]>
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> |
85208be0 ED |
29 | #include "i915_drv.h" |
30 | #include "intel_drv.h" | |
eb48eb00 DV |
31 | #include "../../../platform/x86/intel_ips.h" |
32 | #include <linux/module.h> | |
85208be0 | 33 | |
dc39fff7 BW |
34 | /** |
35 | * RC6 is a special power stage which allows the GPU to enter an very | |
36 | * low-voltage mode when idle, using down to 0V while at this stage. This | |
37 | * stage is entered automatically when the GPU is idle when RC6 support is | |
38 | * enabled, and as soon as new workload arises GPU wakes up automatically as well. | |
39 | * | |
40 | * There are different RC6 modes available in Intel GPU, which differentiate | |
41 | * among each other with the latency required to enter and leave RC6 and | |
42 | * voltage consumed by the GPU in different states. | |
43 | * | |
44 | * The combination of the following flags define which states GPU is allowed | |
45 | * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and | |
46 | * RC6pp is deepest RC6. Their support by hardware varies according to the | |
47 | * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one | |
48 | * which brings the most power savings; deeper states save more power, but | |
49 | * require higher latency to switch to and wake up. | |
50 | */ | |
51 | #define INTEL_RC6_ENABLE (1<<0) | |
52 | #define INTEL_RC6p_ENABLE (1<<1) | |
53 | #define INTEL_RC6pp_ENABLE (1<<2) | |
54 | ||
da2078cd DL |
55 | static void gen9_init_clock_gating(struct drm_device *dev) |
56 | { | |
acd5c346 DL |
57 | struct drm_i915_private *dev_priv = dev->dev_private; |
58 | ||
77719d28 DL |
59 | /* WaEnableLbsSlaRetryTimerDecrement:skl */ |
60 | I915_WRITE(BDW_SCRATCH1, I915_READ(BDW_SCRATCH1) | | |
61 | GEN9_LBS_SLA_RETRY_TIMER_DECREMENT_ENABLE); | |
6381b550 NH |
62 | |
63 | /* WaDisableKillLogic:bxt,skl */ | |
64 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | | |
65 | ECOCHK_DIS_TLB); | |
77719d28 | 66 | } |
91e41d16 | 67 | |
45db2194 | 68 | static void skl_init_clock_gating(struct drm_device *dev) |
da2078cd | 69 | { |
acd5c346 | 70 | struct drm_i915_private *dev_priv = dev->dev_private; |
3ca5da43 | 71 | |
77719d28 DL |
72 | gen9_init_clock_gating(dev); |
73 | ||
669506e7 | 74 | if (INTEL_REVID(dev) <= SKL_REVID_B0) { |
3dcd020a HN |
75 | /* |
76 | * WaDisableSDEUnitClockGating:skl | |
9253c2e5 | 77 | * WaSetGAPSunitClckGateDisable:skl |
3dcd020a HN |
78 | */ |
79 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
9253c2e5 | 80 | GEN8_GAPSUNIT_CLOCK_GATE_DISABLE | |
3dcd020a | 81 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); |
f9fc42f4 DL |
82 | |
83 | /* WaDisableVFUnitClockGating:skl */ | |
84 | I915_WRITE(GEN6_UCGCTL2, I915_READ(GEN6_UCGCTL2) | | |
85 | GEN6_VFUNIT_CLOCK_GATE_DISABLE); | |
3dcd020a | 86 | } |
8bc0ccf6 | 87 | |
2caa3b26 | 88 | if (INTEL_REVID(dev) <= SKL_REVID_D0) { |
81e231af DL |
89 | /* WaDisableHDCInvalidation:skl */ |
90 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | | |
91 | BDW_DISABLE_HDC_INVALIDATION); | |
92 | ||
2caa3b26 DL |
93 | /* WaDisableChickenBitTSGBarrierAckForFFSliceCS:skl */ |
94 | I915_WRITE(FF_SLICE_CS_CHICKEN2, | |
f1d3d34d | 95 | _MASKED_BIT_ENABLE(GEN9_TSG_BARRIER_ACK_DISABLE)); |
2caa3b26 | 96 | } |
81e231af | 97 | |
a4106a78 AS |
98 | /* GEN8_L3SQCREG4 has a dependency with WA batch so any new changes |
99 | * involving this register should also be added to WA batch as required. | |
100 | */ | |
8bc0ccf6 DL |
101 | if (INTEL_REVID(dev) <= SKL_REVID_E0) |
102 | /* WaDisableLSQCROPERFforOCL:skl */ | |
103 | I915_WRITE(GEN8_L3SQCREG4, I915_READ(GEN8_L3SQCREG4) | | |
104 | GEN8_LQSC_RO_PERF_DIS); | |
da2078cd DL |
105 | } |
106 | ||
a82abe43 ID |
107 | static void bxt_init_clock_gating(struct drm_device *dev) |
108 | { | |
32608ca2 ID |
109 | struct drm_i915_private *dev_priv = dev->dev_private; |
110 | ||
a82abe43 | 111 | gen9_init_clock_gating(dev); |
32608ca2 ID |
112 | |
113 | /* | |
114 | * FIXME: | |
115 | * GEN8_SDEUNIT_CLOCK_GATE_DISABLE applies on A0 only. | |
868434c5 | 116 | * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only. |
32608ca2 ID |
117 | */ |
118 | /* WaDisableSDEUnitClockGating:bxt */ | |
119 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
868434c5 BW |
120 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE | |
121 | GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ); | |
32608ca2 | 122 | |
e3a29055 RB |
123 | /* FIXME: apply on A0 only */ |
124 | I915_WRITE(TILECTL, I915_READ(TILECTL) | TILECTL_TLBPF); | |
a82abe43 ID |
125 | } |
126 | ||
c921aba8 DV |
127 | static void i915_pineview_get_mem_freq(struct drm_device *dev) |
128 | { | |
50227e1c | 129 | struct drm_i915_private *dev_priv = dev->dev_private; |
c921aba8 DV |
130 | u32 tmp; |
131 | ||
132 | tmp = I915_READ(CLKCFG); | |
133 | ||
134 | switch (tmp & CLKCFG_FSB_MASK) { | |
135 | case CLKCFG_FSB_533: | |
136 | dev_priv->fsb_freq = 533; /* 133*4 */ | |
137 | break; | |
138 | case CLKCFG_FSB_800: | |
139 | dev_priv->fsb_freq = 800; /* 200*4 */ | |
140 | break; | |
141 | case CLKCFG_FSB_667: | |
142 | dev_priv->fsb_freq = 667; /* 167*4 */ | |
143 | break; | |
144 | case CLKCFG_FSB_400: | |
145 | dev_priv->fsb_freq = 400; /* 100*4 */ | |
146 | break; | |
147 | } | |
148 | ||
149 | switch (tmp & CLKCFG_MEM_MASK) { | |
150 | case CLKCFG_MEM_533: | |
151 | dev_priv->mem_freq = 533; | |
152 | break; | |
153 | case CLKCFG_MEM_667: | |
154 | dev_priv->mem_freq = 667; | |
155 | break; | |
156 | case CLKCFG_MEM_800: | |
157 | dev_priv->mem_freq = 800; | |
158 | break; | |
159 | } | |
160 | ||
161 | /* detect pineview DDR3 setting */ | |
162 | tmp = I915_READ(CSHRDDR3CTL); | |
163 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; | |
164 | } | |
165 | ||
166 | static void i915_ironlake_get_mem_freq(struct drm_device *dev) | |
167 | { | |
50227e1c | 168 | struct drm_i915_private *dev_priv = dev->dev_private; |
c921aba8 DV |
169 | u16 ddrpll, csipll; |
170 | ||
171 | ddrpll = I915_READ16(DDRMPLL1); | |
172 | csipll = I915_READ16(CSIPLL0); | |
173 | ||
174 | switch (ddrpll & 0xff) { | |
175 | case 0xc: | |
176 | dev_priv->mem_freq = 800; | |
177 | break; | |
178 | case 0x10: | |
179 | dev_priv->mem_freq = 1066; | |
180 | break; | |
181 | case 0x14: | |
182 | dev_priv->mem_freq = 1333; | |
183 | break; | |
184 | case 0x18: | |
185 | dev_priv->mem_freq = 1600; | |
186 | break; | |
187 | default: | |
188 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", | |
189 | ddrpll & 0xff); | |
190 | dev_priv->mem_freq = 0; | |
191 | break; | |
192 | } | |
193 | ||
20e4d407 | 194 | dev_priv->ips.r_t = dev_priv->mem_freq; |
c921aba8 DV |
195 | |
196 | switch (csipll & 0x3ff) { | |
197 | case 0x00c: | |
198 | dev_priv->fsb_freq = 3200; | |
199 | break; | |
200 | case 0x00e: | |
201 | dev_priv->fsb_freq = 3733; | |
202 | break; | |
203 | case 0x010: | |
204 | dev_priv->fsb_freq = 4266; | |
205 | break; | |
206 | case 0x012: | |
207 | dev_priv->fsb_freq = 4800; | |
208 | break; | |
209 | case 0x014: | |
210 | dev_priv->fsb_freq = 5333; | |
211 | break; | |
212 | case 0x016: | |
213 | dev_priv->fsb_freq = 5866; | |
214 | break; | |
215 | case 0x018: | |
216 | dev_priv->fsb_freq = 6400; | |
217 | break; | |
218 | default: | |
219 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", | |
220 | csipll & 0x3ff); | |
221 | dev_priv->fsb_freq = 0; | |
222 | break; | |
223 | } | |
224 | ||
225 | if (dev_priv->fsb_freq == 3200) { | |
20e4d407 | 226 | dev_priv->ips.c_m = 0; |
c921aba8 | 227 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
20e4d407 | 228 | dev_priv->ips.c_m = 1; |
c921aba8 | 229 | } else { |
20e4d407 | 230 | dev_priv->ips.c_m = 2; |
c921aba8 DV |
231 | } |
232 | } | |
233 | ||
b445e3b0 ED |
234 | static const struct cxsr_latency cxsr_latency_table[] = { |
235 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ | |
236 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ | |
237 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ | |
238 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ | |
239 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ | |
240 | ||
241 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ | |
242 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ | |
243 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ | |
244 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ | |
245 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ | |
246 | ||
247 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ | |
248 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ | |
249 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ | |
250 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ | |
251 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ | |
252 | ||
253 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ | |
254 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ | |
255 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ | |
256 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ | |
257 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ | |
258 | ||
259 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ | |
260 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ | |
261 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ | |
262 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ | |
263 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ | |
264 | ||
265 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ | |
266 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ | |
267 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ | |
268 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ | |
269 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ | |
270 | }; | |
271 | ||
63c62275 | 272 | static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop, |
b445e3b0 ED |
273 | int is_ddr3, |
274 | int fsb, | |
275 | int mem) | |
276 | { | |
277 | const struct cxsr_latency *latency; | |
278 | int i; | |
279 | ||
280 | if (fsb == 0 || mem == 0) | |
281 | return NULL; | |
282 | ||
283 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { | |
284 | latency = &cxsr_latency_table[i]; | |
285 | if (is_desktop == latency->is_desktop && | |
286 | is_ddr3 == latency->is_ddr3 && | |
287 | fsb == latency->fsb_freq && mem == latency->mem_freq) | |
288 | return latency; | |
289 | } | |
290 | ||
291 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
292 | ||
293 | return NULL; | |
294 | } | |
295 | ||
fc1ac8de VS |
296 | static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable) |
297 | { | |
298 | u32 val; | |
299 | ||
300 | mutex_lock(&dev_priv->rps.hw_lock); | |
301 | ||
302 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); | |
303 | if (enable) | |
304 | val &= ~FORCE_DDR_HIGH_FREQ; | |
305 | else | |
306 | val |= FORCE_DDR_HIGH_FREQ; | |
307 | val &= ~FORCE_DDR_LOW_FREQ; | |
308 | val |= FORCE_DDR_FREQ_REQ_ACK; | |
309 | vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val); | |
310 | ||
311 | if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) & | |
312 | FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) | |
313 | DRM_ERROR("timed out waiting for Punit DDR DVFS request\n"); | |
314 | ||
315 | mutex_unlock(&dev_priv->rps.hw_lock); | |
316 | } | |
317 | ||
cfb41411 VS |
318 | static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable) |
319 | { | |
320 | u32 val; | |
321 | ||
322 | mutex_lock(&dev_priv->rps.hw_lock); | |
323 | ||
324 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); | |
325 | if (enable) | |
326 | val |= DSP_MAXFIFO_PM5_ENABLE; | |
327 | else | |
328 | val &= ~DSP_MAXFIFO_PM5_ENABLE; | |
329 | vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val); | |
330 | ||
331 | mutex_unlock(&dev_priv->rps.hw_lock); | |
332 | } | |
333 | ||
f4998963 VS |
334 | #define FW_WM(value, plane) \ |
335 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK) | |
336 | ||
5209b1f4 | 337 | void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable) |
b445e3b0 | 338 | { |
5209b1f4 ID |
339 | struct drm_device *dev = dev_priv->dev; |
340 | u32 val; | |
b445e3b0 | 341 | |
5209b1f4 ID |
342 | if (IS_VALLEYVIEW(dev)) { |
343 | I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0); | |
a7a6c498 | 344 | POSTING_READ(FW_BLC_SELF_VLV); |
852eb00d | 345 | dev_priv->wm.vlv.cxsr = enable; |
5209b1f4 ID |
346 | } else if (IS_G4X(dev) || IS_CRESTLINE(dev)) { |
347 | I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0); | |
a7a6c498 | 348 | POSTING_READ(FW_BLC_SELF); |
5209b1f4 ID |
349 | } else if (IS_PINEVIEW(dev)) { |
350 | val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN; | |
351 | val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0; | |
352 | I915_WRITE(DSPFW3, val); | |
a7a6c498 | 353 | POSTING_READ(DSPFW3); |
5209b1f4 ID |
354 | } else if (IS_I945G(dev) || IS_I945GM(dev)) { |
355 | val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) : | |
356 | _MASKED_BIT_DISABLE(FW_BLC_SELF_EN); | |
357 | I915_WRITE(FW_BLC_SELF, val); | |
a7a6c498 | 358 | POSTING_READ(FW_BLC_SELF); |
5209b1f4 ID |
359 | } else if (IS_I915GM(dev)) { |
360 | val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) : | |
361 | _MASKED_BIT_DISABLE(INSTPM_SELF_EN); | |
362 | I915_WRITE(INSTPM, val); | |
a7a6c498 | 363 | POSTING_READ(INSTPM); |
5209b1f4 ID |
364 | } else { |
365 | return; | |
366 | } | |
b445e3b0 | 367 | |
5209b1f4 ID |
368 | DRM_DEBUG_KMS("memory self-refresh is %s\n", |
369 | enable ? "enabled" : "disabled"); | |
b445e3b0 ED |
370 | } |
371 | ||
fc1ac8de | 372 | |
b445e3b0 ED |
373 | /* |
374 | * Latency for FIFO fetches is dependent on several factors: | |
375 | * - memory configuration (speed, channels) | |
376 | * - chipset | |
377 | * - current MCH state | |
378 | * It can be fairly high in some situations, so here we assume a fairly | |
379 | * pessimal value. It's a tradeoff between extra memory fetches (if we | |
380 | * set this value too high, the FIFO will fetch frequently to stay full) | |
381 | * and power consumption (set it too low to save power and we might see | |
382 | * FIFO underruns and display "flicker"). | |
383 | * | |
384 | * A value of 5us seems to be a good balance; safe for very low end | |
385 | * platforms but not overly aggressive on lower latency configs. | |
386 | */ | |
5aef6003 | 387 | static const int pessimal_latency_ns = 5000; |
b445e3b0 | 388 | |
b5004720 VS |
389 | #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \ |
390 | ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8)) | |
391 | ||
392 | static int vlv_get_fifo_size(struct drm_device *dev, | |
393 | enum pipe pipe, int plane) | |
394 | { | |
395 | struct drm_i915_private *dev_priv = dev->dev_private; | |
396 | int sprite0_start, sprite1_start, size; | |
397 | ||
398 | switch (pipe) { | |
399 | uint32_t dsparb, dsparb2, dsparb3; | |
400 | case PIPE_A: | |
401 | dsparb = I915_READ(DSPARB); | |
402 | dsparb2 = I915_READ(DSPARB2); | |
403 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0); | |
404 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4); | |
405 | break; | |
406 | case PIPE_B: | |
407 | dsparb = I915_READ(DSPARB); | |
408 | dsparb2 = I915_READ(DSPARB2); | |
409 | sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8); | |
410 | sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12); | |
411 | break; | |
412 | case PIPE_C: | |
413 | dsparb2 = I915_READ(DSPARB2); | |
414 | dsparb3 = I915_READ(DSPARB3); | |
415 | sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16); | |
416 | sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20); | |
417 | break; | |
418 | default: | |
419 | return 0; | |
420 | } | |
421 | ||
422 | switch (plane) { | |
423 | case 0: | |
424 | size = sprite0_start; | |
425 | break; | |
426 | case 1: | |
427 | size = sprite1_start - sprite0_start; | |
428 | break; | |
429 | case 2: | |
430 | size = 512 - 1 - sprite1_start; | |
431 | break; | |
432 | default: | |
433 | return 0; | |
434 | } | |
435 | ||
436 | DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n", | |
437 | pipe_name(pipe), plane == 0 ? "primary" : "sprite", | |
438 | plane == 0 ? plane_name(pipe) : sprite_name(pipe, plane - 1), | |
439 | size); | |
440 | ||
441 | return size; | |
442 | } | |
443 | ||
1fa61106 | 444 | static int i9xx_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
445 | { |
446 | struct drm_i915_private *dev_priv = dev->dev_private; | |
447 | uint32_t dsparb = I915_READ(DSPARB); | |
448 | int size; | |
449 | ||
450 | size = dsparb & 0x7f; | |
451 | if (plane) | |
452 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; | |
453 | ||
454 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
455 | plane ? "B" : "A", size); | |
456 | ||
457 | return size; | |
458 | } | |
459 | ||
feb56b93 | 460 | static int i830_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
461 | { |
462 | struct drm_i915_private *dev_priv = dev->dev_private; | |
463 | uint32_t dsparb = I915_READ(DSPARB); | |
464 | int size; | |
465 | ||
466 | size = dsparb & 0x1ff; | |
467 | if (plane) | |
468 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; | |
469 | size >>= 1; /* Convert to cachelines */ | |
470 | ||
471 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
472 | plane ? "B" : "A", size); | |
473 | ||
474 | return size; | |
475 | } | |
476 | ||
1fa61106 | 477 | static int i845_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
478 | { |
479 | struct drm_i915_private *dev_priv = dev->dev_private; | |
480 | uint32_t dsparb = I915_READ(DSPARB); | |
481 | int size; | |
482 | ||
483 | size = dsparb & 0x7f; | |
484 | size >>= 2; /* Convert to cachelines */ | |
485 | ||
486 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
487 | plane ? "B" : "A", | |
488 | size); | |
489 | ||
490 | return size; | |
491 | } | |
492 | ||
b445e3b0 ED |
493 | /* Pineview has different values for various configs */ |
494 | static const struct intel_watermark_params pineview_display_wm = { | |
e0f0273e VS |
495 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
496 | .max_wm = PINEVIEW_MAX_WM, | |
497 | .default_wm = PINEVIEW_DFT_WM, | |
498 | .guard_size = PINEVIEW_GUARD_WM, | |
499 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
500 | }; |
501 | static const struct intel_watermark_params pineview_display_hplloff_wm = { | |
e0f0273e VS |
502 | .fifo_size = PINEVIEW_DISPLAY_FIFO, |
503 | .max_wm = PINEVIEW_MAX_WM, | |
504 | .default_wm = PINEVIEW_DFT_HPLLOFF_WM, | |
505 | .guard_size = PINEVIEW_GUARD_WM, | |
506 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
507 | }; |
508 | static const struct intel_watermark_params pineview_cursor_wm = { | |
e0f0273e VS |
509 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
510 | .max_wm = PINEVIEW_CURSOR_MAX_WM, | |
511 | .default_wm = PINEVIEW_CURSOR_DFT_WM, | |
512 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, | |
513 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
514 | }; |
515 | static const struct intel_watermark_params pineview_cursor_hplloff_wm = { | |
e0f0273e VS |
516 | .fifo_size = PINEVIEW_CURSOR_FIFO, |
517 | .max_wm = PINEVIEW_CURSOR_MAX_WM, | |
518 | .default_wm = PINEVIEW_CURSOR_DFT_WM, | |
519 | .guard_size = PINEVIEW_CURSOR_GUARD_WM, | |
520 | .cacheline_size = PINEVIEW_FIFO_LINE_SIZE, | |
b445e3b0 ED |
521 | }; |
522 | static const struct intel_watermark_params g4x_wm_info = { | |
e0f0273e VS |
523 | .fifo_size = G4X_FIFO_SIZE, |
524 | .max_wm = G4X_MAX_WM, | |
525 | .default_wm = G4X_MAX_WM, | |
526 | .guard_size = 2, | |
527 | .cacheline_size = G4X_FIFO_LINE_SIZE, | |
b445e3b0 ED |
528 | }; |
529 | static const struct intel_watermark_params g4x_cursor_wm_info = { | |
e0f0273e VS |
530 | .fifo_size = I965_CURSOR_FIFO, |
531 | .max_wm = I965_CURSOR_MAX_WM, | |
532 | .default_wm = I965_CURSOR_DFT_WM, | |
533 | .guard_size = 2, | |
534 | .cacheline_size = G4X_FIFO_LINE_SIZE, | |
b445e3b0 ED |
535 | }; |
536 | static const struct intel_watermark_params valleyview_wm_info = { | |
e0f0273e VS |
537 | .fifo_size = VALLEYVIEW_FIFO_SIZE, |
538 | .max_wm = VALLEYVIEW_MAX_WM, | |
539 | .default_wm = VALLEYVIEW_MAX_WM, | |
540 | .guard_size = 2, | |
541 | .cacheline_size = G4X_FIFO_LINE_SIZE, | |
b445e3b0 ED |
542 | }; |
543 | static const struct intel_watermark_params valleyview_cursor_wm_info = { | |
e0f0273e VS |
544 | .fifo_size = I965_CURSOR_FIFO, |
545 | .max_wm = VALLEYVIEW_CURSOR_MAX_WM, | |
546 | .default_wm = I965_CURSOR_DFT_WM, | |
547 | .guard_size = 2, | |
548 | .cacheline_size = G4X_FIFO_LINE_SIZE, | |
b445e3b0 ED |
549 | }; |
550 | static const struct intel_watermark_params i965_cursor_wm_info = { | |
e0f0273e VS |
551 | .fifo_size = I965_CURSOR_FIFO, |
552 | .max_wm = I965_CURSOR_MAX_WM, | |
553 | .default_wm = I965_CURSOR_DFT_WM, | |
554 | .guard_size = 2, | |
555 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 ED |
556 | }; |
557 | static const struct intel_watermark_params i945_wm_info = { | |
e0f0273e VS |
558 | .fifo_size = I945_FIFO_SIZE, |
559 | .max_wm = I915_MAX_WM, | |
560 | .default_wm = 1, | |
561 | .guard_size = 2, | |
562 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 ED |
563 | }; |
564 | static const struct intel_watermark_params i915_wm_info = { | |
e0f0273e VS |
565 | .fifo_size = I915_FIFO_SIZE, |
566 | .max_wm = I915_MAX_WM, | |
567 | .default_wm = 1, | |
568 | .guard_size = 2, | |
569 | .cacheline_size = I915_FIFO_LINE_SIZE, | |
b445e3b0 | 570 | }; |
9d539105 | 571 | static const struct intel_watermark_params i830_a_wm_info = { |
e0f0273e VS |
572 | .fifo_size = I855GM_FIFO_SIZE, |
573 | .max_wm = I915_MAX_WM, | |
574 | .default_wm = 1, | |
575 | .guard_size = 2, | |
576 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
b445e3b0 | 577 | }; |
9d539105 VS |
578 | static const struct intel_watermark_params i830_bc_wm_info = { |
579 | .fifo_size = I855GM_FIFO_SIZE, | |
580 | .max_wm = I915_MAX_WM/2, | |
581 | .default_wm = 1, | |
582 | .guard_size = 2, | |
583 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
584 | }; | |
feb56b93 | 585 | static const struct intel_watermark_params i845_wm_info = { |
e0f0273e VS |
586 | .fifo_size = I830_FIFO_SIZE, |
587 | .max_wm = I915_MAX_WM, | |
588 | .default_wm = 1, | |
589 | .guard_size = 2, | |
590 | .cacheline_size = I830_FIFO_LINE_SIZE, | |
b445e3b0 ED |
591 | }; |
592 | ||
b445e3b0 ED |
593 | /** |
594 | * intel_calculate_wm - calculate watermark level | |
595 | * @clock_in_khz: pixel clock | |
596 | * @wm: chip FIFO params | |
597 | * @pixel_size: display pixel size | |
598 | * @latency_ns: memory latency for the platform | |
599 | * | |
600 | * Calculate the watermark level (the level at which the display plane will | |
601 | * start fetching from memory again). Each chip has a different display | |
602 | * FIFO size and allocation, so the caller needs to figure that out and pass | |
603 | * in the correct intel_watermark_params structure. | |
604 | * | |
605 | * As the pixel clock runs, the FIFO will be drained at a rate that depends | |
606 | * on the pixel size. When it reaches the watermark level, it'll start | |
607 | * fetching FIFO line sized based chunks from memory until the FIFO fills | |
608 | * past the watermark point. If the FIFO drains completely, a FIFO underrun | |
609 | * will occur, and a display engine hang could result. | |
610 | */ | |
611 | static unsigned long intel_calculate_wm(unsigned long clock_in_khz, | |
612 | const struct intel_watermark_params *wm, | |
613 | int fifo_size, | |
614 | int pixel_size, | |
615 | unsigned long latency_ns) | |
616 | { | |
617 | long entries_required, wm_size; | |
618 | ||
619 | /* | |
620 | * Note: we need to make sure we don't overflow for various clock & | |
621 | * latency values. | |
622 | * clocks go from a few thousand to several hundred thousand. | |
623 | * latency is usually a few thousand | |
624 | */ | |
625 | entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) / | |
626 | 1000; | |
627 | entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size); | |
628 | ||
629 | DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required); | |
630 | ||
631 | wm_size = fifo_size - (entries_required + wm->guard_size); | |
632 | ||
633 | DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size); | |
634 | ||
635 | /* Don't promote wm_size to unsigned... */ | |
636 | if (wm_size > (long)wm->max_wm) | |
637 | wm_size = wm->max_wm; | |
638 | if (wm_size <= 0) | |
639 | wm_size = wm->default_wm; | |
d6feb196 VS |
640 | |
641 | /* | |
642 | * Bspec seems to indicate that the value shouldn't be lower than | |
643 | * 'burst size + 1'. Certainly 830 is quite unhappy with low values. | |
644 | * Lets go for 8 which is the burst size since certain platforms | |
645 | * already use a hardcoded 8 (which is what the spec says should be | |
646 | * done). | |
647 | */ | |
648 | if (wm_size <= 8) | |
649 | wm_size = 8; | |
650 | ||
b445e3b0 ED |
651 | return wm_size; |
652 | } | |
653 | ||
654 | static struct drm_crtc *single_enabled_crtc(struct drm_device *dev) | |
655 | { | |
656 | struct drm_crtc *crtc, *enabled = NULL; | |
657 | ||
70e1e0ec | 658 | for_each_crtc(dev, crtc) { |
3490ea5d | 659 | if (intel_crtc_active(crtc)) { |
b445e3b0 ED |
660 | if (enabled) |
661 | return NULL; | |
662 | enabled = crtc; | |
663 | } | |
664 | } | |
665 | ||
666 | return enabled; | |
667 | } | |
668 | ||
46ba614c | 669 | static void pineview_update_wm(struct drm_crtc *unused_crtc) |
b445e3b0 | 670 | { |
46ba614c | 671 | struct drm_device *dev = unused_crtc->dev; |
b445e3b0 ED |
672 | struct drm_i915_private *dev_priv = dev->dev_private; |
673 | struct drm_crtc *crtc; | |
674 | const struct cxsr_latency *latency; | |
675 | u32 reg; | |
676 | unsigned long wm; | |
677 | ||
678 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3, | |
679 | dev_priv->fsb_freq, dev_priv->mem_freq); | |
680 | if (!latency) { | |
681 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
5209b1f4 | 682 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
683 | return; |
684 | } | |
685 | ||
686 | crtc = single_enabled_crtc(dev); | |
687 | if (crtc) { | |
241bfc38 | 688 | const struct drm_display_mode *adjusted_mode; |
59bea882 | 689 | int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
241bfc38 DL |
690 | int clock; |
691 | ||
6e3c9717 | 692 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 693 | clock = adjusted_mode->crtc_clock; |
b445e3b0 ED |
694 | |
695 | /* Display SR */ | |
696 | wm = intel_calculate_wm(clock, &pineview_display_wm, | |
697 | pineview_display_wm.fifo_size, | |
698 | pixel_size, latency->display_sr); | |
699 | reg = I915_READ(DSPFW1); | |
700 | reg &= ~DSPFW_SR_MASK; | |
f4998963 | 701 | reg |= FW_WM(wm, SR); |
b445e3b0 ED |
702 | I915_WRITE(DSPFW1, reg); |
703 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); | |
704 | ||
705 | /* cursor SR */ | |
706 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, | |
707 | pineview_display_wm.fifo_size, | |
708 | pixel_size, latency->cursor_sr); | |
709 | reg = I915_READ(DSPFW3); | |
710 | reg &= ~DSPFW_CURSOR_SR_MASK; | |
f4998963 | 711 | reg |= FW_WM(wm, CURSOR_SR); |
b445e3b0 ED |
712 | I915_WRITE(DSPFW3, reg); |
713 | ||
714 | /* Display HPLL off SR */ | |
715 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, | |
716 | pineview_display_hplloff_wm.fifo_size, | |
717 | pixel_size, latency->display_hpll_disable); | |
718 | reg = I915_READ(DSPFW3); | |
719 | reg &= ~DSPFW_HPLL_SR_MASK; | |
f4998963 | 720 | reg |= FW_WM(wm, HPLL_SR); |
b445e3b0 ED |
721 | I915_WRITE(DSPFW3, reg); |
722 | ||
723 | /* cursor HPLL off SR */ | |
724 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, | |
725 | pineview_display_hplloff_wm.fifo_size, | |
726 | pixel_size, latency->cursor_hpll_disable); | |
727 | reg = I915_READ(DSPFW3); | |
728 | reg &= ~DSPFW_HPLL_CURSOR_MASK; | |
f4998963 | 729 | reg |= FW_WM(wm, HPLL_CURSOR); |
b445e3b0 ED |
730 | I915_WRITE(DSPFW3, reg); |
731 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); | |
732 | ||
5209b1f4 | 733 | intel_set_memory_cxsr(dev_priv, true); |
b445e3b0 | 734 | } else { |
5209b1f4 | 735 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
736 | } |
737 | } | |
738 | ||
739 | static bool g4x_compute_wm0(struct drm_device *dev, | |
740 | int plane, | |
741 | const struct intel_watermark_params *display, | |
742 | int display_latency_ns, | |
743 | const struct intel_watermark_params *cursor, | |
744 | int cursor_latency_ns, | |
745 | int *plane_wm, | |
746 | int *cursor_wm) | |
747 | { | |
748 | struct drm_crtc *crtc; | |
4fe8590a | 749 | const struct drm_display_mode *adjusted_mode; |
b445e3b0 ED |
750 | int htotal, hdisplay, clock, pixel_size; |
751 | int line_time_us, line_count; | |
752 | int entries, tlb_miss; | |
753 | ||
754 | crtc = intel_get_crtc_for_plane(dev, plane); | |
3490ea5d | 755 | if (!intel_crtc_active(crtc)) { |
b445e3b0 ED |
756 | *cursor_wm = cursor->guard_size; |
757 | *plane_wm = display->guard_size; | |
758 | return false; | |
759 | } | |
760 | ||
6e3c9717 | 761 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 762 | clock = adjusted_mode->crtc_clock; |
fec8cba3 | 763 | htotal = adjusted_mode->crtc_htotal; |
6e3c9717 | 764 | hdisplay = to_intel_crtc(crtc)->config->pipe_src_w; |
59bea882 | 765 | pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
b445e3b0 ED |
766 | |
767 | /* Use the small buffer method to calculate plane watermark */ | |
768 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; | |
769 | tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8; | |
770 | if (tlb_miss > 0) | |
771 | entries += tlb_miss; | |
772 | entries = DIV_ROUND_UP(entries, display->cacheline_size); | |
773 | *plane_wm = entries + display->guard_size; | |
774 | if (*plane_wm > (int)display->max_wm) | |
775 | *plane_wm = display->max_wm; | |
776 | ||
777 | /* Use the large buffer method to calculate cursor watermark */ | |
922044c9 | 778 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 | 779 | line_count = (cursor_latency_ns / line_time_us + 1000) / 1000; |
3dd512fb | 780 | entries = line_count * crtc->cursor->state->crtc_w * pixel_size; |
b445e3b0 ED |
781 | tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8; |
782 | if (tlb_miss > 0) | |
783 | entries += tlb_miss; | |
784 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); | |
785 | *cursor_wm = entries + cursor->guard_size; | |
786 | if (*cursor_wm > (int)cursor->max_wm) | |
787 | *cursor_wm = (int)cursor->max_wm; | |
788 | ||
789 | return true; | |
790 | } | |
791 | ||
792 | /* | |
793 | * Check the wm result. | |
794 | * | |
795 | * If any calculated watermark values is larger than the maximum value that | |
796 | * can be programmed into the associated watermark register, that watermark | |
797 | * must be disabled. | |
798 | */ | |
799 | static bool g4x_check_srwm(struct drm_device *dev, | |
800 | int display_wm, int cursor_wm, | |
801 | const struct intel_watermark_params *display, | |
802 | const struct intel_watermark_params *cursor) | |
803 | { | |
804 | DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n", | |
805 | display_wm, cursor_wm); | |
806 | ||
807 | if (display_wm > display->max_wm) { | |
808 | DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n", | |
809 | display_wm, display->max_wm); | |
810 | return false; | |
811 | } | |
812 | ||
813 | if (cursor_wm > cursor->max_wm) { | |
814 | DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n", | |
815 | cursor_wm, cursor->max_wm); | |
816 | return false; | |
817 | } | |
818 | ||
819 | if (!(display_wm || cursor_wm)) { | |
820 | DRM_DEBUG_KMS("SR latency is 0, disabling\n"); | |
821 | return false; | |
822 | } | |
823 | ||
824 | return true; | |
825 | } | |
826 | ||
827 | static bool g4x_compute_srwm(struct drm_device *dev, | |
828 | int plane, | |
829 | int latency_ns, | |
830 | const struct intel_watermark_params *display, | |
831 | const struct intel_watermark_params *cursor, | |
832 | int *display_wm, int *cursor_wm) | |
833 | { | |
834 | struct drm_crtc *crtc; | |
4fe8590a | 835 | const struct drm_display_mode *adjusted_mode; |
b445e3b0 ED |
836 | int hdisplay, htotal, pixel_size, clock; |
837 | unsigned long line_time_us; | |
838 | int line_count, line_size; | |
839 | int small, large; | |
840 | int entries; | |
841 | ||
842 | if (!latency_ns) { | |
843 | *display_wm = *cursor_wm = 0; | |
844 | return false; | |
845 | } | |
846 | ||
847 | crtc = intel_get_crtc_for_plane(dev, plane); | |
6e3c9717 | 848 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 849 | clock = adjusted_mode->crtc_clock; |
fec8cba3 | 850 | htotal = adjusted_mode->crtc_htotal; |
6e3c9717 | 851 | hdisplay = to_intel_crtc(crtc)->config->pipe_src_w; |
59bea882 | 852 | pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
b445e3b0 | 853 | |
922044c9 | 854 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 ED |
855 | line_count = (latency_ns / line_time_us + 1000) / 1000; |
856 | line_size = hdisplay * pixel_size; | |
857 | ||
858 | /* Use the minimum of the small and large buffer method for primary */ | |
859 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; | |
860 | large = line_count * line_size; | |
861 | ||
862 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); | |
863 | *display_wm = entries + display->guard_size; | |
864 | ||
865 | /* calculate the self-refresh watermark for display cursor */ | |
3dd512fb | 866 | entries = line_count * pixel_size * crtc->cursor->state->crtc_w; |
b445e3b0 ED |
867 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); |
868 | *cursor_wm = entries + cursor->guard_size; | |
869 | ||
870 | return g4x_check_srwm(dev, | |
871 | *display_wm, *cursor_wm, | |
872 | display, cursor); | |
873 | } | |
874 | ||
15665979 VS |
875 | #define FW_WM_VLV(value, plane) \ |
876 | (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV) | |
877 | ||
0018fda1 VS |
878 | static void vlv_write_wm_values(struct intel_crtc *crtc, |
879 | const struct vlv_wm_values *wm) | |
880 | { | |
881 | struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); | |
882 | enum pipe pipe = crtc->pipe; | |
883 | ||
884 | I915_WRITE(VLV_DDL(pipe), | |
885 | (wm->ddl[pipe].cursor << DDL_CURSOR_SHIFT) | | |
886 | (wm->ddl[pipe].sprite[1] << DDL_SPRITE_SHIFT(1)) | | |
887 | (wm->ddl[pipe].sprite[0] << DDL_SPRITE_SHIFT(0)) | | |
888 | (wm->ddl[pipe].primary << DDL_PLANE_SHIFT)); | |
889 | ||
ae80152d | 890 | I915_WRITE(DSPFW1, |
15665979 VS |
891 | FW_WM(wm->sr.plane, SR) | |
892 | FW_WM(wm->pipe[PIPE_B].cursor, CURSORB) | | |
893 | FW_WM_VLV(wm->pipe[PIPE_B].primary, PLANEB) | | |
894 | FW_WM_VLV(wm->pipe[PIPE_A].primary, PLANEA)); | |
ae80152d | 895 | I915_WRITE(DSPFW2, |
15665979 VS |
896 | FW_WM_VLV(wm->pipe[PIPE_A].sprite[1], SPRITEB) | |
897 | FW_WM(wm->pipe[PIPE_A].cursor, CURSORA) | | |
898 | FW_WM_VLV(wm->pipe[PIPE_A].sprite[0], SPRITEA)); | |
ae80152d | 899 | I915_WRITE(DSPFW3, |
15665979 | 900 | FW_WM(wm->sr.cursor, CURSOR_SR)); |
ae80152d VS |
901 | |
902 | if (IS_CHERRYVIEW(dev_priv)) { | |
903 | I915_WRITE(DSPFW7_CHV, | |
15665979 VS |
904 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) | |
905 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC)); | |
ae80152d | 906 | I915_WRITE(DSPFW8_CHV, |
15665979 VS |
907 | FW_WM_VLV(wm->pipe[PIPE_C].sprite[1], SPRITEF) | |
908 | FW_WM_VLV(wm->pipe[PIPE_C].sprite[0], SPRITEE)); | |
ae80152d | 909 | I915_WRITE(DSPFW9_CHV, |
15665979 VS |
910 | FW_WM_VLV(wm->pipe[PIPE_C].primary, PLANEC) | |
911 | FW_WM(wm->pipe[PIPE_C].cursor, CURSORC)); | |
ae80152d | 912 | I915_WRITE(DSPHOWM, |
15665979 VS |
913 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
914 | FW_WM(wm->pipe[PIPE_C].sprite[1] >> 8, SPRITEF_HI) | | |
915 | FW_WM(wm->pipe[PIPE_C].sprite[0] >> 8, SPRITEE_HI) | | |
916 | FW_WM(wm->pipe[PIPE_C].primary >> 8, PLANEC_HI) | | |
917 | FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) | | |
918 | FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) | | |
919 | FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) | | |
920 | FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) | | |
921 | FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) | | |
922 | FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI)); | |
ae80152d VS |
923 | } else { |
924 | I915_WRITE(DSPFW7, | |
15665979 VS |
925 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) | |
926 | FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC)); | |
ae80152d | 927 | I915_WRITE(DSPHOWM, |
15665979 VS |
928 | FW_WM(wm->sr.plane >> 9, SR_HI) | |
929 | FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) | | |
930 | FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) | | |
931 | FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) | | |
932 | FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) | | |
933 | FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) | | |
934 | FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI)); | |
ae80152d VS |
935 | } |
936 | ||
2cb389b7 VS |
937 | /* zero (unused) WM1 watermarks */ |
938 | I915_WRITE(DSPFW4, 0); | |
