbo. Apart from this there is no problem even | |
5329 | * if the Turbo is left enabled in the Control register, as the | |
5330 | * Up/Down interrupts would remain masked. | |
5331 | */ | |
dc97997a | 5332 | gen9_disable_rps(dev_priv); |
23eafea6 SAK |
5333 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
5334 | return; | |
5335 | } | |
5336 | ||
0beb059a AG |
5337 | /* Program defaults and thresholds for RPS*/ |
5338 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
5339 | GEN9_FREQUENCY(dev_priv->rps.rp1_freq)); | |
5340 | ||
5341 | /* 1 second timeout*/ | |
5342 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, | |
5343 | GT_INTERVAL_FROM_US(dev_priv, 1000000)); | |
5344 | ||
b6fef0ef | 5345 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa); |
b6fef0ef | 5346 | |
0beb059a AG |
5347 | /* Leaning on the below call to gen6_set_rps to program/setup the |
5348 | * Up/Down EI & threshold registers, as well as the RP_CONTROL, | |
5349 | * RP_INTERRUPT_LIMITS & RPNSWREQ registers */ | |
3a45b05c | 5350 | reset_rps(dev_priv, gen6_set_rps); |
b6fef0ef JB |
5351 | |
5352 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); | |
5353 | } | |
5354 | ||
dc97997a | 5355 | static void gen9_enable_rc6(struct drm_i915_private *dev_priv) |
20e49366 | 5356 | { |
e2f80391 | 5357 | struct intel_engine_cs *engine; |
3b3f1650 | 5358 | enum intel_engine_id id; |
20e49366 | 5359 | uint32_t rc6_mask = 0; |
20e49366 ZW |
5360 | |
5361 | /* 1a: Software RC state - RC0 */ | |
5362 | I915_WRITE(GEN6_RC_STATE, 0); | |
5363 | ||
5364 | /* 1b: Get forcewake during program sequence. Although the driver | |
5365 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 5366 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
20e49366 ZW |
5367 | |
5368 | /* 2a: Disable RC states. */ | |
5369 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
5370 | ||
5371 | /* 2b: Program RC6 thresholds.*/ | |
63a4dec2 SAK |
5372 | |
5373 | /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */ | |
dc97997a | 5374 | if (IS_SKYLAKE(dev_priv)) |
63a4dec2 SAK |
5375 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16); |
5376 | else | |
5377 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16); | |
20e49366 ZW |
5378 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ |
5379 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
3b3f1650 | 5380 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 5381 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
97c322e7 | 5382 | |
1a3d1898 | 5383 | if (HAS_GUC(dev_priv)) |
97c322e7 SAK |
5384 | I915_WRITE(GUC_MAX_IDLE_COUNT, 0xA); |
5385 | ||
20e49366 | 5386 | I915_WRITE(GEN6_RC_SLEEP, 0); |
20e49366 | 5387 | |
38c23527 ZW |
5388 | /* 2c: Program Coarse Power Gating Policies. */ |
5389 | I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25); | |
5390 | I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25); | |
5391 | ||
20e49366 | 5392 | /* 3a: Enable RC6 */ |
dc97997a | 5393 | if (intel_enable_rc6() & INTEL_RC6_ENABLE) |
20e49366 | 5394 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
87ad3212 | 5395 | DRM_INFO("RC6 %s\n", onoff(rc6_mask & GEN6_RC_CTL_RC6_ENABLE)); |
4ff40a41 | 5396 | /* WaRsUseTimeoutMode:bxt */ |
9fc736e8 | 5397 | if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) { |
3e7732a0 | 5398 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us */ |
e3429cd2 SAK |
5399 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
5400 | GEN7_RC_CTL_TO_MODE | | |
5401 | rc6_mask); | |
3e7732a0 SAK |
5402 | } else { |
5403 | I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */ | |
e3429cd2 SAK |
5404 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
5405 | GEN6_RC_CTL_EI_MODE(1) | | |
5406 | rc6_mask); | |
3e7732a0 | 5407 | } |
20e49366 | 5408 | |
cb07bae0 SK |
5409 | /* |
5410 | * 3b: Enable Coarse Power Gating only when RC6 is enabled. | |
f2d2fe95 | 5411 | * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6. |
cb07bae0 | 5412 | */ |
dc97997a | 5413 | if (NEEDS_WaRsDisableCoarsePowerGating(dev_priv)) |
f2d2fe95 SAK |
5414 | I915_WRITE(GEN9_PG_ENABLE, 0); |
5415 | else | |
5416 | I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? | |
5417 | (GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE) : 0); | |
38c23527 | 5418 | |
59bad947 | 5419 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
20e49366 ZW |
5420 | } |
5421 | ||
dc97997a | 5422 | static void gen8_enable_rps(struct drm_i915_private *dev_priv) |
6edee7f3 | 5423 | { |
e2f80391 | 5424 | struct intel_engine_cs *engine; |
3b3f1650 | 5425 | enum intel_engine_id id; |
93ee2920 | 5426 | uint32_t rc6_mask = 0; |
6edee7f3 BW |
5427 | |
5428 | /* 1a: Software RC state - RC0 */ | |
5429 | I915_WRITE(GEN6_RC_STATE, 0); | |
5430 | ||
5431 | /* 1c & 1d: Get forcewake during program sequence. Although the driver | |
5432 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 5433 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
6edee7f3 BW |
5434 | |
5435 | /* 2a: Disable RC states. */ | |
5436 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
5437 | ||
6edee7f3 BW |
5438 | /* 2b: Program RC6 thresholds.*/ |
5439 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); | |
5440 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
5441 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
3b3f1650 | 5442 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 5443 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
6edee7f3 | 5444 | I915_WRITE(GEN6_RC_SLEEP, 0); |
dc97997a | 5445 | if (IS_BROADWELL(dev_priv)) |
0d68b25e TR |
5446 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */ |
5447 | else | |
5448 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */ | |
6edee7f3 BW |
5449 | |
5450 | /* 3: Enable RC6 */ | |
dc97997a | 5451 | if (intel_enable_rc6() & INTEL_RC6_ENABLE) |
6edee7f3 | 5452 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; |
dc97997a CW |
5453 | intel_print_rc6_info(dev_priv, rc6_mask); |
5454 | if (IS_BROADWELL(dev_priv)) | |
0d68b25e TR |
5455 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | |
5456 | GEN7_RC_CTL_TO_MODE | | |
5457 | rc6_mask); | |
5458 | else | |
5459 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | | |
5460 | GEN6_RC_CTL_EI_MODE(1) | | |
5461 | rc6_mask); | |
6edee7f3 BW |
5462 | |
5463 | /* 4 Program defaults and thresholds for RPS*/ | |
f9bdc585 BW |
5464 | I915_WRITE(GEN6_RPNSWREQ, |
5465 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); | |
5466 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
5467 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); | |
7526ed79 DV |
5468 | /* NB: Docs say 1s, and 1000000 - which aren't equivalent */ |
5469 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */ | |
5470 | ||
5471 | /* Docs recommend 900MHz, and 300 MHz respectively */ | |
5472 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, | |
5473 | dev_priv->rps.max_freq_softlimit << 24 | | |
5474 | dev_priv->rps.min_freq_softlimit << 16); | |
5475 | ||
5476 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */ | |
5477 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/ | |
5478 | I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */ | |
5479 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */ | |
5480 | ||
5481 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
6edee7f3 BW |
5482 | |
5483 | /* 5: Enable RPS */ | |
7526ed79 DV |
5484 | I915_WRITE(GEN6_RP_CONTROL, |
5485 | GEN6_RP_MEDIA_TURBO | | |
5486 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
5487 | GEN6_RP_MEDIA_IS_GFX | | |
5488 | GEN6_RP_ENABLE | | |
5489 | GEN6_RP_UP_BUSY_AVG | | |
5490 | GEN6_RP_DOWN_IDLE_AVG); | |
5491 | ||
5492 | /* 6: Ring frequency + overclocking (our driver does this later */ | |
5493 | ||
3a45b05c | 5494 | reset_rps(dev_priv, gen6_set_rps); |
7526ed79 | 5495 | |
59bad947 | 5496 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
6edee7f3 BW |
5497 | } |
5498 | ||
dc97997a | 5499 | static void gen6_enable_rps(struct drm_i915_private *dev_priv) |
2b4e57bd | 5500 | { |
e2f80391 | 5501 | struct intel_engine_cs *engine; |
3b3f1650 | 5502 | enum intel_engine_id id; |
99ac9612 | 5503 | u32 rc6vids, rc6_mask = 0; |
2b4e57bd | 5504 | u32 gtfifodbg; |
2b4e57bd | 5505 | int rc6_mode; |
b4ac5afc | 5506 | int ret; |
2b4e57bd | 5507 | |
4fc688ce | 5508 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 5509 | |
2b4e57bd ED |
5510 | /* Here begins a magic sequence of register writes to enable |
5511 | * auto-downclocking. | |
5512 | * | |
5513 | * Perhaps there might be some value in exposing these to | |
5514 | * userspace... | |
5515 | */ | |
5516 | I915_WRITE(GEN6_RC_STATE, 0); | |
2b4e57bd ED |
5517 | |
5518 | /* Clear the DBG now so we don't confuse earlier errors */ | |
297b32ec VS |
5519 | gtfifodbg = I915_READ(GTFIFODBG); |
5520 | if (gtfifodbg) { | |
2b4e57bd ED |
5521 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); |
5522 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
5523 | } | |
5524 | ||
59bad947 | 5525 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
2b4e57bd ED |
5526 | |
5527 | /* disable the counters and set deterministic thresholds */ | |
5528 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
5529 | ||
5530 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); | |
5531 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); | |
5532 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); | |
5533 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
5534 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
5535 | ||
3b3f1650 | 5536 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 5537 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
2b4e57bd ED |
5538 | |
5539 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
5540 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); | |
dc97997a | 5541 | if (IS_IVYBRIDGE(dev_priv)) |
351aa566 SM |
5542 | I915_WRITE(GEN6_RC6_THRESHOLD, 125000); |
5543 | else | |
5544 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); | |
0920a487 | 5545 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
2b4e57bd ED |
5546 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
5547 | ||
5a7dc92a | 5548 | /* Check if we are enabling RC6 */ |
dc97997a | 5549 | rc6_mode = intel_enable_rc6(); |
2b4e57bd ED |
5550 | if (rc6_mode & INTEL_RC6_ENABLE) |
5551 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; | |
5552 | ||
5a7dc92a | 5553 | /* We don't use those on Haswell */ |
dc97997a | 5554 | if (!IS_HASWELL(dev_priv)) { |
5a7dc92a ED |
5555 | if (rc6_mode & INTEL_RC6p_ENABLE) |
5556 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; | |
2b4e57bd | 5557 | |
5a7dc92a ED |
5558 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
5559 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; | |
5560 | } | |
2b4e57bd | 5561 | |
dc97997a | 5562 | intel_print_rc6_info(dev_priv, rc6_mask); |
2b4e57bd ED |
5563 | |
5564 | I915_WRITE(GEN6_RC_CONTROL, | |
5565 | rc6_mask | | |
5566 | GEN6_RC_CTL_EI_MODE(1) | | |
5567 | GEN6_RC_CTL_HW_ENABLE); | |
5568 | ||
dd75fdc8 CW |
5569 | /* Power down if completely idle for over 50ms */ |
5570 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000); | |
2b4e57bd | 5571 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
2b4e57bd | 5572 | |
3a45b05c | 5573 | reset_rps(dev_priv, gen6_set_rps); |
2b4e57bd | 5574 | |
31643d54 BW |
5575 | rc6vids = 0; |
5576 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); | |
dc97997a | 5577 | if (IS_GEN6(dev_priv) && ret) { |
31643d54 | 5578 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); |
dc97997a | 5579 | } else if (IS_GEN6(dev_priv) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { |
31643d54 BW |
5580 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", |
5581 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); | |
5582 | rc6vids &= 0xffff00; | |
5583 | rc6vids |= GEN6_ENCODE_RC6_VID(450); | |
5584 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); | |
5585 | if (ret) | |
5586 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); | |
5587 | } | |
5588 | ||
59bad947 | 5589 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
2b4e57bd ED |
5590 | } |
5591 | ||
fb7404e8 | 5592 | static void gen6_update_ring_freq(struct drm_i915_private *dev_priv) |
2b4e57bd ED |
5593 | { |
5594 | int min_freq = 15; | |
3ebecd07 CW |
5595 | unsigned int gpu_freq; |
5596 | unsigned int max_ia_freq, min_ring_freq; | |
4c8c7743 | 5597 | unsigned int max_gpu_freq, min_gpu_freq; |
2b4e57bd | 5598 | int scaling_factor = 180; |
eda79642 | 5599 | struct cpufreq_policy *policy; |
2b4e57bd | 5600 | |
4fc688ce | 5601 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 5602 | |
eda79642 BW |
5603 | policy = cpufreq_cpu_get(0); |
5604 | if (policy) { | |
5605 | max_ia_freq = policy->cpuinfo.max_freq; | |
5606 | cpufreq_cpu_put(policy); | |
5607 | } else { | |
5608 | /* | |
5609 | * Default to measured freq if none found, PCU will ensure we | |
5610 | * don't go over | |
5611 | */ | |
2b4e57bd | 5612 | max_ia_freq = tsc_khz; |
eda79642 | 5613 | } |
2b4e57bd ED |
5614 | |
5615 | /* Convert from kHz to MHz */ | |
5616 | max_ia_freq /= 1000; | |
5617 | ||
153b4b95 | 5618 | min_ring_freq = I915_READ(DCLK) & 0xf; |
f6aca45c BW |
5619 | /* convert DDR frequency from units of 266.6MHz to bandwidth */ |
5620 | min_ring_freq = mult_frac(min_ring_freq, 8, 3); | |
3ebecd07 | 5621 | |
dc97997a | 5622 | if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) { |
4c8c7743 AG |
5623 | /* Convert GT frequency to 50 HZ units */ |
5624 | min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER; | |
5625 | max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER; | |
5626 | } else { | |
5627 | min_gpu_freq = dev_priv->rps.min_freq; | |
5628 | max_gpu_freq = dev_priv->rps.max_freq; | |
5629 | } | |
5630 | ||
2b4e57bd ED |
5631 | /* |
5632 | * For each potential GPU frequency, load a ring frequency we'd like | |
5633 | * to use for memory access. We do this by specifying the IA frequency | |
5634 | * the PCU should use as a reference to determine the ring frequency. | |
5635 | */ | |
4c8c7743 AG |
5636 | for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) { |
5637 | int diff = max_gpu_freq - gpu_freq; | |
3ebecd07 CW |
5638 | unsigned int ia_freq = 0, ring_freq = 0; |
