]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/gpu/drm/i915/intel_pm.c
drm/i915: quirk backlight present on Macbook 4, 1
[mirror_ubuntu-bionic-kernel.git] / drivers / gpu / drm / i915 / intel_pm.c
CommitLineData
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
a82abe43
ID
55static void bxt_init_clock_gating(struct drm_device *dev)
56{
32608ca2
ID
57 struct drm_i915_private *dev_priv = dev->dev_private;
58
a7546159
NH
59 /* WaDisableSDEUnitClockGating:bxt */
60 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
61 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
62
32608ca2
ID
63 /*
64 * FIXME:
868434c5 65 * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only.
32608ca2 66 */
32608ca2 67 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
868434c5 68 GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ);
a82abe43
ID
69}
70
c921aba8
DV
71static void i915_pineview_get_mem_freq(struct drm_device *dev)
72{
50227e1c 73 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV
74 u32 tmp;
75
76 tmp = I915_READ(CLKCFG);
77
78 switch (tmp & CLKCFG_FSB_MASK) {
79 case CLKCFG_FSB_533:
80 dev_priv->fsb_freq = 533; /* 133*4 */
81 break;
82 case CLKCFG_FSB_800:
83 dev_priv->fsb_freq = 800; /* 200*4 */
84 break;
85 case CLKCFG_FSB_667:
86 dev_priv->fsb_freq = 667; /* 167*4 */
87 break;
88 case CLKCFG_FSB_400:
89 dev_priv->fsb_freq = 400; /* 100*4 */
90 break;
91 }
92
93 switch (tmp & CLKCFG_MEM_MASK) {
94 case CLKCFG_MEM_533:
95 dev_priv->mem_freq = 533;
96 break;
97 case CLKCFG_MEM_667:
98 dev_priv->mem_freq = 667;
99 break;
100 case CLKCFG_MEM_800:
101 dev_priv->mem_freq = 800;
102 break;
103 }
104
105 /* detect pineview DDR3 setting */
106 tmp = I915_READ(CSHRDDR3CTL);
107 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
108}
109
110static void i915_ironlake_get_mem_freq(struct drm_device *dev)
111{
50227e1c 112 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV
113 u16 ddrpll, csipll;
114
115 ddrpll = I915_READ16(DDRMPLL1);
116 csipll = I915_READ16(CSIPLL0);
117
118 switch (ddrpll & 0xff) {
119 case 0xc:
120 dev_priv->mem_freq = 800;
121 break;
122 case 0x10:
123 dev_priv->mem_freq = 1066;
124 break;
125 case 0x14:
126 dev_priv->mem_freq = 1333;
127 break;
128 case 0x18:
129 dev_priv->mem_freq = 1600;
130 break;
131 default:
132 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
133 ddrpll & 0xff);
134 dev_priv->mem_freq = 0;
135 break;
136 }
137
20e4d407 138 dev_priv->ips.r_t = dev_priv->mem_freq;
c921aba8
DV
139
140 switch (csipll & 0x3ff) {
141 case 0x00c:
142 dev_priv->fsb_freq = 3200;
143 break;
144 case 0x00e:
145 dev_priv->fsb_freq = 3733;
146 break;
147 case 0x010:
148 dev_priv->fsb_freq = 4266;
149 break;
150 case 0x012:
151 dev_priv->fsb_freq = 4800;
152 break;
153 case 0x014:
154 dev_priv->fsb_freq = 5333;
155 break;
156 case 0x016:
157 dev_priv->fsb_freq = 5866;
158 break;
159 case 0x018:
160 dev_priv->fsb_freq = 6400;
161 break;
162 default:
163 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
164 csipll & 0x3ff);
165 dev_priv->fsb_freq = 0;
166 break;
167 }
168
169 if (dev_priv->fsb_freq == 3200) {
20e4d407 170 dev_priv->ips.c_m = 0;
c921aba8 171 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
20e4d407 172 dev_priv->ips.c_m = 1;
c921aba8 173 } else {
20e4d407 174 dev_priv->ips.c_m = 2;
c921aba8
DV
175 }
176}
177
b445e3b0
ED
178static const struct cxsr_latency cxsr_latency_table[] = {
179 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
180 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
181 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
182 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
183 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
184
185 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
186 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
187 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
188 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
189 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
190
191 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
192 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
193 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
194 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
195 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
196
197 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
198 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
199 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
200 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
201 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
202
203 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
204 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
205 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
206 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
207 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
208
209 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
210 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
211 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
212 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
213 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
214};
215
63c62275 216static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
b445e3b0
ED
217 int is_ddr3,
218 int fsb,
219 int mem)
220{
221 const struct cxsr_latency *latency;
222 int i;
223
224 if (fsb == 0 || mem == 0)
225 return NULL;
226
227 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
228 latency = &cxsr_latency_table[i];
229 if (is_desktop == latency->is_desktop &&
230 is_ddr3 == latency->is_ddr3 &&
231 fsb == latency->fsb_freq && mem == latency->mem_freq)
232 return latency;
233 }
234
235 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
236
237 return NULL;
238}
239
fc1ac8de
VS
240static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable)
241{
242 u32 val;
243
244 mutex_lock(&dev_priv->rps.hw_lock);
245
246 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
247 if (enable)
248 val &= ~FORCE_DDR_HIGH_FREQ;
249 else
250 val |= FORCE_DDR_HIGH_FREQ;
251 val &= ~FORCE_DDR_LOW_FREQ;
252 val |= FORCE_DDR_FREQ_REQ_ACK;
253 vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);
254
255 if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
256 FORCE_DDR_FREQ_REQ_ACK) == 0, 3))
257 DRM_ERROR("timed out waiting for Punit DDR DVFS request\n");
258
259 mutex_unlock(&dev_priv->rps.hw_lock);
260}
261
cfb41411
VS
262static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable)
263{
264 u32 val;
265
266 mutex_lock(&dev_priv->rps.hw_lock);
267
268 val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ);
269 if (enable)
270 val |= DSP_MAXFIFO_PM5_ENABLE;
271 else
272 val &= ~DSP_MAXFIFO_PM5_ENABLE;
273 vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val);
274
275 mutex_unlock(&dev_priv->rps.hw_lock);
276}
277
f4998963
VS
278#define FW_WM(value, plane) \
279 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK)
280
5209b1f4 281void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable)
b445e3b0 282{
5209b1f4
ID
283 struct drm_device *dev = dev_priv->dev;
284 u32 val;
b445e3b0 285
5209b1f4
ID
286 if (IS_VALLEYVIEW(dev)) {
287 I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0);
a7a6c498 288 POSTING_READ(FW_BLC_SELF_VLV);
852eb00d 289 dev_priv->wm.vlv.cxsr = enable;
5209b1f4
ID
290 } else if (IS_G4X(dev) || IS_CRESTLINE(dev)) {
291 I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0);
a7a6c498 292 POSTING_READ(FW_BLC_SELF);
5209b1f4
ID
293 } else if (IS_PINEVIEW(dev)) {
294 val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN;
295 val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0;
296 I915_WRITE(DSPFW3, val);
a7a6c498 297 POSTING_READ(DSPFW3);
5209b1f4
ID
298 } else if (IS_I945G(dev) || IS_I945GM(dev)) {
299 val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) :
300 _MASKED_BIT_DISABLE(FW_BLC_SELF_EN);
301 I915_WRITE(FW_BLC_SELF, val);
a7a6c498 302 POSTING_READ(FW_BLC_SELF);
5209b1f4
ID
303 } else if (IS_I915GM(dev)) {
304 val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) :
305 _MASKED_BIT_DISABLE(INSTPM_SELF_EN);
306 I915_WRITE(INSTPM, val);
a7a6c498 307 POSTING_READ(INSTPM);
5209b1f4
ID
308 } else {
309 return;
310 }
b445e3b0 311
5209b1f4
ID
312 DRM_DEBUG_KMS("memory self-refresh is %s\n",
313 enable ? "enabled" : "disabled");
b445e3b0
ED
314}
315
fc1ac8de 316
b445e3b0
ED
317/*
318 * Latency for FIFO fetches is dependent on several factors:
319 * - memory configuration (speed, channels)
320 * - chipset
321 * - current MCH state
322 * It can be fairly high in some situations, so here we assume a fairly
323 * pessimal value. It's a tradeoff between extra memory fetches (if we
324 * set this value too high, the FIFO will fetch frequently to stay full)
325 * and power consumption (set it too low to save power and we might see
326 * FIFO underruns and display "flicker").
327 *
328 * A value of 5us seems to be a good balance; safe for very low end
329 * platforms but not overly aggressive on lower latency configs.
330 */
5aef6003 331static const int pessimal_latency_ns = 5000;
b445e3b0 332
b5004720
VS
333#define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \
334 ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8))
335
336static int vlv_get_fifo_size(struct drm_device *dev,
337 enum pipe pipe, int plane)
338{
339 struct drm_i915_private *dev_priv = dev->dev_private;
340 int sprite0_start, sprite1_start, size;
341
342 switch (pipe) {
343 uint32_t dsparb, dsparb2, dsparb3;
344 case PIPE_A:
345 dsparb = I915_READ(DSPARB);
346 dsparb2 = I915_READ(DSPARB2);
347 sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0);
348 sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4);
349 break;
350 case PIPE_B:
351 dsparb = I915_READ(DSPARB);
352 dsparb2 = I915_READ(DSPARB2);
353 sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8);
354 sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12);
355 break;
356 case PIPE_C:
357 dsparb2 = I915_READ(DSPARB2);
358 dsparb3 = I915_READ(DSPARB3);
359 sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16);
360 sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20);
361 break;
362 default:
363 return 0;
364 }
365
366 switch (plane) {
367 case 0:
368 size = sprite0_start;
369 break;
370 case 1:
371 size = sprite1_start - sprite0_start;
372 break;
373 case 2:
374 size = 512 - 1 - sprite1_start;
375 break;
376 default:
377 return 0;
378 }
379
380 DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n",
381 pipe_name(pipe), plane == 0 ? "primary" : "sprite",
382 plane == 0 ? plane_name(pipe) : sprite_name(pipe, plane - 1),
383 size);
384
385 return size;
386}
387
1fa61106 388static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED
389{
390 struct drm_i915_private *dev_priv = dev->dev_private;
391 uint32_t dsparb = I915_READ(DSPARB);
392 int size;
393
394 size = dsparb & 0x7f;
395 if (plane)
396 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
397
398 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
399 plane ? "B" : "A", size);
400
401 return size;
402}
403
feb56b93 404static int i830_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED
405{
406 struct drm_i915_private *dev_priv = dev->dev_private;
407 uint32_t dsparb = I915_READ(DSPARB);
408 int size;
409
410 size = dsparb & 0x1ff;
411 if (plane)
412 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
413 size >>= 1; /* Convert to cachelines */
414
415 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
416 plane ? "B" : "A", size);
417
418 return size;
419}
420
1fa61106 421static int i845_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED
422{
423 struct drm_i915_private *dev_priv = dev->dev_private;
424 uint32_t dsparb = I915_READ(DSPARB);
425 int size;
426
427 size = dsparb & 0x7f;
428 size >>= 2; /* Convert to cachelines */
429
430 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
431 plane ? "B" : "A",
432 size);
433
434 return size;
435}
436
b445e3b0
ED
437/* Pineview has different values for various configs */
438static const struct intel_watermark_params pineview_display_wm = {
e0f0273e
VS
439 .fifo_size = PINEVIEW_DISPLAY_FIFO,
440 .max_wm = PINEVIEW_MAX_WM,
441 .default_wm = PINEVIEW_DFT_WM,
442 .guard_size = PINEVIEW_GUARD_WM,
443 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED
444};
445static const struct intel_watermark_params pineview_display_hplloff_wm = {
e0f0273e
VS
446 .fifo_size = PINEVIEW_DISPLAY_FIFO,
447 .max_wm = PINEVIEW_MAX_WM,
448 .default_wm = PINEVIEW_DFT_HPLLOFF_WM,
449 .guard_size = PINEVIEW_GUARD_WM,
450 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED
451};
452static const struct intel_watermark_params pineview_cursor_wm = {
e0f0273e
VS
453 .fifo_size = PINEVIEW_CURSOR_FIFO,
454 .max_wm = PINEVIEW_CURSOR_MAX_WM,
455 .default_wm = PINEVIEW_CURSOR_DFT_WM,
456 .guard_size = PINEVIEW_CURSOR_GUARD_WM,
457 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED
458};
459static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
e0f0273e
VS
460 .fifo_size = PINEVIEW_CURSOR_FIFO,
461 .max_wm = PINEVIEW_CURSOR_MAX_WM,
462 .default_wm = PINEVIEW_CURSOR_DFT_WM,
463 .guard_size = PINEVIEW_CURSOR_GUARD_WM,
464 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED
465};
466static const struct intel_watermark_params g4x_wm_info = {
e0f0273e
VS
467 .fifo_size = G4X_FIFO_SIZE,
468 .max_wm = G4X_MAX_WM,
469 .default_wm = G4X_MAX_WM,
470 .guard_size = 2,
471 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED
472};
473static const struct intel_watermark_params g4x_cursor_wm_info = {
e0f0273e
VS
474 .fifo_size = I965_CURSOR_FIFO,
475 .max_wm = I965_CURSOR_MAX_WM,
476 .default_wm = I965_CURSOR_DFT_WM,
477 .guard_size = 2,
478 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED
479};
480static const struct intel_watermark_params valleyview_wm_info = {
e0f0273e
VS
481 .fifo_size = VALLEYVIEW_FIFO_SIZE,
482 .max_wm = VALLEYVIEW_MAX_WM,
483 .default_wm = VALLEYVIEW_MAX_WM,
484 .guard_size = 2,
485 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED
486};
487static const struct intel_watermark_params valleyview_cursor_wm_info = {
e0f0273e
VS
488 .fifo_size = I965_CURSOR_FIFO,
489 .max_wm = VALLEYVIEW_CURSOR_MAX_WM,
490 .default_wm = I965_CURSOR_DFT_WM,
491 .guard_size = 2,
492 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED
493};
494static const struct intel_watermark_params i965_cursor_wm_info = {
e0f0273e
VS
495 .fifo_size = I965_CURSOR_FIFO,
496 .max_wm = I965_CURSOR_MAX_WM,
497 .default_wm = I965_CURSOR_DFT_WM,
498 .guard_size = 2,
499 .cacheline_size = I915_FIFO_LINE_SIZE,
b445e3b0
ED
500};
501static const struct intel_watermark_params i945_wm_info = {
e0f0273e
VS
502 .fifo_size = I945_FIFO_SIZE,
503 .max_wm = I915_MAX_WM,
504 .default_wm = 1,
505 .guard_size = 2,
506 .cacheline_size = I915_FIFO_LINE_SIZE,
b445e3b0
ED
507};
508static const struct intel_watermark_params i915_wm_info = {
e0f0273e
VS
509 .fifo_size = I915_FIFO_SIZE,
510 .max_wm = I915_MAX_WM,
511 .default_wm = 1,
512 .guard_size = 2,
513 .cacheline_size = I915_FIFO_LINE_SIZE,
b445e3b0 514};
9d539105 515static const struct intel_watermark_params i830_a_wm_info = {
e0f0273e
VS
516 .fifo_size = I855GM_FIFO_SIZE,
517 .max_wm = I915_MAX_WM,
518 .default_wm = 1,
519 .guard_size = 2,
520 .cacheline_size = I830_FIFO_LINE_SIZE,
b445e3b0 521};
9d539105
VS
522static const struct intel_watermark_params i830_bc_wm_info = {
523 .fifo_size = I855GM_FIFO_SIZE,
524 .max_wm = I915_MAX_WM/2,
525 .default_wm = 1,
526 .guard_size = 2,
527 .cacheline_size = I830_FIFO_LINE_SIZE,
528};
feb56b93 529static const struct intel_watermark_params i845_wm_info = {
e0f0273e
VS
530 .fifo_size = I830_FIFO_SIZE,
531 .max_wm = I915_MAX_WM,
532 .default_wm = 1,
533 .guard_size = 2,
534 .cacheline_size = I830_FIFO_LINE_SIZE,
b445e3b0
ED
535};
536
b445e3b0
ED
537/**
538 * intel_calculate_wm - calculate watermark level
539 * @clock_in_khz: pixel clock
540 * @wm: chip FIFO params
541 * @pixel_size: display pixel size
542 * @latency_ns: memory latency for the platform
543 *
544 * Calculate the watermark level (the level at which the display plane will
545 * start fetching from memory again). Each chip has a different display
546 * FIFO size and allocation, so the caller needs to figure that out and pass
547 * in the correct intel_watermark_params structure.
548 *
549 * As the pixel clock runs, the FIFO will be drained at a rate that depends
550 * on the pixel size. When it reaches the watermark level, it'll start
551 * fetching FIFO line sized based chunks from memory until the FIFO fills
552 * past the watermark point. If the FIFO drains completely, a FIFO underrun
553 * will occur, and a display engine hang could result.
554 */
555static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
556 const struct intel_watermark_params *wm,
557 int fifo_size,
558 int pixel_size,
559 unsigned long latency_ns)
560{
561 long entries_required, wm_size;
562
563 /*
564 * Note: we need to make sure we don't overflow for various clock &
565 * latency values.
566 * clocks go from a few thousand to several hundred thousand.
567 * latency is usually a few thousand
568 */
569 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
570 1000;
571 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
572
573 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
574
575 wm_size = fifo_size - (entries_required + wm->guard_size);
576
577 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
578
579 /* Don't promote wm_size to unsigned... */
580 if (wm_size > (long)wm->max_wm)
581 wm_size = wm->max_wm;
582 if (wm_size <= 0)
583 wm_size = wm->default_wm;
d6feb196
VS
584
585 /*
586 * Bspec seems to indicate that the value shouldn't be lower than
587 * 'burst size + 1'. Certainly 830 is quite unhappy with low values.
588 * Lets go for 8 which is the burst size since certain platforms
589 * already use a hardcoded 8 (which is what the spec says should be
590 * done).
591 */
592 if (wm_size <= 8)
593 wm_size = 8;
594
b445e3b0
ED
595 return wm_size;
596}
597
598static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
599{
600 struct drm_crtc *crtc, *enabled = NULL;
601
70e1e0ec 602 for_each_crtc(dev, crtc) {
3490ea5d 603 if (intel_crtc_active(crtc)) {
b445e3b0
ED
604 if (enabled)
605 return NULL;
606 enabled = crtc;
607 }
608 }
609
610 return enabled;
611}
612
46ba614c 613static void pineview_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 614{
46ba614c 615 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
616 struct drm_i915_private *dev_priv = dev->dev_private;
617 struct drm_crtc *crtc;
618 const struct cxsr_latency *latency;
619 u32 reg;
620 unsigned long wm;
621
622 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
623 dev_priv->fsb_freq, dev_priv->mem_freq);
624 if (!latency) {
625 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
5209b1f4 626 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED
627 return;
628 }
629
630 crtc = single_enabled_crtc(dev);
631 if (crtc) {
7c5f93b0 632 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
59bea882 633 int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
7c5f93b0 634 int clock = adjusted_mode->crtc_clock;
b445e3b0
ED
635
636 /* Display SR */
637 wm = intel_calculate_wm(clock, &pineview_display_wm,
638 pineview_display_wm.fifo_size,
639 pixel_size, latency->display_sr);
640 reg = I915_READ(DSPFW1);
641 reg &= ~DSPFW_SR_MASK;
f4998963 642 reg |= FW_WM(wm, SR);
b445e3b0
ED
643 I915_WRITE(DSPFW1, reg);
644 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
645
646 /* cursor SR */
647 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
648 pineview_display_wm.fifo_size,
649 pixel_size, latency->cursor_sr);
650 reg = I915_READ(DSPFW3);
651 reg &= ~DSPFW_CURSOR_SR_MASK;
f4998963 652 reg |= FW_WM(wm, CURSOR_SR);
b445e3b0
ED
653 I915_WRITE(DSPFW3, reg);
654
655 /* Display HPLL off SR */
656 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
657 pineview_display_hplloff_wm.fifo_size,
658 pixel_size, latency->display_hpll_disable);
659 reg = I915_READ(DSPFW3);
660 reg &= ~DSPFW_HPLL_SR_MASK;
f4998963 661 reg |= FW_WM(wm, HPLL_SR);
b445e3b0
ED
662 I915_WRITE(DSPFW3, reg);
663
664 /* cursor HPLL off SR */
665 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
666 pineview_display_hplloff_wm.fifo_size,
667 pixel_size, latency->cursor_hpll_disable);
668 reg = I915_READ(DSPFW3);
669 reg &= ~DSPFW_HPLL_CURSOR_MASK;
f4998963 670 reg |= FW_WM(wm, HPLL_CURSOR);
b445e3b0
ED
671 I915_WRITE(DSPFW3, reg);
672 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
673
5209b1f4 674 intel_set_memory_cxsr(dev_priv, true);
b445e3b0 675 } else {
5209b1f4 676 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED
677 }
678}
679
680static bool g4x_compute_wm0(struct drm_device *dev,
681 int plane,
682 const struct intel_watermark_params *display,
683 int display_latency_ns,
684 const struct intel_watermark_params *cursor,
685 int cursor_latency_ns,
686 int *plane_wm,
687 int *cursor_wm)
688{
689 struct drm_crtc *crtc;
4fe8590a 690 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED
691 int htotal, hdisplay, clock, pixel_size;
692 int line_time_us, line_count;
693 int entries, tlb_miss;
694
695 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 696 if (!intel_crtc_active(crtc)) {
b445e3b0
ED
697 *cursor_wm = cursor->guard_size;
698 *plane_wm = display->guard_size;
699 return false;
700 }
701
6e3c9717 702 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 703 clock = adjusted_mode->crtc_clock;
fec8cba3 704 htotal = adjusted_mode->crtc_htotal;
6e3c9717 705 hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
59bea882 706 pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
b445e3b0
ED
707
708 /* Use the small buffer method to calculate plane watermark */
709 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
710 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
711 if (tlb_miss > 0)
712 entries += tlb_miss;
713 entries = DIV_ROUND_UP(entries, display->cacheline_size);
714 *plane_wm = entries + display->guard_size;
715 if (*plane_wm > (int)display->max_wm)
716 *plane_wm = display->max_wm;
717
718 /* Use the large buffer method to calculate cursor watermark */
922044c9 719 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0 720 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
3dd512fb 721 entries = line_count * crtc->cursor->state->crtc_w * pixel_size;
b445e3b0
ED
722 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
723 if (tlb_miss > 0)
724 entries += tlb_miss;
725 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
726 *cursor_wm = entries + cursor->guard_size;
727 if (*cursor_wm > (int)cursor->max_wm)
728 *cursor_wm = (int)cursor->max_wm;
729
730 return true;
731}
732
733/*
734 * Check the wm result.
735 *
736 * If any calculated watermark values is larger than the maximum value that
737 * can be programmed into the associated watermark register, that watermark
738 * must be disabled.
739 */
740static bool g4x_check_srwm(struct drm_device *dev,
741 int display_wm, int cursor_wm,
742 const struct intel_watermark_params *display,
743 const struct intel_watermark_params *cursor)
744{
745 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
746 display_wm, cursor_wm);
747
748 if (display_wm > display->max_wm) {
749 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
750 display_wm, display->max_wm);
751 return false;
752 }
753
754 if (cursor_wm > cursor->max_wm) {
755 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
756 cursor_wm, cursor->max_wm);
757 return false;
758 }
759
760 if (!(display_wm || cursor_wm)) {
761 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
762 return false;
763 }
764
765 return true;
766}
767
768static bool g4x_compute_srwm(struct drm_device *dev,
769 int plane,
770 int latency_ns,
771 const struct intel_watermark_params *display,
772 const struct intel_watermark_params *cursor,
773 int *display_wm, int *cursor_wm)
774{
775 struct drm_crtc *crtc;
4fe8590a 776 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED
777 int hdisplay, htotal, pixel_size, clock;
778 unsigned long line_time_us;
779 int line_count, line_size;
780 int small, large;
781 int entries;
782
783 if (!latency_ns) {
784 *display_wm = *cursor_wm = 0;
785 return false;
786 }
787
788 crtc = intel_get_crtc_for_plane(dev, plane);
6e3c9717 789 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 790 clock = adjusted_mode->crtc_clock;
fec8cba3 791 htotal = adjusted_mode->crtc_htotal;
6e3c9717 792 hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
59bea882 793 pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
b445e3b0 794
922044c9 795 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED
796 line_count = (latency_ns / line_time_us + 1000) / 1000;
797 line_size = hdisplay * pixel_size;
798
799 /* Use the minimum of the small and large buffer method for primary */
800 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
801 large = line_count * line_size;
802
803 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
804 *display_wm = entries + display->guard_size;
805
806 /* calculate the self-refresh watermark for display cursor */
3dd512fb 807 entries = line_count * pixel_size * crtc->cursor->state->crtc_w;
b445e3b0
ED
808 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
809 *cursor_wm = entries + cursor->guard_size;
810
811 return g4x_check_srwm(dev,
812 *display_wm, *cursor_wm,
813 display, cursor);
814}
815
15665979
VS
816#define FW_WM_VLV(value, plane) \
817 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV)
818
0018fda1
VS
819static void vlv_write_wm_values(struct intel_crtc *crtc,
820 const struct vlv_wm_values *wm)
821{
822 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
823 enum pipe pipe = crtc->pipe;
824
825 I915_WRITE(VLV_DDL(pipe),
826 (wm->ddl[pipe].cursor << DDL_CURSOR_SHIFT) |
827 (wm->ddl[pipe].sprite[1] << DDL_SPRITE_SHIFT(1)) |
828 (wm->ddl[pipe].sprite[0] << DDL_SPRITE_SHIFT(0)) |
829 (wm->ddl[pipe].primary << DDL_PLANE_SHIFT));
830
ae80152d 831 I915_WRITE(DSPFW1,
15665979
VS
832 FW_WM(wm->sr.plane, SR) |
833 FW_WM(wm->pipe[PIPE_B].cursor, CURSORB) |
834 FW_WM_VLV(wm->pipe[PIPE_B].primary, PLANEB) |
835 FW_WM_VLV(wm->pipe[PIPE_A].primary, PLANEA));
ae80152d 836 I915_WRITE(DSPFW2,
15665979
VS
837 FW_WM_VLV(wm->pipe[PIPE_A].sprite[1], SPRITEB) |
838 FW_WM(wm->pipe[PIPE_A].cursor, CURSORA) |
839 FW_WM_VLV(wm->pipe[PIPE_A].sprite[0], SPRITEA));
ae80152d 840 I915_WRITE(DSPFW3,
15665979 841 FW_WM(wm->sr.cursor, CURSOR_SR));
ae80152d
VS
842
843 if (IS_CHERRYVIEW(dev_priv)) {
844 I915_WRITE(DSPFW7_CHV,
15665979
VS
845 FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) |
846 FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC));
ae80152d 847 I915_WRITE(DSPFW8_CHV,
15665979
VS
848 FW_WM_VLV(wm->pipe[PIPE_C].sprite[1], SPRITEF) |
849 FW_WM_VLV(wm->pipe[PIPE_C].sprite[0], SPRITEE));
ae80152d 850 I915_WRITE(DSPFW9_CHV,
15665979
VS
851 FW_WM_VLV(wm->pipe[PIPE_C].primary, PLANEC) |
852 FW_WM(wm->pipe[PIPE_C].cursor, CURSORC));
ae80152d 853 I915_WRITE(DSPHOWM,
15665979
VS
854 FW_WM(wm->sr.plane >> 9, SR_HI) |
855 FW_WM(wm->pipe[PIPE_C].sprite[1] >> 8, SPRITEF_HI) |
856 FW_WM(wm->pipe[PIPE_C].sprite[0] >> 8, SPRITEE_HI) |
857 FW_WM(wm->pipe[PIPE_C].primary >> 8, PLANEC_HI) |
858 FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) |
859 FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) |
860 FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) |
861 FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) |
862 FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) |
863 FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI));
ae80152d
VS
864 } else {
865 I915_WRITE(DSPFW7,
15665979
VS
866 FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) |
867 FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC));
ae80152d 868 I915_WRITE(DSPHOWM,
15665979
VS
869 FW_WM(wm->sr.plane >> 9, SR_HI) |
870 FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) |
871 FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) |
872 FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) |
873 FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) |
874 FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) |
875 FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI));
ae80152d
VS
876 }
877
2cb389b7
VS
878 /* zero (unused) WM1 watermarks */
879 I915_WRITE(DSPFW4, 0);
880 I915_WRITE(DSPFW5, 0);
881 I915_WRITE(DSPFW6, 0);
882 I915_WRITE(DSPHOWM1, 0);
883
ae80152d 884 POSTING_READ(DSPFW1);
0018fda1
VS
885}
886
15665979
VS
887#undef FW_WM_VLV
888
6eb1a681
VS
889enum vlv_wm_level {
890 VLV_WM_LEVEL_PM2,
891 VLV_WM_LEVEL_PM5,
892 VLV_WM_LEVEL_DDR_DVFS,
6eb1a681
VS
893};
894
262cd2e1
VS
895/* latency must be in 0.1us units. */
896static unsigned int vlv_wm_method2(unsigned int pixel_rate,
897 unsigned int pipe_htotal,
898 unsigned int horiz_pixels,
899 unsigned int bytes_per_pixel,
900 unsigned int latency)
901{
902 unsigned int ret;
903
904 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
905 ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
906 ret = DIV_ROUND_UP(ret, 64);
907
908 return ret;
909}
910
911static void vlv_setup_wm_latency(struct drm_device *dev)
912{
913 struct drm_i915_private *dev_priv = dev->dev_private;
914
915 /* all latencies in usec */
916 dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3;
917
58590c14
VS
918 dev_priv->wm.max_level = VLV_WM_LEVEL_PM2;
919
262cd2e1
VS
920 if (IS_CHERRYVIEW(dev_priv)) {
921 dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12;
922 dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33;
58590c14
VS
923
924 dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS;
262cd2e1
VS
925 }
926}
927
928static uint16_t vlv_compute_wm_level(struct intel_plane *plane,
929 struct intel_crtc *crtc,
930 const struct intel_plane_state *state,
931 int level)
932{
933 struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
934 int clock, htotal, pixel_size, width, wm;
935
936 if (dev_priv->wm.pri_latency[level] == 0)
937 return USHRT_MAX;
938
939 if (!state->visible)
940 return 0;
941
942 pixel_size = drm_format_plane_cpp(state->base.fb->pixel_format, 0);
943 clock = crtc->config->base.adjusted_mode.crtc_clock;
944 htotal = crtc->config->base.adjusted_mode.crtc_htotal;
945 width = crtc->config->pipe_src_w;
946 if (WARN_ON(htotal == 0))
947 htotal = 1;
948
949 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
950 /*
951 * FIXME the formula gives values that are
952 * too big for the cursor FIFO, and hence we
953 * would never be able to use cursors. For
954 * now just hardcode the watermark.