939 | I915_WRITE(DSPFW5, 0); | |
940 | I915_WRITE(DSPFW6, 0); | |
941 | I915_WRITE(DSPHOWM1, 0); | |
942 | ||
ae80152d | 943 | POSTING_READ(DSPFW1); |
0018fda1 VS |
944 | } |
945 | ||
15665979 VS |
946 | #undef FW_WM_VLV |
947 | ||
6eb1a681 VS |
948 | enum vlv_wm_level { |
949 | VLV_WM_LEVEL_PM2, | |
950 | VLV_WM_LEVEL_PM5, | |
951 | VLV_WM_LEVEL_DDR_DVFS, | |
952 | CHV_WM_NUM_LEVELS, | |
953 | VLV_WM_NUM_LEVELS = 1, | |
954 | }; | |
955 | ||
262cd2e1 VS |
956 | /* latency must be in 0.1us units. */ |
957 | static unsigned int vlv_wm_method2(unsigned int pixel_rate, | |
958 | unsigned int pipe_htotal, | |
959 | unsigned int horiz_pixels, | |
960 | unsigned int bytes_per_pixel, | |
961 | unsigned int latency) | |
962 | { | |
963 | unsigned int ret; | |
964 | ||
965 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); | |
966 | ret = (ret + 1) * horiz_pixels * bytes_per_pixel; | |
967 | ret = DIV_ROUND_UP(ret, 64); | |
968 | ||
969 | return ret; | |
970 | } | |
971 | ||
972 | static void vlv_setup_wm_latency(struct drm_device *dev) | |
973 | { | |
974 | struct drm_i915_private *dev_priv = dev->dev_private; | |
975 | ||
976 | /* all latencies in usec */ | |
977 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3; | |
978 | ||
979 | if (IS_CHERRYVIEW(dev_priv)) { | |
980 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12; | |
981 | dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33; | |
982 | } | |
983 | } | |
984 | ||
985 | static uint16_t vlv_compute_wm_level(struct intel_plane *plane, | |
986 | struct intel_crtc *crtc, | |
987 | const struct intel_plane_state *state, | |
988 | int level) | |
989 | { | |
990 | struct drm_i915_private *dev_priv = to_i915(plane->base.dev); | |
991 | int clock, htotal, pixel_size, width, wm; | |
992 | ||
993 | if (dev_priv->wm.pri_latency[level] == 0) | |
994 | return USHRT_MAX; | |
995 | ||
996 | if (!state->visible) | |
997 | return 0; | |
998 | ||
999 | pixel_size = drm_format_plane_cpp(state->base.fb->pixel_format, 0); | |
1000 | clock = crtc->config->base.adjusted_mode.crtc_clock; | |
1001 | htotal = crtc->config->base.adjusted_mode.crtc_htotal; | |
1002 | width = crtc->config->pipe_src_w; | |
1003 | if (WARN_ON(htotal == 0)) | |
1004 | htotal = 1; | |
1005 | ||
1006 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { | |
1007 | /* | |
1008 | * FIXME the formula gives values that are | |
1009 | * too big for the cursor FIFO, and hence we | |
1010 | * would never be able to use cursors. For | |
1011 | * now just hardcode the watermark. | |
1012 | */ | |
1013 | wm = 63; | |
1014 | } else { | |
1015 | wm = vlv_wm_method2(clock, htotal, width, pixel_size, | |
1016 | dev_priv->wm.pri_latency[level] * 10); | |
1017 | } | |
1018 | ||
1019 | return min_t(int, wm, USHRT_MAX); | |
1020 | } | |
1021 | ||
54f1b6e1 VS |
1022 | static void vlv_compute_fifo(struct intel_crtc *crtc) |
1023 | { | |
1024 | struct drm_device *dev = crtc->base.dev; | |
1025 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1026 | struct intel_plane *plane; | |
1027 | unsigned int total_rate = 0; | |
1028 | const int fifo_size = 512 - 1; | |
1029 | int fifo_extra, fifo_left = fifo_size; | |
1030 | ||
1031 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1032 | struct intel_plane_state *state = | |
1033 | to_intel_plane_state(plane->base.state); | |
1034 | ||
1035 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) | |
1036 | continue; | |
1037 | ||
1038 | if (state->visible) { | |
1039 | wm_state->num_active_planes++; | |
1040 | total_rate += drm_format_plane_cpp(state->base.fb->pixel_format, 0); | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1045 | struct intel_plane_state *state = | |
1046 | to_intel_plane_state(plane->base.state); | |
1047 | unsigned int rate; | |
1048 | ||
1049 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { | |
1050 | plane->wm.fifo_size = 63; | |
1051 | continue; | |
1052 | } | |
1053 | ||
1054 | if (!state->visible) { | |
1055 | plane->wm.fifo_size = 0; | |
1056 | continue; | |
1057 | } | |
1058 | ||
1059 | rate = drm_format_plane_cpp(state->base.fb->pixel_format, 0); | |
1060 | plane->wm.fifo_size = fifo_size * rate / total_rate; | |
1061 | fifo_left -= plane->wm.fifo_size; | |
1062 | } | |
1063 | ||
1064 | fifo_extra = DIV_ROUND_UP(fifo_left, wm_state->num_active_planes ?: 1); | |
1065 | ||
1066 | /* spread the remainder evenly */ | |
1067 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1068 | int plane_extra; | |
1069 | ||
1070 | if (fifo_left == 0) | |
1071 | break; | |
1072 | ||
1073 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) | |
1074 | continue; | |
1075 | ||
1076 | /* give it all to the first plane if none are active */ | |
1077 | if (plane->wm.fifo_size == 0 && | |
1078 | wm_state->num_active_planes) | |
1079 | continue; | |
1080 | ||
1081 | plane_extra = min(fifo_extra, fifo_left); | |
1082 | plane->wm.fifo_size += plane_extra; | |
1083 | fifo_left -= plane_extra; | |
1084 | } | |
1085 | ||
1086 | WARN_ON(fifo_left != 0); | |
1087 | } | |
1088 | ||
262cd2e1 VS |
1089 | static void vlv_invert_wms(struct intel_crtc *crtc) |
1090 | { | |
1091 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1092 | int level; | |
1093 | ||
1094 | for (level = 0; level < wm_state->num_levels; level++) { | |
1095 | struct drm_device *dev = crtc->base.dev; | |
1096 | const int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1; | |
1097 | struct intel_plane *plane; | |
1098 | ||
1099 | wm_state->sr[level].plane = sr_fifo_size - wm_state->sr[level].plane; | |
1100 | wm_state->sr[level].cursor = 63 - wm_state->sr[level].cursor; | |
1101 | ||
1102 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1103 | switch (plane->base.type) { | |
1104 | int sprite; | |
1105 | case DRM_PLANE_TYPE_CURSOR: | |
1106 | wm_state->wm[level].cursor = plane->wm.fifo_size - | |
1107 | wm_state->wm[level].cursor; | |
1108 | break; | |
1109 | case DRM_PLANE_TYPE_PRIMARY: | |
1110 | wm_state->wm[level].primary = plane->wm.fifo_size - | |
1111 | wm_state->wm[level].primary; | |
1112 | break; | |
1113 | case DRM_PLANE_TYPE_OVERLAY: | |
1114 | sprite = plane->plane; | |
1115 | wm_state->wm[level].sprite[sprite] = plane->wm.fifo_size - | |
1116 | wm_state->wm[level].sprite[sprite]; | |
1117 | break; | |
1118 | } | |
1119 | } | |
1120 | } | |
1121 | } | |
1122 | ||
26e1fe4f | 1123 | static void vlv_compute_wm(struct intel_crtc *crtc) |
262cd2e1 VS |
1124 | { |
1125 | struct drm_device *dev = crtc->base.dev; | |
1126 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1127 | struct intel_plane *plane; | |
1128 | int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1; | |
1129 | int level; | |
1130 | ||
1131 | memset(wm_state, 0, sizeof(*wm_state)); | |
1132 | ||
852eb00d | 1133 | wm_state->cxsr = crtc->pipe != PIPE_C && crtc->wm.cxsr_allowed; |
262cd2e1 VS |
1134 | if (IS_CHERRYVIEW(dev)) |
1135 | wm_state->num_levels = CHV_WM_NUM_LEVELS; | |
1136 | else | |
1137 | wm_state->num_levels = VLV_WM_NUM_LEVELS; | |
1138 | ||
1139 | wm_state->num_active_planes = 0; | |
262cd2e1 | 1140 | |
54f1b6e1 | 1141 | vlv_compute_fifo(crtc); |
262cd2e1 VS |
1142 | |
1143 | if (wm_state->num_active_planes != 1) | |
1144 | wm_state->cxsr = false; | |
1145 | ||
1146 | if (wm_state->cxsr) { | |
1147 | for (level = 0; level < wm_state->num_levels; level++) { | |
1148 | wm_state->sr[level].plane = sr_fifo_size; | |
1149 | wm_state->sr[level].cursor = 63; | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1154 | struct intel_plane_state *state = | |
1155 | to_intel_plane_state(plane->base.state); | |
1156 | ||
1157 | if (!state->visible) | |
1158 | continue; | |
1159 | ||
1160 | /* normal watermarks */ | |
1161 | for (level = 0; level < wm_state->num_levels; level++) { | |
1162 | int wm = vlv_compute_wm_level(plane, crtc, state, level); | |
1163 | int max_wm = plane->base.type == DRM_PLANE_TYPE_CURSOR ? 63 : 511; | |
1164 | ||
1165 | /* hack */ | |
1166 | if (WARN_ON(level == 0 && wm > max_wm)) | |
1167 | wm = max_wm; | |
1168 | ||
1169 | if (wm > plane->wm.fifo_size) | |
1170 | break; | |
1171 | ||
1172 | switch (plane->base.type) { | |
1173 | int sprite; | |
1174 | case DRM_PLANE_TYPE_CURSOR: | |
1175 | wm_state->wm[level].cursor = wm; | |
1176 | break; | |
1177 | case DRM_PLANE_TYPE_PRIMARY: | |
1178 | wm_state->wm[level].primary = wm; | |
1179 | break; | |
1180 | case DRM_PLANE_TYPE_OVERLAY: | |
1181 | sprite = plane->plane; | |
1182 | wm_state->wm[level].sprite[sprite] = wm; | |
1183 | break; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | wm_state->num_levels = level; | |
1188 | ||
1189 | if (!wm_state->cxsr) | |
1190 | continue; | |
1191 | ||
1192 | /* maxfifo watermarks */ | |
1193 | switch (plane->base.type) { | |
1194 | int sprite, level; | |
1195 | case DRM_PLANE_TYPE_CURSOR: | |
1196 | for (level = 0; level < wm_state->num_levels; level++) | |
1197 | wm_state->sr[level].cursor = | |
1198 | wm_state->sr[level].cursor; | |
1199 | break; | |
1200 | case DRM_PLANE_TYPE_PRIMARY: | |
1201 | for (level = 0; level < wm_state->num_levels; level++) | |
1202 | wm_state->sr[level].plane = | |
1203 | min(wm_state->sr[level].plane, | |
1204 | wm_state->wm[level].primary); | |
1205 | break; | |
1206 | case DRM_PLANE_TYPE_OVERLAY: | |
1207 | sprite = plane->plane; | |
1208 | for (level = 0; level < wm_state->num_levels; level++) | |
1209 | wm_state->sr[level].plane = | |
1210 | min(wm_state->sr[level].plane, | |
1211 | wm_state->wm[level].sprite[sprite]); | |
1212 | break; | |
1213 | } | |
1214 | } | |
1215 | ||
1216 | /* clear any (partially) filled invalid levels */ | |
1217 | for (level = wm_state->num_levels; level < CHV_WM_NUM_LEVELS; level++) { | |
1218 | memset(&wm_state->wm[level], 0, sizeof(wm_state->wm[level])); | |
1219 | memset(&wm_state->sr[level], 0, sizeof(wm_state->sr[level])); | |
1220 | } | |
1221 | ||
1222 | vlv_invert_wms(crtc); | |
1223 | } | |
1224 | ||
54f1b6e1 VS |
1225 | #define VLV_FIFO(plane, value) \ |
1226 | (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV) | |
1227 | ||
1228 | static void vlv_pipe_set_fifo_size(struct intel_crtc *crtc) | |
1229 | { | |
1230 | struct drm_device *dev = crtc->base.dev; | |
1231 | struct drm_i915_private *dev_priv = to_i915(dev); | |
1232 | struct intel_plane *plane; | |
1233 | int sprite0_start = 0, sprite1_start = 0, fifo_size = 0; | |
1234 | ||
1235 | for_each_intel_plane_on_crtc(dev, crtc, plane) { | |
1236 | if (plane->base.type == DRM_PLANE_TYPE_CURSOR) { | |
1237 | WARN_ON(plane->wm.fifo_size != 63); | |
1238 | continue; | |
1239 | } | |
1240 | ||
1241 | if (plane->base.type == DRM_PLANE_TYPE_PRIMARY) | |
1242 | sprite0_start = plane->wm.fifo_size; | |
1243 | else if (plane->plane == 0) | |
1244 | sprite1_start = sprite0_start + plane->wm.fifo_size; | |
1245 | else | |
1246 | fifo_size = sprite1_start + plane->wm.fifo_size; | |
1247 | } | |
1248 | ||
1249 | WARN_ON(fifo_size != 512 - 1); | |
1250 | ||
1251 | DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n", | |
1252 | pipe_name(crtc->pipe), sprite0_start, | |
1253 | sprite1_start, fifo_size); | |
1254 | ||
1255 | switch (crtc->pipe) { | |
1256 | uint32_t dsparb, dsparb2, dsparb3; | |
1257 | case PIPE_A: | |
1258 | dsparb = I915_READ(DSPARB); | |
1259 | dsparb2 = I915_READ(DSPARB2); | |
1260 | ||
1261 | dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) | | |
1262 | VLV_FIFO(SPRITEB, 0xff)); | |
1263 | dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) | | |
1264 | VLV_FIFO(SPRITEB, sprite1_start)); | |
1265 | ||
1266 | dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) | | |
1267 | VLV_FIFO(SPRITEB_HI, 0x1)); | |
1268 | dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) | | |
1269 | VLV_FIFO(SPRITEB_HI, sprite1_start >> 8)); | |
1270 | ||
1271 | I915_WRITE(DSPARB, dsparb); | |
1272 | I915_WRITE(DSPARB2, dsparb2); | |
1273 | break; | |
1274 | case PIPE_B: | |
1275 | dsparb = I915_READ(DSPARB); | |
1276 | dsparb2 = I915_READ(DSPARB2); | |
1277 | ||
1278 | dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) | | |
1279 | VLV_FIFO(SPRITED, 0xff)); | |
1280 | dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) | | |
1281 | VLV_FIFO(SPRITED, sprite1_start)); | |
1282 | ||
1283 | dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) | | |
1284 | VLV_FIFO(SPRITED_HI, 0xff)); | |
1285 | dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) | | |
1286 | VLV_FIFO(SPRITED_HI, sprite1_start >> 8)); | |
1287 | ||
1288 | I915_WRITE(DSPARB, dsparb); | |
1289 | I915_WRITE(DSPARB2, dsparb2); | |
1290 | break; | |
1291 | case PIPE_C: | |
1292 | dsparb3 = I915_READ(DSPARB3); | |
1293 | dsparb2 = I915_READ(DSPARB2); | |
1294 | ||
1295 | dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) | | |
1296 | VLV_FIFO(SPRITEF, 0xff)); | |
1297 | dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) | | |
1298 | VLV_FIFO(SPRITEF, sprite1_start)); | |
1299 | ||
1300 | dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) | | |
1301 | VLV_FIFO(SPRITEF_HI, 0xff)); | |
1302 | dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) | | |
1303 | VLV_FIFO(SPRITEF_HI, sprite1_start >> 8)); | |
1304 | ||
1305 | I915_WRITE(DSPARB3, dsparb3); | |
1306 | I915_WRITE(DSPARB2, dsparb2); | |
1307 | break; | |
1308 | default: | |
1309 | break; | |
1310 | } | |
1311 | } | |
1312 | ||
1313 | #undef VLV_FIFO | |
1314 | ||
262cd2e1 VS |
1315 | static void vlv_merge_wm(struct drm_device *dev, |
1316 | struct vlv_wm_values *wm) | |
1317 | { | |
1318 | struct intel_crtc *crtc; | |
1319 | int num_active_crtcs = 0; | |
1320 | ||
1321 | if (IS_CHERRYVIEW(dev)) | |
1322 | wm->level = VLV_WM_LEVEL_DDR_DVFS; | |
1323 | else | |
1324 | wm->level = VLV_WM_LEVEL_PM2; | |
1325 | wm->cxsr = true; | |
1326 | ||
1327 | for_each_intel_crtc(dev, crtc) { | |
1328 | const struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1329 | ||
1330 | if (!crtc->active) | |
1331 | continue; | |
1332 | ||
1333 | if (!wm_state->cxsr) | |
1334 | wm->cxsr = false; | |
1335 | ||
1336 | num_active_crtcs++; | |
1337 | wm->level = min_t(int, wm->level, wm_state->num_levels - 1); | |
1338 | } | |
1339 | ||
1340 | if (num_active_crtcs != 1) | |
1341 | wm->cxsr = false; | |
1342 | ||
6f9c784b VS |
1343 | if (num_active_crtcs > 1) |
1344 | wm->level = VLV_WM_LEVEL_PM2; | |
1345 | ||
262cd2e1 VS |
1346 | for_each_intel_crtc(dev, crtc) { |
1347 | struct vlv_wm_state *wm_state = &crtc->wm_state; | |
1348 | enum pipe pipe = crtc->pipe; | |
1349 | ||
1350 | if (!crtc->active) | |
1351 | continue; | |
1352 | ||
1353 | wm->pipe[pipe] = wm_state->wm[wm->level]; | |
1354 | if (wm->cxsr) | |
1355 | wm->sr = wm_state->sr[wm->level]; | |
1356 | ||
1357 | wm->ddl[pipe].primary = DDL_PRECISION_HIGH | 2; | |
1358 | wm->ddl[pipe].sprite[0] = DDL_PRECISION_HIGH | 2; | |
1359 | wm->ddl[pipe].sprite[1] = DDL_PRECISION_HIGH | 2; | |
1360 | wm->ddl[pipe].cursor = DDL_PRECISION_HIGH | 2; | |
1361 | } | |
1362 | } | |
1363 | ||
1364 | static void vlv_update_wm(struct drm_crtc *crtc) | |
1365 | { | |
1366 | struct drm_device *dev = crtc->dev; | |
1367 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1368 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
1369 | enum pipe pipe = intel_crtc->pipe; | |
1370 | struct vlv_wm_values wm = {}; | |
1371 | ||
26e1fe4f | 1372 | vlv_compute_wm(intel_crtc); |
262cd2e1 VS |
1373 | vlv_merge_wm(dev, &wm); |
1374 | ||
54f1b6e1 VS |
1375 | if (memcmp(&dev_priv->wm.vlv, &wm, sizeof(wm)) == 0) { |
1376 | /* FIXME should be part of crtc atomic commit */ | |
1377 | vlv_pipe_set_fifo_size(intel_crtc); | |
262cd2e1 | 1378 | return; |
54f1b6e1 | 1379 | } |
262cd2e1 VS |
1380 | |
1381 | if (wm.level < VLV_WM_LEVEL_DDR_DVFS && | |
1382 | dev_priv->wm.vlv.level >= VLV_WM_LEVEL_DDR_DVFS) | |
1383 | chv_set_memory_dvfs(dev_priv, false); | |
1384 | ||
1385 | if (wm.level < VLV_WM_LEVEL_PM5 && | |
1386 | dev_priv->wm.vlv.level >= VLV_WM_LEVEL_PM5) | |
1387 | chv_set_memory_pm5(dev_priv, false); | |
1388 | ||
852eb00d | 1389 | if (!wm.cxsr && dev_priv->wm.vlv.cxsr) |
262cd2e1 | 1390 | intel_set_memory_cxsr(dev_priv, false); |
262cd2e1 | 1391 | |
54f1b6e1 VS |
1392 | /* FIXME should be part of crtc atomic commit */ |
1393 | vlv_pipe_set_fifo_size(intel_crtc); | |
1394 | ||
262cd2e1 VS |
1395 | vlv_write_wm_values(intel_crtc, &wm); |
1396 | ||
1397 | DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, " | |
1398 | "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n", | |
1399 | pipe_name(pipe), wm.pipe[pipe].primary, wm.pipe[pipe].cursor, | |
1400 | wm.pipe[pipe].sprite[0], wm.pipe[pipe].sprite[1], | |
1401 | wm.sr.plane, wm.sr.cursor, wm.level, wm.cxsr); | |
1402 | ||
852eb00d | 1403 | if (wm.cxsr && !dev_priv->wm.vlv.cxsr) |
262cd2e1 | 1404 | intel_set_memory_cxsr(dev_priv, true); |
262cd2e1 VS |
1405 | |
1406 | if (wm.level >= VLV_WM_LEVEL_PM5 && | |
1407 | dev_priv->wm.vlv.level < VLV_WM_LEVEL_PM5) | |
1408 | chv_set_memory_pm5(dev_priv, true); | |
1409 | ||
1410 | if (wm.level >= VLV_WM_LEVEL_DDR_DVFS && | |
1411 | dev_priv->wm.vlv.level < VLV_WM_LEVEL_DDR_DVFS) | |
1412 | chv_set_memory_dvfs(dev_priv, true); | |
1413 | ||
1414 | dev_priv->wm.vlv = wm; | |
3c2777fd VS |
1415 | } |
1416 | ||
ae80152d VS |
1417 | #define single_plane_enabled(mask) is_power_of_2(mask) |
1418 | ||
46ba614c | 1419 | static void g4x_update_wm(struct drm_crtc *crtc) |
b445e3b0 | 1420 | { |
46ba614c | 1421 | struct drm_device *dev = crtc->dev; |
b445e3b0 ED |
1422 | static const int sr_latency_ns = 12000; |
1423 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1424 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; | |
1425 | int plane_sr, cursor_sr; | |
1426 | unsigned int enabled = 0; | |
9858425c | 1427 | bool cxsr_enabled; |
b445e3b0 | 1428 | |
51cea1f4 | 1429 | if (g4x_compute_wm0(dev, PIPE_A, |
5aef6003 CW |
1430 | &g4x_wm_info, pessimal_latency_ns, |
1431 | &g4x_cursor_wm_info, pessimal_latency_ns, | |
b445e3b0 | 1432 | &planea_wm, &cursora_wm)) |
51cea1f4 | 1433 | enabled |= 1 << PIPE_A; |
b445e3b0 | 1434 | |
51cea1f4 | 1435 | if (g4x_compute_wm0(dev, PIPE_B, |
5aef6003 CW |
1436 | &g4x_wm_info, pessimal_latency_ns, |
1437 | &g4x_cursor_wm_info, pessimal_latency_ns, | |
b445e3b0 | 1438 | &planeb_wm, &cursorb_wm)) |
51cea1f4 | 1439 | enabled |= 1 << PIPE_B; |
b445e3b0 | 1440 | |
b445e3b0 ED |
1441 | if (single_plane_enabled(enabled) && |
1442 | g4x_compute_srwm(dev, ffs(enabled) - 1, | |
1443 | sr_latency_ns, | |
1444 | &g4x_wm_info, | |
1445 | &g4x_cursor_wm_info, | |
52bd02d8 | 1446 | &plane_sr, &cursor_sr)) { |
9858425c | 1447 | cxsr_enabled = true; |
52bd02d8 | 1448 | } else { |
9858425c | 1449 | cxsr_enabled = false; |
5209b1f4 | 1450 | intel_set_memory_cxsr(dev_priv, false); |
52bd02d8 CW |
1451 | plane_sr = cursor_sr = 0; |
1452 | } | |
b445e3b0 | 1453 | |
a5043453 VS |
1454 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, " |
1455 | "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", | |
b445e3b0 ED |
1456 | planea_wm, cursora_wm, |
1457 | planeb_wm, cursorb_wm, | |
1458 | plane_sr, cursor_sr); | |
1459 | ||
1460 | I915_WRITE(DSPFW1, | |
f4998963 VS |
1461 | FW_WM(plane_sr, SR) | |
1462 | FW_WM(cursorb_wm, CURSORB) | | |
1463 | FW_WM(planeb_wm, PLANEB) | | |
1464 | FW_WM(planea_wm, PLANEA)); | |
b445e3b0 | 1465 | I915_WRITE(DSPFW2, |
8c919b28 | 1466 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
f4998963 | 1467 | FW_WM(cursora_wm, CURSORA)); |
b445e3b0 ED |
1468 | /* HPLL off in SR has some issues on G4x... disable it */ |
1469 | I915_WRITE(DSPFW3, | |
8c919b28 | 1470 | (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) | |
f4998963 | 1471 | FW_WM(cursor_sr, CURSOR_SR)); |
9858425c ID |
1472 | |
1473 | if (cxsr_enabled) | |
1474 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
1475 | } |
1476 | ||
46ba614c | 1477 | static void i965_update_wm(struct drm_crtc *unused_crtc) |
b445e3b0 | 1478 | { |
46ba614c | 1479 | struct drm_device *dev = unused_crtc->dev; |
b445e3b0 ED |
1480 | struct drm_i915_private *dev_priv = dev->dev_private; |
1481 | struct drm_crtc *crtc; | |
1482 | int srwm = 1; | |
1483 | int cursor_sr = 16; | |
9858425c | 1484 | bool cxsr_enabled; |
b445e3b0 ED |
1485 | |
1486 | /* Calc sr entries for one plane configs */ | |
1487 | crtc = single_enabled_crtc(dev); | |
1488 | if (crtc) { | |
1489 | /* self-refresh has much higher latency */ | |
1490 | static const int sr_latency_ns = 12000; | |
4fe8590a | 1491 | const struct drm_display_mode *adjusted_mode = |
6e3c9717 | 1492 | &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 1493 | int clock = adjusted_mode->crtc_clock; |
fec8cba3 | 1494 | int htotal = adjusted_mode->crtc_htotal; |
6e3c9717 | 1495 | int hdisplay = to_intel_crtc(crtc)->config->pipe_src_w; |
59bea882 | 1496 | int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8; |
b445e3b0 ED |
1497 | unsigned long line_time_us; |
1498 | int entries; | |
1499 | ||
922044c9 | 1500 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 ED |
1501 | |
1502 | /* Use ns/us then divide to preserve precision */ | |
1503 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
1504 | pixel_size * hdisplay; | |
1505 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); | |
1506 | srwm = I965_FIFO_SIZE - entries; | |
1507 | if (srwm < 0) | |
1508 | srwm = 1; | |
1509 | srwm &= 0x1ff; | |
1510 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", | |
1511 | entries, srwm); | |
1512 | ||
1513 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
3dd512fb | 1514 | pixel_size * crtc->cursor->state->crtc_w; |
b445e3b0 ED |
1515 | entries = DIV_ROUND_UP(entries, |
1516 | i965_cursor_wm_info.cacheline_size); | |
1517 | cursor_sr = i965_cursor_wm_info.fifo_size - | |
1518 | (entries + i965_cursor_wm_info.guard_size); | |
1519 | ||
1520 | if (cursor_sr > i965_cursor_wm_info.max_wm) | |
1521 | cursor_sr = i965_cursor_wm_info.max_wm; | |
1522 | ||
1523 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " | |
1524 | "cursor %d\n", srwm, cursor_sr); | |
1525 | ||
9858425c | 1526 | cxsr_enabled = true; |
b445e3b0 | 1527 | } else { |
9858425c | 1528 | cxsr_enabled = false; |
b445e3b0 | 1529 | /* Turn off self refresh if both pipes are enabled */ |
5209b1f4 | 1530 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
1531 | } |
1532 | ||
1533 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", | |
1534 | srwm); | |
1535 | ||
1536 | /* 965 has limitations... */ | |
f4998963 VS |
1537 | I915_WRITE(DSPFW1, FW_WM(srwm, SR) | |
1538 | FW_WM(8, CURSORB) | | |
1539 | FW_WM(8, PLANEB) | | |
1540 | FW_WM(8, PLANEA)); | |
1541 | I915_WRITE(DSPFW2, FW_WM(8, CURSORA) | | |
1542 | FW_WM(8, PLANEC_OLD)); | |
b445e3b0 | 1543 | /* update cursor SR watermark */ |
f4998963 | 1544 | I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR)); |
9858425c ID |
1545 | |
1546 | if (cxsr_enabled) | |
1547 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
1548 | } |
1549 | ||
f4998963 VS |
1550 | #undef FW_WM |
1551 | ||
46ba614c | 1552 | static void i9xx_update_wm(struct drm_crtc *unused_crtc) |
b445e3b0 | 1553 | { |
46ba614c | 1554 | struct drm_device *dev = unused_crtc->dev; |
b445e3b0 ED |
1555 | struct drm_i915_private *dev_priv = dev->dev_private; |
1556 | const struct intel_watermark_params *wm_info; | |
1557 | uint32_t fwater_lo; | |
1558 | uint32_t fwater_hi; | |
1559 | int cwm, srwm = 1; | |
1560 | int fifo_size; | |
1561 | int planea_wm, planeb_wm; | |
1562 | struct drm_crtc *crtc, *enabled = NULL; | |
1563 | ||
1564 | if (IS_I945GM(dev)) | |
1565 | wm_info = &i945_wm_info; | |
1566 | else if (!IS_GEN2(dev)) | |
1567 | wm_info = &i915_wm_info; | |
1568 | else | |
9d539105 | 1569 | wm_info = &i830_a_wm_info; |
b445e3b0 ED |
1570 | |
1571 | fifo_size = dev_priv->display.get_fifo_size(dev, 0); | |
1572 | crtc = intel_get_crtc_for_plane(dev, 0); | |
3490ea5d | 1573 | if (intel_crtc_active(crtc)) { |
241bfc38 | 1574 | const struct drm_display_mode *adjusted_mode; |
59bea882 | 1575 | int cpp = crtc->primary->state->fb->bits_per_pixel / 8; |
b9e0bda3 CW |
1576 | if (IS_GEN2(dev)) |
1577 | cpp = 4; | |
1578 | ||
6e3c9717 | 1579 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 1580 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
b9e0bda3 | 1581 | wm_info, fifo_size, cpp, |
5aef6003 | 1582 | pessimal_latency_ns); |
b445e3b0 | 1583 | enabled = crtc; |
9d539105 | 1584 | } else { |
b445e3b0 | 1585 | planea_wm = fifo_size - wm_info->guard_size; |
9d539105 VS |
1586 | if (planea_wm > (long)wm_info->max_wm) |
1587 | planea_wm = wm_info->max_wm; | |
1588 | } | |
1589 | ||
1590 | if (IS_GEN2(dev)) | |
1591 | wm_info = &i830_bc_wm_info; | |
b445e3b0 ED |
1592 | |
1593 | fifo_size = dev_priv->display.get_fifo_size(dev, 1); | |
1594 | crtc = intel_get_crtc_for_plane(dev, 1); | |
3490ea5d | 1595 | if (intel_crtc_active(crtc)) { |
241bfc38 | 1596 | const struct drm_display_mode *adjusted_mode; |
59bea882 | 1597 | int cpp = crtc->primary->state->fb->bits_per_pixel / 8; |
b9e0bda3 CW |
1598 | if (IS_GEN2(dev)) |
1599 | cpp = 4; | |
1600 | ||
6e3c9717 | 1601 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 1602 | planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
b9e0bda3 | 1603 | wm_info, fifo_size, cpp, |
5aef6003 | 1604 | pessimal_latency_ns); |
b445e3b0 ED |
1605 | if (enabled == NULL) |
1606 | enabled = crtc; | |
1607 | else | |
1608 | enabled = NULL; | |
9d539105 | 1609 | } else { |
b445e3b0 | 1610 | planeb_wm = fifo_size - wm_info->guard_size; |
9d539105 VS |
1611 | if (planeb_wm > (long)wm_info->max_wm) |
1612 | planeb_wm = wm_info->max_wm; | |
1613 | } | |
b445e3b0 ED |
1614 | |
1615 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); | |
1616 | ||
2ab1bc9d | 1617 | if (IS_I915GM(dev) && enabled) { |
2ff8fde1 | 1618 | struct drm_i915_gem_object *obj; |
2ab1bc9d | 1619 | |
59bea882 | 1620 | obj = intel_fb_obj(enabled->primary->state->fb); |
2ab1bc9d DV |
1621 | |
1622 | /* self-refresh seems busted with untiled */ | |
2ff8fde1 | 1623 | if (obj->tiling_mode == I915_TILING_NONE) |
2ab1bc9d DV |
1624 | enabled = NULL; |
1625 | } | |
1626 | ||
b445e3b0 ED |
1627 | /* |
1628 | * Overlay gets an aggressive default since video jitter is bad. | |
1629 | */ | |
1630 | cwm = 2; | |
1631 | ||
1632 | /* Play safe and disable self-refresh before adjusting watermarks. */ | |
5209b1f4 | 1633 | intel_set_memory_cxsr(dev_priv, false); |
b445e3b0 ED |
1634 | |
1635 | /* Calc sr entries for one plane configs */ | |
1636 | if (HAS_FW_BLC(dev) && enabled) { | |
1637 | /* self-refresh has much higher latency */ | |
1638 | static const int sr_latency_ns = 6000; | |
4fe8590a | 1639 | const struct drm_display_mode *adjusted_mode = |
6e3c9717 | 1640 | &to_intel_crtc(enabled)->config->base.adjusted_mode; |
241bfc38 | 1641 | int clock = adjusted_mode->crtc_clock; |
fec8cba3 | 1642 | int htotal = adjusted_mode->crtc_htotal; |
6e3c9717 | 1643 | int hdisplay = to_intel_crtc(enabled)->config->pipe_src_w; |
59bea882 | 1644 | int pixel_size = enabled->primary->state->fb->bits_per_pixel / 8; |
b445e3b0 ED |
1645 | unsigned long line_time_us; |
1646 | int entries; | |
1647 | ||
922044c9 | 1648 | line_time_us = max(htotal * 1000 / clock, 1); |
b445e3b0 ED |
1649 | |
1650 | /* Use ns/us then divide to preserve precision */ | |
1651 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
1652 | pixel_size * hdisplay; | |
1653 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); | |
1654 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); | |
1655 | srwm = wm_info->fifo_size - entries; | |
1656 | if (srwm < 0) | |
1657 | srwm = 1; | |
1658 | ||
1659 | if (IS_I945G(dev) || IS_I945GM(dev)) | |
1660 | I915_WRITE(FW_BLC_SELF, | |
1661 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); | |
1662 | else if (IS_I915GM(dev)) | |
1663 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); | |
1664 | } | |
1665 | ||
1666 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", | |
1667 | planea_wm, planeb_wm, cwm, srwm); | |
1668 | ||
1669 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); | |
1670 | fwater_hi = (cwm & 0x1f); | |
1671 | ||
1672 | /* Set request length to 8 cachelines per fetch */ | |
1673 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); | |
1674 | fwater_hi = fwater_hi | (1 << 8); | |
1675 | ||
1676 | I915_WRITE(FW_BLC, fwater_lo); | |
1677 | I915_WRITE(FW_BLC2, fwater_hi); | |
1678 | ||
5209b1f4 ID |
1679 | if (enabled) |
1680 | intel_set_memory_cxsr(dev_priv, true); | |
b445e3b0 ED |
1681 | } |
1682 | ||
feb56b93 | 1683 | static void i845_update_wm(struct drm_crtc *unused_crtc) |
b445e3b0 | 1684 | { |
46ba614c | 1685 | struct drm_device *dev = unused_crtc->dev; |
b445e3b0 ED |
1686 | struct drm_i915_private *dev_priv = dev->dev_private; |
1687 | struct drm_crtc *crtc; | |
241bfc38 | 1688 | const struct drm_display_mode *adjusted_mode; |
b445e3b0 ED |
1689 | uint32_t fwater_lo; |
1690 | int planea_wm; | |
1691 | ||
1692 | crtc = single_enabled_crtc(dev); | |
1693 | if (crtc == NULL) | |
1694 | return; | |
1695 | ||
6e3c9717 | 1696 | adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode; |
241bfc38 | 1697 | planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock, |
feb56b93 | 1698 | &i845_wm_info, |
b445e3b0 | 1699 | dev_priv->display.get_fifo_size(dev, 0), |
5aef6003 | 1700 | 4, pessimal_latency_ns); |
b445e3b0 ED |
1701 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
1702 | fwater_lo |= (3<<8) | planea_wm; | |
1703 | ||
1704 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); | |
1705 | ||
1706 | I915_WRITE(FW_BLC, fwater_lo); | |
1707 | } | |
1708 | ||
8cfb3407 | 1709 | uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state *pipe_config) |
801bcfff | 1710 | { |
fd4daa9c | 1711 | uint32_t pixel_rate; |
801bcfff | 1712 | |
8cfb3407 | 1713 | pixel_rate = pipe_config->base.adjusted_mode.crtc_clock; |
801bcfff PZ |
1714 | |
1715 | /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to | |
1716 | * adjust the pixel_rate here. */ | |
1717 | ||
8cfb3407 | 1718 | if (pipe_config->pch_pfit.enabled) { |
801bcfff | 1719 | uint64_t pipe_w, pipe_h, pfit_w, pfit_h; |
8cfb3407 VS |
1720 | uint32_t pfit_size = pipe_config->pch_pfit.size; |
1721 | ||
1722 | pipe_w = pipe_config->pipe_src_w; | |
1723 | pipe_h = pipe_config->pipe_src_h; | |
801bcfff | 1724 | |
801bcfff PZ |
1725 | pfit_w = (pfit_size >> 16) & 0xFFFF; |
1726 | pfit_h = pfit_size & 0xFFFF; | |
1727 | if (pipe_w < pfit_w) | |
1728 | pipe_w = pfit_w; | |
1729 | if (pipe_h < pfit_h) | |
1730 | pipe_h = pfit_h; | |
1731 | ||
1732 | pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h, | |
1733 | pfit_w * pfit_h); | |
1734 | } | |
1735 | ||
1736 | return pixel_rate; | |
1737 | } | |
1738 | ||
37126462 | 1739 | /* latency must be in 0.1us units. */ |
23297044 | 1740 | static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, |
801bcfff PZ |
1741 | uint32_t latency) |
1742 | { | |
1743 | uint64_t ret; | |
1744 | ||
3312ba65 VS |
1745 | if (WARN(latency == 0, "Latency value missing\n")) |
1746 | return UINT_MAX; | |
1747 | ||
801bcfff PZ |
1748 | ret = (uint64_t) pixel_rate * bytes_per_pixel * latency; |
1749 | ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2; | |
1750 | ||
1751 | return ret; | |
1752 | } | |
1753 | ||
37126462 | 1754 | /* latency must be in 0.1us units. */ |
23297044 | 1755 | static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, |
801bcfff PZ |
1756 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, |
1757 | uint32_t latency) | |
1758 | { | |
1759 | uint32_t ret; | |
1760 | ||
3312ba65 VS |
1761 | if (WARN(latency == 0, "Latency value missing\n")) |
1762 | return UINT_MAX; | |
1763 | ||
801bcfff PZ |
1764 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); |
1765 | ret = (ret + 1) * horiz_pixels * bytes_per_pixel; | |
1766 | ret = DIV_ROUND_UP(ret, 64) + 2; | |
1767 | return ret; | |
1768 | } | |
1769 | ||
23297044 | 1770 | static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels, |
cca32e9a PZ |
1771 | uint8_t bytes_per_pixel) |
1772 | { | |
1773 | return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2; | |
1774 | } | |
1775 | ||
2ac96d2a PB |
1776 | struct skl_pipe_wm_parameters { |
1777 | bool active; | |
1778 | uint32_t pipe_htotal; | |
1779 | uint32_t pixel_rate; /* in KHz */ | |
1780 | struct intel_plane_wm_parameters plane[I915_MAX_PLANES]; | |
1781 | struct intel_plane_wm_parameters cursor; | |
1782 | }; | |
1783 | ||
820c1980 | 1784 | struct ilk_pipe_wm_parameters { |
801bcfff | 1785 | bool active; |
801bcfff PZ |
1786 | uint32_t pipe_htotal; |
1787 | uint32_t pixel_rate; | |
c35426d2 VS |
1788 | struct intel_plane_wm_parameters pri; |
1789 | struct intel_plane_wm_parameters spr; | |
1790 | struct intel_plane_wm_parameters cur; | |
801bcfff PZ |
1791 | }; |
1792 | ||
820c1980 | 1793 | struct ilk_wm_maximums { |
cca32e9a PZ |
1794 | uint16_t pri; |
1795 | uint16_t spr; | |
1796 | uint16_t cur; | |
1797 | uint16_t fbc; | |
1798 | }; | |
1799 | ||
240264f4 VS |
1800 | /* used in computing the new watermarks state */ |
1801 | struct intel_wm_config { | |
1802 | unsigned int num_pipes_active; | |
1803 | bool sprites_enabled; | |
1804 | bool sprites_scaled; | |
240264f4 VS |
1805 | }; |
1806 | ||
37126462 VS |
1807 | /* |
1808 | * For both WM_PIPE and WM_LP. | |
1809 | * mem_value must be in 0.1us units. | |
1810 | */ | |
820c1980 | 1811 | static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters *params, |
cca32e9a PZ |
1812 | uint32_t mem_value, |
1813 | bool is_lp) | |
801bcfff | 1814 | { |
cca32e9a PZ |
1815 | uint32_t method1, method2; |
1816 | ||
c35426d2 | 1817 | if (!params->active || !params->pri.enabled) |
801bcfff PZ |
1818 | return 0; |
1819 | ||
23297044 | 1820 | method1 = ilk_wm_method1(params->pixel_rate, |
c35426d2 | 1821 | params->pri.bytes_per_pixel, |
cca32e9a PZ |
1822 | mem_value); |
1823 | ||
1824 | if (!is_lp) | |
1825 | return method1; | |
1826 | ||
23297044 | 1827 | method2 = ilk_wm_method2(params->pixel_rate, |
cca32e9a | 1828 | params->pipe_htotal, |
c35426d2 VS |
1829 | params->pri.horiz_pixels, |
1830 | params->pri.bytes_per_pixel, | |
cca32e9a PZ |
1831 | mem_value); |
1832 | ||
1833 | return min(method1, method2); | |
801bcfff PZ |
1834 | } |
1835 | ||
37126462 VS |
1836 | /* |
1837 | * For both WM_PIPE and WM_LP. | |
1838 | * mem_value must be in 0.1us units. | |
1839 | */ | |
820c1980 | 1840 | static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters *params, |
801bcfff PZ |
1841 | uint32_t mem_value) |
1842 | { | |
1843 | uint32_t method1, method2; | |
1844 | ||
c35426d2 | 1845 | if (!params->active || !params->spr.enabled) |
801bcfff PZ |
1846 | return 0; |
1847 | ||
23297044 | 1848 | method1 = ilk_wm_method1(params->pixel_rate, |
c35426d2 | 1849 | params->spr.bytes_per_pixel, |
801bcfff | 1850 | mem_value); |
23297044 | 1851 | method2 = ilk_wm_method2(params->pixel_rate, |
801bcfff | 1852 | params->pipe_htotal, |
c35426d2 VS |
1853 | params->spr.horiz_pixels, |
1854 | params->spr.bytes_per_pixel, | |
801bcfff PZ |
1855 | mem_value); |
1856 | return min(method1, method2); | |
1857 | } | |
1858 | ||
37126462 VS |
1859 | /* |
1860 | * For both WM_PIPE and WM_LP. | |
1861 | * mem_value must be in 0.1us units. | |
1862 | */ | |
820c1980 | 1863 | static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters *params, |
801bcfff PZ |
1864 | uint32_t mem_value) |
1865 | { | |
c35426d2 | 1866 | if (!params->active || !params->cur.enabled) |
801bcfff PZ |
1867 | return 0; |
1868 | ||
23297044 | 1869 | return ilk_wm_method2(params->pixel_rate, |
801bcfff | 1870 | params->pipe_htotal, |
c35426d2 VS |
1871 | params->cur.horiz_pixels, |
1872 | params->cur.bytes_per_pixel, | |
801bcfff PZ |
1873 | mem_value); |
1874 | } | |
1875 | ||
cca32e9a | 1876 | /* Only for WM_LP. */ |
820c1980 | 1877 | static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters *params, |
1fda9882 | 1878 | uint32_t pri_val) |
cca32e9a | 1879 | { |
c35426d2 | 1880 | if (!params->active || !params->pri.enabled) |
cca32e9a PZ |
1881 | return 0; |
1882 | ||
23297044 | 1883 | return ilk_wm_fbc(pri_val, |
c35426d2 VS |
1884 | params->pri.horiz_pixels, |
1885 | params->pri.bytes_per_pixel); | |
cca32e9a PZ |
1886 | } |
1887 | ||
158ae64f VS |
1888 | static unsigned int ilk_display_fifo_size(const struct drm_device *dev) |
1889 | { | |
416f4727 VS |
1890 | if (INTEL_INFO(dev)->gen >= 8) |
1891 | return 3072; | |
1892 | else if (INTEL_INFO(dev)->gen >= 7) | |
158ae64f VS |
1893 | return 768; |
1894 | else | |
1895 | return 512; | |
1896 | } | |
1897 | ||
4e975081 VS |
1898 | static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev, |
1899 | int level, bool is_sprite) | |
1900 | { | |
1901 | if (INTEL_INFO(dev)->gen >= 8) | |
1902 | /* BDW primary/sprite plane watermarks */ | |
1903 | return level == 0 ? 255 : 2047; | |
1904 | else if (INTEL_INFO(dev)->gen >= 7) | |
1905 | /* IVB/HSW primary/sprite plane watermarks */ | |
1906 | return level == 0 ? 127 : 1023; | |
1907 | else if (!is_sprite) | |
1908 | /* ILK/SNB primary plane watermarks */ | |
1909 | return level == 0 ? 127 : 511; | |