5639 | ||
dc97997a | 5640 | if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) { |
4c8c7743 AG |
5641 | /* |
5642 | * ring_freq = 2 * GT. ring_freq is in 100MHz units | |
5643 | * No floor required for ring frequency on SKL. | |
5644 | */ | |
5645 | ring_freq = gpu_freq; | |
dc97997a | 5646 | } else if (INTEL_INFO(dev_priv)->gen >= 8) { |
46c764d4 BW |
5647 | /* max(2 * GT, DDR). NB: GT is 50MHz units */ |
5648 | ring_freq = max(min_ring_freq, gpu_freq); | |
dc97997a | 5649 | } else if (IS_HASWELL(dev_priv)) { |
f6aca45c | 5650 | ring_freq = mult_frac(gpu_freq, 5, 4); |
3ebecd07 CW |
5651 | ring_freq = max(min_ring_freq, ring_freq); |
5652 | /* leave ia_freq as the default, chosen by cpufreq */ | |
5653 | } else { | |
5654 | /* On older processors, there is no separate ring | |
5655 | * clock domain, so in order to boost the bandwidth | |
5656 | * of the ring, we need to upclock the CPU (ia_freq). | |
5657 | * | |
5658 | * For GPU frequencies less than 750MHz, | |
5659 | * just use the lowest ring freq. | |
5660 | */ | |
5661 | if (gpu_freq < min_freq) | |
5662 | ia_freq = 800; | |
5663 | else | |
5664 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); | |
5665 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); | |
5666 | } | |
2b4e57bd | 5667 | |
42c0526c BW |
5668 | sandybridge_pcode_write(dev_priv, |
5669 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, | |
3ebecd07 CW |
5670 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
5671 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | | |
5672 | gpu_freq); | |
2b4e57bd | 5673 | } |
2b4e57bd ED |
5674 | } |
5675 | ||
03af2045 | 5676 | static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv) |
2b6b3a09 D |
5677 | { |
5678 | u32 val, rp0; | |
5679 | ||
5b5929cb | 5680 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); |
2b6b3a09 | 5681 | |
43b67998 | 5682 | switch (INTEL_INFO(dev_priv)->sseu.eu_total) { |
5b5929cb JN |
5683 | case 8: |
5684 | /* (2 * 4) config */ | |
5685 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT); | |
5686 | break; | |
5687 | case 12: | |
5688 | /* (2 * 6) config */ | |
5689 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT); | |
5690 | break; | |
5691 | case 16: | |
5692 | /* (2 * 8) config */ | |
5693 | default: | |
5694 | /* Setting (2 * 8) Min RP0 for any other combination */ | |
5695 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT); | |
5696 | break; | |
095acd5f | 5697 | } |
5b5929cb JN |
5698 | |
5699 | rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK); | |
5700 | ||
2b6b3a09 D |
5701 | return rp0; |
5702 | } | |
5703 | ||
5704 | static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
5705 | { | |
5706 | u32 val, rpe; | |
5707 | ||
5708 | val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG); | |
5709 | rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK; | |
5710 | ||
5711 | return rpe; | |
5712 | } | |
5713 | ||
7707df4a D |
5714 | static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5715 | { | |
5716 | u32 val, rp1; | |
5717 | ||
5b5929cb JN |
5718 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); |
5719 | rp1 = (val & FB_GFX_FREQ_FUSE_MASK); | |
5720 | ||
7707df4a D |
5721 | return rp1; |
5722 | } | |
5723 | ||
f8f2b001 D |
5724 | static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5725 | { | |
5726 | u32 val, rp1; | |
5727 | ||
5728 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); | |
5729 | ||
5730 | rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT; | |
5731 | ||
5732 | return rp1; | |
5733 | } | |
5734 | ||
03af2045 | 5735 | static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv) |
0a073b84 JB |
5736 | { |
5737 | u32 val, rp0; | |
5738 | ||
64936258 | 5739 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
0a073b84 JB |
5740 | |
5741 | rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT; | |
5742 | /* Clamp to max */ | |
5743 | rp0 = min_t(u32, rp0, 0xea); | |
5744 | ||
5745 | return rp0; | |
5746 | } | |
5747 | ||
5748 | static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
5749 | { | |
5750 | u32 val, rpe; | |
5751 | ||
64936258 | 5752 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO); |
0a073b84 | 5753 | rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT; |
64936258 | 5754 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI); |
0a073b84 JB |
5755 | rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5; |
5756 | ||
5757 | return rpe; | |
5758 | } | |
5759 | ||
03af2045 | 5760 | static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv) |
0a073b84 | 5761 | { |
36146035 ID |
5762 | u32 val; |
5763 | ||
5764 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff; | |
5765 | /* | |
5766 | * According to the BYT Punit GPU turbo HAS 1.1.6.3 the minimum value | |
5767 | * for the minimum frequency in GPLL mode is 0xc1. Contrary to this on | |
5768 | * a BYT-M B0 the above register contains 0xbf. Moreover when setting | |
5769 | * a frequency Punit will not allow values below 0xc0. Clamp it 0xc0 | |
5770 | * to make sure it matches what Punit accepts. | |
5771 | */ | |
5772 | return max_t(u32, val, 0xc0); | |
0a073b84 JB |
5773 | } |
5774 | ||
ae48434c ID |
5775 | /* Check that the pctx buffer wasn't move under us. */ |
5776 | static void valleyview_check_pctx(struct drm_i915_private *dev_priv) | |
5777 | { | |
5778 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; | |
5779 | ||
5780 | WARN_ON(pctx_addr != dev_priv->mm.stolen_base + | |
5781 | dev_priv->vlv_pctx->stolen->start); | |
5782 | } | |
5783 | ||
38807746 D |
5784 | |
5785 | /* Check that the pcbr address is not empty. */ | |
5786 | static void cherryview_check_pctx(struct drm_i915_private *dev_priv) | |
5787 | { | |
5788 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; | |
5789 | ||
5790 | WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0); | |
5791 | } | |
5792 | ||
dc97997a | 5793 | static void cherryview_setup_pctx(struct drm_i915_private *dev_priv) |
38807746 | 5794 | { |
62106b4f | 5795 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
72e96d64 | 5796 | unsigned long pctx_paddr, paddr; |
38807746 D |
5797 | u32 pcbr; |
5798 | int pctx_size = 32*1024; | |
5799 | ||
38807746 D |
5800 | pcbr = I915_READ(VLV_PCBR); |
5801 | if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) { | |
ce611ef8 | 5802 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
38807746 | 5803 | paddr = (dev_priv->mm.stolen_base + |
62106b4f | 5804 | (ggtt->stolen_size - pctx_size)); |
38807746 D |
5805 | |
5806 | pctx_paddr = (paddr & (~4095)); | |
5807 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
5808 | } | |
ce611ef8 VS |
5809 | |
5810 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); | |
38807746 D |
5811 | } |
5812 | ||
dc97997a | 5813 | static void valleyview_setup_pctx(struct drm_i915_private *dev_priv) |
c9cddffc | 5814 | { |
c9cddffc JB |
5815 | struct drm_i915_gem_object *pctx; |
5816 | unsigned long pctx_paddr; | |
5817 | u32 pcbr; | |
5818 | int pctx_size = 24*1024; | |
5819 | ||
5820 | pcbr = I915_READ(VLV_PCBR); | |
5821 | if (pcbr) { | |
5822 | /* BIOS set it up already, grab the pre-alloc'd space */ | |
5823 | int pcbr_offset; | |
5824 | ||
5825 | pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base; | |
91c8a326 | 5826 | pctx = i915_gem_object_create_stolen_for_preallocated(&dev_priv->drm, |
c9cddffc | 5827 | pcbr_offset, |
190d6cd5 | 5828 | I915_GTT_OFFSET_NONE, |
c9cddffc JB |
5829 | pctx_size); |
5830 | goto out; | |
5831 | } | |
5832 | ||
ce611ef8 VS |
5833 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
5834 | ||
c9cddffc JB |
5835 | /* |
5836 | * From the Gunit register HAS: | |
5837 | * The Gfx driver is expected to program this register and ensure | |
5838 | * proper allocation within Gfx stolen memory. For example, this | |
5839 | * register should be programmed such than the PCBR range does not | |
5840 | * overlap with other ranges, such as the frame buffer, protected | |
5841 | * memory, or any other relevant ranges. | |
5842 | */ | |
91c8a326 | 5843 | pctx = i915_gem_object_create_stolen(&dev_priv->drm, pctx_size); |
c9cddffc JB |
5844 | if (!pctx) { |
5845 | DRM_DEBUG("not enough stolen space for PCTX, disabling\n"); | |
ee504898 | 5846 | goto out; |
c9cddffc JB |
5847 | } |
5848 | ||
5849 | pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start; | |
5850 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
5851 | ||
5852 | out: | |
ce611ef8 | 5853 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
c9cddffc JB |
5854 | dev_priv->vlv_pctx = pctx; |
5855 | } | |
5856 | ||
dc97997a | 5857 | static void valleyview_cleanup_pctx(struct drm_i915_private *dev_priv) |
ae48434c | 5858 | { |
ae48434c ID |
5859 | if (WARN_ON(!dev_priv->vlv_pctx)) |
5860 | return; | |
5861 | ||
f0cd5182 | 5862 | i915_gem_object_put(dev_priv->vlv_pctx); |
ae48434c ID |
5863 | dev_priv->vlv_pctx = NULL; |
5864 | } | |
5865 | ||
c30fec65 VS |
5866 | static void vlv_init_gpll_ref_freq(struct drm_i915_private *dev_priv) |
5867 | { | |
5868 | dev_priv->rps.gpll_ref_freq = | |
5869 | vlv_get_cck_clock(dev_priv, "GPLL ref", | |
5870 | CCK_GPLL_CLOCK_CONTROL, | |
5871 | dev_priv->czclk_freq); | |
5872 | ||
5873 | DRM_DEBUG_DRIVER("GPLL reference freq: %d kHz\n", | |
5874 | dev_priv->rps.gpll_ref_freq); | |
5875 | } | |
5876 | ||
dc97997a | 5877 | static void valleyview_init_gt_powersave(struct drm_i915_private *dev_priv) |
4e80519e | 5878 | { |
2bb25c17 | 5879 | u32 val; |
4e80519e | 5880 | |
dc97997a | 5881 | valleyview_setup_pctx(dev_priv); |
4e80519e | 5882 | |
c30fec65 VS |
5883 | vlv_init_gpll_ref_freq(dev_priv); |
5884 | ||
2bb25c17 VS |
5885 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5886 | switch ((val >> 6) & 3) { | |
5887 | case 0: | |
5888 | case 1: | |
5889 | dev_priv->mem_freq = 800; | |
5890 | break; | |
5891 | case 2: | |
5892 | dev_priv->mem_freq = 1066; | |
5893 | break; | |
5894 | case 3: | |
5895 | dev_priv->mem_freq = 1333; | |
5896 | break; | |
5897 | } | |
80b83b62 | 5898 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
2bb25c17 | 5899 | |
4e80519e ID |
5900 | dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv); |
5901 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; | |
5902 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5903 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
4e80519e ID |
5904 | dev_priv->rps.max_freq); |
5905 | ||
5906 | dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv); | |
5907 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5908 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
4e80519e ID |
5909 | dev_priv->rps.efficient_freq); |
5910 | ||
f8f2b001 D |
5911 | dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv); |
5912 | DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5913 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
f8f2b001 D |
5914 | dev_priv->rps.rp1_freq); |
5915 | ||
4e80519e ID |
5916 | dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv); |
5917 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5918 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
4e80519e | 5919 | dev_priv->rps.min_freq); |
4e80519e ID |
5920 | } |
5921 | ||
dc97997a | 5922 | static void cherryview_init_gt_powersave(struct drm_i915_private *dev_priv) |
38807746 | 5923 | { |
2bb25c17 | 5924 | u32 val; |
2b6b3a09 | 5925 | |
dc97997a | 5926 | cherryview_setup_pctx(dev_priv); |
2b6b3a09 | 5927 | |
c30fec65 VS |
5928 | vlv_init_gpll_ref_freq(dev_priv); |
5929 | ||
a580516d | 5930 | mutex_lock(&dev_priv->sb_lock); |
c6e8f39d | 5931 | val = vlv_cck_read(dev_priv, CCK_FUSE_REG); |
a580516d | 5932 | mutex_unlock(&dev_priv->sb_lock); |
c6e8f39d | 5933 | |
2bb25c17 | 5934 | switch ((val >> 2) & 0x7) { |
2bb25c17 | 5935 | case 3: |
2bb25c17 VS |
5936 | dev_priv->mem_freq = 2000; |
5937 | break; | |
bfa7df01 | 5938 | default: |
2bb25c17 VS |
5939 | dev_priv->mem_freq = 1600; |
5940 | break; | |
5941 | } | |
80b83b62 | 5942 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
2bb25c17 | 5943 | |
2b6b3a09 D |
5944 | dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv); |
5945 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; | |
5946 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5947 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
2b6b3a09 D |
5948 | dev_priv->rps.max_freq); |
5949 | ||
5950 | dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv); | |
5951 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5952 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
2b6b3a09 D |
5953 | dev_priv->rps.efficient_freq); |
5954 | ||
7707df4a D |
5955 | dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv); |
5956 | DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5957 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
7707df4a D |
5958 | dev_priv->rps.rp1_freq); |
5959 | ||
5b7c91b7 D |
5960 | /* PUnit validated range is only [RPe, RP0] */ |
5961 | dev_priv->rps.min_freq = dev_priv->rps.efficient_freq; | |
2b6b3a09 | 5962 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
7c59a9c1 | 5963 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
2b6b3a09 D |
5964 | dev_priv->rps.min_freq); |
5965 | ||
1c14762d VS |
5966 | WARN_ONCE((dev_priv->rps.max_freq | |
5967 | dev_priv->rps.efficient_freq | | |
5968 | dev_priv->rps.rp1_freq | | |
5969 | dev_priv->rps.min_freq) & 1, | |
5970 | "Odd GPU freq values\n"); | |
38807746 D |
5971 | } |
5972 | ||
dc97997a | 5973 | static void valleyview_cleanup_gt_powersave(struct drm_i915_private *dev_priv) |
4e80519e | 5974 | { |
dc97997a | 5975 | valleyview_cleanup_pctx(dev_priv); |
4e80519e ID |
5976 | } |
5977 | ||
dc97997a | 5978 | static void cherryview_enable_rps(struct drm_i915_private *dev_priv) |
38807746 | 5979 | { |
e2f80391 | 5980 | struct intel_engine_cs *engine; |
3b3f1650 | 5981 | enum intel_engine_id id; |
2b6b3a09 | 5982 | u32 gtfifodbg, val, rc6_mode = 0, pcbr; |
38807746 D |
5983 | |
5984 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
5985 | ||
297b32ec VS |
5986 | gtfifodbg = I915_READ(GTFIFODBG) & ~(GT_FIFO_SBDEDICATE_FREE_ENTRY_CHV | |
5987 | GT_FIFO_FREE_ENTRIES_CHV); | |
38807746 D |
5988 | if (gtfifodbg) { |
5989 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", | |
5990 | gtfifodbg); | |