955 */
956 wm = 63;
957 } else {
958 wm = vlv_wm_method2(clock, htotal, width, pixel_size,
959 dev_priv->wm.pri_latency[level] * 10);
960 }
961
962 return min_t(int, wm, USHRT_MAX);
963}
964
54f1b6e1
VS
965static void vlv_compute_fifo(struct intel_crtc *crtc)
966{
967 struct drm_device *dev = crtc->base.dev;
968 struct vlv_wm_state *wm_state = &crtc->wm_state;
969 struct intel_plane *plane;
970 unsigned int total_rate = 0;
971 const int fifo_size = 512 - 1;
972 int fifo_extra, fifo_left = fifo_size;
973
974 for_each_intel_plane_on_crtc(dev, crtc, plane) {
975 struct intel_plane_state *state =
976 to_intel_plane_state(plane->base.state);
977
978 if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
979 continue;
980
981 if (state->visible) {
982 wm_state->num_active_planes++;
983 total_rate += drm_format_plane_cpp(state->base.fb->pixel_format, 0);
984 }
985 }
986
987 for_each_intel_plane_on_crtc(dev, crtc, plane) {
988 struct intel_plane_state *state =
989 to_intel_plane_state(plane->base.state);
990 unsigned int rate;
991
992 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
993 plane->wm.fifo_size = 63;
994 continue;
995 }
996
997 if (!state->visible) {
998 plane->wm.fifo_size = 0;
999 continue;
1000 }
1001
1002 rate = drm_format_plane_cpp(state->base.fb->pixel_format, 0);
1003 plane->wm.fifo_size = fifo_size * rate / total_rate;
1004 fifo_left -= plane->wm.fifo_size;
1005 }
1006
1007 fifo_extra = DIV_ROUND_UP(fifo_left, wm_state->num_active_planes ?: 1);
1008
1009 /* spread the remainder evenly */
1010 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1011 int plane_extra;
1012
1013 if (fifo_left == 0)
1014 break;
1015
1016 if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
1017 continue;
1018
1019 /* give it all to the first plane if none are active */
1020 if (plane->wm.fifo_size == 0 &&
1021 wm_state->num_active_planes)
1022 continue;
1023
1024 plane_extra = min(fifo_extra, fifo_left);
1025 plane->wm.fifo_size += plane_extra;
1026 fifo_left -= plane_extra;
1027 }
1028
1029 WARN_ON(fifo_left != 0);
1030}
1031
262cd2e1
VS
1032static void vlv_invert_wms(struct intel_crtc *crtc)
1033{
1034 struct vlv_wm_state *wm_state = &crtc->wm_state;
1035 int level;
1036
1037 for (level = 0; level < wm_state->num_levels; level++) {
1038 struct drm_device *dev = crtc->base.dev;
1039 const int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1;
1040 struct intel_plane *plane;
1041
1042 wm_state->sr[level].plane = sr_fifo_size - wm_state->sr[level].plane;
1043 wm_state->sr[level].cursor = 63 - wm_state->sr[level].cursor;
1044
1045 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1046 switch (plane->base.type) {
1047 int sprite;
1048 case DRM_PLANE_TYPE_CURSOR:
1049 wm_state->wm[level].cursor = plane->wm.fifo_size -
1050 wm_state->wm[level].cursor;
1051 break;
1052 case DRM_PLANE_TYPE_PRIMARY:
1053 wm_state->wm[level].primary = plane->wm.fifo_size -
1054 wm_state->wm[level].primary;
1055 break;
1056 case DRM_PLANE_TYPE_OVERLAY:
1057 sprite = plane->plane;
1058 wm_state->wm[level].sprite[sprite] = plane->wm.fifo_size -
1059 wm_state->wm[level].sprite[sprite];
1060 break;
1061 }
1062 }
1063 }
1064}
1065
26e1fe4f 1066static void vlv_compute_wm(struct intel_crtc *crtc)
262cd2e1
VS
1067{
1068 struct drm_device *dev = crtc->base.dev;
1069 struct vlv_wm_state *wm_state = &crtc->wm_state;
1070 struct intel_plane *plane;
1071 int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1;
1072 int level;
1073
1074 memset(wm_state, 0, sizeof(*wm_state));
1075
852eb00d 1076 wm_state->cxsr = crtc->pipe != PIPE_C && crtc->wm.cxsr_allowed;
58590c14 1077 wm_state->num_levels = to_i915(dev)->wm.max_level + 1;
262cd2e1
VS
1078
1079 wm_state->num_active_planes = 0;
262cd2e1 1080
54f1b6e1 1081 vlv_compute_fifo(crtc);
262cd2e1
VS
1082
1083 if (wm_state->num_active_planes != 1)
1084 wm_state->cxsr = false;
1085
1086 if (wm_state->cxsr) {
1087 for (level = 0; level < wm_state->num_levels; level++) {
1088 wm_state->sr[level].plane = sr_fifo_size;
1089 wm_state->sr[level].cursor = 63;
1090 }
1091 }
1092
1093 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1094 struct intel_plane_state *state =
1095 to_intel_plane_state(plane->base.state);
1096
1097 if (!state->visible)
1098 continue;
1099
1100 /* normal watermarks */
1101 for (level = 0; level < wm_state->num_levels; level++) {
1102 int wm = vlv_compute_wm_level(plane, crtc, state, level);
1103 int max_wm = plane->base.type == DRM_PLANE_TYPE_CURSOR ? 63 : 511;
1104
1105 /* hack */
1106 if (WARN_ON(level == 0 && wm > max_wm))
1107 wm = max_wm;
1108
1109 if (wm > plane->wm.fifo_size)
1110 break;
1111
1112 switch (plane->base.type) {
1113 int sprite;
1114 case DRM_PLANE_TYPE_CURSOR:
1115 wm_state->wm[level].cursor = wm;
1116 break;
1117 case DRM_PLANE_TYPE_PRIMARY:
1118 wm_state->wm[level].primary = wm;
1119 break;
1120 case DRM_PLANE_TYPE_OVERLAY:
1121 sprite = plane->plane;
1122 wm_state->wm[level].sprite[sprite] = wm;
1123 break;
1124 }
1125 }
1126
1127 wm_state->num_levels = level;
1128
1129 if (!wm_state->cxsr)
1130 continue;
1131
1132 /* maxfifo watermarks */
1133 switch (plane->base.type) {
1134 int sprite, level;
1135 case DRM_PLANE_TYPE_CURSOR:
1136 for (level = 0; level < wm_state->num_levels; level++)
1137 wm_state->sr[level].cursor =
5a37ed0a 1138 wm_state->wm[level].cursor;
262cd2e1
VS
1139 break;
1140 case DRM_PLANE_TYPE_PRIMARY:
1141 for (level = 0; level < wm_state->num_levels; level++)
1142 wm_state->sr[level].plane =
1143 min(wm_state->sr[level].plane,
1144 wm_state->wm[level].primary);
1145 break;
1146 case DRM_PLANE_TYPE_OVERLAY:
1147 sprite = plane->plane;
1148 for (level = 0; level < wm_state->num_levels; level++)
1149 wm_state->sr[level].plane =
1150 min(wm_state->sr[level].plane,
1151 wm_state->wm[level].sprite[sprite]);
1152 break;
1153 }
1154 }
1155
1156 /* clear any (partially) filled invalid levels */
58590c14 1157 for (level = wm_state->num_levels; level < to_i915(dev)->wm.max_level + 1; level++) {
262cd2e1
VS
1158 memset(&wm_state->wm[level], 0, sizeof(wm_state->wm[level]));
1159 memset(&wm_state->sr[level], 0, sizeof(wm_state->sr[level]));
1160 }
1161
1162 vlv_invert_wms(crtc);
1163}
1164
54f1b6e1
VS
1165#define VLV_FIFO(plane, value) \
1166 (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV)
1167
1168static void vlv_pipe_set_fifo_size(struct intel_crtc *crtc)
1169{
1170 struct drm_device *dev = crtc->base.dev;
1171 struct drm_i915_private *dev_priv = to_i915(dev);
1172 struct intel_plane *plane;
1173 int sprite0_start = 0, sprite1_start = 0, fifo_size = 0;
1174
1175 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1176 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
1177 WARN_ON(plane->wm.fifo_size != 63);
1178 continue;
1179 }
1180
1181 if (plane->base.type == DRM_PLANE_TYPE_PRIMARY)
1182 sprite0_start = plane->wm.fifo_size;
1183 else if (plane->plane == 0)
1184 sprite1_start = sprite0_start + plane->wm.fifo_size;
1185 else
1186 fifo_size = sprite1_start + plane->wm.fifo_size;
1187 }
1188
1189 WARN_ON(fifo_size != 512 - 1);
1190
1191 DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n",
1192 pipe_name(crtc->pipe), sprite0_start,
1193 sprite1_start, fifo_size);
1194
1195 switch (crtc->pipe) {
1196 uint32_t dsparb, dsparb2, dsparb3;
1197 case PIPE_A:
1198 dsparb = I915_READ(DSPARB);
1199 dsparb2 = I915_READ(DSPARB2);
1200
1201 dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) |
1202 VLV_FIFO(SPRITEB, 0xff));
1203 dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) |
1204 VLV_FIFO(SPRITEB, sprite1_start));
1205
1206 dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) |
1207 VLV_FIFO(SPRITEB_HI, 0x1));
1208 dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) |
1209 VLV_FIFO(SPRITEB_HI, sprite1_start >> 8));
1210
1211 I915_WRITE(DSPARB, dsparb);
1212 I915_WRITE(DSPARB2, dsparb2);
1213 break;
1214 case PIPE_B:
1215 dsparb = I915_READ(DSPARB);
1216 dsparb2 = I915_READ(DSPARB2);
1217
1218 dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) |
1219 VLV_FIFO(SPRITED, 0xff));
1220 dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) |
1221 VLV_FIFO(SPRITED, sprite1_start));
1222
1223 dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) |
1224 VLV_FIFO(SPRITED_HI, 0xff));
1225 dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) |
1226 VLV_FIFO(SPRITED_HI, sprite1_start >> 8));
1227
1228 I915_WRITE(DSPARB, dsparb);
1229 I915_WRITE(DSPARB2, dsparb2);
1230 break;
1231 case PIPE_C:
1232 dsparb3 = I915_READ(DSPARB3);
1233 dsparb2 = I915_READ(DSPARB2);
1234
1235 dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) |
1236 VLV_FIFO(SPRITEF, 0xff));
1237 dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) |
1238 VLV_FIFO(SPRITEF, sprite1_start));
1239
1240 dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) |
1241 VLV_FIFO(SPRITEF_HI, 0xff));
1242 dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) |
1243 VLV_FIFO(SPRITEF_HI, sprite1_start >> 8));
1244
1245 I915_WRITE(DSPARB3, dsparb3);
1246 I915_WRITE(DSPARB2, dsparb2);
1247 break;
1248 default:
1249 break;
1250 }
1251}
1252
1253#undef VLV_FIFO
1254
262cd2e1
VS
1255static void vlv_merge_wm(struct drm_device *dev,
1256 struct vlv_wm_values *wm)
1257{
1258 struct intel_crtc *crtc;
1259 int num_active_crtcs = 0;
1260
58590c14 1261 wm->level = to_i915(dev)->wm.max_level;
262cd2e1
VS
1262 wm->cxsr = true;
1263
1264 for_each_intel_crtc(dev, crtc) {
1265 const struct vlv_wm_state *wm_state = &crtc->wm_state;
1266
1267 if (!crtc->active)
1268 continue;
1269
1270 if (!wm_state->cxsr)
1271 wm->cxsr = false;
1272
1273 num_active_crtcs++;
1274 wm->level = min_t(int, wm->level, wm_state->num_levels - 1);
1275 }
1276
1277 if (num_active_crtcs != 1)
1278 wm->cxsr = false;
1279
6f9c784b
VS
1280 if (num_active_crtcs > 1)
1281 wm->level = VLV_WM_LEVEL_PM2;
1282
262cd2e1
VS
1283 for_each_intel_crtc(dev, crtc) {
1284 struct vlv_wm_state *wm_state = &crtc->wm_state;
1285 enum pipe pipe = crtc->pipe;
1286
1287 if (!crtc->active)
1288 continue;
1289
1290 wm->pipe[pipe] = wm_state->wm[wm->level];
1291 if (wm->cxsr)
1292 wm->sr = wm_state->sr[wm->level];
1293
1294 wm->ddl[pipe].primary = DDL_PRECISION_HIGH | 2;
1295 wm->ddl[pipe].sprite[0] = DDL_PRECISION_HIGH | 2;
1296 wm->ddl[pipe].sprite[1] = DDL_PRECISION_HIGH | 2;
1297 wm->ddl[pipe].cursor = DDL_PRECISION_HIGH | 2;
1298 }
1299}
1300
1301static void vlv_update_wm(struct drm_crtc *crtc)
1302{
1303 struct drm_device *dev = crtc->dev;
1304 struct drm_i915_private *dev_priv = dev->dev_private;
1305 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
1306 enum pipe pipe = intel_crtc->pipe;
1307 struct vlv_wm_values wm = {};
1308
26e1fe4f 1309 vlv_compute_wm(intel_crtc);
262cd2e1
VS
1310 vlv_merge_wm(dev, &wm);
1311
54f1b6e1
VS
1312 if (memcmp(&dev_priv->wm.vlv, &wm, sizeof(wm)) == 0) {
1313 /* FIXME should be part of crtc atomic commit */
1314 vlv_pipe_set_fifo_size(intel_crtc);
262cd2e1 1315 return;
54f1b6e1 1316 }
262cd2e1
VS
1317
1318 if (wm.level < VLV_WM_LEVEL_DDR_DVFS &&
1319 dev_priv->wm.vlv.level >= VLV_WM_LEVEL_DDR_DVFS)
1320 chv_set_memory_dvfs(dev_priv, false);
1321
1322 if (wm.level < VLV_WM_LEVEL_PM5 &&
1323 dev_priv->wm.vlv.level >= VLV_WM_LEVEL_PM5)
1324 chv_set_memory_pm5(dev_priv, false);
1325
852eb00d 1326 if (!wm.cxsr && dev_priv->wm.vlv.cxsr)
262cd2e1 1327 intel_set_memory_cxsr(dev_priv, false);
262cd2e1 1328
54f1b6e1
VS
1329 /* FIXME should be part of crtc atomic commit */
1330 vlv_pipe_set_fifo_size(intel_crtc);
1331
262cd2e1
VS
1332 vlv_write_wm_values(intel_crtc, &wm);
1333
1334 DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, "
1335 "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n",
1336 pipe_name(pipe), wm.pipe[pipe].primary, wm.pipe[pipe].cursor,
1337 wm.pipe[pipe].sprite[0], wm.pipe[pipe].sprite[1],
1338 wm.sr.plane, wm.sr.cursor, wm.level, wm.cxsr);
1339
852eb00d 1340 if (wm.cxsr && !dev_priv->wm.vlv.cxsr)
262cd2e1 1341 intel_set_memory_cxsr(dev_priv, true);
262cd2e1
VS
1342
1343 if (wm.level >= VLV_WM_LEVEL_PM5 &&
1344 dev_priv->wm.vlv.level < VLV_WM_LEVEL_PM5)
1345 chv_set_memory_pm5(dev_priv, true);
1346
1347 if (wm.level >= VLV_WM_LEVEL_DDR_DVFS &&
1348 dev_priv->wm.vlv.level < VLV_WM_LEVEL_DDR_DVFS)
1349 chv_set_memory_dvfs(dev_priv, true);
1350
1351 dev_priv->wm.vlv = wm;
3c2777fd
VS
1352}
1353
ae80152d
VS
1354#define single_plane_enabled(mask) is_power_of_2(mask)
1355
46ba614c 1356static void g4x_update_wm(struct drm_crtc *crtc)
b445e3b0 1357{
46ba614c 1358 struct drm_device *dev = crtc->dev;
b445e3b0
ED
1359 static const int sr_latency_ns = 12000;
1360 struct drm_i915_private *dev_priv = dev->dev_private;
1361 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1362 int plane_sr, cursor_sr;
1363 unsigned int enabled = 0;
9858425c 1364 bool cxsr_enabled;
b445e3b0 1365
51cea1f4 1366 if (g4x_compute_wm0(dev, PIPE_A,
5aef6003
CW
1367 &g4x_wm_info, pessimal_latency_ns,
1368 &g4x_cursor_wm_info, pessimal_latency_ns,
b445e3b0 1369 &planea_wm, &cursora_wm))
51cea1f4 1370 enabled |= 1 << PIPE_A;
b445e3b0 1371
51cea1f4 1372 if (g4x_compute_wm0(dev, PIPE_B,
5aef6003
CW
1373 &g4x_wm_info, pessimal_latency_ns,
1374 &g4x_cursor_wm_info, pessimal_latency_ns,
b445e3b0 1375 &planeb_wm, &cursorb_wm))
51cea1f4 1376 enabled |= 1 << PIPE_B;
b445e3b0 1377
b445e3b0
ED
1378 if (single_plane_enabled(enabled) &&
1379 g4x_compute_srwm(dev, ffs(enabled) - 1,
1380 sr_latency_ns,
1381 &g4x_wm_info,
1382 &g4x_cursor_wm_info,
52bd02d8 1383 &plane_sr, &cursor_sr)) {
9858425c 1384 cxsr_enabled = true;
52bd02d8 1385 } else {
9858425c 1386 cxsr_enabled = false;
5209b1f4 1387 intel_set_memory_cxsr(dev_priv, false);
52bd02d8
CW
1388 plane_sr = cursor_sr = 0;
1389 }
b445e3b0 1390
a5043453
VS
1391 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, "
1392 "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
b445e3b0
ED
1393 planea_wm, cursora_wm,
1394 planeb_wm, cursorb_wm,
1395 plane_sr, cursor_sr);
1396
1397 I915_WRITE(DSPFW1,
f4998963
VS
1398 FW_WM(plane_sr, SR) |
1399 FW_WM(cursorb_wm, CURSORB) |
1400 FW_WM(planeb_wm, PLANEB) |
1401 FW_WM(planea_wm, PLANEA));
b445e3b0 1402 I915_WRITE(DSPFW2,
8c919b28 1403 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
f4998963 1404 FW_WM(cursora_wm, CURSORA));
b445e3b0
ED
1405 /* HPLL off in SR has some issues on G4x... disable it */
1406 I915_WRITE(DSPFW3,
8c919b28 1407 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
f4998963 1408 FW_WM(cursor_sr, CURSOR_SR));
9858425c
ID
1409
1410 if (cxsr_enabled)
1411 intel_set_memory_cxsr(dev_priv, true);
b445e3b0
ED
1412}
1413
46ba614c 1414static void i965_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1415{
46ba614c 1416 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1417 struct drm_i915_private *dev_priv = dev->dev_private;
1418 struct drm_crtc *crtc;
1419 int srwm = 1;
1420 int cursor_sr = 16;
9858425c 1421 bool cxsr_enabled;
b445e3b0
ED
1422
1423 /* Calc sr entries for one plane configs */
1424 crtc = single_enabled_crtc(dev);
1425 if (crtc) {
1426 /* self-refresh has much higher latency */
1427 static const int sr_latency_ns = 12000;
124abe07 1428 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1429 int clock = adjusted_mode->crtc_clock;
fec8cba3 1430 int htotal = adjusted_mode->crtc_htotal;
6e3c9717 1431 int hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
59bea882 1432 int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
b445e3b0
ED
1433 unsigned long line_time_us;
1434 int entries;
1435
922044c9 1436 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED
1437
1438 /* Use ns/us then divide to preserve precision */
1439 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1440 pixel_size * hdisplay;
1441 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1442 srwm = I965_FIFO_SIZE - entries;
1443 if (srwm < 0)
1444 srwm = 1;
1445 srwm &= 0x1ff;
1446 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1447 entries, srwm);
1448
1449 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
3dd512fb 1450 pixel_size * crtc->cursor->state->crtc_w;
b445e3b0
ED
1451 entries = DIV_ROUND_UP(entries,
1452 i965_cursor_wm_info.cacheline_size);
1453 cursor_sr = i965_cursor_wm_info.fifo_size -
1454 (entries + i965_cursor_wm_info.guard_size);
1455
1456 if (cursor_sr > i965_cursor_wm_info.max_wm)
1457 cursor_sr = i965_cursor_wm_info.max_wm;
1458
1459 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1460 "cursor %d\n", srwm, cursor_sr);
1461
9858425c 1462 cxsr_enabled = true;
b445e3b0 1463 } else {
9858425c 1464 cxsr_enabled = false;
b445e3b0 1465 /* Turn off self refresh if both pipes are enabled */
5209b1f4 1466 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED
1467 }
1468
1469 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1470 srwm);
1471
1472 /* 965 has limitations... */
f4998963
VS
1473 I915_WRITE(DSPFW1, FW_WM(srwm, SR) |
1474 FW_WM(8, CURSORB) |
1475 FW_WM(8, PLANEB) |
1476 FW_WM(8, PLANEA));
1477 I915_WRITE(DSPFW2, FW_WM(8, CURSORA) |
1478 FW_WM(8, PLANEC_OLD));
b445e3b0 1479 /* update cursor SR watermark */
f4998963 1480 I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR));
9858425c
ID
1481
1482 if (cxsr_enabled)
1483 intel_set_memory_cxsr(dev_priv, true);
b445e3b0
ED
1484}
1485
f4998963
VS
1486#undef FW_WM
1487
46ba614c 1488static void i9xx_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1489{
46ba614c 1490 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1491 struct drm_i915_private *dev_priv = dev->dev_private;
1492 const struct intel_watermark_params *wm_info;
1493 uint32_t fwater_lo;
1494 uint32_t fwater_hi;
1495 int cwm, srwm = 1;
1496 int fifo_size;
1497 int planea_wm, planeb_wm;
1498 struct drm_crtc *crtc, *enabled = NULL;
1499
1500 if (IS_I945GM(dev))
1501 wm_info = &i945_wm_info;
1502 else if (!IS_GEN2(dev))
1503 wm_info = &i915_wm_info;
1504 else
9d539105 1505 wm_info = &i830_a_wm_info;
b445e3b0
ED
1506
1507 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1508 crtc = intel_get_crtc_for_plane(dev, 0);
3490ea5d 1509 if (intel_crtc_active(crtc)) {
241bfc38 1510 const struct drm_display_mode *adjusted_mode;
59bea882 1511 int cpp = crtc->primary->state->fb->bits_per_pixel / 8;
b9e0bda3
CW
1512 if (IS_GEN2(dev))
1513 cpp = 4;
1514
6e3c9717 1515 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1516 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1517 wm_info, fifo_size, cpp,
5aef6003 1518 pessimal_latency_ns);
b445e3b0 1519 enabled = crtc;
9d539105 1520 } else {
b445e3b0 1521 planea_wm = fifo_size - wm_info->guard_size;
9d539105
VS
1522 if (planea_wm > (long)wm_info->max_wm)
1523 planea_wm = wm_info->max_wm;
1524 }
1525
1526 if (IS_GEN2(dev))
1527 wm_info = &i830_bc_wm_info;
b445e3b0
ED
1528
1529 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1530 crtc = intel_get_crtc_for_plane(dev, 1);
3490ea5d 1531 if (intel_crtc_active(crtc)) {
241bfc38 1532 const struct drm_display_mode *adjusted_mode;
59bea882 1533 int cpp = crtc->primary->state->fb->bits_per_pixel / 8;
b9e0bda3
CW
1534 if (IS_GEN2(dev))
1535 cpp = 4;
1536
6e3c9717 1537 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1538 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1539 wm_info, fifo_size, cpp,
5aef6003 1540 pessimal_latency_ns);
b445e3b0
ED
1541 if (enabled == NULL)
1542 enabled = crtc;
1543 else
1544 enabled = NULL;
9d539105 1545 } else {
b445e3b0 1546 planeb_wm = fifo_size - wm_info->guard_size;
9d539105
VS
1547 if (planeb_wm > (long)wm_info->max_wm)
1548 planeb_wm = wm_info->max_wm;
1549 }
b445e3b0
ED
1550
1551 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1552
2ab1bc9d 1553 if (IS_I915GM(dev) && enabled) {
2ff8fde1 1554 struct drm_i915_gem_object *obj;
2ab1bc9d 1555
59bea882 1556 obj = intel_fb_obj(enabled->primary->state->fb);
2ab1bc9d
DV
1557
1558 /* self-refresh seems busted with untiled */
2ff8fde1 1559 if (obj->tiling_mode == I915_TILING_NONE)
2ab1bc9d
DV
1560 enabled = NULL;
1561 }
1562
b445e3b0
ED
1563 /*
1564 * Overlay gets an aggressive default since video jitter is bad.
1565 */
1566 cwm = 2;
1567
1568 /* Play safe and disable self-refresh before adjusting watermarks. */
5209b1f4 1569 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED
1570
1571 /* Calc sr entries for one plane configs */
1572 if (HAS_FW_BLC(dev) && enabled) {
1573 /* self-refresh has much higher latency */
1574 static const int sr_latency_ns = 6000;
124abe07 1575 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(enabled)->config->base.adjusted_mode;
241bfc38 1576 int clock = adjusted_mode->crtc_clock;
fec8cba3 1577 int htotal = adjusted_mode->crtc_htotal;
6e3c9717 1578 int hdisplay = to_intel_crtc(enabled)->config->pipe_src_w;
59bea882 1579 int pixel_size = enabled->primary->state->fb->bits_per_pixel / 8;
b445e3b0
ED
1580 unsigned long line_time_us;
1581 int entries;
1582
922044c9 1583 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED
1584
1585 /* Use ns/us then divide to preserve precision */
1586 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1587 pixel_size * hdisplay;
1588 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1589 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1590 srwm = wm_info->fifo_size - entries;
1591 if (srwm < 0)
1592 srwm = 1;
1593
1594 if (IS_I945G(dev) || IS_I945GM(dev))
1595 I915_WRITE(FW_BLC_SELF,
1596 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1597 else if (IS_I915GM(dev))
1598 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1599 }
1600
1601 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1602 planea_wm, planeb_wm, cwm, srwm);
1603
1604 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1605 fwater_hi = (cwm & 0x1f);
1606
1607 /* Set request length to 8 cachelines per fetch */
1608 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1609 fwater_hi = fwater_hi | (1 << 8);
1610
1611 I915_WRITE(FW_BLC, fwater_lo);
1612 I915_WRITE(FW_BLC2, fwater_hi);
1613
5209b1f4
ID
1614 if (enabled)
1615 intel_set_memory_cxsr(dev_priv, true);
b445e3b0
ED
1616}
1617
feb56b93 1618static void i845_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1619{
46ba614c 1620 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1621 struct drm_i915_private *dev_priv = dev->dev_private;
1622 struct drm_crtc *crtc;
241bfc38 1623 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED
1624 uint32_t fwater_lo;
1625 int planea_wm;
1626
1627 crtc = single_enabled_crtc(dev);
1628 if (crtc == NULL)
1629 return;
1630
6e3c9717 1631 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1632 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
feb56b93 1633 &i845_wm_info,
b445e3b0 1634 dev_priv->display.get_fifo_size(dev, 0),
5aef6003 1635 4, pessimal_latency_ns);
b445e3b0
ED
1636 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1637 fwater_lo |= (3<<8) | planea_wm;
1638
1639 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1640
1641 I915_WRITE(FW_BLC, fwater_lo);
1642}
1643
8cfb3407 1644uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state *pipe_config)
801bcfff 1645{
fd4daa9c 1646 uint32_t pixel_rate;
801bcfff 1647
8cfb3407 1648 pixel_rate = pipe_config->base.adjusted_mode.crtc_clock;
801bcfff
PZ
1649
1650 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1651 * adjust the pixel_rate here. */
1652
8cfb3407 1653 if (pipe_config->pch_pfit.enabled) {
801bcfff 1654 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
8cfb3407
VS
1655 uint32_t pfit_size = pipe_config->pch_pfit.size;
1656
1657 pipe_w = pipe_config->pipe_src_w;
1658 pipe_h = pipe_config->pipe_src_h;
801bcfff 1659
801bcfff
PZ
1660 pfit_w = (pfit_size >> 16) & 0xFFFF;
1661 pfit_h = pfit_size & 0xFFFF;
1662 if (pipe_w < pfit_w)
1663 pipe_w = pfit_w;
1664 if (pipe_h < pfit_h)
1665 pipe_h = pfit_h;
1666
1667 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
1668 pfit_w * pfit_h);
1669 }
1670
1671 return pixel_rate;
1672}
1673
37126462 1674/* latency must be in 0.1us units. */
23297044 1675static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
801bcfff
PZ
1676 uint32_t latency)
1677{
1678 uint64_t ret;
1679
3312ba65
VS
1680 if (WARN(latency == 0, "Latency value missing\n"))
1681 return UINT_MAX;
1682
801bcfff
PZ
1683 ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
1684 ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
1685
1686 return ret;
1687}
1688
37126462 1689/* latency must be in 0.1us units. */
23297044 1690static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
801bcfff
PZ
1691 uint32_t horiz_pixels, uint8_t bytes_per_pixel,
1692 uint32_t latency)
1693{
1694 uint32_t ret;
1695
3312ba65
VS
1696 if (WARN(latency == 0, "Latency value missing\n"))
1697 return UINT_MAX;
1698
801bcfff
PZ
1699 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
1700 ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
1701 ret = DIV_ROUND_UP(ret, 64) + 2;
1702 return ret;
1703}
1704
23297044 1705static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
cca32e9a
PZ
1706 uint8_t bytes_per_pixel)
1707{
1708 return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
1709}
1710
2791a16c
PZ
1711struct skl_pipe_wm_parameters {
1712 bool active;
1713 uint32_t pipe_htotal;
1714 uint32_t pixel_rate; /* in KHz */
1715 struct intel_plane_wm_parameters plane[I915_MAX_PLANES];
1716};
1717
820c1980 1718struct ilk_wm_maximums {
cca32e9a
PZ
1719 uint16_t pri;
1720 uint16_t spr;
1721 uint16_t cur;
1722 uint16_t fbc;
1723};
1724
261a27d1
MR
1725/* used in computing the new watermarks state */
1726struct intel_wm_config {
1727 unsigned int num_pipes_active;
1728 bool sprites_enabled;
1729 bool sprites_scaled;
1730};
1731
37126462
VS
1732/*
1733 * For both WM_PIPE and WM_LP.
1734 * mem_value must be in 0.1us units.
1735 */
7221fc33 1736static uint32_t ilk_compute_pri_wm(const struct intel_crtc_state *cstate,
43d59eda 1737 const struct intel_plane_state *pstate,
cca32e9a
PZ
1738 uint32_t mem_value,
1739 bool is_lp)
801bcfff 1740{
43d59eda 1741 int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
cca32e9a
PZ
1742 uint32_t method1, method2;
1743
7221fc33 1744 if (!cstate->base.active || !pstate->visible)
801bcfff
PZ
1745 return 0;
1746
7221fc33 1747 method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), bpp, mem_value);
cca32e9a
PZ
1748
1749 if (!is_lp)
1750 return method1;
1751
7221fc33
MR
1752 method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
1753 cstate->base.adjusted_mode.crtc_htotal,
43d59eda
MR
1754 drm_rect_width(&pstate->dst),
1755 bpp,
cca32e9a
PZ
1756 mem_value);
1757
1758 return min(method1, method2);
801bcfff
PZ
1759}
1760
37126462
VS
1761/*
1762 * For both WM_PIPE and WM_LP.
1763 * mem_value must be in 0.1us units.
1764 */
7221fc33 1765static uint32_t ilk_compute_spr_wm(const struct intel_crtc_state *cstate,
43d59eda 1766 const struct intel_plane_state *pstate,
801bcfff
PZ
1767 uint32_t mem_value)
1768{
43d59eda 1769 int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
801bcfff
PZ
1770 uint32_t method1, method2;
1771
7221fc33 1772 if (!cstate->base.active || !pstate->visible)
801bcfff
PZ
1773 return 0;
1774
7221fc33
MR
1775 method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), bpp, mem_value);
1776 method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
1777 cstate->base.adjusted_mode.crtc_htotal,
43d59eda
MR
1778 drm_rect_width(&pstate->dst),
1779 bpp,
801bcfff
PZ
1780 mem_value);
1781 return min(method1, method2);
1782}
1783
37126462
VS
1784/*
1785 * For both WM_PIPE and WM_LP.
1786 * mem_value must be in 0.1us units.
1787 */
7221fc33 1788static uint32_t ilk_compute_cur_wm(const struct intel_crtc_state *cstate,
43d59eda 1789 const struct intel_plane_state *pstate,
801bcfff
PZ
1790 uint32_t mem_value)
1791{
43d59eda
MR
1792 int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
1793
7221fc33 1794 if (!cstate->base.active || !pstate->visible)
801bcfff
PZ
1795 return 0;
1796
7221fc33
MR
1797 return ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
1798 cstate->base.adjusted_mode.crtc_htotal,
43d59eda
MR
1799 drm_rect_width(&pstate->dst),
1800 bpp,
801bcfff
PZ
1801 mem_value);
1802}
1803
cca32e9a 1804/* Only for WM_LP. */
7221fc33 1805static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate,
43d59eda 1806 const struct intel_plane_state *pstate,
1fda9882 1807 uint32_t pri_val)
cca32e9a 1808{
43d59eda
MR
1809 int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
1810
7221fc33 1811 if (!cstate->base.active || !pstate->visible)
cca32e9a
PZ
1812 return 0;
1813
43d59eda 1814 return ilk_wm_fbc(pri_val, drm_rect_width(&pstate->dst), bpp);
cca32e9a
PZ
1815}
1816
158ae64f
VS
1817static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
1818{
416f4727
VS
1819 if (INTEL_INFO(dev)->gen >= 8)
1820 return 3072;
1821 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS
1822 return 768;
1823 else
1824 return 512;
1825}
1826
4e975081
VS
1827static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev,
1828 int level, bool is_sprite)
1829{
1830 if (INTEL_INFO(dev)->gen >= 8)
1831 /* BDW primary/sprite plane watermarks */
1832 return level == 0 ? 255 : 2047;
1833 else if (INTEL_INFO(dev)->gen >= 7)
1834 /* IVB/HSW primary/sprite plane watermarks */
1835 return level == 0 ? 127 : 1023;
1836 else if (!is_sprite)
1837 /* ILK/SNB primary plane watermarks */
1838 return level == 0 ? 127 : 511;
1839 else
1840 /* ILK/SNB sprite plane watermarks */
1841 return level == 0 ? 63 : 255;
1842}
1843
1844static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev,
1845 int level)
1846{
1847 if (INTEL_INFO(dev)->gen >= 7)
1848 return level == 0 ? 63 : 255;
1849 else
1850 return level == 0 ? 31 : 63;
1851}
1852
1853static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev)
1854{
1855 if (INTEL_INFO(dev)->gen >= 8)
1856 return 31;
1857 else
1858 return 15;
1859}
1860
158ae64f
VS
1861/* Calculate the maximum primary/sprite plane watermark */
1862static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
1863 int level,
240264f4 1864 const struct intel_wm_config *config,
158ae64f
VS
1865 enum intel_ddb_partitioning ddb_partitioning,
1866 bool is_sprite)
1867{
1868 unsigned int fifo_size = ilk_display_fifo_size(dev);
158ae64f
VS
1869
1870 /* if sprites aren't enabled, sprites get nothing */
240264f4 1871 if (is_sprite && !config->sprites_enabled)
158ae64f
VS
1872 return 0;
1873
1874 /* HSW allows LP1+ watermarks even with multiple pipes */
240264f4 1875 if (level == 0 || config->num_pipes_active > 1) {
158ae64f
VS
1876 fifo_size /= INTEL_INFO(dev)->num_pipes;
1877
1878 /*
1879 * For some reason the non self refresh
1880 * FIFO size is only half of the self
1881 * refresh FIFO size on ILK/SNB.
1882 */
1883 if (INTEL_INFO(dev)->gen <= 6)
1884 fifo_size /= 2;
1885 }
1886
240264f4 1887 if (config->sprites_enabled) {
158ae64f
VS
1888 /* level 0 is always calculated with 1:1 split */
1889 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
1890 if (is_sprite)
1891 fifo_size *= 5;
1892 fifo_size /= 6;
1893 } else {
1894 fifo_size /= 2;
1895 }
1896 }
1897
1898 /* clamp to max that the registers can hold */
4e975081 1899 return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite));
158ae64f
VS
1900}
1901
1902/* Calculate the maximum cursor plane watermark */
1903static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
240264f4
VS
1904 int level,
1905 const struct intel_wm_config *config)
158ae64f
VS
1906{
1907 /* HSW LP1+ watermarks w/ multiple pipes */
240264f4 1908 if (level > 0 && config->num_pipes_active > 1)
158ae64f
VS
1909 return 64;
1910
1911 /* otherwise just report max that registers can hold */
4e975081 1912 return ilk_cursor_wm_reg_max(dev, level);
158ae64f
VS
1913}
1914
d34ff9c6 1915static void ilk_compute_wm_maximums(const struct drm_device *dev,
34982fe1
VS
1916 int level,
1917 const struct intel_wm_config *config,
1918 enum intel_ddb_partitioning ddb_partitioning,
820c1980 1919 struct ilk_wm_maximums *max)
158ae64f 1920{
240264f4
VS
1921 max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
1922 max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
1923 max->cur = ilk_cursor_wm_max(dev, level, config);
4e975081 1924 max->fbc = ilk_fbc_wm_reg_max(dev);
158ae64f
VS
1925}
1926
a3cb4048
VS
1927static void ilk_compute_wm_reg_maximums(struct drm_device *dev,
1928 int level,
1929 struct ilk_wm_maximums *max)
1930{
1931 max->pri = ilk_plane_wm_reg_max(dev, level, false);
1932 max->spr = ilk_plane_wm_reg_max(dev, level, true);
1933 max->cur = ilk_cursor_wm_reg_max(dev, level);
1934 max->fbc = ilk_fbc_wm_reg_max(dev);
1935}
1936
d9395655 1937static bool ilk_validate_wm_level(int level,
820c1980 1938 const struct ilk_wm_maximums *max,
d9395655 1939 struct intel_wm_level *result)
a9786a11
VS
1940{
1941 bool ret;
1942
1943 /* already determined to be invalid? */
1944 if (!result->enable)
1945 return false;
1946
1947 result->enable = result->pri_val <= max->pri &&
1948 result->spr_val <= max->spr &&
1949 result->cur_val <= max->cur;
1950
1951 ret = result->enable;
1952
1953 /*
1954 * HACK until we can pre-compute everything,
1955 * and thus fail gracefully if LP0 watermarks
1956 * are exceeded...