1910 | else | |
1911 | /* ILK/SNB sprite plane watermarks */ | |
1912 | return level == 0 ? 63 : 255; | |
1913 | } | |
1914 | ||
1915 | static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev, | |
1916 | int level) | |
1917 | { | |
1918 | if (INTEL_INFO(dev)->gen >= 7) | |
1919 | return level == 0 ? 63 : 255; | |
1920 | else | |
1921 | return level == 0 ? 31 : 63; | |
1922 | } | |
1923 | ||
1924 | static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev) | |
1925 | { | |
1926 | if (INTEL_INFO(dev)->gen >= 8) | |
1927 | return 31; | |
1928 | else | |
1929 | return 15; | |
1930 | } | |
1931 | ||
158ae64f VS |
1932 | /* Calculate the maximum primary/sprite plane watermark */ |
1933 | static unsigned int ilk_plane_wm_max(const struct drm_device *dev, | |
1934 | int level, | |
240264f4 | 1935 | const struct intel_wm_config *config, |
158ae64f VS |
1936 | enum intel_ddb_partitioning ddb_partitioning, |
1937 | bool is_sprite) | |
1938 | { | |
1939 | unsigned int fifo_size = ilk_display_fifo_size(dev); | |
158ae64f VS |
1940 | |
1941 | /* if sprites aren't enabled, sprites get nothing */ | |
240264f4 | 1942 | if (is_sprite && !config->sprites_enabled) |
158ae64f VS |
1943 | return 0; |
1944 | ||
1945 | /* HSW allows LP1+ watermarks even with multiple pipes */ | |
240264f4 | 1946 | if (level == 0 || config->num_pipes_active > 1) { |
158ae64f VS |
1947 | fifo_size /= INTEL_INFO(dev)->num_pipes; |
1948 | ||
1949 | /* | |
1950 | * For some reason the non self refresh | |
1951 | * FIFO size is only half of the self | |
1952 | * refresh FIFO size on ILK/SNB. | |
1953 | */ | |
1954 | if (INTEL_INFO(dev)->gen <= 6) | |
1955 | fifo_size /= 2; | |
1956 | } | |
1957 | ||
240264f4 | 1958 | if (config->sprites_enabled) { |
158ae64f VS |
1959 | /* level 0 is always calculated with 1:1 split */ |
1960 | if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) { | |
1961 | if (is_sprite) | |
1962 | fifo_size *= 5; | |
1963 | fifo_size /= 6; | |
1964 | } else { | |
1965 | fifo_size /= 2; | |
1966 | } | |
1967 | } | |
1968 | ||
1969 | /* clamp to max that the registers can hold */ | |
4e975081 | 1970 | return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite)); |
158ae64f VS |
1971 | } |
1972 | ||
1973 | /* Calculate the maximum cursor plane watermark */ | |
1974 | static unsigned int ilk_cursor_wm_max(const struct drm_device *dev, | |
240264f4 VS |
1975 | int level, |
1976 | const struct intel_wm_config *config) | |
158ae64f VS |
1977 | { |
1978 | /* HSW LP1+ watermarks w/ multiple pipes */ | |
240264f4 | 1979 | if (level > 0 && config->num_pipes_active > 1) |
158ae64f VS |
1980 | return 64; |
1981 | ||
1982 | /* otherwise just report max that registers can hold */ | |
4e975081 | 1983 | return ilk_cursor_wm_reg_max(dev, level); |
158ae64f VS |
1984 | } |
1985 | ||
d34ff9c6 | 1986 | static void ilk_compute_wm_maximums(const struct drm_device *dev, |
34982fe1 VS |
1987 | int level, |
1988 | const struct intel_wm_config *config, | |
1989 | enum intel_ddb_partitioning ddb_partitioning, | |
820c1980 | 1990 | struct ilk_wm_maximums *max) |
158ae64f | 1991 | { |
240264f4 VS |
1992 | max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false); |
1993 | max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true); | |
1994 | max->cur = ilk_cursor_wm_max(dev, level, config); | |
4e975081 | 1995 | max->fbc = ilk_fbc_wm_reg_max(dev); |
158ae64f VS |
1996 | } |
1997 | ||
a3cb4048 VS |
1998 | static void ilk_compute_wm_reg_maximums(struct drm_device *dev, |
1999 | int level, | |
2000 | struct ilk_wm_maximums *max) | |
2001 | { | |
2002 | max->pri = ilk_plane_wm_reg_max(dev, level, false); | |
2003 | max->spr = ilk_plane_wm_reg_max(dev, level, true); | |
2004 | max->cur = ilk_cursor_wm_reg_max(dev, level); | |
2005 | max->fbc = ilk_fbc_wm_reg_max(dev); | |
2006 | } | |
2007 | ||
d9395655 | 2008 | static bool ilk_validate_wm_level(int level, |
820c1980 | 2009 | const struct ilk_wm_maximums *max, |
d9395655 | 2010 | struct intel_wm_level *result) |
a9786a11 VS |
2011 | { |
2012 | bool ret; | |
2013 | ||
2014 | /* already determined to be invalid? */ | |
2015 | if (!result->enable) | |
2016 | return false; | |
2017 | ||
2018 | result->enable = result->pri_val <= max->pri && | |
2019 | result->spr_val <= max->spr && | |
2020 | result->cur_val <= max->cur; | |
2021 | ||
2022 | ret = result->enable; | |
2023 | ||
2024 | /* | |
2025 | * HACK until we can pre-compute everything, | |
2026 | * and thus fail gracefully if LP0 watermarks | |
2027 | * are exceeded... | |
2028 | */ | |
2029 | if (level == 0 && !result->enable) { | |
2030 | if (result->pri_val > max->pri) | |
2031 | DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n", | |
2032 | level, result->pri_val, max->pri); | |
2033 | if (result->spr_val > max->spr) | |
2034 | DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n", | |
2035 | level, result->spr_val, max->spr); | |
2036 | if (result->cur_val > max->cur) | |
2037 | DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n", | |
2038 | level, result->cur_val, max->cur); | |
2039 | ||
2040 | result->pri_val = min_t(uint32_t, result->pri_val, max->pri); | |
2041 | result->spr_val = min_t(uint32_t, result->spr_val, max->spr); | |
2042 | result->cur_val = min_t(uint32_t, result->cur_val, max->cur); | |
2043 | result->enable = true; | |
2044 | } | |
2045 | ||
a9786a11 VS |
2046 | return ret; |
2047 | } | |
2048 | ||
d34ff9c6 | 2049 | static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv, |
6f5ddd17 | 2050 | int level, |
820c1980 | 2051 | const struct ilk_pipe_wm_parameters *p, |
1fd527cc | 2052 | struct intel_wm_level *result) |
6f5ddd17 VS |
2053 | { |
2054 | uint16_t pri_latency = dev_priv->wm.pri_latency[level]; | |
2055 | uint16_t spr_latency = dev_priv->wm.spr_latency[level]; | |
2056 | uint16_t cur_latency = dev_priv->wm.cur_latency[level]; | |
2057 | ||
2058 | /* WM1+ latency values stored in 0.5us units */ | |
2059 | if (level > 0) { | |
2060 | pri_latency *= 5; | |
2061 | spr_latency *= 5; | |
2062 | cur_latency *= 5; | |
2063 | } | |
2064 | ||
2065 | result->pri_val = ilk_compute_pri_wm(p, pri_latency, level); | |
2066 | result->spr_val = ilk_compute_spr_wm(p, spr_latency); | |
2067 | result->cur_val = ilk_compute_cur_wm(p, cur_latency); | |
2068 | result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val); | |
2069 | result->enable = true; | |
2070 | } | |
2071 | ||
801bcfff PZ |
2072 | static uint32_t |
2073 | hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc) | |
1f8eeabf ED |
2074 | { |
2075 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1011d8c4 | 2076 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
6e3c9717 | 2077 | struct drm_display_mode *mode = &intel_crtc->config->base.adjusted_mode; |
85a02deb | 2078 | u32 linetime, ips_linetime; |
1f8eeabf | 2079 | |
3ef00284 | 2080 | if (!intel_crtc->active) |
801bcfff | 2081 | return 0; |
1011d8c4 | 2082 | |
1f8eeabf ED |
2083 | /* The WM are computed with base on how long it takes to fill a single |
2084 | * row at the given clock rate, multiplied by 8. | |
2085 | * */ | |
fec8cba3 JB |
2086 | linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8, |
2087 | mode->crtc_clock); | |
2088 | ips_linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8, | |
05024da3 | 2089 | dev_priv->cdclk_freq); |
1f8eeabf | 2090 | |
801bcfff PZ |
2091 | return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) | |
2092 | PIPE_WM_LINETIME_TIME(linetime); | |
1f8eeabf ED |
2093 | } |
2094 | ||
2af30a5c | 2095 | static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[8]) |
12b134df VS |
2096 | { |
2097 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2098 | ||
2af30a5c PB |
2099 | if (IS_GEN9(dev)) { |
2100 | uint32_t val; | |
4f947386 | 2101 | int ret, i; |
367294be | 2102 | int level, max_level = ilk_wm_max_level(dev); |
2af30a5c PB |
2103 | |
2104 | /* read the first set of memory latencies[0:3] */ | |
2105 | val = 0; /* data0 to be programmed to 0 for first set */ | |
2106 | mutex_lock(&dev_priv->rps.hw_lock); | |
2107 | ret = sandybridge_pcode_read(dev_priv, | |
2108 | GEN9_PCODE_READ_MEM_LATENCY, | |
2109 | &val); | |
2110 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2111 | ||
2112 | if (ret) { | |
2113 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); | |
2114 | return; | |
2115 | } | |
2116 | ||
2117 | wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK; | |
2118 | wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & | |
2119 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2120 | wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & | |
2121 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2122 | wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & | |
2123 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2124 | ||
2125 | /* read the second set of memory latencies[4:7] */ | |
2126 | val = 1; /* data0 to be programmed to 1 for second set */ | |
2127 | mutex_lock(&dev_priv->rps.hw_lock); | |
2128 | ret = sandybridge_pcode_read(dev_priv, | |
2129 | GEN9_PCODE_READ_MEM_LATENCY, | |
2130 | &val); | |
2131 | mutex_unlock(&dev_priv->rps.hw_lock); | |
2132 | if (ret) { | |
2133 | DRM_ERROR("SKL Mailbox read error = %d\n", ret); | |
2134 | return; | |
2135 | } | |
2136 | ||
2137 | wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK; | |
2138 | wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) & | |
2139 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2140 | wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) & | |
2141 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2142 | wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) & | |
2143 | GEN9_MEM_LATENCY_LEVEL_MASK; | |
2144 | ||
367294be | 2145 | /* |
6f97235b DL |
2146 | * WaWmMemoryReadLatency:skl |
2147 | * | |
367294be VK |
2148 | * punit doesn't take into account the read latency so we need |
2149 | * to add 2us to the various latency levels we retrieve from | |
2150 | * the punit. | |
2151 | * - W0 is a bit special in that it's the only level that | |
2152 | * can't be disabled if we want to have display working, so | |
2153 | * we always add 2us there. | |
2154 | * - For levels >=1, punit returns 0us latency when they are | |
2155 | * disabled, so we respect that and don't add 2us then | |
4f947386 VK |
2156 | * |
2157 | * Additionally, if a level n (n > 1) has a 0us latency, all | |
2158 | * levels m (m >= n) need to be disabled. We make sure to | |
2159 | * sanitize the values out of the punit to satisfy this | |
2160 | * requirement. | |
367294be VK |
2161 | */ |
2162 | wm[0] += 2; | |
2163 | for (level = 1; level <= max_level; level++) | |
2164 | if (wm[level] != 0) | |
2165 | wm[level] += 2; | |
4f947386 VK |
2166 | else { |
2167 | for (i = level + 1; i <= max_level; i++) | |
2168 | wm[i] = 0; | |
367294be | 2169 | |
4f947386 VK |
2170 | break; |
2171 | } | |
2af30a5c | 2172 | } else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
12b134df VS |
2173 | uint64_t sskpd = I915_READ64(MCH_SSKPD); |
2174 | ||
2175 | wm[0] = (sskpd >> 56) & 0xFF; | |
2176 | if (wm[0] == 0) | |
2177 | wm[0] = sskpd & 0xF; | |
e5d5019e VS |
2178 | wm[1] = (sskpd >> 4) & 0xFF; |
2179 | wm[2] = (sskpd >> 12) & 0xFF; | |
2180 | wm[3] = (sskpd >> 20) & 0x1FF; | |
2181 | wm[4] = (sskpd >> 32) & 0x1FF; | |
63cf9a13 VS |
2182 | } else if (INTEL_INFO(dev)->gen >= 6) { |
2183 | uint32_t sskpd = I915_READ(MCH_SSKPD); | |
2184 | ||
2185 | wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK; | |
2186 | wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK; | |
2187 | wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK; | |
2188 | wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK; | |
3a88d0ac VS |
2189 | } else if (INTEL_INFO(dev)->gen >= 5) { |
2190 | uint32_t mltr = I915_READ(MLTR_ILK); | |
2191 | ||
2192 | /* ILK primary LP0 latency is 700 ns */ | |
2193 | wm[0] = 7; | |
2194 | wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK; | |
2195 | wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK; | |
12b134df VS |
2196 | } |
2197 | } | |
2198 | ||
53615a5e VS |
2199 | static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5]) |
2200 | { | |
2201 | /* ILK sprite LP0 latency is 1300 ns */ | |
2202 | if (INTEL_INFO(dev)->gen == 5) | |
2203 | wm[0] = 13; | |
2204 | } | |
2205 | ||
2206 | static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5]) | |
2207 | { | |
2208 | /* ILK cursor LP0 latency is 1300 ns */ | |
2209 | if (INTEL_INFO(dev)->gen == 5) | |
2210 | wm[0] = 13; | |
2211 | ||
2212 | /* WaDoubleCursorLP3Latency:ivb */ | |
2213 | if (IS_IVYBRIDGE(dev)) | |
2214 | wm[3] *= 2; | |
2215 | } | |
2216 | ||
546c81fd | 2217 | int ilk_wm_max_level(const struct drm_device *dev) |
26ec971e | 2218 | { |
26ec971e | 2219 | /* how many WM levels are we expecting */ |
b6e742f6 | 2220 | if (INTEL_INFO(dev)->gen >= 9) |
2af30a5c PB |
2221 | return 7; |
2222 | else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) | |
ad0d6dc4 | 2223 | return 4; |
26ec971e | 2224 | else if (INTEL_INFO(dev)->gen >= 6) |
ad0d6dc4 | 2225 | return 3; |
26ec971e | 2226 | else |
ad0d6dc4 VS |
2227 | return 2; |
2228 | } | |
7526ed79 | 2229 | |
ad0d6dc4 VS |
2230 | static void intel_print_wm_latency(struct drm_device *dev, |
2231 | const char *name, | |
2af30a5c | 2232 | const uint16_t wm[8]) |
ad0d6dc4 VS |
2233 | { |
2234 | int level, max_level = ilk_wm_max_level(dev); | |
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 | */ | |
2249 | if (IS_GEN9(dev)) | |
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 | { | |
2263 | int level, max_level = ilk_wm_max_level(dev_priv->dev); | |
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 | ||
2275 | static void snb_wm_latency_quirk(struct drm_device *dev) | |
2276 | { | |
2277 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2278 | bool changed; | |
2279 | ||
2280 | /* | |
2281 | * The BIOS provided WM memory latency values are often | |
2282 | * inadequate for high resolution displays. Adjust them. | |
2283 | */ | |
2284 | changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) | | |
2285 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) | | |
2286 | ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12); | |
2287 | ||
2288 | if (!changed) | |
2289 | return; | |
2290 | ||
2291 | DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n"); | |
2292 | intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency); | |
2293 | intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency); | |
2294 | intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency); | |
2295 | } | |
2296 | ||
fa50ad61 | 2297 | static void ilk_setup_wm_latency(struct drm_device *dev) |
53615a5e VS |
2298 | { |
2299 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2300 | ||
2301 | intel_read_wm_latency(dev, dev_priv->wm.pri_latency); | |
2302 | ||
2303 | memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency, | |
2304 | sizeof(dev_priv->wm.pri_latency)); | |
2305 | memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency, | |
2306 | sizeof(dev_priv->wm.pri_latency)); | |
2307 | ||
2308 | intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency); | |
2309 | intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency); | |
26ec971e VS |
2310 | |
2311 | intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency); | |
2312 | intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency); | |
2313 | intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency); | |
e95a2f75 VS |
2314 | |
2315 | if (IS_GEN6(dev)) | |
2316 | snb_wm_latency_quirk(dev); | |
53615a5e VS |
2317 | } |
2318 | ||
2af30a5c PB |
2319 | static void skl_setup_wm_latency(struct drm_device *dev) |
2320 | { | |
2321 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2322 | ||
2323 | intel_read_wm_latency(dev, dev_priv->wm.skl_latency); | |
2324 | intel_print_wm_latency(dev, "Gen9 Plane", dev_priv->wm.skl_latency); | |
2325 | } | |
2326 | ||
820c1980 | 2327 | static void ilk_compute_wm_parameters(struct drm_crtc *crtc, |
2a44b76b | 2328 | struct ilk_pipe_wm_parameters *p) |
1011d8c4 | 2329 | { |
7c4a395f VS |
2330 | struct drm_device *dev = crtc->dev; |
2331 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
2332 | enum pipe pipe = intel_crtc->pipe; | |
7c4a395f | 2333 | struct drm_plane *plane; |
1011d8c4 | 2334 | |
3ef00284 | 2335 | if (!intel_crtc->active) |
2a44b76b | 2336 | return; |
801bcfff | 2337 | |
2a44b76b | 2338 | p->active = true; |
6e3c9717 | 2339 | p->pipe_htotal = intel_crtc->config->base.adjusted_mode.crtc_htotal; |
8cfb3407 | 2340 | p->pixel_rate = ilk_pipe_pixel_rate(intel_crtc->config); |
c9f038a1 | 2341 | |
54da691d | 2342 | if (crtc->primary->state->fb) |
c9f038a1 MR |
2343 | p->pri.bytes_per_pixel = |
2344 | crtc->primary->state->fb->bits_per_pixel / 8; | |
54da691d TG |
2345 | else |
2346 | p->pri.bytes_per_pixel = 4; | |
2347 | ||
2348 | p->cur.bytes_per_pixel = 4; | |
2349 | /* | |
2350 | * TODO: for now, assume primary and cursor planes are always enabled. | |
2351 | * Setting them to false makes the screen flicker. | |
2352 | */ | |
2353 | p->pri.enabled = true; | |
2354 | p->cur.enabled = true; | |
c9f038a1 | 2355 | |
6e3c9717 | 2356 | p->pri.horiz_pixels = intel_crtc->config->pipe_src_w; |
3dd512fb | 2357 | p->cur.horiz_pixels = intel_crtc->base.cursor->state->crtc_w; |
7c4a395f | 2358 | |
af2b653b | 2359 | drm_for_each_legacy_plane(plane, &dev->mode_config.plane_list) { |
801bcfff | 2360 | struct intel_plane *intel_plane = to_intel_plane(plane); |
801bcfff | 2361 | |
2a44b76b | 2362 | if (intel_plane->pipe == pipe) { |
7c4a395f | 2363 | p->spr = intel_plane->wm; |
2a44b76b VS |
2364 | break; |
2365 | } | |
2366 | } | |
2367 | } | |
2368 | ||
2369 | static void ilk_compute_wm_config(struct drm_device *dev, | |
2370 | struct intel_wm_config *config) | |
2371 | { | |
2372 | struct intel_crtc *intel_crtc; | |
2373 | ||
2374 | /* Compute the currently _active_ config */ | |
d3fcc808 | 2375 | for_each_intel_crtc(dev, intel_crtc) { |
2a44b76b | 2376 | const struct intel_pipe_wm *wm = &intel_crtc->wm.active; |
cca32e9a | 2377 | |
2a44b76b VS |
2378 | if (!wm->pipe_enabled) |
2379 | continue; | |
cca32e9a | 2380 | |
2a44b76b VS |
2381 | config->sprites_enabled |= wm->sprites_enabled; |
2382 | config->sprites_scaled |= wm->sprites_scaled; | |
2383 | config->num_pipes_active++; | |
cca32e9a | 2384 | } |
801bcfff PZ |
2385 | } |
2386 | ||
0b2ae6d7 VS |
2387 | /* Compute new watermarks for the pipe */ |
2388 | static bool intel_compute_pipe_wm(struct drm_crtc *crtc, | |
820c1980 | 2389 | const struct ilk_pipe_wm_parameters *params, |
0b2ae6d7 VS |
2390 | struct intel_pipe_wm *pipe_wm) |
2391 | { | |
2392 | struct drm_device *dev = crtc->dev; | |
d34ff9c6 | 2393 | const struct drm_i915_private *dev_priv = dev->dev_private; |
0b2ae6d7 VS |
2394 | int level, max_level = ilk_wm_max_level(dev); |
2395 | /* LP0 watermark maximums depend on this pipe alone */ | |
2396 | struct intel_wm_config config = { | |
2397 | .num_pipes_active = 1, | |
2398 | .sprites_enabled = params->spr.enabled, | |
2399 | .sprites_scaled = params->spr.scaled, | |
2400 | }; | |
820c1980 | 2401 | struct ilk_wm_maximums max; |
0b2ae6d7 | 2402 | |
2a44b76b VS |
2403 | pipe_wm->pipe_enabled = params->active; |
2404 | pipe_wm->sprites_enabled = params->spr.enabled; | |
2405 | pipe_wm->sprites_scaled = params->spr.scaled; | |
2406 | ||
7b39a0b7 VS |
2407 | /* ILK/SNB: LP2+ watermarks only w/o sprites */ |
2408 | if (INTEL_INFO(dev)->gen <= 6 && params->spr.enabled) | |
2409 | max_level = 1; | |
2410 | ||
2411 | /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */ | |
2412 | if (params->spr.scaled) | |
2413 | max_level = 0; | |
2414 | ||
a3cb4048 | 2415 | ilk_compute_wm_level(dev_priv, 0, params, &pipe_wm->wm[0]); |
0b2ae6d7 | 2416 | |
a42a5719 | 2417 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
ce0e0713 | 2418 | pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc); |
0b2ae6d7 | 2419 | |
a3cb4048 VS |
2420 | /* LP0 watermarks always use 1/2 DDB partitioning */ |
2421 | ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max); | |
2422 | ||
0b2ae6d7 | 2423 | /* At least LP0 must be valid */ |
a3cb4048 VS |
2424 | if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) |
2425 | return false; | |
2426 | ||
2427 | ilk_compute_wm_reg_maximums(dev, 1, &max); | |
2428 | ||
2429 | for (level = 1; level <= max_level; level++) { | |
2430 | struct intel_wm_level wm = {}; | |
2431 | ||
2432 | ilk_compute_wm_level(dev_priv, level, params, &wm); | |
2433 | ||
2434 | /* | |
2435 | * Disable any watermark level that exceeds the | |
2436 | * register maximums since such watermarks are | |
2437 | * always invalid. | |
2438 | */ | |
2439 | if (!ilk_validate_wm_level(level, &max, &wm)) | |
2440 | break; | |
2441 | ||
2442 | pipe_wm->wm[level] = wm; | |
2443 | } | |
2444 | ||
2445 | return true; | |
0b2ae6d7 VS |
2446 | } |
2447 | ||
2448 | /* | |
2449 | * Merge the watermarks from all active pipes for a specific level. | |
2450 | */ | |
2451 | static void ilk_merge_wm_level(struct drm_device *dev, | |
2452 | int level, | |
2453 | struct intel_wm_level *ret_wm) | |
2454 | { | |
2455 | const struct intel_crtc *intel_crtc; | |
2456 | ||
d52fea5b VS |
2457 | ret_wm->enable = true; |
2458 | ||
d3fcc808 | 2459 | for_each_intel_crtc(dev, intel_crtc) { |
fe392efd VS |
2460 | const struct intel_pipe_wm *active = &intel_crtc->wm.active; |
2461 | const struct intel_wm_level *wm = &active->wm[level]; | |
2462 | ||
2463 | if (!active->pipe_enabled) | |
2464 | continue; | |
0b2ae6d7 | 2465 | |
d52fea5b VS |
2466 | /* |
2467 | * The watermark values may have been used in the past, | |
2468 | * so we must maintain them in the registers for some | |
2469 | * time even if the level is now disabled. | |
2470 | */ | |
0b2ae6d7 | 2471 | if (!wm->enable) |
d52fea5b | 2472 | ret_wm->enable = false; |
0b2ae6d7 VS |
2473 | |
2474 | ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val); | |
2475 | ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val); | |
2476 | ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val); | |
2477 | ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val); | |
2478 | } | |
0b2ae6d7 VS |
2479 | } |
2480 | ||
2481 | /* | |
2482 | * Merge all low power watermarks for all active pipes. | |
2483 | */ | |
2484 | static void ilk_wm_merge(struct drm_device *dev, | |
0ba22e26 | 2485 | const struct intel_wm_config *config, |
820c1980 | 2486 | const struct ilk_wm_maximums *max, |
0b2ae6d7 VS |
2487 | struct intel_pipe_wm *merged) |
2488 | { | |
7733b49b | 2489 | struct drm_i915_private *dev_priv = dev->dev_private; |
0b2ae6d7 | 2490 | int level, max_level = ilk_wm_max_level(dev); |
d52fea5b | 2491 | int last_enabled_level = max_level; |
0b2ae6d7 | 2492 | |
0ba22e26 VS |
2493 | /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */ |
2494 | if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) && | |
2495 | config->num_pipes_active > 1) | |
2496 | return; | |
2497 | ||
6c8b6c28 VS |
2498 | /* ILK: FBC WM must be disabled always */ |
2499 | merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6; | |
0b2ae6d7 VS |
2500 | |
2501 | /* merge each WM1+ level */ | |
2502 | for (level = 1; level <= max_level; level++) { | |
2503 | struct intel_wm_level *wm = &merged->wm[level]; | |
2504 | ||
2505 | ilk_merge_wm_level(dev, level, wm); | |
2506 | ||
d52fea5b VS |
2507 | if (level > last_enabled_level) |
2508 | wm->enable = false; | |
2509 | else if (!ilk_validate_wm_level(level, max, wm)) | |
2510 | /* make sure all following levels get disabled */ | |
2511 | last_enabled_level = level - 1; | |
0b2ae6d7 VS |
2512 | |
2513 | /* | |
2514 | * The spec says it is preferred to disable | |
2515 | * FBC WMs instead of disabling a WM level. | |
2516 | */ | |
2517 | if (wm->fbc_val > max->fbc) { | |
d52fea5b VS |
2518 | if (wm->enable) |
2519 | merged->fbc_wm_enabled = false; | |
0b2ae6d7 VS |
2520 | wm->fbc_val = 0; |
2521 | } | |
2522 | } | |
6c8b6c28 VS |
2523 | |
2524 | /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */ | |
2525 | /* | |
2526 | * FIXME this is racy. FBC might get enabled later. | |
2527 | * What we should check here is whether FBC can be | |
2528 | * enabled sometime later. | |
2529 | */ | |
7733b49b PZ |
2530 | if (IS_GEN5(dev) && !merged->fbc_wm_enabled && |
2531 | intel_fbc_enabled(dev_priv)) { | |
6c8b6c28 VS |
2532 | for (level = 2; level <= max_level; level++) { |
2533 | struct intel_wm_level *wm = &merged->wm[level]; | |
2534 | ||
2535 | wm->enable = false; | |
2536 | } | |
2537 | } | |
0b2ae6d7 VS |
2538 | } |
2539 | ||
b380ca3c VS |
2540 | static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm) |
2541 | { | |
2542 | /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */ | |
2543 | return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable); | |
2544 | } | |
2545 | ||
a68d68ee VS |
2546 | /* The value we need to program into the WM_LPx latency field */ |
2547 | static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level) | |
2548 | { | |
2549 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2550 | ||
a42a5719 | 2551 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
a68d68ee VS |
2552 | return 2 * level; |
2553 | else | |
2554 | return dev_priv->wm.pri_latency[level]; | |
2555 | } | |
2556 | ||
820c1980 | 2557 | static void ilk_compute_wm_results(struct drm_device *dev, |
0362c781 | 2558 | const struct intel_pipe_wm *merged, |
609cedef | 2559 | enum intel_ddb_partitioning partitioning, |
820c1980 | 2560 | struct ilk_wm_values *results) |
801bcfff | 2561 | { |
0b2ae6d7 VS |
2562 | struct intel_crtc *intel_crtc; |
2563 | int level, wm_lp; | |
cca32e9a | 2564 | |
0362c781 | 2565 | results->enable_fbc_wm = merged->fbc_wm_enabled; |
609cedef | 2566 | results->partitioning = partitioning; |
cca32e9a | 2567 | |
0b2ae6d7 | 2568 | /* LP1+ register values */ |
cca32e9a | 2569 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { |
1fd527cc | 2570 | const struct intel_wm_level *r; |
801bcfff | 2571 | |
b380ca3c | 2572 | level = ilk_wm_lp_to_level(wm_lp, merged); |
0b2ae6d7 | 2573 | |
0362c781 | 2574 | r = &merged->wm[level]; |
cca32e9a | 2575 | |
d52fea5b VS |
2576 | /* |
2577 | * Maintain the watermark values even if the level is | |
2578 | * disabled. Doing otherwise could cause underruns. | |
2579 | */ | |
2580 | results->wm_lp[wm_lp - 1] = | |
a68d68ee | 2581 | (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) | |
416f4727 VS |
2582 | (r->pri_val << WM1_LP_SR_SHIFT) | |
2583 | r->cur_val; | |
2584 | ||
d52fea5b VS |
2585 | if (r->enable) |
2586 | results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN; | |
2587 | ||
416f4727 VS |
2588 | if (INTEL_INFO(dev)->gen >= 8) |
2589 | results->wm_lp[wm_lp - 1] |= | |
2590 | r->fbc_val << WM1_LP_FBC_SHIFT_BDW; | |
2591 | else | |
2592 | results->wm_lp[wm_lp - 1] |= | |
2593 | r->fbc_val << WM1_LP_FBC_SHIFT; | |
2594 | ||
d52fea5b VS |
2595 | /* |
2596 | * Always set WM1S_LP_EN when spr_val != 0, even if the | |
2597 | * level is disabled. Doing otherwise could cause underruns. | |
2598 | */ | |
6cef2b8a VS |
2599 | if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) { |
2600 | WARN_ON(wm_lp != 1); | |
2601 | results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val; | |
2602 | } else | |
2603 | results->wm_lp_spr[wm_lp - 1] = r->spr_val; | |
cca32e9a | 2604 | } |
801bcfff | 2605 | |
0b2ae6d7 | 2606 | /* LP0 register values */ |
d3fcc808 | 2607 | for_each_intel_crtc(dev, intel_crtc) { |
0b2ae6d7 VS |
2608 | enum pipe pipe = intel_crtc->pipe; |
2609 | const struct intel_wm_level *r = | |
2610 | &intel_crtc->wm.active.wm[0]; | |
2611 | ||
2612 | if (WARN_ON(!r->enable)) | |
2613 | continue; | |
2614 | ||
2615 | results->wm_linetime[pipe] = intel_crtc->wm.active.linetime; | |
1011d8c4 | 2616 | |
0b2ae6d7 VS |
2617 | results->wm_pipe[pipe] = |
2618 | (r->pri_val << WM0_PIPE_PLANE_SHIFT) | | |
2619 | (r->spr_val << WM0_PIPE_SPRITE_SHIFT) | | |
2620 | r->cur_val; | |
801bcfff PZ |
2621 | } |
2622 | } | |
2623 | ||
861f3389 PZ |
2624 | /* Find the result with the highest level enabled. Check for enable_fbc_wm in |
2625 | * case both are at the same level. Prefer r1 in case they're the same. */ | |
820c1980 | 2626 | static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev, |
198a1e9b VS |
2627 | struct intel_pipe_wm *r1, |
2628 | struct intel_pipe_wm *r2) | |
861f3389 | 2629 | { |
198a1e9b VS |
2630 | int level, max_level = ilk_wm_max_level(dev); |
2631 | int level1 = 0, level2 = 0; | |
861f3389 | 2632 | |
198a1e9b VS |
2633 | for (level = 1; level <= max_level; level++) { |
2634 | if (r1->wm[level].enable) | |
2635 | level1 = level; | |
2636 | if (r2->wm[level].enable) | |
2637 | level2 = level; | |
861f3389 PZ |
2638 | } |
2639 | ||
198a1e9b VS |
2640 | if (level1 == level2) { |
2641 | if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled) | |
861f3389 PZ |
2642 | return r2; |
2643 | else | |
2644 | return r1; | |
198a1e9b | 2645 | } else if (level1 > level2) { |
861f3389 PZ |
2646 | return r1; |
2647 | } else { | |
2648 | return r2; | |
2649 | } | |
2650 | } | |
2651 | ||
49a687c4 VS |
2652 | /* dirty bits used to track which watermarks need changes */ |
2653 | #define WM_DIRTY_PIPE(pipe) (1 << (pipe)) | |
2654 | #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe))) | |
2655 | #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp))) | |
2656 | #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3)) | |
2657 | #define WM_DIRTY_FBC (1 << 24) | |
2658 | #define WM_DIRTY_DDB (1 << 25) | |
2659 | ||
055e393f | 2660 | static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv, |
820c1980 ID |
2661 | const struct ilk_wm_values *old, |
2662 | const struct ilk_wm_values *new) | |
49a687c4 VS |
2663 | { |
2664 | unsigned int dirty = 0; | |
2665 | enum pipe pipe; | |
2666 | int wm_lp; | |
2667 | ||
055e393f | 2668 | for_each_pipe(dev_priv, pipe) { |
49a687c4 VS |
2669 | if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) { |
2670 | dirty |= WM_DIRTY_LINETIME(pipe); | |
2671 | /* Must disable LP1+ watermarks too */ | |
2672 | dirty |= WM_DIRTY_LP_ALL; | |
2673 | } | |
2674 | ||
2675 | if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) { | |
2676 | dirty |= WM_DIRTY_PIPE(pipe); | |
2677 | /* Must disable LP1+ watermarks too */ | |
2678 | dirty |= WM_DIRTY_LP_ALL; | |
2679 | } | |
2680 | } | |
2681 | ||
2682 | if (old->enable_fbc_wm != new->enable_fbc_wm) { | |
2683 | dirty |= WM_DIRTY_FBC; | |
2684 | /* Must disable LP1+ watermarks too */ | |
2685 | dirty |= WM_DIRTY_LP_ALL; | |
2686 | } | |
2687 | ||
2688 | if (old->partitioning != new->partitioning) { | |
2689 | dirty |= WM_DIRTY_DDB; | |
2690 | /* Must disable LP1+ watermarks too */ | |
2691 | dirty |= WM_DIRTY_LP_ALL; | |
2692 | } | |
2693 | ||
2694 | /* LP1+ watermarks already deemed dirty, no need to continue */ | |
2695 | if (dirty & WM_DIRTY_LP_ALL) | |
2696 | return dirty; | |
2697 | ||
2698 | /* Find the lowest numbered LP1+ watermark in need of an update... */ | |
2699 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { | |
2700 | if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] || | |
2701 | old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1]) | |
2702 | break; | |
2703 | } | |
2704 | ||
2705 | /* ...and mark it and all higher numbered LP1+ watermarks as dirty */ | |
2706 | for (; wm_lp <= 3; wm_lp++) | |
2707 | dirty |= WM_DIRTY_LP(wm_lp); | |
2708 | ||
2709 | return dirty; | |
2710 | } | |
2711 | ||
8553c18e VS |
2712 | static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv, |
2713 | unsigned int dirty) | |
801bcfff | 2714 | { |
820c1980 | 2715 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
8553c18e | 2716 | bool changed = false; |
801bcfff | 2717 | |
facd619b VS |
2718 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) { |
2719 | previous->wm_lp[2] &= ~WM1_LP_SR_EN; | |
2720 | I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]); | |
8553c18e | 2721 | changed = true; |
facd619b VS |
2722 | } |
2723 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) { | |
2724 | previous->wm_lp[1] &= ~WM1_LP_SR_EN; | |
2725 | I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]); | |
8553c18e | 2726 | changed = true; |
facd619b VS |
2727 | } |
2728 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) { | |
2729 | previous->wm_lp[0] &= ~WM1_LP_SR_EN; | |
2730 | I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]); | |
8553c18e | 2731 | changed = true; |
facd619b | 2732 | } |
801bcfff | 2733 | |
facd619b VS |
2734 | /* |
2735 | * Don't touch WM1S_LP_EN here. | |
2736 | * Doing so could cause underruns. | |
2737 | */ | |
6cef2b8a | 2738 | |
8553c18e VS |
2739 | return changed; |
2740 | } | |
2741 | ||
2742 | /* | |
2743 | * The spec says we shouldn't write when we don't need, because every write | |
2744 | * causes WMs to be re-evaluated, expending some power. | |
2745 | */ | |
820c1980 ID |
2746 | static void ilk_write_wm_values(struct drm_i915_private *dev_priv, |
2747 | struct ilk_wm_values *results) | |
8553c18e VS |
2748 | { |
2749 | struct drm_device *dev = dev_priv->dev; | |
820c1980 | 2750 | struct ilk_wm_values *previous = &dev_priv->wm.hw; |
8553c18e VS |
2751 | unsigned int dirty; |
2752 | uint32_t val; | |
2753 | ||
055e393f | 2754 | dirty = ilk_compute_wm_dirty(dev_priv, previous, results); |
8553c18e VS |
2755 | if (!dirty) |
2756 | return; | |
2757 | ||
2758 | _ilk_disable_lp_wm(dev_priv, dirty); | |
2759 | ||
49a687c4 | 2760 | if (dirty & WM_DIRTY_PIPE(PIPE_A)) |
801bcfff | 2761 | I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]); |
49a687c4 | 2762 | if (dirty & WM_DIRTY_PIPE(PIPE_B)) |
801bcfff | 2763 | I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]); |
49a687c4 | 2764 | if (dirty & WM_DIRTY_PIPE(PIPE_C)) |
801bcfff PZ |
2765 | I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]); |
2766 | ||
49a687c4 | 2767 | if (dirty & WM_DIRTY_LINETIME(PIPE_A)) |
801bcfff | 2768 | I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]); |
49a687c4 | 2769 | if (dirty & WM_DIRTY_LINETIME(PIPE_B)) |
801bcfff | 2770 | I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]); |
49a687c4 | 2771 | if (dirty & WM_DIRTY_LINETIME(PIPE_C)) |
801bcfff PZ |
2772 | I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]); |
2773 | ||
49a687c4 | 2774 | if (dirty & WM_DIRTY_DDB) { |
a42a5719 | 2775 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) { |
ac9545fd VS |
2776 | val = I915_READ(WM_MISC); |
2777 | if (results->partitioning == INTEL_DDB_PART_1_2) | |
2778 | val &= ~WM_MISC_DATA_PARTITION_5_6; | |
2779 | else | |
2780 | val |= WM_MISC_DATA_PARTITION_5_6; | |
2781 | I915_WRITE(WM_MISC, val); | |
2782 | } else { | |
2783 | val = I915_READ(DISP_ARB_CTL2); | |
2784 | if (results->partitioning == INTEL_DDB_PART_1_2) | |
2785 | val &= ~DISP_DATA_PARTITION_5_6; | |
2786 | else | |
2787 | val |= DISP_DATA_PARTITION_5_6; | |
2788 | I915_WRITE(DISP_ARB_CTL2, val); | |
2789 | } | |
1011d8c4 PZ |
2790 | } |
2791 | ||
49a687c4 | 2792 | if (dirty & WM_DIRTY_FBC) { |
cca32e9a PZ |
2793 | val = I915_READ(DISP_ARB_CTL); |
2794 | if (results->enable_fbc_wm) | |
2795 | val &= ~DISP_FBC_WM_DIS; | |
2796 | else | |
2797 | val |= DISP_FBC_WM_DIS; | |
2798 | I915_WRITE(DISP_ARB_CTL, val); | |
2799 | } | |
2800 | ||
954911eb ID |
2801 | if (dirty & WM_DIRTY_LP(1) && |
2802 | previous->wm_lp_spr[0] != results->wm_lp_spr[0]) | |
2803 | I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]); | |
2804 | ||
2805 | if (INTEL_INFO(dev)->gen >= 7) { | |
6cef2b8a VS |
2806 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1]) |
2807 | I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]); | |
2808 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2]) | |
2809 | I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]); | |
2810 | } | |
801bcfff | 2811 | |
facd619b | 2812 | if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0]) |
801bcfff | 2813 | I915_WRITE(WM1_LP_ILK, results->wm_lp[0]); |
facd619b | 2814 | if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1]) |
801bcfff | 2815 | I915_WRITE(WM2_LP_ILK, results->wm_lp[1]); |
facd619b | 2816 | if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2]) |
801bcfff | 2817 | I915_WRITE(WM3_LP_ILK, results->wm_lp[2]); |
609cedef VS |
2818 | |
2819 | dev_priv->wm.hw = *results; | |
801bcfff PZ |
2820 | } |
2821 | ||
8553c18e VS |
2822 | static bool ilk_disable_lp_wm(struct drm_device *dev) |
2823 | { | |
2824 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2825 | ||
2826 | return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL); | |
2827 | } | |
2828 | ||
b9cec075 DL |
2829 | /* |
2830 | * On gen9, we need to allocate Display Data Buffer (DDB) portions to the | |
2831 | * different active planes. | |
2832 | */ | |
2833 | ||
2834 | #define SKL_DDB_SIZE 896 /* in blocks */ | |
43d735a6 | 2835 | #define BXT_DDB_SIZE 512 |
b9cec075 DL |
2836 | |
2837 | static void | |
2838 | skl_ddb_get_pipe_allocation_limits(struct drm_device *dev, | |