5991 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
5992 | } | |
5993 | ||
5994 | cherryview_check_pctx(dev_priv); | |
5995 | ||
5996 | /* 1a & 1b: Get forcewake during program sequence. Although the driver | |
5997 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 5998 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
38807746 | 5999 | |
160614a2 VS |
6000 | /* Disable RC states. */ |
6001 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
6002 | ||
38807746 D |
6003 | /* 2a: Program RC6 thresholds.*/ |
6004 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); | |
6005 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
6006 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
6007 | ||
3b3f1650 | 6008 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 6009 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
38807746 D |
6010 | I915_WRITE(GEN6_RC_SLEEP, 0); |
6011 | ||
f4f71c7d D |
6012 | /* TO threshold set to 500 us ( 0x186 * 1.28 us) */ |
6013 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x186); | |
38807746 D |
6014 | |
6015 | /* allows RC6 residency counter to work */ | |
6016 | I915_WRITE(VLV_COUNTER_CONTROL, | |
6017 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH | | |
6018 | VLV_MEDIA_RC6_COUNT_EN | | |
6019 | VLV_RENDER_RC6_COUNT_EN)); | |
6020 | ||
6021 | /* For now we assume BIOS is allocating and populating the PCBR */ | |
6022 | pcbr = I915_READ(VLV_PCBR); | |
6023 | ||
38807746 | 6024 | /* 3: Enable RC6 */ |
dc97997a CW |
6025 | if ((intel_enable_rc6() & INTEL_RC6_ENABLE) && |
6026 | (pcbr >> VLV_PCBR_ADDR_SHIFT)) | |
af5a75a3 | 6027 | rc6_mode = GEN7_RC_CTL_TO_MODE; |
38807746 D |
6028 | |
6029 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); | |
6030 | ||
2b6b3a09 | 6031 | /* 4 Program defaults and thresholds for RPS*/ |
3cbdb48f | 6032 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
2b6b3a09 D |
6033 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
6034 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
6035 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
6036 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
6037 | ||
6038 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
6039 | ||
6040 | /* 5: Enable RPS */ | |
6041 | I915_WRITE(GEN6_RP_CONTROL, | |
6042 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
eb973a5e | 6043 | GEN6_RP_MEDIA_IS_GFX | |
2b6b3a09 D |
6044 | GEN6_RP_ENABLE | |
6045 | GEN6_RP_UP_BUSY_AVG | | |
6046 | GEN6_RP_DOWN_IDLE_AVG); | |
6047 | ||
3ef62342 D |
6048 | /* Setting Fixed Bias */ |
6049 | val = VLV_OVERRIDE_EN | | |
6050 | VLV_SOC_TDP_EN | | |
6051 | CHV_BIAS_CPU_50_SOC_50; | |
6052 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); | |
6053 | ||
2b6b3a09 D |
6054 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
6055 | ||
8d40c3ae VS |
6056 | /* RPS code assumes GPLL is used */ |
6057 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); | |
6058 | ||
742f491d | 6059 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE)); |
2b6b3a09 D |
6060 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
6061 | ||
3a45b05c | 6062 | reset_rps(dev_priv, valleyview_set_rps); |
2b6b3a09 | 6063 | |
59bad947 | 6064 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
38807746 D |
6065 | } |
6066 | ||
dc97997a | 6067 | static void valleyview_enable_rps(struct drm_i915_private *dev_priv) |
0a073b84 | 6068 | { |
e2f80391 | 6069 | struct intel_engine_cs *engine; |
3b3f1650 | 6070 | enum intel_engine_id id; |
2a5913a8 | 6071 | u32 gtfifodbg, val, rc6_mode = 0; |
0a073b84 JB |
6072 | |
6073 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
6074 | ||
ae48434c ID |
6075 | valleyview_check_pctx(dev_priv); |
6076 | ||
297b32ec VS |
6077 | gtfifodbg = I915_READ(GTFIFODBG); |
6078 | if (gtfifodbg) { | |
f7d85c1e JB |
6079 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
6080 | gtfifodbg); | |
0a073b84 JB |
6081 | I915_WRITE(GTFIFODBG, gtfifodbg); |
6082 | } | |
6083 | ||
c8d9a590 | 6084 | /* If VLV, Forcewake all wells, else re-direct to regular path */ |
59bad947 | 6085 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
0a073b84 | 6086 | |
160614a2 VS |
6087 | /* Disable RC states. */ |
6088 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
6089 | ||
cad725fe | 6090 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
0a073b84 JB |
6091 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
6092 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
6093 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
6094 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
6095 | ||
6096 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
6097 | ||
6098 | I915_WRITE(GEN6_RP_CONTROL, | |
6099 | GEN6_RP_MEDIA_TURBO | | |
6100 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
6101 | GEN6_RP_MEDIA_IS_GFX | | |
6102 | GEN6_RP_ENABLE | | |
6103 | GEN6_RP_UP_BUSY_AVG | | |
6104 | GEN6_RP_DOWN_IDLE_CONT); | |
6105 | ||
6106 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000); | |
6107 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
6108 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
6109 | ||
3b3f1650 | 6110 | for_each_engine(engine, dev_priv, id) |
e2f80391 | 6111 | I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10); |
0a073b84 | 6112 | |
2f0aa304 | 6113 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x557); |
0a073b84 JB |
6114 | |
6115 | /* allows RC6 residency counter to work */ | |
49798eb2 | 6116 | I915_WRITE(VLV_COUNTER_CONTROL, |
31685c25 D |
6117 | _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN | |
6118 | VLV_RENDER_RC0_COUNT_EN | | |
49798eb2 JB |
6119 | VLV_MEDIA_RC6_COUNT_EN | |
6120 | VLV_RENDER_RC6_COUNT_EN)); | |
31685c25 | 6121 | |
dc97997a | 6122 | if (intel_enable_rc6() & INTEL_RC6_ENABLE) |
6b88f295 | 6123 | rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL; |
dc39fff7 | 6124 | |
dc97997a | 6125 | intel_print_rc6_info(dev_priv, rc6_mode); |
dc39fff7 | 6126 | |
a2b23fe0 | 6127 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
0a073b84 | 6128 | |
3ef62342 D |
6129 | /* Setting Fixed Bias */ |
6130 | val = VLV_OVERRIDE_EN | | |
6131 | VLV_SOC_TDP_EN | | |
6132 | VLV_BIAS_CPU_125_SOC_875; | |
6133 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); | |
6134 | ||
64936258 | 6135 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
0a073b84 | 6136 | |
8d40c3ae VS |
6137 | /* RPS code assumes GPLL is used */ |
6138 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); | |
6139 | ||
742f491d | 6140 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE)); |
0a073b84 JB |
6141 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
6142 | ||
3a45b05c | 6143 | reset_rps(dev_priv, valleyview_set_rps); |
0a073b84 | 6144 | |
59bad947 | 6145 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
0a073b84 JB |
6146 | } |
6147 | ||
dde18883 ED |
6148 | static unsigned long intel_pxfreq(u32 vidfreq) |
6149 | { | |
6150 | unsigned long freq; | |
6151 | int div = (vidfreq & 0x3f0000) >> 16; | |
6152 | int post = (vidfreq & 0x3000) >> 12; | |
6153 | int pre = (vidfreq & 0x7); | |
6154 | ||
6155 | if (!pre) | |
6156 | return 0; | |
6157 | ||
6158 | freq = ((div * 133333) / ((1<<post) * pre)); | |
6159 | ||
6160 | return freq; | |
6161 | } | |
6162 | ||
eb48eb00 DV |
6163 | static const struct cparams { |
6164 | u16 i; | |
6165 | u16 t; | |
6166 | u16 m; | |
6167 | u16 c; | |
6168 | } cparams[] = { | |
6169 | { 1, 1333, 301, 28664 }, | |
6170 | { 1, 1066, 294, 24460 }, | |
6171 | { 1, 800, 294, 25192 }, | |
6172 | { 0, 1333, 276, 27605 }, | |
6173 | { 0, 1066, 276, 27605 }, | |
6174 | { 0, 800, 231, 23784 }, | |
6175 | }; | |
6176 | ||
f531dcb2 | 6177 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
6178 | { |
6179 | u64 total_count, diff, ret; | |
6180 | u32 count1, count2, count3, m = 0, c = 0; | |
6181 | unsigned long now = jiffies_to_msecs(jiffies), diff1; | |
6182 | int i; | |
6183 | ||
02d71956 DV |
6184 | assert_spin_locked(&mchdev_lock); |
6185 | ||
20e4d407 | 6186 | diff1 = now - dev_priv->ips.last_time1; |
eb48eb00 DV |
6187 | |
6188 | /* Prevent division-by-zero if we are asking too fast. | |
6189 | * Also, we don't get interesting results if we are polling | |
6190 | * faster than once in 10ms, so just return the saved value | |
6191 | * in such cases. | |
6192 | */ | |
6193 | if (diff1 <= 10) | |
20e4d407 | 6194 | return dev_priv->ips.chipset_power; |
eb48eb00 DV |
6195 | |
6196 | count1 = I915_READ(DMIEC); | |
6197 | count2 = I915_READ(DDREC); | |
6198 | count3 = I915_READ(CSIEC); | |
6199 | ||
6200 | total_count = count1 + count2 + count3; | |
6201 | ||
6202 | /* FIXME: handle per-counter overflow */ | |
20e4d407 DV |
6203 | if (total_count < dev_priv->ips.last_count1) { |
6204 | diff = ~0UL - dev_priv->ips.last_count1; | |
eb48eb00 DV |
6205 | diff += total_count; |
6206 | } else { | |
20e4d407 | 6207 | diff = total_count - dev_priv->ips.last_count1; |
eb48eb00 DV |
6208 | } |
6209 | ||
6210 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { | |
20e4d407 DV |
6211 | if (cparams[i].i == dev_priv->ips.c_m && |
6212 | cparams[i].t == dev_priv->ips.r_t) { | |
eb48eb00 DV |
6213 | m = cparams[i].m; |
6214 | c = cparams[i].c; | |
6215 | break; | |
6216 | } | |
6217 | } | |
6218 | ||
6219 | diff = div_u64(diff, diff1); | |
6220 | ret = ((m * diff) + c); | |
6221 | ret = div_u64(ret, 10); | |
6222 | ||
20e4d407 DV |
6223 | dev_priv->ips.last_count1 = total_count; |
6224 | dev_priv->ips.last_time1 = now; | |
eb48eb00 | 6225 | |
20e4d407 | 6226 | dev_priv->ips.chipset_power = ret; |
eb48eb00 DV |
6227 | |
6228 | return ret; | |
6229 | } | |
6230 | ||
f531dcb2 CW |
6231 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
6232 | { | |
6233 | unsigned long val; | |
6234 | ||
dc97997a | 6235 | if (INTEL_INFO(dev_priv)->gen != 5) |
f531dcb2 CW |
6236 | return 0; |
6237 | ||
6238 | spin_lock_irq(&mchdev_lock); | |
6239 | ||
6240 | val = __i915_chipset_val(dev_priv); | |
6241 | ||
6242 | spin_unlock_irq(&mchdev_lock); | |
6243 | ||
6244 | return val; | |
6245 | } | |
6246 | ||
eb48eb00 DV |
6247 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
6248 | { | |
6249 | unsigned long m, x, b; | |
6250 | u32 tsfs; | |
6251 | ||
6252 | tsfs = I915_READ(TSFS); | |
6253 | ||
6254 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); | |
6255 | x = I915_READ8(TR1); | |
6256 | ||
6257 | b = tsfs & TSFS_INTR_MASK; | |
6258 | ||
6259 | return ((m * x) / 127) - b; | |
6260 | } | |
6261 | ||
d972d6ee MK |
6262 | static int _pxvid_to_vd(u8 pxvid) |
6263 | { | |
6264 | if (pxvid == 0) | |
6265 | return 0; | |
6266 | ||
6267 | if (pxvid >= 8 && pxvid < 31) | |
6268 | pxvid = 31; | |
6269 | ||
6270 | return (pxvid + 2) * 125; | |
6271 | } | |
6272 | ||
6273 | static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) | |
eb48eb00 | 6274 | { |
d972d6ee MK |
6275 | const int vd = _pxvid_to_vd(pxvid); |
6276 | const int vm = vd - 1125; | |
6277 | ||
dc97997a | 6278 | if (INTEL_INFO(dev_priv)->is_mobile) |
d972d6ee MK |
6279 | return vm > 0 ? vm : 0; |
6280 | ||
6281 | return vd; | |
eb48eb00 DV |
6282 | } |
6283 | ||
02d71956 | 6284 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 | 6285 | { |
5ed0bdf2 | 6286 | u64 now, diff, diffms; |
eb48eb00 DV |
6287 | u32 count; |
6288 | ||
02d71956 | 6289 | assert_spin_locked(&mchdev_lock); |
eb48eb00 | 6290 | |
5ed0bdf2 TG |
6291 | now = ktime_get_raw_ns(); |
6292 | diffms = now - dev_priv->ips.last_time2; | |
6293 | do_div(diffms, NSEC_PER_MSEC); | |
eb48eb00 DV |
6294 | |
6295 | /* Don't divide by 0 */ | |
eb48eb00 DV |
6296 | if (!diffms) |
6297 | return; | |
6298 | ||
6299 | count = I915_READ(GFXEC); | |
6300 | ||
20e4d407 DV |
6301 | if (count < dev_priv->ips.last_count2) { |
6302 | diff = ~0UL - dev_priv->ips.last_count2; | |
eb48eb00 DV |
6303 | diff += count; |
6304 | } else { | |
20e4d407 | 6305 | diff = count - dev_priv->ips.last_count2; |
eb48eb00 DV |
6306 | } |
6307 | ||
20e4d407 DV |
6308 | dev_priv->ips.last_count2 = count; |
6309 | dev_priv->ips.last_time2 = now; | |
eb48eb00 DV |
6310 | |
6311 | /* More magic constants... */ | |
6312 | diff = diff * 1181; | |
6313 | diff = div_u64(diff, diffms * 10); | |
20e4d407 | 6314 | dev_priv->ips.gfx_power = diff; |
eb48eb00 DV |
6315 | } |
6316 | ||
02d71956 DV |
6317 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
6318 | { | |
dc97997a | 6319 | if (INTEL_INFO(dev_priv)->gen != 5) |
02d71956 DV |
6320 | return; |
6321 | ||
9270388e | 6322 | spin_lock_irq(&mchdev_lock); |
02d71956 DV |
6323 | |
6324 | __i915_update_gfx_val(dev_priv); | |
6325 | ||
9270388e | 6326 | spin_unlock_irq(&mchdev_lock); |
02d71956 DV |
6327 | } |
6328 | ||
f531dcb2 | 6329 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
6330 | { |
6331 | unsigned long t, corr, state1, corr2, state2; | |
6332 | u32 pxvid, ext_v; | |
6333 | ||
02d71956 DV |
6334 | assert_spin_locked(&mchdev_lock); |
6335 | ||
616847e7 | 6336 | pxvid = I915_READ(PXVFREQ(dev_priv->rps.cur_freq)); |
eb48eb00 DV |
6337 | pxvid = (pxvid >> 24) & 0x7f; |
6338 | ext_v = pvid_to_extvid(dev_priv, pxvid); | |
6339 | ||
6340 | state1 = ext_v; | |
6341 | ||
6342 | t = i915_mch_val(dev_priv); | |
6343 | ||
6344 | /* Revel in the empirically derived constants */ | |
6345 | ||
6346 | /* Correction factor in 1/100000 units */ | |
6347 | if (t > 80) | |
6348 | corr = ((t * 2349) + 135940); | |
6349 | else if (t >= 50) | |
6350 | corr = ((t * 964) + 29317); | |
6351 | else /* < 50 */ | |
6352 | corr = ((t * 301) + 1004); | |
6353 | ||
6354 | corr = corr * ((150142 * state1) / 10000 - 78642); | |
6355 | corr /= 100000; | |
20e4d407 | 6356 | corr2 = (corr * dev_priv->ips.corr); |
eb48eb00 DV |
6357 | |
6358 | state2 = (corr2 * state1) / 10000; | |
6359 | state2 /= 100; /* convert to mW */ | |
6360 | ||
02d71956 | 6361 | __i915_update_gfx_val(dev_priv); |
eb48eb00 | 6362 | |
20e4d407 | 6363 | return dev_priv->ips.gfx_power + state2; |
eb48eb00 DV |
6364 | } |
6365 | ||
f531dcb2 CW |
6366 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
6367 | { | |
6368 | unsigned long val; | |
6369 | ||
dc97997a | 6370 | if (INTEL_INFO(dev_priv)->gen != 5) |
f531dcb2 CW |
6371 | return 0; |
6372 | ||
6373 | spin_lock_irq(&mchdev_lock); | |
6374 | ||
6375 | val = __i915_gfx_val(dev_priv); | |
6376 | ||