1957 */
1958 if (level == 0 && !result->enable) {
1959 if (result->pri_val > max->pri)
1960 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1961 level, result->pri_val, max->pri);
1962 if (result->spr_val > max->spr)
1963 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1964 level, result->spr_val, max->spr);
1965 if (result->cur_val > max->cur)
1966 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
1967 level, result->cur_val, max->cur);
1968
1969 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
1970 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
1971 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
1972 result->enable = true;
1973 }
1974
a9786a11
VS
1975 return ret;
1976}
1977
d34ff9c6 1978static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
43d59eda 1979 const struct intel_crtc *intel_crtc,
6f5ddd17 1980 int level,
7221fc33 1981 struct intel_crtc_state *cstate,
1fd527cc 1982 struct intel_wm_level *result)
6f5ddd17 1983{
261a27d1 1984 struct intel_plane *intel_plane;
6f5ddd17
VS
1985 uint16_t pri_latency = dev_priv->wm.pri_latency[level];
1986 uint16_t spr_latency = dev_priv->wm.spr_latency[level];
1987 uint16_t cur_latency = dev_priv->wm.cur_latency[level];
1988
1989 /* WM1+ latency values stored in 0.5us units */
1990 if (level > 0) {
1991 pri_latency *= 5;
1992 spr_latency *= 5;
1993 cur_latency *= 5;
1994 }
1995
261a27d1
MR
1996 for_each_intel_plane_on_crtc(dev_priv->dev, intel_crtc, intel_plane) {
1997 struct intel_plane_state *pstate =
1998 to_intel_plane_state(intel_plane->base.state);
1999
2000 switch (intel_plane->base.type) {
2001 case DRM_PLANE_TYPE_PRIMARY:
2002 result->pri_val = ilk_compute_pri_wm(cstate, pstate,
2003 pri_latency,
2004 level);
2005 result->fbc_val = ilk_compute_fbc_wm(cstate, pstate,
2006 result->pri_val);
2007 break;
2008 case DRM_PLANE_TYPE_OVERLAY:
2009 result->spr_val = ilk_compute_spr_wm(cstate, pstate,
2010 spr_latency);
2011 break;
2012 case DRM_PLANE_TYPE_CURSOR:
2013 result->cur_val = ilk_compute_cur_wm(cstate, pstate,
2014 cur_latency);
2015 break;
2016 }
2017 }
2018
6f5ddd17
VS
2019 result->enable = true;
2020}
2021
801bcfff
PZ
2022static uint32_t
2023hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
1f8eeabf
ED
2024{
2025 struct drm_i915_private *dev_priv = dev->dev_private;
1011d8c4 2026 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
7c5f93b0 2027 const struct drm_display_mode *adjusted_mode = &intel_crtc->config->base.adjusted_mode;
85a02deb 2028 u32 linetime, ips_linetime;
1f8eeabf 2029
3ef00284 2030 if (!intel_crtc->active)
801bcfff 2031 return 0;
1011d8c4 2032
1f8eeabf
ED
2033 /* The WM are computed with base on how long it takes to fill a single
2034 * row at the given clock rate, multiplied by 8.
2035 * */
124abe07
VS
2036 linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
2037 adjusted_mode->crtc_clock);
2038 ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
05024da3 2039 dev_priv->cdclk_freq);
1f8eeabf 2040
801bcfff
PZ
2041 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
2042 PIPE_WM_LINETIME_TIME(linetime);
1f8eeabf
ED
2043}
2044
2af30a5c 2045static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[8])
12b134df
VS
2046{
2047 struct drm_i915_private *dev_priv = dev->dev_private;
2048
2af30a5c
PB
2049 if (IS_GEN9(dev)) {
2050 uint32_t val;
4f947386 2051 int ret, i;
367294be 2052 int level, max_level = ilk_wm_max_level(dev);
2af30a5c
PB
2053
2054 /* read the first set of memory latencies[0:3] */
2055 val = 0; /* data0 to be programmed to 0 for first set */
2056 mutex_lock(&dev_priv->rps.hw_lock);
2057 ret = sandybridge_pcode_read(dev_priv,
2058 GEN9_PCODE_READ_MEM_LATENCY,
2059 &val);
2060 mutex_unlock(&dev_priv->rps.hw_lock);
2061
2062 if (ret) {
2063 DRM_ERROR("SKL Mailbox read error = %d\n", ret);
2064 return;
2065 }
2066
2067 wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
2068 wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
2069 GEN9_MEM_LATENCY_LEVEL_MASK;
2070 wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
2071 GEN9_MEM_LATENCY_LEVEL_MASK;
2072 wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
2073 GEN9_MEM_LATENCY_LEVEL_MASK;
2074
2075 /* read the second set of memory latencies[4:7] */
2076 val = 1; /* data0 to be programmed to 1 for second set */
2077 mutex_lock(&dev_priv->rps.hw_lock);
2078 ret = sandybridge_pcode_read(dev_priv,
2079 GEN9_PCODE_READ_MEM_LATENCY,
2080 &val);
2081 mutex_unlock(&dev_priv->rps.hw_lock);
2082 if (ret) {
2083 DRM_ERROR("SKL Mailbox read error = %d\n", ret);
2084 return;
2085 }
2086
2087 wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
2088 wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
2089 GEN9_MEM_LATENCY_LEVEL_MASK;
2090 wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
2091 GEN9_MEM_LATENCY_LEVEL_MASK;
2092 wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
2093 GEN9_MEM_LATENCY_LEVEL_MASK;
2094
367294be 2095 /*
6f97235b
DL
2096 * WaWmMemoryReadLatency:skl
2097 *
367294be
VK
2098 * punit doesn't take into account the read latency so we need
2099 * to add 2us to the various latency levels we retrieve from
2100 * the punit.
2101 * - W0 is a bit special in that it's the only level that
2102 * can't be disabled if we want to have display working, so
2103 * we always add 2us there.
2104 * - For levels >=1, punit returns 0us latency when they are
2105 * disabled, so we respect that and don't add 2us then
4f947386
VK
2106 *
2107 * Additionally, if a level n (n > 1) has a 0us latency, all
2108 * levels m (m >= n) need to be disabled. We make sure to
2109 * sanitize the values out of the punit to satisfy this
2110 * requirement.
367294be
VK
2111 */
2112 wm[0] += 2;
2113 for (level = 1; level <= max_level; level++)
2114 if (wm[level] != 0)
2115 wm[level] += 2;
4f947386
VK
2116 else {
2117 for (i = level + 1; i <= max_level; i++)
2118 wm[i] = 0;
367294be 2119
4f947386
VK
2120 break;
2121 }
2af30a5c 2122 } else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
12b134df
VS
2123 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2124
2125 wm[0] = (sskpd >> 56) & 0xFF;
2126 if (wm[0] == 0)
2127 wm[0] = sskpd & 0xF;
e5d5019e
VS
2128 wm[1] = (sskpd >> 4) & 0xFF;
2129 wm[2] = (sskpd >> 12) & 0xFF;
2130 wm[3] = (sskpd >> 20) & 0x1FF;
2131 wm[4] = (sskpd >> 32) & 0x1FF;
63cf9a13
VS
2132 } else if (INTEL_INFO(dev)->gen >= 6) {
2133 uint32_t sskpd = I915_READ(MCH_SSKPD);
2134
2135 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2136 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2137 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2138 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
3a88d0ac
VS
2139 } else if (INTEL_INFO(dev)->gen >= 5) {
2140 uint32_t mltr = I915_READ(MLTR_ILK);
2141
2142 /* ILK primary LP0 latency is 700 ns */
2143 wm[0] = 7;
2144 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2145 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
12b134df
VS
2146 }
2147}
2148
53615a5e
VS
2149static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2150{
2151 /* ILK sprite LP0 latency is 1300 ns */
2152 if (INTEL_INFO(dev)->gen == 5)
2153 wm[0] = 13;
2154}
2155
2156static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2157{
2158 /* ILK cursor LP0 latency is 1300 ns */
2159 if (INTEL_INFO(dev)->gen == 5)
2160 wm[0] = 13;
2161
2162 /* WaDoubleCursorLP3Latency:ivb */
2163 if (IS_IVYBRIDGE(dev))
2164 wm[3] *= 2;
2165}
2166
546c81fd 2167int ilk_wm_max_level(const struct drm_device *dev)
26ec971e 2168{
26ec971e 2169 /* how many WM levels are we expecting */
b6e742f6 2170 if (INTEL_INFO(dev)->gen >= 9)
2af30a5c
PB
2171 return 7;
2172 else if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ad0d6dc4 2173 return 4;
26ec971e 2174 else if (INTEL_INFO(dev)->gen >= 6)
ad0d6dc4 2175 return 3;
26ec971e 2176 else
ad0d6dc4
VS
2177 return 2;
2178}
7526ed79 2179
ad0d6dc4
VS
2180static void intel_print_wm_latency(struct drm_device *dev,
2181 const char *name,
2af30a5c 2182 const uint16_t wm[8])
ad0d6dc4
VS
2183{
2184 int level, max_level = ilk_wm_max_level(dev);
26ec971e
VS
2185
2186 for (level = 0; level <= max_level; level++) {
2187 unsigned int latency = wm[level];
2188
2189 if (latency == 0) {
2190 DRM_ERROR("%s WM%d latency not provided\n",
2191 name, level);
2192 continue;
2193 }
2194
2af30a5c
PB
2195 /*
2196 * - latencies are in us on gen9.
2197 * - before then, WM1+ latency values are in 0.5us units
2198 */
2199 if (IS_GEN9(dev))
2200 latency *= 10;
2201 else if (level > 0)
26ec971e
VS
2202 latency *= 5;
2203
2204 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2205 name, level, wm[level],
2206 latency / 10, latency % 10);
2207 }
2208}
2209
e95a2f75
VS
2210static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
2211 uint16_t wm[5], uint16_t min)
2212{
2213 int level, max_level = ilk_wm_max_level(dev_priv->dev);
2214
2215 if (wm[0] >= min)
2216 return false;
2217
2218 wm[0] = max(wm[0], min);
2219 for (level = 1; level <= max_level; level++)
2220 wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
2221
2222 return true;
2223}
2224
2225static void snb_wm_latency_quirk(struct drm_device *dev)
2226{
2227 struct drm_i915_private *dev_priv = dev->dev_private;
2228 bool changed;
2229
2230 /*
2231 * The BIOS provided WM memory latency values are often
2232 * inadequate for high resolution displays. Adjust them.
2233 */
2234 changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
2235 ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
2236 ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
2237
2238 if (!changed)
2239 return;
2240
2241 DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
2242 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2243 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2244 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
2245}
2246
fa50ad61 2247static void ilk_setup_wm_latency(struct drm_device *dev)
53615a5e
VS
2248{
2249 struct drm_i915_private *dev_priv = dev->dev_private;
2250
2251 intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2252
2253 memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2254 sizeof(dev_priv->wm.pri_latency));
2255 memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2256 sizeof(dev_priv->wm.pri_latency));
2257
2258 intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2259 intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
26ec971e
VS
2260
2261 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2262 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2263 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
e95a2f75
VS
2264
2265 if (IS_GEN6(dev))
2266 snb_wm_latency_quirk(dev);
53615a5e
VS
2267}
2268
2af30a5c
PB
2269static void skl_setup_wm_latency(struct drm_device *dev)
2270{
2271 struct drm_i915_private *dev_priv = dev->dev_private;
2272
2273 intel_read_wm_latency(dev, dev_priv->wm.skl_latency);
2274 intel_print_wm_latency(dev, "Gen9 Plane", dev_priv->wm.skl_latency);
2275}
2276
261a27d1
MR
2277static void ilk_compute_wm_config(struct drm_device *dev,
2278 struct intel_wm_config *config)
2279{
2280 struct intel_crtc *intel_crtc;
2281
2282 /* Compute the currently _active_ config */
2283 for_each_intel_crtc(dev, intel_crtc) {
2284 const struct intel_pipe_wm *wm = &intel_crtc->wm.active;
2285
2286 if (!wm->pipe_enabled)
2287 continue;
2288
2289 config->sprites_enabled |= wm->sprites_enabled;
2290 config->sprites_scaled |= wm->sprites_scaled;
2291 config->num_pipes_active++;
2292 }
2293}
2294
0b2ae6d7 2295/* Compute new watermarks for the pipe */
261a27d1
MR
2296static bool intel_compute_pipe_wm(struct intel_crtc_state *cstate,
2297 struct intel_pipe_wm *pipe_wm)
0b2ae6d7 2298{
261a27d1
MR
2299 struct drm_crtc *crtc = cstate->base.crtc;
2300 struct drm_device *dev = crtc->dev;
d34ff9c6 2301 const struct drm_i915_private *dev_priv = dev->dev_private;
261a27d1 2302 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
43d59eda
MR
2303 struct intel_plane *intel_plane;
2304 struct intel_plane_state *sprstate = NULL;
0b2ae6d7
VS
2305 int level, max_level = ilk_wm_max_level(dev);
2306 /* LP0 watermark maximums depend on this pipe alone */
2307 struct intel_wm_config config = {
2308 .num_pipes_active = 1,
0b2ae6d7 2309 };
820c1980 2310 struct ilk_wm_maximums max;
0b2ae6d7 2311
43d59eda 2312 for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) {
261a27d1
MR
2313 if (intel_plane->base.type == DRM_PLANE_TYPE_OVERLAY) {
2314 sprstate = to_intel_plane_state(intel_plane->base.state);
2315 break;
2316 }
43d59eda
MR
2317 }
2318
2319 config.sprites_enabled = sprstate->visible;
2320 config.sprites_scaled = sprstate->visible &&
2321 (drm_rect_width(&sprstate->dst) != drm_rect_width(&sprstate->src) >> 16 ||
2322 drm_rect_height(&sprstate->dst) != drm_rect_height(&sprstate->src) >> 16);
2323
7221fc33 2324 pipe_wm->pipe_enabled = cstate->base.active;
261a27d1 2325 pipe_wm->sprites_enabled = sprstate->visible;
43d59eda 2326 pipe_wm->sprites_scaled = config.sprites_scaled;
2a44b76b 2327
7b39a0b7 2328 /* ILK/SNB: LP2+ watermarks only w/o sprites */
43d59eda 2329 if (INTEL_INFO(dev)->gen <= 6 && sprstate->visible)
7b39a0b7
VS
2330 max_level = 1;
2331
2332 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
43d59eda 2333 if (config.sprites_scaled)
7b39a0b7
VS
2334 max_level = 0;
2335
261a27d1 2336 ilk_compute_wm_level(dev_priv, intel_crtc, 0, cstate, &pipe_wm->wm[0]);
0b2ae6d7 2337
a42a5719 2338 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
261a27d1 2339 pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
0b2ae6d7 2340
a3cb4048
VS
2341 /* LP0 watermarks always use 1/2 DDB partitioning */
2342 ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
2343
0b2ae6d7 2344 /* At least LP0 must be valid */
a3cb4048 2345 if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0]))
261a27d1 2346 return false;
a3cb4048
VS
2347
2348 ilk_compute_wm_reg_maximums(dev, 1, &max);
2349
2350 for (level = 1; level <= max_level; level++) {
2351 struct intel_wm_level wm = {};
2352
261a27d1 2353 ilk_compute_wm_level(dev_priv, intel_crtc, level, cstate, &wm);
a3cb4048
VS
2354
2355 /*
2356 * Disable any watermark level that exceeds the
2357 * register maximums since such watermarks are
2358 * always invalid.
2359 */
2360 if (!ilk_validate_wm_level(level, &max, &wm))
2361 break;
2362
2363 pipe_wm->wm[level] = wm;
2364 }
2365
261a27d1 2366 return true;
0b2ae6d7
VS
2367}
2368
2369/*
2370 * Merge the watermarks from all active pipes for a specific level.
2371 */
2372static void ilk_merge_wm_level(struct drm_device *dev,
2373 int level,
2374 struct intel_wm_level *ret_wm)
2375{
2376 const struct intel_crtc *intel_crtc;
2377
d52fea5b
VS
2378 ret_wm->enable = true;
2379
d3fcc808 2380 for_each_intel_crtc(dev, intel_crtc) {
261a27d1 2381 const struct intel_pipe_wm *active = &intel_crtc->wm.active;
fe392efd
VS
2382 const struct intel_wm_level *wm = &active->wm[level];
2383
2384 if (!active->pipe_enabled)
2385 continue;
0b2ae6d7 2386
d52fea5b
VS
2387 /*
2388 * The watermark values may have been used in the past,
2389 * so we must maintain them in the registers for some
2390 * time even if the level is now disabled.
2391 */
0b2ae6d7 2392 if (!wm->enable)
d52fea5b 2393 ret_wm->enable = false;
0b2ae6d7
VS
2394
2395 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2396 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2397 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2398 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2399 }
0b2ae6d7
VS
2400}
2401
2402/*
2403 * Merge all low power watermarks for all active pipes.
2404 */
2405static void ilk_wm_merge(struct drm_device *dev,
0ba22e26 2406 const struct intel_wm_config *config,
820c1980 2407 const struct ilk_wm_maximums *max,
0b2ae6d7
VS
2408 struct intel_pipe_wm *merged)
2409{
7733b49b 2410 struct drm_i915_private *dev_priv = dev->dev_private;
0b2ae6d7 2411 int level, max_level = ilk_wm_max_level(dev);
d52fea5b 2412 int last_enabled_level = max_level;
0b2ae6d7 2413
0ba22e26
VS
2414 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2415 if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) &&
2416 config->num_pipes_active > 1)
2417 return;
2418
6c8b6c28
VS
2419 /* ILK: FBC WM must be disabled always */
2420 merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6;
0b2ae6d7
VS
2421
2422 /* merge each WM1+ level */
2423 for (level = 1; level <= max_level; level++) {
2424 struct intel_wm_level *wm = &merged->wm[level];
2425
2426 ilk_merge_wm_level(dev, level, wm);
2427
d52fea5b
VS
2428 if (level > last_enabled_level)
2429 wm->enable = false;
2430 else if (!ilk_validate_wm_level(level, max, wm))
2431 /* make sure all following levels get disabled */
2432 last_enabled_level = level - 1;
0b2ae6d7
VS
2433
2434 /*
2435 * The spec says it is preferred to disable
2436 * FBC WMs instead of disabling a WM level.
2437 */
2438 if (wm->fbc_val > max->fbc) {
d52fea5b
VS
2439 if (wm->enable)
2440 merged->fbc_wm_enabled = false;
0b2ae6d7
VS
2441 wm->fbc_val = 0;
2442 }
2443 }
6c8b6c28
VS
2444
2445 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2446 /*
2447 * FIXME this is racy. FBC might get enabled later.
2448 * What we should check here is whether FBC can be
2449 * enabled sometime later.
2450 */
7733b49b
PZ
2451 if (IS_GEN5(dev) && !merged->fbc_wm_enabled &&
2452 intel_fbc_enabled(dev_priv)) {
6c8b6c28
VS
2453 for (level = 2; level <= max_level; level++) {
2454 struct intel_wm_level *wm = &merged->wm[level];
2455
2456 wm->enable = false;
2457 }
2458 }
0b2ae6d7
VS
2459}
2460
b380ca3c
VS
2461static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2462{
2463 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2464 return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2465}
2466
a68d68ee
VS
2467/* The value we need to program into the WM_LPx latency field */
2468static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level)
2469{
2470 struct drm_i915_private *dev_priv = dev->dev_private;
2471
a42a5719 2472 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
a68d68ee
VS
2473 return 2 * level;
2474 else
2475 return dev_priv->wm.pri_latency[level];
2476}
2477
820c1980 2478static void ilk_compute_wm_results(struct drm_device *dev,
0362c781 2479 const struct intel_pipe_wm *merged,
609cedef 2480 enum intel_ddb_partitioning partitioning,
820c1980 2481 struct ilk_wm_values *results)
801bcfff 2482{
0b2ae6d7
VS
2483 struct intel_crtc *intel_crtc;
2484 int level, wm_lp;
cca32e9a 2485
0362c781 2486 results->enable_fbc_wm = merged->fbc_wm_enabled;
609cedef 2487 results->partitioning = partitioning;
cca32e9a 2488
0b2ae6d7 2489 /* LP1+ register values */
cca32e9a 2490 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
1fd527cc 2491 const struct intel_wm_level *r;
801bcfff 2492
b380ca3c 2493 level = ilk_wm_lp_to_level(wm_lp, merged);
0b2ae6d7 2494
0362c781 2495 r = &merged->wm[level];
cca32e9a 2496
d52fea5b
VS
2497 /*
2498 * Maintain the watermark values even if the level is
2499 * disabled. Doing otherwise could cause underruns.
2500 */
2501 results->wm_lp[wm_lp - 1] =
a68d68ee 2502 (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) |
416f4727
VS
2503 (r->pri_val << WM1_LP_SR_SHIFT) |
2504 r->cur_val;
2505
d52fea5b
VS
2506 if (r->enable)
2507 results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN;
2508
416f4727
VS
2509 if (INTEL_INFO(dev)->gen >= 8)
2510 results->wm_lp[wm_lp - 1] |=
2511 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
2512 else
2513 results->wm_lp[wm_lp - 1] |=
2514 r->fbc_val << WM1_LP_FBC_SHIFT;
2515
d52fea5b
VS
2516 /*
2517 * Always set WM1S_LP_EN when spr_val != 0, even if the
2518 * level is disabled. Doing otherwise could cause underruns.
2519 */
6cef2b8a
VS
2520 if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) {
2521 WARN_ON(wm_lp != 1);
2522 results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
2523 } else
2524 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
cca32e9a 2525 }
801bcfff 2526
0b2ae6d7 2527 /* LP0 register values */
d3fcc808 2528 for_each_intel_crtc(dev, intel_crtc) {
0b2ae6d7 2529 enum pipe pipe = intel_crtc->pipe;
261a27d1
MR
2530 const struct intel_wm_level *r =
2531 &intel_crtc->wm.active.wm[0];
0b2ae6d7
VS
2532
2533 if (WARN_ON(!r->enable))
2534 continue;
2535
261a27d1 2536 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
1011d8c4 2537
0b2ae6d7
VS
2538 results->wm_pipe[pipe] =
2539 (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2540 (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2541 r->cur_val;
801bcfff
PZ
2542 }
2543}
2544
861f3389
PZ
2545/* Find the result with the highest level enabled. Check for enable_fbc_wm in
2546 * case both are at the same level. Prefer r1 in case they're the same. */
820c1980 2547static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev,
198a1e9b
VS
2548 struct intel_pipe_wm *r1,
2549 struct intel_pipe_wm *r2)
861f3389 2550{
198a1e9b
VS
2551 int level, max_level = ilk_wm_max_level(dev);
2552 int level1 = 0, level2 = 0;
861f3389 2553
198a1e9b
VS
2554 for (level = 1; level <= max_level; level++) {
2555 if (r1->wm[level].enable)
2556 level1 = level;
2557 if (r2->wm[level].enable)
2558 level2 = level;
861f3389
PZ
2559 }
2560
198a1e9b
VS
2561 if (level1 == level2) {
2562 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
861f3389
PZ
2563 return r2;
2564 else
2565 return r1;
198a1e9b 2566 } else if (level1 > level2) {
861f3389
PZ
2567 return r1;
2568 } else {
2569 return r2;
2570 }
2571}
2572
49a687c4
VS
2573/* dirty bits used to track which watermarks need changes */
2574#define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2575#define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2576#define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2577#define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2578#define WM_DIRTY_FBC (1 << 24)
2579#define WM_DIRTY_DDB (1 << 25)
2580
055e393f 2581static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv,
820c1980
ID
2582 const struct ilk_wm_values *old,
2583 const struct ilk_wm_values *new)
49a687c4
VS
2584{
2585 unsigned int dirty = 0;
2586 enum pipe pipe;
2587 int wm_lp;
2588
055e393f 2589 for_each_pipe(dev_priv, pipe) {
49a687c4
VS
2590 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
2591 dirty |= WM_DIRTY_LINETIME(pipe);
2592 /* Must disable LP1+ watermarks too */
2593 dirty |= WM_DIRTY_LP_ALL;
2594 }
2595
2596 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
2597 dirty |= WM_DIRTY_PIPE(pipe);
2598 /* Must disable LP1+ watermarks too */
2599 dirty |= WM_DIRTY_LP_ALL;
2600 }
2601 }
2602
2603 if (old->enable_fbc_wm != new->enable_fbc_wm) {
2604 dirty |= WM_DIRTY_FBC;
2605 /* Must disable LP1+ watermarks too */
2606 dirty |= WM_DIRTY_LP_ALL;
2607 }
2608
2609 if (old->partitioning != new->partitioning) {
2610 dirty |= WM_DIRTY_DDB;
2611 /* Must disable LP1+ watermarks too */
2612 dirty |= WM_DIRTY_LP_ALL;
2613 }
2614
2615 /* LP1+ watermarks already deemed dirty, no need to continue */
2616 if (dirty & WM_DIRTY_LP_ALL)
2617 return dirty;
2618
2619 /* Find the lowest numbered LP1+ watermark in need of an update... */
2620 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2621 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
2622 old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
2623 break;
2624 }
2625
2626 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2627 for (; wm_lp <= 3; wm_lp++)
2628 dirty |= WM_DIRTY_LP(wm_lp);
2629
2630 return dirty;
2631}
2632
8553c18e
VS
2633static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
2634 unsigned int dirty)
801bcfff 2635{
820c1980 2636 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e 2637 bool changed = false;
801bcfff 2638
facd619b
VS
2639 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
2640 previous->wm_lp[2] &= ~WM1_LP_SR_EN;
2641 I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
8553c18e 2642 changed = true;
facd619b
VS
2643 }
2644 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
2645 previous->wm_lp[1] &= ~WM1_LP_SR_EN;
2646 I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
8553c18e 2647 changed = true;
facd619b
VS
2648 }
2649 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
2650 previous->wm_lp[0] &= ~WM1_LP_SR_EN;
2651 I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
8553c18e 2652 changed = true;
facd619b 2653 }
801bcfff 2654
facd619b
VS
2655 /*
2656 * Don't touch WM1S_LP_EN here.
2657 * Doing so could cause underruns.
2658 */
6cef2b8a 2659
8553c18e
VS
2660 return changed;
2661}
2662
2663/*
2664 * The spec says we shouldn't write when we don't need, because every write
2665 * causes WMs to be re-evaluated, expending some power.
2666 */
820c1980
ID
2667static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
2668 struct ilk_wm_values *results)
8553c18e
VS
2669{
2670 struct drm_device *dev = dev_priv->dev;
820c1980 2671 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e
VS
2672 unsigned int dirty;
2673 uint32_t val;
2674
055e393f 2675 dirty = ilk_compute_wm_dirty(dev_priv, previous, results);
8553c18e
VS
2676 if (!dirty)
2677 return;
2678
2679 _ilk_disable_lp_wm(dev_priv, dirty);
2680
49a687c4 2681 if (dirty & WM_DIRTY_PIPE(PIPE_A))
801bcfff 2682 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
49a687c4 2683 if (dirty & WM_DIRTY_PIPE(PIPE_B))
801bcfff 2684 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
49a687c4 2685 if (dirty & WM_DIRTY_PIPE(PIPE_C))
801bcfff
PZ
2686 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2687
49a687c4 2688 if (dirty & WM_DIRTY_LINETIME(PIPE_A))
801bcfff 2689 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
49a687c4 2690 if (dirty & WM_DIRTY_LINETIME(PIPE_B))
801bcfff 2691 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
49a687c4 2692 if (dirty & WM_DIRTY_LINETIME(PIPE_C))
801bcfff
PZ
2693 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2694
49a687c4 2695 if (dirty & WM_DIRTY_DDB) {
a42a5719 2696 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
ac9545fd
VS
2697 val = I915_READ(WM_MISC);
2698 if (results->partitioning == INTEL_DDB_PART_1_2)
2699 val &= ~WM_MISC_DATA_PARTITION_5_6;
2700 else
2701 val |= WM_MISC_DATA_PARTITION_5_6;
2702 I915_WRITE(WM_MISC, val);
2703 } else {
2704 val = I915_READ(DISP_ARB_CTL2);
2705 if (results->partitioning == INTEL_DDB_PART_1_2)
2706 val &= ~DISP_DATA_PARTITION_5_6;
2707 else
2708 val |= DISP_DATA_PARTITION_5_6;
2709 I915_WRITE(DISP_ARB_CTL2, val);
2710 }
1011d8c4
PZ
2711 }
2712
49a687c4 2713 if (dirty & WM_DIRTY_FBC) {
cca32e9a
PZ
2714 val = I915_READ(DISP_ARB_CTL);
2715 if (results->enable_fbc_wm)
2716 val &= ~DISP_FBC_WM_DIS;
2717 else
2718 val |= DISP_FBC_WM_DIS;
2719 I915_WRITE(DISP_ARB_CTL, val);
2720 }
2721
954911eb
ID
2722 if (dirty & WM_DIRTY_LP(1) &&
2723 previous->wm_lp_spr[0] != results->wm_lp_spr[0])
2724 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2725
2726 if (INTEL_INFO(dev)->gen >= 7) {
6cef2b8a
VS
2727 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
2728 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2729 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
2730 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2731 }
801bcfff 2732
facd619b 2733 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
801bcfff 2734 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
facd619b 2735 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
801bcfff 2736 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
facd619b 2737 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
801bcfff 2738 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
609cedef
VS
2739
2740 dev_priv->wm.hw = *results;
801bcfff
PZ
2741}
2742
8553c18e
VS
2743static bool ilk_disable_lp_wm(struct drm_device *dev)
2744{
2745 struct drm_i915_private *dev_priv = dev->dev_private;
2746
2747 return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
2748}
2749
b9cec075
DL
2750/*
2751 * On gen9, we need to allocate Display Data Buffer (DDB) portions to the
2752 * different active planes.
2753 */
2754
2755#define SKL_DDB_SIZE 896 /* in blocks */
43d735a6 2756#define BXT_DDB_SIZE 512
b9cec075
DL
2757
2758static void
2759skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
2791a16c 2760 struct drm_crtc *for_crtc,
b9cec075 2761 const struct intel_wm_config *config,
2791a16c 2762 const struct skl_pipe_wm_parameters *params,
b9cec075
DL
2763 struct skl_ddb_entry *alloc /* out */)
2764{
2765 struct drm_crtc *crtc;
2766 unsigned int pipe_size, ddb_size;
2767 int nth_active_pipe;
2768
2791a16c 2769 if (!params->active) {
b9cec075
DL
2770 alloc->start = 0;
2771 alloc->end = 0;
2772 return;
2773 }
2774
43d735a6
DL
2775 if (IS_BROXTON(dev))
2776 ddb_size = BXT_DDB_SIZE;
2777 else
2778 ddb_size = SKL_DDB_SIZE;
b9cec075
DL
2779
2780 ddb_size -= 4; /* 4 blocks for bypass path allocation */
2781
2782 nth_active_pipe = 0;
2783 for_each_crtc(dev, crtc) {
3ef00284 2784 if (!to_intel_crtc(crtc)->active)
b9cec075
DL
2785 continue;
2786
2787 if (crtc == for_crtc)
2788 break;
2789
2790 nth_active_pipe++;
2791 }
2792
2793 pipe_size = ddb_size / config->num_pipes_active;
2794 alloc->start = nth_active_pipe * ddb_size / config->num_pipes_active;
16160e3d 2795 alloc->end = alloc->start + pipe_size;
b9cec075
DL
2796}
2797
2798static unsigned int skl_cursor_allocation(const struct intel_wm_config *config)
2799{
2800 if (config->num_pipes_active == 1)
2801 return 32;
2802
2803 return 8;
2804}
2805
a269c583
DL
2806static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg)
2807{
2808 entry->start = reg & 0x3ff;
2809 entry->end = (reg >> 16) & 0x3ff;
16160e3d
DL
2810 if (entry->end)
2811 entry->end += 1;
a269c583
DL
2812}
2813
08db6652
DL
2814void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv,
2815 struct skl_ddb_allocation *ddb /* out */)
a269c583 2816{
a269c583
DL
2817 enum pipe pipe;
2818 int plane;
2819 u32 val;
2820
b10f1b20
ML
2821 memset(ddb, 0, sizeof(*ddb));
2822
a269c583 2823 for_each_pipe(dev_priv, pipe) {
b10f1b20
ML
2824 if (!intel_display_power_is_enabled(dev_priv, POWER_DOMAIN_PIPE(pipe)))
2825 continue;
2826
dd740780 2827 for_each_plane(dev_priv, pipe, plane) {
a269c583
DL
2828 val = I915_READ(PLANE_BUF_CFG(pipe, plane));
2829 skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane],
2830 val);
2831 }
2832
2833 val = I915_READ(CUR_BUF_CFG(pipe));
4969d33e
MR
2834 skl_ddb_entry_init_from_hw(&ddb->plane[pipe][PLANE_CURSOR],
2835 val);
a269c583
DL
2836 }
2837}
2838
b9cec075 2839static unsigned int
2791a16c 2840skl_plane_relative_data_rate(const struct intel_plane_wm_parameters *p, int y)
b9cec075 2841{
2cd601c6
CK
2842
2843 /* for planar format */
2791a16c 2844 if (p->y_bytes_per_pixel) {
2cd601c6 2845 if (y) /* y-plane data rate */
2791a16c 2846 return p->horiz_pixels * p->vert_pixels * p->y_bytes_per_pixel;
2cd601c6 2847 else /* uv-plane data rate */
2791a16c 2848 return (p->horiz_pixels/2) * (p->vert_pixels/2) * p->bytes_per_pixel;
2cd601c6
CK
2849 }
2850
2851 /* for packed formats */
2791a16c 2852 return p->horiz_pixels * p->vert_pixels * p->bytes_per_pixel;
b9cec075
DL
2853}
2854
2855/*
2856 * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching
2857 * a 8192x4096@32bpp framebuffer:
2858 * 3 * 4096 * 8192 * 4 < 2^32
2859 */
2860static unsigned int
2791a16c
PZ
2861skl_get_total_relative_data_rate(struct intel_crtc *intel_crtc,
2862 const struct skl_pipe_wm_parameters *params)
b9cec075
DL
2863{
2864 unsigned int total_data_rate = 0;
2791a16c 2865 int plane;
b9cec075 2866
2791a16c
PZ
2867 for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) {
2868 const struct intel_plane_wm_parameters *p;
b9cec075 2869
2791a16c
PZ
2870 p = &params->plane[plane];
2871 if (!p->enabled)
b9cec075
DL
2872 continue;
2873
2791a16c
PZ
2874 total_data_rate += skl_plane_relative_data_rate(p, 0); /* packed/uv */
2875 if (p->y_bytes_per_pixel) {
2876 total_data_rate += skl_plane_relative_data_rate(p, 1); /* y-plane */
2877 }
b9cec075
DL
2878 }
2879
2880 return total_data_rate;
2881}
2882
2883static void
2791a16c 2884skl_allocate_pipe_ddb(struct drm_crtc *crtc,
261a27d1 2885 const struct intel_wm_config *config,
2791a16c 2886 const struct skl_pipe_wm_parameters *params,
b9cec075
DL
2887 struct skl_ddb_allocation *ddb /* out */)
2888{
2889 struct drm_device *dev = crtc->dev;
2791a16c 2890 struct drm_i915_private *dev_priv = dev->dev_private;
b9cec075
DL
2891 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2892 enum pipe pipe = intel_crtc->pipe;
34bb56af 2893 struct skl_ddb_entry *alloc = &ddb->pipe[pipe];
b9cec075 2894 uint16_t alloc_size, start, cursor_blocks;
80958155 2895 uint16_t minimum[I915_MAX_PLANES];
2cd601c6 2896 uint16_t y_minimum[I915_MAX_PLANES];
b9cec075 2897 unsigned int total_data_rate;
2791a16c 2898 int plane;
b9cec075 2899
2791a16c 2900 skl_ddb_get_pipe_allocation_limits(dev, crtc, config, params, alloc);
34bb56af 2901 alloc_size = skl_ddb_entry_size(alloc);
b9cec075
DL
2902 if (alloc_size == 0) {
2903 memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe]));
4969d33e
MR
2904 memset(&ddb->plane[pipe][PLANE_CURSOR], 0,
2905 sizeof(ddb->plane[pipe][PLANE_CURSOR]));
b9cec075
DL
2906 return;
2907 }
2908
2909 cursor_blocks = skl_cursor_allocation(config);
4969d33e
MR
2910 ddb->plane[pipe][PLANE_CURSOR].start = alloc->end - cursor_blocks;
2911 ddb->plane[pipe][PLANE_CURSOR].end = alloc->end;
b9cec075
DL
2912
2913 alloc_size -= cursor_blocks;
34bb56af 2914 alloc->end -= cursor_blocks;
b9cec075 2915
80958155 2916 /* 1. Allocate the mininum required blocks for each active plane */
2791a16c
PZ
2917 for_each_plane(dev_priv, pipe, plane) {
2918 const struct intel_plane_wm_parameters *p;
80958155 2919
2791a16c
PZ
2920 p = &params->plane[plane];
2921 if (!p->enabled)
80958155
DL
2922 continue;
2923
2791a16c
PZ
2924 minimum[plane] = 8;
2925 alloc_size -= minimum[plane];
2926 y_minimum[plane] = p->y_bytes_per_pixel ? 8 : 0;
2927 alloc_size -= y_minimum[plane];
80958155
DL
2928 }
2929
b9cec075 2930 /*
80958155
DL
2931 * 2. Distribute the remaining space in proportion to the amount of
2932 * data each plane needs to fetch from memory.
b9cec075
DL
2933 *
2934 * FIXME: we may not allocate every single block here.