2839 | struct drm_crtc *for_crtc, | |
2840 | const struct intel_wm_config *config, | |
2841 | const struct skl_pipe_wm_parameters *params, | |
2842 | struct skl_ddb_entry *alloc /* out */) | |
2843 | { | |
2844 | struct drm_crtc *crtc; | |
2845 | unsigned int pipe_size, ddb_size; | |
2846 | int nth_active_pipe; | |
2847 | ||
2848 | if (!params->active) { | |
2849 | alloc->start = 0; | |
2850 | alloc->end = 0; | |
2851 | return; | |
2852 | } | |
2853 | ||
43d735a6 DL |
2854 | if (IS_BROXTON(dev)) |
2855 | ddb_size = BXT_DDB_SIZE; | |
2856 | else | |
2857 | ddb_size = SKL_DDB_SIZE; | |
b9cec075 DL |
2858 | |
2859 | ddb_size -= 4; /* 4 blocks for bypass path allocation */ | |
2860 | ||
2861 | nth_active_pipe = 0; | |
2862 | for_each_crtc(dev, crtc) { | |
3ef00284 | 2863 | if (!to_intel_crtc(crtc)->active) |
b9cec075 DL |
2864 | continue; |
2865 | ||
2866 | if (crtc == for_crtc) | |
2867 | break; | |
2868 | ||
2869 | nth_active_pipe++; | |
2870 | } | |
2871 | ||
2872 | pipe_size = ddb_size / config->num_pipes_active; | |
2873 | alloc->start = nth_active_pipe * ddb_size / config->num_pipes_active; | |
16160e3d | 2874 | alloc->end = alloc->start + pipe_size; |
b9cec075 DL |
2875 | } |
2876 | ||
2877 | static unsigned int skl_cursor_allocation(const struct intel_wm_config *config) | |
2878 | { | |
2879 | if (config->num_pipes_active == 1) | |
2880 | return 32; | |
2881 | ||
2882 | return 8; | |
2883 | } | |
2884 | ||
a269c583 DL |
2885 | static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg) |
2886 | { | |
2887 | entry->start = reg & 0x3ff; | |
2888 | entry->end = (reg >> 16) & 0x3ff; | |
16160e3d DL |
2889 | if (entry->end) |
2890 | entry->end += 1; | |
a269c583 DL |
2891 | } |
2892 | ||
08db6652 DL |
2893 | void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv, |
2894 | struct skl_ddb_allocation *ddb /* out */) | |
a269c583 | 2895 | { |
a269c583 DL |
2896 | enum pipe pipe; |
2897 | int plane; | |
2898 | u32 val; | |
2899 | ||
2900 | for_each_pipe(dev_priv, pipe) { | |
dd740780 | 2901 | for_each_plane(dev_priv, pipe, plane) { |
a269c583 DL |
2902 | val = I915_READ(PLANE_BUF_CFG(pipe, plane)); |
2903 | skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane], | |
2904 | val); | |
2905 | } | |
2906 | ||
2907 | val = I915_READ(CUR_BUF_CFG(pipe)); | |
2908 | skl_ddb_entry_init_from_hw(&ddb->cursor[pipe], val); | |
2909 | } | |
2910 | } | |
2911 | ||
b9cec075 | 2912 | static unsigned int |
2cd601c6 | 2913 | skl_plane_relative_data_rate(const struct intel_plane_wm_parameters *p, int y) |
b9cec075 | 2914 | { |
2cd601c6 CK |
2915 | |
2916 | /* for planar format */ | |
2917 | if (p->y_bytes_per_pixel) { | |
2918 | if (y) /* y-plane data rate */ | |
2919 | return p->horiz_pixels * p->vert_pixels * p->y_bytes_per_pixel; | |
2920 | else /* uv-plane data rate */ | |
2921 | return (p->horiz_pixels/2) * (p->vert_pixels/2) * p->bytes_per_pixel; | |
2922 | } | |
2923 | ||
2924 | /* for packed formats */ | |
b9cec075 DL |
2925 | return p->horiz_pixels * p->vert_pixels * p->bytes_per_pixel; |
2926 | } | |
2927 | ||
2928 | /* | |
2929 | * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching | |
2930 | * a 8192x4096@32bpp framebuffer: | |
2931 | * 3 * 4096 * 8192 * 4 < 2^32 | |
2932 | */ | |
2933 | static unsigned int | |
2934 | skl_get_total_relative_data_rate(struct intel_crtc *intel_crtc, | |
2935 | const struct skl_pipe_wm_parameters *params) | |
2936 | { | |
2937 | unsigned int total_data_rate = 0; | |
2938 | int plane; | |
2939 | ||
2940 | for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) { | |
2941 | const struct intel_plane_wm_parameters *p; | |
2942 | ||
2943 | p = ¶ms->plane[plane]; | |
2944 | if (!p->enabled) | |
2945 | continue; | |
2946 | ||
2cd601c6 CK |
2947 | total_data_rate += skl_plane_relative_data_rate(p, 0); /* packed/uv */ |
2948 | if (p->y_bytes_per_pixel) { | |
2949 | total_data_rate += skl_plane_relative_data_rate(p, 1); /* y-plane */ | |
2950 | } | |
b9cec075 DL |
2951 | } |
2952 | ||
2953 | return total_data_rate; | |
2954 | } | |
2955 | ||
2956 | static void | |
2957 | skl_allocate_pipe_ddb(struct drm_crtc *crtc, | |
2958 | const struct intel_wm_config *config, | |
2959 | const struct skl_pipe_wm_parameters *params, | |
2960 | struct skl_ddb_allocation *ddb /* out */) | |
2961 | { | |
2962 | struct drm_device *dev = crtc->dev; | |
dd740780 | 2963 | struct drm_i915_private *dev_priv = dev->dev_private; |
b9cec075 DL |
2964 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
2965 | enum pipe pipe = intel_crtc->pipe; | |
34bb56af | 2966 | struct skl_ddb_entry *alloc = &ddb->pipe[pipe]; |
b9cec075 | 2967 | uint16_t alloc_size, start, cursor_blocks; |
80958155 | 2968 | uint16_t minimum[I915_MAX_PLANES]; |
2cd601c6 | 2969 | uint16_t y_minimum[I915_MAX_PLANES]; |
b9cec075 DL |
2970 | unsigned int total_data_rate; |
2971 | int plane; | |
2972 | ||
34bb56af DL |
2973 | skl_ddb_get_pipe_allocation_limits(dev, crtc, config, params, alloc); |
2974 | alloc_size = skl_ddb_entry_size(alloc); | |
b9cec075 DL |
2975 | if (alloc_size == 0) { |
2976 | memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe])); | |
2977 | memset(&ddb->cursor[pipe], 0, sizeof(ddb->cursor[pipe])); | |
2978 | return; | |
2979 | } | |
2980 | ||
2981 | cursor_blocks = skl_cursor_allocation(config); | |
34bb56af DL |
2982 | ddb->cursor[pipe].start = alloc->end - cursor_blocks; |
2983 | ddb->cursor[pipe].end = alloc->end; | |
b9cec075 DL |
2984 | |
2985 | alloc_size -= cursor_blocks; | |
34bb56af | 2986 | alloc->end -= cursor_blocks; |
b9cec075 | 2987 | |
80958155 | 2988 | /* 1. Allocate the mininum required blocks for each active plane */ |
dd740780 | 2989 | for_each_plane(dev_priv, pipe, plane) { |
80958155 DL |
2990 | const struct intel_plane_wm_parameters *p; |
2991 | ||
2992 | p = ¶ms->plane[plane]; | |
2993 | if (!p->enabled) | |
2994 | continue; | |
2995 | ||
2996 | minimum[plane] = 8; | |
2997 | alloc_size -= minimum[plane]; | |
2cd601c6 CK |
2998 | y_minimum[plane] = p->y_bytes_per_pixel ? 8 : 0; |
2999 | alloc_size -= y_minimum[plane]; | |
80958155 DL |
3000 | } |
3001 | ||
b9cec075 | 3002 | /* |
80958155 DL |
3003 | * 2. Distribute the remaining space in proportion to the amount of |
3004 | * data each plane needs to fetch from memory. | |
b9cec075 DL |
3005 | * |
3006 | * FIXME: we may not allocate every single block here. | |
3007 | */ | |
3008 | total_data_rate = skl_get_total_relative_data_rate(intel_crtc, params); | |
3009 | ||
34bb56af | 3010 | start = alloc->start; |
b9cec075 DL |
3011 | for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) { |
3012 | const struct intel_plane_wm_parameters *p; | |
2cd601c6 CK |
3013 | unsigned int data_rate, y_data_rate; |
3014 | uint16_t plane_blocks, y_plane_blocks = 0; | |
b9cec075 DL |
3015 | |
3016 | p = ¶ms->plane[plane]; | |
3017 | if (!p->enabled) | |
3018 | continue; | |
3019 | ||
2cd601c6 | 3020 | data_rate = skl_plane_relative_data_rate(p, 0); |
b9cec075 DL |
3021 | |
3022 | /* | |
2cd601c6 | 3023 | * allocation for (packed formats) or (uv-plane part of planar format): |
b9cec075 DL |
3024 | * promote the expression to 64 bits to avoid overflowing, the |
3025 | * result is < available as data_rate / total_data_rate < 1 | |
3026 | */ | |
80958155 DL |
3027 | plane_blocks = minimum[plane]; |
3028 | plane_blocks += div_u64((uint64_t)alloc_size * data_rate, | |
3029 | total_data_rate); | |
b9cec075 DL |
3030 | |
3031 | ddb->plane[pipe][plane].start = start; | |
16160e3d | 3032 | ddb->plane[pipe][plane].end = start + plane_blocks; |
b9cec075 DL |
3033 | |
3034 | start += plane_blocks; | |
2cd601c6 CK |
3035 | |
3036 | /* | |
3037 | * allocation for y_plane part of planar format: | |
3038 | */ | |
3039 | if (p->y_bytes_per_pixel) { | |
3040 | y_data_rate = skl_plane_relative_data_rate(p, 1); | |
3041 | y_plane_blocks = y_minimum[plane]; | |
3042 | y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate, | |
3043 | total_data_rate); | |
3044 | ||
3045 | ddb->y_plane[pipe][plane].start = start; | |
3046 | ddb->y_plane[pipe][plane].end = start + y_plane_blocks; | |
3047 | ||
3048 | start += y_plane_blocks; | |
3049 | } | |
3050 | ||
b9cec075 DL |
3051 | } |
3052 | ||
3053 | } | |
3054 | ||
5cec258b | 3055 | static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state *config) |
2d41c0b5 PB |
3056 | { |
3057 | /* TODO: Take into account the scalers once we support them */ | |
2d112de7 | 3058 | return config->base.adjusted_mode.crtc_clock; |
2d41c0b5 PB |
3059 | } |
3060 | ||
3061 | /* | |
3062 | * The max latency should be 257 (max the punit can code is 255 and we add 2us | |
3063 | * for the read latency) and bytes_per_pixel should always be <= 8, so that | |
3064 | * should allow pixel_rate up to ~2 GHz which seems sufficient since max | |
3065 | * 2xcdclk is 1350 MHz and the pixel rate should never exceed that. | |
3066 | */ | |
3067 | static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, | |
3068 | uint32_t latency) | |
3069 | { | |
3070 | uint32_t wm_intermediate_val, ret; | |
3071 | ||
3072 | if (latency == 0) | |
3073 | return UINT_MAX; | |
3074 | ||
d4c2aa60 | 3075 | wm_intermediate_val = latency * pixel_rate * bytes_per_pixel / 512; |
2d41c0b5 PB |
3076 | ret = DIV_ROUND_UP(wm_intermediate_val, 1000); |
3077 | ||
3078 | return ret; | |
3079 | } | |
3080 | ||
3081 | static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, | |
3082 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, | |
0fda6568 | 3083 | uint64_t tiling, uint32_t latency) |
2d41c0b5 | 3084 | { |
d4c2aa60 TU |
3085 | uint32_t ret; |
3086 | uint32_t plane_bytes_per_line, plane_blocks_per_line; | |
3087 | uint32_t wm_intermediate_val; | |
2d41c0b5 PB |
3088 | |
3089 | if (latency == 0) | |
3090 | return UINT_MAX; | |
3091 | ||
3092 | plane_bytes_per_line = horiz_pixels * bytes_per_pixel; | |
0fda6568 TU |
3093 | |
3094 | if (tiling == I915_FORMAT_MOD_Y_TILED || | |
3095 | tiling == I915_FORMAT_MOD_Yf_TILED) { | |
3096 | plane_bytes_per_line *= 4; | |
3097 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); | |
3098 | plane_blocks_per_line /= 4; | |
3099 | } else { | |
3100 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); | |
3101 | } | |
3102 | ||
2d41c0b5 PB |
3103 | wm_intermediate_val = latency * pixel_rate; |
3104 | ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) * | |
d4c2aa60 | 3105 | plane_blocks_per_line; |
2d41c0b5 PB |
3106 | |
3107 | return ret; | |
3108 | } | |
3109 | ||
2d41c0b5 PB |
3110 | static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation *new_ddb, |
3111 | const struct intel_crtc *intel_crtc) | |
3112 | { | |
3113 | struct drm_device *dev = intel_crtc->base.dev; | |
3114 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3115 | const struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb; | |
3116 | enum pipe pipe = intel_crtc->pipe; | |
3117 | ||
3118 | if (memcmp(new_ddb->plane[pipe], cur_ddb->plane[pipe], | |
3119 | sizeof(new_ddb->plane[pipe]))) | |
3120 | return true; | |
3121 | ||
3122 | if (memcmp(&new_ddb->cursor[pipe], &cur_ddb->cursor[pipe], | |
3123 | sizeof(new_ddb->cursor[pipe]))) | |
3124 | return true; | |
3125 | ||
3126 | return false; | |
3127 | } | |
3128 | ||
3129 | static void skl_compute_wm_global_parameters(struct drm_device *dev, | |
3130 | struct intel_wm_config *config) | |
3131 | { | |
3132 | struct drm_crtc *crtc; | |
3133 | struct drm_plane *plane; | |
3134 | ||
3135 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) | |
3ef00284 | 3136 | config->num_pipes_active += to_intel_crtc(crtc)->active; |
2d41c0b5 PB |
3137 | |
3138 | /* FIXME: I don't think we need those two global parameters on SKL */ | |
3139 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { | |
3140 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
3141 | ||
3142 | config->sprites_enabled |= intel_plane->wm.enabled; | |
3143 | config->sprites_scaled |= intel_plane->wm.scaled; | |
3144 | } | |
3145 | } | |
3146 | ||
3147 | static void skl_compute_wm_pipe_parameters(struct drm_crtc *crtc, | |
3148 | struct skl_pipe_wm_parameters *p) | |
3149 | { | |
3150 | struct drm_device *dev = crtc->dev; | |
3151 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3152 | enum pipe pipe = intel_crtc->pipe; | |
3153 | struct drm_plane *plane; | |
0fda6568 | 3154 | struct drm_framebuffer *fb; |
2d41c0b5 PB |
3155 | int i = 1; /* Index for sprite planes start */ |
3156 | ||
3ef00284 | 3157 | p->active = intel_crtc->active; |
2d41c0b5 | 3158 | if (p->active) { |
6e3c9717 ACO |
3159 | p->pipe_htotal = intel_crtc->config->base.adjusted_mode.crtc_htotal; |
3160 | p->pixel_rate = skl_pipe_pixel_rate(intel_crtc->config); | |
2d41c0b5 | 3161 | |
0fda6568 | 3162 | fb = crtc->primary->state->fb; |
2cd601c6 | 3163 | /* For planar: Bpp is for uv plane, y_Bpp is for y plane */ |
c9f038a1 MR |
3164 | if (fb) { |
3165 | p->plane[0].enabled = true; | |
2cd601c6 CK |
3166 | p->plane[0].bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ? |
3167 | drm_format_plane_cpp(fb->pixel_format, 1) : fb->bits_per_pixel / 8; | |
3168 | p->plane[0].y_bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ? | |
3169 | drm_format_plane_cpp(fb->pixel_format, 0) : 0; | |
0fda6568 | 3170 | p->plane[0].tiling = fb->modifier[0]; |
c9f038a1 MR |
3171 | } else { |
3172 | p->plane[0].enabled = false; | |
3173 | p->plane[0].bytes_per_pixel = 0; | |
2cd601c6 | 3174 | p->plane[0].y_bytes_per_pixel = 0; |
c9f038a1 MR |
3175 | p->plane[0].tiling = DRM_FORMAT_MOD_NONE; |
3176 | } | |
3177 | p->plane[0].horiz_pixels = intel_crtc->config->pipe_src_w; | |
3178 | p->plane[0].vert_pixels = intel_crtc->config->pipe_src_h; | |
1fc0a8f7 | 3179 | p->plane[0].rotation = crtc->primary->state->rotation; |
2d41c0b5 | 3180 | |
c9f038a1 | 3181 | fb = crtc->cursor->state->fb; |
2cd601c6 | 3182 | p->cursor.y_bytes_per_pixel = 0; |
c9f038a1 MR |
3183 | if (fb) { |
3184 | p->cursor.enabled = true; | |
3185 | p->cursor.bytes_per_pixel = fb->bits_per_pixel / 8; | |
3186 | p->cursor.horiz_pixels = crtc->cursor->state->crtc_w; | |
3187 | p->cursor.vert_pixels = crtc->cursor->state->crtc_h; | |
3188 | } else { | |
3189 | p->cursor.enabled = false; | |
3190 | p->cursor.bytes_per_pixel = 0; | |
3191 | p->cursor.horiz_pixels = 64; | |
3192 | p->cursor.vert_pixels = 64; | |
3193 | } | |
2d41c0b5 PB |
3194 | } |
3195 | ||
3196 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { | |
3197 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
3198 | ||
a712f8eb SJ |
3199 | if (intel_plane->pipe == pipe && |
3200 | plane->type == DRM_PLANE_TYPE_OVERLAY) | |
2d41c0b5 PB |
3201 | p->plane[i++] = intel_plane->wm; |
3202 | } | |
3203 | } | |
3204 | ||
d4c2aa60 TU |
3205 | static bool skl_compute_plane_wm(const struct drm_i915_private *dev_priv, |
3206 | struct skl_pipe_wm_parameters *p, | |
afb024aa DL |
3207 | struct intel_plane_wm_parameters *p_params, |
3208 | uint16_t ddb_allocation, | |
d4c2aa60 | 3209 | int level, |
afb024aa DL |
3210 | uint16_t *out_blocks, /* out */ |
3211 | uint8_t *out_lines /* out */) | |
2d41c0b5 | 3212 | { |
d4c2aa60 TU |
3213 | uint32_t latency = dev_priv->wm.skl_latency[level]; |
3214 | uint32_t method1, method2; | |
3215 | uint32_t plane_bytes_per_line, plane_blocks_per_line; | |
3216 | uint32_t res_blocks, res_lines; | |
3217 | uint32_t selected_result; | |
2cd601c6 | 3218 | uint8_t bytes_per_pixel; |
2d41c0b5 | 3219 | |
d4c2aa60 | 3220 | if (latency == 0 || !p->active || !p_params->enabled) |
2d41c0b5 PB |
3221 | return false; |
3222 | ||
2cd601c6 CK |
3223 | bytes_per_pixel = p_params->y_bytes_per_pixel ? |
3224 | p_params->y_bytes_per_pixel : | |
3225 | p_params->bytes_per_pixel; | |
2d41c0b5 | 3226 | method1 = skl_wm_method1(p->pixel_rate, |
2cd601c6 | 3227 | bytes_per_pixel, |
d4c2aa60 | 3228 | latency); |
2d41c0b5 PB |
3229 | method2 = skl_wm_method2(p->pixel_rate, |
3230 | p->pipe_htotal, | |
3231 | p_params->horiz_pixels, | |
2cd601c6 | 3232 | bytes_per_pixel, |
0fda6568 | 3233 | p_params->tiling, |
d4c2aa60 | 3234 | latency); |
2d41c0b5 | 3235 | |
2cd601c6 | 3236 | plane_bytes_per_line = p_params->horiz_pixels * bytes_per_pixel; |
d4c2aa60 | 3237 | plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512); |
2d41c0b5 | 3238 | |
0fda6568 TU |
3239 | if (p_params->tiling == I915_FORMAT_MOD_Y_TILED || |
3240 | p_params->tiling == I915_FORMAT_MOD_Yf_TILED) { | |
1fc0a8f7 TU |
3241 | uint32_t min_scanlines = 4; |
3242 | uint32_t y_tile_minimum; | |
3243 | if (intel_rotation_90_or_270(p_params->rotation)) { | |
3244 | switch (p_params->bytes_per_pixel) { | |
3245 | case 1: | |
3246 | min_scanlines = 16; | |
3247 | break; | |
3248 | case 2: | |
3249 | min_scanlines = 8; | |
3250 | break; | |
3251 | case 8: | |
3252 | WARN(1, "Unsupported pixel depth for rotation"); | |
2f0b5790 | 3253 | } |
1fc0a8f7 TU |
3254 | } |
3255 | y_tile_minimum = plane_blocks_per_line * min_scanlines; | |
0fda6568 TU |
3256 | selected_result = max(method2, y_tile_minimum); |
3257 | } else { | |
3258 | if ((ddb_allocation / plane_blocks_per_line) >= 1) | |
3259 | selected_result = min(method1, method2); | |
3260 | else | |
3261 | selected_result = method1; | |
3262 | } | |
2d41c0b5 | 3263 | |
d4c2aa60 TU |
3264 | res_blocks = selected_result + 1; |
3265 | res_lines = DIV_ROUND_UP(selected_result, plane_blocks_per_line); | |
e6d66171 | 3266 | |
0fda6568 TU |
3267 | if (level >= 1 && level <= 7) { |
3268 | if (p_params->tiling == I915_FORMAT_MOD_Y_TILED || | |
3269 | p_params->tiling == I915_FORMAT_MOD_Yf_TILED) | |
3270 | res_lines += 4; | |
3271 | else | |
3272 | res_blocks++; | |
3273 | } | |
e6d66171 | 3274 | |
d4c2aa60 | 3275 | if (res_blocks >= ddb_allocation || res_lines > 31) |
e6d66171 DL |
3276 | return false; |
3277 | ||
3278 | *out_blocks = res_blocks; | |
3279 | *out_lines = res_lines; | |
2d41c0b5 PB |
3280 | |
3281 | return true; | |
3282 | } | |
3283 | ||
3284 | static void skl_compute_wm_level(const struct drm_i915_private *dev_priv, | |
3285 | struct skl_ddb_allocation *ddb, | |
3286 | struct skl_pipe_wm_parameters *p, | |
3287 | enum pipe pipe, | |
3288 | int level, | |
3289 | int num_planes, | |
3290 | struct skl_wm_level *result) | |
3291 | { | |
2d41c0b5 PB |
3292 | uint16_t ddb_blocks; |
3293 | int i; | |
3294 | ||
3295 | for (i = 0; i < num_planes; i++) { | |
3296 | ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]); | |
3297 | ||
d4c2aa60 TU |
3298 | result->plane_en[i] = skl_compute_plane_wm(dev_priv, |
3299 | p, &p->plane[i], | |
2d41c0b5 | 3300 | ddb_blocks, |
d4c2aa60 | 3301 | level, |
2d41c0b5 PB |
3302 | &result->plane_res_b[i], |
3303 | &result->plane_res_l[i]); | |
3304 | } | |
3305 | ||
3306 | ddb_blocks = skl_ddb_entry_size(&ddb->cursor[pipe]); | |
d4c2aa60 TU |
3307 | result->cursor_en = skl_compute_plane_wm(dev_priv, p, &p->cursor, |
3308 | ddb_blocks, level, | |
3309 | &result->cursor_res_b, | |
2d41c0b5 PB |
3310 | &result->cursor_res_l); |
3311 | } | |
3312 | ||
407b50f3 DL |
3313 | static uint32_t |
3314 | skl_compute_linetime_wm(struct drm_crtc *crtc, struct skl_pipe_wm_parameters *p) | |
3315 | { | |
3ef00284 | 3316 | if (!to_intel_crtc(crtc)->active) |
407b50f3 DL |
3317 | return 0; |
3318 | ||
661abfc0 MK |
3319 | if (WARN_ON(p->pixel_rate == 0)) |
3320 | return 0; | |
407b50f3 | 3321 | |
661abfc0 | 3322 | return DIV_ROUND_UP(8 * p->pipe_htotal * 1000, p->pixel_rate); |
407b50f3 DL |
3323 | } |
3324 | ||
3325 | static void skl_compute_transition_wm(struct drm_crtc *crtc, | |
3326 | struct skl_pipe_wm_parameters *params, | |
9414f563 | 3327 | struct skl_wm_level *trans_wm /* out */) |
407b50f3 | 3328 | { |
9414f563 DL |
3329 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3330 | int i; | |
3331 | ||
407b50f3 DL |
3332 | if (!params->active) |
3333 | return; | |
9414f563 DL |
3334 | |
3335 | /* Until we know more, just disable transition WMs */ | |
3336 | for (i = 0; i < intel_num_planes(intel_crtc); i++) | |
3337 | trans_wm->plane_en[i] = false; | |
3338 | trans_wm->cursor_en = false; | |
407b50f3 DL |
3339 | } |
3340 | ||
2d41c0b5 PB |
3341 | static void skl_compute_pipe_wm(struct drm_crtc *crtc, |
3342 | struct skl_ddb_allocation *ddb, | |
3343 | struct skl_pipe_wm_parameters *params, | |
3344 | struct skl_pipe_wm *pipe_wm) | |
3345 | { | |
3346 | struct drm_device *dev = crtc->dev; | |
3347 | const struct drm_i915_private *dev_priv = dev->dev_private; | |
3348 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3349 | int level, max_level = ilk_wm_max_level(dev); | |
3350 | ||
3351 | for (level = 0; level <= max_level; level++) { | |
3352 | skl_compute_wm_level(dev_priv, ddb, params, intel_crtc->pipe, | |
3353 | level, intel_num_planes(intel_crtc), | |
3354 | &pipe_wm->wm[level]); | |
3355 | } | |
3356 | pipe_wm->linetime = skl_compute_linetime_wm(crtc, params); | |
3357 | ||
9414f563 | 3358 | skl_compute_transition_wm(crtc, params, &pipe_wm->trans_wm); |
2d41c0b5 PB |
3359 | } |
3360 | ||
3361 | static void skl_compute_wm_results(struct drm_device *dev, | |
3362 | struct skl_pipe_wm_parameters *p, | |
3363 | struct skl_pipe_wm *p_wm, | |
3364 | struct skl_wm_values *r, | |
3365 | struct intel_crtc *intel_crtc) | |
3366 | { | |
3367 | int level, max_level = ilk_wm_max_level(dev); | |
3368 | enum pipe pipe = intel_crtc->pipe; | |
9414f563 DL |
3369 | uint32_t temp; |
3370 | int i; | |
2d41c0b5 PB |
3371 | |
3372 | for (level = 0; level <= max_level; level++) { | |
2d41c0b5 PB |
3373 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { |
3374 | temp = 0; | |
2d41c0b5 PB |
3375 | |
3376 | temp |= p_wm->wm[level].plane_res_l[i] << | |
3377 | PLANE_WM_LINES_SHIFT; | |
3378 | temp |= p_wm->wm[level].plane_res_b[i]; | |
3379 | if (p_wm->wm[level].plane_en[i]) | |
3380 | temp |= PLANE_WM_EN; | |
3381 | ||
3382 | r->plane[pipe][i][level] = temp; | |
2d41c0b5 PB |
3383 | } |
3384 | ||
3385 | temp = 0; | |
2d41c0b5 PB |
3386 | |
3387 | temp |= p_wm->wm[level].cursor_res_l << PLANE_WM_LINES_SHIFT; | |
3388 | temp |= p_wm->wm[level].cursor_res_b; | |
3389 | ||
3390 | if (p_wm->wm[level].cursor_en) | |
3391 | temp |= PLANE_WM_EN; | |
3392 | ||
3393 | r->cursor[pipe][level] = temp; | |
2d41c0b5 PB |
3394 | |
3395 | } | |
3396 | ||
9414f563 DL |
3397 | /* transition WMs */ |
3398 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { | |
3399 | temp = 0; | |
3400 | temp |= p_wm->trans_wm.plane_res_l[i] << PLANE_WM_LINES_SHIFT; | |
3401 | temp |= p_wm->trans_wm.plane_res_b[i]; | |
3402 | if (p_wm->trans_wm.plane_en[i]) | |
3403 | temp |= PLANE_WM_EN; | |
3404 | ||
3405 | r->plane_trans[pipe][i] = temp; | |
3406 | } | |
3407 | ||
3408 | temp = 0; | |
3409 | temp |= p_wm->trans_wm.cursor_res_l << PLANE_WM_LINES_SHIFT; | |
3410 | temp |= p_wm->trans_wm.cursor_res_b; | |
3411 | if (p_wm->trans_wm.cursor_en) | |
3412 | temp |= PLANE_WM_EN; | |
3413 | ||
3414 | r->cursor_trans[pipe] = temp; | |
3415 | ||
2d41c0b5 PB |
3416 | r->wm_linetime[pipe] = p_wm->linetime; |
3417 | } | |
3418 | ||
16160e3d DL |
3419 | static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, uint32_t reg, |
3420 | const struct skl_ddb_entry *entry) | |
3421 | { | |
3422 | if (entry->end) | |
3423 | I915_WRITE(reg, (entry->end - 1) << 16 | entry->start); | |
3424 | else | |
3425 | I915_WRITE(reg, 0); | |
3426 | } | |
3427 | ||
2d41c0b5 PB |
3428 | static void skl_write_wm_values(struct drm_i915_private *dev_priv, |
3429 | const struct skl_wm_values *new) | |
3430 | { | |
3431 | struct drm_device *dev = dev_priv->dev; | |
3432 | struct intel_crtc *crtc; | |
3433 | ||
3434 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, base.head) { | |
3435 | int i, level, max_level = ilk_wm_max_level(dev); | |
3436 | enum pipe pipe = crtc->pipe; | |
3437 | ||
5d374d96 DL |
3438 | if (!new->dirty[pipe]) |
3439 | continue; | |
8211bd5b | 3440 | |
5d374d96 | 3441 | I915_WRITE(PIPE_WM_LINETIME(pipe), new->wm_linetime[pipe]); |
8211bd5b | 3442 | |
5d374d96 DL |
3443 | for (level = 0; level <= max_level; level++) { |
3444 | for (i = 0; i < intel_num_planes(crtc); i++) | |
3445 | I915_WRITE(PLANE_WM(pipe, i, level), | |
3446 | new->plane[pipe][i][level]); | |
3447 | I915_WRITE(CUR_WM(pipe, level), | |
3448 | new->cursor[pipe][level]); | |
2d41c0b5 | 3449 | } |
5d374d96 DL |
3450 | for (i = 0; i < intel_num_planes(crtc); i++) |
3451 | I915_WRITE(PLANE_WM_TRANS(pipe, i), | |
3452 | new->plane_trans[pipe][i]); | |
3453 | I915_WRITE(CUR_WM_TRANS(pipe), new->cursor_trans[pipe]); | |
3454 | ||
2cd601c6 | 3455 | for (i = 0; i < intel_num_planes(crtc); i++) { |
5d374d96 DL |
3456 | skl_ddb_entry_write(dev_priv, |
3457 | PLANE_BUF_CFG(pipe, i), | |
3458 | &new->ddb.plane[pipe][i]); | |
2cd601c6 CK |
3459 | skl_ddb_entry_write(dev_priv, |
3460 | PLANE_NV12_BUF_CFG(pipe, i), | |
3461 | &new->ddb.y_plane[pipe][i]); | |
3462 | } | |
5d374d96 DL |
3463 | |
3464 | skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe), | |
3465 | &new->ddb.cursor[pipe]); | |
2d41c0b5 | 3466 | } |
2d41c0b5 PB |
3467 | } |
3468 | ||
0e8fb7ba DL |
3469 | /* |
3470 | * When setting up a new DDB allocation arrangement, we need to correctly | |
3471 | * sequence the times at which the new allocations for the pipes are taken into | |
3472 | * account or we'll have pipes fetching from space previously allocated to | |
3473 | * another pipe. | |
3474 | * | |
3475 | * Roughly the sequence looks like: | |
3476 | * 1. re-allocate the pipe(s) with the allocation being reduced and not | |
3477 | * overlapping with a previous light-up pipe (another way to put it is: | |
3478 | * pipes with their new allocation strickly included into their old ones). | |
3479 | * 2. re-allocate the other pipes that get their allocation reduced | |
3480 | * 3. allocate the pipes having their allocation increased | |
3481 | * | |
3482 | * Steps 1. and 2. are here to take care of the following case: | |
3483 | * - Initially DDB looks like this: | |
3484 | * | B | C | | |
3485 | * - enable pipe A. | |
3486 | * - pipe B has a reduced DDB allocation that overlaps with the old pipe C | |
3487 | * allocation | |
3488 | * | A | B | C | | |
3489 | * | |
3490 | * We need to sequence the re-allocation: C, B, A (and not B, C, A). | |
3491 | */ | |
3492 | ||
d21b795c DL |
3493 | static void |
3494 | skl_wm_flush_pipe(struct drm_i915_private *dev_priv, enum pipe pipe, int pass) | |
0e8fb7ba | 3495 | { |
0e8fb7ba DL |
3496 | int plane; |
3497 | ||
d21b795c DL |
3498 | DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe), pass); |
3499 | ||
dd740780 | 3500 | for_each_plane(dev_priv, pipe, plane) { |
0e8fb7ba DL |
3501 | I915_WRITE(PLANE_SURF(pipe, plane), |
3502 | I915_READ(PLANE_SURF(pipe, plane))); | |
3503 | } | |
3504 | I915_WRITE(CURBASE(pipe), I915_READ(CURBASE(pipe))); | |
3505 | } | |
3506 | ||
3507 | static bool | |
3508 | skl_ddb_allocation_included(const struct skl_ddb_allocation *old, | |
3509 | const struct skl_ddb_allocation *new, | |
3510 | enum pipe pipe) | |
3511 | { | |
3512 | uint16_t old_size, new_size; | |
3513 | ||
3514 | old_size = skl_ddb_entry_size(&old->pipe[pipe]); | |
3515 | new_size = skl_ddb_entry_size(&new->pipe[pipe]); | |
3516 | ||
3517 | return old_size != new_size && | |
3518 | new->pipe[pipe].start >= old->pipe[pipe].start && | |
3519 | new->pipe[pipe].end <= old->pipe[pipe].end; | |
3520 | } | |
3521 | ||
3522 | static void skl_flush_wm_values(struct drm_i915_private *dev_priv, | |
3523 | struct skl_wm_values *new_values) | |
3524 | { | |
3525 | struct drm_device *dev = dev_priv->dev; | |
3526 | struct skl_ddb_allocation *cur_ddb, *new_ddb; | |
c929cb45 | 3527 | bool reallocated[I915_MAX_PIPES] = {}; |
0e8fb7ba DL |
3528 | struct intel_crtc *crtc; |
3529 | enum pipe pipe; | |
3530 | ||
3531 | new_ddb = &new_values->ddb; | |
3532 | cur_ddb = &dev_priv->wm.skl_hw.ddb; | |
3533 | ||
3534 | /* | |
3535 | * First pass: flush the pipes with the new allocation contained into | |
3536 | * the old space. | |
3537 | * | |
3538 | * We'll wait for the vblank on those pipes to ensure we can safely | |
3539 | * re-allocate the freed space without this pipe fetching from it. | |
3540 | */ | |
3541 | for_each_intel_crtc(dev, crtc) { | |
3542 | if (!crtc->active) | |
3543 | continue; | |
3544 | ||
3545 | pipe = crtc->pipe; | |
3546 | ||
3547 | if (!skl_ddb_allocation_included(cur_ddb, new_ddb, pipe)) | |
3548 | continue; | |
3549 | ||
d21b795c | 3550 | skl_wm_flush_pipe(dev_priv, pipe, 1); |
0e8fb7ba DL |
3551 | intel_wait_for_vblank(dev, pipe); |
3552 | ||
3553 | reallocated[pipe] = true; | |
3554 | } | |
3555 | ||
3556 | ||
3557 | /* | |
3558 | * Second pass: flush the pipes that are having their allocation | |
3559 | * reduced, but overlapping with a previous allocation. | |
3560 | * | |
3561 | * Here as well we need to wait for the vblank to make sure the freed | |
3562 | * space is not used anymore. | |
3563 | */ | |
3564 | for_each_intel_crtc(dev, crtc) { | |
3565 | if (!crtc->active) | |
3566 | continue; | |
3567 | ||
3568 | pipe = crtc->pipe; | |
3569 | ||
3570 | if (reallocated[pipe]) | |
3571 | continue; | |
3572 | ||
3573 | if (skl_ddb_entry_size(&new_ddb->pipe[pipe]) < | |
3574 | skl_ddb_entry_size(&cur_ddb->pipe[pipe])) { | |
d21b795c | 3575 | skl_wm_flush_pipe(dev_priv, pipe, 2); |
0e8fb7ba | 3576 | intel_wait_for_vblank(dev, pipe); |
d9d8e6b3 | 3577 | reallocated[pipe] = true; |
0e8fb7ba | 3578 | } |
0e8fb7ba DL |
3579 | } |
3580 | ||
3581 | /* | |
3582 | * Third pass: flush the pipes that got more space allocated. | |
3583 | * | |
3584 | * We don't need to actively wait for the update here, next vblank | |
3585 | * will just get more DDB space with the correct WM values. | |
3586 | */ | |
3587 | for_each_intel_crtc(dev, crtc) { | |
3588 | if (!crtc->active) | |
3589 | continue; | |
3590 | ||
3591 | pipe = crtc->pipe; | |
3592 | ||
3593 | /* | |
3594 | * At this point, only the pipes more space than before are | |
3595 | * left to re-allocate. | |
3596 | */ | |
3597 | if (reallocated[pipe]) | |
3598 | continue; | |
3599 | ||
d21b795c | 3600 | skl_wm_flush_pipe(dev_priv, pipe, 3); |
0e8fb7ba DL |
3601 | } |
3602 | } | |
3603 | ||
2d41c0b5 PB |
3604 | static bool skl_update_pipe_wm(struct drm_crtc *crtc, |
3605 | struct skl_pipe_wm_parameters *params, | |
3606 | struct intel_wm_config *config, | |
3607 | struct skl_ddb_allocation *ddb, /* out */ | |
3608 | struct skl_pipe_wm *pipe_wm /* out */) | |
3609 | { | |
3610 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3611 | ||
3612 | skl_compute_wm_pipe_parameters(crtc, params); | |
b9cec075 | 3613 | skl_allocate_pipe_ddb(crtc, config, params, ddb); |
2d41c0b5 PB |
3614 | skl_compute_pipe_wm(crtc, ddb, params, pipe_wm); |
3615 | ||
3616 | if (!memcmp(&intel_crtc->wm.skl_active, pipe_wm, sizeof(*pipe_wm))) | |
3617 | return false; | |
3618 | ||
3619 | intel_crtc->wm.skl_active = *pipe_wm; | |
2cd601c6 | 3620 | |
2d41c0b5 PB |
3621 | return true; |
3622 | } | |
3623 | ||
3624 | static void skl_update_other_pipe_wm(struct drm_device *dev, | |
3625 | struct drm_crtc *crtc, | |
3626 | struct intel_wm_config *config, | |
3627 | struct skl_wm_values *r) | |
3628 | { | |
3629 | struct intel_crtc *intel_crtc; | |
3630 | struct intel_crtc *this_crtc = to_intel_crtc(crtc); | |
3631 | ||
3632 | /* | |
3633 | * If the WM update hasn't changed the allocation for this_crtc (the | |
3634 | * crtc we are currently computing the new WM values for), other | |
3635 | * enabled crtcs will keep the same allocation and we don't need to | |
3636 | * recompute anything for them. | |
3637 | */ | |
3638 | if (!skl_ddb_allocation_changed(&r->ddb, this_crtc)) | |
3639 | return; | |
3640 | ||
3641 | /* | |
3642 | * Otherwise, because of this_crtc being freshly enabled/disabled, the | |
3643 | * other active pipes need new DDB allocation and WM values. | |
3644 | */ | |
3645 | list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, | |
3646 | base.head) { | |
3647 | struct skl_pipe_wm_parameters params = {}; | |
3648 | struct skl_pipe_wm pipe_wm = {}; | |
3649 | bool wm_changed; | |
3650 | ||
3651 | if (this_crtc->pipe == intel_crtc->pipe) | |
3652 | continue; | |
3653 | ||
3654 | if (!intel_crtc->active) | |
3655 | continue; | |
3656 | ||
3657 | wm_changed = skl_update_pipe_wm(&intel_crtc->base, | |
3658 | ¶ms, config, | |
3659 | &r->ddb, &pipe_wm); | |
3660 | ||
3661 | /* | |
3662 | * If we end up re-computing the other pipe WM values, it's | |
3663 | * because it was really needed, so we expect the WM values to | |
3664 | * be different. | |
3665 | */ | |
3666 | WARN_ON(!wm_changed); | |
3667 | ||
3668 | skl_compute_wm_results(dev, ¶ms, &pipe_wm, r, intel_crtc); | |
3669 | r->dirty[intel_crtc->pipe] = true; | |
3670 | } | |
3671 | } | |
3672 | ||
3673 | static void skl_update_wm(struct drm_crtc *crtc) | |
3674 | { | |
3675 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3676 | struct drm_device *dev = crtc->dev; | |
3677 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3678 | struct skl_pipe_wm_parameters params = {}; | |
3679 | struct skl_wm_values *results = &dev_priv->wm.skl_results; | |
3680 | struct skl_pipe_wm pipe_wm = {}; | |
3681 | struct intel_wm_config config = {}; | |
3682 | ||
3683 | memset(results, 0, sizeof(*results)); | |
3684 | ||
3685 | skl_compute_wm_global_parameters(dev, &config); | |
3686 | ||
3687 | if (!skl_update_pipe_wm(crtc, ¶ms, &config, | |
3688 | &results->ddb, &pipe_wm)) | |
3689 | return; | |
3690 | ||
3691 | skl_compute_wm_results(dev, ¶ms, &pipe_wm, results, intel_crtc); | |
3692 | results->dirty[intel_crtc->pipe] = true; | |
3693 | ||
3694 | skl_update_other_pipe_wm(dev, crtc, &config, results); | |
3695 | skl_write_wm_values(dev_priv, results); | |
0e8fb7ba | 3696 | skl_flush_wm_values(dev_priv, results); |
53b0deb4 DL |
3697 | |
3698 | /* store the new configuration */ | |
3699 | dev_priv->wm.skl_hw = *results; | |
2d41c0b5 PB |
3700 | } |
3701 | ||
3702 | static void | |
3703 | skl_update_sprite_wm(struct drm_plane *plane, struct drm_crtc *crtc, | |
3704 | uint32_t sprite_width, uint32_t sprite_height, | |
3705 | int pixel_size, bool enabled, bool scaled) | |
3706 | { | |
3707 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
0fda6568 | 3708 | struct drm_framebuffer *fb = plane->state->fb; |
2d41c0b5 PB |
3709 | |
3710 | intel_plane->wm.enabled = enabled; | |
3711 | intel_plane->wm.scaled = scaled; | |
3712 | intel_plane->wm.horiz_pixels = sprite_width; | |
3713 | intel_plane->wm.vert_pixels = sprite_height; | |
0fda6568 | 3714 | intel_plane->wm.tiling = DRM_FORMAT_MOD_NONE; |
2cd601c6 CK |
3715 | |
3716 | /* For planar: Bpp is for UV plane, y_Bpp is for Y plane */ | |
3717 | intel_plane->wm.bytes_per_pixel = | |
3718 | (fb && fb->pixel_format == DRM_FORMAT_NV12) ? | |
3719 | drm_format_plane_cpp(plane->state->fb->pixel_format, 1) : pixel_size; | |
3720 | intel_plane->wm.y_bytes_per_pixel = | |
3721 | (fb && fb->pixel_format == DRM_FORMAT_NV12) ? | |
3722 | drm_format_plane_cpp(plane->state->fb->pixel_format, 0) : 0; | |
3723 | ||
0fda6568 TU |
3724 | /* |
3725 | * Framebuffer can be NULL on plane disable, but it does not | |
3726 | * matter for watermarks if we assume no tiling in that case. | |
3727 | */ | |
3728 | if (fb) | |
3729 | intel_plane->wm.tiling = fb->modifier[0]; | |
1fc0a8f7 | 3730 | intel_plane->wm.rotation = plane->state->rotation; |
2d41c0b5 PB |
3731 | |
3732 | skl_update_wm(crtc); | |
3733 | } | |
3734 | ||
820c1980 | 3735 | static void ilk_update_wm(struct drm_crtc *crtc) |
801bcfff | 3736 | { |
7c4a395f | 3737 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
46ba614c | 3738 | struct drm_device *dev = crtc->dev; |
801bcfff | 3739 | struct drm_i915_private *dev_priv = dev->dev_private; |
820c1980 ID |
3740 | struct ilk_wm_maximums max; |
3741 | struct ilk_pipe_wm_parameters params = {}; | |
3742 | struct ilk_wm_values results = {}; | |
77c122bc | 3743 | enum intel_ddb_partitioning partitioning; |
7c4a395f | 3744 | struct intel_pipe_wm pipe_wm = {}; |
198a1e9b | 3745 | struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm; |
a485bfb8 | 3746 | struct intel_wm_config config = {}; |
7c4a395f | 3747 | |
2a44b76b | 3748 | ilk_compute_wm_parameters(crtc, ¶ms); |
7c4a395f VS |
3749 | |
3750 | intel_compute_pipe_wm(crtc, ¶ms, &pipe_wm); | |
3751 | ||
3752 | if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm))) | |
3753 | return; | |
861f3389 | 3754 | |
7c4a395f | 3755 | intel_crtc->wm.active = pipe_wm; |
861f3389 | 3756 | |
2a44b76b VS |
3757 | ilk_compute_wm_config(dev, &config); |
3758 | ||
34982fe1 | 3759 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max); |
0ba22e26 | 3760 | ilk_wm_merge(dev, &config, &max, &lp_wm_1_2); |
a485bfb8 VS |
3761 | |
3762 | /* 5/6 split only in single pipe config on IVB+ */ | |
ec98c8d1 VS |
3763 | if (INTEL_INFO(dev)->gen >= 7 && |
3764 | config.num_pipes_active == 1 && config.sprites_enabled) { | |
34982fe1 | 3765 | ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max); |
0ba22e26 | 3766 | ilk_wm_merge(dev, &config, &max, &lp_wm_5_6); |
0362c781 | 3767 | |
820c1980 | 3768 | best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6); |