6377 | spin_unlock_irq(&mchdev_lock); | |
6378 | ||
6379 | return val; | |
6380 | } | |
6381 | ||
eb48eb00 DV |
6382 | /** |
6383 | * i915_read_mch_val - return value for IPS use | |
6384 | * | |
6385 | * Calculate and return a value for the IPS driver to use when deciding whether | |
6386 | * we have thermal and power headroom to increase CPU or GPU power budget. | |
6387 | */ | |
6388 | unsigned long i915_read_mch_val(void) | |
6389 | { | |
6390 | struct drm_i915_private *dev_priv; | |
6391 | unsigned long chipset_val, graphics_val, ret = 0; | |
6392 | ||
9270388e | 6393 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
6394 | if (!i915_mch_dev) |
6395 | goto out_unlock; | |
6396 | dev_priv = i915_mch_dev; | |
6397 | ||
f531dcb2 CW |
6398 | chipset_val = __i915_chipset_val(dev_priv); |
6399 | graphics_val = __i915_gfx_val(dev_priv); | |
eb48eb00 DV |
6400 | |
6401 | ret = chipset_val + graphics_val; | |
6402 | ||
6403 | out_unlock: | |
9270388e | 6404 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6405 | |
6406 | return ret; | |
6407 | } | |
6408 | EXPORT_SYMBOL_GPL(i915_read_mch_val); | |
6409 | ||
6410 | /** | |
6411 | * i915_gpu_raise - raise GPU frequency limit | |
6412 | * | |
6413 | * Raise the limit; IPS indicates we have thermal headroom. | |
6414 | */ | |
6415 | bool i915_gpu_raise(void) | |
6416 | { | |
6417 | struct drm_i915_private *dev_priv; | |
6418 | bool ret = true; | |
6419 | ||
9270388e | 6420 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
6421 | if (!i915_mch_dev) { |
6422 | ret = false; | |
6423 | goto out_unlock; | |
6424 | } | |
6425 | dev_priv = i915_mch_dev; | |
6426 | ||
20e4d407 DV |
6427 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
6428 | dev_priv->ips.max_delay--; | |
eb48eb00 DV |
6429 | |
6430 | out_unlock: | |
9270388e | 6431 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6432 | |
6433 | return ret; | |
6434 | } | |
6435 | EXPORT_SYMBOL_GPL(i915_gpu_raise); | |
6436 | ||
6437 | /** | |
6438 | * i915_gpu_lower - lower GPU frequency limit | |
6439 | * | |
6440 | * IPS indicates we're close to a thermal limit, so throttle back the GPU | |
6441 | * frequency maximum. | |
6442 | */ | |
6443 | bool i915_gpu_lower(void) | |
6444 | { | |
6445 | struct drm_i915_private *dev_priv; | |
6446 | bool ret = true; | |
6447 | ||
9270388e | 6448 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
6449 | if (!i915_mch_dev) { |
6450 | ret = false; | |
6451 | goto out_unlock; | |
6452 | } | |
6453 | dev_priv = i915_mch_dev; | |
6454 | ||
20e4d407 DV |
6455 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
6456 | dev_priv->ips.max_delay++; | |
eb48eb00 DV |
6457 | |
6458 | out_unlock: | |
9270388e | 6459 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6460 | |
6461 | return ret; | |
6462 | } | |
6463 | EXPORT_SYMBOL_GPL(i915_gpu_lower); | |
6464 | ||
6465 | /** | |
6466 | * i915_gpu_busy - indicate GPU business to IPS | |
6467 | * | |
6468 | * Tell the IPS driver whether or not the GPU is busy. | |
6469 | */ | |
6470 | bool i915_gpu_busy(void) | |
6471 | { | |
eb48eb00 DV |
6472 | bool ret = false; |
6473 | ||
9270388e | 6474 | spin_lock_irq(&mchdev_lock); |
dcff85c8 CW |
6475 | if (i915_mch_dev) |
6476 | ret = i915_mch_dev->gt.awake; | |
9270388e | 6477 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6478 | |
6479 | return ret; | |
6480 | } | |
6481 | EXPORT_SYMBOL_GPL(i915_gpu_busy); | |
6482 | ||
6483 | /** | |
6484 | * i915_gpu_turbo_disable - disable graphics turbo | |
6485 | * | |
6486 | * Disable graphics turbo by resetting the max frequency and setting the | |
6487 | * current frequency to the default. | |
6488 | */ | |
6489 | bool i915_gpu_turbo_disable(void) | |
6490 | { | |
6491 | struct drm_i915_private *dev_priv; | |
6492 | bool ret = true; | |
6493 | ||
9270388e | 6494 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
6495 | if (!i915_mch_dev) { |
6496 | ret = false; | |
6497 | goto out_unlock; | |
6498 | } | |
6499 | dev_priv = i915_mch_dev; | |
6500 | ||
20e4d407 | 6501 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
eb48eb00 | 6502 | |
91d14251 | 6503 | if (!ironlake_set_drps(dev_priv, dev_priv->ips.fstart)) |
eb48eb00 DV |
6504 | ret = false; |
6505 | ||
6506 | out_unlock: | |
9270388e | 6507 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6508 | |
6509 | return ret; | |
6510 | } | |
6511 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); | |
6512 | ||
6513 | /** | |
6514 | * Tells the intel_ips driver that the i915 driver is now loaded, if | |
6515 | * IPS got loaded first. | |
6516 | * | |
6517 | * This awkward dance is so that neither module has to depend on the | |
6518 | * other in order for IPS to do the appropriate communication of | |
6519 | * GPU turbo limits to i915. | |
6520 | */ | |
6521 | static void | |
6522 | ips_ping_for_i915_load(void) | |
6523 | { | |
6524 | void (*link)(void); | |
6525 | ||
6526 | link = symbol_get(ips_link_to_i915_driver); | |
6527 | if (link) { | |
6528 | link(); | |
6529 | symbol_put(ips_link_to_i915_driver); | |
6530 | } | |
6531 | } | |
6532 | ||
6533 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) | |
6534 | { | |
02d71956 DV |
6535 | /* We only register the i915 ips part with intel-ips once everything is |
6536 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ | |
9270388e | 6537 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 6538 | i915_mch_dev = dev_priv; |
9270388e | 6539 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6540 | |
6541 | ips_ping_for_i915_load(); | |
6542 | } | |
6543 | ||
6544 | void intel_gpu_ips_teardown(void) | |
6545 | { | |
9270388e | 6546 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 6547 | i915_mch_dev = NULL; |
9270388e | 6548 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 | 6549 | } |
76c3552f | 6550 | |
dc97997a | 6551 | static void intel_init_emon(struct drm_i915_private *dev_priv) |
dde18883 | 6552 | { |
dde18883 ED |
6553 | u32 lcfuse; |
6554 | u8 pxw[16]; | |
6555 | int i; | |
6556 | ||
6557 | /* Disable to program */ | |
6558 | I915_WRITE(ECR, 0); | |
6559 | POSTING_READ(ECR); | |
6560 | ||
6561 | /* Program energy weights for various events */ | |
6562 | I915_WRITE(SDEW, 0x15040d00); | |
6563 | I915_WRITE(CSIEW0, 0x007f0000); | |
6564 | I915_WRITE(CSIEW1, 0x1e220004); | |
6565 | I915_WRITE(CSIEW2, 0x04000004); | |
6566 | ||
6567 | for (i = 0; i < 5; i++) | |
616847e7 | 6568 | I915_WRITE(PEW(i), 0); |
dde18883 | 6569 | for (i = 0; i < 3; i++) |
616847e7 | 6570 | I915_WRITE(DEW(i), 0); |
dde18883 ED |
6571 | |
6572 | /* Program P-state weights to account for frequency power adjustment */ | |
6573 | for (i = 0; i < 16; i++) { | |
616847e7 | 6574 | u32 pxvidfreq = I915_READ(PXVFREQ(i)); |
dde18883 ED |
6575 | unsigned long freq = intel_pxfreq(pxvidfreq); |
6576 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> | |
6577 | PXVFREQ_PX_SHIFT; | |
6578 | unsigned long val; | |
6579 | ||
6580 | val = vid * vid; | |
6581 | val *= (freq / 1000); | |
6582 | val *= 255; | |
6583 | val /= (127*127*900); | |
6584 | if (val > 0xff) | |
6585 | DRM_ERROR("bad pxval: %ld\n", val); | |
6586 | pxw[i] = val; | |
6587 | } | |
6588 | /* Render standby states get 0 weight */ | |
6589 | pxw[14] = 0; | |
6590 | pxw[15] = 0; | |
6591 | ||
6592 | for (i = 0; i < 4; i++) { | |
6593 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | | |
6594 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); | |
616847e7 | 6595 | I915_WRITE(PXW(i), val); |
dde18883 ED |
6596 | } |
6597 | ||
6598 | /* Adjust magic regs to magic values (more experimental results) */ | |
6599 | I915_WRITE(OGW0, 0); | |
6600 | I915_WRITE(OGW1, 0); | |
6601 | I915_WRITE(EG0, 0x00007f00); | |
6602 | I915_WRITE(EG1, 0x0000000e); | |
6603 | I915_WRITE(EG2, 0x000e0000); | |
6604 | I915_WRITE(EG3, 0x68000300); | |
6605 | I915_WRITE(EG4, 0x42000000); | |
6606 | I915_WRITE(EG5, 0x00140031); | |
6607 | I915_WRITE(EG6, 0); | |
6608 | I915_WRITE(EG7, 0); | |
6609 | ||
6610 | for (i = 0; i < 8; i++) | |
616847e7 | 6611 | I915_WRITE(PXWL(i), 0); |
dde18883 ED |
6612 | |
6613 | /* Enable PMON + select events */ | |
6614 | I915_WRITE(ECR, 0x80000019); | |
6615 | ||
6616 | lcfuse = I915_READ(LCFUSE02); | |
6617 | ||
20e4d407 | 6618 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
dde18883 ED |
6619 | } |
6620 | ||
dc97997a | 6621 | void intel_init_gt_powersave(struct drm_i915_private *dev_priv) |
ae48434c | 6622 | { |
b268c699 ID |
6623 | /* |
6624 | * RPM depends on RC6 to save restore the GT HW context, so make RC6 a | |
6625 | * requirement. | |
6626 | */ | |
6627 | if (!i915.enable_rc6) { | |
6628 | DRM_INFO("RC6 disabled, disabling runtime PM support\n"); | |
6629 | intel_runtime_pm_get(dev_priv); | |
6630 | } | |
e6069ca8 | 6631 | |
b5163dbb | 6632 | mutex_lock(&dev_priv->drm.struct_mutex); |
773ea9a8 CW |
6633 | mutex_lock(&dev_priv->rps.hw_lock); |
6634 | ||
6635 | /* Initialize RPS limits (for userspace) */ | |
dc97997a CW |
6636 | if (IS_CHERRYVIEW(dev_priv)) |
6637 | cherryview_init_gt_powersave(dev_priv); | |
6638 | else if (IS_VALLEYVIEW(dev_priv)) | |
6639 | valleyview_init_gt_powersave(dev_priv); | |
2a13ae79 | 6640 | else if (INTEL_GEN(dev_priv) >= 6) |
773ea9a8 CW |
6641 | gen6_init_rps_frequencies(dev_priv); |
6642 | ||
6643 | /* Derive initial user preferences/limits from the hardware limits */ | |
6644 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; | |
6645 | dev_priv->rps.cur_freq = dev_priv->rps.idle_freq; | |
6646 | ||
6647 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; | |
6648 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; | |
6649 | ||
6650 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) | |
6651 | dev_priv->rps.min_freq_softlimit = | |
6652 | max_t(int, | |
6653 | dev_priv->rps.efficient_freq, | |
6654 | intel_freq_opcode(dev_priv, 450)); | |
6655 | ||
99ac9612 CW |
6656 | /* After setting max-softlimit, find the overclock max freq */ |
6657 | if (IS_GEN6(dev_priv) || | |
6658 | IS_IVYBRIDGE(dev_priv) || IS_HASWELL(dev_priv)) { | |
6659 | u32 params = 0; | |
6660 | ||
6661 | sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, ¶ms); | |
6662 | if (params & BIT(31)) { /* OC supported */ | |
6663 | DRM_DEBUG_DRIVER("Overclocking supported, max: %dMHz, overclock: %dMHz\n", | |
6664 | (dev_priv->rps.max_freq & 0xff) * 50, | |
6665 | (params & 0xff) * 50); | |
6666 | dev_priv->rps.max_freq = params & 0xff; | |
6667 | } | |
6668 | } | |
6669 | ||
29ecd78d CW |
6670 | /* Finally allow us to boost to max by default */ |
6671 | dev_priv->rps.boost_freq = dev_priv->rps.max_freq; | |
6672 | ||
773ea9a8 | 6673 | mutex_unlock(&dev_priv->rps.hw_lock); |
b5163dbb | 6674 | mutex_unlock(&dev_priv->drm.struct_mutex); |
54b4f68f CW |
6675 | |
6676 | intel_autoenable_gt_powersave(dev_priv); | |
ae48434c ID |
6677 | } |
6678 | ||
dc97997a | 6679 | void intel_cleanup_gt_powersave(struct drm_i915_private *dev_priv) |
ae48434c | 6680 | { |
8dac1e1f | 6681 | if (IS_VALLEYVIEW(dev_priv)) |
dc97997a | 6682 | valleyview_cleanup_gt_powersave(dev_priv); |
b268c699 ID |
6683 | |
6684 | if (!i915.enable_rc6) | |
6685 | intel_runtime_pm_put(dev_priv); | |
ae48434c ID |
6686 | } |
6687 | ||
54b4f68f CW |
6688 | /** |
6689 | * intel_suspend_gt_powersave - suspend PM work and helper threads | |
6690 | * @dev_priv: i915 device | |
6691 | * | |
6692 | * We don't want to disable RC6 or other features here, we just want | |
6693 | * to make sure any work we've queued has finished and won't bother | |
6694 | * us while we're suspended. | |
6695 | */ | |
6696 | void intel_suspend_gt_powersave(struct drm_i915_private *dev_priv) | |
6697 | { | |
6698 | if (INTEL_GEN(dev_priv) < 6) | |
6699 | return; | |
6700 | ||
6701 | if (cancel_delayed_work_sync(&dev_priv->rps.autoenable_work)) | |
6702 | intel_runtime_pm_put(dev_priv); | |
6703 | ||
6704 | /* gen6_rps_idle() will be called later to disable interrupts */ | |
6705 | } | |
6706 | ||
b7137e0c CW |
6707 | void intel_sanitize_gt_powersave(struct drm_i915_private *dev_priv) |
6708 | { | |
6709 | dev_priv->rps.enabled = true; /* force disabling */ | |
6710 | intel_disable_gt_powersave(dev_priv); | |
54b4f68f CW |
6711 | |
6712 | gen6_reset_rps_interrupts(dev_priv); | |
156c7ca0 JB |
6713 | } |
6714 | ||
dc97997a | 6715 | void intel_disable_gt_powersave(struct drm_i915_private *dev_priv) |
8090c6b9 | 6716 | { |
b7137e0c CW |
6717 | if (!READ_ONCE(dev_priv->rps.enabled)) |
6718 | return; | |
e494837a | 6719 | |
b7137e0c | 6720 | mutex_lock(&dev_priv->rps.hw_lock); |
e534770a | 6721 | |
b7137e0c CW |
6722 | if (INTEL_GEN(dev_priv) >= 9) { |
6723 | gen9_disable_rc6(dev_priv); | |
6724 | gen9_disable_rps(dev_priv); | |
6725 | } else if (IS_CHERRYVIEW(dev_priv)) { | |
6726 | cherryview_disable_rps(dev_priv); | |
6727 | } else if (IS_VALLEYVIEW(dev_priv)) { | |
6728 | valleyview_disable_rps(dev_priv); | |
6729 | } else if (INTEL_GEN(dev_priv) >= 6) { | |
6730 | gen6_disable_rps(dev_priv); | |
6731 | } else if (IS_IRONLAKE_M(dev_priv)) { | |
6732 | ironlake_disable_drps(dev_priv); | |
930ebb46 | 6733 | } |
b7137e0c CW |
6734 | |
6735 | dev_priv->rps.enabled = false; | |
6736 | mutex_unlock(&dev_priv->rps.hw_lock); | |
8090c6b9 DV |
6737 | } |
6738 | ||
b7137e0c | 6739 | void intel_enable_gt_powersave(struct drm_i915_private *dev_priv) |
1a01ab3b | 6740 | { |
54b4f68f CW |
6741 | /* We shouldn't be disabling as we submit, so this should be less |
6742 | * racy than it appears! | |
6743 | */ | |
b7137e0c CW |
6744 | if (READ_ONCE(dev_priv->rps.enabled)) |
6745 | return; | |
1a01ab3b | 6746 | |
b7137e0c CW |
6747 | /* Powersaving is controlled by the host when inside a VM */ |
6748 | if (intel_vgpu_active(dev_priv)) | |
6749 | return; | |
0a073b84 | 6750 | |
b7137e0c | 6751 | mutex_lock(&dev_priv->rps.hw_lock); |
dc97997a CW |
6752 | |
6753 | if (IS_CHERRYVIEW(dev_priv)) { | |
6754 | cherryview_enable_rps(dev_priv); | |
6755 | } else if (IS_VALLEYVIEW(dev_priv)) { | |
6756 | valleyview_enable_rps(dev_priv); | |
b7137e0c | 6757 | } else if (INTEL_GEN(dev_priv) >= 9) { |
dc97997a CW |
6758 | gen9_enable_rc6(dev_priv); |
6759 | gen9_enable_rps(dev_priv); | |
6760 | if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) | |
fb7404e8 | 6761 | gen6_update_ring_freq(dev_priv); |
dc97997a CW |
6762 | } else if (IS_BROADWELL(dev_priv)) { |