2935 */
2791a16c 2936 total_data_rate = skl_get_total_relative_data_rate(intel_crtc, params);
b9cec075 2937
34bb56af 2938 start = alloc->start;
2791a16c
PZ
2939 for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) {
2940 const struct intel_plane_wm_parameters *p;
2cd601c6
CK
2941 unsigned int data_rate, y_data_rate;
2942 uint16_t plane_blocks, y_plane_blocks = 0;
b9cec075 2943
2791a16c
PZ
2944 p = &params->plane[plane];
2945 if (!p->enabled)
b9cec075
DL
2946 continue;
2947
2791a16c 2948 data_rate = skl_plane_relative_data_rate(p, 0);
b9cec075
DL
2949
2950 /*
2cd601c6 2951 * allocation for (packed formats) or (uv-plane part of planar format):
b9cec075
DL
2952 * promote the expression to 64 bits to avoid overflowing, the
2953 * result is < available as data_rate / total_data_rate < 1
2954 */
2791a16c 2955 plane_blocks = minimum[plane];
80958155
DL
2956 plane_blocks += div_u64((uint64_t)alloc_size * data_rate,
2957 total_data_rate);
b9cec075 2958
2791a16c
PZ
2959 ddb->plane[pipe][plane].start = start;
2960 ddb->plane[pipe][plane].end = start + plane_blocks;
b9cec075
DL
2961
2962 start += plane_blocks;
2cd601c6
CK
2963
2964 /*
2965 * allocation for y_plane part of planar format:
2966 */
2791a16c
PZ
2967 if (p->y_bytes_per_pixel) {
2968 y_data_rate = skl_plane_relative_data_rate(p, 1);
2969 y_plane_blocks = y_minimum[plane];
2cd601c6
CK
2970 y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate,
2971 total_data_rate);
2972
2791a16c
PZ
2973 ddb->y_plane[pipe][plane].start = start;
2974 ddb->y_plane[pipe][plane].end = start + y_plane_blocks;
2cd601c6
CK
2975
2976 start += y_plane_blocks;
2977 }
2978
b9cec075
DL
2979 }
2980
2981}
2982
5cec258b 2983static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state *config)
2d41c0b5
PB
2984{
2985 /* TODO: Take into account the scalers once we support them */
2d112de7 2986 return config->base.adjusted_mode.crtc_clock;
2d41c0b5
PB
2987}
2988
2989/*
2990 * The max latency should be 257 (max the punit can code is 255 and we add 2us
2991 * for the read latency) and bytes_per_pixel should always be <= 8, so that
2992 * should allow pixel_rate up to ~2 GHz which seems sufficient since max
2993 * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
2994*/
2995static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
2996 uint32_t latency)
2997{
2998 uint32_t wm_intermediate_val, ret;
2999
3000 if (latency == 0)
3001 return UINT_MAX;
3002
d4c2aa60 3003 wm_intermediate_val = latency * pixel_rate * bytes_per_pixel / 512;
2d41c0b5
PB
3004 ret = DIV_ROUND_UP(wm_intermediate_val, 1000);
3005
3006 return ret;
3007}
3008
3009static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
3010 uint32_t horiz_pixels, uint8_t bytes_per_pixel,
0fda6568 3011 uint64_t tiling, uint32_t latency)
2d41c0b5 3012{
d4c2aa60
TU
3013 uint32_t ret;
3014 uint32_t plane_bytes_per_line, plane_blocks_per_line;
3015 uint32_t wm_intermediate_val;
2d41c0b5
PB
3016
3017 if (latency == 0)
3018 return UINT_MAX;
3019
3020 plane_bytes_per_line = horiz_pixels * bytes_per_pixel;
0fda6568
TU
3021
3022 if (tiling == I915_FORMAT_MOD_Y_TILED ||
3023 tiling == I915_FORMAT_MOD_Yf_TILED) {
3024 plane_bytes_per_line *= 4;
3025 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
3026 plane_blocks_per_line /= 4;
3027 } else {
3028 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
3029 }
3030
2d41c0b5
PB
3031 wm_intermediate_val = latency * pixel_rate;
3032 ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) *
d4c2aa60 3033 plane_blocks_per_line;
2d41c0b5
PB
3034
3035 return ret;
3036}
3037
2d41c0b5
PB
3038static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation *new_ddb,
3039 const struct intel_crtc *intel_crtc)
3040{
3041 struct drm_device *dev = intel_crtc->base.dev;
3042 struct drm_i915_private *dev_priv = dev->dev_private;
3043 const struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb;
3044 enum pipe pipe = intel_crtc->pipe;
3045
3046 if (memcmp(new_ddb->plane[pipe], cur_ddb->plane[pipe],
3047 sizeof(new_ddb->plane[pipe])))
3048 return true;
3049
4969d33e
MR
3050 if (memcmp(&new_ddb->plane[pipe][PLANE_CURSOR], &cur_ddb->plane[pipe][PLANE_CURSOR],
3051 sizeof(new_ddb->plane[pipe][PLANE_CURSOR])))
2d41c0b5
PB
3052 return true;
3053
3054 return false;
3055}
3056
261a27d1
MR
3057static void skl_compute_wm_global_parameters(struct drm_device *dev,
3058 struct intel_wm_config *config)
3059{
3060 struct drm_crtc *crtc;
2791a16c 3061 struct drm_plane *plane;
261a27d1
MR
3062
3063 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
3064 config->num_pipes_active += to_intel_crtc(crtc)->active;
2791a16c
PZ
3065
3066 /* FIXME: I don't think we need those two global parameters on SKL */
3067 list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
3068 struct intel_plane *intel_plane = to_intel_plane(plane);
3069
3070 config->sprites_enabled |= intel_plane->wm.enabled;
3071 config->sprites_scaled |= intel_plane->wm.scaled;
3072 }
3073}
3074
3075static void skl_compute_wm_pipe_parameters(struct drm_crtc *crtc,
3076 struct skl_pipe_wm_parameters *p)
3077{
3078 struct drm_device *dev = crtc->dev;
3079 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3080 enum pipe pipe = intel_crtc->pipe;
3081 struct drm_plane *plane;
3082 struct drm_framebuffer *fb;
3083 int i = 1; /* Index for sprite planes start */
3084
3085 p->active = intel_crtc->active;
3086 if (p->active) {
3087 p->pipe_htotal = intel_crtc->config->base.adjusted_mode.crtc_htotal;
3088 p->pixel_rate = skl_pipe_pixel_rate(intel_crtc->config);
3089
3090 fb = crtc->primary->state->fb;
3091 /* For planar: Bpp is for uv plane, y_Bpp is for y plane */
3092 if (fb) {
3093 p->plane[0].enabled = true;
3094 p->plane[0].bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ?
3095 drm_format_plane_cpp(fb->pixel_format, 1) :
3096 drm_format_plane_cpp(fb->pixel_format, 0);
3097 p->plane[0].y_bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ?
3098 drm_format_plane_cpp(fb->pixel_format, 0) : 0;
3099 p->plane[0].tiling = fb->modifier[0];
3100 } else {
3101 p->plane[0].enabled = false;
3102 p->plane[0].bytes_per_pixel = 0;
3103 p->plane[0].y_bytes_per_pixel = 0;
3104 p->plane[0].tiling = DRM_FORMAT_MOD_NONE;
3105 }
3106 p->plane[0].horiz_pixels = intel_crtc->config->pipe_src_w;
3107 p->plane[0].vert_pixels = intel_crtc->config->pipe_src_h;
3108 p->plane[0].rotation = crtc->primary->state->rotation;
3109
3110 fb = crtc->cursor->state->fb;
3111 p->plane[PLANE_CURSOR].y_bytes_per_pixel = 0;
3112 if (fb) {
3113 p->plane[PLANE_CURSOR].enabled = true;
3114 p->plane[PLANE_CURSOR].bytes_per_pixel = fb->bits_per_pixel / 8;
3115 p->plane[PLANE_CURSOR].horiz_pixels = crtc->cursor->state->crtc_w;
3116 p->plane[PLANE_CURSOR].vert_pixels = crtc->cursor->state->crtc_h;
3117 } else {
3118 p->plane[PLANE_CURSOR].enabled = false;
3119 p->plane[PLANE_CURSOR].bytes_per_pixel = 0;
3120 p->plane[PLANE_CURSOR].horiz_pixels = 64;
3121 p->plane[PLANE_CURSOR].vert_pixels = 64;
3122 }
3123 }
3124
3125 list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
3126 struct intel_plane *intel_plane = to_intel_plane(plane);
3127
3128 if (intel_plane->pipe == pipe &&
3129 plane->type == DRM_PLANE_TYPE_OVERLAY)
3130 p->plane[i++] = intel_plane->wm;
3131 }
261a27d1
MR
3132}
3133
d4c2aa60 3134static bool skl_compute_plane_wm(const struct drm_i915_private *dev_priv,
2791a16c
PZ
3135 struct skl_pipe_wm_parameters *p,
3136 struct intel_plane_wm_parameters *p_params,
afb024aa 3137 uint16_t ddb_allocation,
d4c2aa60 3138 int level,
afb024aa
DL
3139 uint16_t *out_blocks, /* out */
3140 uint8_t *out_lines /* out */)
2d41c0b5 3141{
d4c2aa60
TU
3142 uint32_t latency = dev_priv->wm.skl_latency[level];
3143 uint32_t method1, method2;
3144 uint32_t plane_bytes_per_line, plane_blocks_per_line;
3145 uint32_t res_blocks, res_lines;
3146 uint32_t selected_result;
2cd601c6 3147 uint8_t bytes_per_pixel;
2d41c0b5 3148
2791a16c 3149 if (latency == 0 || !p->active || !p_params->enabled)
2d41c0b5
PB
3150 return false;
3151
2791a16c
PZ
3152 bytes_per_pixel = p_params->y_bytes_per_pixel ?
3153 p_params->y_bytes_per_pixel :
3154 p_params->bytes_per_pixel;
3155 method1 = skl_wm_method1(p->pixel_rate,
2cd601c6 3156 bytes_per_pixel,
d4c2aa60 3157 latency);
2791a16c
PZ
3158 method2 = skl_wm_method2(p->pixel_rate,
3159 p->pipe_htotal,
3160 p_params->horiz_pixels,
2cd601c6 3161 bytes_per_pixel,
2791a16c 3162 p_params->tiling,
d4c2aa60 3163 latency);
2d41c0b5 3164
2791a16c 3165 plane_bytes_per_line = p_params->horiz_pixels * bytes_per_pixel;
d4c2aa60 3166 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
2d41c0b5 3167
2791a16c
PZ
3168 if (p_params->tiling == I915_FORMAT_MOD_Y_TILED ||
3169 p_params->tiling == I915_FORMAT_MOD_Yf_TILED) {
1fc0a8f7
TU
3170 uint32_t min_scanlines = 4;
3171 uint32_t y_tile_minimum;
2791a16c
PZ
3172 if (intel_rotation_90_or_270(p_params->rotation)) {
3173 switch (p_params->bytes_per_pixel) {
1fc0a8f7
TU
3174 case 1:
3175 min_scanlines = 16;
3176 break;
3177 case 2:
3178 min_scanlines = 8;
3179 break;
3180 case 8:
3181 WARN(1, "Unsupported pixel depth for rotation");
2f0b5790 3182 }
1fc0a8f7
TU
3183 }
3184 y_tile_minimum = plane_blocks_per_line * min_scanlines;
0fda6568
TU
3185 selected_result = max(method2, y_tile_minimum);
3186 } else {
3187 if ((ddb_allocation / plane_blocks_per_line) >= 1)
3188 selected_result = min(method1, method2);
3189 else
3190 selected_result = method1;
3191 }
2d41c0b5 3192
d4c2aa60
TU
3193 res_blocks = selected_result + 1;
3194 res_lines = DIV_ROUND_UP(selected_result, plane_blocks_per_line);
e6d66171 3195
0fda6568 3196 if (level >= 1 && level <= 7) {
2791a16c
PZ
3197 if (p_params->tiling == I915_FORMAT_MOD_Y_TILED ||
3198 p_params->tiling == I915_FORMAT_MOD_Yf_TILED)
0fda6568
TU
3199 res_lines += 4;
3200 else
3201 res_blocks++;
3202 }
e6d66171 3203
d4c2aa60 3204 if (res_blocks >= ddb_allocation || res_lines > 31)
e6d66171
DL
3205 return false;
3206
3207 *out_blocks = res_blocks;
3208 *out_lines = res_lines;
2d41c0b5
PB
3209
3210 return true;
3211}
3212
3213static void skl_compute_wm_level(const struct drm_i915_private *dev_priv,
3214 struct skl_ddb_allocation *ddb,
2791a16c
PZ
3215 struct skl_pipe_wm_parameters *p,
3216 enum pipe pipe,
2d41c0b5 3217 int level,
2791a16c 3218 int num_planes,
2d41c0b5
PB
3219 struct skl_wm_level *result)
3220{
2d41c0b5 3221 uint16_t ddb_blocks;
2791a16c 3222 int i;
2d41c0b5 3223
2791a16c 3224 for (i = 0; i < num_planes; i++) {
2d41c0b5
PB
3225 ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]);
3226
d4c2aa60 3227 result->plane_en[i] = skl_compute_plane_wm(dev_priv,
2791a16c 3228 p, &p->plane[i],
2d41c0b5 3229 ddb_blocks,
d4c2aa60 3230 level,
2d41c0b5
PB
3231 &result->plane_res_b[i],
3232 &result->plane_res_l[i]);
3233 }
2791a16c
PZ
3234
3235 ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][PLANE_CURSOR]);
3236 result->plane_en[PLANE_CURSOR] = skl_compute_plane_wm(dev_priv, p,
3237 &p->plane[PLANE_CURSOR],
3238 ddb_blocks, level,
3239 &result->plane_res_b[PLANE_CURSOR],
3240 &result->plane_res_l[PLANE_CURSOR]);
2d41c0b5
PB
3241}
3242
407b50f3 3243static uint32_t
2791a16c 3244skl_compute_linetime_wm(struct drm_crtc *crtc, struct skl_pipe_wm_parameters *p)
407b50f3 3245{
2791a16c 3246 if (!to_intel_crtc(crtc)->active)
407b50f3
DL
3247 return 0;
3248
2791a16c 3249 if (WARN_ON(p->pixel_rate == 0))
661abfc0 3250 return 0;
407b50f3 3251
2791a16c 3252 return DIV_ROUND_UP(8 * p->pipe_htotal * 1000, p->pixel_rate);
407b50f3
DL
3253}
3254
2791a16c
PZ
3255static void skl_compute_transition_wm(struct drm_crtc *crtc,
3256 struct skl_pipe_wm_parameters *params,
9414f563 3257 struct skl_wm_level *trans_wm /* out */)
407b50f3 3258{
9414f563 3259 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2791a16c 3260 int i;
9414f563 3261
2791a16c 3262 if (!params->active)
407b50f3 3263 return;
9414f563
DL
3264
3265 /* Until we know more, just disable transition WMs */
2791a16c 3266 for (i = 0; i < intel_num_planes(intel_crtc); i++)
9414f563 3267 trans_wm->plane_en[i] = false;
2791a16c 3268 trans_wm->plane_en[PLANE_CURSOR] = false;
407b50f3
DL
3269}
3270
2791a16c 3271static void skl_compute_pipe_wm(struct drm_crtc *crtc,
2d41c0b5 3272 struct skl_ddb_allocation *ddb,
2791a16c 3273 struct skl_pipe_wm_parameters *params,
2d41c0b5
PB
3274 struct skl_pipe_wm *pipe_wm)
3275{
2791a16c 3276 struct drm_device *dev = crtc->dev;
2d41c0b5 3277 const struct drm_i915_private *dev_priv = dev->dev_private;
2791a16c 3278 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2d41c0b5
PB
3279 int level, max_level = ilk_wm_max_level(dev);
3280
3281 for (level = 0; level <= max_level; level++) {
2791a16c
PZ
3282 skl_compute_wm_level(dev_priv, ddb, params, intel_crtc->pipe,
3283 level, intel_num_planes(intel_crtc),
3284 &pipe_wm->wm[level]);
2d41c0b5 3285 }
2791a16c 3286 pipe_wm->linetime = skl_compute_linetime_wm(crtc, params);
2d41c0b5 3287
2791a16c 3288 skl_compute_transition_wm(crtc, params, &pipe_wm->trans_wm);
2d41c0b5
PB
3289}
3290
3291static void skl_compute_wm_results(struct drm_device *dev,
2791a16c 3292 struct skl_pipe_wm_parameters *p,
2d41c0b5
PB
3293 struct skl_pipe_wm *p_wm,
3294 struct skl_wm_values *r,
3295 struct intel_crtc *intel_crtc)
3296{
3297 int level, max_level = ilk_wm_max_level(dev);
3298 enum pipe pipe = intel_crtc->pipe;
9414f563
DL
3299 uint32_t temp;
3300 int i;
2d41c0b5
PB
3301
3302 for (level = 0; level <= max_level; level++) {
2d41c0b5
PB
3303 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
3304 temp = 0;
2d41c0b5
PB
3305
3306 temp |= p_wm->wm[level].plane_res_l[i] <<
3307 PLANE_WM_LINES_SHIFT;
3308 temp |= p_wm->wm[level].plane_res_b[i];
3309 if (p_wm->wm[level].plane_en[i])
3310 temp |= PLANE_WM_EN;
3311
3312 r->plane[pipe][i][level] = temp;
2d41c0b5
PB
3313 }
3314
3315 temp = 0;
2d41c0b5 3316
4969d33e
MR
3317 temp |= p_wm->wm[level].plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT;
3318 temp |= p_wm->wm[level].plane_res_b[PLANE_CURSOR];
2d41c0b5 3319
4969d33e 3320 if (p_wm->wm[level].plane_en[PLANE_CURSOR])
2d41c0b5
PB
3321 temp |= PLANE_WM_EN;
3322
4969d33e 3323 r->plane[pipe][PLANE_CURSOR][level] = temp;
2d41c0b5
PB
3324
3325 }
3326
9414f563
DL
3327 /* transition WMs */
3328 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
3329 temp = 0;
3330 temp |= p_wm->trans_wm.plane_res_l[i] << PLANE_WM_LINES_SHIFT;
3331 temp |= p_wm->trans_wm.plane_res_b[i];
3332 if (p_wm->trans_wm.plane_en[i])
3333 temp |= PLANE_WM_EN;
3334
3335 r->plane_trans[pipe][i] = temp;
3336 }
3337
3338 temp = 0;
4969d33e
MR
3339 temp |= p_wm->trans_wm.plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT;
3340 temp |= p_wm->trans_wm.plane_res_b[PLANE_CURSOR];
3341 if (p_wm->trans_wm.plane_en[PLANE_CURSOR])
9414f563
DL
3342 temp |= PLANE_WM_EN;
3343
4969d33e 3344 r->plane_trans[pipe][PLANE_CURSOR] = temp;
9414f563 3345
2d41c0b5
PB
3346 r->wm_linetime[pipe] = p_wm->linetime;
3347}
3348
16160e3d
DL
3349static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, uint32_t reg,
3350 const struct skl_ddb_entry *entry)
3351{
3352 if (entry->end)
3353 I915_WRITE(reg, (entry->end - 1) << 16 | entry->start);
3354 else
3355 I915_WRITE(reg, 0);
3356}
3357
2d41c0b5
PB
3358static void skl_write_wm_values(struct drm_i915_private *dev_priv,
3359 const struct skl_wm_values *new)
3360{
3361 struct drm_device *dev = dev_priv->dev;
3362 struct intel_crtc *crtc;
3363
3364 list_for_each_entry(crtc, &dev->mode_config.crtc_list, base.head) {
3365 int i, level, max_level = ilk_wm_max_level(dev);
3366 enum pipe pipe = crtc->pipe;
3367
5d374d96
DL
3368 if (!new->dirty[pipe])
3369 continue;
8211bd5b 3370
5d374d96 3371 I915_WRITE(PIPE_WM_LINETIME(pipe), new->wm_linetime[pipe]);
8211bd5b 3372
5d374d96
DL
3373 for (level = 0; level <= max_level; level++) {
3374 for (i = 0; i < intel_num_planes(crtc); i++)
3375 I915_WRITE(PLANE_WM(pipe, i, level),
3376 new->plane[pipe][i][level]);
3377 I915_WRITE(CUR_WM(pipe, level),
4969d33e 3378 new->plane[pipe][PLANE_CURSOR][level]);
2d41c0b5 3379 }
5d374d96
DL
3380 for (i = 0; i < intel_num_planes(crtc); i++)
3381 I915_WRITE(PLANE_WM_TRANS(pipe, i),
3382 new->plane_trans[pipe][i]);
4969d33e
MR
3383 I915_WRITE(CUR_WM_TRANS(pipe),
3384 new->plane_trans[pipe][PLANE_CURSOR]);
5d374d96 3385
2cd601c6 3386 for (i = 0; i < intel_num_planes(crtc); i++) {
5d374d96
DL
3387 skl_ddb_entry_write(dev_priv,
3388 PLANE_BUF_CFG(pipe, i),
3389 &new->ddb.plane[pipe][i]);
2cd601c6
CK
3390 skl_ddb_entry_write(dev_priv,
3391 PLANE_NV12_BUF_CFG(pipe, i),
3392 &new->ddb.y_plane[pipe][i]);
3393 }
5d374d96
DL
3394
3395 skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe),
4969d33e 3396 &new->ddb.plane[pipe][PLANE_CURSOR]);
2d41c0b5 3397 }
2d41c0b5
PB
3398}
3399
0e8fb7ba
DL
3400/*
3401 * When setting up a new DDB allocation arrangement, we need to correctly
3402 * sequence the times at which the new allocations for the pipes are taken into
3403 * account or we'll have pipes fetching from space previously allocated to
3404 * another pipe.
3405 *
3406 * Roughly the sequence looks like:
3407 * 1. re-allocate the pipe(s) with the allocation being reduced and not
3408 * overlapping with a previous light-up pipe (another way to put it is:
3409 * pipes with their new allocation strickly included into their old ones).
3410 * 2. re-allocate the other pipes that get their allocation reduced
3411 * 3. allocate the pipes having their allocation increased
3412 *
3413 * Steps 1. and 2. are here to take care of the following case:
3414 * - Initially DDB looks like this:
3415 * | B | C |
3416 * - enable pipe A.
3417 * - pipe B has a reduced DDB allocation that overlaps with the old pipe C
3418 * allocation
3419 * | A | B | C |
3420 *
3421 * We need to sequence the re-allocation: C, B, A (and not B, C, A).
3422 */
3423
d21b795c
DL
3424static void
3425skl_wm_flush_pipe(struct drm_i915_private *dev_priv, enum pipe pipe, int pass)
0e8fb7ba 3426{
0e8fb7ba
DL
3427 int plane;
3428
d21b795c
DL
3429 DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe), pass);
3430
dd740780 3431 for_each_plane(dev_priv, pipe, plane) {
0e8fb7ba
DL
3432 I915_WRITE(PLANE_SURF(pipe, plane),
3433 I915_READ(PLANE_SURF(pipe, plane)));
3434 }
3435 I915_WRITE(CURBASE(pipe), I915_READ(CURBASE(pipe)));
3436}
3437
3438static bool
3439skl_ddb_allocation_included(const struct skl_ddb_allocation *old,
3440 const struct skl_ddb_allocation *new,
3441 enum pipe pipe)
3442{
3443 uint16_t old_size, new_size;
3444
3445 old_size = skl_ddb_entry_size(&old->pipe[pipe]);
3446 new_size = skl_ddb_entry_size(&new->pipe[pipe]);
3447
3448 return old_size != new_size &&
3449 new->pipe[pipe].start >= old->pipe[pipe].start &&
3450 new->pipe[pipe].end <= old->pipe[pipe].end;
3451}
3452
3453static void skl_flush_wm_values(struct drm_i915_private *dev_priv,
3454 struct skl_wm_values *new_values)
3455{
3456 struct drm_device *dev = dev_priv->dev;
3457 struct skl_ddb_allocation *cur_ddb, *new_ddb;
c929cb45 3458 bool reallocated[I915_MAX_PIPES] = {};
0e8fb7ba
DL
3459 struct intel_crtc *crtc;
3460 enum pipe pipe;
3461
3462 new_ddb = &new_values->ddb;
3463 cur_ddb = &dev_priv->wm.skl_hw.ddb;
3464
3465 /*
3466 * First pass: flush the pipes with the new allocation contained into
3467 * the old space.
3468 *
3469 * We'll wait for the vblank on those pipes to ensure we can safely
3470 * re-allocate the freed space without this pipe fetching from it.
3471 */
3472 for_each_intel_crtc(dev, crtc) {
3473 if (!crtc->active)
3474 continue;
3475
3476 pipe = crtc->pipe;
3477
3478 if (!skl_ddb_allocation_included(cur_ddb, new_ddb, pipe))
3479 continue;
3480
d21b795c 3481 skl_wm_flush_pipe(dev_priv, pipe, 1);
0e8fb7ba
DL
3482 intel_wait_for_vblank(dev, pipe);
3483
3484 reallocated[pipe] = true;
3485 }
3486
3487
3488 /*
3489 * Second pass: flush the pipes that are having their allocation
3490 * reduced, but overlapping with a previous allocation.
3491 *
3492 * Here as well we need to wait for the vblank to make sure the freed
3493 * space is not used anymore.
3494 */
3495 for_each_intel_crtc(dev, crtc) {
3496 if (!crtc->active)
3497 continue;
3498
3499 pipe = crtc->pipe;
3500
3501 if (reallocated[pipe])
3502 continue;
3503
3504 if (skl_ddb_entry_size(&new_ddb->pipe[pipe]) <
3505 skl_ddb_entry_size(&cur_ddb->pipe[pipe])) {
d21b795c 3506 skl_wm_flush_pipe(dev_priv, pipe, 2);
0e8fb7ba 3507 intel_wait_for_vblank(dev, pipe);
d9d8e6b3 3508 reallocated[pipe] = true;
0e8fb7ba 3509 }
0e8fb7ba
DL
3510 }
3511
3512 /*
3513 * Third pass: flush the pipes that got more space allocated.
3514 *
3515 * We don't need to actively wait for the update here, next vblank
3516 * will just get more DDB space with the correct WM values.
3517 */
3518 for_each_intel_crtc(dev, crtc) {
3519 if (!crtc->active)
3520 continue;
3521
3522 pipe = crtc->pipe;
3523
3524 /*
3525 * At this point, only the pipes more space than before are
3526 * left to re-allocate.
3527 */
3528 if (reallocated[pipe])
3529 continue;
3530
d21b795c 3531 skl_wm_flush_pipe(dev_priv, pipe, 3);
0e8fb7ba
DL
3532 }
3533}
3534
2d41c0b5 3535static bool skl_update_pipe_wm(struct drm_crtc *crtc,
2791a16c 3536 struct skl_pipe_wm_parameters *params,
261a27d1 3537 struct intel_wm_config *config,
2d41c0b5
PB
3538 struct skl_ddb_allocation *ddb, /* out */
3539 struct skl_pipe_wm *pipe_wm /* out */)
3540{
3541 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3542
2791a16c
PZ
3543 skl_compute_wm_pipe_parameters(crtc, params);
3544 skl_allocate_pipe_ddb(crtc, config, params, ddb);
3545 skl_compute_pipe_wm(crtc, ddb, params, pipe_wm);
2d41c0b5 3546
261a27d1 3547 if (!memcmp(&intel_crtc->wm.skl_active, pipe_wm, sizeof(*pipe_wm)))
2d41c0b5
PB
3548 return false;
3549
261a27d1 3550 intel_crtc->wm.skl_active = *pipe_wm;
2cd601c6 3551
2d41c0b5
PB
3552 return true;
3553}
3554
3555static void skl_update_other_pipe_wm(struct drm_device *dev,
3556 struct drm_crtc *crtc,
261a27d1 3557 struct intel_wm_config *config,
2d41c0b5
PB
3558 struct skl_wm_values *r)
3559{
3560 struct intel_crtc *intel_crtc;
3561 struct intel_crtc *this_crtc = to_intel_crtc(crtc);
3562
3563 /*
3564 * If the WM update hasn't changed the allocation for this_crtc (the
3565 * crtc we are currently computing the new WM values for), other
3566 * enabled crtcs will keep the same allocation and we don't need to
3567 * recompute anything for them.
3568 */
3569 if (!skl_ddb_allocation_changed(&r->ddb, this_crtc))
3570 return;
3571
3572 /*
3573 * Otherwise, because of this_crtc being freshly enabled/disabled, the
3574 * other active pipes need new DDB allocation and WM values.
3575 */
3576 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list,
3577 base.head) {
2791a16c 3578 struct skl_pipe_wm_parameters params = {};
2d41c0b5
PB
3579 struct skl_pipe_wm pipe_wm = {};
3580 bool wm_changed;
3581
3582 if (this_crtc->pipe == intel_crtc->pipe)
3583 continue;
3584
3585 if (!intel_crtc->active)
3586 continue;
3587
2791a16c
PZ
3588 wm_changed = skl_update_pipe_wm(&intel_crtc->base,
3589 &params, config,
2d41c0b5
PB
3590 &r->ddb, &pipe_wm);
3591
3592 /*
3593 * If we end up re-computing the other pipe WM values, it's
3594 * because it was really needed, so we expect the WM values to
3595 * be different.
3596 */
3597 WARN_ON(!wm_changed);
3598
2791a16c 3599 skl_compute_wm_results(dev, &params, &pipe_wm, r, intel_crtc);
2d41c0b5
PB
3600 r->dirty[intel_crtc->pipe] = true;
3601 }
3602}
3603
adda50b8
BP
3604static void skl_clear_wm(struct skl_wm_values *watermarks, enum pipe pipe)
3605{
3606 watermarks->wm_linetime[pipe] = 0;
3607 memset(watermarks->plane[pipe], 0,
3608 sizeof(uint32_t) * 8 * I915_MAX_PLANES);
adda50b8
BP
3609 memset(watermarks->plane_trans[pipe],
3610 0, sizeof(uint32_t) * I915_MAX_PLANES);
4969d33e 3611 watermarks->plane_trans[pipe][PLANE_CURSOR] = 0;
adda50b8
BP
3612
3613 /* Clear ddb entries for pipe */
3614 memset(&watermarks->ddb.pipe[pipe], 0, sizeof(struct skl_ddb_entry));
3615 memset(&watermarks->ddb.plane[pipe], 0,
3616 sizeof(struct skl_ddb_entry) * I915_MAX_PLANES);
3617 memset(&watermarks->ddb.y_plane[pipe], 0,
3618 sizeof(struct skl_ddb_entry) * I915_MAX_PLANES);
4969d33e
MR
3619 memset(&watermarks->ddb.plane[pipe][PLANE_CURSOR], 0,
3620 sizeof(struct skl_ddb_entry));
adda50b8
BP
3621
3622}
3623
2d41c0b5
PB
3624static void skl_update_wm(struct drm_crtc *crtc)
3625{
3626 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3627 struct drm_device *dev = crtc->dev;
3628 struct drm_i915_private *dev_priv = dev->dev_private;
2791a16c 3629 struct skl_pipe_wm_parameters params = {};
2d41c0b5 3630 struct skl_wm_values *results = &dev_priv->wm.skl_results;
261a27d1
MR
3631 struct skl_pipe_wm pipe_wm = {};
3632 struct intel_wm_config config = {};
2d41c0b5 3633
adda50b8
BP
3634
3635 /* Clear all dirty flags */
3636 memset(results->dirty, 0, sizeof(bool) * I915_MAX_PIPES);
3637
3638 skl_clear_wm(results, intel_crtc->pipe);
2d41c0b5 3639
261a27d1
MR
3640 skl_compute_wm_global_parameters(dev, &config);
3641
2791a16c
PZ
3642 if (!skl_update_pipe_wm(crtc, &params, &config,
3643 &results->ddb, &pipe_wm))
2d41c0b5
PB
3644 return;
3645
2791a16c 3646 skl_compute_wm_results(dev, &params, &pipe_wm, results, intel_crtc);
2d41c0b5
PB
3647 results->dirty[intel_crtc->pipe] = true;
3648
261a27d1 3649 skl_update_other_pipe_wm(dev, crtc, &config, results);
2d41c0b5 3650 skl_write_wm_values(dev_priv, results);
0e8fb7ba 3651 skl_flush_wm_values(dev_priv, results);
53b0deb4
DL
3652
3653 /* store the new configuration */
3654 dev_priv->wm.skl_hw = *results;
2d41c0b5
PB
3655}
3656
2791a16c
PZ
3657static void
3658skl_update_sprite_wm(struct drm_plane *plane, struct drm_crtc *crtc,
3659 uint32_t sprite_width, uint32_t sprite_height,
3660 int pixel_size, bool enabled, bool scaled)
3661{
3662 struct intel_plane *intel_plane = to_intel_plane(plane);
3663 struct drm_framebuffer *fb = plane->state->fb;
3664
3665 intel_plane->wm.enabled = enabled;
3666 intel_plane->wm.scaled = scaled;
3667 intel_plane->wm.horiz_pixels = sprite_width;
3668 intel_plane->wm.vert_pixels = sprite_height;
3669 intel_plane->wm.tiling = DRM_FORMAT_MOD_NONE;
3670
3671 /* For planar: Bpp is for UV plane, y_Bpp is for Y plane */
3672 intel_plane->wm.bytes_per_pixel =
3673 (fb && fb->pixel_format == DRM_FORMAT_NV12) ?
3674 drm_format_plane_cpp(plane->state->fb->pixel_format, 1) : pixel_size;
3675 intel_plane->wm.y_bytes_per_pixel =
3676 (fb && fb->pixel_format == DRM_FORMAT_NV12) ?
3677 drm_format_plane_cpp(plane->state->fb->pixel_format, 0) : 0;
3678
3679 /*
3680 * Framebuffer can be NULL on plane disable, but it does not
3681 * matter for watermarks if we assume no tiling in that case.