861f3389 | 3769 | } else { |
198a1e9b | 3770 | best_lp_wm = &lp_wm_1_2; |
861f3389 PZ |
3771 | } |
3772 | ||
198a1e9b | 3773 | partitioning = (best_lp_wm == &lp_wm_1_2) ? |
77c122bc | 3774 | INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6; |
801bcfff | 3775 | |
820c1980 | 3776 | ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results); |
609cedef | 3777 | |
820c1980 | 3778 | ilk_write_wm_values(dev_priv, &results); |
1011d8c4 PZ |
3779 | } |
3780 | ||
ed57cb8a DL |
3781 | static void |
3782 | ilk_update_sprite_wm(struct drm_plane *plane, | |
3783 | struct drm_crtc *crtc, | |
3784 | uint32_t sprite_width, uint32_t sprite_height, | |
3785 | int pixel_size, bool enabled, bool scaled) | |
526682e9 | 3786 | { |
8553c18e | 3787 | struct drm_device *dev = plane->dev; |
adf3d35e | 3788 | struct intel_plane *intel_plane = to_intel_plane(plane); |
526682e9 | 3789 | |
adf3d35e VS |
3790 | intel_plane->wm.enabled = enabled; |
3791 | intel_plane->wm.scaled = scaled; | |
3792 | intel_plane->wm.horiz_pixels = sprite_width; | |
ed57cb8a | 3793 | intel_plane->wm.vert_pixels = sprite_width; |
adf3d35e | 3794 | intel_plane->wm.bytes_per_pixel = pixel_size; |
526682e9 | 3795 | |
8553c18e VS |
3796 | /* |
3797 | * IVB workaround: must disable low power watermarks for at least | |
3798 | * one frame before enabling scaling. LP watermarks can be re-enabled | |
3799 | * when scaling is disabled. | |
3800 | * | |
3801 | * WaCxSRDisabledForSpriteScaling:ivb | |
3802 | */ | |
3803 | if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev)) | |
3804 | intel_wait_for_vblank(dev, intel_plane->pipe); | |
3805 | ||
820c1980 | 3806 | ilk_update_wm(crtc); |
526682e9 PZ |
3807 | } |
3808 | ||
3078999f PB |
3809 | static void skl_pipe_wm_active_state(uint32_t val, |
3810 | struct skl_pipe_wm *active, | |
3811 | bool is_transwm, | |
3812 | bool is_cursor, | |
3813 | int i, | |
3814 | int level) | |
3815 | { | |
3816 | bool is_enabled = (val & PLANE_WM_EN) != 0; | |
3817 | ||
3818 | if (!is_transwm) { | |
3819 | if (!is_cursor) { | |
3820 | active->wm[level].plane_en[i] = is_enabled; | |
3821 | active->wm[level].plane_res_b[i] = | |
3822 | val & PLANE_WM_BLOCKS_MASK; | |
3823 | active->wm[level].plane_res_l[i] = | |
3824 | (val >> PLANE_WM_LINES_SHIFT) & | |
3825 | PLANE_WM_LINES_MASK; | |
3826 | } else { | |
3827 | active->wm[level].cursor_en = is_enabled; | |
3828 | active->wm[level].cursor_res_b = | |
3829 | val & PLANE_WM_BLOCKS_MASK; | |
3830 | active->wm[level].cursor_res_l = | |
3831 | (val >> PLANE_WM_LINES_SHIFT) & | |
3832 | PLANE_WM_LINES_MASK; | |
3833 | } | |
3834 | } else { | |
3835 | if (!is_cursor) { | |
3836 | active->trans_wm.plane_en[i] = is_enabled; | |
3837 | active->trans_wm.plane_res_b[i] = | |
3838 | val & PLANE_WM_BLOCKS_MASK; | |
3839 | active->trans_wm.plane_res_l[i] = | |
3840 | (val >> PLANE_WM_LINES_SHIFT) & | |
3841 | PLANE_WM_LINES_MASK; | |
3842 | } else { | |
3843 | active->trans_wm.cursor_en = is_enabled; | |
3844 | active->trans_wm.cursor_res_b = | |
3845 | val & PLANE_WM_BLOCKS_MASK; | |
3846 | active->trans_wm.cursor_res_l = | |
3847 | (val >> PLANE_WM_LINES_SHIFT) & | |
3848 | PLANE_WM_LINES_MASK; | |
3849 | } | |
3850 | } | |
3851 | } | |
3852 | ||
3853 | static void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc) | |
3854 | { | |
3855 | struct drm_device *dev = crtc->dev; | |
3856 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3857 | struct skl_wm_values *hw = &dev_priv->wm.skl_hw; | |
3858 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
3859 | struct skl_pipe_wm *active = &intel_crtc->wm.skl_active; | |
3860 | enum pipe pipe = intel_crtc->pipe; | |
3861 | int level, i, max_level; | |
3862 | uint32_t temp; | |
3863 | ||
3864 | max_level = ilk_wm_max_level(dev); | |
3865 | ||
3866 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); | |
3867 | ||
3868 | for (level = 0; level <= max_level; level++) { | |
3869 | for (i = 0; i < intel_num_planes(intel_crtc); i++) | |
3870 | hw->plane[pipe][i][level] = | |
3871 | I915_READ(PLANE_WM(pipe, i, level)); | |
3872 | hw->cursor[pipe][level] = I915_READ(CUR_WM(pipe, level)); | |
3873 | } | |
3874 | ||
3875 | for (i = 0; i < intel_num_planes(intel_crtc); i++) | |
3876 | hw->plane_trans[pipe][i] = I915_READ(PLANE_WM_TRANS(pipe, i)); | |
3877 | hw->cursor_trans[pipe] = I915_READ(CUR_WM_TRANS(pipe)); | |
3878 | ||
3ef00284 | 3879 | if (!intel_crtc->active) |
3078999f PB |
3880 | return; |
3881 | ||
3882 | hw->dirty[pipe] = true; | |
3883 | ||
3884 | active->linetime = hw->wm_linetime[pipe]; | |
3885 | ||
3886 | for (level = 0; level <= max_level; level++) { | |
3887 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { | |
3888 | temp = hw->plane[pipe][i][level]; | |
3889 | skl_pipe_wm_active_state(temp, active, false, | |
3890 | false, i, level); | |
3891 | } | |
3892 | temp = hw->cursor[pipe][level]; | |
3893 | skl_pipe_wm_active_state(temp, active, false, true, i, level); | |
3894 | } | |
3895 | ||
3896 | for (i = 0; i < intel_num_planes(intel_crtc); i++) { | |
3897 | temp = hw->plane_trans[pipe][i]; | |
3898 | skl_pipe_wm_active_state(temp, active, true, false, i, 0); | |
3899 | } | |
3900 | ||
3901 | temp = hw->cursor_trans[pipe]; | |
3902 | skl_pipe_wm_active_state(temp, active, true, true, i, 0); | |
3903 | } | |
3904 | ||
3905 | void skl_wm_get_hw_state(struct drm_device *dev) | |
3906 | { | |
a269c583 DL |
3907 | struct drm_i915_private *dev_priv = dev->dev_private; |
3908 | struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb; | |
3078999f PB |
3909 | struct drm_crtc *crtc; |
3910 | ||
a269c583 | 3911 | skl_ddb_get_hw_state(dev_priv, ddb); |
3078999f PB |
3912 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) |
3913 | skl_pipe_wm_get_hw_state(crtc); | |
3914 | } | |
3915 | ||
243e6a44 VS |
3916 | static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc) |
3917 | { | |
3918 | struct drm_device *dev = crtc->dev; | |
3919 | struct drm_i915_private *dev_priv = dev->dev_private; | |
820c1980 | 3920 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
243e6a44 VS |
3921 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
3922 | struct intel_pipe_wm *active = &intel_crtc->wm.active; | |
3923 | enum pipe pipe = intel_crtc->pipe; | |
3924 | static const unsigned int wm0_pipe_reg[] = { | |
3925 | [PIPE_A] = WM0_PIPEA_ILK, | |
3926 | [PIPE_B] = WM0_PIPEB_ILK, | |
3927 | [PIPE_C] = WM0_PIPEC_IVB, | |
3928 | }; | |
3929 | ||
3930 | hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]); | |
a42a5719 | 3931 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
ce0e0713 | 3932 | hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe)); |
243e6a44 | 3933 | |
3ef00284 | 3934 | active->pipe_enabled = intel_crtc->active; |
2a44b76b VS |
3935 | |
3936 | if (active->pipe_enabled) { | |
243e6a44 VS |
3937 | u32 tmp = hw->wm_pipe[pipe]; |
3938 | ||
3939 | /* | |
3940 | * For active pipes LP0 watermark is marked as | |
3941 | * enabled, and LP1+ watermaks as disabled since | |
3942 | * we can't really reverse compute them in case | |
3943 | * multiple pipes are active. | |
3944 | */ | |
3945 | active->wm[0].enable = true; | |
3946 | active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT; | |
3947 | active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT; | |
3948 | active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK; | |
3949 | active->linetime = hw->wm_linetime[pipe]; | |
3950 | } else { | |
3951 | int level, max_level = ilk_wm_max_level(dev); | |
3952 | ||
3953 | /* | |
3954 | * For inactive pipes, all watermark levels | |
3955 | * should be marked as enabled but zeroed, | |
3956 | * which is what we'd compute them to. | |
3957 | */ | |
3958 | for (level = 0; level <= max_level; level++) | |
3959 | active->wm[level].enable = true; | |
3960 | } | |
3961 | } | |
3962 | ||
6eb1a681 VS |
3963 | #define _FW_WM(value, plane) \ |
3964 | (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT) | |
3965 | #define _FW_WM_VLV(value, plane) \ | |
3966 | (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT) | |
3967 | ||
3968 | static void vlv_read_wm_values(struct drm_i915_private *dev_priv, | |
3969 | struct vlv_wm_values *wm) | |
3970 | { | |
3971 | enum pipe pipe; | |
3972 | uint32_t tmp; | |
3973 | ||
3974 | for_each_pipe(dev_priv, pipe) { | |
3975 | tmp = I915_READ(VLV_DDL(pipe)); | |
3976 | ||
3977 | wm->ddl[pipe].primary = | |
3978 | (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
3979 | wm->ddl[pipe].cursor = | |
3980 | (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
3981 | wm->ddl[pipe].sprite[0] = | |
3982 | (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
3983 | wm->ddl[pipe].sprite[1] = | |
3984 | (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK); | |
3985 | } | |
3986 | ||
3987 | tmp = I915_READ(DSPFW1); | |
3988 | wm->sr.plane = _FW_WM(tmp, SR); | |
3989 | wm->pipe[PIPE_B].cursor = _FW_WM(tmp, CURSORB); | |
3990 | wm->pipe[PIPE_B].primary = _FW_WM_VLV(tmp, PLANEB); | |
3991 | wm->pipe[PIPE_A].primary = _FW_WM_VLV(tmp, PLANEA); | |
3992 | ||
3993 | tmp = I915_READ(DSPFW2); | |
3994 | wm->pipe[PIPE_A].sprite[1] = _FW_WM_VLV(tmp, SPRITEB); | |
3995 | wm->pipe[PIPE_A].cursor = _FW_WM(tmp, CURSORA); | |
3996 | wm->pipe[PIPE_A].sprite[0] = _FW_WM_VLV(tmp, SPRITEA); | |
3997 | ||
3998 | tmp = I915_READ(DSPFW3); | |
3999 | wm->sr.cursor = _FW_WM(tmp, CURSOR_SR); | |
4000 | ||
4001 | if (IS_CHERRYVIEW(dev_priv)) { | |
4002 | tmp = I915_READ(DSPFW7_CHV); | |
4003 | wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED); | |
4004 | wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC); | |
4005 | ||
4006 | tmp = I915_READ(DSPFW8_CHV); | |
4007 | wm->pipe[PIPE_C].sprite[1] = _FW_WM_VLV(tmp, SPRITEF); | |
4008 | wm->pipe[PIPE_C].sprite[0] = _FW_WM_VLV(tmp, SPRITEE); | |
4009 | ||
4010 | tmp = I915_READ(DSPFW9_CHV); | |
4011 | wm->pipe[PIPE_C].primary = _FW_WM_VLV(tmp, PLANEC); | |
4012 | wm->pipe[PIPE_C].cursor = _FW_WM(tmp, CURSORC); | |
4013 | ||
4014 | tmp = I915_READ(DSPHOWM); | |
4015 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; | |
4016 | wm->pipe[PIPE_C].sprite[1] |= _FW_WM(tmp, SPRITEF_HI) << 8; | |
4017 | wm->pipe[PIPE_C].sprite[0] |= _FW_WM(tmp, SPRITEE_HI) << 8; | |
4018 | wm->pipe[PIPE_C].primary |= _FW_WM(tmp, PLANEC_HI) << 8; | |
4019 | wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8; | |
4020 | wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8; | |
4021 | wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8; | |
4022 | wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8; | |
4023 | wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8; | |
4024 | wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8; | |
4025 | } else { | |
4026 | tmp = I915_READ(DSPFW7); | |
4027 | wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED); | |
4028 | wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC); | |
4029 | ||
4030 | tmp = I915_READ(DSPHOWM); | |
4031 | wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9; | |
4032 | wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8; | |
4033 | wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8; | |
4034 | wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8; | |
4035 | wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8; | |
4036 | wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8; | |
4037 | wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8; | |
4038 | } | |
4039 | } | |
4040 | ||
4041 | #undef _FW_WM | |
4042 | #undef _FW_WM_VLV | |
4043 | ||
4044 | void vlv_wm_get_hw_state(struct drm_device *dev) | |
4045 | { | |
4046 | struct drm_i915_private *dev_priv = to_i915(dev); | |
4047 | struct vlv_wm_values *wm = &dev_priv->wm.vlv; | |
4048 | struct intel_plane *plane; | |
4049 | enum pipe pipe; | |
4050 | u32 val; | |
4051 | ||
4052 | vlv_read_wm_values(dev_priv, wm); | |
4053 | ||
4054 | for_each_intel_plane(dev, plane) { | |
4055 | switch (plane->base.type) { | |
4056 | int sprite; | |
4057 | case DRM_PLANE_TYPE_CURSOR: | |
4058 | plane->wm.fifo_size = 63; | |
4059 | break; | |
4060 | case DRM_PLANE_TYPE_PRIMARY: | |
4061 | plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, 0); | |
4062 | break; | |
4063 | case DRM_PLANE_TYPE_OVERLAY: | |
4064 | sprite = plane->plane; | |
4065 | plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, sprite + 1); | |
4066 | break; | |
4067 | } | |
4068 | } | |
4069 | ||
4070 | wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN; | |
4071 | wm->level = VLV_WM_LEVEL_PM2; | |
4072 | ||
4073 | if (IS_CHERRYVIEW(dev_priv)) { | |
4074 | mutex_lock(&dev_priv->rps.hw_lock); | |
4075 | ||
4076 | val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ); | |
4077 | if (val & DSP_MAXFIFO_PM5_ENABLE) | |
4078 | wm->level = VLV_WM_LEVEL_PM5; | |
4079 | ||
4080 | val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2); | |
4081 | if ((val & FORCE_DDR_HIGH_FREQ) == 0) | |
4082 | wm->level = VLV_WM_LEVEL_DDR_DVFS; | |
4083 | ||
4084 | mutex_unlock(&dev_priv->rps.hw_lock); | |
4085 | } | |
4086 | ||
4087 | for_each_pipe(dev_priv, pipe) | |
4088 | DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n", | |
4089 | pipe_name(pipe), wm->pipe[pipe].primary, wm->pipe[pipe].cursor, | |
4090 | wm->pipe[pipe].sprite[0], wm->pipe[pipe].sprite[1]); | |
4091 | ||
4092 | DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n", | |
4093 | wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr); | |
4094 | } | |
4095 | ||
243e6a44 VS |
4096 | void ilk_wm_get_hw_state(struct drm_device *dev) |
4097 | { | |
4098 | struct drm_i915_private *dev_priv = dev->dev_private; | |
820c1980 | 4099 | struct ilk_wm_values *hw = &dev_priv->wm.hw; |
243e6a44 VS |
4100 | struct drm_crtc *crtc; |
4101 | ||
70e1e0ec | 4102 | for_each_crtc(dev, crtc) |
243e6a44 VS |
4103 | ilk_pipe_wm_get_hw_state(crtc); |
4104 | ||
4105 | hw->wm_lp[0] = I915_READ(WM1_LP_ILK); | |
4106 | hw->wm_lp[1] = I915_READ(WM2_LP_ILK); | |
4107 | hw->wm_lp[2] = I915_READ(WM3_LP_ILK); | |
4108 | ||
4109 | hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK); | |
cfa7698b VS |
4110 | if (INTEL_INFO(dev)->gen >= 7) { |
4111 | hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB); | |
4112 | hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB); | |
4113 | } | |
243e6a44 | 4114 | |
a42a5719 | 4115 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) |
ac9545fd VS |
4116 | hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ? |
4117 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; | |
4118 | else if (IS_IVYBRIDGE(dev)) | |
4119 | hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ? | |
4120 | INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2; | |
243e6a44 VS |
4121 | |
4122 | hw->enable_fbc_wm = | |
4123 | !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS); | |
4124 | } | |
4125 | ||
b445e3b0 ED |
4126 | /** |
4127 | * intel_update_watermarks - update FIFO watermark values based on current modes | |
4128 | * | |
4129 | * Calculate watermark values for the various WM regs based on current mode | |
4130 | * and plane configuration. | |
4131 | * | |
4132 | * There are several cases to deal with here: | |
4133 | * - normal (i.e. non-self-refresh) | |
4134 | * - self-refresh (SR) mode | |
4135 | * - lines are large relative to FIFO size (buffer can hold up to 2) | |
4136 | * - lines are small relative to FIFO size (buffer can hold more than 2 | |
4137 | * lines), so need to account for TLB latency | |
4138 | * | |
4139 | * The normal calculation is: | |
4140 | * watermark = dotclock * bytes per pixel * latency | |
4141 | * where latency is platform & configuration dependent (we assume pessimal | |
4142 | * values here). | |
4143 | * | |
4144 | * The SR calculation is: | |
4145 | * watermark = (trunc(latency/line time)+1) * surface width * | |
4146 | * bytes per pixel | |
4147 | * where | |
4148 | * line time = htotal / dotclock | |
4149 | * surface width = hdisplay for normal plane and 64 for cursor | |
4150 | * and latency is assumed to be high, as above. | |
4151 | * | |
4152 | * The final value programmed to the register should always be rounded up, | |
4153 | * and include an extra 2 entries to account for clock crossings. | |
4154 | * | |
4155 | * We don't use the sprite, so we can ignore that. And on Crestline we have | |
4156 | * to set the non-SR watermarks to 8. | |
4157 | */ | |
46ba614c | 4158 | void intel_update_watermarks(struct drm_crtc *crtc) |
b445e3b0 | 4159 | { |
46ba614c | 4160 | struct drm_i915_private *dev_priv = crtc->dev->dev_private; |
b445e3b0 ED |
4161 | |
4162 | if (dev_priv->display.update_wm) | |
46ba614c | 4163 | dev_priv->display.update_wm(crtc); |
b445e3b0 ED |
4164 | } |
4165 | ||
adf3d35e VS |
4166 | void intel_update_sprite_watermarks(struct drm_plane *plane, |
4167 | struct drm_crtc *crtc, | |
ed57cb8a DL |
4168 | uint32_t sprite_width, |
4169 | uint32_t sprite_height, | |
4170 | int pixel_size, | |
39db4a4d | 4171 | bool enabled, bool scaled) |
b445e3b0 | 4172 | { |
adf3d35e | 4173 | struct drm_i915_private *dev_priv = plane->dev->dev_private; |
b445e3b0 ED |
4174 | |
4175 | if (dev_priv->display.update_sprite_wm) | |
ed57cb8a DL |
4176 | dev_priv->display.update_sprite_wm(plane, crtc, |
4177 | sprite_width, sprite_height, | |
39db4a4d | 4178 | pixel_size, enabled, scaled); |
b445e3b0 ED |
4179 | } |
4180 | ||
9270388e DV |
4181 | /** |
4182 | * Lock protecting IPS related data structures | |
9270388e DV |
4183 | */ |
4184 | DEFINE_SPINLOCK(mchdev_lock); | |
4185 | ||
4186 | /* Global for IPS driver to get at the current i915 device. Protected by | |
4187 | * mchdev_lock. */ | |
4188 | static struct drm_i915_private *i915_mch_dev; | |
4189 | ||
2b4e57bd ED |
4190 | bool ironlake_set_drps(struct drm_device *dev, u8 val) |
4191 | { | |
4192 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4193 | u16 rgvswctl; | |
4194 | ||
9270388e DV |
4195 | assert_spin_locked(&mchdev_lock); |
4196 | ||
2b4e57bd ED |
4197 | rgvswctl = I915_READ16(MEMSWCTL); |
4198 | if (rgvswctl & MEMCTL_CMD_STS) { | |
4199 | DRM_DEBUG("gpu busy, RCS change rejected\n"); | |
4200 | return false; /* still busy with another command */ | |
4201 | } | |
4202 | ||
4203 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | | |
4204 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; | |
4205 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
4206 | POSTING_READ16(MEMSWCTL); | |
4207 | ||
4208 | rgvswctl |= MEMCTL_CMD_STS; | |
4209 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
4210 | ||
4211 | return true; | |
4212 | } | |
4213 | ||
8090c6b9 | 4214 | static void ironlake_enable_drps(struct drm_device *dev) |
2b4e57bd ED |
4215 | { |
4216 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4217 | u32 rgvmodectl = I915_READ(MEMMODECTL); | |
4218 | u8 fmax, fmin, fstart, vstart; | |
4219 | ||
9270388e DV |
4220 | spin_lock_irq(&mchdev_lock); |
4221 | ||
2b4e57bd ED |
4222 | /* Enable temp reporting */ |
4223 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); | |
4224 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); | |
4225 | ||
4226 | /* 100ms RC evaluation intervals */ | |
4227 | I915_WRITE(RCUPEI, 100000); | |
4228 | I915_WRITE(RCDNEI, 100000); | |
4229 | ||
4230 | /* Set max/min thresholds to 90ms and 80ms respectively */ | |
4231 | I915_WRITE(RCBMAXAVG, 90000); | |
4232 | I915_WRITE(RCBMINAVG, 80000); | |
4233 | ||
4234 | I915_WRITE(MEMIHYST, 1); | |
4235 | ||
4236 | /* Set up min, max, and cur for interrupt handling */ | |
4237 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; | |
4238 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); | |
4239 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> | |
4240 | MEMMODE_FSTART_SHIFT; | |
4241 | ||
4242 | vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >> | |
4243 | PXVFREQ_PX_SHIFT; | |
4244 | ||
20e4d407 DV |
4245 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
4246 | dev_priv->ips.fstart = fstart; | |
2b4e57bd | 4247 | |
20e4d407 DV |
4248 | dev_priv->ips.max_delay = fstart; |
4249 | dev_priv->ips.min_delay = fmin; | |
4250 | dev_priv->ips.cur_delay = fstart; | |
2b4e57bd ED |
4251 | |
4252 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", | |
4253 | fmax, fmin, fstart); | |
4254 | ||
4255 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); | |
4256 | ||
4257 | /* | |
4258 | * Interrupts will be enabled in ironlake_irq_postinstall | |
4259 | */ | |
4260 | ||
4261 | I915_WRITE(VIDSTART, vstart); | |
4262 | POSTING_READ(VIDSTART); | |
4263 | ||
4264 | rgvmodectl |= MEMMODE_SWMODE_EN; | |
4265 | I915_WRITE(MEMMODECTL, rgvmodectl); | |
4266 | ||
9270388e | 4267 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
2b4e57bd | 4268 | DRM_ERROR("stuck trying to change perf mode\n"); |
6adfb1ef | 4269 | msleep(1); |
2b4e57bd ED |
4270 | |
4271 | ironlake_set_drps(dev, fstart); | |
4272 | ||
20e4d407 | 4273 | dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) + |
2b4e57bd | 4274 | I915_READ(0x112e0); |
20e4d407 DV |
4275 | dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies); |
4276 | dev_priv->ips.last_count2 = I915_READ(0x112f4); | |
5ed0bdf2 | 4277 | dev_priv->ips.last_time2 = ktime_get_raw_ns(); |
9270388e DV |
4278 | |
4279 | spin_unlock_irq(&mchdev_lock); | |
2b4e57bd ED |
4280 | } |
4281 | ||
8090c6b9 | 4282 | static void ironlake_disable_drps(struct drm_device *dev) |
2b4e57bd ED |
4283 | { |
4284 | struct drm_i915_private *dev_priv = dev->dev_private; | |
9270388e DV |
4285 | u16 rgvswctl; |
4286 | ||
4287 | spin_lock_irq(&mchdev_lock); | |
4288 | ||
4289 | rgvswctl = I915_READ16(MEMSWCTL); | |
2b4e57bd ED |
4290 | |
4291 | /* Ack interrupts, disable EFC interrupt */ | |
4292 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); | |
4293 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); | |
4294 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); | |
4295 | I915_WRITE(DEIIR, DE_PCU_EVENT); | |
4296 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); | |
4297 | ||
4298 | /* Go back to the starting frequency */ | |
20e4d407 | 4299 | ironlake_set_drps(dev, dev_priv->ips.fstart); |
6adfb1ef | 4300 | msleep(1); |
2b4e57bd ED |
4301 | rgvswctl |= MEMCTL_CMD_STS; |
4302 | I915_WRITE(MEMSWCTL, rgvswctl); | |
6adfb1ef | 4303 | msleep(1); |
2b4e57bd | 4304 | |
9270388e | 4305 | spin_unlock_irq(&mchdev_lock); |
2b4e57bd ED |
4306 | } |
4307 | ||
acbe9475 DV |
4308 | /* There's a funny hw issue where the hw returns all 0 when reading from |
4309 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value | |
4310 | * ourselves, instead of doing a rmw cycle (which might result in us clearing | |
4311 | * all limits and the gpu stuck at whatever frequency it is at atm). | |
4312 | */ | |
74ef1173 | 4313 | static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val) |
2b4e57bd | 4314 | { |
7b9e0ae6 | 4315 | u32 limits; |
2b4e57bd | 4316 | |
20b46e59 DV |
4317 | /* Only set the down limit when we've reached the lowest level to avoid |
4318 | * getting more interrupts, otherwise leave this clear. This prevents a | |
4319 | * race in the hw when coming out of rc6: There's a tiny window where | |
4320 | * the hw runs at the minimal clock before selecting the desired | |
4321 | * frequency, if the down threshold expires in that window we will not | |
4322 | * receive a down interrupt. */ | |
74ef1173 AG |
4323 | if (IS_GEN9(dev_priv->dev)) { |
4324 | limits = (dev_priv->rps.max_freq_softlimit) << 23; | |
4325 | if (val <= dev_priv->rps.min_freq_softlimit) | |
4326 | limits |= (dev_priv->rps.min_freq_softlimit) << 14; | |
4327 | } else { | |
4328 | limits = dev_priv->rps.max_freq_softlimit << 24; | |
4329 | if (val <= dev_priv->rps.min_freq_softlimit) | |
4330 | limits |= dev_priv->rps.min_freq_softlimit << 16; | |
4331 | } | |
20b46e59 DV |
4332 | |
4333 | return limits; | |
4334 | } | |
4335 | ||
dd75fdc8 CW |
4336 | static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val) |
4337 | { | |
4338 | int new_power; | |
8a586437 AG |
4339 | u32 threshold_up = 0, threshold_down = 0; /* in % */ |
4340 | u32 ei_up = 0, ei_down = 0; | |
dd75fdc8 CW |
4341 | |
4342 | new_power = dev_priv->rps.power; | |
4343 | switch (dev_priv->rps.power) { | |
4344 | case LOW_POWER: | |
b39fb297 | 4345 | if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq) |
dd75fdc8 CW |
4346 | new_power = BETWEEN; |
4347 | break; | |
4348 | ||
4349 | case BETWEEN: | |
b39fb297 | 4350 | if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq) |
dd75fdc8 | 4351 | new_power = LOW_POWER; |
b39fb297 | 4352 | else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq) |
dd75fdc8 CW |
4353 | new_power = HIGH_POWER; |
4354 | break; | |
4355 | ||
4356 | case HIGH_POWER: | |
b39fb297 | 4357 | if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq) |
dd75fdc8 CW |
4358 | new_power = BETWEEN; |
4359 | break; | |
4360 | } | |
4361 | /* Max/min bins are special */ | |
aed242ff | 4362 | if (val <= dev_priv->rps.min_freq_softlimit) |
dd75fdc8 | 4363 | new_power = LOW_POWER; |
aed242ff | 4364 | if (val >= dev_priv->rps.max_freq_softlimit) |
dd75fdc8 CW |
4365 | new_power = HIGH_POWER; |
4366 | if (new_power == dev_priv->rps.power) | |
4367 | return; | |
4368 | ||
4369 | /* Note the units here are not exactly 1us, but 1280ns. */ | |
4370 | switch (new_power) { | |
4371 | case LOW_POWER: | |
4372 | /* Upclock if more than 95% busy over 16ms */ | |
8a586437 AG |
4373 | ei_up = 16000; |
4374 | threshold_up = 95; | |
dd75fdc8 CW |
4375 | |
4376 | /* Downclock if less than 85% busy over 32ms */ | |
8a586437 AG |
4377 | ei_down = 32000; |
4378 | threshold_down = 85; | |
dd75fdc8 CW |
4379 | break; |
4380 | ||
4381 | case BETWEEN: | |
4382 | /* Upclock if more than 90% busy over 13ms */ | |
8a586437 AG |
4383 | ei_up = 13000; |
4384 | threshold_up = 90; | |
dd75fdc8 CW |
4385 | |
4386 | /* Downclock if less than 75% busy over 32ms */ | |
8a586437 AG |
4387 | ei_down = 32000; |
4388 | threshold_down = 75; | |
dd75fdc8 CW |
4389 | break; |
4390 | ||
4391 | case HIGH_POWER: | |
4392 | /* Upclock if more than 85% busy over 10ms */ | |
8a586437 AG |
4393 | ei_up = 10000; |
4394 | threshold_up = 85; | |
dd75fdc8 CW |
4395 | |
4396 | /* Downclock if less than 60% busy over 32ms */ | |
8a586437 AG |
4397 | ei_down = 32000; |
4398 | threshold_down = 60; | |
dd75fdc8 CW |
4399 | break; |
4400 | } | |
4401 | ||
8a586437 AG |
4402 | I915_WRITE(GEN6_RP_UP_EI, |
4403 | GT_INTERVAL_FROM_US(dev_priv, ei_up)); | |
4404 | I915_WRITE(GEN6_RP_UP_THRESHOLD, | |
4405 | GT_INTERVAL_FROM_US(dev_priv, (ei_up * threshold_up / 100))); | |
4406 | ||
4407 | I915_WRITE(GEN6_RP_DOWN_EI, | |
4408 | GT_INTERVAL_FROM_US(dev_priv, ei_down)); | |
4409 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, | |
4410 | GT_INTERVAL_FROM_US(dev_priv, (ei_down * threshold_down / 100))); | |
4411 | ||
4412 | I915_WRITE(GEN6_RP_CONTROL, | |
4413 | GEN6_RP_MEDIA_TURBO | | |
4414 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
4415 | GEN6_RP_MEDIA_IS_GFX | | |
4416 | GEN6_RP_ENABLE | | |
4417 | GEN6_RP_UP_BUSY_AVG | | |
4418 | GEN6_RP_DOWN_IDLE_AVG); | |
4419 | ||
dd75fdc8 | 4420 | dev_priv->rps.power = new_power; |
8fb55197 CW |
4421 | dev_priv->rps.up_threshold = threshold_up; |
4422 | dev_priv->rps.down_threshold = threshold_down; | |
dd75fdc8 CW |
4423 | dev_priv->rps.last_adj = 0; |
4424 | } | |
4425 | ||
2876ce73 CW |
4426 | static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val) |
4427 | { | |
4428 | u32 mask = 0; | |
4429 | ||
4430 | if (val > dev_priv->rps.min_freq_softlimit) | |
6f4b12f8 | 4431 | mask |= GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT; |
2876ce73 | 4432 | if (val < dev_priv->rps.max_freq_softlimit) |
6f4b12f8 | 4433 | mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD; |
2876ce73 | 4434 | |
7b3c29f6 CW |
4435 | mask &= dev_priv->pm_rps_events; |
4436 | ||
59d02a1f | 4437 | return gen6_sanitize_rps_pm_mask(dev_priv, ~mask); |
2876ce73 CW |
4438 | } |
4439 | ||
b8a5ff8d JM |
4440 | /* gen6_set_rps is called to update the frequency request, but should also be |
4441 | * called when the range (min_delay and max_delay) is modified so that we can | |
4442 | * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */ | |
ffe02b40 | 4443 | static void gen6_set_rps(struct drm_device *dev, u8 val) |
20b46e59 DV |
4444 | { |
4445 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7b9e0ae6 | 4446 | |
4fc688ce | 4447 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
aed242ff CW |
4448 | WARN_ON(val > dev_priv->rps.max_freq); |
4449 | WARN_ON(val < dev_priv->rps.min_freq); | |
004777cb | 4450 | |
eb64cad1 CW |
4451 | /* min/max delay may still have been modified so be sure to |
4452 | * write the limits value. | |
4453 | */ | |
4454 | if (val != dev_priv->rps.cur_freq) { | |
4455 | gen6_set_rps_thresholds(dev_priv, val); | |
b8a5ff8d | 4456 | |
5704195c AG |
4457 | if (IS_GEN9(dev)) |
4458 | I915_WRITE(GEN6_RPNSWREQ, | |
4459 | GEN9_FREQUENCY(val)); | |
4460 | else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) | |
eb64cad1 CW |
4461 | I915_WRITE(GEN6_RPNSWREQ, |
4462 | HSW_FREQUENCY(val)); | |
4463 | else | |
4464 | I915_WRITE(GEN6_RPNSWREQ, | |
4465 | GEN6_FREQUENCY(val) | | |
4466 | GEN6_OFFSET(0) | | |
4467 | GEN6_AGGRESSIVE_TURBO); | |
b8a5ff8d | 4468 | } |
7b9e0ae6 | 4469 | |
7b9e0ae6 CW |
4470 | /* Make sure we continue to get interrupts |
4471 | * until we hit the minimum or maximum frequencies. | |
4472 | */ | |
74ef1173 | 4473 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val)); |
2876ce73 | 4474 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
7b9e0ae6 | 4475 | |
d5570a72 BW |
4476 | POSTING_READ(GEN6_RPNSWREQ); |
4477 | ||
b39fb297 | 4478 | dev_priv->rps.cur_freq = val; |
be2cde9a | 4479 | trace_intel_gpu_freq_change(val * 50); |
2b4e57bd ED |
4480 | } |
4481 | ||
ffe02b40 VS |
4482 | static void valleyview_set_rps(struct drm_device *dev, u8 val) |
4483 | { | |
4484 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4485 | ||
4486 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
aed242ff CW |
4487 | WARN_ON(val > dev_priv->rps.max_freq); |
4488 | WARN_ON(val < dev_priv->rps.min_freq); | |
ffe02b40 VS |
4489 | |
4490 | if (WARN_ONCE(IS_CHERRYVIEW(dev) && (val & 1), | |
4491 | "Odd GPU freq value\n")) | |
4492 | val &= ~1; | |
4493 | ||
cd25dd5b D |
4494 | I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val)); |
4495 | ||
8fb55197 | 4496 | if (val != dev_priv->rps.cur_freq) { |
ffe02b40 | 4497 | vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val); |
8fb55197 CW |
4498 | if (!IS_CHERRYVIEW(dev_priv)) |
4499 | gen6_set_rps_thresholds(dev_priv, val); | |
4500 | } | |
ffe02b40 | 4501 | |
ffe02b40 VS |
4502 | dev_priv->rps.cur_freq = val; |
4503 | trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val)); | |
4504 | } | |
4505 | ||
a7f6e231 | 4506 | /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down |
76c3552f D |
4507 | * |
4508 | * * If Gfx is Idle, then | |
a7f6e231 D |
4509 | * 1. Forcewake Media well. |
4510 | * 2. Request idle freq. | |
4511 | * 3. Release Forcewake of Media well. | |
76c3552f D |
4512 | */ |
4513 | static void vlv_set_rps_idle(struct drm_i915_private *dev_priv) | |
4514 | { | |
aed242ff | 4515 | u32 val = dev_priv->rps.idle_freq; |
5549d25f | 4516 | |
aed242ff | 4517 | if (dev_priv->rps.cur_freq <= val) |
76c3552f D |
4518 | return; |
4519 | ||
a7f6e231 D |
4520 | /* Wake up the media well, as that takes a lot less |
4521 | * power than the Render well. */ | |
4522 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA); | |
4523 | valleyview_set_rps(dev_priv->dev, val); | |
4524 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA); | |
76c3552f D |
4525 | } |
4526 | ||
43cf3bf0 CW |
4527 | void gen6_rps_busy(struct drm_i915_private *dev_priv) |
4528 | { | |
4529 | mutex_lock(&dev_priv->rps.hw_lock); | |
4530 | if (dev_priv->rps.enabled) { | |
4531 | if (dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED)) | |
4532 | gen6_rps_reset_ei(dev_priv); | |
4533 | I915_WRITE(GEN6_PMINTRMSK, | |
4534 | gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq)); | |
4535 | } | |
4536 | mutex_unlock(&dev_priv->rps.hw_lock); | |
4537 | } | |
4538 | ||
b29c19b6 CW |
4539 | void gen6_rps_idle(struct drm_i915_private *dev_priv) |
4540 | { | |
691bb717 DL |
4541 | struct drm_device *dev = dev_priv->dev; |
4542 | ||
b29c19b6 | 4543 | mutex_lock(&dev_priv->rps.hw_lock); |
c0951f0c | 4544 | if (dev_priv->rps.enabled) { |
21a11fff | 4545 | if (IS_VALLEYVIEW(dev)) |
76c3552f | 4546 | vlv_set_rps_idle(dev_priv); |
7526ed79 | 4547 | else |
aed242ff | 4548 | gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq); |
c0951f0c | 4549 | dev_priv->rps.last_adj = 0; |
43cf3bf0 | 4550 | I915_WRITE(GEN6_PMINTRMSK, 0xffffffff); |
c0951f0c | 4551 | } |
8d3afd7d | 4552 | mutex_unlock(&dev_priv->rps.hw_lock); |
1854d5ca | 4553 | |
8d3afd7d | 4554 | spin_lock(&dev_priv->rps.client_lock); |
1854d5ca CW |
4555 | while (!list_empty(&dev_priv->rps.clients)) |
4556 | list_del_init(dev_priv->rps.clients.next); | |
8d3afd7d | 4557 | spin_unlock(&dev_priv->rps.client_lock); |
b29c19b6 CW |
4558 | } |
4559 | ||
1854d5ca | 4560 | void gen6_rps_boost(struct drm_i915_private *dev_priv, |
e61b9958 CW |
4561 | struct intel_rps_client *rps, |
4562 | unsigned long submitted) | |
b29c19b6 | 4563 | { |
8d3afd7d CW |
4564 | /* This is intentionally racy! We peek at the state here, then |
4565 | * validate inside the RPS worker. | |
4566 | */ | |
4567 | if (!(dev_priv->mm.busy && | |
4568 | dev_priv->rps.enabled && | |
4569 | dev_priv->rps.cur_freq < dev_priv->rps.max_freq_softlimit)) | |
4570 | return; | |
43cf3bf0 | 4571 | |
e61b9958 CW |
4572 | /* Force a RPS boost (and don't count it against the client) if |
4573 | * the GPU is severely congested. | |
4574 | */ | |
d0bc54f2 | 4575 | if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES)) |
e61b9958 CW |
4576 | rps = NULL; |
4577 | ||
8d3afd7d CW |
4578 | spin_lock(&dev_priv->rps.client_lock); |
4579 | if (rps == NULL || list_empty(&rps->link)) { | |
4580 | spin_lock_irq(&dev_priv->irq_lock); | |
4581 | if (dev_priv->rps.interrupts_enabled) { | |
4582 | dev_priv->rps.client_boost = true; | |
4583 | queue_work(dev_priv->wq, &dev_priv->rps.work); | |
4584 | } | |
4585 | spin_unlock_irq(&dev_priv->irq_lock); | |
1854d5ca | 4586 | |
2e1b8730 CW |
4587 | if (rps != NULL) { |
4588 | list_add(&rps->link, &dev_priv->rps.clients); | |
4589 | rps->boosts++; | |
1854d5ca CW |
4590 | } else |
4591 | dev_priv->rps.boosts++; | |
c0951f0c | 4592 | } |
8d3afd7d | 4593 | spin_unlock(&dev_priv->rps.client_lock); |
b29c19b6 CW |
4594 | } |
4595 | ||
ffe02b40 | 4596 | void intel_set_rps(struct drm_device *dev, u8 val) |
0a073b84 | 4597 | { |
ffe02b40 VS |
4598 | if (IS_VALLEYVIEW(dev)) |
4599 | valleyview_set_rps(dev, val); | |
4600 | else | |
4601 | gen6_set_rps(dev, val); | |
0a073b84 JB |
4602 | } |
4603 | ||
20e49366 ZW |
4604 | static void gen9_disable_rps(struct drm_device *dev) |
4605 | { | |
4606 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4607 | ||
4608 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
38c23527 | 4609 | I915_WRITE(GEN9_PG_ENABLE, 0); |
20e49366 ZW |
4610 | } |
4611 | ||
44fc7d5c | 4612 | static void gen6_disable_rps(struct drm_device *dev) |
d20d4f0c JB |
4613 | { |
4614 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4615 | ||
4616 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
44fc7d5c | 4617 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
44fc7d5c DV |
4618 | } |
4619 | ||
38807746 D |
4620 | static void cherryview_disable_rps(struct drm_device *dev) |
4621 | { | |
4622 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4623 | ||
4624 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
4625 | } | |
4626 | ||
44fc7d5c DV |
4627 | static void valleyview_disable_rps(struct drm_device *dev) |
4628 | { | |
4629 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4630 | ||
98a2e5f9 D |
4631 | /* we're doing forcewake before Disabling RC6, |
4632 | * This what the BIOS expects when going into suspend */ | |
59bad947 | 4633 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
98a2e5f9 | 4634 | |
44fc7d5c | 4635 | I915_WRITE(GEN6_RC_CONTROL, 0); |
d20d4f0c | 4636 | |
59bad947 | 4637 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
d20d4f0c JB |
4638 | } |
4639 | ||
dc39fff7 BW |
4640 | static void intel_print_rc6_info(struct drm_device *dev, u32 mode) |
4641 | { | |
91ca689a ID |
4642 | if (IS_VALLEYVIEW(dev)) { |
4643 | if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1))) | |
4644 | mode = GEN6_RC_CTL_RC6_ENABLE; | |
4645 | else | |
4646 | mode = 0; | |
4647 | } | |
58abf1da RV |
4648 | if (HAS_RC6p(dev)) |
4649 | DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n", | |
4650 | (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off", | |
4651 | (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off", | |
4652 | (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off"); | |
4653 | ||
4654 | else | |