6763 | gen8_enable_rps(dev_priv); | |
fb7404e8 | 6764 | gen6_update_ring_freq(dev_priv); |
b7137e0c | 6765 | } else if (INTEL_GEN(dev_priv) >= 6) { |
dc97997a | 6766 | gen6_enable_rps(dev_priv); |
fb7404e8 | 6767 | gen6_update_ring_freq(dev_priv); |
b7137e0c CW |
6768 | } else if (IS_IRONLAKE_M(dev_priv)) { |
6769 | ironlake_enable_drps(dev_priv); | |
6770 | intel_init_emon(dev_priv); | |
0a073b84 | 6771 | } |
aed242ff CW |
6772 | |
6773 | WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq); | |
6774 | WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq); | |
6775 | ||
6776 | WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq); | |
6777 | WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq); | |
6778 | ||
54b4f68f | 6779 | dev_priv->rps.enabled = true; |
b7137e0c CW |
6780 | mutex_unlock(&dev_priv->rps.hw_lock); |
6781 | } | |
3cc134e3 | 6782 | |
54b4f68f CW |
6783 | static void __intel_autoenable_gt_powersave(struct work_struct *work) |
6784 | { | |
6785 | struct drm_i915_private *dev_priv = | |
6786 | container_of(work, typeof(*dev_priv), rps.autoenable_work.work); | |
6787 | struct intel_engine_cs *rcs; | |
6788 | struct drm_i915_gem_request *req; | |
6789 | ||
6790 | if (READ_ONCE(dev_priv->rps.enabled)) | |
6791 | goto out; | |
6792 | ||
3b3f1650 | 6793 | rcs = dev_priv->engine[RCS]; |
54b4f68f CW |
6794 | if (rcs->last_context) |
6795 | goto out; | |
6796 | ||
6797 | if (!rcs->init_context) | |
6798 | goto out; | |
6799 | ||
6800 | mutex_lock(&dev_priv->drm.struct_mutex); | |
6801 | ||
6802 | req = i915_gem_request_alloc(rcs, dev_priv->kernel_context); | |
6803 | if (IS_ERR(req)) | |
6804 | goto unlock; | |
6805 | ||
6806 | if (!i915.enable_execlists && i915_switch_context(req) == 0) | |
6807 | rcs->init_context(req); | |
6808 | ||
6809 | /* Mark the device busy, calling intel_enable_gt_powersave() */ | |
6810 | i915_add_request_no_flush(req); | |
6811 | ||
6812 | unlock: | |
6813 | mutex_unlock(&dev_priv->drm.struct_mutex); | |
6814 | out: | |
6815 | intel_runtime_pm_put(dev_priv); | |
6816 | } | |
6817 | ||
6818 | void intel_autoenable_gt_powersave(struct drm_i915_private *dev_priv) | |
6819 | { | |
6820 | if (READ_ONCE(dev_priv->rps.enabled)) | |
6821 | return; | |
6822 | ||
6823 | if (IS_IRONLAKE_M(dev_priv)) { | |
6824 | ironlake_enable_drps(dev_priv); | |
54b4f68f | 6825 | intel_init_emon(dev_priv); |
54b4f68f CW |
6826 | } else if (INTEL_INFO(dev_priv)->gen >= 6) { |
6827 | /* | |
6828 | * PCU communication is slow and this doesn't need to be | |
6829 | * done at any specific time, so do this out of our fast path | |
6830 | * to make resume and init faster. | |
6831 | * | |
6832 | * We depend on the HW RC6 power context save/restore | |
6833 | * mechanism when entering D3 through runtime PM suspend. So | |
6834 | * disable RPM until RPS/RC6 is properly setup. We can only | |
6835 | * get here via the driver load/system resume/runtime resume | |
6836 | * paths, so the _noresume version is enough (and in case of | |
6837 | * runtime resume it's necessary). | |
6838 | */ | |
6839 | if (queue_delayed_work(dev_priv->wq, | |
6840 | &dev_priv->rps.autoenable_work, | |
6841 | round_jiffies_up_relative(HZ))) | |
6842 | intel_runtime_pm_get_noresume(dev_priv); | |
6843 | } | |
6844 | } | |
6845 | ||
46f16e63 | 6846 | static void ibx_init_clock_gating(struct drm_i915_private *dev_priv) |
3107bd48 | 6847 | { |
3107bd48 DV |
6848 | /* |
6849 | * On Ibex Peak and Cougar Point, we need to disable clock | |
6850 | * gating for the panel power sequencer or it will fail to | |
6851 | * start up when no ports are active. | |
6852 | */ | |
6853 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); | |
6854 | } | |
6855 | ||
46f16e63 | 6856 | static void g4x_disable_trickle_feed(struct drm_i915_private *dev_priv) |
0e088b8f | 6857 | { |
b12ce1d8 | 6858 | enum pipe pipe; |
0e088b8f | 6859 | |
055e393f | 6860 | for_each_pipe(dev_priv, pipe) { |
0e088b8f VS |
6861 | I915_WRITE(DSPCNTR(pipe), |
6862 | I915_READ(DSPCNTR(pipe)) | | |
6863 | DISPPLANE_TRICKLE_FEED_DISABLE); | |
b12ce1d8 VS |
6864 | |
6865 | I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe))); | |
6866 | POSTING_READ(DSPSURF(pipe)); | |
0e088b8f VS |
6867 | } |
6868 | } | |
6869 | ||
46f16e63 | 6870 | static void ilk_init_lp_watermarks(struct drm_i915_private *dev_priv) |
017636cc | 6871 | { |
017636cc VS |
6872 | I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN); |
6873 | I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN); | |
6874 | I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN); | |
6875 | ||
6876 | /* | |
6877 | * Don't touch WM1S_LP_EN here. | |
6878 | * Doing so could cause underruns. | |
6879 | */ | |
6880 | } | |
6881 | ||
46f16e63 | 6882 | static void ironlake_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 6883 | { |
231e54f6 | 6884 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 6885 | |
f1e8fa56 DL |
6886 | /* |
6887 | * Required for FBC | |
6888 | * WaFbcDisableDpfcClockGating:ilk | |
6889 | */ | |
4d47e4f5 DL |
6890 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
6891 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | | |
6892 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; | |
6f1d69b0 ED |
6893 | |
6894 | I915_WRITE(PCH_3DCGDIS0, | |
6895 | MARIUNIT_CLOCK_GATE_DISABLE | | |
6896 | SVSMUNIT_CLOCK_GATE_DISABLE); | |
6897 | I915_WRITE(PCH_3DCGDIS1, | |
6898 | VFMUNIT_CLOCK_GATE_DISABLE); | |
6899 | ||
6f1d69b0 ED |
6900 | /* |
6901 | * According to the spec the following bits should be set in | |
6902 | * order to enable memory self-refresh | |
6903 | * The bit 22/21 of 0x42004 | |
6904 | * The bit 5 of 0x42020 | |
6905 | * The bit 15 of 0x45000 | |
6906 | */ | |
6907 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
6908 | (I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6909 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); | |
4d47e4f5 | 6910 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
6f1d69b0 ED |
6911 | I915_WRITE(DISP_ARB_CTL, |
6912 | (I915_READ(DISP_ARB_CTL) | | |
6913 | DISP_FBC_WM_DIS)); | |
017636cc | 6914 | |
46f16e63 | 6915 | ilk_init_lp_watermarks(dev_priv); |
6f1d69b0 ED |
6916 | |
6917 | /* | |
6918 | * Based on the document from hardware guys the following bits | |
6919 | * should be set unconditionally in order to enable FBC. | |
6920 | * The bit 22 of 0x42000 | |
6921 | * The bit 22 of 0x42004 | |
6922 | * The bit 7,8,9 of 0x42020. | |
6923 | */ | |
50a0bc90 | 6924 | if (IS_IRONLAKE_M(dev_priv)) { |
4bb35334 | 6925 | /* WaFbcAsynchFlipDisableFbcQueue:ilk */ |
6f1d69b0 ED |
6926 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
6927 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
6928 | ILK_FBCQ_DIS); | |
6929 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
6930 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6931 | ILK_DPARB_GATE); | |
6f1d69b0 ED |
6932 | } |
6933 | ||
4d47e4f5 DL |
6934 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6935 | ||
6f1d69b0 ED |
6936 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6937 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6938 | ILK_ELPIN_409_SELECT); | |
6939 | I915_WRITE(_3D_CHICKEN2, | |
6940 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | | |
6941 | _3D_CHICKEN2_WM_READ_PIPELINED); | |
4358a374 | 6942 | |
ecdb4eb7 | 6943 | /* WaDisableRenderCachePipelinedFlush:ilk */ |
4358a374 DV |
6944 | I915_WRITE(CACHE_MODE_0, |
6945 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
3107bd48 | 6946 | |
4e04632e AG |
6947 | /* WaDisable_RenderCache_OperationalFlush:ilk */ |
6948 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6949 | ||
46f16e63 | 6950 | g4x_disable_trickle_feed(dev_priv); |
bdad2b2f | 6951 | |
46f16e63 | 6952 | ibx_init_clock_gating(dev_priv); |
3107bd48 DV |
6953 | } |
6954 | ||
46f16e63 | 6955 | static void cpt_init_clock_gating(struct drm_i915_private *dev_priv) |
3107bd48 | 6956 | { |
3107bd48 | 6957 | int pipe; |
3f704fa2 | 6958 | uint32_t val; |
3107bd48 DV |
6959 | |
6960 | /* | |
6961 | * On Ibex Peak and Cougar Point, we need to disable clock | |
6962 | * gating for the panel power sequencer or it will fail to | |
6963 | * start up when no ports are active. | |
6964 | */ | |
cd664078 JB |
6965 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE | |
6966 | PCH_DPLUNIT_CLOCK_GATE_DISABLE | | |
6967 | PCH_CPUNIT_CLOCK_GATE_DISABLE); | |
3107bd48 DV |
6968 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | |
6969 | DPLS_EDP_PPS_FIX_DIS); | |
335c07b7 TI |
6970 | /* The below fixes the weird display corruption, a few pixels shifted |
6971 | * downward, on (only) LVDS of some HP laptops with IVY. | |
6972 | */ | |
055e393f | 6973 | for_each_pipe(dev_priv, pipe) { |
dc4bd2d1 PZ |
6974 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
6975 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; | |
6976 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; | |
41aa3448 | 6977 | if (dev_priv->vbt.fdi_rx_polarity_inverted) |
3f704fa2 | 6978 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
dc4bd2d1 PZ |
6979 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
6980 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; | |
6981 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; | |
3f704fa2 PZ |
6982 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
6983 | } | |
3107bd48 | 6984 | /* WADP0ClockGatingDisable */ |
055e393f | 6985 | for_each_pipe(dev_priv, pipe) { |
3107bd48 DV |
6986 | I915_WRITE(TRANS_CHICKEN1(pipe), |
6987 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); | |
6988 | } | |
6f1d69b0 ED |
6989 | } |
6990 | ||
46f16e63 | 6991 | static void gen6_check_mch_setup(struct drm_i915_private *dev_priv) |
1d7aaa0c | 6992 | { |
1d7aaa0c DV |
6993 | uint32_t tmp; |
6994 | ||
6995 | tmp = I915_READ(MCH_SSKPD); | |
df662a28 DV |
6996 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) |
6997 | DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n", | |
6998 | tmp); | |
1d7aaa0c DV |
6999 | } |
7000 | ||
46f16e63 | 7001 | static void gen6_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7002 | { |
231e54f6 | 7003 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 7004 | |
231e54f6 | 7005 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6f1d69b0 ED |
7006 | |
7007 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
7008 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
7009 | ILK_ELPIN_409_SELECT); | |
7010 | ||
ecdb4eb7 | 7011 | /* WaDisableHiZPlanesWhenMSAAEnabled:snb */ |
4283908e DV |
7012 | I915_WRITE(_3D_CHICKEN, |
7013 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); | |
7014 | ||
4e04632e AG |
7015 | /* WaDisable_RenderCache_OperationalFlush:snb */ |
7016 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
7017 | ||
8d85d272 VS |
7018 | /* |
7019 | * BSpec recoomends 8x4 when MSAA is used, | |
7020 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
7021 | * |
7022 | * Note that PS/WM thread counts depend on the WIZ hashing | |
7023 | * disable bit, which we don't touch here, but it's good | |
7024 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
8d85d272 VS |
7025 | */ |
7026 | I915_WRITE(GEN6_GT_MODE, | |
98533251 | 7027 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
8d85d272 | 7028 | |
46f16e63 | 7029 | ilk_init_lp_watermarks(dev_priv); |
6f1d69b0 | 7030 | |
6f1d69b0 | 7031 | I915_WRITE(CACHE_MODE_0, |
50743298 | 7032 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
6f1d69b0 ED |
7033 | |
7034 | I915_WRITE(GEN6_UCGCTL1, | |
7035 | I915_READ(GEN6_UCGCTL1) | | |
7036 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | | |
7037 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
7038 | ||
7039 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock | |
7040 | * gating disable must be set. Failure to set it results in | |
7041 | * flickering pixels due to Z write ordering failures after | |
7042 | * some amount of runtime in the Mesa "fire" demo, and Unigine | |
7043 | * Sanctuary and Tropics, and apparently anything else with | |
7044 | * alpha test or pixel discard. | |
7045 | * | |
7046 | * According to the spec, bit 11 (RCCUNIT) must also be set, | |
7047 | * but we didn't debug actual testcases to find it out. | |
0f846f81 | 7048 | * |
ef59318c VS |
7049 | * WaDisableRCCUnitClockGating:snb |
7050 | * WaDisableRCPBUnitClockGating:snb | |
6f1d69b0 ED |
7051 | */ |
7052 | I915_WRITE(GEN6_UCGCTL2, | |
7053 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | | |
7054 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); | |
7055 | ||
5eb146dd | 7056 | /* WaStripsFansDisableFastClipPerformanceFix:snb */ |
743b57d8 VS |
7057 | I915_WRITE(_3D_CHICKEN3, |
7058 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL)); | |
6f1d69b0 | 7059 | |
e927ecde VS |
7060 | /* |
7061 | * Bspec says: | |
7062 | * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and | |
7063 | * 3DSTATE_SF number of SF output attributes is more than 16." | |
7064 | */ | |
7065 | I915_WRITE(_3D_CHICKEN3, | |
7066 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH)); | |
7067 | ||
6f1d69b0 ED |
7068 | /* |
7069 | * According to the spec the following bits should be | |
7070 | * set in order to enable memory self-refresh and fbc: | |
7071 | * The bit21 and bit22 of 0x42000 | |
7072 | * The bit21 and bit22 of 0x42004 | |
7073 | * The bit5 and bit7 of 0x42020 | |
7074 | * The bit14 of 0x70180 | |
7075 | * The bit14 of 0x71180 | |
4bb35334 DL |
7076 | * |
7077 | * WaFbcAsynchFlipDisableFbcQueue:snb | |
6f1d69b0 ED |
7078 | */ |
7079 | I915_WRITE(ILK_DISPLAY_CHICKEN1, | |
7080 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
7081 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); | |
7082 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
7083 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
7084 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); | |
231e54f6 DL |
7085 | I915_WRITE(ILK_DSPCLK_GATE_D, |
7086 | I915_READ(ILK_DSPCLK_GATE_D) | | |
7087 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | | |
7088 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); | |
6f1d69b0 | 7089 | |
46f16e63 | 7090 | g4x_disable_trickle_feed(dev_priv); |
f8f2ac9a | 7091 | |
46f16e63 | 7092 | cpt_init_clock_gating(dev_priv); |