3682 */
3683 if (fb)
3684 intel_plane->wm.tiling = fb->modifier[0];
3685 intel_plane->wm.rotation = plane->state->rotation;
3686
3687 skl_update_wm(crtc);
3688}
3689
261a27d1 3690static void ilk_update_wm(struct drm_crtc *crtc)
801bcfff 3691{
261a27d1
MR
3692 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3693 struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state);
3694 struct drm_device *dev = crtc->dev;
3695 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 3696 struct ilk_wm_maximums max;
820c1980 3697 struct ilk_wm_values results = {};
77c122bc 3698 enum intel_ddb_partitioning partitioning;
261a27d1
MR
3699 struct intel_pipe_wm pipe_wm = {};
3700 struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
3701 struct intel_wm_config config = {};
861f3389 3702
261a27d1
MR
3703 WARN_ON(cstate->base.active != intel_crtc->active);
3704
261a27d1
MR
3705 intel_compute_pipe_wm(cstate, &pipe_wm);
3706
3707 if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
3708 return;
3709
3710 intel_crtc->wm.active = pipe_wm;
3711
3712 ilk_compute_wm_config(dev, &config);
3713
3714 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
3715 ilk_wm_merge(dev, &config, &max, &lp_wm_1_2);
a485bfb8
VS
3716
3717 /* 5/6 split only in single pipe config on IVB+ */
ec98c8d1 3718 if (INTEL_INFO(dev)->gen >= 7 &&
261a27d1
MR
3719 config.num_pipes_active == 1 && config.sprites_enabled) {
3720 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
3721 ilk_wm_merge(dev, &config, &max, &lp_wm_5_6);
0362c781 3722
820c1980 3723 best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
861f3389 3724 } else {
198a1e9b 3725 best_lp_wm = &lp_wm_1_2;
861f3389
PZ
3726 }
3727
198a1e9b 3728 partitioning = (best_lp_wm == &lp_wm_1_2) ?
77c122bc 3729 INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
801bcfff 3730
820c1980 3731 ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results);
609cedef 3732
820c1980 3733 ilk_write_wm_values(dev_priv, &results);
1011d8c4
PZ
3734}
3735
2791a16c
PZ
3736static void
3737ilk_update_sprite_wm(struct drm_plane *plane,
3738 struct drm_crtc *crtc,
3739 uint32_t sprite_width, uint32_t sprite_height,
3740 int pixel_size, bool enabled, bool scaled)
3741{
3742 struct drm_device *dev = plane->dev;
3743 struct intel_plane *intel_plane = to_intel_plane(plane);
3744
3745 /*
3746 * IVB workaround: must disable low power watermarks for at least
3747 * one frame before enabling scaling. LP watermarks can be re-enabled
3748 * when scaling is disabled.
3749 *
3750 * WaCxSRDisabledForSpriteScaling:ivb
3751 */
3752 if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev))
3753 intel_wait_for_vblank(dev, intel_plane->pipe);
3754
3755 ilk_update_wm(crtc);
3756}
3757
3078999f
PB
3758static void skl_pipe_wm_active_state(uint32_t val,
3759 struct skl_pipe_wm *active,
3760 bool is_transwm,
3761 bool is_cursor,
3762 int i,
3763 int level)
3764{
3765 bool is_enabled = (val & PLANE_WM_EN) != 0;
3766
3767 if (!is_transwm) {
3768 if (!is_cursor) {
3769 active->wm[level].plane_en[i] = is_enabled;
3770 active->wm[level].plane_res_b[i] =
3771 val & PLANE_WM_BLOCKS_MASK;
3772 active->wm[level].plane_res_l[i] =
3773 (val >> PLANE_WM_LINES_SHIFT) &
3774 PLANE_WM_LINES_MASK;
3775 } else {
4969d33e
MR
3776 active->wm[level].plane_en[PLANE_CURSOR] = is_enabled;
3777 active->wm[level].plane_res_b[PLANE_CURSOR] =
3078999f 3778 val & PLANE_WM_BLOCKS_MASK;
4969d33e 3779 active->wm[level].plane_res_l[PLANE_CURSOR] =
3078999f
PB
3780 (val >> PLANE_WM_LINES_SHIFT) &
3781 PLANE_WM_LINES_MASK;
3782 }
3783 } else {
3784 if (!is_cursor) {
3785 active->trans_wm.plane_en[i] = is_enabled;
3786 active->trans_wm.plane_res_b[i] =
3787 val & PLANE_WM_BLOCKS_MASK;
3788 active->trans_wm.plane_res_l[i] =
3789 (val >> PLANE_WM_LINES_SHIFT) &
3790 PLANE_WM_LINES_MASK;
3791 } else {
4969d33e
MR
3792 active->trans_wm.plane_en[PLANE_CURSOR] = is_enabled;
3793 active->trans_wm.plane_res_b[PLANE_CURSOR] =
3078999f 3794 val & PLANE_WM_BLOCKS_MASK;
4969d33e 3795 active->trans_wm.plane_res_l[PLANE_CURSOR] =
3078999f
PB
3796 (val >> PLANE_WM_LINES_SHIFT) &
3797 PLANE_WM_LINES_MASK;
3798 }
3799 }
3800}
3801
3802static void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc)
3803{
3804 struct drm_device *dev = crtc->dev;
3805 struct drm_i915_private *dev_priv = dev->dev_private;
3806 struct skl_wm_values *hw = &dev_priv->wm.skl_hw;
3807 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
261a27d1 3808 struct skl_pipe_wm *active = &intel_crtc->wm.skl_active;
3078999f
PB
3809 enum pipe pipe = intel_crtc->pipe;
3810 int level, i, max_level;
3811 uint32_t temp;
3812
3813 max_level = ilk_wm_max_level(dev);
3814
3815 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
3816
3817 for (level = 0; level <= max_level; level++) {
3818 for (i = 0; i < intel_num_planes(intel_crtc); i++)
3819 hw->plane[pipe][i][level] =
3820 I915_READ(PLANE_WM(pipe, i, level));
4969d33e 3821 hw->plane[pipe][PLANE_CURSOR][level] = I915_READ(CUR_WM(pipe, level));
3078999f
PB
3822 }
3823
3824 for (i = 0; i < intel_num_planes(intel_crtc); i++)
3825 hw->plane_trans[pipe][i] = I915_READ(PLANE_WM_TRANS(pipe, i));
4969d33e 3826 hw->plane_trans[pipe][PLANE_CURSOR] = I915_READ(CUR_WM_TRANS(pipe));
3078999f 3827
3ef00284 3828 if (!intel_crtc->active)
3078999f
PB
3829 return;
3830
3831 hw->dirty[pipe] = true;
3832
3833 active->linetime = hw->wm_linetime[pipe];
3834
3835 for (level = 0; level <= max_level; level++) {
3836 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
3837 temp = hw->plane[pipe][i][level];
3838 skl_pipe_wm_active_state(temp, active, false,
3839 false, i, level);
3840 }
4969d33e 3841 temp = hw->plane[pipe][PLANE_CURSOR][level];
3078999f
PB
3842 skl_pipe_wm_active_state(temp, active, false, true, i, level);
3843 }
3844
3845 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
3846 temp = hw->plane_trans[pipe][i];
3847 skl_pipe_wm_active_state(temp, active, true, false, i, 0);
3848 }
3849
4969d33e 3850 temp = hw->plane_trans[pipe][PLANE_CURSOR];
3078999f
PB
3851 skl_pipe_wm_active_state(temp, active, true, true, i, 0);
3852}
3853
3854void skl_wm_get_hw_state(struct drm_device *dev)
3855{
a269c583
DL
3856 struct drm_i915_private *dev_priv = dev->dev_private;
3857 struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb;
3078999f
PB
3858 struct drm_crtc *crtc;
3859
a269c583 3860 skl_ddb_get_hw_state(dev_priv, ddb);
3078999f
PB
3861 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
3862 skl_pipe_wm_get_hw_state(crtc);
3863}
3864
243e6a44
VS
3865static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
3866{
3867 struct drm_device *dev = crtc->dev;
3868 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 3869 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44 3870 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
261a27d1 3871 struct intel_pipe_wm *active = &intel_crtc->wm.active;
243e6a44
VS
3872 enum pipe pipe = intel_crtc->pipe;
3873 static const unsigned int wm0_pipe_reg[] = {
3874 [PIPE_A] = WM0_PIPEA_ILK,
3875 [PIPE_B] = WM0_PIPEB_ILK,
3876 [PIPE_C] = WM0_PIPEC_IVB,
3877 };
3878
3879 hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
a42a5719 3880 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 3881 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
243e6a44 3882
3ef00284 3883 active->pipe_enabled = intel_crtc->active;
2a44b76b
VS
3884
3885 if (active->pipe_enabled) {
243e6a44
VS
3886 u32 tmp = hw->wm_pipe[pipe];
3887
3888 /*
3889 * For active pipes LP0 watermark is marked as
3890 * enabled, and LP1+ watermaks as disabled since
3891 * we can't really reverse compute them in case
3892 * multiple pipes are active.
3893 */
3894 active->wm[0].enable = true;
3895 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
3896 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
3897 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
3898 active->linetime = hw->wm_linetime[pipe];
3899 } else {
3900 int level, max_level = ilk_wm_max_level(dev);
3901
3902 /*
3903 * For inactive pipes, all watermark levels
3904 * should be marked as enabled but zeroed,
3905 * which is what we'd compute them to.
3906 */
3907 for (level = 0; level <= max_level; level++)
3908 active->wm[level].enable = true;
3909 }
3910}
3911
6eb1a681
VS
3912#define _FW_WM(value, plane) \
3913 (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT)
3914#define _FW_WM_VLV(value, plane) \
3915 (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT)
3916
3917static void vlv_read_wm_values(struct drm_i915_private *dev_priv,
3918 struct vlv_wm_values *wm)
3919{
3920 enum pipe pipe;
3921 uint32_t tmp;
3922
3923 for_each_pipe(dev_priv, pipe) {
3924 tmp = I915_READ(VLV_DDL(pipe));
3925
3926 wm->ddl[pipe].primary =
3927 (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
3928 wm->ddl[pipe].cursor =
3929 (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
3930 wm->ddl[pipe].sprite[0] =
3931 (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
3932 wm->ddl[pipe].sprite[1] =
3933 (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
3934 }
3935
3936 tmp = I915_READ(DSPFW1);
3937 wm->sr.plane = _FW_WM(tmp, SR);
3938 wm->pipe[PIPE_B].cursor = _FW_WM(tmp, CURSORB);
3939 wm->pipe[PIPE_B].primary = _FW_WM_VLV(tmp, PLANEB);
3940 wm->pipe[PIPE_A].primary = _FW_WM_VLV(tmp, PLANEA);
3941
3942 tmp = I915_READ(DSPFW2);
3943 wm->pipe[PIPE_A].sprite[1] = _FW_WM_VLV(tmp, SPRITEB);
3944 wm->pipe[PIPE_A].cursor = _FW_WM(tmp, CURSORA);
3945 wm->pipe[PIPE_A].sprite[0] = _FW_WM_VLV(tmp, SPRITEA);
3946
3947 tmp = I915_READ(DSPFW3);
3948 wm->sr.cursor = _FW_WM(tmp, CURSOR_SR);
3949
3950 if (IS_CHERRYVIEW(dev_priv)) {
3951 tmp = I915_READ(DSPFW7_CHV);
3952 wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED);
3953 wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC);
3954
3955 tmp = I915_READ(DSPFW8_CHV);
3956 wm->pipe[PIPE_C].sprite[1] = _FW_WM_VLV(tmp, SPRITEF);
3957 wm->pipe[PIPE_C].sprite[0] = _FW_WM_VLV(tmp, SPRITEE);
3958
3959 tmp = I915_READ(DSPFW9_CHV);
3960 wm->pipe[PIPE_C].primary = _FW_WM_VLV(tmp, PLANEC);
3961 wm->pipe[PIPE_C].cursor = _FW_WM(tmp, CURSORC);
3962
3963 tmp = I915_READ(DSPHOWM);
3964 wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
3965 wm->pipe[PIPE_C].sprite[1] |= _FW_WM(tmp, SPRITEF_HI) << 8;
3966 wm->pipe[PIPE_C].sprite[0] |= _FW_WM(tmp, SPRITEE_HI) << 8;
3967 wm->pipe[PIPE_C].primary |= _FW_WM(tmp, PLANEC_HI) << 8;
3968 wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8;
3969 wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
3970 wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8;
3971 wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
3972 wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
3973 wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8;
3974 } else {
3975 tmp = I915_READ(DSPFW7);
3976 wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED);
3977 wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC);
3978
3979 tmp = I915_READ(DSPHOWM);
3980 wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
3981 wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8;
3982 wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
3983 wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8;
3984 wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
3985 wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
3986 wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8;
3987 }
3988}
3989
3990#undef _FW_WM
3991#undef _FW_WM_VLV
3992
3993void vlv_wm_get_hw_state(struct drm_device *dev)
3994{
3995 struct drm_i915_private *dev_priv = to_i915(dev);
3996 struct vlv_wm_values *wm = &dev_priv->wm.vlv;
3997 struct intel_plane *plane;
3998 enum pipe pipe;
3999 u32 val;
4000
4001 vlv_read_wm_values(dev_priv, wm);
4002
4003 for_each_intel_plane(dev, plane) {
4004 switch (plane->base.type) {
4005 int sprite;
4006 case DRM_PLANE_TYPE_CURSOR:
4007 plane->wm.fifo_size = 63;
4008 break;
4009 case DRM_PLANE_TYPE_PRIMARY:
4010 plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, 0);
4011 break;
4012 case DRM_PLANE_TYPE_OVERLAY:
4013 sprite = plane->plane;
4014 plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, sprite + 1);
4015 break;
4016 }
4017 }
4018
4019 wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN;
4020 wm->level = VLV_WM_LEVEL_PM2;
4021
4022 if (IS_CHERRYVIEW(dev_priv)) {
4023 mutex_lock(&dev_priv->rps.hw_lock);
4024
4025 val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ);
4026 if (val & DSP_MAXFIFO_PM5_ENABLE)
4027 wm->level = VLV_WM_LEVEL_PM5;
4028
58590c14
VS
4029 /*
4030 * If DDR DVFS is disabled in the BIOS, Punit
4031 * will never ack the request. So if that happens
4032 * assume we don't have to enable/disable DDR DVFS
4033 * dynamically. To test that just set the REQ_ACK
4034 * bit to poke the Punit, but don't change the
4035 * HIGH/LOW bits so that we don't actually change
4036 * the current state.
4037 */
6eb1a681 4038 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
58590c14
VS
4039 val |= FORCE_DDR_FREQ_REQ_ACK;
4040 vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);
4041
4042 if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
4043 FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) {
4044 DRM_DEBUG_KMS("Punit not acking DDR DVFS request, "
4045 "assuming DDR DVFS is disabled\n");
4046 dev_priv->wm.max_level = VLV_WM_LEVEL_PM5;
4047 } else {
4048 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
4049 if ((val & FORCE_DDR_HIGH_FREQ) == 0)
4050 wm->level = VLV_WM_LEVEL_DDR_DVFS;
4051 }
6eb1a681
VS
4052
4053 mutex_unlock(&dev_priv->rps.hw_lock);
4054 }
4055
4056 for_each_pipe(dev_priv, pipe)
4057 DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n",
4058 pipe_name(pipe), wm->pipe[pipe].primary, wm->pipe[pipe].cursor,
4059 wm->pipe[pipe].sprite[0], wm->pipe[pipe].sprite[1]);
4060
4061 DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n",
4062 wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr);
4063}
4064
243e6a44
VS
4065void ilk_wm_get_hw_state(struct drm_device *dev)
4066{
4067 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 4068 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS
4069 struct drm_crtc *crtc;
4070
70e1e0ec 4071 for_each_crtc(dev, crtc)
243e6a44
VS
4072 ilk_pipe_wm_get_hw_state(crtc);
4073
4074 hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
4075 hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
4076 hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
4077
4078 hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
cfa7698b
VS
4079 if (INTEL_INFO(dev)->gen >= 7) {
4080 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
4081 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
4082 }
243e6a44 4083
a42a5719 4084 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ac9545fd
VS
4085 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
4086 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
4087 else if (IS_IVYBRIDGE(dev))
4088 hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
4089 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
243e6a44
VS
4090
4091 hw->enable_fbc_wm =
4092 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
4093}
4094
b445e3b0
ED
4095/**
4096 * intel_update_watermarks - update FIFO watermark values based on current modes
4097 *
4098 * Calculate watermark values for the various WM regs based on current mode
4099 * and plane configuration.
4100 *
4101 * There are several cases to deal with here:
4102 * - normal (i.e. non-self-refresh)
4103 * - self-refresh (SR) mode
4104 * - lines are large relative to FIFO size (buffer can hold up to 2)
4105 * - lines are small relative to FIFO size (buffer can hold more than 2
4106 * lines), so need to account for TLB latency
4107 *
4108 * The normal calculation is:
4109 * watermark = dotclock * bytes per pixel * latency
4110 * where latency is platform & configuration dependent (we assume pessimal
4111 * values here).
4112 *
4113 * The SR calculation is:
4114 * watermark = (trunc(latency/line time)+1) * surface width *
4115 * bytes per pixel
4116 * where
4117 * line time = htotal / dotclock
4118 * surface width = hdisplay for normal plane and 64 for cursor
4119 * and latency is assumed to be high, as above.
4120 *
4121 * The final value programmed to the register should always be rounded up,
4122 * and include an extra 2 entries to account for clock crossings.
4123 *
4124 * We don't use the sprite, so we can ignore that. And on Crestline we have
4125 * to set the non-SR watermarks to 8.
4126 */
46ba614c 4127void intel_update_watermarks(struct drm_crtc *crtc)
b445e3b0 4128{
46ba614c 4129 struct drm_i915_private *dev_priv = crtc->dev->dev_private;
b445e3b0
ED
4130
4131 if (dev_priv->display.update_wm)
46ba614c 4132 dev_priv->display.update_wm(crtc);
b445e3b0
ED
4133}
4134
2791a16c
PZ
4135void intel_update_sprite_watermarks(struct drm_plane *plane,
4136 struct drm_crtc *crtc,
4137 uint32_t sprite_width,
4138 uint32_t sprite_height,
4139 int pixel_size,
4140 bool enabled, bool scaled)
4141{
4142 struct drm_i915_private *dev_priv = plane->dev->dev_private;
4143
4144 if (dev_priv->display.update_sprite_wm)
4145 dev_priv->display.update_sprite_wm(plane, crtc,
4146 sprite_width, sprite_height,
4147 pixel_size, enabled, scaled);
4148}
4149
9270388e
DV
4150/**
4151 * Lock protecting IPS related data structures
9270388e
DV
4152 */
4153DEFINE_SPINLOCK(mchdev_lock);
4154
4155/* Global for IPS driver to get at the current i915 device. Protected by
4156 * mchdev_lock. */
4157static struct drm_i915_private *i915_mch_dev;
4158
2b4e57bd
ED
4159bool ironlake_set_drps(struct drm_device *dev, u8 val)
4160{
4161 struct drm_i915_private *dev_priv = dev->dev_private;
4162 u16 rgvswctl;
4163
9270388e
DV
4164 assert_spin_locked(&mchdev_lock);
4165
2b4e57bd
ED
4166 rgvswctl = I915_READ16(MEMSWCTL);
4167 if (rgvswctl & MEMCTL_CMD_STS) {
4168 DRM_DEBUG("gpu busy, RCS change rejected\n");
4169 return false; /* still busy with another command */
4170 }
4171
4172 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
4173 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
4174 I915_WRITE16(MEMSWCTL, rgvswctl);
4175 POSTING_READ16(MEMSWCTL);
4176
4177 rgvswctl |= MEMCTL_CMD_STS;
4178 I915_WRITE16(MEMSWCTL, rgvswctl);
4179
4180 return true;
4181}
4182
8090c6b9 4183static void ironlake_enable_drps(struct drm_device *dev)
2b4e57bd
ED
4184{
4185 struct drm_i915_private *dev_priv = dev->dev_private;
4186 u32 rgvmodectl = I915_READ(MEMMODECTL);
4187 u8 fmax, fmin, fstart, vstart;
4188
9270388e
DV
4189 spin_lock_irq(&mchdev_lock);
4190
2b4e57bd
ED
4191 /* Enable temp reporting */
4192 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
4193 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
4194
4195 /* 100ms RC evaluation intervals */
4196 I915_WRITE(RCUPEI, 100000);
4197 I915_WRITE(RCDNEI, 100000);
4198
4199 /* Set max/min thresholds to 90ms and 80ms respectively */
4200 I915_WRITE(RCBMAXAVG, 90000);
4201 I915_WRITE(RCBMINAVG, 80000);
4202
4203 I915_WRITE(MEMIHYST, 1);
4204
4205 /* Set up min, max, and cur for interrupt handling */
4206 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
4207 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
4208 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
4209 MEMMODE_FSTART_SHIFT;
4210
616847e7 4211 vstart = (I915_READ(PXVFREQ(fstart)) & PXVFREQ_PX_MASK) >>
2b4e57bd
ED
4212 PXVFREQ_PX_SHIFT;
4213
20e4d407
DV
4214 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
4215 dev_priv->ips.fstart = fstart;
2b4e57bd 4216
20e4d407
DV
4217 dev_priv->ips.max_delay = fstart;
4218 dev_priv->ips.min_delay = fmin;
4219 dev_priv->ips.cur_delay = fstart;
2b4e57bd
ED
4220
4221 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
4222 fmax, fmin, fstart);
4223
4224 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
4225
4226 /*
4227 * Interrupts will be enabled in ironlake_irq_postinstall
4228 */
4229
4230 I915_WRITE(VIDSTART, vstart);
4231 POSTING_READ(VIDSTART);
4232
4233 rgvmodectl |= MEMMODE_SWMODE_EN;
4234 I915_WRITE(MEMMODECTL, rgvmodectl);
4235
9270388e 4236 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2b4e57bd 4237 DRM_ERROR("stuck trying to change perf mode\n");
dd92d8de 4238 mdelay(1);
2b4e57bd
ED
4239
4240 ironlake_set_drps(dev, fstart);
4241
7d81c3e0
VS
4242 dev_priv->ips.last_count1 = I915_READ(DMIEC) +
4243 I915_READ(DDREC) + I915_READ(CSIEC);
20e4d407 4244 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
7d81c3e0 4245 dev_priv->ips.last_count2 = I915_READ(GFXEC);
5ed0bdf2 4246 dev_priv->ips.last_time2 = ktime_get_raw_ns();
9270388e
DV
4247
4248 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED
4249}
4250
8090c6b9 4251static void ironlake_disable_drps(struct drm_device *dev)
2b4e57bd
ED
4252{
4253 struct drm_i915_private *dev_priv = dev->dev_private;
9270388e
DV
4254 u16 rgvswctl;
4255
4256 spin_lock_irq(&mchdev_lock);
4257
4258 rgvswctl = I915_READ16(MEMSWCTL);
2b4e57bd
ED
4259
4260 /* Ack interrupts, disable EFC interrupt */
4261 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
4262 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
4263 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
4264 I915_WRITE(DEIIR, DE_PCU_EVENT);
4265 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
4266
4267 /* Go back to the starting frequency */
20e4d407 4268 ironlake_set_drps(dev, dev_priv->ips.fstart);
dd92d8de 4269 mdelay(1);
2b4e57bd
ED
4270 rgvswctl |= MEMCTL_CMD_STS;
4271 I915_WRITE(MEMSWCTL, rgvswctl);
dd92d8de 4272 mdelay(1);
2b4e57bd 4273
9270388e 4274 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED
4275}
4276
acbe9475
DV
4277/* There's a funny hw issue where the hw returns all 0 when reading from
4278 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
4279 * ourselves, instead of doing a rmw cycle (which might result in us clearing
4280 * all limits and the gpu stuck at whatever frequency it is at atm).
4281 */
74ef1173 4282static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val)
2b4e57bd 4283{
7b9e0ae6 4284 u32 limits;
2b4e57bd 4285
20b46e59
DV
4286 /* Only set the down limit when we've reached the lowest level to avoid
4287 * getting more interrupts, otherwise leave this clear. This prevents a
4288 * race in the hw when coming out of rc6: There's a tiny window where
4289 * the hw runs at the minimal clock before selecting the desired
4290 * frequency, if the down threshold expires in that window we will not
4291 * receive a down interrupt. */
74ef1173
AG
4292 if (IS_GEN9(dev_priv->dev)) {
4293 limits = (dev_priv->rps.max_freq_softlimit) << 23;
4294 if (val <= dev_priv->rps.min_freq_softlimit)
4295 limits |= (dev_priv->rps.min_freq_softlimit) << 14;
4296 } else {
4297 limits = dev_priv->rps.max_freq_softlimit << 24;
4298 if (val <= dev_priv->rps.min_freq_softlimit)
4299 limits |= dev_priv->rps.min_freq_softlimit << 16;
4300 }
20b46e59
DV
4301
4302 return limits;
4303}
4304
dd75fdc8
CW
4305static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
4306{
4307 int new_power;
8a586437
AG
4308 u32 threshold_up = 0, threshold_down = 0; /* in % */
4309 u32 ei_up = 0, ei_down = 0;
dd75fdc8
CW
4310
4311 new_power = dev_priv->rps.power;
4312 switch (dev_priv->rps.power) {
4313 case LOW_POWER:
b39fb297 4314 if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq)
dd75fdc8
CW
4315 new_power = BETWEEN;
4316 break;
4317
4318 case BETWEEN:
b39fb297 4319 if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq)
dd75fdc8 4320 new_power = LOW_POWER;
b39fb297 4321 else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq)
dd75fdc8
CW
4322 new_power = HIGH_POWER;
4323 break;
4324
4325 case HIGH_POWER:
b39fb297 4326 if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq)
dd75fdc8
CW
4327 new_power = BETWEEN;
4328 break;
4329 }
4330 /* Max/min bins are special */
aed242ff 4331 if (val <= dev_priv->rps.min_freq_softlimit)
dd75fdc8 4332 new_power = LOW_POWER;
aed242ff 4333 if (val >= dev_priv->rps.max_freq_softlimit)
dd75fdc8
CW
4334 new_power = HIGH_POWER;
4335 if (new_power == dev_priv->rps.power)
4336 return;
4337
4338 /* Note the units here are not exactly 1us, but 1280ns. */
4339 switch (new_power) {
4340 case LOW_POWER:
4341 /* Upclock if more than 95% busy over 16ms */
8a586437
AG
4342 ei_up = 16000;
4343 threshold_up = 95;
dd75fdc8
CW
4344
4345 /* Downclock if less than 85% busy over 32ms */
8a586437
AG
4346 ei_down = 32000;
4347 threshold_down = 85;
dd75fdc8
CW
4348 break;
4349
4350 case BETWEEN:
4351 /* Upclock if more than 90% busy over 13ms */
8a586437
AG
4352 ei_up = 13000;
4353 threshold_up = 90;
dd75fdc8
CW
4354
4355 /* Downclock if less than 75% busy over 32ms */
8a586437
AG
4356 ei_down = 32000;
4357 threshold_down = 75;
dd75fdc8
CW
4358 break;
4359
4360 case HIGH_POWER:
4361 /* Upclock if more than 85% busy over 10ms */
8a586437
AG
4362 ei_up = 10000;
4363 threshold_up = 85;
dd75fdc8
CW
4364
4365 /* Downclock if less than 60% busy over 32ms */
8a586437
AG
4366 ei_down = 32000;
4367 threshold_down = 60;
dd75fdc8
CW
4368 break;
4369 }
4370
8a586437
AG
4371 I915_WRITE(GEN6_RP_UP_EI,
4372 GT_INTERVAL_FROM_US(dev_priv, ei_up));
4373 I915_WRITE(GEN6_RP_UP_THRESHOLD,
4374 GT_INTERVAL_FROM_US(dev_priv, (ei_up * threshold_up / 100)));
4375
4376 I915_WRITE(GEN6_RP_DOWN_EI,
4377 GT_INTERVAL_FROM_US(dev_priv, ei_down));
4378 I915_WRITE(GEN6_RP_DOWN_THRESHOLD,
4379 GT_INTERVAL_FROM_US(dev_priv, (ei_down * threshold_down / 100)));
4380
4381 I915_WRITE(GEN6_RP_CONTROL,
4382 GEN6_RP_MEDIA_TURBO |
4383 GEN6_RP_MEDIA_HW_NORMAL_MODE |
4384 GEN6_RP_MEDIA_IS_GFX |
4385 GEN6_RP_ENABLE |
4386 GEN6_RP_UP_BUSY_AVG |
4387 GEN6_RP_DOWN_IDLE_AVG);
4388
dd75fdc8 4389 dev_priv->rps.power = new_power;
8fb55197
CW
4390 dev_priv->rps.up_threshold = threshold_up;
4391 dev_priv->rps.down_threshold = threshold_down;
dd75fdc8
CW
4392 dev_priv->rps.last_adj = 0;
4393}
4394
2876ce73
CW
4395static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val)
4396{
4397 u32 mask = 0;
4398
4399 if (val > dev_priv->rps.min_freq_softlimit)
6f4b12f8 4400 mask |= GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT;
2876ce73 4401 if (val < dev_priv->rps.max_freq_softlimit)
6f4b12f8 4402 mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD;
2876ce73 4403
7b3c29f6
CW
4404 mask &= dev_priv->pm_rps_events;
4405
59d02a1f 4406 return gen6_sanitize_rps_pm_mask(dev_priv, ~mask);
2876ce73
CW
4407}
4408
b8a5ff8d
JM
4409/* gen6_set_rps is called to update the frequency request, but should also be
4410 * called when the range (min_delay and max_delay) is modified so that we can
4411 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
ffe02b40 4412static void gen6_set_rps(struct drm_device *dev, u8 val)
20b46e59
DV
4413{
4414 struct drm_i915_private *dev_priv = dev->dev_private;
7b9e0ae6 4415
23eafea6
SAK
4416 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
4417 if (IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0))
4418 return;
4419
4fc688ce 4420 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
aed242ff
CW
4421 WARN_ON(val > dev_priv->rps.max_freq);
4422 WARN_ON(val < dev_priv->rps.min_freq);
004777cb 4423
eb64cad1
CW
4424 /* min/max delay may still have been modified so be sure to
4425 * write the limits value.
4426 */
4427 if (val != dev_priv->rps.cur_freq) {
4428 gen6_set_rps_thresholds(dev_priv, val);
b8a5ff8d 4429
5704195c
AG
4430 if (IS_GEN9(dev))
4431 I915_WRITE(GEN6_RPNSWREQ,
4432 GEN9_FREQUENCY(val));
4433 else if (IS_HASWELL(dev) || IS_BROADWELL(dev))
eb64cad1
CW
4434 I915_WRITE(GEN6_RPNSWREQ,
4435 HSW_FREQUENCY(val));
4436 else
4437 I915_WRITE(GEN6_RPNSWREQ,
4438 GEN6_FREQUENCY(val) |
4439 GEN6_OFFSET(0) |
4440 GEN6_AGGRESSIVE_TURBO);
b8a5ff8d 4441 }
7b9e0ae6 4442
7b9e0ae6
CW
4443 /* Make sure we continue to get interrupts
4444 * until we hit the minimum or maximum frequencies.
4445 */
74ef1173 4446 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val));
2876ce73 4447 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
7b9e0ae6 4448
d5570a72
BW
4449 POSTING_READ(GEN6_RPNSWREQ);
4450
b39fb297 4451 dev_priv->rps.cur_freq = val;
be2cde9a 4452 trace_intel_gpu_freq_change(val * 50);
2b4e57bd
ED
4453}
4454
ffe02b40
VS
4455static void valleyview_set_rps(struct drm_device *dev, u8 val)
4456{
4457 struct drm_i915_private *dev_priv = dev->dev_private;
4458
4459 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
aed242ff
CW
4460 WARN_ON(val > dev_priv->rps.max_freq);
4461 WARN_ON(val < dev_priv->rps.min_freq);
ffe02b40
VS
4462
4463 if (WARN_ONCE(IS_CHERRYVIEW(dev) && (val & 1),
4464 "Odd GPU freq value\n"))
4465 val &= ~1;
4466
cd25dd5b
D
4467 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
4468
8fb55197 4469 if (val != dev_priv->rps.cur_freq) {
ffe02b40 4470 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
8fb55197
CW
4471 if (!IS_CHERRYVIEW(dev_priv))
4472 gen6_set_rps_thresholds(dev_priv, val);
4473 }
ffe02b40 4474
ffe02b40
VS
4475 dev_priv->rps.cur_freq = val;
4476 trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val));
4477}
4478
a7f6e231 4479/* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down
76c3552f
D
4480 *
4481 * * If Gfx is Idle, then
a7f6e231
D
4482 * 1. Forcewake Media well.
4483 * 2. Request idle freq.
4484 * 3. Release Forcewake of Media well.
76c3552f
D
4485*/
4486static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
4487{
aed242ff 4488 u32 val = dev_priv->rps.idle_freq;
5549d25f 4489
aed242ff 4490 if (dev_priv->rps.cur_freq <= val)
76c3552f
D
4491 return;
4492
a7f6e231
D
4493 /* Wake up the media well, as that takes a lot less
4494 * power than the Render well. */
4495 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA);
4496 valleyview_set_rps(dev_priv->dev, val);
4497 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA);
76c3552f
D
4498}
4499
43cf3bf0
CW
4500void gen6_rps_busy(struct drm_i915_private *dev_priv)
4501{
4502 mutex_lock(&dev_priv->rps.hw_lock);
4503 if (dev_priv->rps.enabled) {
4504 if (dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED))
4505 gen6_rps_reset_ei(dev_priv);
4506 I915_WRITE(GEN6_PMINTRMSK,
4507 gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq));
4508 }
4509 mutex_unlock(&dev_priv->rps.hw_lock);
4510}
4511
b29c19b6
CW
4512void gen6_rps_idle(struct drm_i915_private *dev_priv)
4513{
691bb717
DL
4514 struct drm_device *dev = dev_priv->dev;
4515
b29c19b6 4516 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 4517 if (dev_priv->rps.enabled) {
21a11fff 4518 if (IS_VALLEYVIEW(dev))
76c3552f 4519 vlv_set_rps_idle(dev_priv);
7526ed79 4520 else
aed242ff 4521 gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq);
c0951f0c 4522 dev_priv->rps.last_adj = 0;
43cf3bf0 4523 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
c0951f0c 4524 }
8d3afd7d 4525 mutex_unlock(&dev_priv->rps.hw_lock);
1854d5ca 4526
8d3afd7d 4527 spin_lock(&dev_priv->rps.client_lock);
1854d5ca
CW
4528 while (!list_empty(&dev_priv->rps.clients))
4529 list_del_init(dev_priv->rps.clients.next);
8d3afd7d 4530 spin_unlock(&dev_priv->rps.client_lock);
b29c19b6
CW
4531}
4532
1854d5ca 4533void gen6_rps_boost(struct drm_i915_private *dev_priv,
e61b9958
CW
4534 struct intel_rps_client *rps,
4535 unsigned long submitted)
b29c19b6 4536{
8d3afd7d
CW
4537 /* This is intentionally racy! We peek at the state here, then
4538 * validate inside the RPS worker.
4539 */
4540 if (!(dev_priv->mm.busy &&
4541 dev_priv->rps.enabled &&
4542 dev_priv->rps.cur_freq < dev_priv->rps.max_freq_softlimit))
4543 return;
43cf3bf0 4544
e61b9958
CW
4545 /* Force a RPS boost (and don't count it against the client) if
4546 * the GPU is severely congested.