4655 | DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n", | |
4656 | (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off"); | |
dc39fff7 BW |
4657 | } |
4658 | ||
e6069ca8 | 4659 | static int sanitize_rc6_option(const struct drm_device *dev, int enable_rc6) |
2b4e57bd | 4660 | { |
e7d66d89 DV |
4661 | /* No RC6 before Ironlake and code is gone for ilk. */ |
4662 | if (INTEL_INFO(dev)->gen < 6) | |
e6069ca8 ID |
4663 | return 0; |
4664 | ||
456470eb | 4665 | /* Respect the kernel parameter if it is set */ |
e6069ca8 ID |
4666 | if (enable_rc6 >= 0) { |
4667 | int mask; | |
4668 | ||
58abf1da | 4669 | if (HAS_RC6p(dev)) |
e6069ca8 ID |
4670 | mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE | |
4671 | INTEL_RC6pp_ENABLE; | |
4672 | else | |
4673 | mask = INTEL_RC6_ENABLE; | |
4674 | ||
4675 | if ((enable_rc6 & mask) != enable_rc6) | |
8dfd1f04 DV |
4676 | DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n", |
4677 | enable_rc6 & mask, enable_rc6, mask); | |
e6069ca8 ID |
4678 | |
4679 | return enable_rc6 & mask; | |
4680 | } | |
2b4e57bd | 4681 | |
8bade1ad | 4682 | if (IS_IVYBRIDGE(dev)) |
cca84a1f | 4683 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); |
8bade1ad BW |
4684 | |
4685 | return INTEL_RC6_ENABLE; | |
2b4e57bd ED |
4686 | } |
4687 | ||
e6069ca8 ID |
4688 | int intel_enable_rc6(const struct drm_device *dev) |
4689 | { | |
4690 | return i915.enable_rc6; | |
4691 | } | |
4692 | ||
93ee2920 | 4693 | static void gen6_init_rps_frequencies(struct drm_device *dev) |
3280e8b0 | 4694 | { |
93ee2920 TR |
4695 | struct drm_i915_private *dev_priv = dev->dev_private; |
4696 | uint32_t rp_state_cap; | |
4697 | u32 ddcc_status = 0; | |
4698 | int ret; | |
4699 | ||
3280e8b0 BW |
4700 | /* All of these values are in units of 50MHz */ |
4701 | dev_priv->rps.cur_freq = 0; | |
93ee2920 | 4702 | /* static values from HW: RP0 > RP1 > RPn (min_freq) */ |
35040562 BP |
4703 | if (IS_BROXTON(dev)) { |
4704 | rp_state_cap = I915_READ(BXT_RP_STATE_CAP); | |
4705 | dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff; | |
4706 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; | |
4707 | dev_priv->rps.min_freq = (rp_state_cap >> 0) & 0xff; | |
4708 | } else { | |
4709 | rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); | |
4710 | dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff; | |
4711 | dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff; | |
4712 | dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff; | |
4713 | } | |
4714 | ||
3280e8b0 BW |
4715 | /* hw_max = RP0 until we check for overclocking */ |
4716 | dev_priv->rps.max_freq = dev_priv->rps.rp0_freq; | |
4717 | ||
93ee2920 | 4718 | dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq; |
c5e0688c | 4719 | if (IS_HASWELL(dev) || IS_BROADWELL(dev) || IS_SKYLAKE(dev)) { |
93ee2920 TR |
4720 | ret = sandybridge_pcode_read(dev_priv, |
4721 | HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL, | |
4722 | &ddcc_status); | |
4723 | if (0 == ret) | |
4724 | dev_priv->rps.efficient_freq = | |
46efa4ab TR |
4725 | clamp_t(u8, |
4726 | ((ddcc_status >> 8) & 0xff), | |
4727 | dev_priv->rps.min_freq, | |
4728 | dev_priv->rps.max_freq); | |
93ee2920 TR |
4729 | } |
4730 | ||
c5e0688c AG |
4731 | if (IS_SKYLAKE(dev)) { |
4732 | /* Store the frequency values in 16.66 MHZ units, which is | |
4733 | the natural hardware unit for SKL */ | |
4734 | dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER; | |
4735 | dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER; | |
4736 | dev_priv->rps.min_freq *= GEN9_FREQ_SCALER; | |
4737 | dev_priv->rps.max_freq *= GEN9_FREQ_SCALER; | |
4738 | dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER; | |
4739 | } | |
4740 | ||
aed242ff CW |
4741 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; |
4742 | ||
3280e8b0 BW |
4743 | /* Preserve min/max settings in case of re-init */ |
4744 | if (dev_priv->rps.max_freq_softlimit == 0) | |
4745 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; | |
4746 | ||
93ee2920 TR |
4747 | if (dev_priv->rps.min_freq_softlimit == 0) { |
4748 | if (IS_HASWELL(dev) || IS_BROADWELL(dev)) | |
4749 | dev_priv->rps.min_freq_softlimit = | |
813b5e69 VS |
4750 | max_t(int, dev_priv->rps.efficient_freq, |
4751 | intel_freq_opcode(dev_priv, 450)); | |
93ee2920 TR |
4752 | else |
4753 | dev_priv->rps.min_freq_softlimit = | |
4754 | dev_priv->rps.min_freq; | |
4755 | } | |
3280e8b0 BW |
4756 | } |
4757 | ||
b6fef0ef | 4758 | /* See the Gen9_GT_PM_Programming_Guide doc for the below */ |
20e49366 | 4759 | static void gen9_enable_rps(struct drm_device *dev) |
b6fef0ef JB |
4760 | { |
4761 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4762 | ||
4763 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); | |
4764 | ||
ba1c554c DL |
4765 | gen6_init_rps_frequencies(dev); |
4766 | ||
0beb059a AG |
4767 | /* Program defaults and thresholds for RPS*/ |
4768 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
4769 | GEN9_FREQUENCY(dev_priv->rps.rp1_freq)); | |
4770 | ||
4771 | /* 1 second timeout*/ | |
4772 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, | |
4773 | GT_INTERVAL_FROM_US(dev_priv, 1000000)); | |
4774 | ||
b6fef0ef | 4775 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa); |
b6fef0ef | 4776 | |
0beb059a AG |
4777 | /* Leaning on the below call to gen6_set_rps to program/setup the |
4778 | * Up/Down EI & threshold registers, as well as the RP_CONTROL, | |
4779 | * RP_INTERRUPT_LIMITS & RPNSWREQ registers */ | |
4780 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ | |
4781 | gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit); | |
b6fef0ef JB |
4782 | |
4783 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); | |
4784 | } | |
4785 | ||
4786 | static void gen9_enable_rc6(struct drm_device *dev) | |
20e49366 ZW |
4787 | { |
4788 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4789 | struct intel_engine_cs *ring; | |
4790 | uint32_t rc6_mask = 0; | |
4791 | int unused; | |
4792 | ||
4793 | /* 1a: Software RC state - RC0 */ | |
4794 | I915_WRITE(GEN6_RC_STATE, 0); | |
4795 | ||
4796 | /* 1b: Get forcewake during program sequence. Although the driver | |
4797 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 4798 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
20e49366 ZW |
4799 | |
4800 | /* 2a: Disable RC states. */ | |
4801 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
4802 | ||
4803 | /* 2b: Program RC6 thresholds.*/ | |
4804 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16); | |
4805 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
4806 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
4807 | for_each_ring(ring, dev_priv, unused) | |
4808 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
4809 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
4810 | I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */ | |
4811 | ||
38c23527 ZW |
4812 | /* 2c: Program Coarse Power Gating Policies. */ |
4813 | I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25); | |
4814 | I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25); | |
4815 | ||
20e49366 ZW |
4816 | /* 3a: Enable RC6 */ |
4817 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) | |
4818 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; | |
4819 | DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? | |
4820 | "on" : "off"); | |
4821 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | | |
4822 | GEN6_RC_CTL_EI_MODE(1) | | |
4823 | rc6_mask); | |
4824 | ||
cb07bae0 SK |
4825 | /* |
4826 | * 3b: Enable Coarse Power Gating only when RC6 is enabled. | |
4827 | * WaDisableRenderPowerGating:skl,bxt - Render PG need to be disabled with RC6. | |
4828 | */ | |
a4104c55 | 4829 | I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? |
cb07bae0 | 4830 | GEN9_MEDIA_PG_ENABLE : 0); |
a4104c55 | 4831 | |
38c23527 | 4832 | |
59bad947 | 4833 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
20e49366 ZW |
4834 | |
4835 | } | |
4836 | ||
6edee7f3 BW |
4837 | static void gen8_enable_rps(struct drm_device *dev) |
4838 | { | |
4839 | struct drm_i915_private *dev_priv = dev->dev_private; | |
a4872ba6 | 4840 | struct intel_engine_cs *ring; |
93ee2920 | 4841 | uint32_t rc6_mask = 0; |
6edee7f3 BW |
4842 | int unused; |
4843 | ||
4844 | /* 1a: Software RC state - RC0 */ | |
4845 | I915_WRITE(GEN6_RC_STATE, 0); | |
4846 | ||
4847 | /* 1c & 1d: Get forcewake during program sequence. Although the driver | |
4848 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 4849 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
6edee7f3 BW |
4850 | |
4851 | /* 2a: Disable RC states. */ | |
4852 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
4853 | ||
93ee2920 TR |
4854 | /* Initialize rps frequencies */ |
4855 | gen6_init_rps_frequencies(dev); | |
6edee7f3 BW |
4856 | |
4857 | /* 2b: Program RC6 thresholds.*/ | |
4858 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); | |
4859 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
4860 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
4861 | for_each_ring(ring, dev_priv, unused) | |
4862 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
4863 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
0d68b25e TR |
4864 | if (IS_BROADWELL(dev)) |
4865 | I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */ | |
4866 | else | |
4867 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */ | |
6edee7f3 BW |
4868 | |
4869 | /* 3: Enable RC6 */ | |
4870 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) | |
4871 | rc6_mask = GEN6_RC_CTL_RC6_ENABLE; | |
abbf9d2c | 4872 | intel_print_rc6_info(dev, rc6_mask); |
0d68b25e TR |
4873 | if (IS_BROADWELL(dev)) |
4874 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | | |
4875 | GEN7_RC_CTL_TO_MODE | | |
4876 | rc6_mask); | |
4877 | else | |
4878 | I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE | | |
4879 | GEN6_RC_CTL_EI_MODE(1) | | |
4880 | rc6_mask); | |
6edee7f3 BW |
4881 | |
4882 | /* 4 Program defaults and thresholds for RPS*/ | |
f9bdc585 BW |
4883 | I915_WRITE(GEN6_RPNSWREQ, |
4884 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); | |
4885 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
4886 | HSW_FREQUENCY(dev_priv->rps.rp1_freq)); | |
7526ed79 DV |
4887 | /* NB: Docs say 1s, and 1000000 - which aren't equivalent */ |
4888 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */ | |
4889 | ||
4890 | /* Docs recommend 900MHz, and 300 MHz respectively */ | |
4891 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, | |
4892 | dev_priv->rps.max_freq_softlimit << 24 | | |
4893 | dev_priv->rps.min_freq_softlimit << 16); | |
4894 | ||
4895 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */ | |
4896 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/ | |
4897 | I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */ | |
4898 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */ | |
4899 | ||
4900 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
6edee7f3 BW |
4901 | |
4902 | /* 5: Enable RPS */ | |
7526ed79 DV |
4903 | I915_WRITE(GEN6_RP_CONTROL, |
4904 | GEN6_RP_MEDIA_TURBO | | |
4905 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
4906 | GEN6_RP_MEDIA_IS_GFX | | |
4907 | GEN6_RP_ENABLE | | |
4908 | GEN6_RP_UP_BUSY_AVG | | |
4909 | GEN6_RP_DOWN_IDLE_AVG); | |
4910 | ||
4911 | /* 6: Ring frequency + overclocking (our driver does this later */ | |
4912 | ||
c7f3153a | 4913 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
aed242ff | 4914 | gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq); |
7526ed79 | 4915 | |
59bad947 | 4916 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
6edee7f3 BW |
4917 | } |
4918 | ||
79f5b2c7 | 4919 | static void gen6_enable_rps(struct drm_device *dev) |
2b4e57bd | 4920 | { |
79f5b2c7 | 4921 | struct drm_i915_private *dev_priv = dev->dev_private; |
a4872ba6 | 4922 | struct intel_engine_cs *ring; |
d060c169 | 4923 | u32 rc6vids, pcu_mbox = 0, rc6_mask = 0; |
2b4e57bd | 4924 | u32 gtfifodbg; |
2b4e57bd | 4925 | int rc6_mode; |
42c0526c | 4926 | int i, ret; |
2b4e57bd | 4927 | |
4fc688ce | 4928 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 4929 | |
2b4e57bd ED |
4930 | /* Here begins a magic sequence of register writes to enable |
4931 | * auto-downclocking. | |
4932 | * | |
4933 | * Perhaps there might be some value in exposing these to | |
4934 | * userspace... | |
4935 | */ | |
4936 | I915_WRITE(GEN6_RC_STATE, 0); | |
2b4e57bd ED |
4937 | |
4938 | /* Clear the DBG now so we don't confuse earlier errors */ | |
4939 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { | |
4940 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); | |
4941 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
4942 | } | |
4943 | ||
59bad947 | 4944 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
2b4e57bd | 4945 | |
93ee2920 TR |
4946 | /* Initialize rps frequencies */ |
4947 | gen6_init_rps_frequencies(dev); | |
dd0a1aa1 | 4948 | |
2b4e57bd ED |
4949 | /* disable the counters and set deterministic thresholds */ |
4950 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
4951 | ||
4952 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); | |
4953 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); | |
4954 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); | |
4955 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
4956 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
4957 | ||
b4519513 CW |
4958 | for_each_ring(ring, dev_priv, i) |
4959 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
2b4e57bd ED |
4960 | |
4961 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
4962 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); | |
29c78f60 | 4963 | if (IS_IVYBRIDGE(dev)) |
351aa566 SM |
4964 | I915_WRITE(GEN6_RC6_THRESHOLD, 125000); |
4965 | else | |
4966 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); | |
0920a487 | 4967 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
2b4e57bd ED |
4968 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
4969 | ||
5a7dc92a | 4970 | /* Check if we are enabling RC6 */ |
2b4e57bd ED |
4971 | rc6_mode = intel_enable_rc6(dev_priv->dev); |
4972 | if (rc6_mode & INTEL_RC6_ENABLE) | |
4973 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; | |
4974 | ||
5a7dc92a ED |
4975 | /* We don't use those on Haswell */ |
4976 | if (!IS_HASWELL(dev)) { | |
4977 | if (rc6_mode & INTEL_RC6p_ENABLE) | |
4978 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; | |
2b4e57bd | 4979 | |
5a7dc92a ED |
4980 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
4981 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; | |
4982 | } | |
2b4e57bd | 4983 | |
dc39fff7 | 4984 | intel_print_rc6_info(dev, rc6_mask); |
2b4e57bd ED |
4985 | |
4986 | I915_WRITE(GEN6_RC_CONTROL, | |
4987 | rc6_mask | | |
4988 | GEN6_RC_CTL_EI_MODE(1) | | |
4989 | GEN6_RC_CTL_HW_ENABLE); | |
4990 | ||
dd75fdc8 CW |
4991 | /* Power down if completely idle for over 50ms */ |
4992 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000); | |
2b4e57bd | 4993 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
2b4e57bd | 4994 | |
42c0526c | 4995 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0); |
d060c169 | 4996 | if (ret) |
42c0526c | 4997 | DRM_DEBUG_DRIVER("Failed to set the min frequency\n"); |
d060c169 BW |
4998 | |
4999 | ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox); | |
5000 | if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */ | |
5001 | DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n", | |
b39fb297 | 5002 | (dev_priv->rps.max_freq_softlimit & 0xff) * 50, |
d060c169 | 5003 | (pcu_mbox & 0xff) * 50); |
b39fb297 | 5004 | dev_priv->rps.max_freq = pcu_mbox & 0xff; |
2b4e57bd ED |
5005 | } |
5006 | ||
dd75fdc8 | 5007 | dev_priv->rps.power = HIGH_POWER; /* force a reset */ |
aed242ff | 5008 | gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq); |
2b4e57bd | 5009 | |
31643d54 BW |
5010 | rc6vids = 0; |
5011 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); | |
5012 | if (IS_GEN6(dev) && ret) { | |
5013 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); | |
5014 | } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { | |
5015 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", | |
5016 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); | |
5017 | rc6vids &= 0xffff00; | |
5018 | rc6vids |= GEN6_ENCODE_RC6_VID(450); | |
5019 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); | |
5020 | if (ret) | |
5021 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); | |
5022 | } | |
5023 | ||
59bad947 | 5024 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
2b4e57bd ED |
5025 | } |
5026 | ||
c2bc2fc5 | 5027 | static void __gen6_update_ring_freq(struct drm_device *dev) |
2b4e57bd | 5028 | { |
79f5b2c7 | 5029 | struct drm_i915_private *dev_priv = dev->dev_private; |
2b4e57bd | 5030 | int min_freq = 15; |
3ebecd07 CW |
5031 | unsigned int gpu_freq; |
5032 | unsigned int max_ia_freq, min_ring_freq; | |
4c8c7743 | 5033 | unsigned int max_gpu_freq, min_gpu_freq; |
2b4e57bd | 5034 | int scaling_factor = 180; |
eda79642 | 5035 | struct cpufreq_policy *policy; |
2b4e57bd | 5036 | |
4fc688ce | 5037 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 5038 | |
eda79642 BW |
5039 | policy = cpufreq_cpu_get(0); |
5040 | if (policy) { | |
5041 | max_ia_freq = policy->cpuinfo.max_freq; | |
5042 | cpufreq_cpu_put(policy); | |
5043 | } else { | |
5044 | /* | |
5045 | * Default to measured freq if none found, PCU will ensure we | |
5046 | * don't go over | |
5047 | */ | |
2b4e57bd | 5048 | max_ia_freq = tsc_khz; |
eda79642 | 5049 | } |
2b4e57bd ED |
5050 | |
5051 | /* Convert from kHz to MHz */ | |
5052 | max_ia_freq /= 1000; | |
5053 | ||
153b4b95 | 5054 | min_ring_freq = I915_READ(DCLK) & 0xf; |
f6aca45c BW |
5055 | /* convert DDR frequency from units of 266.6MHz to bandwidth */ |
5056 | min_ring_freq = mult_frac(min_ring_freq, 8, 3); | |
3ebecd07 | 5057 | |
4c8c7743 AG |
5058 | if (IS_SKYLAKE(dev)) { |
5059 | /* Convert GT frequency to 50 HZ units */ | |
5060 | min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER; | |
5061 | max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER; | |
5062 | } else { | |
5063 | min_gpu_freq = dev_priv->rps.min_freq; | |
5064 | max_gpu_freq = dev_priv->rps.max_freq; | |
5065 | } | |
5066 | ||
2b4e57bd ED |
5067 | /* |
5068 | * For each potential GPU frequency, load a ring frequency we'd like | |
5069 | * to use for memory access. We do this by specifying the IA frequency | |
5070 | * the PCU should use as a reference to determine the ring frequency. | |
5071 | */ | |
4c8c7743 AG |
5072 | for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) { |
5073 | int diff = max_gpu_freq - gpu_freq; | |
3ebecd07 CW |
5074 | unsigned int ia_freq = 0, ring_freq = 0; |
5075 | ||
4c8c7743 AG |
5076 | if (IS_SKYLAKE(dev)) { |
5077 | /* | |
5078 | * ring_freq = 2 * GT. ring_freq is in 100MHz units | |
5079 | * No floor required for ring frequency on SKL. | |
5080 | */ | |
5081 | ring_freq = gpu_freq; | |
5082 | } else if (INTEL_INFO(dev)->gen >= 8) { | |
46c764d4 BW |
5083 | /* max(2 * GT, DDR). NB: GT is 50MHz units */ |
5084 | ring_freq = max(min_ring_freq, gpu_freq); | |
5085 | } else if (IS_HASWELL(dev)) { | |
f6aca45c | 5086 | ring_freq = mult_frac(gpu_freq, 5, 4); |
3ebecd07 CW |
5087 | ring_freq = max(min_ring_freq, ring_freq); |
5088 | /* leave ia_freq as the default, chosen by cpufreq */ | |
5089 | } else { | |
5090 | /* On older processors, there is no separate ring | |
5091 | * clock domain, so in order to boost the bandwidth | |
5092 | * of the ring, we need to upclock the CPU (ia_freq). | |
5093 | * | |
5094 | * For GPU frequencies less than 750MHz, | |
5095 | * just use the lowest ring freq. | |
5096 | */ | |
5097 | if (gpu_freq < min_freq) | |
5098 | ia_freq = 800; | |
5099 | else | |
5100 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); | |
5101 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); | |
5102 | } | |
2b4e57bd | 5103 | |
42c0526c BW |
5104 | sandybridge_pcode_write(dev_priv, |
5105 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, | |
3ebecd07 CW |
5106 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
5107 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | | |
5108 | gpu_freq); | |
2b4e57bd | 5109 | } |
2b4e57bd ED |
5110 | } |
5111 | ||
c2bc2fc5 ID |
5112 | void gen6_update_ring_freq(struct drm_device *dev) |
5113 | { | |
5114 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5115 | ||
97d3308a | 5116 | if (!HAS_CORE_RING_FREQ(dev)) |
c2bc2fc5 ID |
5117 | return; |
5118 | ||
5119 | mutex_lock(&dev_priv->rps.hw_lock); | |
5120 | __gen6_update_ring_freq(dev); | |
5121 | mutex_unlock(&dev_priv->rps.hw_lock); | |
5122 | } | |
5123 | ||
03af2045 | 5124 | static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv) |
2b6b3a09 | 5125 | { |
095acd5f | 5126 | struct drm_device *dev = dev_priv->dev; |
2b6b3a09 D |
5127 | u32 val, rp0; |
5128 | ||
095acd5f D |
5129 | if (dev->pdev->revision >= 0x20) { |
5130 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); | |
2b6b3a09 | 5131 | |
095acd5f D |
5132 | switch (INTEL_INFO(dev)->eu_total) { |
5133 | case 8: | |
5134 | /* (2 * 4) config */ | |
5135 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT); | |
5136 | break; | |
5137 | case 12: | |
5138 | /* (2 * 6) config */ | |
5139 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT); | |
5140 | break; | |
5141 | case 16: | |
5142 | /* (2 * 8) config */ | |
5143 | default: | |
5144 | /* Setting (2 * 8) Min RP0 for any other combination */ | |
5145 | rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT); | |
5146 | break; | |
5147 | } | |
5148 | rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK); | |
5149 | } else { | |
5150 | /* For pre-production hardware */ | |
5151 | val = vlv_punit_read(dev_priv, PUNIT_GPU_STATUS_REG); | |
5152 | rp0 = (val >> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT) & | |
5153 | PUNIT_GPU_STATUS_MAX_FREQ_MASK; | |
5154 | } | |
2b6b3a09 D |
5155 | return rp0; |
5156 | } | |
5157 | ||
5158 | static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
5159 | { | |
5160 | u32 val, rpe; | |
5161 | ||
5162 | val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG); | |
5163 | rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK; | |
5164 | ||
5165 | return rpe; | |
5166 | } | |
5167 | ||
7707df4a D |
5168 | static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5169 | { | |
095acd5f | 5170 | struct drm_device *dev = dev_priv->dev; |
7707df4a D |
5171 | u32 val, rp1; |
5172 | ||
095acd5f D |
5173 | if (dev->pdev->revision >= 0x20) { |
5174 | val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE); | |
5175 | rp1 = (val & FB_GFX_FREQ_FUSE_MASK); | |
5176 | } else { | |
5177 | /* For pre-production hardware */ | |
5178 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); | |
5179 | rp1 = ((val >> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT) & | |
5180 | PUNIT_GPU_STATUS_MAX_FREQ_MASK); | |
5181 | } | |
7707df4a D |
5182 | return rp1; |
5183 | } | |
5184 | ||
f8f2b001 D |
5185 | static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv) |
5186 | { | |
5187 | u32 val, rp1; | |
5188 | ||
5189 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); | |
5190 | ||
5191 | rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT; | |
5192 | ||
5193 | return rp1; | |
5194 | } | |
5195 | ||
03af2045 | 5196 | static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv) |
0a073b84 JB |
5197 | { |
5198 | u32 val, rp0; | |
5199 | ||
64936258 | 5200 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
0a073b84 JB |
5201 | |
5202 | rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT; | |
5203 | /* Clamp to max */ | |
5204 | rp0 = min_t(u32, rp0, 0xea); | |
5205 | ||
5206 | return rp0; | |
5207 | } | |
5208 | ||
5209 | static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
5210 | { | |
5211 | u32 val, rpe; | |
5212 | ||
64936258 | 5213 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO); |
0a073b84 | 5214 | rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT; |
64936258 | 5215 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI); |
0a073b84 JB |
5216 | rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5; |
5217 | ||
5218 | return rpe; | |
5219 | } | |
5220 | ||
03af2045 | 5221 | static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv) |
0a073b84 | 5222 | { |
64936258 | 5223 | return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff; |
0a073b84 JB |
5224 | } |
5225 | ||
ae48434c ID |
5226 | /* Check that the pctx buffer wasn't move under us. */ |
5227 | static void valleyview_check_pctx(struct drm_i915_private *dev_priv) | |
5228 | { | |
5229 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; | |
5230 | ||
5231 | WARN_ON(pctx_addr != dev_priv->mm.stolen_base + | |
5232 | dev_priv->vlv_pctx->stolen->start); | |
5233 | } | |
5234 | ||
38807746 D |
5235 | |
5236 | /* Check that the pcbr address is not empty. */ | |
5237 | static void cherryview_check_pctx(struct drm_i915_private *dev_priv) | |
5238 | { | |
5239 | unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095; | |
5240 | ||
5241 | WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0); | |
5242 | } | |
5243 | ||
5244 | static void cherryview_setup_pctx(struct drm_device *dev) | |
5245 | { | |
5246 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5247 | unsigned long pctx_paddr, paddr; | |
5248 | struct i915_gtt *gtt = &dev_priv->gtt; | |
5249 | u32 pcbr; | |
5250 | int pctx_size = 32*1024; | |
5251 | ||
5252 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); | |
5253 | ||
5254 | pcbr = I915_READ(VLV_PCBR); | |
5255 | if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) { | |
ce611ef8 | 5256 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
38807746 D |
5257 | paddr = (dev_priv->mm.stolen_base + |
5258 | (gtt->stolen_size - pctx_size)); | |
5259 | ||
5260 | pctx_paddr = (paddr & (~4095)); | |
5261 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
5262 | } | |
ce611ef8 VS |
5263 | |
5264 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); | |
38807746 D |
5265 | } |
5266 | ||
c9cddffc JB |
5267 | static void valleyview_setup_pctx(struct drm_device *dev) |
5268 | { | |
5269 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5270 | struct drm_i915_gem_object *pctx; | |
5271 | unsigned long pctx_paddr; | |
5272 | u32 pcbr; | |
5273 | int pctx_size = 24*1024; | |
5274 | ||
17b0c1f7 ID |
5275 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
5276 | ||
c9cddffc JB |
5277 | pcbr = I915_READ(VLV_PCBR); |
5278 | if (pcbr) { | |
5279 | /* BIOS set it up already, grab the pre-alloc'd space */ | |
5280 | int pcbr_offset; | |
5281 | ||
5282 | pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base; | |
5283 | pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev, | |
5284 | pcbr_offset, | |
190d6cd5 | 5285 | I915_GTT_OFFSET_NONE, |
c9cddffc JB |
5286 | pctx_size); |
5287 | goto out; | |
5288 | } | |
5289 | ||
ce611ef8 VS |
5290 | DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n"); |
5291 | ||
c9cddffc JB |
5292 | /* |
5293 | * From the Gunit register HAS: | |
5294 | * The Gfx driver is expected to program this register and ensure | |
5295 | * proper allocation within Gfx stolen memory. For example, this | |
5296 | * register should be programmed such than the PCBR range does not | |
5297 | * overlap with other ranges, such as the frame buffer, protected | |
5298 | * memory, or any other relevant ranges. | |
5299 | */ | |
5300 | pctx = i915_gem_object_create_stolen(dev, pctx_size); | |
5301 | if (!pctx) { | |
5302 | DRM_DEBUG("not enough stolen space for PCTX, disabling\n"); | |
5303 | return; | |
5304 | } | |
5305 | ||
5306 | pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start; | |
5307 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
5308 | ||
5309 | out: | |
ce611ef8 | 5310 | DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR)); |
c9cddffc JB |
5311 | dev_priv->vlv_pctx = pctx; |
5312 | } | |
5313 | ||
ae48434c ID |
5314 | static void valleyview_cleanup_pctx(struct drm_device *dev) |
5315 | { | |
5316 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5317 | ||
5318 | if (WARN_ON(!dev_priv->vlv_pctx)) | |
5319 | return; | |
5320 | ||
5321 | drm_gem_object_unreference(&dev_priv->vlv_pctx->base); | |
5322 | dev_priv->vlv_pctx = NULL; | |
5323 | } | |
5324 | ||
4e80519e ID |
5325 | static void valleyview_init_gt_powersave(struct drm_device *dev) |
5326 | { | |
5327 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2bb25c17 | 5328 | u32 val; |
4e80519e ID |
5329 | |
5330 | valleyview_setup_pctx(dev); | |
5331 | ||
5332 | mutex_lock(&dev_priv->rps.hw_lock); | |
5333 | ||
2bb25c17 VS |
5334 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5335 | switch ((val >> 6) & 3) { | |
5336 | case 0: | |
5337 | case 1: | |
5338 | dev_priv->mem_freq = 800; | |
5339 | break; | |
5340 | case 2: | |
5341 | dev_priv->mem_freq = 1066; | |
5342 | break; | |
5343 | case 3: | |
5344 | dev_priv->mem_freq = 1333; | |
5345 | break; | |
5346 | } | |
80b83b62 | 5347 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
2bb25c17 | 5348 | |
4e80519e ID |
5349 | dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv); |
5350 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; | |
5351 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5352 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
4e80519e ID |
5353 | dev_priv->rps.max_freq); |
5354 | ||
5355 | dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv); | |
5356 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5357 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
4e80519e ID |
5358 | dev_priv->rps.efficient_freq); |
5359 | ||
f8f2b001 D |
5360 | dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv); |
5361 | DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5362 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
f8f2b001 D |
5363 | dev_priv->rps.rp1_freq); |
5364 | ||
4e80519e ID |
5365 | dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv); |
5366 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5367 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
4e80519e ID |
5368 | dev_priv->rps.min_freq); |
5369 | ||
aed242ff CW |
5370 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; |
5371 | ||
4e80519e ID |
5372 | /* Preserve min/max settings in case of re-init */ |
5373 | if (dev_priv->rps.max_freq_softlimit == 0) | |
5374 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; | |
5375 | ||
5376 | if (dev_priv->rps.min_freq_softlimit == 0) | |
5377 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; | |
5378 | ||
5379 | mutex_unlock(&dev_priv->rps.hw_lock); | |
5380 | } | |
5381 | ||
38807746 D |
5382 | static void cherryview_init_gt_powersave(struct drm_device *dev) |
5383 | { | |
2b6b3a09 | 5384 | struct drm_i915_private *dev_priv = dev->dev_private; |
2bb25c17 | 5385 | u32 val; |
2b6b3a09 | 5386 | |
38807746 | 5387 | cherryview_setup_pctx(dev); |
2b6b3a09 D |
5388 | |
5389 | mutex_lock(&dev_priv->rps.hw_lock); | |
5390 | ||
a580516d | 5391 | mutex_lock(&dev_priv->sb_lock); |
c6e8f39d | 5392 | val = vlv_cck_read(dev_priv, CCK_FUSE_REG); |
a580516d | 5393 | mutex_unlock(&dev_priv->sb_lock); |
c6e8f39d | 5394 | |
2bb25c17 VS |
5395 | switch ((val >> 2) & 0x7) { |
5396 | case 0: | |
5397 | case 1: | |
5398 | dev_priv->rps.cz_freq = 200; | |
5399 | dev_priv->mem_freq = 1600; | |
5400 | break; | |
5401 | case 2: | |
5402 | dev_priv->rps.cz_freq = 267; | |
5403 | dev_priv->mem_freq = 1600; | |
5404 | break; | |
5405 | case 3: | |
5406 | dev_priv->rps.cz_freq = 333; | |
5407 | dev_priv->mem_freq = 2000; | |
5408 | break; | |
5409 | case 4: | |
5410 | dev_priv->rps.cz_freq = 320; | |
5411 | dev_priv->mem_freq = 1600; | |
5412 | break; | |
5413 | case 5: | |
5414 | dev_priv->rps.cz_freq = 400; | |
5415 | dev_priv->mem_freq = 1600; | |
5416 | break; | |
5417 | } | |
80b83b62 | 5418 | DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq); |
2bb25c17 | 5419 | |
2b6b3a09 D |
5420 | dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv); |
5421 | dev_priv->rps.rp0_freq = dev_priv->rps.max_freq; | |
5422 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5423 | intel_gpu_freq(dev_priv, dev_priv->rps.max_freq), |
2b6b3a09 D |
5424 | dev_priv->rps.max_freq); |
5425 | ||
5426 | dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv); | |
5427 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5428 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
2b6b3a09 D |
5429 | dev_priv->rps.efficient_freq); |
5430 | ||
7707df4a D |
5431 | dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv); |
5432 | DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5433 | intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq), |
7707df4a D |
5434 | dev_priv->rps.rp1_freq); |
5435 | ||
5b7c91b7 D |
5436 | /* PUnit validated range is only [RPe, RP0] */ |
5437 | dev_priv->rps.min_freq = dev_priv->rps.efficient_freq; | |
2b6b3a09 | 5438 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", |
7c59a9c1 | 5439 | intel_gpu_freq(dev_priv, dev_priv->rps.min_freq), |
2b6b3a09 D |
5440 | dev_priv->rps.min_freq); |
5441 | ||
1c14762d VS |
5442 | WARN_ONCE((dev_priv->rps.max_freq | |
5443 | dev_priv->rps.efficient_freq | | |
5444 | dev_priv->rps.rp1_freq | | |
5445 | dev_priv->rps.min_freq) & 1, | |
5446 | "Odd GPU freq values\n"); | |
5447 | ||
aed242ff CW |
5448 | dev_priv->rps.idle_freq = dev_priv->rps.min_freq; |
5449 | ||
2b6b3a09 D |
5450 | /* Preserve min/max settings in case of re-init */ |
5451 | if (dev_priv->rps.max_freq_softlimit == 0) | |
5452 | dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq; | |
5453 | ||
5454 | if (dev_priv->rps.min_freq_softlimit == 0) | |
5455 | dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq; | |
5456 | ||
5457 | mutex_unlock(&dev_priv->rps.hw_lock); | |
38807746 D |
5458 | } |
5459 | ||
4e80519e ID |
5460 | static void valleyview_cleanup_gt_powersave(struct drm_device *dev) |
5461 | { | |
5462 | valleyview_cleanup_pctx(dev); | |
5463 | } | |
5464 | ||
38807746 D |
5465 | static void cherryview_enable_rps(struct drm_device *dev) |
5466 | { | |
5467 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5468 | struct intel_engine_cs *ring; | |
2b6b3a09 | 5469 | u32 gtfifodbg, val, rc6_mode = 0, pcbr; |
38807746 D |
5470 | int i; |
5471 | ||
5472 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
5473 | ||
5474 | gtfifodbg = I915_READ(GTFIFODBG); | |
5475 | if (gtfifodbg) { | |
5476 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", | |
5477 | gtfifodbg); | |
5478 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
5479 | } | |
5480 | ||
5481 | cherryview_check_pctx(dev_priv); | |
5482 | ||
5483 | /* 1a & 1b: Get forcewake during program sequence. Although the driver | |
5484 | * hasn't enabled a state yet where we need forcewake, BIOS may have.*/ | |
59bad947 | 5485 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
38807746 | 5486 | |
160614a2 VS |
5487 | /* Disable RC states. */ |
5488 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
5489 | ||
38807746 D |
5490 | /* 2a: Program RC6 thresholds.*/ |
5491 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16); | |
5492 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */ | |
5493 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */ | |
5494 | ||
5495 | for_each_ring(ring, dev_priv, i) | |
5496 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
5497 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
5498 | ||
f4f71c7d D |
5499 | /* TO threshold set to 500 us ( 0x186 * 1.28 us) */ |
5500 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x186); | |
38807746 D |
5501 | |
5502 | /* allows RC6 residency counter to work */ | |
5503 | I915_WRITE(VLV_COUNTER_CONTROL, | |
5504 | _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH | | |
5505 | VLV_MEDIA_RC6_COUNT_EN | | |
5506 | VLV_RENDER_RC6_COUNT_EN)); | |
5507 | ||
5508 | /* For now we assume BIOS is allocating and populating the PCBR */ | |
5509 | pcbr = I915_READ(VLV_PCBR); | |
5510 | ||
38807746 D |
5511 | /* 3: Enable RC6 */ |
5512 | if ((intel_enable_rc6(dev) & INTEL_RC6_ENABLE) && | |
5513 | (pcbr >> VLV_PCBR_ADDR_SHIFT)) | |
af5a75a3 | 5514 | rc6_mode = GEN7_RC_CTL_TO_MODE; |
38807746 D |
5515 | |
5516 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); | |
5517 | ||
2b6b3a09 | 5518 | /* 4 Program defaults and thresholds for RPS*/ |
3cbdb48f | 5519 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
2b6b3a09 D |
5520 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
5521 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
5522 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
5523 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
5524 | ||
5525 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
5526 | ||
5527 | /* 5: Enable RPS */ | |
5528 | I915_WRITE(GEN6_RP_CONTROL, | |