1d7aaa0c | 7093 | |
46f16e63 | 7094 | gen6_check_mch_setup(dev_priv); |
6f1d69b0 ED |
7095 | } |
7096 | ||
7097 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) | |
7098 | { | |
7099 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); | |
7100 | ||
3aad9059 | 7101 | /* |
46680e0a | 7102 | * WaVSThreadDispatchOverride:ivb,vlv |
3aad9059 VS |
7103 | * |
7104 | * This actually overrides the dispatch | |
7105 | * mode for all thread types. | |
7106 | */ | |
6f1d69b0 ED |
7107 | reg &= ~GEN7_FF_SCHED_MASK; |
7108 | reg |= GEN7_FF_TS_SCHED_HW; | |
7109 | reg |= GEN7_FF_VS_SCHED_HW; | |
7110 | reg |= GEN7_FF_DS_SCHED_HW; | |
7111 | ||
7112 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); | |
7113 | } | |
7114 | ||
46f16e63 | 7115 | static void lpt_init_clock_gating(struct drm_i915_private *dev_priv) |
17a303ec | 7116 | { |
17a303ec PZ |
7117 | /* |
7118 | * TODO: this bit should only be enabled when really needed, then | |
7119 | * disabled when not needed anymore in order to save power. | |
7120 | */ | |
4f8036a2 | 7121 | if (HAS_PCH_LPT_LP(dev_priv)) |
17a303ec PZ |
7122 | I915_WRITE(SOUTH_DSPCLK_GATE_D, |
7123 | I915_READ(SOUTH_DSPCLK_GATE_D) | | |
7124 | PCH_LP_PARTITION_LEVEL_DISABLE); | |
0a790cdb PZ |
7125 | |
7126 | /* WADPOClockGatingDisable:hsw */ | |
36c0d0cf VS |
7127 | I915_WRITE(TRANS_CHICKEN1(PIPE_A), |
7128 | I915_READ(TRANS_CHICKEN1(PIPE_A)) | | |
0a790cdb | 7129 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); |
17a303ec PZ |
7130 | } |
7131 | ||
712bf364 | 7132 | static void lpt_suspend_hw(struct drm_i915_private *dev_priv) |
7d708ee4 | 7133 | { |
4f8036a2 | 7134 | if (HAS_PCH_LPT_LP(dev_priv)) { |
7d708ee4 ID |
7135 | uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D); |
7136 | ||
7137 | val &= ~PCH_LP_PARTITION_LEVEL_DISABLE; | |
7138 | I915_WRITE(SOUTH_DSPCLK_GATE_D, val); | |
7139 | } | |
7140 | } | |
7141 | ||
450174fe ID |
7142 | static void gen8_set_l3sqc_credits(struct drm_i915_private *dev_priv, |
7143 | int general_prio_credits, | |
7144 | int high_prio_credits) | |
7145 | { | |
7146 | u32 misccpctl; | |
7147 | ||
7148 | /* WaTempDisableDOPClkGating:bdw */ | |
7149 | misccpctl = I915_READ(GEN7_MISCCPCTL); | |
7150 | I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE); | |
7151 | ||
7152 | I915_WRITE(GEN8_L3SQCREG1, | |
7153 | L3_GENERAL_PRIO_CREDITS(general_prio_credits) | | |
7154 | L3_HIGH_PRIO_CREDITS(high_prio_credits)); | |
7155 | ||
7156 | /* | |
7157 | * Wait at least 100 clocks before re-enabling clock gating. | |
7158 | * See the definition of L3SQCREG1 in BSpec. | |
7159 | */ | |
7160 | POSTING_READ(GEN8_L3SQCREG1); | |
7161 | udelay(1); | |
7162 | I915_WRITE(GEN7_MISCCPCTL, misccpctl); | |
7163 | } | |
7164 | ||
46f16e63 | 7165 | static void kabylake_init_clock_gating(struct drm_i915_private *dev_priv) |
9498dba7 | 7166 | { |
46f16e63 | 7167 | gen9_init_clock_gating(dev_priv); |
9498dba7 MK |
7168 | |
7169 | /* WaDisableSDEUnitClockGating:kbl */ | |
7170 | if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0)) | |
7171 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
7172 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
8aeb7f62 MK |
7173 | |
7174 | /* WaDisableGamClockGating:kbl */ | |
7175 | if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0)) | |
7176 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | | |
7177 | GEN6_GAMUNIT_CLOCK_GATE_DISABLE); | |
031cd8c8 MK |
7178 | |
7179 | /* WaFbcNukeOnHostModify:kbl */ | |
7180 | I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) | | |
7181 | ILK_DPFC_NUKE_ON_ANY_MODIFICATION); | |
9498dba7 MK |
7182 | } |
7183 | ||
46f16e63 | 7184 | static void skylake_init_clock_gating(struct drm_i915_private *dev_priv) |
dc00b6a0 | 7185 | { |
46f16e63 | 7186 | gen9_init_clock_gating(dev_priv); |
44fff99f MK |
7187 | |
7188 | /* WAC6entrylatency:skl */ | |
7189 | I915_WRITE(FBC_LLC_READ_CTRL, I915_READ(FBC_LLC_READ_CTRL) | | |
7190 | FBC_LLC_FULLY_OPEN); | |
031cd8c8 MK |
7191 | |
7192 | /* WaFbcNukeOnHostModify:skl */ | |
7193 | I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) | | |
7194 | ILK_DPFC_NUKE_ON_ANY_MODIFICATION); | |
dc00b6a0 DV |
7195 | } |
7196 | ||
46f16e63 | 7197 | static void broadwell_init_clock_gating(struct drm_i915_private *dev_priv) |
1020a5c2 | 7198 | { |
07d27e20 | 7199 | enum pipe pipe; |
1020a5c2 | 7200 | |
46f16e63 | 7201 | ilk_init_lp_watermarks(dev_priv); |
50ed5fbd | 7202 | |
ab57fff1 | 7203 | /* WaSwitchSolVfFArbitrationPriority:bdw */ |
50ed5fbd | 7204 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
fe4ab3ce | 7205 | |
ab57fff1 | 7206 | /* WaPsrDPAMaskVBlankInSRD:bdw */ |
fe4ab3ce BW |
7207 | I915_WRITE(CHICKEN_PAR1_1, |
7208 | I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD); | |
7209 | ||
ab57fff1 | 7210 | /* WaPsrDPRSUnmaskVBlankInSRD:bdw */ |
055e393f | 7211 | for_each_pipe(dev_priv, pipe) { |
07d27e20 | 7212 | I915_WRITE(CHICKEN_PIPESL_1(pipe), |
c7c65622 | 7213 | I915_READ(CHICKEN_PIPESL_1(pipe)) | |
8f670bb1 | 7214 | BDW_DPRS_MASK_VBLANK_SRD); |
fe4ab3ce | 7215 | } |
63801f21 | 7216 | |
ab57fff1 BW |
7217 | /* WaVSRefCountFullforceMissDisable:bdw */ |
7218 | /* WaDSRefCountFullforceMissDisable:bdw */ | |
7219 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
7220 | I915_READ(GEN7_FF_THREAD_MODE) & | |
7221 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); | |
36075a4c | 7222 | |
295e8bb7 VS |
7223 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
7224 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); | |
4f1ca9e9 VS |
7225 | |
7226 | /* WaDisableSDEUnitClockGating:bdw */ | |
7227 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
7228 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
5d708680 | 7229 | |
450174fe ID |
7230 | /* WaProgramL3SqcReg1Default:bdw */ |
7231 | gen8_set_l3sqc_credits(dev_priv, 30, 2); | |
4d487cff | 7232 | |
6d50b065 VS |
7233 | /* |
7234 | * WaGttCachingOffByDefault:bdw | |
7235 | * GTT cache may not work with big pages, so if those | |
7236 | * are ever enabled GTT cache may need to be disabled. | |
7237 | */ | |
7238 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); | |
7239 | ||
17e0adf0 MK |
7240 | /* WaKVMNotificationOnConfigChange:bdw */ |
7241 | I915_WRITE(CHICKEN_PAR2_1, I915_READ(CHICKEN_PAR2_1) | |
7242 | | KVM_CONFIG_CHANGE_NOTIFICATION_SELECT); | |
7243 | ||
46f16e63 | 7244 | lpt_init_clock_gating(dev_priv); |
1020a5c2 BW |
7245 | } |
7246 | ||
46f16e63 | 7247 | static void haswell_init_clock_gating(struct drm_i915_private *dev_priv) |
cad2a2d7 | 7248 | { |
46f16e63 | 7249 | ilk_init_lp_watermarks(dev_priv); |
cad2a2d7 | 7250 | |
f3fc4884 FJ |
7251 | /* L3 caching of data atomics doesn't work -- disable it. */ |
7252 | I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE); | |
7253 | I915_WRITE(HSW_ROW_CHICKEN3, | |
7254 | _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE)); | |
7255 | ||
ecdb4eb7 | 7256 | /* This is required by WaCatErrorRejectionIssue:hsw */ |
cad2a2d7 ED |
7257 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
7258 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
7259 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
7260 | ||
e36ea7ff VS |
7261 | /* WaVSRefCountFullforceMissDisable:hsw */ |
7262 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
7263 | I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME); | |
cad2a2d7 | 7264 | |
4e04632e AG |
7265 | /* WaDisable_RenderCache_OperationalFlush:hsw */ |
7266 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
7267 | ||
fe27c606 CW |
7268 | /* enable HiZ Raw Stall Optimization */ |
7269 | I915_WRITE(CACHE_MODE_0_GEN7, | |
7270 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); | |
7271 | ||
ecdb4eb7 | 7272 | /* WaDisable4x2SubspanOptimization:hsw */ |
cad2a2d7 ED |
7273 | I915_WRITE(CACHE_MODE_1, |
7274 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
1544d9d5 | 7275 | |
a12c4967 VS |
7276 | /* |
7277 | * BSpec recommends 8x4 when MSAA is used, | |
7278 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
7279 | * |
7280 | * Note that PS/WM thread counts depend on the WIZ hashing | |
7281 | * disable bit, which we don't touch here, but it's good | |
7282 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
a12c4967 VS |
7283 | */ |
7284 | I915_WRITE(GEN7_GT_MODE, | |
98533251 | 7285 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
a12c4967 | 7286 | |
94411593 KG |
7287 | /* WaSampleCChickenBitEnable:hsw */ |
7288 | I915_WRITE(HALF_SLICE_CHICKEN3, | |
7289 | _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE)); | |
7290 | ||
ecdb4eb7 | 7291 | /* WaSwitchSolVfFArbitrationPriority:hsw */ |
e3dff585 BW |
7292 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
7293 | ||
90a88643 PZ |
7294 | /* WaRsPkgCStateDisplayPMReq:hsw */ |
7295 | I915_WRITE(CHICKEN_PAR1_1, | |
7296 | I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES); | |
1544d9d5 | 7297 | |
46f16e63 | 7298 | lpt_init_clock_gating(dev_priv); |
cad2a2d7 ED |
7299 | } |
7300 | ||
46f16e63 | 7301 | static void ivybridge_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7302 | { |
20848223 | 7303 | uint32_t snpcr; |
6f1d69b0 | 7304 | |
46f16e63 | 7305 | ilk_init_lp_watermarks(dev_priv); |
6f1d69b0 | 7306 | |
231e54f6 | 7307 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
6f1d69b0 | 7308 | |
ecdb4eb7 | 7309 | /* WaDisableEarlyCull:ivb */ |
87f8020e JB |
7310 | I915_WRITE(_3D_CHICKEN3, |
7311 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
7312 | ||
ecdb4eb7 | 7313 | /* WaDisableBackToBackFlipFix:ivb */ |
6f1d69b0 ED |
7314 | I915_WRITE(IVB_CHICKEN3, |
7315 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
7316 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
7317 | ||
ecdb4eb7 | 7318 | /* WaDisablePSDDualDispatchEnable:ivb */ |
50a0bc90 | 7319 | if (IS_IVB_GT1(dev_priv)) |
12f3382b JB |
7320 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
7321 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 7322 | |
4e04632e AG |
7323 | /* WaDisable_RenderCache_OperationalFlush:ivb */ |
7324 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
7325 | ||
ecdb4eb7 | 7326 | /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */ |
6f1d69b0 ED |
7327 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
7328 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); | |
7329 | ||
ecdb4eb7 | 7330 | /* WaApplyL3ControlAndL3ChickenMode:ivb */ |
6f1d69b0 ED |
7331 | I915_WRITE(GEN7_L3CNTLREG1, |
7332 | GEN7_WA_FOR_GEN7_L3_CONTROL); | |
7333 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, | |
8ab43976 | 7334 | GEN7_WA_L3_CHICKEN_MODE); |
50a0bc90 | 7335 | if (IS_IVB_GT1(dev_priv)) |
8ab43976 JB |
7336 | I915_WRITE(GEN7_ROW_CHICKEN2, |
7337 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
412236c2 VS |
7338 | else { |
7339 | /* must write both registers */ | |
7340 | I915_WRITE(GEN7_ROW_CHICKEN2, | |
7341 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
8ab43976 JB |
7342 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, |
7343 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
412236c2 | 7344 | } |
6f1d69b0 | 7345 | |
ecdb4eb7 | 7346 | /* WaForceL3Serialization:ivb */ |
61939d97 JB |
7347 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
7348 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
7349 | ||
1b80a19a | 7350 | /* |
0f846f81 | 7351 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
ecdb4eb7 | 7352 | * This implements the WaDisableRCZUnitClockGating:ivb workaround. |
0f846f81 JB |
7353 | */ |
7354 | I915_WRITE(GEN6_UCGCTL2, | |
28acf3b2 | 7355 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
0f846f81 | 7356 | |
ecdb4eb7 | 7357 | /* This is required by WaCatErrorRejectionIssue:ivb */ |
6f1d69b0 ED |
7358 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
7359 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
7360 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
7361 | ||
46f16e63 | 7362 | g4x_disable_trickle_feed(dev_priv); |
6f1d69b0 ED |
7363 | |
7364 | gen7_setup_fixed_func_scheduler(dev_priv); | |
97e1930f | 7365 | |
22721343 CW |
7366 | if (0) { /* causes HiZ corruption on ivb:gt1 */ |
7367 | /* enable HiZ Raw Stall Optimization */ | |
7368 | I915_WRITE(CACHE_MODE_0_GEN7, | |
7369 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); | |
7370 | } | |
116f2b6d | 7371 | |
ecdb4eb7 | 7372 | /* WaDisable4x2SubspanOptimization:ivb */ |
97e1930f DV |
7373 | I915_WRITE(CACHE_MODE_1, |
7374 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
20848223 | 7375 | |
a607c1a4 VS |
7376 | /* |
7377 | * BSpec recommends 8x4 when MSAA is used, | |
7378 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
7379 | * |
7380 | * Note that PS/WM thread counts depend on the WIZ hashing | |
7381 | * disable bit, which we don't touch here, but it's good | |
7382 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
a607c1a4 VS |
7383 | */ |
7384 | I915_WRITE(GEN7_GT_MODE, | |
98533251 | 7385 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
a607c1a4 | 7386 | |
20848223 BW |
7387 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); |
7388 | snpcr &= ~GEN6_MBC_SNPCR_MASK; | |
7389 | snpcr |= GEN6_MBC_SNPCR_MED; | |
7390 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); | |
3107bd48 | 7391 | |
6e266956 | 7392 | if (!HAS_PCH_NOP(dev_priv)) |
46f16e63 | 7393 | cpt_init_clock_gating(dev_priv); |
1d7aaa0c | 7394 | |
46f16e63 | 7395 | gen6_check_mch_setup(dev_priv); |
6f1d69b0 ED |
7396 | } |
7397 | ||
46f16e63 | 7398 | static void valleyview_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7399 | { |
ecdb4eb7 | 7400 | /* WaDisableEarlyCull:vlv */ |
87f8020e JB |
7401 | I915_WRITE(_3D_CHICKEN3, |
7402 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
7403 | ||
ecdb4eb7 | 7404 | /* WaDisableBackToBackFlipFix:vlv */ |
6f1d69b0 ED |
7405 | I915_WRITE(IVB_CHICKEN3, |
7406 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
7407 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
7408 | ||
fad7d36e | 7409 | /* WaPsdDispatchEnable:vlv */ |
ecdb4eb7 | 7410 | /* WaDisablePSDDualDispatchEnable:vlv */ |
12f3382b | 7411 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
d3bc0303 JB |
7412 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
7413 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 7414 | |