4547 */
d0bc54f2 4548 if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES))
e61b9958
CW
4549 rps = NULL;
4550
8d3afd7d
CW
4551 spin_lock(&dev_priv->rps.client_lock);
4552 if (rps == NULL || list_empty(&rps->link)) {
4553 spin_lock_irq(&dev_priv->irq_lock);
4554 if (dev_priv->rps.interrupts_enabled) {
4555 dev_priv->rps.client_boost = true;
4556 queue_work(dev_priv->wq, &dev_priv->rps.work);
4557 }
4558 spin_unlock_irq(&dev_priv->irq_lock);
1854d5ca 4559
2e1b8730
CW
4560 if (rps != NULL) {
4561 list_add(&rps->link, &dev_priv->rps.clients);
4562 rps->boosts++;
1854d5ca
CW
4563 } else
4564 dev_priv->rps.boosts++;
c0951f0c 4565 }
8d3afd7d 4566 spin_unlock(&dev_priv->rps.client_lock);
b29c19b6
CW
4567}
4568
ffe02b40 4569void intel_set_rps(struct drm_device *dev, u8 val)
0a073b84 4570{
ffe02b40
VS
4571 if (IS_VALLEYVIEW(dev))
4572 valleyview_set_rps(dev, val);
4573 else
4574 gen6_set_rps(dev, val);
0a073b84
JB
4575}
4576
20e49366
ZW
4577static void gen9_disable_rps(struct drm_device *dev)
4578{
4579 struct drm_i915_private *dev_priv = dev->dev_private;
4580
4581 I915_WRITE(GEN6_RC_CONTROL, 0);
38c23527 4582 I915_WRITE(GEN9_PG_ENABLE, 0);
20e49366
ZW
4583}
4584
44fc7d5c 4585static void gen6_disable_rps(struct drm_device *dev)
d20d4f0c
JB
4586{
4587 struct drm_i915_private *dev_priv = dev->dev_private;
4588
4589 I915_WRITE(GEN6_RC_CONTROL, 0);
44fc7d5c 4590 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
44fc7d5c
DV
4591}
4592
38807746
D
4593static void cherryview_disable_rps(struct drm_device *dev)
4594{
4595 struct drm_i915_private *dev_priv = dev->dev_private;
4596
4597 I915_WRITE(GEN6_RC_CONTROL, 0);
4598}
4599
44fc7d5c
DV
4600static void valleyview_disable_rps(struct drm_device *dev)
4601{
4602 struct drm_i915_private *dev_priv = dev->dev_private;
4603
98a2e5f9
D
4604 /* we're doing forcewake before Disabling RC6,
4605 * This what the BIOS expects when going into suspend */
59bad947 4606 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
98a2e5f9 4607
44fc7d5c 4608 I915_WRITE(GEN6_RC_CONTROL, 0);
d20d4f0c 4609
59bad947 4610 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
d20d4f0c
JB
4611}
4612
dc39fff7
BW
4613static void intel_print_rc6_info(struct drm_device *dev, u32 mode)
4614{
91ca689a
ID
4615 if (IS_VALLEYVIEW(dev)) {
4616 if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1)))
4617 mode = GEN6_RC_CTL_RC6_ENABLE;
4618 else
4619 mode = 0;
4620 }
58abf1da
RV
4621 if (HAS_RC6p(dev))
4622 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n",
4623 (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
4624 (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
4625 (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
4626
4627 else
4628 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n",
4629 (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off");
dc39fff7
BW
4630}
4631
e6069ca8 4632static int sanitize_rc6_option(const struct drm_device *dev, int enable_rc6)
2b4e57bd 4633{
e7d66d89
DV
4634 /* No RC6 before Ironlake and code is gone for ilk. */
4635 if (INTEL_INFO(dev)->gen < 6)
e6069ca8
ID
4636 return 0;
4637
456470eb 4638 /* Respect the kernel parameter if it is set */
e6069ca8
ID
4639 if (enable_rc6 >= 0) {
4640 int mask;
4641
58abf1da 4642 if (HAS_RC6p(dev))
e6069ca8
ID
4643 mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE |
4644 INTEL_RC6pp_ENABLE;
4645 else
4646 mask = INTEL_RC6_ENABLE;
4647
4648 if ((enable_rc6 & mask) != enable_rc6)
8dfd1f04
DV
4649 DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
4650 enable_rc6 & mask, enable_rc6, mask);
e6069ca8
ID
4651
4652 return enable_rc6 & mask;
4653 }
2b4e57bd 4654
8bade1ad 4655 if (IS_IVYBRIDGE(dev))
cca84a1f 4656 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
8bade1ad
BW
4657
4658 return INTEL_RC6_ENABLE;
2b4e57bd
ED
4659}
4660
e6069ca8
ID
4661int intel_enable_rc6(const struct drm_device *dev)
4662{
4663 return i915.enable_rc6;
4664}
4665
93ee2920 4666static void gen6_init_rps_frequencies(struct drm_device *dev)
3280e8b0 4667{
93ee2920
TR
4668 struct drm_i915_private *dev_priv = dev->dev_private;
4669 uint32_t rp_state_cap;
4670 u32 ddcc_status = 0;
4671 int ret;
4672
3280e8b0
BW
4673 /* All of these values are in units of 50MHz */
4674 dev_priv->rps.cur_freq = 0;
93ee2920 4675 /* static values from HW: RP0 > RP1 > RPn (min_freq) */
35040562
BP
4676 if (IS_BROXTON(dev)) {
4677 rp_state_cap = I915_READ(BXT_RP_STATE_CAP);
4678 dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff;
4679 dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff;
4680 dev_priv->rps.min_freq = (rp_state_cap >> 0) & 0xff;
4681 } else {
4682 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
4683 dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff;
4684 dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff;
4685 dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff;
4686 }
4687
3280e8b0
BW
4688 /* hw_max = RP0 until we check for overclocking */
4689 dev_priv->rps.max_freq = dev_priv->rps.rp0_freq;
4690
93ee2920 4691 dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq;
c5e0688c 4692 if (IS_HASWELL(dev) || IS_BROADWELL(dev) || IS_SKYLAKE(dev)) {
93ee2920
TR
4693 ret = sandybridge_pcode_read(dev_priv,
4694 HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL,
4695 &ddcc_status);
4696 if (0 == ret)
4697 dev_priv->rps.efficient_freq =
46efa4ab
TR
4698 clamp_t(u8,
4699 ((ddcc_status >> 8) & 0xff),
4700 dev_priv->rps.min_freq,
4701 dev_priv->rps.max_freq);
93ee2920
TR
4702 }
4703
c5e0688c
AG
4704 if (IS_SKYLAKE(dev)) {
4705 /* Store the frequency values in 16.66 MHZ units, which is
4706 the natural hardware unit for SKL */
4707 dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER;
4708 dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER;
4709 dev_priv->rps.min_freq *= GEN9_FREQ_SCALER;
4710 dev_priv->rps.max_freq *= GEN9_FREQ_SCALER;
4711 dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER;
4712 }
4713
aed242ff
CW
4714 dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
4715
3280e8b0
BW
4716 /* Preserve min/max settings in case of re-init */
4717 if (dev_priv->rps.max_freq_softlimit == 0)
4718 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
4719
93ee2920
TR
4720 if (dev_priv->rps.min_freq_softlimit == 0) {
4721 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
4722 dev_priv->rps.min_freq_softlimit =
813b5e69
VS
4723 max_t(int, dev_priv->rps.efficient_freq,
4724 intel_freq_opcode(dev_priv, 450));
93ee2920
TR
4725 else
4726 dev_priv->rps.min_freq_softlimit =
4727 dev_priv->rps.min_freq;
4728 }
3280e8b0
BW
4729}
4730
b6fef0ef 4731/* See the Gen9_GT_PM_Programming_Guide doc for the below */
20e49366 4732static void gen9_enable_rps(struct drm_device *dev)
b6fef0ef
JB
4733{
4734 struct drm_i915_private *dev_priv = dev->dev_private;
4735
4736 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
4737
ba1c554c
DL
4738 gen6_init_rps_frequencies(dev);
4739
23eafea6
SAK
4740 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
4741 if (IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) {
4742 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
4743 return;
4744 }
4745
0beb059a
AG
4746 /* Program defaults and thresholds for RPS*/
4747 I915_WRITE(GEN6_RC_VIDEO_FREQ,
4748 GEN9_FREQUENCY(dev_priv->rps.rp1_freq));
4749
4750 /* 1 second timeout*/
4751 I915_WRITE(GEN6_RP_DOWN_TIMEOUT,
4752 GT_INTERVAL_FROM_US(dev_priv, 1000000));
4753
b6fef0ef 4754 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa);
b6fef0ef 4755
0beb059a
AG
4756 /* Leaning on the below call to gen6_set_rps to program/setup the
4757 * Up/Down EI & threshold registers, as well as the RP_CONTROL,
4758 * RP_INTERRUPT_LIMITS & RPNSWREQ registers */
4759 dev_priv->rps.power = HIGH_POWER; /* force a reset */
4760 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
b6fef0ef
JB
4761
4762 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
4763}
4764
4765static void gen9_enable_rc6(struct drm_device *dev)
20e49366
ZW
4766{
4767 struct drm_i915_private *dev_priv = dev->dev_private;
4768 struct intel_engine_cs *ring;
4769 uint32_t rc6_mask = 0;
4770 int unused;
4771
4772 /* 1a: Software RC state - RC0 */
4773 I915_WRITE(GEN6_RC_STATE, 0);
4774
4775 /* 1b: Get forcewake during program sequence. Although the driver
4776 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
59bad947 4777 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
20e49366
ZW
4778
4779 /* 2a: Disable RC states. */
4780 I915_WRITE(GEN6_RC_CONTROL, 0);
4781
4782 /* 2b: Program RC6 thresholds.*/
63a4dec2
SAK
4783
4784 /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */
4785 if (IS_SKYLAKE(dev) && !((IS_SKL_GT3(dev) || IS_SKL_GT4(dev)) &&
4786 (INTEL_REVID(dev) <= SKL_REVID_E0)))
4787 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16);
4788 else
4789 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16);
20e49366
ZW
4790 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
4791 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
4792 for_each_ring(ring, dev_priv, unused)
4793 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
97c322e7
SAK
4794
4795 if (HAS_GUC_UCODE(dev))
4796 I915_WRITE(GUC_MAX_IDLE_COUNT, 0xA);
4797
20e49366 4798 I915_WRITE(GEN6_RC_SLEEP, 0);
20e49366 4799
38c23527
ZW
4800 /* 2c: Program Coarse Power Gating Policies. */
4801 I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25);
4802 I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25);
4803
20e49366
ZW
4804 /* 3a: Enable RC6 */
4805 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
4806 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
4807 DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ?
4808 "on" : "off");
3e7732a0 4809 /* WaRsUseTimeoutMode */
e3429cd2 4810 if ((IS_SKYLAKE(dev) && INTEL_REVID(dev) <= SKL_REVID_D0) ||
3e7732a0
SAK
4811 (IS_BROXTON(dev) && INTEL_REVID(dev) <= BXT_REVID_A0)) {
4812 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us */
e3429cd2
SAK
4813 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
4814 GEN7_RC_CTL_TO_MODE |
4815 rc6_mask);
3e7732a0
SAK
4816 } else {
4817 I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */
e3429cd2
SAK
4818 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
4819 GEN6_RC_CTL_EI_MODE(1) |
4820 rc6_mask);
3e7732a0 4821 }
20e49366 4822
cb07bae0
SK
4823 /*
4824 * 3b: Enable Coarse Power Gating only when RC6 is enabled.
f2d2fe95 4825 * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6.
cb07bae0 4826 */
f2d2fe95
SAK
4827 if ((IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) ||
4828 ((IS_SKL_GT3(dev) || IS_SKL_GT4(dev)) && (INTEL_REVID(dev) <= SKL_REVID_E0)))
4829 I915_WRITE(GEN9_PG_ENABLE, 0);
4830 else
4831 I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ?
4832 (GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE) : 0);
38c23527 4833
59bad947 4834 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
20e49366
ZW
4835
4836}
4837
6edee7f3
BW
4838static void gen8_enable_rps(struct drm_device *dev)
4839{
4840 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 4841 struct intel_engine_cs *ring;
93ee2920 4842 uint32_t rc6_mask = 0;
6edee7f3
BW
4843 int unused;
4844
4845 /* 1a: Software RC state - RC0 */
4846 I915_WRITE(GEN6_RC_STATE, 0);
4847
4848 /* 1c & 1d: Get forcewake during program sequence. Although the driver
4849 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
59bad947 4850 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW
4851
4852 /* 2a: Disable RC states. */
4853 I915_WRITE(GEN6_RC_CONTROL, 0);
4854
93ee2920
TR
4855 /* Initialize rps frequencies */
4856 gen6_init_rps_frequencies(dev);
6edee7f3
BW
4857
4858 /* 2b: Program RC6 thresholds.*/
4859 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
4860 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
4861 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
4862 for_each_ring(ring, dev_priv, unused)
4863 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
4864 I915_WRITE(GEN6_RC_SLEEP, 0);
0d68b25e
TR
4865 if (IS_BROADWELL(dev))
4866 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */
4867 else
4868 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
6edee7f3
BW
4869
4870 /* 3: Enable RC6 */
4871 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
4872 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
abbf9d2c 4873 intel_print_rc6_info(dev, rc6_mask);
0d68b25e
TR
4874 if (IS_BROADWELL(dev))
4875 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
4876 GEN7_RC_CTL_TO_MODE |
4877 rc6_mask);
4878 else
4879 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
4880 GEN6_RC_CTL_EI_MODE(1) |
4881 rc6_mask);
6edee7f3
BW
4882
4883 /* 4 Program defaults and thresholds for RPS*/
f9bdc585
BW
4884 I915_WRITE(GEN6_RPNSWREQ,
4885 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
4886 I915_WRITE(GEN6_RC_VIDEO_FREQ,
4887 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
7526ed79
DV
4888 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
4889 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
4890
4891 /* Docs recommend 900MHz, and 300 MHz respectively */
4892 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
4893 dev_priv->rps.max_freq_softlimit << 24 |
4894 dev_priv->rps.min_freq_softlimit << 16);
4895
4896 I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
4897 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
4898 I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
4899 I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
4900
4901 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
6edee7f3
BW
4902
4903 /* 5: Enable RPS */
7526ed79
DV
4904 I915_WRITE(GEN6_RP_CONTROL,
4905 GEN6_RP_MEDIA_TURBO |
4906 GEN6_RP_MEDIA_HW_NORMAL_MODE |
4907 GEN6_RP_MEDIA_IS_GFX |
4908 GEN6_RP_ENABLE |
4909 GEN6_RP_UP_BUSY_AVG |
4910 GEN6_RP_DOWN_IDLE_AVG);
4911
4912 /* 6: Ring frequency + overclocking (our driver does this later */
4913
c7f3153a 4914 dev_priv->rps.power = HIGH_POWER; /* force a reset */
aed242ff 4915 gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq);
7526ed79 4916
59bad947 4917 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW
4918}
4919
79f5b2c7 4920static void gen6_enable_rps(struct drm_device *dev)
2b4e57bd 4921{
79f5b2c7 4922 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 4923 struct intel_engine_cs *ring;
d060c169 4924 u32 rc6vids, pcu_mbox = 0, rc6_mask = 0;
2b4e57bd 4925 u32 gtfifodbg;
2b4e57bd 4926 int rc6_mode;
42c0526c 4927 int i, ret;
2b4e57bd 4928
4fc688ce 4929 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 4930
2b4e57bd
ED
4931 /* Here begins a magic sequence of register writes to enable
4932 * auto-downclocking.
4933 *
4934 * Perhaps there might be some value in exposing these to
4935 * userspace...
4936 */
4937 I915_WRITE(GEN6_RC_STATE, 0);
2b4e57bd
ED
4938
4939 /* Clear the DBG now so we don't confuse earlier errors */
4940 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
4941 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
4942 I915_WRITE(GTFIFODBG, gtfifodbg);
4943 }
4944
59bad947 4945 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
2b4e57bd 4946
93ee2920
TR
4947 /* Initialize rps frequencies */
4948 gen6_init_rps_frequencies(dev);
dd0a1aa1 4949
2b4e57bd
ED
4950 /* disable the counters and set deterministic thresholds */
4951 I915_WRITE(GEN6_RC_CONTROL, 0);
4952
4953 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
4954 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
4955 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
4956 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
4957 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
4958
b4519513
CW
4959 for_each_ring(ring, dev_priv, i)
4960 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2b4e57bd
ED
4961
4962 I915_WRITE(GEN6_RC_SLEEP, 0);
4963 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
29c78f60 4964 if (IS_IVYBRIDGE(dev))
351aa566
SM
4965 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
4966 else
4967 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
0920a487 4968 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2b4e57bd
ED
4969 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
4970
5a7dc92a 4971 /* Check if we are enabling RC6 */
2b4e57bd
ED
4972 rc6_mode = intel_enable_rc6(dev_priv->dev);
4973 if (rc6_mode & INTEL_RC6_ENABLE)
4974 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
4975
5a7dc92a
ED
4976 /* We don't use those on Haswell */
4977 if (!IS_HASWELL(dev)) {
4978 if (rc6_mode & INTEL_RC6p_ENABLE)
4979 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2b4e57bd 4980
5a7dc92a
ED
4981 if (rc6_mode & INTEL_RC6pp_ENABLE)
4982 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
4983 }
2b4e57bd 4984
dc39fff7 4985 intel_print_rc6_info(dev, rc6_mask);
2b4e57bd
ED
4986
4987 I915_WRITE(GEN6_RC_CONTROL,
4988 rc6_mask |
4989 GEN6_RC_CTL_EI_MODE(1) |
4990 GEN6_RC_CTL_HW_ENABLE);
4991
dd75fdc8
CW
4992 /* Power down if completely idle for over 50ms */
4993 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
2b4e57bd 4994 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2b4e57bd 4995
42c0526c 4996 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
d060c169 4997 if (ret)
42c0526c 4998 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
d060c169
BW
4999
5000 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
5001 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
5002 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
b39fb297 5003 (dev_priv->rps.max_freq_softlimit & 0xff) * 50,
d060c169 5004 (pcu_mbox & 0xff) * 50);
b39fb297 5005 dev_priv->rps.max_freq = pcu_mbox & 0xff;
2b4e57bd
ED
5006 }
5007
dd75fdc8 5008 dev_priv->rps.power = HIGH_POWER; /* force a reset */
aed242ff 5009 gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq);
2b4e57bd 5010
31643d54
BW
5011 rc6vids = 0;
5012 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
5013 if (IS_GEN6(dev) && ret) {
5014 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
5015 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
5016 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
5017 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
5018 rc6vids &= 0xffff00;
5019 rc6vids |= GEN6_ENCODE_RC6_VID(450);
5020 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
5021 if (ret)
5022 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
5023 }
5024
59bad947 5025 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
2b4e57bd
ED
5026}
5027
c2bc2fc5 5028static void __gen6_update_ring_freq(struct drm_device *dev)
2b4e57bd 5029{
79f5b2c7 5030 struct drm_i915_private *dev_priv = dev->dev_private;
2b4e57bd 5031 int min_freq = 15;
3ebecd07
CW
5032 unsigned int gpu_freq;
5033 unsigned int max_ia_freq, min_ring_freq;
4c8c7743 5034 unsigned int max_gpu_freq, min_gpu_freq;
2b4e57bd 5035 int scaling_factor = 180;
eda79642 5036 struct cpufreq_policy *policy;
2b4e57bd 5037
4fc688ce 5038 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 5039
eda79642
BW
5040 policy = cpufreq_cpu_get(0);
5041 if (policy) {
5042 max_ia_freq = policy->cpuinfo.max_freq;
5043 cpufreq_cpu_put(policy);
5044 } else {
5045 /*
5046 * Default to measured freq if none found, PCU will ensure we
5047 * don't go over
5048 */
2b4e57bd 5049 max_ia_freq = tsc_khz;
eda79642 5050 }
2b4e57bd
ED
5051
5052 /* Convert from kHz to MHz */
5053 max_ia_freq /= 1000;
5054
153b4b95 5055 min_ring_freq = I915_READ(DCLK) & 0xf;
f6aca45c
BW
5056 /* convert DDR frequency from units of 266.6MHz to bandwidth */
5057 min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3ebecd07 5058
4c8c7743
AG
5059 if (IS_SKYLAKE(dev)) {
5060 /* Convert GT frequency to 50 HZ units */
5061 min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER;
5062 max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER;
5063 } else {
5064 min_gpu_freq = dev_priv->rps.min_freq;
5065 max_gpu_freq = dev_priv->rps.max_freq;
5066 }
5067
2b4e57bd
ED
5068 /*
5069 * For each potential GPU frequency, load a ring frequency we'd like
5070 * to use for memory access. We do this by specifying the IA frequency
5071 * the PCU should use as a reference to determine the ring frequency.
5072 */
4c8c7743
AG
5073 for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) {
5074 int diff = max_gpu_freq - gpu_freq;
3ebecd07
CW
5075 unsigned int ia_freq = 0, ring_freq = 0;
5076
4c8c7743
AG
5077 if (IS_SKYLAKE(dev)) {
5078 /*
5079 * ring_freq = 2 * GT. ring_freq is in 100MHz units
5080 * No floor required for ring frequency on SKL.
5081 */
5082 ring_freq = gpu_freq;
5083 } else if (INTEL_INFO(dev)->gen >= 8) {
46c764d4
BW
5084 /* max(2 * GT, DDR). NB: GT is 50MHz units */
5085 ring_freq = max(min_ring_freq, gpu_freq);
5086 } else if (IS_HASWELL(dev)) {
f6aca45c 5087 ring_freq = mult_frac(gpu_freq, 5, 4);
3ebecd07
CW
5088 ring_freq = max(min_ring_freq, ring_freq);
5089 /* leave ia_freq as the default, chosen by cpufreq */
5090 } else {
5091 /* On older processors, there is no separate ring
5092 * clock domain, so in order to boost the bandwidth
5093 * of the ring, we need to upclock the CPU (ia_freq).
5094 *
5095 * For GPU frequencies less than 750MHz,
5096 * just use the lowest ring freq.
5097 */
5098 if (gpu_freq < min_freq)
5099 ia_freq = 800;
5100 else
5101 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
5102 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
5103 }
2b4e57bd 5104
42c0526c
BW
5105 sandybridge_pcode_write(dev_priv,
5106 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3ebecd07
CW
5107 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
5108 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
5109 gpu_freq);
2b4e57bd 5110 }
2b4e57bd
ED
5111}
5112
c2bc2fc5
ID
5113void gen6_update_ring_freq(struct drm_device *dev)
5114{
5115 struct drm_i915_private *dev_priv = dev->dev_private;
5116
97d3308a 5117 if (!HAS_CORE_RING_FREQ(dev))
c2bc2fc5
ID
5118 return;
5119
5120 mutex_lock(&dev_priv->rps.hw_lock);
5121 __gen6_update_ring_freq(dev);
5122 mutex_unlock(&dev_priv->rps.hw_lock);
5123}
5124
03af2045 5125static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv)
2b6b3a09 5126{
095acd5f 5127 struct drm_device *dev = dev_priv->dev;
2b6b3a09
D
5128 u32 val, rp0;
5129
5b5929cb 5130 val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE);
2b6b3a09 5131
5b5929cb
JN
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;
095acd5f 5147 }
5b5929cb
JN
5148
5149 rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK);
5150
2b6b3a09
D
5151 return rp0;
5152}
5153
5154static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv)
5155{
5156 u32 val, rpe;
5157
5158 val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG);
5159 rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK;
5160
5161 return rpe;
5162}
5163
7707df4a
D
5164static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv)
5165{
5166 u32 val, rp1;
5167
5b5929cb
JN
5168 val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE);
5169 rp1 = (val & FB_GFX_FREQ_FUSE_MASK);
5170
7707df4a
D
5171 return rp1;
5172}
5173
f8f2b001
D
5174static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv)
5175{
5176 u32 val, rp1;
5177
5178 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
5179
5180 rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT;
5181
5182 return rp1;
5183}
5184
03af2045 5185static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
0a073b84
JB
5186{
5187 u32 val, rp0;
5188
64936258 5189 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
0a073b84
JB
5190
5191 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
5192 /* Clamp to max */
5193 rp0 = min_t(u32, rp0, 0xea);
5194
5195 return rp0;
5196}
5197
5198static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
5199{
5200 u32 val, rpe;
5201
64936258 5202 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
0a073b84 5203 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
64936258 5204 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
0a073b84
JB
5205 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
5206
5207 return rpe;
5208}
5209
03af2045 5210static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
0a073b84 5211{
64936258 5212 return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
0a073b84
JB
5213}
5214
ae48434c
ID
5215/* Check that the pctx buffer wasn't move under us. */
5216static void valleyview_check_pctx(struct drm_i915_private *dev_priv)
5217{
5218 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
5219
5220 WARN_ON(pctx_addr != dev_priv->mm.stolen_base +
5221 dev_priv->vlv_pctx->stolen->start);
5222}
5223
38807746
D
5224
5225/* Check that the pcbr address is not empty. */
5226static void cherryview_check_pctx(struct drm_i915_private *dev_priv)
5227{
5228 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
5229
5230 WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0);
5231}
5232
5233static void cherryview_setup_pctx(struct drm_device *dev)
5234{
5235 struct drm_i915_private *dev_priv = dev->dev_private;
5236 unsigned long pctx_paddr, paddr;
5237 struct i915_gtt *gtt = &dev_priv->gtt;
5238 u32 pcbr;
5239 int pctx_size = 32*1024;
5240
5241 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
5242
5243 pcbr = I915_READ(VLV_PCBR);
5244 if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) {
ce611ef8 5245 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
38807746
D
5246 paddr = (dev_priv->mm.stolen_base +
5247 (gtt->stolen_size - pctx_size));
5248
5249 pctx_paddr = (paddr & (~4095));
5250 I915_WRITE(VLV_PCBR, pctx_paddr);
5251 }
ce611ef8
VS
5252
5253 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR));
38807746
D
5254}
5255
c9cddffc
JB
5256static void valleyview_setup_pctx(struct drm_device *dev)
5257{
5258 struct drm_i915_private *dev_priv = dev->dev_private;
5259 struct drm_i915_gem_object *pctx;
5260 unsigned long pctx_paddr;
5261 u32 pcbr;
5262 int pctx_size = 24*1024;
5263
17b0c1f7
ID
5264 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
5265
c9cddffc
JB
5266 pcbr = I915_READ(VLV_PCBR);
5267 if (pcbr) {
5268 /* BIOS set it up already, grab the pre-alloc'd space */
5269 int pcbr_offset;
5270
5271 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
5272 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
5273 pcbr_offset,
190d6cd5 5274 I915_GTT_OFFSET_NONE,
c9cddffc
JB
5275 pctx_size);
5276 goto out;
5277 }
5278
ce611ef8
VS
5279 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
5280
c9cddffc
JB
5281 /*
5282 * From the Gunit register HAS:
5283 * The Gfx driver is expected to program this register and ensure
5284 * proper allocation within Gfx stolen memory. For example, this
5285 * register should be programmed such than the PCBR range does not
5286 * overlap with other ranges, such as the frame buffer, protected
5287 * memory, or any other relevant ranges.
5288 */
5289 pctx = i915_gem_object_create_stolen(dev, pctx_size);
5290 if (!pctx) {
5291 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
5292 return;
5293 }
5294
5295 pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
5296 I915_WRITE(VLV_PCBR, pctx_paddr);
5297
5298out:
ce611ef8 5299 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR));
c9cddffc
JB
5300 dev_priv->vlv_pctx = pctx;
5301}
5302
ae48434c
ID
5303static void valleyview_cleanup_pctx(struct drm_device *dev)
5304{
5305 struct drm_i915_private *dev_priv = dev->dev_private;
5306
5307 if (WARN_ON(!dev_priv->vlv_pctx))
5308 return;
5309
5310 drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
5311 dev_priv->vlv_pctx = NULL;
5312}
5313
4e80519e
ID
5314static void valleyview_init_gt_powersave(struct drm_device *dev)
5315{
5316 struct drm_i915_private *dev_priv = dev->dev_private;
2bb25c17 5317 u32 val;
4e80519e
ID
5318
5319 valleyview_setup_pctx(dev);
5320
5321 mutex_lock(&dev_priv->rps.hw_lock);
5322
2bb25c17
VS
5323 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5324 switch ((val >> 6) & 3) {
5325 case 0:
5326 case 1:
5327 dev_priv->mem_freq = 800;
5328 break;
5329 case 2:
5330 dev_priv->mem_freq = 1066;
5331 break;
5332 case 3:
5333 dev_priv->mem_freq = 1333;
5334 break;
5335 }
80b83b62 5336 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq);
2bb25c17 5337
4e80519e
ID
5338 dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv);
5339 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
5340 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
7c59a9c1 5341 intel_gpu_freq(dev_priv, dev_priv->rps.max_freq),
4e80519e
ID
5342 dev_priv->rps.max_freq);
5343
5344 dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv);
5345 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
7c59a9c1 5346 intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
4e80519e
ID
5347 dev_priv->rps.efficient_freq);
5348
f8f2b001
D
5349 dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv);
5350 DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n",
7c59a9c1 5351 intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq),
f8f2b001
D
5352 dev_priv->rps.rp1_freq);
5353
4e80519e
ID
5354 dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv);
5355 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
7c59a9c1 5356 intel_gpu_freq(dev_priv, dev_priv->rps.min_freq),
4e80519e
ID
5357 dev_priv->rps.min_freq);
5358
aed242ff
CW
5359 dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
5360
4e80519e
ID
5361 /* Preserve min/max settings in case of re-init */
5362 if (dev_priv->rps.max_freq_softlimit == 0)
5363 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
5364
5365 if (dev_priv->rps.min_freq_softlimit == 0)
5366 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
5367
5368 mutex_unlock(&dev_priv->rps.hw_lock);
5369}
5370
38807746
D
5371static void cherryview_init_gt_powersave(struct drm_device *dev)
5372{
2b6b3a09 5373 struct drm_i915_private *dev_priv = dev->dev_private;
2bb25c17 5374 u32 val;
2b6b3a09 5375
38807746 5376 cherryview_setup_pctx(dev);
2b6b3a09
D
5377
5378 mutex_lock(&dev_priv->rps.hw_lock);
5379
a580516d 5380 mutex_lock(&dev_priv->sb_lock);
c6e8f39d 5381 val = vlv_cck_read(dev_priv, CCK_FUSE_REG);
a580516d 5382 mutex_unlock(&dev_priv->sb_lock);
c6e8f39d 5383
2bb25c17 5384 switch ((val >> 2) & 0x7) {
2bb25c17 5385 case 3:
2bb25c17
VS
5386 dev_priv->mem_freq = 2000;
5387 break;
bfa7df01 5388 default:
2bb25c17
VS
5389 dev_priv->mem_freq = 1600;
5390 break;
5391 }
80b83b62 5392 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq);
2bb25c17 5393
2b6b3a09
D
5394 dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv);
5395 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
5396 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
7c59a9c1 5397 intel_gpu_freq(dev_priv, dev_priv->rps.max_freq),
2b6b3a09
D
5398 dev_priv->rps.max_freq);
5399
5400 dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv);
5401 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
7c59a9c1 5402 intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
2b6b3a09
D
5403 dev_priv->rps.efficient_freq);
5404
7707df4a
D
5405 dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv);
5406 DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n",
7c59a9c1 5407 intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq),
7707df4a
D
5408 dev_priv->rps.rp1_freq);
5409
5b7c91b7
D
5410 /* PUnit validated range is only [RPe, RP0] */
5411 dev_priv->rps.min_freq = dev_priv->rps.efficient_freq;
2b6b3a09 5412 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
7c59a9c1 5413 intel_gpu_freq(dev_priv, dev_priv->rps.min_freq),
2b6b3a09
D
5414 dev_priv->rps.min_freq);
5415
1c14762d
VS
5416 WARN_ONCE((dev_priv->rps.max_freq |
5417 dev_priv->rps.efficient_freq |
5418 dev_priv->rps.rp1_freq |
5419 dev_priv->rps.min_freq) & 1,
5420 "Odd GPU freq values\n");
5421
aed242ff
CW
5422 dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
5423
2b6b3a09
D
5424 /* Preserve min/max settings in case of re-init */
5425 if (dev_priv->rps.max_freq_softlimit == 0)
5426 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
5427
5428 if (dev_priv->rps.min_freq_softlimit == 0)
5429 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
5430
5431 mutex_unlock(&dev_priv->rps.hw_lock);
38807746
D
5432}
5433
4e80519e
ID
5434static void valleyview_cleanup_gt_powersave(struct drm_device *dev)
5435{
5436 valleyview_cleanup_pctx(dev);
5437}
5438
38807746
D
5439static void cherryview_enable_rps(struct drm_device *dev)
5440{
5441 struct drm_i915_private *dev_priv = dev->dev_private;
5442 struct intel_engine_cs *ring;
2b6b3a09 5443 u32 gtfifodbg, val, rc6_mode = 0, pcbr;
38807746
D
5444 int i;
5445
5446 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
5447
5448 gtfifodbg = I915_READ(GTFIFODBG);
5449 if (gtfifodbg) {
5450 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5451 gtfifodbg);
5452 I915_WRITE(GTFIFODBG, gtfifodbg);
5453 }
5454
5455 cherryview_check_pctx(dev_priv);
5456
5457 /* 1a & 1b: Get forcewake during program sequence. Although the driver
5458 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
59bad947 5459 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
38807746 5460
160614a2
VS
5461 /* Disable RC states. */
5462 I915_WRITE(GEN6_RC_CONTROL, 0);
5463
38807746
D
5464 /* 2a: Program RC6 thresholds.*/
5465 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
5466 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
5467 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
5468
5469 for_each_ring(ring, dev_priv, i)
5470 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
5471 I915_WRITE(GEN6_RC_SLEEP, 0);
5472
f4f71c7d
D
5473 /* TO threshold set to 500 us ( 0x186 * 1.28 us) */
5474 I915_WRITE(GEN6_RC6_THRESHOLD, 0x186);
38807746
D
5475
5476 /* allows RC6 residency counter to work */
5477 I915_WRITE(VLV_COUNTER_CONTROL,
5478 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
5479 VLV_MEDIA_RC6_COUNT_EN |
5480 VLV_RENDER_RC6_COUNT_EN));
5481
5482 /* For now we assume BIOS is allocating and populating the PCBR */
5483 pcbr = I915_READ(VLV_PCBR);
5484
38807746
D
5485 /* 3: Enable RC6 */
5486 if ((intel_enable_rc6(dev) & INTEL_RC6_ENABLE) &&
5487 (pcbr >> VLV_PCBR_ADDR_SHIFT))
af5a75a3 5488 rc6_mode = GEN7_RC_CTL_TO_MODE;
38807746
D
5489
5490 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
5491
2b6b3a09 5492 /* 4 Program defaults and thresholds for RPS*/
3cbdb48f 5493 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
2b6b3a09
D
5494 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
5495 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
5496 I915_WRITE(GEN6_RP_UP_EI, 66000);
5497 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
5498
5499 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
5500
5501 /* 5: Enable RPS */
5502 I915_WRITE(GEN6_RP_CONTROL,
5503 GEN6_RP_MEDIA_HW_NORMAL_MODE |
eb973a5e 5504 GEN6_RP_MEDIA_IS_GFX |
2b6b3a09
D
5505 GEN6_RP_ENABLE |
5506 GEN6_RP_UP_BUSY_AVG |
5507 GEN6_RP_DOWN_IDLE_AVG);
5508
3ef62342
D
5509 /* Setting Fixed Bias */
5510 val = VLV_OVERRIDE_EN |
5511 VLV_SOC_TDP_EN |
5512 CHV_BIAS_CPU_50_SOC_50;
5513 vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val);
5514
2b6b3a09
D
5515 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5516
8d40c3ae
VS
5517 /* RPS code assumes GPLL is used */
5518 WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n");
5519
742f491d 5520 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
2b6b3a09
D
5521 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
5522
5523 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
5524 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
7c59a9c1 5525 intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
2b6b3a09
D
5526 dev_priv->rps.cur_freq);
5527
5528 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
7c59a9c1 5529 intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
2b6b3a09
D
5530 dev_priv->rps.efficient_freq);
5531
5532 valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
5533
59bad947 5534 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
38807746
D
5535}
5536
0a073b84
JB
5537static void valleyview_enable_rps(struct drm_device *dev)
5538{
5539 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 5540 struct intel_engine_cs *ring;
2a5913a8 5541 u32 gtfifodbg, val, rc6_mode = 0;
0a073b84
JB
5542 int i;
5543
5544 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
5545
ae48434c
ID
5546 valleyview_check_pctx(dev_priv);
5547
0a073b84 5548 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
f7d85c1e
JB
5549 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5550 gtfifodbg);
0a073b84
JB
5551 I915_WRITE(GTFIFODBG, gtfifodbg);
5552 }
5553
c8d9a590 5554 /* If VLV, Forcewake all wells, else re-direct to regular path */
59bad947 5555 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
0a073b84 5556
160614a2
VS
5557 /* Disable RC states. */
5558 I915_WRITE(GEN6_RC_CONTROL, 0);
5559
cad725fe 5560 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
0a073b84
JB
5561 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
5562 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
5563 I915_WRITE(GEN6_RP_UP_EI, 66000);
5564 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
5565
5566 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
5567
5568 I915_WRITE(GEN6_RP_CONTROL,
5569 GEN6_RP_MEDIA_TURBO |
5570 GEN6_RP_MEDIA_HW_NORMAL_MODE |
5571 GEN6_RP_MEDIA_IS_GFX |
5572 GEN6_RP_ENABLE |
5573 GEN6_RP_UP_BUSY_AVG |
5574 GEN6_RP_DOWN_IDLE_CONT);
5575
5576 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
5577 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
5578 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
5579
5580 for_each_ring(ring, dev_priv, i)
5581 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
5582
2f0aa304 5583 I915_WRITE(GEN6_RC6_THRESHOLD, 0x557);
0a073b84
JB
5584
5585 /* allows RC6 residency counter to work */
49798eb2 5586 I915_WRITE(VLV_COUNTER_CONTROL,
31685c25
D
5587 _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN |
5588 VLV_RENDER_RC0_COUNT_EN |
49798eb2
JB
5589 VLV_MEDIA_RC6_COUNT_EN |
5590 VLV_RENDER_RC6_COUNT_EN));
31685c25 5591
a2b23fe0 5592 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
6b88f295 5593 rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
dc39fff7
BW
5594
5595 intel_print_rc6_info(dev, rc6_mode);
5596
a2b23fe0 5597 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
0a073b84 5598
3ef62342
D
5599 /* Setting Fixed Bias */
5600 val = VLV_OVERRIDE_EN |
5601 VLV_SOC_TDP_EN |
5602 VLV_BIAS_CPU_125_SOC_875;
5603 vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val);
5604
64936258 5605 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
0a073b84 5606
8d40c3ae
VS
5607 /* RPS code assumes GPLL is used */
5608 WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n");
5609
742f491d 5610 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
0a073b84
JB
5611 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
5612
b39fb297 5613 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
73008b98 5614 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
7c59a9c1 5615 intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
b39fb297 5616 dev_priv->rps.cur_freq);
0a073b84 5617
73008b98 5618 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
7c59a9c1 5619 intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
b39fb297 5620 dev_priv->rps.efficient_freq);
0a073b84 5621
b39fb297 5622 valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
0a073b84 5623
59bad947 5624 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
0a073b84
JB
5625}
5626
dde18883
ED
5627static unsigned long intel_pxfreq(u32 vidfreq)
5628{
5629 unsigned long freq;
5630 int div = (vidfreq & 0x3f0000) >> 16;
5631 int post = (vidfreq & 0x3000) >> 12;
5632 int pre = (vidfreq & 0x7);
5633
5634 if (!pre)
5635 return 0;
5636
5637 freq = ((div * 133333) / ((1<<post) * pre));
5638
5639 return freq;
5640}
5641
eb48eb00
DV
5642static const struct cparams {
5643 u16 i;
5644 u16 t;
5645 u16 m;
5646 u16 c;
5647} cparams[] = {
5648 { 1, 1333, 301, 28664 },
5649 { 1, 1066, 294, 24460 },
5650 { 1, 800, 294, 25192 },
5651 { 0, 1333, 276, 27605 },
5652 { 0, 1066, 276, 27605 },
5653 { 0, 800, 231, 23784 },
5654};
5655
f531dcb2 5656static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
eb48eb00
DV
5657{
5658 u64 total_count, diff, ret;
5659 u32 count1, count2, count3, m = 0, c = 0;
5660 unsigned long now = jiffies_to_msecs(jiffies), diff1;
5661 int i;
5662
02d71956
DV
5663 assert_spin_locked(&mchdev_lock);
5664
20e4d407 5665 diff1 = now - dev_priv->ips.last_time1;
eb48eb00
DV
5666
5667 /* Prevent division-by-zero if we are asking too fast.