5529 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
eb973a5e | 5530 | GEN6_RP_MEDIA_IS_GFX | |
2b6b3a09 D |
5531 | GEN6_RP_ENABLE | |
5532 | GEN6_RP_UP_BUSY_AVG | | |
5533 | GEN6_RP_DOWN_IDLE_AVG); | |
5534 | ||
3ef62342 D |
5535 | /* Setting Fixed Bias */ |
5536 | val = VLV_OVERRIDE_EN | | |
5537 | VLV_SOC_TDP_EN | | |
5538 | CHV_BIAS_CPU_50_SOC_50; | |
5539 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); | |
5540 | ||
2b6b3a09 D |
5541 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
5542 | ||
8d40c3ae VS |
5543 | /* RPS code assumes GPLL is used */ |
5544 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); | |
5545 | ||
c8e9627d | 5546 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & GPLLENABLE ? "yes" : "no"); |
2b6b3a09 D |
5547 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
5548 | ||
5549 | dev_priv->rps.cur_freq = (val >> 8) & 0xff; | |
5550 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", | |
7c59a9c1 | 5551 | intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq), |
2b6b3a09 D |
5552 | dev_priv->rps.cur_freq); |
5553 | ||
5554 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", | |
7c59a9c1 | 5555 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
2b6b3a09 D |
5556 | dev_priv->rps.efficient_freq); |
5557 | ||
5558 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq); | |
5559 | ||
59bad947 | 5560 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
38807746 D |
5561 | } |
5562 | ||
0a073b84 JB |
5563 | static void valleyview_enable_rps(struct drm_device *dev) |
5564 | { | |
5565 | struct drm_i915_private *dev_priv = dev->dev_private; | |
a4872ba6 | 5566 | struct intel_engine_cs *ring; |
2a5913a8 | 5567 | u32 gtfifodbg, val, rc6_mode = 0; |
0a073b84 JB |
5568 | int i; |
5569 | ||
5570 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
5571 | ||
ae48434c ID |
5572 | valleyview_check_pctx(dev_priv); |
5573 | ||
0a073b84 | 5574 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { |
f7d85c1e JB |
5575 | DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n", |
5576 | gtfifodbg); | |
0a073b84 JB |
5577 | I915_WRITE(GTFIFODBG, gtfifodbg); |
5578 | } | |
5579 | ||
c8d9a590 | 5580 | /* If VLV, Forcewake all wells, else re-direct to regular path */ |
59bad947 | 5581 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
0a073b84 | 5582 | |
160614a2 VS |
5583 | /* Disable RC states. */ |
5584 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
5585 | ||
cad725fe | 5586 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); |
0a073b84 JB |
5587 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
5588 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
5589 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
5590 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
5591 | ||
5592 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
5593 | ||
5594 | I915_WRITE(GEN6_RP_CONTROL, | |
5595 | GEN6_RP_MEDIA_TURBO | | |
5596 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
5597 | GEN6_RP_MEDIA_IS_GFX | | |
5598 | GEN6_RP_ENABLE | | |
5599 | GEN6_RP_UP_BUSY_AVG | | |
5600 | GEN6_RP_DOWN_IDLE_CONT); | |
5601 | ||
5602 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000); | |
5603 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
5604 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
5605 | ||
5606 | for_each_ring(ring, dev_priv, i) | |
5607 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
5608 | ||
2f0aa304 | 5609 | I915_WRITE(GEN6_RC6_THRESHOLD, 0x557); |
0a073b84 JB |
5610 | |
5611 | /* allows RC6 residency counter to work */ | |
49798eb2 | 5612 | I915_WRITE(VLV_COUNTER_CONTROL, |
31685c25 D |
5613 | _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN | |
5614 | VLV_RENDER_RC0_COUNT_EN | | |
49798eb2 JB |
5615 | VLV_MEDIA_RC6_COUNT_EN | |
5616 | VLV_RENDER_RC6_COUNT_EN)); | |
31685c25 | 5617 | |
a2b23fe0 | 5618 | if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE) |
6b88f295 | 5619 | rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL; |
dc39fff7 BW |
5620 | |
5621 | intel_print_rc6_info(dev, rc6_mode); | |
5622 | ||
a2b23fe0 | 5623 | I915_WRITE(GEN6_RC_CONTROL, rc6_mode); |
0a073b84 | 5624 | |
3ef62342 D |
5625 | /* Setting Fixed Bias */ |
5626 | val = VLV_OVERRIDE_EN | | |
5627 | VLV_SOC_TDP_EN | | |
5628 | VLV_BIAS_CPU_125_SOC_875; | |
5629 | vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val); | |
5630 | ||
64936258 | 5631 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
0a073b84 | 5632 | |
8d40c3ae VS |
5633 | /* RPS code assumes GPLL is used */ |
5634 | WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n"); | |
5635 | ||
c8e9627d | 5636 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & GPLLENABLE ? "yes" : "no"); |
0a073b84 JB |
5637 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); |
5638 | ||
b39fb297 | 5639 | dev_priv->rps.cur_freq = (val >> 8) & 0xff; |
73008b98 | 5640 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", |
7c59a9c1 | 5641 | intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq), |
b39fb297 | 5642 | dev_priv->rps.cur_freq); |
0a073b84 | 5643 | |
73008b98 | 5644 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", |
7c59a9c1 | 5645 | intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq), |
b39fb297 | 5646 | dev_priv->rps.efficient_freq); |
0a073b84 | 5647 | |
b39fb297 | 5648 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq); |
0a073b84 | 5649 | |
59bad947 | 5650 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
0a073b84 JB |
5651 | } |
5652 | ||
dde18883 ED |
5653 | static unsigned long intel_pxfreq(u32 vidfreq) |
5654 | { | |
5655 | unsigned long freq; | |
5656 | int div = (vidfreq & 0x3f0000) >> 16; | |
5657 | int post = (vidfreq & 0x3000) >> 12; | |
5658 | int pre = (vidfreq & 0x7); | |
5659 | ||
5660 | if (!pre) | |
5661 | return 0; | |
5662 | ||
5663 | freq = ((div * 133333) / ((1<<post) * pre)); | |
5664 | ||
5665 | return freq; | |
5666 | } | |
5667 | ||
eb48eb00 DV |
5668 | static const struct cparams { |
5669 | u16 i; | |
5670 | u16 t; | |
5671 | u16 m; | |
5672 | u16 c; | |
5673 | } cparams[] = { | |
5674 | { 1, 1333, 301, 28664 }, | |
5675 | { 1, 1066, 294, 24460 }, | |
5676 | { 1, 800, 294, 25192 }, | |
5677 | { 0, 1333, 276, 27605 }, | |
5678 | { 0, 1066, 276, 27605 }, | |
5679 | { 0, 800, 231, 23784 }, | |
5680 | }; | |
5681 | ||
f531dcb2 | 5682 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
5683 | { |
5684 | u64 total_count, diff, ret; | |
5685 | u32 count1, count2, count3, m = 0, c = 0; | |
5686 | unsigned long now = jiffies_to_msecs(jiffies), diff1; | |
5687 | int i; | |
5688 | ||
02d71956 DV |
5689 | assert_spin_locked(&mchdev_lock); |
5690 | ||
20e4d407 | 5691 | diff1 = now - dev_priv->ips.last_time1; |
eb48eb00 DV |
5692 | |
5693 | /* Prevent division-by-zero if we are asking too fast. | |
5694 | * Also, we don't get interesting results if we are polling | |
5695 | * faster than once in 10ms, so just return the saved value | |
5696 | * in such cases. | |
5697 | */ | |
5698 | if (diff1 <= 10) | |
20e4d407 | 5699 | return dev_priv->ips.chipset_power; |
eb48eb00 DV |
5700 | |
5701 | count1 = I915_READ(DMIEC); | |
5702 | count2 = I915_READ(DDREC); | |
5703 | count3 = I915_READ(CSIEC); | |
5704 | ||
5705 | total_count = count1 + count2 + count3; | |
5706 | ||
5707 | /* FIXME: handle per-counter overflow */ | |
20e4d407 DV |
5708 | if (total_count < dev_priv->ips.last_count1) { |
5709 | diff = ~0UL - dev_priv->ips.last_count1; | |
eb48eb00 DV |
5710 | diff += total_count; |
5711 | } else { | |
20e4d407 | 5712 | diff = total_count - dev_priv->ips.last_count1; |
eb48eb00 DV |
5713 | } |
5714 | ||
5715 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { | |
20e4d407 DV |
5716 | if (cparams[i].i == dev_priv->ips.c_m && |
5717 | cparams[i].t == dev_priv->ips.r_t) { | |
eb48eb00 DV |
5718 | m = cparams[i].m; |
5719 | c = cparams[i].c; | |
5720 | break; | |
5721 | } | |
5722 | } | |
5723 | ||
5724 | diff = div_u64(diff, diff1); | |
5725 | ret = ((m * diff) + c); | |
5726 | ret = div_u64(ret, 10); | |
5727 | ||
20e4d407 DV |
5728 | dev_priv->ips.last_count1 = total_count; |
5729 | dev_priv->ips.last_time1 = now; | |
eb48eb00 | 5730 | |
20e4d407 | 5731 | dev_priv->ips.chipset_power = ret; |
eb48eb00 DV |
5732 | |
5733 | return ret; | |
5734 | } | |
5735 | ||
f531dcb2 CW |
5736 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
5737 | { | |
3d13ef2e | 5738 | struct drm_device *dev = dev_priv->dev; |
f531dcb2 CW |
5739 | unsigned long val; |
5740 | ||
3d13ef2e | 5741 | if (INTEL_INFO(dev)->gen != 5) |
f531dcb2 CW |
5742 | return 0; |
5743 | ||
5744 | spin_lock_irq(&mchdev_lock); | |
5745 | ||
5746 | val = __i915_chipset_val(dev_priv); | |
5747 | ||
5748 | spin_unlock_irq(&mchdev_lock); | |
5749 | ||
5750 | return val; | |
5751 | } | |
5752 | ||
eb48eb00 DV |
5753 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
5754 | { | |
5755 | unsigned long m, x, b; | |
5756 | u32 tsfs; | |
5757 | ||
5758 | tsfs = I915_READ(TSFS); | |
5759 | ||
5760 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); | |
5761 | x = I915_READ8(TR1); | |
5762 | ||
5763 | b = tsfs & TSFS_INTR_MASK; | |
5764 | ||
5765 | return ((m * x) / 127) - b; | |
5766 | } | |
5767 | ||
d972d6ee MK |
5768 | static int _pxvid_to_vd(u8 pxvid) |
5769 | { | |
5770 | if (pxvid == 0) | |
5771 | return 0; | |
5772 | ||
5773 | if (pxvid >= 8 && pxvid < 31) | |
5774 | pxvid = 31; | |
5775 | ||
5776 | return (pxvid + 2) * 125; | |
5777 | } | |
5778 | ||
5779 | static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) | |
eb48eb00 | 5780 | { |
3d13ef2e | 5781 | struct drm_device *dev = dev_priv->dev; |
d972d6ee MK |
5782 | const int vd = _pxvid_to_vd(pxvid); |
5783 | const int vm = vd - 1125; | |
5784 | ||
3d13ef2e | 5785 | if (INTEL_INFO(dev)->is_mobile) |
d972d6ee MK |
5786 | return vm > 0 ? vm : 0; |
5787 | ||
5788 | return vd; | |
eb48eb00 DV |
5789 | } |
5790 | ||
02d71956 | 5791 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 | 5792 | { |
5ed0bdf2 | 5793 | u64 now, diff, diffms; |
eb48eb00 DV |
5794 | u32 count; |
5795 | ||
02d71956 | 5796 | assert_spin_locked(&mchdev_lock); |
eb48eb00 | 5797 | |
5ed0bdf2 TG |
5798 | now = ktime_get_raw_ns(); |
5799 | diffms = now - dev_priv->ips.last_time2; | |
5800 | do_div(diffms, NSEC_PER_MSEC); | |
eb48eb00 DV |
5801 | |
5802 | /* Don't divide by 0 */ | |
eb48eb00 DV |
5803 | if (!diffms) |
5804 | return; | |
5805 | ||
5806 | count = I915_READ(GFXEC); | |
5807 | ||
20e4d407 DV |
5808 | if (count < dev_priv->ips.last_count2) { |
5809 | diff = ~0UL - dev_priv->ips.last_count2; | |
eb48eb00 DV |
5810 | diff += count; |
5811 | } else { | |
20e4d407 | 5812 | diff = count - dev_priv->ips.last_count2; |
eb48eb00 DV |
5813 | } |
5814 | ||
20e4d407 DV |
5815 | dev_priv->ips.last_count2 = count; |
5816 | dev_priv->ips.last_time2 = now; | |
eb48eb00 DV |
5817 | |
5818 | /* More magic constants... */ | |
5819 | diff = diff * 1181; | |
5820 | diff = div_u64(diff, diffms * 10); | |
20e4d407 | 5821 | dev_priv->ips.gfx_power = diff; |
eb48eb00 DV |
5822 | } |
5823 | ||
02d71956 DV |
5824 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
5825 | { | |
3d13ef2e DL |
5826 | struct drm_device *dev = dev_priv->dev; |
5827 | ||
5828 | if (INTEL_INFO(dev)->gen != 5) | |
02d71956 DV |
5829 | return; |
5830 | ||
9270388e | 5831 | spin_lock_irq(&mchdev_lock); |
02d71956 DV |
5832 | |
5833 | __i915_update_gfx_val(dev_priv); | |
5834 | ||
9270388e | 5835 | spin_unlock_irq(&mchdev_lock); |
02d71956 DV |
5836 | } |
5837 | ||
f531dcb2 | 5838 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
5839 | { |
5840 | unsigned long t, corr, state1, corr2, state2; | |
5841 | u32 pxvid, ext_v; | |
5842 | ||
02d71956 DV |
5843 | assert_spin_locked(&mchdev_lock); |
5844 | ||
b39fb297 | 5845 | pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_freq * 4)); |
eb48eb00 DV |
5846 | pxvid = (pxvid >> 24) & 0x7f; |
5847 | ext_v = pvid_to_extvid(dev_priv, pxvid); | |
5848 | ||
5849 | state1 = ext_v; | |
5850 | ||
5851 | t = i915_mch_val(dev_priv); | |
5852 | ||
5853 | /* Revel in the empirically derived constants */ | |
5854 | ||
5855 | /* Correction factor in 1/100000 units */ | |
5856 | if (t > 80) | |
5857 | corr = ((t * 2349) + 135940); | |
5858 | else if (t >= 50) | |
5859 | corr = ((t * 964) + 29317); | |
5860 | else /* < 50 */ | |
5861 | corr = ((t * 301) + 1004); | |
5862 | ||
5863 | corr = corr * ((150142 * state1) / 10000 - 78642); | |
5864 | corr /= 100000; | |
20e4d407 | 5865 | corr2 = (corr * dev_priv->ips.corr); |
eb48eb00 DV |
5866 | |
5867 | state2 = (corr2 * state1) / 10000; | |
5868 | state2 /= 100; /* convert to mW */ | |
5869 | ||
02d71956 | 5870 | __i915_update_gfx_val(dev_priv); |
eb48eb00 | 5871 | |
20e4d407 | 5872 | return dev_priv->ips.gfx_power + state2; |
eb48eb00 DV |
5873 | } |
5874 | ||
f531dcb2 CW |
5875 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
5876 | { | |
3d13ef2e | 5877 | struct drm_device *dev = dev_priv->dev; |
f531dcb2 CW |
5878 | unsigned long val; |
5879 | ||
3d13ef2e | 5880 | if (INTEL_INFO(dev)->gen != 5) |
f531dcb2 CW |
5881 | return 0; |
5882 | ||
5883 | spin_lock_irq(&mchdev_lock); | |
5884 | ||
5885 | val = __i915_gfx_val(dev_priv); | |
5886 | ||
5887 | spin_unlock_irq(&mchdev_lock); | |
5888 | ||
5889 | return val; | |
5890 | } | |
5891 | ||
eb48eb00 DV |
5892 | /** |
5893 | * i915_read_mch_val - return value for IPS use | |
5894 | * | |
5895 | * Calculate and return a value for the IPS driver to use when deciding whether | |
5896 | * we have thermal and power headroom to increase CPU or GPU power budget. | |
5897 | */ | |
5898 | unsigned long i915_read_mch_val(void) | |
5899 | { | |
5900 | struct drm_i915_private *dev_priv; | |
5901 | unsigned long chipset_val, graphics_val, ret = 0; | |
5902 | ||
9270388e | 5903 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
5904 | if (!i915_mch_dev) |
5905 | goto out_unlock; | |
5906 | dev_priv = i915_mch_dev; | |
5907 | ||
f531dcb2 CW |
5908 | chipset_val = __i915_chipset_val(dev_priv); |
5909 | graphics_val = __i915_gfx_val(dev_priv); | |
eb48eb00 DV |
5910 | |
5911 | ret = chipset_val + graphics_val; | |
5912 | ||
5913 | out_unlock: | |
9270388e | 5914 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
5915 | |
5916 | return ret; | |
5917 | } | |
5918 | EXPORT_SYMBOL_GPL(i915_read_mch_val); | |
5919 | ||
5920 | /** | |
5921 | * i915_gpu_raise - raise GPU frequency limit | |
5922 | * | |
5923 | * Raise the limit; IPS indicates we have thermal headroom. | |
5924 | */ | |
5925 | bool i915_gpu_raise(void) | |
5926 | { | |
5927 | struct drm_i915_private *dev_priv; | |
5928 | bool ret = true; | |
5929 | ||
9270388e | 5930 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
5931 | if (!i915_mch_dev) { |
5932 | ret = false; | |
5933 | goto out_unlock; | |
5934 | } | |
5935 | dev_priv = i915_mch_dev; | |
5936 | ||
20e4d407 DV |
5937 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
5938 | dev_priv->ips.max_delay--; | |
eb48eb00 DV |
5939 | |
5940 | out_unlock: | |
9270388e | 5941 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
5942 | |
5943 | return ret; | |
5944 | } | |
5945 | EXPORT_SYMBOL_GPL(i915_gpu_raise); | |
5946 | ||
5947 | /** | |
5948 | * i915_gpu_lower - lower GPU frequency limit | |
5949 | * | |
5950 | * IPS indicates we're close to a thermal limit, so throttle back the GPU | |
5951 | * frequency maximum. | |
5952 | */ | |
5953 | bool i915_gpu_lower(void) | |
5954 | { | |
5955 | struct drm_i915_private *dev_priv; | |
5956 | bool ret = true; | |
5957 | ||
9270388e | 5958 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
5959 | if (!i915_mch_dev) { |
5960 | ret = false; | |
5961 | goto out_unlock; | |
5962 | } | |
5963 | dev_priv = i915_mch_dev; | |
5964 | ||
20e4d407 DV |
5965 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
5966 | dev_priv->ips.max_delay++; | |
eb48eb00 DV |
5967 | |
5968 | out_unlock: | |
9270388e | 5969 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
5970 | |
5971 | return ret; | |
5972 | } | |
5973 | EXPORT_SYMBOL_GPL(i915_gpu_lower); | |
5974 | ||
5975 | /** | |
5976 | * i915_gpu_busy - indicate GPU business to IPS | |
5977 | * | |
5978 | * Tell the IPS driver whether or not the GPU is busy. | |
5979 | */ | |
5980 | bool i915_gpu_busy(void) | |
5981 | { | |
5982 | struct drm_i915_private *dev_priv; | |
a4872ba6 | 5983 | struct intel_engine_cs *ring; |
eb48eb00 | 5984 | bool ret = false; |
f047e395 | 5985 | int i; |
eb48eb00 | 5986 | |
9270388e | 5987 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
5988 | if (!i915_mch_dev) |
5989 | goto out_unlock; | |
5990 | dev_priv = i915_mch_dev; | |
5991 | ||
f047e395 CW |
5992 | for_each_ring(ring, dev_priv, i) |
5993 | ret |= !list_empty(&ring->request_list); | |
eb48eb00 DV |
5994 | |
5995 | out_unlock: | |
9270388e | 5996 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
5997 | |
5998 | return ret; | |
5999 | } | |
6000 | EXPORT_SYMBOL_GPL(i915_gpu_busy); | |
6001 | ||
6002 | /** | |
6003 | * i915_gpu_turbo_disable - disable graphics turbo | |
6004 | * | |
6005 | * Disable graphics turbo by resetting the max frequency and setting the | |
6006 | * current frequency to the default. | |
6007 | */ | |
6008 | bool i915_gpu_turbo_disable(void) | |
6009 | { | |
6010 | struct drm_i915_private *dev_priv; | |
6011 | bool ret = true; | |
6012 | ||
9270388e | 6013 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
6014 | if (!i915_mch_dev) { |
6015 | ret = false; | |
6016 | goto out_unlock; | |
6017 | } | |
6018 | dev_priv = i915_mch_dev; | |
6019 | ||
20e4d407 | 6020 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
eb48eb00 | 6021 | |
20e4d407 | 6022 | if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart)) |
eb48eb00 DV |
6023 | ret = false; |
6024 | ||
6025 | out_unlock: | |
9270388e | 6026 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6027 | |
6028 | return ret; | |
6029 | } | |
6030 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); | |
6031 | ||
6032 | /** | |
6033 | * Tells the intel_ips driver that the i915 driver is now loaded, if | |
6034 | * IPS got loaded first. | |
6035 | * | |
6036 | * This awkward dance is so that neither module has to depend on the | |
6037 | * other in order for IPS to do the appropriate communication of | |
6038 | * GPU turbo limits to i915. | |
6039 | */ | |
6040 | static void | |
6041 | ips_ping_for_i915_load(void) | |
6042 | { | |
6043 | void (*link)(void); | |
6044 | ||
6045 | link = symbol_get(ips_link_to_i915_driver); | |
6046 | if (link) { | |
6047 | link(); | |
6048 | symbol_put(ips_link_to_i915_driver); | |
6049 | } | |
6050 | } | |
6051 | ||
6052 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) | |
6053 | { | |
02d71956 DV |
6054 | /* We only register the i915 ips part with intel-ips once everything is |
6055 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ | |
9270388e | 6056 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 6057 | i915_mch_dev = dev_priv; |
9270388e | 6058 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
6059 | |
6060 | ips_ping_for_i915_load(); | |
6061 | } | |
6062 | ||
6063 | void intel_gpu_ips_teardown(void) | |
6064 | { | |
9270388e | 6065 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 6066 | i915_mch_dev = NULL; |
9270388e | 6067 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 | 6068 | } |
76c3552f | 6069 | |
8090c6b9 | 6070 | static void intel_init_emon(struct drm_device *dev) |
dde18883 ED |
6071 | { |
6072 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6073 | u32 lcfuse; | |
6074 | u8 pxw[16]; | |
6075 | int i; | |
6076 | ||
6077 | /* Disable to program */ | |
6078 | I915_WRITE(ECR, 0); | |
6079 | POSTING_READ(ECR); | |
6080 | ||
6081 | /* Program energy weights for various events */ | |
6082 | I915_WRITE(SDEW, 0x15040d00); | |
6083 | I915_WRITE(CSIEW0, 0x007f0000); | |
6084 | I915_WRITE(CSIEW1, 0x1e220004); | |
6085 | I915_WRITE(CSIEW2, 0x04000004); | |
6086 | ||
6087 | for (i = 0; i < 5; i++) | |
6088 | I915_WRITE(PEW + (i * 4), 0); | |
6089 | for (i = 0; i < 3; i++) | |
6090 | I915_WRITE(DEW + (i * 4), 0); | |
6091 | ||
6092 | /* Program P-state weights to account for frequency power adjustment */ | |
6093 | for (i = 0; i < 16; i++) { | |
6094 | u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4)); | |
6095 | unsigned long freq = intel_pxfreq(pxvidfreq); | |
6096 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> | |
6097 | PXVFREQ_PX_SHIFT; | |
6098 | unsigned long val; | |
6099 | ||
6100 | val = vid * vid; | |
6101 | val *= (freq / 1000); | |
6102 | val *= 255; | |
6103 | val /= (127*127*900); | |
6104 | if (val > 0xff) | |
6105 | DRM_ERROR("bad pxval: %ld\n", val); | |
6106 | pxw[i] = val; | |
6107 | } | |
6108 | /* Render standby states get 0 weight */ | |
6109 | pxw[14] = 0; | |
6110 | pxw[15] = 0; | |
6111 | ||
6112 | for (i = 0; i < 4; i++) { | |
6113 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | | |
6114 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); | |
6115 | I915_WRITE(PXW + (i * 4), val); | |
6116 | } | |
6117 | ||
6118 | /* Adjust magic regs to magic values (more experimental results) */ | |
6119 | I915_WRITE(OGW0, 0); | |
6120 | I915_WRITE(OGW1, 0); | |
6121 | I915_WRITE(EG0, 0x00007f00); | |
6122 | I915_WRITE(EG1, 0x0000000e); | |
6123 | I915_WRITE(EG2, 0x000e0000); | |
6124 | I915_WRITE(EG3, 0x68000300); | |
6125 | I915_WRITE(EG4, 0x42000000); | |
6126 | I915_WRITE(EG5, 0x00140031); | |
6127 | I915_WRITE(EG6, 0); | |
6128 | I915_WRITE(EG7, 0); | |
6129 | ||
6130 | for (i = 0; i < 8; i++) | |
6131 | I915_WRITE(PXWL + (i * 4), 0); | |
6132 | ||
6133 | /* Enable PMON + select events */ | |
6134 | I915_WRITE(ECR, 0x80000019); | |
6135 | ||
6136 | lcfuse = I915_READ(LCFUSE02); | |
6137 | ||
20e4d407 | 6138 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
dde18883 ED |
6139 | } |
6140 | ||
ae48434c ID |
6141 | void intel_init_gt_powersave(struct drm_device *dev) |
6142 | { | |
e6069ca8 ID |
6143 | i915.enable_rc6 = sanitize_rc6_option(dev, i915.enable_rc6); |
6144 | ||
38807746 D |
6145 | if (IS_CHERRYVIEW(dev)) |
6146 | cherryview_init_gt_powersave(dev); | |
6147 | else if (IS_VALLEYVIEW(dev)) | |
4e80519e | 6148 | valleyview_init_gt_powersave(dev); |
ae48434c ID |
6149 | } |
6150 | ||
6151 | void intel_cleanup_gt_powersave(struct drm_device *dev) | |
6152 | { | |
38807746 D |
6153 | if (IS_CHERRYVIEW(dev)) |
6154 | return; | |
6155 | else if (IS_VALLEYVIEW(dev)) | |
4e80519e | 6156 | valleyview_cleanup_gt_powersave(dev); |
ae48434c ID |
6157 | } |
6158 | ||
dbea3cea ID |
6159 | static void gen6_suspend_rps(struct drm_device *dev) |
6160 | { | |
6161 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6162 | ||
6163 | flush_delayed_work(&dev_priv->rps.delayed_resume_work); | |
6164 | ||
4c2a8897 | 6165 | gen6_disable_rps_interrupts(dev); |
dbea3cea ID |
6166 | } |
6167 | ||
156c7ca0 JB |
6168 | /** |
6169 | * intel_suspend_gt_powersave - suspend PM work and helper threads | |
6170 | * @dev: drm device | |
6171 | * | |
6172 | * We don't want to disable RC6 or other features here, we just want | |
6173 | * to make sure any work we've queued has finished and won't bother | |
6174 | * us while we're suspended. | |
6175 | */ | |
6176 | void intel_suspend_gt_powersave(struct drm_device *dev) | |
6177 | { | |
6178 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6179 | ||
d4d70aa5 ID |
6180 | if (INTEL_INFO(dev)->gen < 6) |
6181 | return; | |
6182 | ||
dbea3cea | 6183 | gen6_suspend_rps(dev); |
b47adc17 D |
6184 | |
6185 | /* Force GPU to min freq during suspend */ | |
6186 | gen6_rps_idle(dev_priv); | |
156c7ca0 JB |
6187 | } |
6188 | ||
8090c6b9 DV |
6189 | void intel_disable_gt_powersave(struct drm_device *dev) |
6190 | { | |
1a01ab3b JB |
6191 | struct drm_i915_private *dev_priv = dev->dev_private; |
6192 | ||
930ebb46 | 6193 | if (IS_IRONLAKE_M(dev)) { |
8090c6b9 | 6194 | ironlake_disable_drps(dev); |
38807746 | 6195 | } else if (INTEL_INFO(dev)->gen >= 6) { |
10d8d366 | 6196 | intel_suspend_gt_powersave(dev); |
e494837a | 6197 | |
4fc688ce | 6198 | mutex_lock(&dev_priv->rps.hw_lock); |
20e49366 ZW |
6199 | if (INTEL_INFO(dev)->gen >= 9) |
6200 | gen9_disable_rps(dev); | |
6201 | else if (IS_CHERRYVIEW(dev)) | |
38807746 D |
6202 | cherryview_disable_rps(dev); |
6203 | else if (IS_VALLEYVIEW(dev)) | |
d20d4f0c JB |
6204 | valleyview_disable_rps(dev); |
6205 | else | |
6206 | gen6_disable_rps(dev); | |
e534770a | 6207 | |
c0951f0c | 6208 | dev_priv->rps.enabled = false; |
4fc688ce | 6209 | mutex_unlock(&dev_priv->rps.hw_lock); |
930ebb46 | 6210 | } |
8090c6b9 DV |
6211 | } |
6212 | ||
1a01ab3b JB |
6213 | static void intel_gen6_powersave_work(struct work_struct *work) |
6214 | { | |
6215 | struct drm_i915_private *dev_priv = | |
6216 | container_of(work, struct drm_i915_private, | |
6217 | rps.delayed_resume_work.work); | |
6218 | struct drm_device *dev = dev_priv->dev; | |
6219 | ||
4fc688ce | 6220 | mutex_lock(&dev_priv->rps.hw_lock); |
0a073b84 | 6221 | |
4c2a8897 | 6222 | gen6_reset_rps_interrupts(dev); |
3cc134e3 | 6223 | |
38807746 D |
6224 | if (IS_CHERRYVIEW(dev)) { |
6225 | cherryview_enable_rps(dev); | |
6226 | } else if (IS_VALLEYVIEW(dev)) { | |
0a073b84 | 6227 | valleyview_enable_rps(dev); |
20e49366 | 6228 | } else if (INTEL_INFO(dev)->gen >= 9) { |
b6fef0ef | 6229 | gen9_enable_rc6(dev); |
20e49366 | 6230 | gen9_enable_rps(dev); |
cc017fb4 AG |
6231 | if (IS_SKYLAKE(dev)) |
6232 | __gen6_update_ring_freq(dev); | |
6edee7f3 BW |
6233 | } else if (IS_BROADWELL(dev)) { |
6234 | gen8_enable_rps(dev); | |
c2bc2fc5 | 6235 | __gen6_update_ring_freq(dev); |
0a073b84 JB |
6236 | } else { |
6237 | gen6_enable_rps(dev); | |
c2bc2fc5 | 6238 | __gen6_update_ring_freq(dev); |
0a073b84 | 6239 | } |
aed242ff CW |
6240 | |
6241 | WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq); | |
6242 | WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq); | |
6243 | ||
6244 | WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq); | |
6245 | WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq); | |
6246 | ||
c0951f0c | 6247 | dev_priv->rps.enabled = true; |
3cc134e3 | 6248 | |
4c2a8897 | 6249 | gen6_enable_rps_interrupts(dev); |
3cc134e3 | 6250 | |
4fc688ce | 6251 | mutex_unlock(&dev_priv->rps.hw_lock); |
c6df39b5 ID |
6252 | |
6253 | intel_runtime_pm_put(dev_priv); | |
1a01ab3b JB |
6254 | } |
6255 | ||
8090c6b9 DV |
6256 | void intel_enable_gt_powersave(struct drm_device *dev) |
6257 | { | |
1a01ab3b JB |
6258 | struct drm_i915_private *dev_priv = dev->dev_private; |
6259 | ||
f61018b1 YZ |
6260 | /* Powersaving is controlled by the host when inside a VM */ |
6261 | if (intel_vgpu_active(dev)) | |
6262 | return; | |
6263 | ||
8090c6b9 | 6264 | if (IS_IRONLAKE_M(dev)) { |
dc1d0136 | 6265 | mutex_lock(&dev->struct_mutex); |
8090c6b9 | 6266 | ironlake_enable_drps(dev); |
8090c6b9 | 6267 | intel_init_emon(dev); |
dc1d0136 | 6268 | mutex_unlock(&dev->struct_mutex); |
38807746 | 6269 | } else if (INTEL_INFO(dev)->gen >= 6) { |
1a01ab3b JB |
6270 | /* |
6271 | * PCU communication is slow and this doesn't need to be | |
6272 | * done at any specific time, so do this out of our fast path | |
6273 | * to make resume and init faster. | |
c6df39b5 ID |
6274 | * |
6275 | * We depend on the HW RC6 power context save/restore | |
6276 | * mechanism when entering D3 through runtime PM suspend. So | |
6277 | * disable RPM until RPS/RC6 is properly setup. We can only | |
6278 | * get here via the driver load/system resume/runtime resume | |
6279 | * paths, so the _noresume version is enough (and in case of | |
6280 | * runtime resume it's necessary). | |
1a01ab3b | 6281 | */ |
c6df39b5 ID |
6282 | if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work, |
6283 | round_jiffies_up_relative(HZ))) | |
6284 | intel_runtime_pm_get_noresume(dev_priv); | |
8090c6b9 DV |
6285 | } |
6286 | } | |
6287 | ||
c6df39b5 ID |
6288 | void intel_reset_gt_powersave(struct drm_device *dev) |
6289 | { | |
6290 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6291 | ||
dbea3cea ID |
6292 | if (INTEL_INFO(dev)->gen < 6) |
6293 | return; | |
6294 | ||
6295 | gen6_suspend_rps(dev); | |
c6df39b5 | 6296 | dev_priv->rps.enabled = false; |
c6df39b5 ID |
6297 | } |
6298 | ||
3107bd48 DV |
6299 | static void ibx_init_clock_gating(struct drm_device *dev) |
6300 | { | |
6301 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6302 | ||
6303 | /* | |
6304 | * On Ibex Peak and Cougar Point, we need to disable clock | |
6305 | * gating for the panel power sequencer or it will fail to | |
6306 | * start up when no ports are active. | |
6307 | */ | |
6308 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); | |
6309 | } | |
6310 | ||
0e088b8f VS |
6311 | static void g4x_disable_trickle_feed(struct drm_device *dev) |
6312 | { | |
6313 | struct drm_i915_private *dev_priv = dev->dev_private; | |
b12ce1d8 | 6314 | enum pipe pipe; |
0e088b8f | 6315 | |
055e393f | 6316 | for_each_pipe(dev_priv, pipe) { |
0e088b8f VS |
6317 | I915_WRITE(DSPCNTR(pipe), |
6318 | I915_READ(DSPCNTR(pipe)) | | |
6319 | DISPPLANE_TRICKLE_FEED_DISABLE); | |
b12ce1d8 VS |
6320 | |
6321 | I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe))); | |
6322 | POSTING_READ(DSPSURF(pipe)); | |
0e088b8f VS |
6323 | } |
6324 | } | |
6325 | ||
017636cc VS |
6326 | static void ilk_init_lp_watermarks(struct drm_device *dev) |
6327 | { | |
6328 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6329 | ||
6330 | I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN); | |
6331 | I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN); | |
6332 | I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN); | |
6333 | ||
6334 | /* | |
6335 | * Don't touch WM1S_LP_EN here. | |
6336 | * Doing so could cause underruns. | |
6337 | */ | |
6338 | } | |
6339 | ||
1fa61106 | 6340 | static void ironlake_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6341 | { |
6342 | struct drm_i915_private *dev_priv = dev->dev_private; | |
231e54f6 | 6343 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 6344 | |
f1e8fa56 DL |
6345 | /* |
6346 | * Required for FBC | |
6347 | * WaFbcDisableDpfcClockGating:ilk | |
6348 | */ | |
4d47e4f5 DL |
6349 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
6350 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | | |
6351 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; | |
6f1d69b0 ED |
6352 | |
6353 | I915_WRITE(PCH_3DCGDIS0, | |
6354 | MARIUNIT_CLOCK_GATE_DISABLE | | |
6355 | SVSMUNIT_CLOCK_GATE_DISABLE); | |
6356 | I915_WRITE(PCH_3DCGDIS1, | |
6357 | VFMUNIT_CLOCK_GATE_DISABLE); | |
6358 | ||
6f1d69b0 ED |
6359 | /* |
6360 | * According to the spec the following bits should be set in | |
6361 | * order to enable memory self-refresh | |
6362 | * The bit 22/21 of 0x42004 | |
6363 | * The bit 5 of 0x42020 | |
6364 | * The bit 15 of 0x45000 | |
6365 | */ | |
6366 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
6367 | (I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6368 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); | |
4d47e4f5 | 6369 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
6f1d69b0 ED |
6370 | I915_WRITE(DISP_ARB_CTL, |
6371 | (I915_READ(DISP_ARB_CTL) | | |
6372 | DISP_FBC_WM_DIS)); | |
017636cc VS |
6373 | |
6374 | ilk_init_lp_watermarks(dev); | |
6f1d69b0 ED |
6375 | |
6376 | /* | |
6377 | * Based on the document from hardware guys the following bits | |
6378 | * should be set unconditionally in order to enable FBC. | |
6379 | * The bit 22 of 0x42000 | |
6380 | * The bit 22 of 0x42004 | |
6381 | * The bit 7,8,9 of 0x42020. | |
6382 | */ | |
6383 | if (IS_IRONLAKE_M(dev)) { | |
4bb35334 | 6384 | /* WaFbcAsynchFlipDisableFbcQueue:ilk */ |
6f1d69b0 ED |
6385 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
6386 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
6387 | ILK_FBCQ_DIS); | |
6388 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
6389 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6390 | ILK_DPARB_GATE); | |
6f1d69b0 ED |
6391 | } |
6392 | ||
4d47e4f5 DL |
6393 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6394 | ||
6f1d69b0 ED |
6395 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
6396 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6397 | ILK_ELPIN_409_SELECT); | |
6398 | I915_WRITE(_3D_CHICKEN2, | |
6399 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | | |
6400 | _3D_CHICKEN2_WM_READ_PIPELINED); | |
4358a374 | 6401 | |
ecdb4eb7 | 6402 | /* WaDisableRenderCachePipelinedFlush:ilk */ |
4358a374 DV |
6403 | I915_WRITE(CACHE_MODE_0, |
6404 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
3107bd48 | 6405 | |
4e04632e AG |
6406 | /* WaDisable_RenderCache_OperationalFlush:ilk */ |
6407 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6408 | ||
0e088b8f | 6409 | g4x_disable_trickle_feed(dev); |
bdad2b2f | 6410 | |
3107bd48 DV |
6411 | ibx_init_clock_gating(dev); |
6412 | } | |
6413 | ||
6414 | static void cpt_init_clock_gating(struct drm_device *dev) | |
6415 | { | |
6416 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6417 | int pipe; | |
3f704fa2 | 6418 | uint32_t val; |
3107bd48 DV |
6419 | |
6420 | /* | |
6421 | * On Ibex Peak and Cougar Point, we need to disable clock | |
6422 | * gating for the panel power sequencer or it will fail to | |
6423 | * start up when no ports are active. | |
6424 | */ | |
cd664078 JB |
6425 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE | |
6426 | PCH_DPLUNIT_CLOCK_GATE_DISABLE | | |
6427 | PCH_CPUNIT_CLOCK_GATE_DISABLE); | |
3107bd48 DV |
6428 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | |
6429 | DPLS_EDP_PPS_FIX_DIS); | |
335c07b7 TI |
6430 | /* The below fixes the weird display corruption, a few pixels shifted |
6431 | * downward, on (only) LVDS of some HP laptops with IVY. | |
6432 | */ | |
055e393f | 6433 | for_each_pipe(dev_priv, pipe) { |
dc4bd2d1 PZ |
6434 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
6435 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; | |
6436 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; | |
41aa3448 | 6437 | if (dev_priv->vbt.fdi_rx_polarity_inverted) |
3f704fa2 | 6438 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
dc4bd2d1 PZ |
6439 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
6440 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; | |
6441 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; | |
3f704fa2 PZ |
6442 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
6443 | } | |
3107bd48 | 6444 | /* WADP0ClockGatingDisable */ |
055e393f | 6445 | for_each_pipe(dev_priv, pipe) { |
3107bd48 DV |
6446 | I915_WRITE(TRANS_CHICKEN1(pipe), |
6447 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); | |
6448 | } | |
6f1d69b0 ED |
6449 | } |
6450 | ||
1d7aaa0c DV |
6451 | static void gen6_check_mch_setup(struct drm_device *dev) |
6452 | { | |
6453 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6454 | uint32_t tmp; | |
6455 | ||
6456 | tmp = I915_READ(MCH_SSKPD); | |
df662a28 DV |
6457 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) |
6458 | DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n", | |
6459 | tmp); | |
1d7aaa0c DV |
6460 | } |
6461 | ||
1fa61106 | 6462 | static void gen6_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6463 | { |
6464 | struct drm_i915_private *dev_priv = dev->dev_private; | |
231e54f6 | 6465 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 6466 | |
231e54f6 | 6467 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6f1d69b0 ED |
6468 | |
6469 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
6470 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6471 | ILK_ELPIN_409_SELECT); | |
6472 | ||
ecdb4eb7 | 6473 | /* WaDisableHiZPlanesWhenMSAAEnabled:snb */ |
4283908e DV |
6474 | I915_WRITE(_3D_CHICKEN, |
6475 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); | |
6476 | ||
4e04632e AG |
6477 | /* WaDisable_RenderCache_OperationalFlush:snb */ |
6478 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6479 | ||
8d85d272 VS |
6480 | /* |
6481 | * BSpec recoomends 8x4 when MSAA is used, | |
6482 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
6483 | * |
6484 | * Note that PS/WM thread counts depend on the WIZ hashing | |
6485 | * disable bit, which we don't touch here, but it's good | |
6486 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
8d85d272 VS |
6487 | */ |
6488 | I915_WRITE(GEN6_GT_MODE, | |
98533251 | 6489 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
8d85d272 | 6490 | |
017636cc | 6491 | ilk_init_lp_watermarks(dev); |
6f1d69b0 | 6492 | |
6f1d69b0 | 6493 | I915_WRITE(CACHE_MODE_0, |
50743298 | 6494 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
6f1d69b0 ED |
6495 | |
6496 | I915_WRITE(GEN6_UCGCTL1, | |
6497 | I915_READ(GEN6_UCGCTL1) | | |
6498 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | | |
6499 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
6500 | ||
6501 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock | |
6502 | * gating disable must be set. Failure to set it results in | |
6503 | * flickering pixels due to Z write ordering failures after | |
6504 | * some amount of runtime in the Mesa "fire" demo, and Unigine | |
6505 | * Sanctuary and Tropics, and apparently anything else with | |
6506 | * alpha test or pixel discard. | |
6507 | * | |
6508 | * According to the spec, bit 11 (RCCUNIT) must also be set, | |
6509 | * but we didn't debug actual testcases to find it out. | |
0f846f81 | 6510 | * |
ef59318c VS |
6511 | * WaDisableRCCUnitClockGating:snb |
6512 | * WaDisableRCPBUnitClockGating:snb | |
6f1d69b0 ED |
6513 | */ |
6514 | I915_WRITE(GEN6_UCGCTL2, | |
6515 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | | |
6516 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); | |
6517 | ||
5eb146dd | 6518 | /* WaStripsFansDisableFastClipPerformanceFix:snb */ |
743b57d8 VS |
6519 | I915_WRITE(_3D_CHICKEN3, |
6520 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL)); | |
6f1d69b0 | 6521 | |
e927ecde VS |
6522 | /* |
6523 | * Bspec says: | |
6524 | * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and | |
6525 | * 3DSTATE_SF number of SF output attributes is more than 16." | |
6526 | */ | |
6527 | I915_WRITE(_3D_CHICKEN3, | |
6528 | _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH)); | |
6529 | ||
6f1d69b0 ED |
6530 | /* |
6531 | * According to the spec the following bits should be | |
6532 | * set in order to enable memory self-refresh and fbc: | |
6533 | * The bit21 and bit22 of 0x42000 | |
6534 | * The bit21 and bit22 of 0x42004 | |
6535 | * The bit5 and bit7 of 0x42020 | |
6536 | * The bit14 of 0x70180 | |
6537 | * The bit14 of 0x71180 | |
4bb35334 DL |
6538 | * |
6539 | * WaFbcAsynchFlipDisableFbcQueue:snb | |
6f1d69b0 ED |
6540 | */ |
6541 | I915_WRITE(ILK_DISPLAY_CHICKEN1, | |
6542 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
6543 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); | |
6544 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
6545 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
6546 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); | |
231e54f6 DL |
6547 | I915_WRITE(ILK_DSPCLK_GATE_D, |
6548 | I915_READ(ILK_DSPCLK_GATE_D) | | |
6549 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | | |
6550 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); | |
6f1d69b0 | 6551 | |
0e088b8f | 6552 | g4x_disable_trickle_feed(dev); |
f8f2ac9a | 6553 | |
3107bd48 | 6554 | cpt_init_clock_gating(dev); |
1d7aaa0c DV |
6555 | |
6556 | gen6_check_mch_setup(dev); | |
6f1d69b0 ED |
6557 | } |
6558 | ||
6559 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) | |
6560 | { | |
6561 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); | |
6562 | ||
3aad9059 | 6563 | /* |
46680e0a | 6564 | * WaVSThreadDispatchOverride:ivb,vlv |
3aad9059 VS |
6565 | * |
6566 | * This actually overrides the dispatch | |
6567 | * mode for all thread types. | |
6568 | */ | |
6f1d69b0 ED |
6569 | reg &= ~GEN7_FF_SCHED_MASK; |
6570 | reg |= GEN7_FF_TS_SCHED_HW; | |
6571 | reg |= GEN7_FF_VS_SCHED_HW; | |
6572 | reg |= GEN7_FF_DS_SCHED_HW; | |
6573 | ||
6574 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); | |
6575 | } | |
6576 | ||
17a303ec PZ |
6577 | static void lpt_init_clock_gating(struct drm_device *dev) |
6578 | { | |
6579 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6580 | ||
6581 | /* | |
6582 | * TODO: this bit should only be enabled when really needed, then | |
6583 | * disabled when not needed anymore in order to save power. | |
6584 | */ | |
6585 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) | |
6586 | I915_WRITE(SOUTH_DSPCLK_GATE_D, | |
6587 | I915_READ(SOUTH_DSPCLK_GATE_D) | | |
6588 | PCH_LP_PARTITION_LEVEL_DISABLE); | |
0a790cdb PZ |
6589 | |
6590 | /* WADPOClockGatingDisable:hsw */ | |
6591 | I915_WRITE(_TRANSA_CHICKEN1, | |
6592 | I915_READ(_TRANSA_CHICKEN1) | | |
6593 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); | |
17a303ec PZ |
6594 | } |
6595 | ||
7d708ee4 ID |
6596 | static void lpt_suspend_hw(struct drm_device *dev) |
6597 | { | |
6598 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6599 | ||
6600 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) { | |
6601 | uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D); | |
6602 | ||
6603 | val &= ~PCH_LP_PARTITION_LEVEL_DISABLE; | |
6604 | I915_WRITE(SOUTH_DSPCLK_GATE_D, val); | |
6605 | } | |
6606 | } | |
6607 | ||
47c2bd97 | 6608 | static void broadwell_init_clock_gating(struct drm_device *dev) |
1020a5c2 BW |
6609 | { |
6610 | struct drm_i915_private *dev_priv = dev->dev_private; | |
07d27e20 | 6611 | enum pipe pipe; |
4d487cff | 6612 | uint32_t misccpctl; |
1020a5c2 | 6613 | |
7ad0dbab | 6614 | ilk_init_lp_watermarks(dev); |
50ed5fbd | 6615 | |
ab57fff1 | 6616 | /* WaSwitchSolVfFArbitrationPriority:bdw */ |
50ed5fbd | 6617 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
fe4ab3ce | 6618 | |
ab57fff1 | 6619 | /* WaPsrDPAMaskVBlankInSRD:bdw */ |
fe4ab3ce BW |
6620 | I915_WRITE(CHICKEN_PAR1_1, |
6621 | I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD); | |
6622 | ||
ab57fff1 | 6623 | /* WaPsrDPRSUnmaskVBlankInSRD:bdw */ |
055e393f | 6624 | for_each_pipe(dev_priv, pipe) { |
07d27e20 | 6625 | I915_WRITE(CHICKEN_PIPESL_1(pipe), |
c7c65622 | 6626 | I915_READ(CHICKEN_PIPESL_1(pipe)) | |
8f670bb1 | 6627 | BDW_DPRS_MASK_VBLANK_SRD); |
fe4ab3ce | 6628 | } |
63801f21 | 6629 | |
ab57fff1 BW |
6630 | /* WaVSRefCountFullforceMissDisable:bdw */ |
6631 | /* WaDSRefCountFullforceMissDisable:bdw */ | |
6632 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
6633 | I915_READ(GEN7_FF_THREAD_MODE) & | |
6634 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); | |
36075a4c | 6635 | |
295e8bb7 VS |
6636 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, |
6637 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); | |
4f1ca9e9 VS |
6638 | |
6639 | /* WaDisableSDEUnitClockGating:bdw */ | |
6640 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
6641 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
5d708680 | 6642 | |
4d487cff VS |
6643 | /* |
6644 | * WaProgramL3SqcReg1Default:bdw | |
6645 | * WaTempDisableDOPClkGating:bdw | |
6646 | */ | |
6647 | misccpctl = I915_READ(GEN7_MISCCPCTL); | |
6648 | I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE); | |
6649 | I915_WRITE(GEN8_L3SQCREG1, BDW_WA_L3SQCREG1_DEFAULT); | |
6650 | I915_WRITE(GEN7_MISCCPCTL, misccpctl); | |
6651 | ||
6d50b065 VS |
6652 | /* |
6653 | * WaGttCachingOffByDefault:bdw | |
6654 | * GTT cache may not work with big pages, so if those | |
6655 | * are ever enabled GTT cache may need to be disabled. | |
6656 | */ | |
6657 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); | |
6658 | ||
89d6b2b8 | 6659 | lpt_init_clock_gating(dev); |
1020a5c2 BW |
6660 | } |
6661 | ||
cad2a2d7 ED |
6662 | static void haswell_init_clock_gating(struct drm_device *dev) |
6663 | { | |
6664 | struct drm_i915_private *dev_priv = dev->dev_private; | |
cad2a2d7 | 6665 | |
017636cc | 6666 | ilk_init_lp_watermarks(dev); |
cad2a2d7 | 6667 | |
f3fc4884 FJ |
6668 | /* L3 caching of data atomics doesn't work -- disable it. */ |
6669 | I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE); | |
6670 | I915_WRITE(HSW_ROW_CHICKEN3, | |
6671 | _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE)); | |
6672 | ||
ecdb4eb7 | 6673 | /* This is required by WaCatErrorRejectionIssue:hsw */ |
cad2a2d7 ED |
6674 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6675 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
6676 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
6677 | ||
e36ea7ff VS |
6678 | /* WaVSRefCountFullforceMissDisable:hsw */ |
6679 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
6680 | I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME); | |
cad2a2d7 | 6681 | |
4e04632e AG |
6682 | /* WaDisable_RenderCache_OperationalFlush:hsw */ |
6683 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6684 | ||
fe27c606 CW |
6685 | /* enable HiZ Raw Stall Optimization */ |
6686 | I915_WRITE(CACHE_MODE_0_GEN7, | |
6687 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); | |
6688 | ||
ecdb4eb7 | 6689 | /* WaDisable4x2SubspanOptimization:hsw */ |
cad2a2d7 ED |
6690 | I915_WRITE(CACHE_MODE_1, |
6691 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
1544d9d5 | 6692 | |
a12c4967 VS |
6693 | /* |
6694 | * BSpec recommends 8x4 when MSAA is used, | |
6695 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
6696 | * |
6697 | * Note that PS/WM thread counts depend on the WIZ hashing | |
6698 | * disable bit, which we don't touch here, but it's good | |
6699 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
a12c4967 VS |
6700 | */ |
6701 | I915_WRITE(GEN7_GT_MODE, | |
98533251 | 6702 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
a12c4967 | 6703 | |
94411593 KG |
6704 | /* WaSampleCChickenBitEnable:hsw */ |
6705 | I915_WRITE(HALF_SLICE_CHICKEN3, | |
6706 | _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE)); | |
6707 | ||
ecdb4eb7 | 6708 | /* WaSwitchSolVfFArbitrationPriority:hsw */ |
e3dff585 BW |
6709 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
6710 | ||
90a88643 PZ |
6711 | /* WaRsPkgCStateDisplayPMReq:hsw */ |
6712 | I915_WRITE(CHICKEN_PAR1_1, | |
6713 | I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES); | |
1544d9d5 | 6714 | |
17a303ec | 6715 | lpt_init_clock_gating(dev); |
cad2a2d7 ED |
6716 | } |
6717 | ||
1fa61106 | 6718 | static void ivybridge_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6719 | { |
6720 | struct drm_i915_private *dev_priv = dev->dev_private; | |
20848223 | 6721 | uint32_t snpcr; |
6f1d69b0 | 6722 | |
017636cc | 6723 | ilk_init_lp_watermarks(dev); |
6f1d69b0 | 6724 | |
231e54f6 | 6725 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
6f1d69b0 | 6726 | |
ecdb4eb7 | 6727 | /* WaDisableEarlyCull:ivb */ |
87f8020e JB |
6728 | I915_WRITE(_3D_CHICKEN3, |
6729 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
6730 | ||
ecdb4eb7 | 6731 | /* WaDisableBackToBackFlipFix:ivb */ |
6f1d69b0 ED |
6732 | I915_WRITE(IVB_CHICKEN3, |
6733 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
6734 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
6735 | ||
ecdb4eb7 | 6736 | /* WaDisablePSDDualDispatchEnable:ivb */ |
12f3382b JB |
6737 | if (IS_IVB_GT1(dev)) |
6738 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, | |
6739 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 6740 | |
4e04632e AG |
6741 | /* WaDisable_RenderCache_OperationalFlush:ivb */ |
6742 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6743 | ||
ecdb4eb7 | 6744 | /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */ |
6f1d69b0 ED |
6745 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
6746 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); | |
6747 | ||
ecdb4eb7 | 6748 | /* WaApplyL3ControlAndL3ChickenMode:ivb */ |
6f1d69b0 ED |
6749 | I915_WRITE(GEN7_L3CNTLREG1, |
6750 | GEN7_WA_FOR_GEN7_L3_CONTROL); | |
6751 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, | |
8ab43976 JB |
6752 | GEN7_WA_L3_CHICKEN_MODE); |
6753 | if (IS_IVB_GT1(dev)) | |
6754 | I915_WRITE(GEN7_ROW_CHICKEN2, | |
6755 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
412236c2 VS |
6756 | else { |
6757 | /* must write both registers */ | |
6758 | I915_WRITE(GEN7_ROW_CHICKEN2, | |
6759 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
8ab43976 JB |
6760 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, |
6761 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
412236c2 | 6762 | } |
6f1d69b0 | 6763 | |
ecdb4eb7 | 6764 | /* WaForceL3Serialization:ivb */ |
61939d97 JB |
6765 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
6766 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
6767 | ||
1b80a19a | 6768 | /* |
0f846f81 | 6769 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
ecdb4eb7 | 6770 | * This implements the WaDisableRCZUnitClockGating:ivb workaround. |
0f846f81 JB |
6771 | */ |
6772 | I915_WRITE(GEN6_UCGCTL2, | |
28acf3b2 | 6773 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
0f846f81 | 6774 | |
ecdb4eb7 | 6775 | /* This is required by WaCatErrorRejectionIssue:ivb */ |
6f1d69b0 ED |
6776 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6777 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
6778 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
6779 | ||
0e088b8f | 6780 | g4x_disable_trickle_feed(dev); |
6f1d69b0 ED |
6781 | |
6782 | gen7_setup_fixed_func_scheduler(dev_priv); | |
97e1930f | 6783 | |
22721343 CW |
6784 | if (0) { /* causes HiZ corruption on ivb:gt1 */ |
6785 | /* enable HiZ Raw Stall Optimization */ | |
6786 | I915_WRITE(CACHE_MODE_0_GEN7, | |
6787 | _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE)); | |
6788 | } | |
116f2b6d | 6789 | |
ecdb4eb7 | 6790 | /* WaDisable4x2SubspanOptimization:ivb */ |
97e1930f DV |
6791 | I915_WRITE(CACHE_MODE_1, |
6792 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
20848223 | 6793 | |
a607c1a4 VS |
6794 | /* |
6795 | * BSpec recommends 8x4 when MSAA is used, | |
6796 | * however in practice 16x4 seems fastest. | |
c5c98a58 VS |
6797 | * |
6798 | * Note that PS/WM thread counts depend on the WIZ hashing | |
6799 | * disable bit, which we don't touch here, but it's good | |
6800 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
a607c1a4 VS |
6801 | */ |
6802 | I915_WRITE(GEN7_GT_MODE, | |
98533251 | 6803 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); |
a607c1a4 | 6804 | |
20848223 BW |
6805 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); |
6806 | snpcr &= ~GEN6_MBC_SNPCR_MASK; | |
6807 | snpcr |= GEN6_MBC_SNPCR_MED; | |
6808 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); | |
3107bd48 | 6809 | |
ab5c608b BW |
6810 | if (!HAS_PCH_NOP(dev)) |
6811 | cpt_init_clock_gating(dev); | |
1d7aaa0c DV |
6812 | |
6813 | gen6_check_mch_setup(dev); | |
6f1d69b0 ED |
6814 | } |
6815 | ||
c6beb13e VS |
6816 | static void vlv_init_display_clock_gating(struct drm_i915_private *dev_priv) |
6817 | { | |
6818 | I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE); | |
6819 | ||
6820 | /* | |
6821 | * Disable trickle feed and enable pnd deadline calculation | |
6822 | */ | |
6823 | I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE); | |
6824 | I915_WRITE(CBR1_VLV, 0); | |
6825 | } | |
6826 | ||
1fa61106 | 6827 | static void valleyview_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6828 | { |
6829 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6f1d69b0 | 6830 | |
c6beb13e | 6831 | vlv_init_display_clock_gating(dev_priv); |
6f1d69b0 | 6832 | |
ecdb4eb7 | 6833 | /* WaDisableEarlyCull:vlv */ |
87f8020e JB |
6834 | I915_WRITE(_3D_CHICKEN3, |
6835 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
6836 | ||
ecdb4eb7 | 6837 | /* WaDisableBackToBackFlipFix:vlv */ |
6f1d69b0 ED |
6838 | I915_WRITE(IVB_CHICKEN3, |
6839 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
6840 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
6841 | ||
fad7d36e | 6842 | /* WaPsdDispatchEnable:vlv */ |
ecdb4eb7 | 6843 | /* WaDisablePSDDualDispatchEnable:vlv */ |
12f3382b | 6844 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
d3bc0303 JB |
6845 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
6846 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 6847 | |
4e04632e AG |
6848 | /* WaDisable_RenderCache_OperationalFlush:vlv */ |
6849 | I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6850 | ||
ecdb4eb7 | 6851 | /* WaForceL3Serialization:vlv */ |
61939d97 JB |
6852 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
6853 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
6854 | ||
ecdb4eb7 | 6855 | /* WaDisableDopClockGating:vlv */ |
8ab43976 JB |
6856 | I915_WRITE(GEN7_ROW_CHICKEN2, |
6857 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
6858 | ||
ecdb4eb7 | 6859 | /* This is required by WaCatErrorRejectionIssue:vlv */ |
6f1d69b0 ED |
6860 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
6861 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
6862 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
6863 | ||
46680e0a VS |
6864 | gen7_setup_fixed_func_scheduler(dev_priv); |
6865 | ||
3c0edaeb | 6866 | /* |
0f846f81 | 6867 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. |
ecdb4eb7 | 6868 | * This implements the WaDisableRCZUnitClockGating:vlv workaround. |
0f846f81 JB |
6869 | */ |
6870 | I915_WRITE(GEN6_UCGCTL2, | |
3c0edaeb | 6871 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE); |
0f846f81 | 6872 | |
c98f5062 AG |
6873 | /* WaDisableL3Bank2xClockGate:vlv |
6874 | * Disabling L3 clock gating- MMIO 940c[25] = 1 | |
6875 | * Set bit 25, to disable L3_BANK_2x_CLK_GATING */ | |
6876 | I915_WRITE(GEN7_UCGCTL4, | |
6877 | I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE); | |
e3f33d46 | 6878 | |
afd58e79 VS |
6879 | /* |
6880 | * BSpec says this must be set, even though | |
6881 | * WaDisable4x2SubspanOptimization isn't listed for VLV. | |
6882 | */ | |
6b26c86d DV |
6883 | I915_WRITE(CACHE_MODE_1, |
6884 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
7983117f | 6885 | |
da2518f9 VS |
6886 | /* |
6887 | * BSpec recommends 8x4 when MSAA is used, | |
6888 | * however in practice 16x4 seems fastest. | |
6889 | * | |
6890 | * Note that PS/WM thread counts depend on the WIZ hashing | |
6891 | * disable bit, which we don't touch here, but it's good | |
6892 | * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM). | |
6893 | */ | |
6894 | I915_WRITE(GEN7_GT_MODE, | |
6895 | _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4)); | |
6896 | ||
031994ee VS |
6897 | /* |
6898 | * WaIncreaseL3CreditsForVLVB0:vlv | |
6899 | * This is the hardware default actually. | |
6900 | */ | |
6901 | I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE); | |
6902 | ||
2d809570 | 6903 | /* |
ecdb4eb7 | 6904 | * WaDisableVLVClockGating_VBIIssue:vlv |
2d809570 JB |
6905 | * Disable clock gating on th GCFG unit to prevent a delay |
6906 | * in the reporting of vblank events. | |
6907 | */ | |
7a0d1eed | 6908 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS); |
6f1d69b0 ED |
6909 | } |
6910 | ||
a4565da8 VS |
6911 | static void cherryview_init_clock_gating(struct drm_device *dev) |
6912 | { | |
6913 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6914 | ||
c6beb13e | 6915 | vlv_init_display_clock_gating(dev_priv); |
dd811e70 | 6916 | |
232ce337 VS |
6917 | /* WaVSRefCountFullforceMissDisable:chv */ |
6918 | /* WaDSRefCountFullforceMissDisable:chv */ | |
6919 | I915_WRITE(GEN7_FF_THREAD_MODE, | |
6920 | I915_READ(GEN7_FF_THREAD_MODE) & | |
6921 | ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME)); | |
acea6f95 VS |
6922 | |
6923 | /* WaDisableSemaphoreAndSyncFlipWait:chv */ | |
6924 | I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL, | |
6925 | _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE)); | |
0846697c VS |
6926 | |
6927 | /* WaDisableCSUnitClockGating:chv */ | |
6928 | I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) | | |
6929 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
c631780f VS |
6930 | |
6931 | /* WaDisableSDEUnitClockGating:chv */ | |
6932 | I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) | | |
6933 | GEN8_SDEUNIT_CLOCK_GATE_DISABLE); | |
6d50b065 VS |
6934 | |
6935 | /* | |
6936 | * GTT cache may not work with big pages, so if those | |
6937 | * are ever enabled GTT cache may need to be disabled. | |
6938 | */ | |
6939 | I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL); | |
a4565da8 VS |
6940 | } |
6941 | ||
1fa61106 | 6942 | static void g4x_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6943 | { |
6944 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6945 | uint32_t dspclk_gate; | |
6946 | ||
6947 | I915_WRITE(RENCLK_GATE_D1, 0); | |
6948 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | | |
6949 | GS_UNIT_CLOCK_GATE_DISABLE | | |
6950 | CL_UNIT_CLOCK_GATE_DISABLE); | |
6951 | I915_WRITE(RAMCLK_GATE_D, 0); | |
6952 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | | |
6953 | OVRUNIT_CLOCK_GATE_DISABLE | | |
6954 | OVCUNIT_CLOCK_GATE_DISABLE; | |
6955 | if (IS_GM45(dev)) | |
6956 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; | |
6957 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); | |
4358a374 DV |
6958 | |
6959 | /* WaDisableRenderCachePipelinedFlush */ | |
6960 | I915_WRITE(CACHE_MODE_0, | |
6961 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
de1aa629 | 6962 | |
4e04632e AG |
6963 | /* WaDisable_RenderCache_OperationalFlush:g4x */ |
6964 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6965 | ||
0e088b8f | 6966 | g4x_disable_trickle_feed(dev); |
6f1d69b0 ED |
6967 | } |
6968 | ||
1fa61106 | 6969 | static void crestline_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6970 | { |
6971 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6972 | ||
6973 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); | |
6974 | I915_WRITE(RENCLK_GATE_D2, 0); | |
6975 | I915_WRITE(DSPCLK_GATE_D, 0); | |
6976 | I915_WRITE(RAMCLK_GATE_D, 0); | |
6977 | I915_WRITE16(DEUC, 0); | |
20f94967 VS |
6978 | I915_WRITE(MI_ARB_STATE, |
6979 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
4e04632e AG |
6980 | |
6981 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ | |
6982 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6f1d69b0 ED |
6983 | } |
6984 | ||
1fa61106 | 6985 | static void broadwater_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
6986 | { |
6987 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6988 | ||
6989 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | | |
6990 | I965_RCC_CLOCK_GATE_DISABLE | | |
6991 | I965_RCPB_CLOCK_GATE_DISABLE | | |
6992 | I965_ISC_CLOCK_GATE_DISABLE | | |
6993 | I965_FBC_CLOCK_GATE_DISABLE); | |
6994 | I915_WRITE(RENCLK_GATE_D2, 0); | |
20f94967 VS |
6995 | I915_WRITE(MI_ARB_STATE, |
6996 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
4e04632e AG |
6997 | |
6998 | /* WaDisable_RenderCache_OperationalFlush:gen4 */ | |
6999 | I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE)); | |
6f1d69b0 ED |
7000 | } |
7001 | ||
1fa61106 | 7002 | static void gen3_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
7003 | { |
7004 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7005 | u32 dstate = I915_READ(D_STATE); | |
7006 | ||
7007 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | | |
7008 | DSTATE_DOT_CLOCK_GATING; | |
7009 | I915_WRITE(D_STATE, dstate); | |
13a86b85 CW |
7010 | |
7011 | if (IS_PINEVIEW(dev)) | |
7012 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); | |
974a3b0f DV |
7013 | |
7014 | /* IIR "flip pending" means done if this bit is set */ | |
7015 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); | |
12fabbcb VS |
7016 | |
7017 | /* interrupts should cause a wake up from C3 */ | |
3299254f | 7018 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN)); |
dbb42748 VS |
7019 | |
7020 | /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ | |
7021 | I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); | |
1038392b VS |
7022 | |
7023 | I915_WRITE(MI_ARB_STATE, | |
7024 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
7025 | } |
7026 | ||
1fa61106 | 7027 | static void i85x_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
7028 | { |
7029 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7030 | ||
7031 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); | |
54e472ae VS |
7032 | |
7033 | /* interrupts should cause a wake up from C3 */ | |
7034 | I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) | | |
7035 | _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE)); | |
1038392b VS |
7036 | |
7037 | I915_WRITE(MEM_MODE, | |
7038 | _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
7039 | } |
7040 | ||
1fa61106 | 7041 | static void i830_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
7042 | { |
7043 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7044 | ||
7045 | I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE); | |
1038392b VS |
7046 | |
7047 | I915_WRITE(MEM_MODE, | |
7048 | _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) | | |
7049 | _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
7050 | } |
7051 | ||
6f1d69b0 ED |
7052 | void intel_init_clock_gating(struct drm_device *dev) |
7053 | { | |
7054 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7055 | ||
c57e3551 DL |
7056 | if (dev_priv->display.init_clock_gating) |
7057 | dev_priv->display.init_clock_gating(dev); | |
6f1d69b0 ED |
7058 | } |
7059 | ||
7d708ee4 ID |
7060 | void intel_suspend_hw(struct drm_device *dev) |
7061 | { | |
7062 | if (HAS_PCH_LPT(dev)) | |
7063 | lpt_suspend_hw(dev); | |
7064 | } | |
7065 | ||
1fa61106 ED |
7066 | /* Set up chip specific power management-related functions */ |
7067 | void intel_init_pm(struct drm_device *dev) | |
7068 | { | |
7069 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7070 | ||
7ff0ebcc | 7071 | intel_fbc_init(dev_priv); |
1fa61106 | 7072 | |
c921aba8 DV |
7073 | /* For cxsr */ |
7074 | if (IS_PINEVIEW(dev)) | |
7075 | i915_pineview_get_mem_freq(dev); | |
7076 | else if (IS_GEN5(dev)) | |
7077 | i915_ironlake_get_mem_freq(dev); | |
7078 | ||
1fa61106 | 7079 | /* For FIFO watermark updates */ |
f5ed50cb | 7080 | if (INTEL_INFO(dev)->gen >= 9) { |
2af30a5c PB |
7081 | skl_setup_wm_latency(dev); |
7082 | ||
a82abe43 ID |
7083 | if (IS_BROXTON(dev)) |
7084 | dev_priv->display.init_clock_gating = | |
7085 | bxt_init_clock_gating; | |
7086 | else if (IS_SKYLAKE(dev)) | |
7087 | dev_priv->display.init_clock_gating = | |
7088 | skl_init_clock_gating; | |
2d41c0b5 PB |
7089 | dev_priv->display.update_wm = skl_update_wm; |
7090 | dev_priv->display.update_sprite_wm = skl_update_sprite_wm; | |
c83155a6 | 7091 | } else if (HAS_PCH_SPLIT(dev)) { |
fa50ad61 | 7092 | ilk_setup_wm_latency(dev); |
53615a5e | 7093 | |
bd602544 VS |
7094 | if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] && |
7095 | dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) || | |
7096 | (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] && | |
7097 | dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) { | |
7098 | dev_priv->display.update_wm = ilk_update_wm; | |
7099 | dev_priv->display.update_sprite_wm = ilk_update_sprite_wm; | |
7100 | } else { | |
7101 | DRM_DEBUG_KMS("Failed to read display plane latency. " | |
7102 | "Disable CxSR\n"); | |
7103 | } | |
7104 | ||
7105 | if (IS_GEN5(dev)) | |
1fa61106 | 7106 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; |
bd602544 | 7107 | else if (IS_GEN6(dev)) |
1fa61106 | 7108 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; |
bd602544 | 7109 | else if (IS_IVYBRIDGE(dev)) |
1fa61106 | 7110 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; |
bd602544 | 7111 | else if (IS_HASWELL(dev)) |
cad2a2d7 | 7112 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; |
bd602544 | 7113 | else if (INTEL_INFO(dev)->gen == 8) |
47c2bd97 | 7114 | dev_priv->display.init_clock_gating = broadwell_init_clock_gating; |
a4565da8 | 7115 | } else if (IS_CHERRYVIEW(dev)) { |
262cd2e1 VS |
7116 | vlv_setup_wm_latency(dev); |
7117 | ||
7118 | dev_priv->display.update_wm = vlv_update_wm; | |
a4565da8 VS |
7119 | dev_priv->display.init_clock_gating = |
7120 | cherryview_init_clock_gating; | |
1fa61106 | 7121 | } else if (IS_VALLEYVIEW(dev)) { |
26e1fe4f VS |
7122 | vlv_setup_wm_latency(dev); |
7123 | ||
7124 | dev_priv->display.update_wm = vlv_update_wm; | |
1fa61106 ED |
7125 | dev_priv->display.init_clock_gating = |
7126 | valleyview_init_clock_gating; | |
1fa61106 ED |
7127 | } else if (IS_PINEVIEW(dev)) { |
7128 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev), | |
7129 | dev_priv->is_ddr3, | |
7130 | dev_priv->fsb_freq, | |
7131 | dev_priv->mem_freq)) { | |
7132 | DRM_INFO("failed to find known CxSR latency " | |
7133 | "(found ddr%s fsb freq %d, mem freq %d), " | |
7134 | "disabling CxSR\n", | |
7135 | (dev_priv->is_ddr3 == 1) ? "3" : "2", | |
7136 | dev_priv->fsb_freq, dev_priv->mem_freq); | |
7137 | /* Disable CxSR and never update its watermark again */ | |
5209b1f4 | 7138 | intel_set_memory_cxsr(dev_priv, false); |
1fa61106 ED |
7139 | dev_priv->display.update_wm = NULL; |
7140 | } else | |
7141 | dev_priv->display.update_wm = pineview_update_wm; | |
7142 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; | |
7143 | } else if (IS_G4X(dev)) { | |
7144 | dev_priv->display.update_wm = g4x_update_wm; | |
7145 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; | |
7146 | } else if (IS_GEN4(dev)) { | |
7147 | dev_priv->display.update_wm = i965_update_wm; | |
7148 | if (IS_CRESTLINE(dev)) | |
7149 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; | |
7150 | else if (IS_BROADWATER(dev)) | |
7151 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; | |
7152 | } else if (IS_GEN3(dev)) { | |
7153 | dev_priv->display.update_wm = i9xx_update_wm; | |
7154 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; | |
7155 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; | |
feb56b93 DV |
7156 | } else if (IS_GEN2(dev)) { |
7157 | if (INTEL_INFO(dev)->num_pipes == 1) { | |
7158 | dev_priv->display.update_wm = i845_update_wm; | |
1fa61106 | 7159 | dev_priv->display.get_fifo_size = i845_get_fifo_size; |
feb56b93 DV |
7160 | } else { |
7161 | dev_priv->display.update_wm = i9xx_update_wm; | |
1fa61106 | 7162 | dev_priv->display.get_fifo_size = i830_get_fifo_size; |
feb56b93 DV |
7163 | } |
7164 | ||
7165 | if (IS_I85X(dev) || IS_I865G(dev)) | |
7166 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; | |
7167 | else | |
7168 | dev_priv->display.init_clock_gating = i830_init_clock_gating; | |
7169 | } else { | |
7170 | DRM_ERROR("unexpected fall-through in intel_init_pm\n"); | |
1fa61106 ED |
7171 | } |
7172 | } | |
7173 | ||
151a49d0 | 7174 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val) |
42c0526c | 7175 | { |
4fc688ce | 7176 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c BW |
7177 | |
7178 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
7179 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); | |
7180 | return -EAGAIN; | |
7181 | } | |
7182 | ||
7183 | I915_WRITE(GEN6_PCODE_DATA, *val); | |
dddab346 | 7184 | I915_WRITE(GEN6_PCODE_DATA1, 0); |
42c0526c BW |
7185 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); |
7186 | ||
7187 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, | |
7188 | 500)) { | |
7189 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); | |
7190 | return -ETIMEDOUT; | |
7191 | } | |
7192 | ||
7193 | *val = I915_READ(GEN6_PCODE_DATA); | |
7194 | I915_WRITE(GEN6_PCODE_DATA, 0); | |
7195 | ||
7196 | return 0; | |
7197 | } | |
7198 | ||
151a49d0 | 7199 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u32 mbox, u32 val) |
42c0526c | 7200 | { |
4fc688ce | 7201 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c BW |
7202 | |
7203 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
7204 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); | |
7205 | return -EAGAIN; | |
7206 | } | |
7207 | ||
7208 | I915_WRITE(GEN6_PCODE_DATA, val); | |
7209 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); | |
7210 | ||
7211 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, | |
7212 | 500)) { | |
7213 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); | |
7214 | return -ETIMEDOUT; | |
7215 | } | |
7216 | ||
7217 | I915_WRITE(GEN6_PCODE_DATA, 0); | |
7218 | ||
7219 | return 0; | |
7220 | } | |
a0e4e199 | 7221 | |
dd06f88c | 7222 | static int vlv_gpu_freq_div(unsigned int czclk_freq) |
855ba3be | 7223 | { |
dd06f88c VS |
7224 | switch (czclk_freq) { |
7225 | case 200: | |
7226 | return 10; | |
7227 | case 267: | |
7228 | return 12; | |
7229 | case 320: | |
7230 | case 333: | |
dd06f88c | 7231 | return 16; |
ab3fb157 VS |
7232 | case 400: |
7233 | return 20; | |
855ba3be JB |
7234 | default: |
7235 | return -1; | |
7236 | } | |
dd06f88c | 7237 | } |
855ba3be | 7238 | |
dd06f88c VS |
7239 | static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val) |
7240 | { | |
7241 | int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->mem_freq, 4); | |
7242 | ||
7243 | div = vlv_gpu_freq_div(czclk_freq); | |
7244 | if (div < 0) | |
7245 | return div; | |
7246 | ||
7247 | return DIV_ROUND_CLOSEST(czclk_freq * (val + 6 - 0xbd), div); | |
855ba3be JB |
7248 | } |
7249 | ||
b55dd647 | 7250 | static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val) |
855ba3be | 7251 | { |
dd06f88c | 7252 | int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->mem_freq, 4); |
855ba3be | 7253 | |
dd06f88c VS |
7254 | mul = vlv_gpu_freq_div(czclk_freq); |
7255 | if (mul < 0) | |
7256 | return mul; | |
855ba3be | 7257 | |
dd06f88c | 7258 | return DIV_ROUND_CLOSEST(mul * val, czclk_freq) + 0xbd - 6; |
855ba3be JB |
7259 | } |
7260 | ||
b55dd647 | 7261 | static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 7262 | { |
dd06f88c | 7263 | int div, czclk_freq = dev_priv->rps.cz_freq; |
22b1b2f8 | 7264 | |
dd06f88c VS |
7265 | div = vlv_gpu_freq_div(czclk_freq) / 2; |
7266 | if (div < 0) | |
7267 | return div; | |
22b1b2f8 | 7268 | |
dd06f88c | 7269 | return DIV_ROUND_CLOSEST(czclk_freq * val, 2 * div) / 2; |
22b1b2f8 D |
7270 | } |
7271 | ||
b55dd647 | 7272 | static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 7273 | { |
dd06f88c | 7274 | int mul, czclk_freq = dev_priv->rps.cz_freq; |
22b1b2f8 | 7275 | |
dd06f88c VS |
7276 | mul = vlv_gpu_freq_div(czclk_freq) / 2; |
7277 | if (mul < 0) | |
7278 | return mul; | |
22b1b2f8 | 7279 | |
1c14762d | 7280 | /* CHV needs even values */ |
dd06f88c | 7281 | return DIV_ROUND_CLOSEST(val * 2 * mul, czclk_freq) * 2; |
22b1b2f8 D |
7282 | } |
7283 | ||
616bc820 | 7284 | int intel_gpu_freq(struct drm_i915_private *dev_priv, int val) |
22b1b2f8 | 7285 | { |
80b6dda4 AG |
7286 | if (IS_GEN9(dev_priv->dev)) |
7287 | return (val * GT_FREQUENCY_MULTIPLIER) / GEN9_FREQ_SCALER; | |
7288 | else if (IS_CHERRYVIEW(dev_priv->dev)) | |
616bc820 | 7289 | return chv_gpu_freq(dev_priv, val); |
22b1b2f8 | 7290 | else if (IS_VALLEYVIEW(dev_priv->dev)) |
616bc820 VS |
7291 | return byt_gpu_freq(dev_priv, val); |
7292 | else | |
7293 | return val * GT_FREQUENCY_MULTIPLIER; | |
22b1b2f8 D |
7294 | } |
7295 | ||
616bc820 VS |
7296 | int intel_freq_opcode(struct drm_i915_private *dev_priv, int val) |
7297 | { | |
80b6dda4 AG |
7298 | if (IS_GEN9(dev_priv->dev)) |
7299 | return (val * GEN9_FREQ_SCALER) / GT_FREQUENCY_MULTIPLIER; | |
7300 | else if (IS_CHERRYVIEW(dev_priv->dev)) | |
616bc820 | 7301 | return chv_freq_opcode(dev_priv, val); |
22b1b2f8 | 7302 | else if (IS_VALLEYVIEW(dev_priv->dev)) |
616bc820 VS |
7303 | return byt_freq_opcode(dev_priv, val); |
7304 | else | |
7305 | return val / GT_FREQUENCY_MULTIPLIER; | |
7306 | } | |
22b1b2f8 | 7307 | |
6ad790c0 CW |
7308 | struct request_boost { |
7309 | struct work_struct work; | |
eed29a5b | 7310 | struct drm_i915_gem_request *req; |
6ad790c0 CW |
7311 | }; |
7312 | ||
7313 | static void __intel_rps_boost_work(struct work_struct *work) | |
7314 | { | |
7315 | struct request_boost *boost = container_of(work, struct request_boost, work); | |
e61b9958 | 7316 | struct drm_i915_gem_request *req = boost->req; |
6ad790c0 | 7317 | |
e61b9958 CW |
7318 | if (!i915_gem_request_completed(req, true)) |
7319 | gen6_rps_boost(to_i915(req->ring->dev), NULL, | |
7320 | req->emitted_jiffies); | |
6ad790c0 | 7321 | |
e61b9958 | 7322 | i915_gem_request_unreference__unlocked(req); |
6ad790c0 CW |
7323 | kfree(boost); |
7324 | } | |
7325 | ||
7326 | void intel_queue_rps_boost_for_request(struct drm_device *dev, | |
eed29a5b | 7327 | struct drm_i915_gem_request *req) |
6ad790c0 CW |
7328 | { |
7329 | struct request_boost *boost; | |
7330 | ||
eed29a5b | 7331 | if (req == NULL || INTEL_INFO(dev)->gen < 6) |
6ad790c0 CW |
7332 | return; |
7333 | ||
e61b9958 CW |
7334 | if (i915_gem_request_completed(req, true)) |
7335 | return; | |
7336 | ||
6ad790c0 CW |
7337 | boost = kmalloc(sizeof(*boost), GFP_ATOMIC); |
7338 | if (boost == NULL) | |
7339 | return; | |
7340 | ||
eed29a5b DV |
7341 | i915_gem_request_reference(req); |
7342 | boost->req = req; | |
6ad790c0 CW |
7343 | |
7344 | INIT_WORK(&boost->work, __intel_rps_boost_work); | |
7345 | queue_work(to_i915(dev)->wq, &boost->work); | |
7346 | } | |
7347 | ||
f742a552 | 7348 | void intel_pm_setup(struct drm_device *dev) |
907b28c5 CW |
7349 | { |
7350 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7351 | ||
f742a552 | 7352 | mutex_init(&dev_priv->rps.hw_lock); |
8d3afd7d | 7353 | spin_lock_init(&dev_priv->rps.client_lock); |
f742a552 | 7354 | |
907b28c5 CW |
7355 | INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work, |
7356 | intel_gen6_powersave_work); | |
1854d5ca | 7357 | INIT_LIST_HEAD(&dev_priv->rps.clients); |
2e1b8730 CW |
7358 | INIT_LIST_HEAD(&dev_priv->rps.semaphores.link); |
7359 | INIT_LIST_HEAD(&dev_priv->rps.mmioflips.link); | |
5d584b2e | 7360 | |
33688d95 | 7361 | dev_priv->pm.suspended = false; |
907b28c5 | 7362 | } |