4e04632e AG |
7415 | /* WaDisable_RenderCache_OperationalFlush:vlv */ |
7416 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
7417 | ||
ecdb4eb7 | 7418 | /* WaForceL3Serialization:vlv */ |
61939d97 JB |
7419 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
7420 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
7421 | ||
ecdb4eb7 | 7422 | /* WaDisableDopClockGating:vlv */ |
8ab43976 JB |
7423 | I915_WRITE(GEN7_ROW_CHICKEN2, |
7424 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
7425 | ||
ecdb4eb7 | 7426 | /* This is required by WaCatErrorRejectionIssue:vlv */ |
6f1d69b0 ED |
7427 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
7428 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
7429 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
7430 | ||
46680e0a VS |
7431 | gen7_setup_fixed_func_scheduler(dev_priv); |
7432 | ||
3c0edaeb | 7433 | /* |
0f846f81 | 7434 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
ecdb4eb7 | 7435 | * This implements the WaDisableRCZUnitClockGating:vlv workaround. |
0f846f81 JB |
7436 | */ |
7437 | I915_WRITE(GEN6_UCGCTL2, | |
3c0edaeb | 7438 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
0f846f81 | 7439 | |
c98f5062 AG |
7440 | /* WaDisableL3Bank2xClockGate:vlv |
7441 | * Disabling L3 clock gating- MMIO 940c[25] = 1 | |
7442 | * Set bit 25, to disable L3_BANK_2x_CLK_GATING */ | |
7443 | I915_WRITE(GEN7_UCGCTL4, | |
7444 | I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE); | |
e3f33d46 | 7445 | |
afd58e79 VS |
7446 | /* |
7447 | * BSpec says this must be set, even though | |
7448 | * WaDisable4x2SubspanOptimization isn't listed for VLV. | |
7449 | */ | |
6b26c86d DV |
7450 | I915_WRITE(CACHE_MODE_1, |
7451 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
7983117f | 7452 | |
da2518f9 VS |
7453 | /* |
7454 | * BSpec recommends 8x4 when MSAA is used, | |
7455 | * however in practice 16x4 seems fastest. | |
7456 | * | |
7457 | * Note that PS/WM thread counts depend on the WIZ hashing | |
7458 | * disable bit, which we don't touch here, but it's good | |
7459 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
7460 | */ | |
7461 | I915_WRITE(GEN7_GT_MODE, | |
7462 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); | |
7463 | ||
031994ee VS |
7464 | /* |
7465 | * WaIncreaseL3CreditsForVLVB0:vlv | |
7466 | * This is the hardware default actually. | |
7467 | */ | |
7468 | I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE); | |
7469 | ||
2d809570 | 7470 | /* |
ecdb4eb7 | 7471 | * WaDisableVLVClockGating_VBIIssue:vlv |
2d809570 JB |
7472 | * Disable clock gating on th GCFG unit to prevent a delay |
7473 | * in the reporting of vblank events. | |
7474 | */ | |
7a0d1eed | 7475 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS); |
6f1d69b0 ED |
7476 | } |
7477 | ||
46f16e63 | 7478 | static void cherryview_init_clock_gating(struct drm_i915_private *dev_priv) |
a4565da8 | 7479 | { |
232ce337 VS |
7480 | /* WaVSRefCountFullforceMissDisable:chv */ |
7481 | /* WaDSRefCountFullforceMissDisable:chv */ | |
7482 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
7483 | I915_READ(GEN7_FF_THREAD_MODE) & | |
7484 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); | |
acea6f95 VS |
7485 | |
7486 | /* WaDisableSemaphoreAndSyncFlipWait:chv */ | |
7487 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, | |
7488 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); | |
0846697c VS |
7489 | |
7490 | /* WaDisableCSUnitClockGating:chv */ | |
7491 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | | |
7492 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
c631780f VS |
7493 | |
7494 | /* WaDisableSDEUnitClockGating:chv */ | |
7495 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
7496 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
6d50b065 | 7497 | |
450174fe ID |
7498 | /* |
7499 | * WaProgramL3SqcReg1Default:chv | |
7500 | * See gfxspecs/Related Documents/Performance Guide/ | |
7501 | * LSQC Setting Recommendations. | |
7502 | */ | |
7503 | gen8_set_l3sqc_credits(dev_priv, 38, 2); | |
7504 | ||
6d50b065 VS |
7505 | /* |
7506 | * GTT cache may not work with big pages, so if those | |
7507 | * are ever enabled GTT cache may need to be disabled. | |
7508 | */ | |
7509 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); | |
a4565da8 VS |
7510 | } |
7511 | ||
46f16e63 | 7512 | static void g4x_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7513 | { |
6f1d69b0 ED |
7514 | uint32_t dspclk_gate; |
7515 | ||
7516 | I915_WRITE(RENCLK_GATE_D1, 0); | |
7517 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | | |
7518 | GS_UNIT_CLOCK_GATE_DISABLE | | |
7519 | CL_UNIT_CLOCK_GATE_DISABLE); | |
7520 | I915_WRITE(RAMCLK_GATE_D, 0); | |
7521 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | | |
7522 | OVRUNIT_CLOCK_GATE_DISABLE | | |
7523 | OVCUNIT_CLOCK_GATE_DISABLE; | |
50a0bc90 | 7524 | if (IS_GM45(dev_priv)) |
6f1d69b0 ED |
7525 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; |
7526 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); | |
4358a374 DV |
7527 | |
7528 | /* WaDisableRenderCachePipelinedFlush */ | |
7529 | I915_WRITE(CACHE_MODE_0, | |
7530 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
de1aa629 | 7531 | |
4e04632e AG |
7532 | /* WaDisable_RenderCache_OperationalFlush:g4x */ |
7533 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
7534 | ||
46f16e63 | 7535 | g4x_disable_trickle_feed(dev_priv); |
6f1d69b0 ED |
7536 | } |
7537 | ||
46f16e63 | 7538 | static void crestline_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7539 | { |
6f1d69b0 ED |
7540 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); |
7541 | I915_WRITE(RENCLK_GATE_D2, 0); | |
7542 | I915_WRITE(DSPCLK_GATE_D, 0); | |
7543 | I915_WRITE(RAMCLK_GATE_D, 0); | |
7544 | I915_WRITE16(DEUC, 0); | |
20f94967 VS |
7545 | I915_WRITE(MI_ARB_STATE, |
7546 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
4e04632e AG |
7547 | |
7548 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ | |
7549 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6f1d69b0 ED |
7550 | } |
7551 | ||
46f16e63 | 7552 | static void broadwater_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7553 | { |
6f1d69b0 ED |
7554 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | |
7555 | I965_RCC_CLOCK_GATE_DISABLE | | |
7556 | I965_RCPB_CLOCK_GATE_DISABLE | | |
7557 | I965_ISC_CLOCK_GATE_DISABLE | | |
7558 | I965_FBC_CLOCK_GATE_DISABLE); | |
7559 | I915_WRITE(RENCLK_GATE_D2, 0); | |
20f94967 VS |
7560 | I915_WRITE(MI_ARB_STATE, |
7561 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
4e04632e AG |
7562 | |
7563 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ | |
7564 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6f1d69b0 ED |
7565 | } |
7566 | ||
46f16e63 | 7567 | static void gen3_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7568 | { |
6f1d69b0 ED |
7569 | u32 dstate = I915_READ(D_STATE); |
7570 | ||
7571 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | | |
7572 | DSTATE_DOT_CLOCK_GATING; | |
7573 | I915_WRITE(D_STATE, dstate); | |
13a86b85 | 7574 | |
9b1e14f4 | 7575 | if (IS_PINEVIEW(dev_priv)) |
13a86b85 | 7576 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); |
974a3b0f DV |
7577 | |
7578 | /* IIR "flip pending" means done if this bit is set */ | |
7579 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); | |
12fabbcb VS |
7580 | |
7581 | /* interrupts should cause a wake up from C3 */ | |
3299254f | 7582 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN)); |
dbb42748 VS |
7583 | |
7584 | /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ | |
7585 | I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); | |
1038392b VS |
7586 | |
7587 | I915_WRITE(MI_ARB_STATE, | |
7588 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
7589 | } |
7590 | ||
46f16e63 | 7591 | static void i85x_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7592 | { |
6f1d69b0 | 7593 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); |
54e472ae VS |
7594 | |
7595 | /* interrupts should cause a wake up from C3 */ | |
7596 | I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) | | |
7597 | _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE)); | |
1038392b VS |
7598 | |
7599 | I915_WRITE(MEM_MODE, | |
7600 | _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
7601 | } |
7602 | ||
46f16e63 | 7603 | static void i830_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7604 | { |
6f1d69b0 | 7605 | I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE); |
1038392b VS |
7606 | |
7607 | I915_WRITE(MEM_MODE, | |
7608 | _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) | | |
7609 | _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
7610 | } |
7611 | ||
46f16e63 | 7612 | void intel_init_clock_gating(struct drm_i915_private *dev_priv) |
6f1d69b0 | 7613 | { |
46f16e63 | 7614 | dev_priv->display.init_clock_gating(dev_priv); |
6f1d69b0 ED |
7615 | } |
7616 | ||
712bf364 | 7617 | void intel_suspend_hw(struct drm_i915_private *dev_priv) |
7d708ee4 | 7618 | { |
712bf364 VS |
7619 | if (HAS_PCH_LPT(dev_priv)) |
7620 | lpt_suspend_hw(dev_priv); | |
7d708ee4 ID |
7621 | } |
7622 | ||
46f16e63 | 7623 | static void nop_init_clock_gating(struct drm_i915_private *dev_priv) |
bb400da9 ID |
7624 | { |
7625 | DRM_DEBUG_KMS("No clock gating settings or workarounds applied.\n"); | |
7626 | } | |
7627 | ||
7628 | /** | |
7629 | * intel_init_clock_gating_hooks - setup the clock gating hooks | |
7630 | * @dev_priv: device private | |
7631 | * | |
7632 | * Setup the hooks that configure which clocks of a given platform can be | |
7633 | * gated and also apply various GT and display specific workarounds for these | |
7634 | * platforms. Note that some GT specific workarounds are applied separately | |
7635 | * when GPU contexts or batchbuffers start their execution. | |
7636 | */ | |
7637 | void intel_init_clock_gating_hooks(struct drm_i915_private *dev_priv) | |
7638 | { | |
7639 | if (IS_SKYLAKE(dev_priv)) | |
dc00b6a0 | 7640 | dev_priv->display.init_clock_gating = skylake_init_clock_gating; |
bb400da9 | 7641 | else if (IS_KABYLAKE(dev_priv)) |
9498dba7 | 7642 | dev_priv->display.init_clock_gating = kabylake_init_clock_gating; |
bb400da9 ID |
7643 | else if (IS_BROXTON(dev_priv)) |
7644 | dev_priv->display.init_clock_gating = bxt_init_clock_gating; | |
7645 | else if (IS_BROADWELL(dev_priv)) | |
7646 | dev_priv->display.init_clock_gating = broadwell_init_clock_gating; | |
7647 | else if (IS_CHERRYVIEW(dev_priv)) | |
7648 | dev_priv->display.init_clock_gating = cherryview_init_clock_gating; | |
7649 | else if (IS_HASWELL(dev_priv)) | |
7650 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; | |
7651 | else if (IS_IVYBRIDGE(dev_priv)) | |
7652 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; | |
7653 | else if (IS_VALLEYVIEW(dev_priv)) | |
7654 | dev_priv->display.init_clock_gating = valleyview_init_clock_gating; | |
7655 | else if (IS_GEN6(dev_priv)) | |
7656 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; | |
7657 | else if (IS_GEN5(dev_priv)) | |
7658 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; | |
7659 | else if (IS_G4X(dev_priv)) | |
7660 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; | |
7661 | else if (IS_CRESTLINE(dev_priv)) | |
7662 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; | |
7663 | else if (IS_BROADWATER(dev_priv)) | |
7664 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; | |
7665 | else if (IS_GEN3(dev_priv)) | |
7666 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; | |
7667 | else if (IS_I85X(dev_priv) || IS_I865G(dev_priv)) | |
7668 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; | |
7669 | else if (IS_GEN2(dev_priv)) | |
7670 | dev_priv->display.init_clock_gating = i830_init_clock_gating; | |
7671 | else { | |
7672 | MISSING_CASE(INTEL_DEVID(dev_priv)); | |
7673 | dev_priv->display.init_clock_gating = nop_init_clock_gating; | |
7674 | } | |
7675 | } | |
7676 | ||
1fa61106 | 7677 | /* Set up chip specific power management-related functions */ |
62d75df7 | 7678 | void intel_init_pm(struct drm_i915_private *dev_priv) |
1fa61106 | 7679 | { |
7ff0ebcc | 7680 | intel_fbc_init(dev_priv); |
1fa61106 | 7681 | |
c921aba8 | 7682 | /* For cxsr */ |
9b1e14f4 | 7683 | if (IS_PINEVIEW(dev_priv)) |
148ac1f3 | 7684 | i915_pineview_get_mem_freq(dev_priv); |
5db94019 | 7685 | else if (IS_GEN5(dev_priv)) |
148ac1f3 | 7686 | i915_ironlake_get_mem_freq(dev_priv); |
c921aba8 | 7687 | |
1fa61106 | 7688 | /* For FIFO watermark updates */ |
62d75df7 | 7689 | if (INTEL_GEN(dev_priv) >= 9) { |
bb726519 | 7690 | skl_setup_wm_latency(dev_priv); |
e62929b3 | 7691 | dev_priv->display.initial_watermarks = skl_initial_wm; |
ccf010fb | 7692 | dev_priv->display.atomic_update_watermarks = skl_atomic_update_crtc_wm; |
98d39494 | 7693 | dev_priv->display.compute_global_watermarks = skl_compute_wm; |
6e266956 | 7694 | } else if (HAS_PCH_SPLIT(dev_priv)) { |
bb726519 | 7695 | ilk_setup_wm_latency(dev_priv); |
53615a5e | 7696 | |
5db94019 | 7697 | if ((IS_GEN5(dev_priv) && dev_priv->wm.pri_latency[1] && |
bd602544 | 7698 | dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) || |
5db94019 | 7699 | (!IS_GEN5(dev_priv) && dev_priv->wm.pri_latency[0] && |
bd602544 | 7700 | dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) { |
86c8bbbe | 7701 | dev_priv->display.compute_pipe_wm = ilk_compute_pipe_wm; |
ed4a6a7c MR |
7702 | dev_priv->display.compute_intermediate_wm = |
7703 | ilk_compute_intermediate_wm; | |
7704 | dev_priv->display.initial_watermarks = | |
7705 | ilk_initial_watermarks; | |
7706 | dev_priv->display.optimize_watermarks = | |
7707 | ilk_optimize_watermarks; | |
bd602544 VS |
7708 | } else { |
7709 | DRM_DEBUG_KMS("Failed to read display plane latency. " | |
7710 | "Disable CxSR\n"); | |
7711 | } | |
920a14b2 | 7712 | } else if (IS_CHERRYVIEW(dev_priv)) { |
bb726519 | 7713 | vlv_setup_wm_latency(dev_priv); |
262cd2e1 | 7714 | dev_priv->display.update_wm = vlv_update_wm; |
11a914c2 | 7715 | } else if (IS_VALLEYVIEW(dev_priv)) { |
bb726519 | 7716 | vlv_setup_wm_latency(dev_priv); |
26e1fe4f | 7717 | dev_priv->display.update_wm = vlv_update_wm; |
9b1e14f4 | 7718 | } else if (IS_PINEVIEW(dev_priv)) { |
50a0bc90 | 7719 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev_priv), |