5668 * Also, we don't get interesting results if we are polling
5669 * faster than once in 10ms, so just return the saved value
5670 * in such cases.
5671 */
5672 if (diff1 <= 10)
20e4d407 5673 return dev_priv->ips.chipset_power;
eb48eb00
DV
5674
5675 count1 = I915_READ(DMIEC);
5676 count2 = I915_READ(DDREC);
5677 count3 = I915_READ(CSIEC);
5678
5679 total_count = count1 + count2 + count3;
5680
5681 /* FIXME: handle per-counter overflow */
20e4d407
DV
5682 if (total_count < dev_priv->ips.last_count1) {
5683 diff = ~0UL - dev_priv->ips.last_count1;
eb48eb00
DV
5684 diff += total_count;
5685 } else {
20e4d407 5686 diff = total_count - dev_priv->ips.last_count1;
eb48eb00
DV
5687 }
5688
5689 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
20e4d407
DV
5690 if (cparams[i].i == dev_priv->ips.c_m &&
5691 cparams[i].t == dev_priv->ips.r_t) {
eb48eb00
DV
5692 m = cparams[i].m;
5693 c = cparams[i].c;
5694 break;
5695 }
5696 }
5697
5698 diff = div_u64(diff, diff1);
5699 ret = ((m * diff) + c);
5700 ret = div_u64(ret, 10);
5701
20e4d407
DV
5702 dev_priv->ips.last_count1 = total_count;
5703 dev_priv->ips.last_time1 = now;
eb48eb00 5704
20e4d407 5705 dev_priv->ips.chipset_power = ret;
eb48eb00
DV
5706
5707 return ret;
5708}
5709
f531dcb2
CW
5710unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
5711{
3d13ef2e 5712 struct drm_device *dev = dev_priv->dev;
f531dcb2
CW
5713 unsigned long val;
5714
3d13ef2e 5715 if (INTEL_INFO(dev)->gen != 5)
f531dcb2
CW
5716 return 0;
5717
5718 spin_lock_irq(&mchdev_lock);
5719
5720 val = __i915_chipset_val(dev_priv);
5721
5722 spin_unlock_irq(&mchdev_lock);
5723
5724 return val;
5725}
5726
eb48eb00
DV
5727unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
5728{
5729 unsigned long m, x, b;
5730 u32 tsfs;
5731
5732 tsfs = I915_READ(TSFS);
5733
5734 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
5735 x = I915_READ8(TR1);
5736
5737 b = tsfs & TSFS_INTR_MASK;
5738
5739 return ((m * x) / 127) - b;
5740}
5741
d972d6ee
MK
5742static int _pxvid_to_vd(u8 pxvid)
5743{
5744 if (pxvid == 0)
5745 return 0;
5746
5747 if (pxvid >= 8 && pxvid < 31)
5748 pxvid = 31;
5749
5750 return (pxvid + 2) * 125;
5751}
5752
5753static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
eb48eb00 5754{
3d13ef2e 5755 struct drm_device *dev = dev_priv->dev;
d972d6ee
MK
5756 const int vd = _pxvid_to_vd(pxvid);
5757 const int vm = vd - 1125;
5758
3d13ef2e 5759 if (INTEL_INFO(dev)->is_mobile)
d972d6ee
MK
5760 return vm > 0 ? vm : 0;
5761
5762 return vd;
eb48eb00
DV
5763}
5764
02d71956 5765static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00 5766{
5ed0bdf2 5767 u64 now, diff, diffms;
eb48eb00
DV
5768 u32 count;
5769
02d71956 5770 assert_spin_locked(&mchdev_lock);
eb48eb00 5771
5ed0bdf2
TG
5772 now = ktime_get_raw_ns();
5773 diffms = now - dev_priv->ips.last_time2;
5774 do_div(diffms, NSEC_PER_MSEC);
eb48eb00
DV
5775
5776 /* Don't divide by 0 */
eb48eb00
DV
5777 if (!diffms)
5778 return;
5779
5780 count = I915_READ(GFXEC);
5781
20e4d407
DV
5782 if (count < dev_priv->ips.last_count2) {
5783 diff = ~0UL - dev_priv->ips.last_count2;
eb48eb00
DV
5784 diff += count;
5785 } else {
20e4d407 5786 diff = count - dev_priv->ips.last_count2;
eb48eb00
DV
5787 }
5788
20e4d407
DV
5789 dev_priv->ips.last_count2 = count;
5790 dev_priv->ips.last_time2 = now;
eb48eb00
DV
5791
5792 /* More magic constants... */
5793 diff = diff * 1181;
5794 diff = div_u64(diff, diffms * 10);
20e4d407 5795 dev_priv->ips.gfx_power = diff;
eb48eb00
DV
5796}
5797
02d71956
DV
5798void i915_update_gfx_val(struct drm_i915_private *dev_priv)
5799{
3d13ef2e
DL
5800 struct drm_device *dev = dev_priv->dev;
5801
5802 if (INTEL_INFO(dev)->gen != 5)
02d71956
DV
5803 return;
5804
9270388e 5805 spin_lock_irq(&mchdev_lock);
02d71956
DV
5806
5807 __i915_update_gfx_val(dev_priv);
5808
9270388e 5809 spin_unlock_irq(&mchdev_lock);
02d71956
DV
5810}
5811
f531dcb2 5812static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV
5813{
5814 unsigned long t, corr, state1, corr2, state2;
5815 u32 pxvid, ext_v;
5816
02d71956
DV
5817 assert_spin_locked(&mchdev_lock);
5818
616847e7 5819 pxvid = I915_READ(PXVFREQ(dev_priv->rps.cur_freq));
eb48eb00
DV
5820 pxvid = (pxvid >> 24) & 0x7f;
5821 ext_v = pvid_to_extvid(dev_priv, pxvid);
5822
5823 state1 = ext_v;
5824
5825 t = i915_mch_val(dev_priv);
5826
5827 /* Revel in the empirically derived constants */
5828
5829 /* Correction factor in 1/100000 units */
5830 if (t > 80)
5831 corr = ((t * 2349) + 135940);
5832 else if (t >= 50)
5833 corr = ((t * 964) + 29317);
5834 else /* < 50 */
5835 corr = ((t * 301) + 1004);
5836
5837 corr = corr * ((150142 * state1) / 10000 - 78642);
5838 corr /= 100000;
20e4d407 5839 corr2 = (corr * dev_priv->ips.corr);
eb48eb00
DV
5840
5841 state2 = (corr2 * state1) / 10000;
5842 state2 /= 100; /* convert to mW */
5843
02d71956 5844 __i915_update_gfx_val(dev_priv);
eb48eb00 5845
20e4d407 5846 return dev_priv->ips.gfx_power + state2;
eb48eb00
DV
5847}
5848
f531dcb2
CW
5849unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
5850{
3d13ef2e 5851 struct drm_device *dev = dev_priv->dev;
f531dcb2
CW
5852 unsigned long val;
5853
3d13ef2e 5854 if (INTEL_INFO(dev)->gen != 5)
f531dcb2
CW
5855 return 0;
5856
5857 spin_lock_irq(&mchdev_lock);
5858
5859 val = __i915_gfx_val(dev_priv);
5860
5861 spin_unlock_irq(&mchdev_lock);
5862
5863 return val;
5864}
5865
eb48eb00
DV
5866/**
5867 * i915_read_mch_val - return value for IPS use
5868 *
5869 * Calculate and return a value for the IPS driver to use when deciding whether
5870 * we have thermal and power headroom to increase CPU or GPU power budget.
5871 */
5872unsigned long i915_read_mch_val(void)
5873{
5874 struct drm_i915_private *dev_priv;
5875 unsigned long chipset_val, graphics_val, ret = 0;
5876
9270388e 5877 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
5878 if (!i915_mch_dev)
5879 goto out_unlock;
5880 dev_priv = i915_mch_dev;
5881
f531dcb2
CW
5882 chipset_val = __i915_chipset_val(dev_priv);
5883 graphics_val = __i915_gfx_val(dev_priv);
eb48eb00
DV
5884
5885 ret = chipset_val + graphics_val;
5886
5887out_unlock:
9270388e 5888 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
5889
5890 return ret;
5891}
5892EXPORT_SYMBOL_GPL(i915_read_mch_val);
5893
5894/**
5895 * i915_gpu_raise - raise GPU frequency limit
5896 *
5897 * Raise the limit; IPS indicates we have thermal headroom.
5898 */
5899bool i915_gpu_raise(void)
5900{
5901 struct drm_i915_private *dev_priv;
5902 bool ret = true;
5903
9270388e 5904 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
5905 if (!i915_mch_dev) {
5906 ret = false;
5907 goto out_unlock;
5908 }
5909 dev_priv = i915_mch_dev;
5910
20e4d407
DV
5911 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
5912 dev_priv->ips.max_delay--;
eb48eb00
DV
5913
5914out_unlock:
9270388e 5915 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
5916
5917 return ret;
5918}
5919EXPORT_SYMBOL_GPL(i915_gpu_raise);
5920
5921/**
5922 * i915_gpu_lower - lower GPU frequency limit
5923 *
5924 * IPS indicates we're close to a thermal limit, so throttle back the GPU
5925 * frequency maximum.
5926 */
5927bool i915_gpu_lower(void)
5928{
5929 struct drm_i915_private *dev_priv;
5930 bool ret = true;
5931
9270388e 5932 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
5933 if (!i915_mch_dev) {
5934 ret = false;
5935 goto out_unlock;
5936 }
5937 dev_priv = i915_mch_dev;
5938
20e4d407
DV
5939 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
5940 dev_priv->ips.max_delay++;
eb48eb00
DV
5941
5942out_unlock:
9270388e 5943 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
5944
5945 return ret;
5946}
5947EXPORT_SYMBOL_GPL(i915_gpu_lower);
5948
5949/**
5950 * i915_gpu_busy - indicate GPU business to IPS
5951 *
5952 * Tell the IPS driver whether or not the GPU is busy.
5953 */
5954bool i915_gpu_busy(void)
5955{
5956 struct drm_i915_private *dev_priv;
a4872ba6 5957 struct intel_engine_cs *ring;
eb48eb00 5958 bool ret = false;
f047e395 5959 int i;
eb48eb00 5960
9270388e 5961 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
5962 if (!i915_mch_dev)
5963 goto out_unlock;
5964 dev_priv = i915_mch_dev;
5965
f047e395
CW
5966 for_each_ring(ring, dev_priv, i)
5967 ret |= !list_empty(&ring->request_list);
eb48eb00
DV
5968
5969out_unlock:
9270388e 5970 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
5971
5972 return ret;
5973}
5974EXPORT_SYMBOL_GPL(i915_gpu_busy);
5975
5976/**
5977 * i915_gpu_turbo_disable - disable graphics turbo
5978 *
5979 * Disable graphics turbo by resetting the max frequency and setting the
5980 * current frequency to the default.
5981 */
5982bool i915_gpu_turbo_disable(void)
5983{
5984 struct drm_i915_private *dev_priv;
5985 bool ret = true;
5986
9270388e 5987 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
5988 if (!i915_mch_dev) {
5989 ret = false;
5990 goto out_unlock;
5991 }
5992 dev_priv = i915_mch_dev;
5993
20e4d407 5994 dev_priv->ips.max_delay = dev_priv->ips.fstart;
eb48eb00 5995
20e4d407 5996 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
eb48eb00
DV
5997 ret = false;
5998
5999out_unlock:
9270388e 6000 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
6001
6002 return ret;
6003}
6004EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
6005
6006/**
6007 * Tells the intel_ips driver that the i915 driver is now loaded, if
6008 * IPS got loaded first.
6009 *
6010 * This awkward dance is so that neither module has to depend on the
6011 * other in order for IPS to do the appropriate communication of
6012 * GPU turbo limits to i915.
6013 */
6014static void
6015ips_ping_for_i915_load(void)
6016{
6017 void (*link)(void);
6018
6019 link = symbol_get(ips_link_to_i915_driver);
6020 if (link) {
6021 link();
6022 symbol_put(ips_link_to_i915_driver);
6023 }
6024}
6025
6026void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
6027{
02d71956
DV
6028 /* We only register the i915 ips part with intel-ips once everything is
6029 * set up, to avoid intel-ips sneaking in and reading bogus values. */
9270388e 6030 spin_lock_irq(&mchdev_lock);
eb48eb00 6031 i915_mch_dev = dev_priv;
9270388e 6032 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
6033
6034 ips_ping_for_i915_load();
6035}
6036
6037void intel_gpu_ips_teardown(void)
6038{
9270388e 6039 spin_lock_irq(&mchdev_lock);
eb48eb00 6040 i915_mch_dev = NULL;
9270388e 6041 spin_unlock_irq(&mchdev_lock);
eb48eb00 6042}
76c3552f 6043
8090c6b9 6044static void intel_init_emon(struct drm_device *dev)
dde18883
ED
6045{
6046 struct drm_i915_private *dev_priv = dev->dev_private;
6047 u32 lcfuse;
6048 u8 pxw[16];
6049 int i;
6050
6051 /* Disable to program */
6052 I915_WRITE(ECR, 0);
6053 POSTING_READ(ECR);
6054
6055 /* Program energy weights for various events */
6056 I915_WRITE(SDEW, 0x15040d00);
6057 I915_WRITE(CSIEW0, 0x007f0000);
6058 I915_WRITE(CSIEW1, 0x1e220004);
6059 I915_WRITE(CSIEW2, 0x04000004);
6060
6061 for (i = 0; i < 5; i++)
616847e7 6062 I915_WRITE(PEW(i), 0);
dde18883 6063 for (i = 0; i < 3; i++)
616847e7 6064 I915_WRITE(DEW(i), 0);
dde18883
ED
6065
6066 /* Program P-state weights to account for frequency power adjustment */
6067 for (i = 0; i < 16; i++) {
616847e7 6068 u32 pxvidfreq = I915_READ(PXVFREQ(i));
dde18883
ED
6069 unsigned long freq = intel_pxfreq(pxvidfreq);
6070 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
6071 PXVFREQ_PX_SHIFT;
6072 unsigned long val;
6073
6074 val = vid * vid;
6075 val *= (freq / 1000);
6076 val *= 255;
6077 val /= (127*127*900);
6078 if (val > 0xff)
6079 DRM_ERROR("bad pxval: %ld\n", val);
6080 pxw[i] = val;
6081 }
6082 /* Render standby states get 0 weight */
6083 pxw[14] = 0;
6084 pxw[15] = 0;
6085
6086 for (i = 0; i < 4; i++) {
6087 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
6088 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
616847e7 6089 I915_WRITE(PXW(i), val);
dde18883
ED
6090 }
6091
6092 /* Adjust magic regs to magic values (more experimental results) */
6093 I915_WRITE(OGW0, 0);
6094 I915_WRITE(OGW1, 0);
6095 I915_WRITE(EG0, 0x00007f00);
6096 I915_WRITE(EG1, 0x0000000e);
6097 I915_WRITE(EG2, 0x000e0000);
6098 I915_WRITE(EG3, 0x68000300);
6099 I915_WRITE(EG4, 0x42000000);
6100 I915_WRITE(EG5, 0x00140031);
6101 I915_WRITE(EG6, 0);
6102 I915_WRITE(EG7, 0);
6103
6104 for (i = 0; i < 8; i++)
616847e7 6105 I915_WRITE(PXWL(i), 0);
dde18883
ED
6106
6107 /* Enable PMON + select events */
6108 I915_WRITE(ECR, 0x80000019);
6109
6110 lcfuse = I915_READ(LCFUSE02);
6111
20e4d407 6112 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
dde18883
ED
6113}
6114
ae48434c
ID
6115void intel_init_gt_powersave(struct drm_device *dev)
6116{
e6069ca8
ID
6117 i915.enable_rc6 = sanitize_rc6_option(dev, i915.enable_rc6);
6118
38807746
D
6119 if (IS_CHERRYVIEW(dev))
6120 cherryview_init_gt_powersave(dev);
6121 else if (IS_VALLEYVIEW(dev))
4e80519e 6122 valleyview_init_gt_powersave(dev);
ae48434c
ID
6123}
6124
6125void intel_cleanup_gt_powersave(struct drm_device *dev)
6126{
38807746
D
6127 if (IS_CHERRYVIEW(dev))
6128 return;
6129 else if (IS_VALLEYVIEW(dev))
4e80519e 6130 valleyview_cleanup_gt_powersave(dev);
ae48434c
ID
6131}
6132
dbea3cea
ID
6133static void gen6_suspend_rps(struct drm_device *dev)
6134{
6135 struct drm_i915_private *dev_priv = dev->dev_private;
6136
6137 flush_delayed_work(&dev_priv->rps.delayed_resume_work);
6138
4c2a8897 6139 gen6_disable_rps_interrupts(dev);
dbea3cea
ID
6140}
6141
156c7ca0
JB
6142/**
6143 * intel_suspend_gt_powersave - suspend PM work and helper threads
6144 * @dev: drm device
6145 *
6146 * We don't want to disable RC6 or other features here, we just want
6147 * to make sure any work we've queued has finished and won't bother
6148 * us while we're suspended.
6149 */
6150void intel_suspend_gt_powersave(struct drm_device *dev)
6151{
6152 struct drm_i915_private *dev_priv = dev->dev_private;
6153
d4d70aa5
ID
6154 if (INTEL_INFO(dev)->gen < 6)
6155 return;
6156
dbea3cea 6157 gen6_suspend_rps(dev);
b47adc17
D
6158
6159 /* Force GPU to min freq during suspend */
6160 gen6_rps_idle(dev_priv);
156c7ca0
JB
6161}
6162
8090c6b9
DV
6163void intel_disable_gt_powersave(struct drm_device *dev)
6164{
1a01ab3b
JB
6165 struct drm_i915_private *dev_priv = dev->dev_private;
6166
930ebb46 6167 if (IS_IRONLAKE_M(dev)) {
8090c6b9 6168 ironlake_disable_drps(dev);
38807746 6169 } else if (INTEL_INFO(dev)->gen >= 6) {
10d8d366 6170 intel_suspend_gt_powersave(dev);
e494837a 6171
4fc688ce 6172 mutex_lock(&dev_priv->rps.hw_lock);
20e49366
ZW
6173 if (INTEL_INFO(dev)->gen >= 9)
6174 gen9_disable_rps(dev);
6175 else if (IS_CHERRYVIEW(dev))
38807746
D
6176 cherryview_disable_rps(dev);
6177 else if (IS_VALLEYVIEW(dev))
d20d4f0c
JB
6178 valleyview_disable_rps(dev);
6179 else
6180 gen6_disable_rps(dev);
e534770a 6181
c0951f0c 6182 dev_priv->rps.enabled = false;
4fc688ce 6183 mutex_unlock(&dev_priv->rps.hw_lock);
930ebb46 6184 }
8090c6b9
DV
6185}
6186
1a01ab3b
JB
6187static void intel_gen6_powersave_work(struct work_struct *work)
6188{
6189 struct drm_i915_private *dev_priv =
6190 container_of(work, struct drm_i915_private,
6191 rps.delayed_resume_work.work);
6192 struct drm_device *dev = dev_priv->dev;
6193
4fc688ce 6194 mutex_lock(&dev_priv->rps.hw_lock);
0a073b84 6195
4c2a8897 6196 gen6_reset_rps_interrupts(dev);
3cc134e3 6197
38807746
D
6198 if (IS_CHERRYVIEW(dev)) {
6199 cherryview_enable_rps(dev);
6200 } else if (IS_VALLEYVIEW(dev)) {
0a073b84 6201 valleyview_enable_rps(dev);
20e49366 6202 } else if (INTEL_INFO(dev)->gen >= 9) {
b6fef0ef 6203 gen9_enable_rc6(dev);
20e49366 6204 gen9_enable_rps(dev);
cc017fb4
AG
6205 if (IS_SKYLAKE(dev))
6206 __gen6_update_ring_freq(dev);
6edee7f3
BW
6207 } else if (IS_BROADWELL(dev)) {
6208 gen8_enable_rps(dev);
c2bc2fc5 6209 __gen6_update_ring_freq(dev);
0a073b84
JB
6210 } else {
6211 gen6_enable_rps(dev);
c2bc2fc5 6212 __gen6_update_ring_freq(dev);
0a073b84 6213 }
aed242ff
CW
6214
6215 WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq);
6216 WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq);
6217
6218 WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq);
6219 WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq);
6220
c0951f0c 6221 dev_priv->rps.enabled = true;
3cc134e3 6222
4c2a8897 6223 gen6_enable_rps_interrupts(dev);
3cc134e3 6224
4fc688ce 6225 mutex_unlock(&dev_priv->rps.hw_lock);
c6df39b5
ID
6226
6227 intel_runtime_pm_put(dev_priv);
1a01ab3b
JB
6228}
6229
8090c6b9
DV
6230void intel_enable_gt_powersave(struct drm_device *dev)
6231{
1a01ab3b
JB
6232 struct drm_i915_private *dev_priv = dev->dev_private;
6233
f61018b1
YZ
6234 /* Powersaving is controlled by the host when inside a VM */
6235 if (intel_vgpu_active(dev))
6236 return;
6237
8090c6b9 6238 if (IS_IRONLAKE_M(dev)) {
dc1d0136 6239 mutex_lock(&dev->struct_mutex);
8090c6b9 6240 ironlake_enable_drps(dev);
8090c6b9 6241 intel_init_emon(dev);
dc1d0136 6242 mutex_unlock(&dev->struct_mutex);
38807746 6243 } else if (INTEL_INFO(dev)->gen >= 6) {
1a01ab3b
JB
6244 /*
6245 * PCU communication is slow and this doesn't need to be
6246 * done at any specific time, so do this out of our fast path
6247 * to make resume and init faster.
c6df39b5
ID
6248 *
6249 * We depend on the HW RC6 power context save/restore
6250 * mechanism when entering D3 through runtime PM suspend. So
6251 * disable RPM until RPS/RC6 is properly setup. We can only
6252 * get here via the driver load/system resume/runtime resume
6253 * paths, so the _noresume version is enough (and in case of
6254 * runtime resume it's necessary).
1a01ab3b 6255 */
c6df39b5
ID
6256 if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
6257 round_jiffies_up_relative(HZ)))
6258 intel_runtime_pm_get_noresume(dev_priv);
8090c6b9
DV
6259 }
6260}
6261
c6df39b5
ID
6262void intel_reset_gt_powersave(struct drm_device *dev)
6263{
6264 struct drm_i915_private *dev_priv = dev->dev_private;
6265
dbea3cea
ID
6266 if (INTEL_INFO(dev)->gen < 6)
6267 return;
6268
6269 gen6_suspend_rps(dev);
c6df39b5 6270 dev_priv->rps.enabled = false;
c6df39b5
ID
6271}
6272
3107bd48
DV
6273static void ibx_init_clock_gating(struct drm_device *dev)
6274{
6275 struct drm_i915_private *dev_priv = dev->dev_private;
6276
6277 /*
6278 * On Ibex Peak and Cougar Point, we need to disable clock
6279 * gating for the panel power sequencer or it will fail to
6280 * start up when no ports are active.
6281 */
6282 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
6283}
6284
0e088b8f
VS
6285static void g4x_disable_trickle_feed(struct drm_device *dev)
6286{
6287 struct drm_i915_private *dev_priv = dev->dev_private;
b12ce1d8 6288 enum pipe pipe;
0e088b8f 6289
055e393f 6290 for_each_pipe(dev_priv, pipe) {
0e088b8f
VS
6291 I915_WRITE(DSPCNTR(pipe),
6292 I915_READ(DSPCNTR(pipe)) |
6293 DISPPLANE_TRICKLE_FEED_DISABLE);
b12ce1d8
VS
6294
6295 I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe)));
6296 POSTING_READ(DSPSURF(pipe));
0e088b8f
VS
6297 }
6298}
6299
017636cc
VS
6300static void ilk_init_lp_watermarks(struct drm_device *dev)
6301{
6302 struct drm_i915_private *dev_priv = dev->dev_private;
6303
6304 I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
6305 I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
6306 I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
6307
6308 /*
6309 * Don't touch WM1S_LP_EN here.
6310 * Doing so could cause underruns.
6311 */
6312}
6313
1fa61106 6314static void ironlake_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6315{
6316 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 6317 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 6318
f1e8fa56
DL
6319 /*
6320 * Required for FBC
6321 * WaFbcDisableDpfcClockGating:ilk
6322 */
4d47e4f5
DL
6323 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
6324 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
6325 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED
6326
6327 I915_WRITE(PCH_3DCGDIS0,
6328 MARIUNIT_CLOCK_GATE_DISABLE |
6329 SVSMUNIT_CLOCK_GATE_DISABLE);
6330 I915_WRITE(PCH_3DCGDIS1,
6331 VFMUNIT_CLOCK_GATE_DISABLE);
6332
6f1d69b0
ED
6333 /*
6334 * According to the spec the following bits should be set in
6335 * order to enable memory self-refresh
6336 * The bit 22/21 of 0x42004
6337 * The bit 5 of 0x42020
6338 * The bit 15 of 0x45000
6339 */
6340 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6341 (I915_READ(ILK_DISPLAY_CHICKEN2) |
6342 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4d47e4f5 6343 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED
6344 I915_WRITE(DISP_ARB_CTL,
6345 (I915_READ(DISP_ARB_CTL) |
6346 DISP_FBC_WM_DIS));
017636cc
VS
6347
6348 ilk_init_lp_watermarks(dev);
6f1d69b0
ED
6349
6350 /*
6351 * Based on the document from hardware guys the following bits
6352 * should be set unconditionally in order to enable FBC.
6353 * The bit 22 of 0x42000
6354 * The bit 22 of 0x42004
6355 * The bit 7,8,9 of 0x42020.
6356 */
6357 if (IS_IRONLAKE_M(dev)) {
4bb35334 6358 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6f1d69b0
ED
6359 I915_WRITE(ILK_DISPLAY_CHICKEN1,
6360 I915_READ(ILK_DISPLAY_CHICKEN1) |
6361 ILK_FBCQ_DIS);
6362 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6363 I915_READ(ILK_DISPLAY_CHICKEN2) |
6364 ILK_DPARB_GATE);
6f1d69b0
ED
6365 }
6366
4d47e4f5
DL
6367 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6368
6f1d69b0
ED
6369 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6370 I915_READ(ILK_DISPLAY_CHICKEN2) |
6371 ILK_ELPIN_409_SELECT);
6372 I915_WRITE(_3D_CHICKEN2,
6373 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
6374 _3D_CHICKEN2_WM_READ_PIPELINED);
4358a374 6375
ecdb4eb7 6376 /* WaDisableRenderCachePipelinedFlush:ilk */
4358a374
DV
6377 I915_WRITE(CACHE_MODE_0,
6378 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3107bd48 6379
4e04632e
AG
6380 /* WaDisable_RenderCache_OperationalFlush:ilk */
6381 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6382
0e088b8f 6383 g4x_disable_trickle_feed(dev);
bdad2b2f 6384
3107bd48
DV
6385 ibx_init_clock_gating(dev);
6386}
6387
6388static void cpt_init_clock_gating(struct drm_device *dev)
6389{
6390 struct drm_i915_private *dev_priv = dev->dev_private;
6391 int pipe;
3f704fa2 6392 uint32_t val;
3107bd48
DV
6393
6394 /*
6395 * On Ibex Peak and Cougar Point, we need to disable clock
6396 * gating for the panel power sequencer or it will fail to
6397 * start up when no ports are active.
6398 */
cd664078
JB
6399 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
6400 PCH_DPLUNIT_CLOCK_GATE_DISABLE |
6401 PCH_CPUNIT_CLOCK_GATE_DISABLE);
3107bd48
DV
6402 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
6403 DPLS_EDP_PPS_FIX_DIS);
335c07b7
TI
6404 /* The below fixes the weird display corruption, a few pixels shifted
6405 * downward, on (only) LVDS of some HP laptops with IVY.
6406 */
055e393f 6407 for_each_pipe(dev_priv, pipe) {
dc4bd2d1
PZ
6408 val = I915_READ(TRANS_CHICKEN2(pipe));
6409 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
6410 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
41aa3448 6411 if (dev_priv->vbt.fdi_rx_polarity_inverted)
3f704fa2 6412 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
dc4bd2d1
PZ
6413 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
6414 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
6415 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3f704fa2
PZ
6416 I915_WRITE(TRANS_CHICKEN2(pipe), val);
6417 }
3107bd48 6418 /* WADP0ClockGatingDisable */
055e393f 6419 for_each_pipe(dev_priv, pipe) {
3107bd48
DV
6420 I915_WRITE(TRANS_CHICKEN1(pipe),
6421 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
6422 }
6f1d69b0
ED
6423}
6424
1d7aaa0c
DV
6425static void gen6_check_mch_setup(struct drm_device *dev)
6426{
6427 struct drm_i915_private *dev_priv = dev->dev_private;
6428 uint32_t tmp;
6429
6430 tmp = I915_READ(MCH_SSKPD);
df662a28
DV
6431 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL)
6432 DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n",
6433 tmp);
1d7aaa0c
DV
6434}
6435
1fa61106 6436static void gen6_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6437{
6438 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 6439 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 6440
231e54f6 6441 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6f1d69b0
ED
6442
6443 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6444 I915_READ(ILK_DISPLAY_CHICKEN2) |
6445 ILK_ELPIN_409_SELECT);
6446
ecdb4eb7 6447 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4283908e
DV
6448 I915_WRITE(_3D_CHICKEN,
6449 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
6450
4e04632e
AG
6451 /* WaDisable_RenderCache_OperationalFlush:snb */
6452 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6453
8d85d272
VS
6454 /*
6455 * BSpec recoomends 8x4 when MSAA is used,
6456 * however in practice 16x4 seems fastest.
c5c98a58
VS
6457 *
6458 * Note that PS/WM thread counts depend on the WIZ hashing
6459 * disable bit, which we don't touch here, but it's good
6460 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
8d85d272
VS
6461 */
6462 I915_WRITE(GEN6_GT_MODE,
98533251 6463 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
8d85d272 6464
017636cc 6465 ilk_init_lp_watermarks(dev);
6f1d69b0 6466
6f1d69b0 6467 I915_WRITE(CACHE_MODE_0,
50743298 6468 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
6f1d69b0
ED
6469
6470 I915_WRITE(GEN6_UCGCTL1,
6471 I915_READ(GEN6_UCGCTL1) |
6472 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
6473 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
6474
6475 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
6476 * gating disable must be set. Failure to set it results in
6477 * flickering pixels due to Z write ordering failures after
6478 * some amount of runtime in the Mesa "fire" demo, and Unigine
6479 * Sanctuary and Tropics, and apparently anything else with
6480 * alpha test or pixel discard.