1fa61106 ED |
7720 | dev_priv->is_ddr3, |
7721 | dev_priv->fsb_freq, | |
7722 | dev_priv->mem_freq)) { | |
7723 | DRM_INFO("failed to find known CxSR latency " | |
7724 | "(found ddr%s fsb freq %d, mem freq %d), " | |
7725 | "disabling CxSR\n", | |
7726 | (dev_priv->is_ddr3 == 1) ? "3" : "2", | |
7727 | dev_priv->fsb_freq, dev_priv->mem_freq); | |
7728 | /* Disable CxSR and never update its watermark again */ | |
5209b1f4 | 7729 | intel_set_memory_cxsr(dev_priv, false); |
1fa61106 ED |
7730 | dev_priv->display.update_wm = NULL; |
7731 | } else | |
7732 | dev_priv->display.update_wm = pineview_update_wm; | |
9beb5fea | 7733 | } else if (IS_G4X(dev_priv)) { |
1fa61106 | 7734 | dev_priv->display.update_wm = g4x_update_wm; |
5db94019 | 7735 | } else if (IS_GEN4(dev_priv)) { |
1fa61106 | 7736 | dev_priv->display.update_wm = i965_update_wm; |
5db94019 | 7737 | } else if (IS_GEN3(dev_priv)) { |
1fa61106 ED |
7738 | dev_priv->display.update_wm = i9xx_update_wm; |
7739 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; | |
5db94019 | 7740 | } else if (IS_GEN2(dev_priv)) { |
62d75df7 | 7741 | if (INTEL_INFO(dev_priv)->num_pipes == 1) { |
feb56b93 | 7742 | dev_priv->display.update_wm = i845_update_wm; |
1fa61106 | 7743 | dev_priv->display.get_fifo_size = i845_get_fifo_size; |
feb56b93 DV |
7744 | } else { |
7745 | dev_priv->display.update_wm = i9xx_update_wm; | |
1fa61106 | 7746 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
feb56b93 | 7747 | } |
feb56b93 DV |
7748 | } else { |
7749 | DRM_ERROR("unexpected fall-through in intel_init_pm\n"); | |
1fa61106 ED |
7750 | } |
7751 | } | |
7752 | ||
87660502 L |
7753 | static inline int gen6_check_mailbox_status(struct drm_i915_private *dev_priv) |
7754 | { | |
7755 | uint32_t flags = | |
7756 | I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_ERROR_MASK; | |
7757 | ||
7758 | switch (flags) { | |
7759 | case GEN6_PCODE_SUCCESS: | |
7760 | return 0; | |
7761 | case GEN6_PCODE_UNIMPLEMENTED_CMD: | |
7762 | case GEN6_PCODE_ILLEGAL_CMD: | |
7763 | return -ENXIO; | |
7764 | case GEN6_PCODE_MIN_FREQ_TABLE_GT_RATIO_OUT_OF_RANGE: | |
7850d1c3 | 7765 | case GEN7_PCODE_MIN_FREQ_TABLE_GT_RATIO_OUT_OF_RANGE: |
87660502 L |
7766 | return -EOVERFLOW; |
7767 | case GEN6_PCODE_TIMEOUT: | |
7768 | return -ETIMEDOUT; | |
7769 | default: | |
7770 | MISSING_CASE(flags) | |
7771 | return 0; | |
7772 | } | |
7773 | } | |
7774 | ||
7775 | static inline int gen7_check_mailbox_status(struct drm_i915_private *dev_priv) | |
7776 | { | |
7777 | uint32_t flags = | |
7778 | I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_ERROR_MASK; | |
7779 | ||
7780 | switch (flags) { | |
7781 | case GEN6_PCODE_SUCCESS: | |
7782 | return 0; | |
7783 | case GEN6_PCODE_ILLEGAL_CMD: | |
7784 | return -ENXIO; | |
7785 | case GEN7_PCODE_TIMEOUT: | |
7786 | return -ETIMEDOUT; | |
7787 | case GEN7_PCODE_ILLEGAL_DATA: | |
7788 | return -EINVAL; | |
7789 | case GEN7_PCODE_MIN_FREQ_TABLE_GT_RATIO_OUT_OF_RANGE: | |
7790 | return -EOVERFLOW; | |
7791 | default: | |
7792 | MISSING_CASE(flags); | |
7793 | return 0; | |
7794 | } | |
7795 | } | |
7796 | ||
151a49d0 | 7797 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val) |
42c0526c | 7798 | { |
87660502 L |
7799 | int status; |
7800 | ||
4fc688ce | 7801 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c | 7802 | |
3f5582dd CW |
7803 | /* GEN6_PCODE_* are outside of the forcewake domain, we can |
7804 | * use te fw I915_READ variants to reduce the amount of work | |
7805 | * required when reading/writing. | |
7806 | */ | |
7807 | ||
7808 | if (I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
42c0526c BW |
7809 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); |
7810 | return -EAGAIN; | |
7811 | } | |
7812 | ||
3f5582dd CW |
7813 | I915_WRITE_FW(GEN6_PCODE_DATA, *val); |
7814 | I915_WRITE_FW(GEN6_PCODE_DATA1, 0); | |
7815 | I915_WRITE_FW(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); | |
42c0526c | 7816 | |
3f5582dd CW |
7817 | if (intel_wait_for_register_fw(dev_priv, |
7818 | GEN6_PCODE_MAILBOX, GEN6_PCODE_READY, 0, | |
7819 | 500)) { | |
42c0526c BW |
7820 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); |
7821 | return -ETIMEDOUT; | |
7822 | } | |
7823 | ||
3f5582dd CW |
7824 | *val = I915_READ_FW(GEN6_PCODE_DATA); |
7825 | I915_WRITE_FW(GEN6_PCODE_DATA, 0); | |
42c0526c | 7826 | |
87660502 L |
7827 | if (INTEL_GEN(dev_priv) > 6) |
7828 | status = gen7_check_mailbox_status(dev_priv); | |
7829 | else | |
7830 | status = gen6_check_mailbox_status(dev_priv); | |
7831 | ||
7832 | if (status) { | |
7833 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed: %d\n", | |
7834 | status); | |
7835 | return status; | |
7836 | } | |
7837 | ||
42c0526c BW |
7838 | return 0; |
7839 | } | |
7840 | ||
3f5582dd | 7841 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, |
87660502 | 7842 | u32 mbox, u32 val) |
42c0526c | 7843 | { |
87660502 L |
7844 | int status; |
7845 | ||
4fc688ce | 7846 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c | 7847 | |
3f5582dd CW |
7848 | /* GEN6_PCODE_* are outside of the forcewake domain, we can |
7849 | * use te fw I915_READ variants to reduce the amount of work | |
7850 | * required when reading/writing. | |
7851 | */ | |
7852 | ||
7853 | if (I915_READ_FW(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
42c0526c BW |
7854 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); |
7855 | return -EAGAIN; | |
7856 | } | |
7857 | ||
3f5582dd CW |
7858 | I915_WRITE_FW(GEN6_PCODE_DATA, val); |
7859 | I915_WRITE_FW(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); | |
42c0526c | 7860 | |
3f5582dd CW |
7861 | if (intel_wait_for_register_fw(dev_priv, |
7862 | GEN6_PCODE_MAILBOX, GEN6_PCODE_READY, 0, | |
7863 | 500)) { | |
42c0526c BW |
7864 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); |
7865 | return -ETIMEDOUT; | |
7866 | } | |
7867 | ||
3f5582dd | 7868 | I915_WRITE_FW(GEN6_PCODE_DATA, 0); |
42c0526c | 7869 | |
87660502 L |
7870 | if (INTEL_GEN(dev_priv) > 6) |
7871 | status = gen7_check_mailbox_status(dev_priv); | |
7872 | else | |
7873 | status = gen6_check_mailbox_status(dev_priv); | |
7874 | ||
7875 | if (status) { | |
7876 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed: %d\n", | |
7877 | status); | |
7878 | return status; | |
7879 | } | |
7880 | ||
42c0526c BW |
7881 | return 0; |
7882 | } | |
a0e4e199 | 7883 | |
2c7d0602 ID |
7884 | static bool skl_pcode_try_request(struct drm_i915_private *dev_priv, u32 mbox, |
7885 | u32 request, u32 reply_mask, u32 reply, | |
7886 | u32 *status) | |
7887 | { | |
7888 | u32 val = request; | |
7889 | ||
7890 | *status = sandybridge_pcode_read(dev_priv, mbox, &val); | |
7891 | ||
7892 | return *status || ((val & reply_mask) == reply); | |
7893 | } | |
7894 | ||
7895 | /** | |
7896 | * skl_pcode_request - send PCODE request until acknowledgment | |
7897 | * @dev_priv: device private | |
7898 | * @mbox: PCODE mailbox ID the request is targeted for | |
7899 | * @request: request ID | |
7900 | * @reply_mask: mask used to check for request acknowledgment | |
7901 | * @reply: value used to check for request acknowledgment | |
7902 | * @timeout_base_ms: timeout for polling with preemption enabled | |
7903 | * | |
7904 | * Keep resending the @request to @mbox until PCODE acknowledges it, PCODE | |
ffd98b12 | 7905 | * reports an error or an overall timeout of @timeout_base_ms+50 ms expires. |
2c7d0602 ID |
7906 | * The request is acknowledged once the PCODE reply dword equals @reply after |
7907 | * applying @reply_mask. Polling is first attempted with preemption enabled | |
ffd98b12 | 7908 | * for @timeout_base_ms and if this times out for another 50 ms with |
2c7d0602 ID |
7909 | * preemption disabled. |
7910 | * | |
7911 | * Returns 0 on success, %-ETIMEDOUT in case of a timeout, <0 in case of some | |
7912 | * other error as reported by PCODE. | |
7913 | */ | |
7914 | int skl_pcode_request(struct drm_i915_private *dev_priv, u32 mbox, u32 request, | |
7915 | u32 reply_mask, u32 reply, int timeout_base_ms) | |
7916 | { | |
7917 | u32 status; | |
7918 | int ret; | |
7919 | ||
7920 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
7921 | ||
7922 | #define COND skl_pcode_try_request(dev_priv, mbox, request, reply_mask, reply, \ | |
7923 | &status) | |
7924 | ||
7925 | /* | |
7926 | * Prime the PCODE by doing a request first. Normally it guarantees | |
7927 | * that a subsequent request, at most @timeout_base_ms later, succeeds. | |
7928 | * _wait_for() doesn't guarantee when its passed condition is evaluated | |
7929 | * first, so send the first request explicitly. | |
7930 | */ | |
7931 | if (COND) { | |
7932 | ret = 0; | |
7933 | goto out; | |
7934 | } | |
7935 | ret = _wait_for(COND, timeout_base_ms * 1000, 10); | |
7936 | if (!ret) | |
7937 | goto out; | |
7938 | ||
7939 | /* | |
7940 | * The above can time out if the number of requests was low (2 in the | |
7941 | * worst case) _and_ PCODE was busy for some reason even after a | |
7942 | * (queued) request and @timeout_base_ms delay. As a workaround retry | |
7943 | * the poll with preemption disabled to maximize the number of | |
ffd98b12 | 7944 | * requests. Increase the timeout from @timeout_base_ms to 50ms to |
2c7d0602 | 7945 | * account for interrupts that could reduce the number of these |
ffd98b12 ID |
7946 | * requests, and for any quirks of the PCODE firmware that delays |
7947 | * the request completion. | |
2c7d0602 ID |
7948 | */ |
7949 | DRM_DEBUG_KMS("PCODE timeout, retrying with preemption disabled\n"); | |
7950 | WARN_ON_ONCE(timeout_base_ms > 3); | |
7951 | preempt_disable(); | |
ffd98b12 | 7952 | ret = wait_for_atomic(COND, 50); |
2c7d0602 ID |
7953 | preempt_enable(); |
7954 | ||
7955 | out: | |
7956 | return ret ? ret : status; | |
7957 | #undef COND | |
7958 | } | |
7959 | ||
dd06f88c VS |
7960 | static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7961 | { | |
c30fec65 VS |
7962 | /* |
7963 | * N = val - 0xb7 | |
7964 | * Slow = Fast = GPLL ref * N | |
7965 | */ | |
7966 | return DIV_ROUND_CLOSEST(dev_priv->rps.gpll_ref_freq * (val - 0xb7), 1000); | |
855ba3be JB |
7967 | } |
7968 | ||
b55dd647 | 7969 | static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val) |
855ba3be | 7970 | { |
c30fec65 | 7971 | return DIV_ROUND_CLOSEST(1000 * val, dev_priv->rps.gpll_ref_freq) + 0xb7; |
855ba3be JB |
7972 | } |
7973 | ||
b55dd647 | 7974 | static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 7975 | { |
c30fec65 VS |
7976 | /* |
7977 | * N = val / 2 | |
7978 | * CU (slow) = CU2x (fast) / 2 = GPLL ref * N / 2 | |
7979 | */ | |
7980 | return DIV_ROUND_CLOSEST(dev_priv->rps.gpll_ref_freq * val, 2 * 2 * 1000); | |
22b1b2f8 D |
7981 | } |
7982 | ||
b55dd647 | 7983 | static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 7984 | { |
1c14762d | 7985 | /* CHV needs even values */ |
c30fec65 | 7986 | return DIV_ROUND_CLOSEST(2 * 1000 * val, dev_priv->rps.gpll_ref_freq) * 2; |
22b1b2f8 D |
7987 | } |
7988 | ||
616bc820 | 7989 | int intel_gpu_freq(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 7990 | { |
2d1fe073 | 7991 | if (IS_GEN9(dev_priv)) |
500a3d2e MK |
7992 | return DIV_ROUND_CLOSEST(val * GT_FREQUENCY_MULTIPLIER, |
7993 | GEN9_FREQ_SCALER); | |
2d1fe073 | 7994 | else if (IS_CHERRYVIEW(dev_priv)) |
616bc820 | 7995 | return chv_gpu_freq(dev_priv, val); |
2d1fe073 | 7996 | else if (IS_VALLEYVIEW(dev_priv)) |
616bc820 VS |
7997 | return byt_gpu_freq(dev_priv, val); |
7998 | else | |
7999 | return val * GT_FREQUENCY_MULTIPLIER; | |
22b1b2f8 D |
8000 | } |
8001 | ||
616bc820 VS |
8002 | int intel_freq_opcode(struct drm_i915_private *dev_priv, int val) |
8003 | { | |
2d1fe073 | 8004 | if (IS_GEN9(dev_priv)) |
500a3d2e MK |
8005 | return DIV_ROUND_CLOSEST(val * GEN9_FREQ_SCALER, |
8006 | GT_FREQUENCY_MULTIPLIER); | |
2d1fe073 | 8007 | else if (IS_CHERRYVIEW(dev_priv)) |
616bc820 | 8008 | return chv_freq_opcode(dev_priv, val); |
2d1fe073 | 8009 | else if (IS_VALLEYVIEW(dev_priv)) |
616bc820 VS |
8010 | return byt_freq_opcode(dev_priv, val); |
8011 | else | |
500a3d2e | 8012 | return DIV_ROUND_CLOSEST(val, GT_FREQUENCY_MULTIPLIER); |
616bc820 | 8013 | } |
22b1b2f8 | 8014 | |
6ad790c0 CW |
8015 | struct request_boost { |
8016 | struct work_struct work; | |
eed29a5b | 8017 | struct drm_i915_gem_request *req; |
6ad790c0 CW |
8018 | }; |
8019 | ||
8020 | static void __intel_rps_boost_work(struct work_struct *work) | |
8021 | { | |
8022 | struct request_boost *boost = container_of(work, struct request_boost, work); | |
e61b9958 | 8023 | struct drm_i915_gem_request *req = boost->req; |
6ad790c0 | 8024 | |
f69a02c9 | 8025 | if (!i915_gem_request_completed(req)) |
c033666a | 8026 | gen6_rps_boost(req->i915, NULL, req->emitted_jiffies); |
6ad790c0 | 8027 | |
e8a261ea | 8028 | i915_gem_request_put(req); |
6ad790c0 CW |
8029 | kfree(boost); |
8030 | } | |
8031 | ||
91d14251 | 8032 | void intel_queue_rps_boost_for_request(struct drm_i915_gem_request *req) |
6ad790c0 CW |
8033 | { |
8034 | struct request_boost *boost; | |
8035 | ||
91d14251 | 8036 | if (req == NULL || INTEL_GEN(req->i915) < 6) |
6ad790c0 CW |
8037 | return; |
8038 | ||
f69a02c9 | 8039 | if (i915_gem_request_completed(req)) |
e61b9958 CW |
8040 | return; |
8041 | ||
6ad790c0 CW |
8042 | boost = kmalloc(sizeof(*boost), GFP_ATOMIC); |
8043 | if (boost == NULL) | |
8044 | return; | |
8045 | ||
e8a261ea | 8046 | boost->req = i915_gem_request_get(req); |
6ad790c0 CW |
8047 | |
8048 | INIT_WORK(&boost->work, __intel_rps_boost_work); | |
91d14251 | 8049 | queue_work(req->i915->wq, &boost->work); |
6ad790c0 CW |
8050 | } |
8051 | ||
f742a552 | 8052 | void intel_pm_setup(struct drm_device *dev) |
907b28c5 | 8053 | { |
fac5e23e | 8054 | struct drm_i915_private *dev_priv = to_i915(dev); |
907b28c5 | 8055 | |
f742a552 | 8056 | mutex_init(&dev_priv->rps.hw_lock); |
8d3afd7d | 8057 | spin_lock_init(&dev_priv->rps.client_lock); |
f742a552 | 8058 | |
54b4f68f CW |
8059 | INIT_DELAYED_WORK(&dev_priv->rps.autoenable_work, |
8060 | __intel_autoenable_gt_powersave); | |
1854d5ca | 8061 | INIT_LIST_HEAD(&dev_priv->rps.clients); |
5d584b2e | 8062 | |
33688d95 | 8063 | dev_priv->pm.suspended = false; |
1f814dac | 8064 | atomic_set(&dev_priv->pm.wakeref_count, 0); |
907b28c5 | 8065 | } |