6481 *
6482 * According to the spec, bit 11 (RCCUNIT) must also be set,
6483 * but we didn't debug actual testcases to find it out.
0f846f81 6484 *
ef59318c
VS
6485 * WaDisableRCCUnitClockGating:snb
6486 * WaDisableRCPBUnitClockGating:snb
6f1d69b0
ED
6487 */
6488 I915_WRITE(GEN6_UCGCTL2,
6489 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
6490 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
6491
5eb146dd 6492 /* WaStripsFansDisableFastClipPerformanceFix:snb */
743b57d8
VS
6493 I915_WRITE(_3D_CHICKEN3,
6494 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL));
6f1d69b0 6495
e927ecde
VS
6496 /*
6497 * Bspec says:
6498 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
6499 * 3DSTATE_SF number of SF output attributes is more than 16."
6500 */
6501 I915_WRITE(_3D_CHICKEN3,
6502 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH));
6503
6f1d69b0
ED
6504 /*
6505 * According to the spec the following bits should be
6506 * set in order to enable memory self-refresh and fbc:
6507 * The bit21 and bit22 of 0x42000
6508 * The bit21 and bit22 of 0x42004
6509 * The bit5 and bit7 of 0x42020
6510 * The bit14 of 0x70180
6511 * The bit14 of 0x71180
4bb35334
DL
6512 *
6513 * WaFbcAsynchFlipDisableFbcQueue:snb
6f1d69b0
ED
6514 */
6515 I915_WRITE(ILK_DISPLAY_CHICKEN1,
6516 I915_READ(ILK_DISPLAY_CHICKEN1) |
6517 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
6518 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6519 I915_READ(ILK_DISPLAY_CHICKEN2) |
6520 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
231e54f6
DL
6521 I915_WRITE(ILK_DSPCLK_GATE_D,
6522 I915_READ(ILK_DSPCLK_GATE_D) |
6523 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
6524 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
6f1d69b0 6525
0e088b8f 6526 g4x_disable_trickle_feed(dev);
f8f2ac9a 6527
3107bd48 6528 cpt_init_clock_gating(dev);
1d7aaa0c
DV
6529
6530 gen6_check_mch_setup(dev);
6f1d69b0
ED
6531}
6532
6533static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
6534{
6535 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
6536
3aad9059 6537 /*
46680e0a 6538 * WaVSThreadDispatchOverride:ivb,vlv
3aad9059
VS
6539 *
6540 * This actually overrides the dispatch
6541 * mode for all thread types.
6542 */
6f1d69b0
ED
6543 reg &= ~GEN7_FF_SCHED_MASK;
6544 reg |= GEN7_FF_TS_SCHED_HW;
6545 reg |= GEN7_FF_VS_SCHED_HW;
6546 reg |= GEN7_FF_DS_SCHED_HW;
6547
6548 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
6549}
6550
17a303ec
PZ
6551static void lpt_init_clock_gating(struct drm_device *dev)
6552{
6553 struct drm_i915_private *dev_priv = dev->dev_private;
6554
6555 /*
6556 * TODO: this bit should only be enabled when really needed, then
6557 * disabled when not needed anymore in order to save power.
6558 */
c2699524 6559 if (HAS_PCH_LPT_LP(dev))
17a303ec
PZ
6560 I915_WRITE(SOUTH_DSPCLK_GATE_D,
6561 I915_READ(SOUTH_DSPCLK_GATE_D) |
6562 PCH_LP_PARTITION_LEVEL_DISABLE);
0a790cdb
PZ
6563
6564 /* WADPOClockGatingDisable:hsw */
36c0d0cf
VS
6565 I915_WRITE(TRANS_CHICKEN1(PIPE_A),
6566 I915_READ(TRANS_CHICKEN1(PIPE_A)) |
0a790cdb 6567 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
17a303ec
PZ
6568}
6569
7d708ee4
ID
6570static void lpt_suspend_hw(struct drm_device *dev)
6571{
6572 struct drm_i915_private *dev_priv = dev->dev_private;
6573
c2699524 6574 if (HAS_PCH_LPT_LP(dev)) {
7d708ee4
ID
6575 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
6576
6577 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
6578 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
6579 }
6580}
6581
47c2bd97 6582static void broadwell_init_clock_gating(struct drm_device *dev)
1020a5c2
BW
6583{
6584 struct drm_i915_private *dev_priv = dev->dev_private;
07d27e20 6585 enum pipe pipe;
4d487cff 6586 uint32_t misccpctl;
1020a5c2 6587
7ad0dbab 6588 ilk_init_lp_watermarks(dev);
50ed5fbd 6589
ab57fff1 6590 /* WaSwitchSolVfFArbitrationPriority:bdw */
50ed5fbd 6591 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
fe4ab3ce 6592
ab57fff1 6593 /* WaPsrDPAMaskVBlankInSRD:bdw */
fe4ab3ce
BW
6594 I915_WRITE(CHICKEN_PAR1_1,
6595 I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
6596
ab57fff1 6597 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
055e393f 6598 for_each_pipe(dev_priv, pipe) {
07d27e20 6599 I915_WRITE(CHICKEN_PIPESL_1(pipe),
c7c65622 6600 I915_READ(CHICKEN_PIPESL_1(pipe)) |
8f670bb1 6601 BDW_DPRS_MASK_VBLANK_SRD);
fe4ab3ce 6602 }
63801f21 6603
ab57fff1
BW
6604 /* WaVSRefCountFullforceMissDisable:bdw */
6605 /* WaDSRefCountFullforceMissDisable:bdw */
6606 I915_WRITE(GEN7_FF_THREAD_MODE,
6607 I915_READ(GEN7_FF_THREAD_MODE) &
6608 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
36075a4c 6609
295e8bb7
VS
6610 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
6611 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
4f1ca9e9
VS
6612
6613 /* WaDisableSDEUnitClockGating:bdw */
6614 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
6615 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
5d708680 6616
4d487cff
VS
6617 /*
6618 * WaProgramL3SqcReg1Default:bdw
6619 * WaTempDisableDOPClkGating:bdw
6620 */
6621 misccpctl = I915_READ(GEN7_MISCCPCTL);
6622 I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE);
6623 I915_WRITE(GEN8_L3SQCREG1, BDW_WA_L3SQCREG1_DEFAULT);
6624 I915_WRITE(GEN7_MISCCPCTL, misccpctl);
6625
6d50b065
VS
6626 /*
6627 * WaGttCachingOffByDefault:bdw
6628 * GTT cache may not work with big pages, so if those
6629 * are ever enabled GTT cache may need to be disabled.
6630 */
6631 I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL);
6632
89d6b2b8 6633 lpt_init_clock_gating(dev);
1020a5c2
BW
6634}
6635
cad2a2d7
ED
6636static void haswell_init_clock_gating(struct drm_device *dev)
6637{
6638 struct drm_i915_private *dev_priv = dev->dev_private;
cad2a2d7 6639
017636cc 6640 ilk_init_lp_watermarks(dev);
cad2a2d7 6641
f3fc4884
FJ
6642 /* L3 caching of data atomics doesn't work -- disable it. */
6643 I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
6644 I915_WRITE(HSW_ROW_CHICKEN3,
6645 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
6646
ecdb4eb7 6647 /* This is required by WaCatErrorRejectionIssue:hsw */
cad2a2d7
ED
6648 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
6649 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
6650 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
6651
e36ea7ff
VS
6652 /* WaVSRefCountFullforceMissDisable:hsw */
6653 I915_WRITE(GEN7_FF_THREAD_MODE,
6654 I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);
cad2a2d7 6655
4e04632e
AG
6656 /* WaDisable_RenderCache_OperationalFlush:hsw */
6657 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6658
fe27c606
CW
6659 /* enable HiZ Raw Stall Optimization */
6660 I915_WRITE(CACHE_MODE_0_GEN7,
6661 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
6662
ecdb4eb7 6663 /* WaDisable4x2SubspanOptimization:hsw */
cad2a2d7
ED
6664 I915_WRITE(CACHE_MODE_1,
6665 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
1544d9d5 6666
a12c4967
VS
6667 /*
6668 * BSpec recommends 8x4 when MSAA is used,
6669 * however in practice 16x4 seems fastest.
c5c98a58
VS
6670 *
6671 * Note that PS/WM thread counts depend on the WIZ hashing
6672 * disable bit, which we don't touch here, but it's good
6673 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a12c4967
VS
6674 */
6675 I915_WRITE(GEN7_GT_MODE,
98533251 6676 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
a12c4967 6677
94411593
KG
6678 /* WaSampleCChickenBitEnable:hsw */
6679 I915_WRITE(HALF_SLICE_CHICKEN3,
6680 _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE));
6681
ecdb4eb7 6682 /* WaSwitchSolVfFArbitrationPriority:hsw */
e3dff585
BW
6683 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
6684
90a88643
PZ
6685 /* WaRsPkgCStateDisplayPMReq:hsw */
6686 I915_WRITE(CHICKEN_PAR1_1,
6687 I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
1544d9d5 6688
17a303ec 6689 lpt_init_clock_gating(dev);
cad2a2d7
ED
6690}
6691
1fa61106 6692static void ivybridge_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6693{
6694 struct drm_i915_private *dev_priv = dev->dev_private;
20848223 6695 uint32_t snpcr;
6f1d69b0 6696
017636cc 6697 ilk_init_lp_watermarks(dev);
6f1d69b0 6698
231e54f6 6699 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 6700
ecdb4eb7 6701 /* WaDisableEarlyCull:ivb */
87f8020e
JB
6702 I915_WRITE(_3D_CHICKEN3,
6703 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
6704
ecdb4eb7 6705 /* WaDisableBackToBackFlipFix:ivb */
6f1d69b0
ED
6706 I915_WRITE(IVB_CHICKEN3,
6707 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
6708 CHICKEN3_DGMG_DONE_FIX_DISABLE);
6709
ecdb4eb7 6710 /* WaDisablePSDDualDispatchEnable:ivb */
12f3382b
JB
6711 if (IS_IVB_GT1(dev))
6712 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
6713 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 6714
4e04632e
AG
6715 /* WaDisable_RenderCache_OperationalFlush:ivb */
6716 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6717
ecdb4eb7 6718 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6f1d69b0
ED
6719 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
6720 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
6721
ecdb4eb7 6722 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6f1d69b0
ED
6723 I915_WRITE(GEN7_L3CNTLREG1,
6724 GEN7_WA_FOR_GEN7_L3_CONTROL);
6725 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
8ab43976
JB
6726 GEN7_WA_L3_CHICKEN_MODE);
6727 if (IS_IVB_GT1(dev))
6728 I915_WRITE(GEN7_ROW_CHICKEN2,
6729 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2
VS
6730 else {
6731 /* must write both registers */
6732 I915_WRITE(GEN7_ROW_CHICKEN2,
6733 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
8ab43976
JB
6734 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
6735 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2 6736 }
6f1d69b0 6737
ecdb4eb7 6738 /* WaForceL3Serialization:ivb */
61939d97
JB
6739 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
6740 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
6741
1b80a19a 6742 /*
0f846f81 6743 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 6744 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
0f846f81
JB
6745 */
6746 I915_WRITE(GEN6_UCGCTL2,
28acf3b2 6747 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 6748
ecdb4eb7 6749 /* This is required by WaCatErrorRejectionIssue:ivb */
6f1d69b0
ED
6750 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
6751 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
6752 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
6753
0e088b8f 6754 g4x_disable_trickle_feed(dev);
6f1d69b0
ED
6755
6756 gen7_setup_fixed_func_scheduler(dev_priv);
97e1930f 6757
22721343
CW
6758 if (0) { /* causes HiZ corruption on ivb:gt1 */
6759 /* enable HiZ Raw Stall Optimization */
6760 I915_WRITE(CACHE_MODE_0_GEN7,
6761 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
6762 }
116f2b6d 6763
ecdb4eb7 6764 /* WaDisable4x2SubspanOptimization:ivb */
97e1930f
DV
6765 I915_WRITE(CACHE_MODE_1,
6766 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
20848223 6767
a607c1a4
VS
6768 /*
6769 * BSpec recommends 8x4 when MSAA is used,
6770 * however in practice 16x4 seems fastest.
c5c98a58
VS
6771 *
6772 * Note that PS/WM thread counts depend on the WIZ hashing
6773 * disable bit, which we don't touch here, but it's good
6774 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a607c1a4
VS
6775 */
6776 I915_WRITE(GEN7_GT_MODE,
98533251 6777 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
a607c1a4 6778
20848223
BW
6779 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
6780 snpcr &= ~GEN6_MBC_SNPCR_MASK;
6781 snpcr |= GEN6_MBC_SNPCR_MED;
6782 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
3107bd48 6783
ab5c608b
BW
6784 if (!HAS_PCH_NOP(dev))
6785 cpt_init_clock_gating(dev);
1d7aaa0c
DV
6786
6787 gen6_check_mch_setup(dev);
6f1d69b0
ED
6788}
6789
c6beb13e
VS
6790static void vlv_init_display_clock_gating(struct drm_i915_private *dev_priv)
6791{
6792 I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
6793
6794 /*
6795 * Disable trickle feed and enable pnd deadline calculation
6796 */
6797 I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
6798 I915_WRITE(CBR1_VLV, 0);
6799}
6800
1fa61106 6801static void valleyview_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6802{
6803 struct drm_i915_private *dev_priv = dev->dev_private;
6f1d69b0 6804
c6beb13e 6805 vlv_init_display_clock_gating(dev_priv);
6f1d69b0 6806
ecdb4eb7 6807 /* WaDisableEarlyCull:vlv */
87f8020e
JB
6808 I915_WRITE(_3D_CHICKEN3,
6809 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
6810
ecdb4eb7 6811 /* WaDisableBackToBackFlipFix:vlv */
6f1d69b0
ED
6812 I915_WRITE(IVB_CHICKEN3,
6813 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
6814 CHICKEN3_DGMG_DONE_FIX_DISABLE);
6815
fad7d36e 6816 /* WaPsdDispatchEnable:vlv */
ecdb4eb7 6817 /* WaDisablePSDDualDispatchEnable:vlv */
12f3382b 6818 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
d3bc0303
JB
6819 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
6820 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 6821
4e04632e
AG
6822 /* WaDisable_RenderCache_OperationalFlush:vlv */
6823 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6824
ecdb4eb7 6825 /* WaForceL3Serialization:vlv */
61939d97
JB
6826 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
6827 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
6828
ecdb4eb7 6829 /* WaDisableDopClockGating:vlv */
8ab43976
JB
6830 I915_WRITE(GEN7_ROW_CHICKEN2,
6831 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
6832
ecdb4eb7 6833 /* This is required by WaCatErrorRejectionIssue:vlv */
6f1d69b0
ED
6834 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
6835 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
6836 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
6837
46680e0a
VS
6838 gen7_setup_fixed_func_scheduler(dev_priv);
6839
3c0edaeb 6840 /*
0f846f81 6841 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 6842 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
0f846f81
JB
6843 */
6844 I915_WRITE(GEN6_UCGCTL2,
3c0edaeb 6845 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 6846
c98f5062
AG
6847 /* WaDisableL3Bank2xClockGate:vlv
6848 * Disabling L3 clock gating- MMIO 940c[25] = 1
6849 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
6850 I915_WRITE(GEN7_UCGCTL4,
6851 I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
e3f33d46 6852
afd58e79
VS
6853 /*
6854 * BSpec says this must be set, even though
6855 * WaDisable4x2SubspanOptimization isn't listed for VLV.
6856 */
6b26c86d
DV
6857 I915_WRITE(CACHE_MODE_1,
6858 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
7983117f 6859
da2518f9
VS
6860 /*
6861 * BSpec recommends 8x4 when MSAA is used,
6862 * however in practice 16x4 seems fastest.
6863 *
6864 * Note that PS/WM thread counts depend on the WIZ hashing
6865 * disable bit, which we don't touch here, but it's good
6866 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6867 */
6868 I915_WRITE(GEN7_GT_MODE,
6869 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
6870
031994ee
VS
6871 /*
6872 * WaIncreaseL3CreditsForVLVB0:vlv
6873 * This is the hardware default actually.
6874 */
6875 I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);
6876
2d809570 6877 /*
ecdb4eb7 6878 * WaDisableVLVClockGating_VBIIssue:vlv
2d809570
JB
6879 * Disable clock gating on th GCFG unit to prevent a delay
6880 * in the reporting of vblank events.
6881 */
7a0d1eed 6882 I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
6f1d69b0
ED
6883}
6884
a4565da8
VS
6885static void cherryview_init_clock_gating(struct drm_device *dev)
6886{
6887 struct drm_i915_private *dev_priv = dev->dev_private;
6888
c6beb13e 6889 vlv_init_display_clock_gating(dev_priv);
dd811e70 6890
232ce337
VS
6891 /* WaVSRefCountFullforceMissDisable:chv */
6892 /* WaDSRefCountFullforceMissDisable:chv */
6893 I915_WRITE(GEN7_FF_THREAD_MODE,
6894 I915_READ(GEN7_FF_THREAD_MODE) &
6895 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
acea6f95
VS
6896
6897 /* WaDisableSemaphoreAndSyncFlipWait:chv */
6898 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
6899 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
0846697c
VS
6900
6901 /* WaDisableCSUnitClockGating:chv */
6902 I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
6903 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
c631780f
VS
6904
6905 /* WaDisableSDEUnitClockGating:chv */
6906 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
6907 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
6d50b065
VS
6908
6909 /*
6910 * GTT cache may not work with big pages, so if those
6911 * are ever enabled GTT cache may need to be disabled.
6912 */
6913 I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL);
a4565da8
VS
6914}
6915
1fa61106 6916static void g4x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6917{
6918 struct drm_i915_private *dev_priv = dev->dev_private;
6919 uint32_t dspclk_gate;
6920
6921 I915_WRITE(RENCLK_GATE_D1, 0);
6922 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
6923 GS_UNIT_CLOCK_GATE_DISABLE |
6924 CL_UNIT_CLOCK_GATE_DISABLE);
6925 I915_WRITE(RAMCLK_GATE_D, 0);
6926 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
6927 OVRUNIT_CLOCK_GATE_DISABLE |
6928 OVCUNIT_CLOCK_GATE_DISABLE;
6929 if (IS_GM45(dev))
6930 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
6931 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4358a374
DV
6932
6933 /* WaDisableRenderCachePipelinedFlush */
6934 I915_WRITE(CACHE_MODE_0,
6935 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
de1aa629 6936
4e04632e
AG
6937 /* WaDisable_RenderCache_OperationalFlush:g4x */
6938 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6939
0e088b8f 6940 g4x_disable_trickle_feed(dev);
6f1d69b0
ED
6941}
6942
1fa61106 6943static void crestline_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6944{
6945 struct drm_i915_private *dev_priv = dev->dev_private;
6946
6947 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
6948 I915_WRITE(RENCLK_GATE_D2, 0);
6949 I915_WRITE(DSPCLK_GATE_D, 0);
6950 I915_WRITE(RAMCLK_GATE_D, 0);
6951 I915_WRITE16(DEUC, 0);
20f94967
VS
6952 I915_WRITE(MI_ARB_STATE,
6953 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
4e04632e
AG
6954
6955 /* WaDisable_RenderCache_OperationalFlush:gen4 */
6956 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6f1d69b0
ED
6957}
6958
1fa61106 6959static void broadwater_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6960{
6961 struct drm_i915_private *dev_priv = dev->dev_private;
6962
6963 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
6964 I965_RCC_CLOCK_GATE_DISABLE |
6965 I965_RCPB_CLOCK_GATE_DISABLE |
6966 I965_ISC_CLOCK_GATE_DISABLE |
6967 I965_FBC_CLOCK_GATE_DISABLE);
6968 I915_WRITE(RENCLK_GATE_D2, 0);
20f94967
VS
6969 I915_WRITE(MI_ARB_STATE,
6970 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
4e04632e
AG
6971
6972 /* WaDisable_RenderCache_OperationalFlush:gen4 */
6973 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6f1d69b0
ED
6974}
6975
1fa61106 6976static void gen3_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
6977{
6978 struct drm_i915_private *dev_priv = dev->dev_private;
6979 u32 dstate = I915_READ(D_STATE);
6980
6981 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
6982 DSTATE_DOT_CLOCK_GATING;
6983 I915_WRITE(D_STATE, dstate);
13a86b85
CW
6984
6985 if (IS_PINEVIEW(dev))
6986 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
974a3b0f
DV
6987
6988 /* IIR "flip pending" means done if this bit is set */
6989 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
12fabbcb
VS
6990
6991 /* interrupts should cause a wake up from C3 */
3299254f 6992 I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN));
dbb42748
VS
6993
6994 /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
6995 I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE));
1038392b
VS
6996
6997 I915_WRITE(MI_ARB_STATE,
6998 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED
6999}
7000
1fa61106 7001static void i85x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
7002{
7003 struct drm_i915_private *dev_priv = dev->dev_private;
7004
7005 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
54e472ae
VS
7006
7007 /* interrupts should cause a wake up from C3 */
7008 I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) |
7009 _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE));
1038392b
VS
7010
7011 I915_WRITE(MEM_MODE,
7012 _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED
7013}
7014
1fa61106 7015static void i830_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
7016{
7017 struct drm_i915_private *dev_priv = dev->dev_private;
7018
7019 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
1038392b
VS
7020
7021 I915_WRITE(MEM_MODE,
7022 _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) |
7023 _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE));
6f1d69b0
ED
7024}
7025
6f1d69b0
ED
7026void intel_init_clock_gating(struct drm_device *dev)
7027{
7028 struct drm_i915_private *dev_priv = dev->dev_private;
7029
c57e3551
DL
7030 if (dev_priv->display.init_clock_gating)
7031 dev_priv->display.init_clock_gating(dev);
6f1d69b0
ED
7032}
7033
7d708ee4
ID
7034void intel_suspend_hw(struct drm_device *dev)
7035{
7036 if (HAS_PCH_LPT(dev))
7037 lpt_suspend_hw(dev);
7038}
7039
1fa61106
ED
7040/* Set up chip specific power management-related functions */
7041void intel_init_pm(struct drm_device *dev)
7042{
7043 struct drm_i915_private *dev_priv = dev->dev_private;
7044
7ff0ebcc 7045 intel_fbc_init(dev_priv);
1fa61106 7046
c921aba8
DV
7047 /* For cxsr */
7048 if (IS_PINEVIEW(dev))
7049 i915_pineview_get_mem_freq(dev);
7050 else if (IS_GEN5(dev))
7051 i915_ironlake_get_mem_freq(dev);
7052
1fa61106 7053 /* For FIFO watermark updates */
f5ed50cb 7054 if (INTEL_INFO(dev)->gen >= 9) {
2af30a5c
PB
7055 skl_setup_wm_latency(dev);
7056
a82abe43
ID
7057 if (IS_BROXTON(dev))
7058 dev_priv->display.init_clock_gating =
7059 bxt_init_clock_gating;
2d41c0b5 7060 dev_priv->display.update_wm = skl_update_wm;
2791a16c 7061 dev_priv->display.update_sprite_wm = skl_update_sprite_wm;
c83155a6 7062 } else if (HAS_PCH_SPLIT(dev)) {
fa50ad61 7063 ilk_setup_wm_latency(dev);
53615a5e 7064
bd602544
VS
7065 if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] &&
7066 dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
7067 (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] &&
7068 dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
7069 dev_priv->display.update_wm = ilk_update_wm;
2791a16c 7070 dev_priv->display.update_sprite_wm = ilk_update_sprite_wm;
bd602544
VS
7071 } else {
7072 DRM_DEBUG_KMS("Failed to read display plane latency. "
7073 "Disable CxSR\n");
7074 }
7075
7076 if (IS_GEN5(dev))
1fa61106 7077 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
bd602544 7078 else if (IS_GEN6(dev))
1fa61106 7079 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
bd602544 7080 else if (IS_IVYBRIDGE(dev))
1fa61106 7081 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
bd602544 7082 else if (IS_HASWELL(dev))
cad2a2d7 7083 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
bd602544 7084 else if (INTEL_INFO(dev)->gen == 8)
47c2bd97 7085 dev_priv->display.init_clock_gating = broadwell_init_clock_gating;
a4565da8 7086 } else if (IS_CHERRYVIEW(dev)) {
262cd2e1
VS
7087 vlv_setup_wm_latency(dev);
7088
7089 dev_priv->display.update_wm = vlv_update_wm;
a4565da8
VS
7090 dev_priv->display.init_clock_gating =
7091 cherryview_init_clock_gating;
1fa61106 7092 } else if (IS_VALLEYVIEW(dev)) {
26e1fe4f
VS
7093 vlv_setup_wm_latency(dev);
7094
7095 dev_priv->display.update_wm = vlv_update_wm;
1fa61106
ED
7096 dev_priv->display.init_clock_gating =
7097 valleyview_init_clock_gating;
1fa61106
ED
7098 } else if (IS_PINEVIEW(dev)) {
7099 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
7100 dev_priv->is_ddr3,
7101 dev_priv->fsb_freq,
7102 dev_priv->mem_freq)) {
7103 DRM_INFO("failed to find known CxSR latency "
7104 "(found ddr%s fsb freq %d, mem freq %d), "
7105 "disabling CxSR\n",
7106 (dev_priv->is_ddr3 == 1) ? "3" : "2",
7107 dev_priv->fsb_freq, dev_priv->mem_freq);
7108 /* Disable CxSR and never update its watermark again */
5209b1f4 7109 intel_set_memory_cxsr(dev_priv, false);
1fa61106
ED
7110 dev_priv->display.update_wm = NULL;
7111 } else
7112 dev_priv->display.update_wm = pineview_update_wm;
7113 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
7114 } else if (IS_G4X(dev)) {
7115 dev_priv->display.update_wm = g4x_update_wm;
7116 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
7117 } else if (IS_GEN4(dev)) {
7118 dev_priv->display.update_wm = i965_update_wm;
7119 if (IS_CRESTLINE(dev))
7120 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
7121 else if (IS_BROADWATER(dev))
7122 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
7123 } else if (IS_GEN3(dev)) {
7124 dev_priv->display.update_wm = i9xx_update_wm;
7125 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
7126 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
feb56b93
DV
7127 } else if (IS_GEN2(dev)) {
7128 if (INTEL_INFO(dev)->num_pipes == 1) {
7129 dev_priv->display.update_wm = i845_update_wm;
1fa61106 7130 dev_priv->display.get_fifo_size = i845_get_fifo_size;
feb56b93
DV
7131 } else {
7132 dev_priv->display.update_wm = i9xx_update_wm;
1fa61106 7133 dev_priv->display.get_fifo_size = i830_get_fifo_size;
feb56b93
DV
7134 }
7135
7136 if (IS_I85X(dev) || IS_I865G(dev))
7137 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
7138 else
7139 dev_priv->display.init_clock_gating = i830_init_clock_gating;
7140 } else {
7141 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
1fa61106
ED
7142 }
7143}
7144
151a49d0 7145int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val)
42c0526c 7146{
4fc688ce 7147 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW
7148
7149 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
7150 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
7151 return -EAGAIN;
7152 }
7153
7154 I915_WRITE(GEN6_PCODE_DATA, *val);
dddab346 7155 I915_WRITE(GEN6_PCODE_DATA1, 0);
42c0526c
BW
7156 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
7157
7158 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
7159 500)) {
7160 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
7161 return -ETIMEDOUT;
7162 }
7163
7164 *val = I915_READ(GEN6_PCODE_DATA);
7165 I915_WRITE(GEN6_PCODE_DATA, 0);
7166
7167 return 0;
7168}
7169
151a49d0 7170int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u32 mbox, u32 val)
42c0526c 7171{
4fc688ce 7172 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW
7173
7174 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
7175 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
7176 return -EAGAIN;
7177 }
7178
7179 I915_WRITE(GEN6_PCODE_DATA, val);
7180 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
7181
7182 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
7183 500)) {
7184 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
7185 return -ETIMEDOUT;
7186 }
7187
7188 I915_WRITE(GEN6_PCODE_DATA, 0);
7189
7190 return 0;
7191}
a0e4e199 7192
dd06f88c 7193static int vlv_gpu_freq_div(unsigned int czclk_freq)
855ba3be 7194{
dd06f88c
VS
7195 switch (czclk_freq) {
7196 case 200:
7197 return 10;
7198 case 267:
7199 return 12;
7200 case 320:
7201 case 333:
dd06f88c 7202 return 16;
ab3fb157
VS
7203 case 400:
7204 return 20;
855ba3be
JB
7205 default:
7206 return -1;
7207 }
dd06f88c 7208}
855ba3be 7209
dd06f88c
VS
7210static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val)
7211{
bfa7df01 7212 int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
dd06f88c
VS
7213
7214 div = vlv_gpu_freq_div(czclk_freq);
7215 if (div < 0)
7216 return div;
7217
7218 return DIV_ROUND_CLOSEST(czclk_freq * (val + 6 - 0xbd), div);
855ba3be
JB
7219}
7220
b55dd647 7221static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val)
855ba3be 7222{
bfa7df01 7223 int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
855ba3be 7224
dd06f88c
VS
7225 mul = vlv_gpu_freq_div(czclk_freq);
7226 if (mul < 0)
7227 return mul;
855ba3be 7228
dd06f88c 7229 return DIV_ROUND_CLOSEST(mul * val, czclk_freq) + 0xbd - 6;
855ba3be
JB
7230}
7231
b55dd647 7232static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val)
22b1b2f8 7233{
bfa7df01 7234 int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
22b1b2f8 7235
dd06f88c
VS
7236 div = vlv_gpu_freq_div(czclk_freq) / 2;
7237 if (div < 0)
7238 return div;
22b1b2f8 7239
dd06f88c 7240 return DIV_ROUND_CLOSEST(czclk_freq * val, 2 * div) / 2;
22b1b2f8
D
7241}
7242
b55dd647 7243static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val)
22b1b2f8 7244{
bfa7df01 7245 int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
22b1b2f8 7246
dd06f88c
VS
7247 mul = vlv_gpu_freq_div(czclk_freq) / 2;
7248 if (mul < 0)
7249 return mul;
22b1b2f8 7250
1c14762d 7251 /* CHV needs even values */
dd06f88c 7252 return DIV_ROUND_CLOSEST(val * 2 * mul, czclk_freq) * 2;
22b1b2f8
D
7253}
7254
616bc820 7255int intel_gpu_freq(struct drm_i915_private *dev_priv, int val)
22b1b2f8 7256{
80b6dda4
AG
7257 if (IS_GEN9(dev_priv->dev))
7258 return (val * GT_FREQUENCY_MULTIPLIER) / GEN9_FREQ_SCALER;
7259 else if (IS_CHERRYVIEW(dev_priv->dev))
616bc820 7260 return chv_gpu_freq(dev_priv, val);
22b1b2f8 7261 else if (IS_VALLEYVIEW(dev_priv->dev))
616bc820
VS
7262 return byt_gpu_freq(dev_priv, val);
7263 else
7264 return val * GT_FREQUENCY_MULTIPLIER;
22b1b2f8
D
7265}
7266
616bc820
VS
7267int intel_freq_opcode(struct drm_i915_private *dev_priv, int val)
7268{
80b6dda4
AG
7269 if (IS_GEN9(dev_priv->dev))
7270 return (val * GEN9_FREQ_SCALER) / GT_FREQUENCY_MULTIPLIER;
7271 else if (IS_CHERRYVIEW(dev_priv->dev))
616bc820 7272 return chv_freq_opcode(dev_priv, val);
22b1b2f8 7273 else if (IS_VALLEYVIEW(dev_priv->dev))
616bc820
VS
7274 return byt_freq_opcode(dev_priv, val);
7275 else
7276 return val / GT_FREQUENCY_MULTIPLIER;
7277}
22b1b2f8 7278
6ad790c0
CW
7279struct request_boost {
7280 struct work_struct work;
eed29a5b 7281 struct drm_i915_gem_request *req;
6ad790c0
CW
7282};
7283
7284static void __intel_rps_boost_work(struct work_struct *work)
7285{
7286 struct request_boost *boost = container_of(work, struct request_boost, work);
e61b9958 7287 struct drm_i915_gem_request *req = boost->req;
6ad790c0 7288
e61b9958
CW
7289 if (!i915_gem_request_completed(req, true))
7290 gen6_rps_boost(to_i915(req->ring->dev), NULL,
7291 req->emitted_jiffies);
6ad790c0 7292
e61b9958 7293 i915_gem_request_unreference__unlocked(req);
6ad790c0
CW
7294 kfree(boost);
7295}
7296
7297void intel_queue_rps_boost_for_request(struct drm_device *dev,
eed29a5b 7298 struct drm_i915_gem_request *req)
6ad790c0
CW
7299{
7300 struct request_boost *boost;
7301
eed29a5b 7302 if (req == NULL || INTEL_INFO(dev)->gen < 6)
6ad790c0
CW
7303 return;
7304
e61b9958
CW
7305 if (i915_gem_request_completed(req, true))
7306 return;
7307
6ad790c0
CW
7308 boost = kmalloc(sizeof(*boost), GFP_ATOMIC);
7309 if (boost == NULL)
7310 return;
7311
eed29a5b
DV
7312 i915_gem_request_reference(req);
7313 boost->req = req;
6ad790c0
CW
7314
7315 INIT_WORK(&boost->work, __intel_rps_boost_work);
7316 queue_work(to_i915(dev)->wq, &boost->work);
7317}
7318
f742a552 7319void intel_pm_setup(struct drm_device *dev)
907b28c5
CW
7320{
7321 struct drm_i915_private *dev_priv = dev->dev_private;
7322
f742a552 7323 mutex_init(&dev_priv->rps.hw_lock);
8d3afd7d 7324 spin_lock_init(&dev_priv->rps.client_lock);
f742a552 7325
907b28c5
CW
7326 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
7327 intel_gen6_powersave_work);
1854d5ca 7328 INIT_LIST_HEAD(&dev_priv->rps.clients);
2e1b8730
CW
7329 INIT_LIST_HEAD(&dev_priv->rps.semaphores.link);
7330 INIT_LIST_HEAD(&dev_priv->rps.mmioflips.link);
5d584b2e 7331
33688d95 7332 dev_priv->pm.suspended = false;
907b28c5 7333}