]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - drivers/gpu/drm/i915/intel_pm.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drm/i915/guc: rename loader entry points
[mirror_ubuntu-jammy-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 34/**
18afd443
JN		 35 * DOC: RC6
36 *
dc39fff7
BW		 37 * RC6 is a special power stage which allows the GPU to enter an very
38 * low-voltage mode when idle, using down to 0V while at this stage. This
39 * stage is entered automatically when the GPU is idle when RC6 support is
40 * enabled, and as soon as new workload arises GPU wakes up automatically as well.
41 *
42 * There are different RC6 modes available in Intel GPU, which differentiate
43 * among each other with the latency required to enter and leave RC6 and
44 * voltage consumed by the GPU in different states.
45 *
46 * The combination of the following flags define which states GPU is allowed
47 * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
48 * RC6pp is deepest RC6. Their support by hardware varies according to the
49 * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
50 * which brings the most power savings; deeper states save more power, but
51 * require higher latency to switch to and wake up.
52 */
53#define INTEL_RC6_ENABLE (1<<0)
54#define INTEL_RC6p_ENABLE (1<<1)
55#define INTEL_RC6pp_ENABLE (1<<2)
56
a82abe43
ID		 57static void bxt_init_clock_gating(struct drm_device *dev)
58{
32608ca2
ID		 59 struct drm_i915_private *dev_priv = dev->dev_private;
60
dc00b6a0
DV		 61 /* See Bspec note for PSR2_CTL bit 31, Wa#828:bxt */
62 I915_WRITE(CHICKEN_PAR1_1,
63 I915_READ(CHICKEN_PAR1_1) | SKL_EDP_PSR_FIX_RDWRAP);
64
a7546159
NH		 65 /* WaDisableSDEUnitClockGating:bxt */
66 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
67 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
68
32608ca2
ID		 69 /*
70 * FIXME:
868434c5 71 * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only.
32608ca2 72 */
32608ca2 73 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
868434c5 74 GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ);
d965e7ac
ID		 75
76 /*
77 * Wa: Backlight PWM may stop in the asserted state, causing backlight
78 * to stay fully on.
79 */
80 if (IS_BXT_REVID(dev_priv, BXT_REVID_B0, REVID_FOREVER))
81 I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) |
82 PWM1_GATING_DIS | PWM2_GATING_DIS);
a82abe43
ID		 83}
84
c921aba8
DV		 85static void i915_pineview_get_mem_freq(struct drm_device *dev)
86{
50227e1c 87 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV		 88 u32 tmp;
89
90 tmp = I915_READ(CLKCFG);
91
92 switch (tmp & CLKCFG_FSB_MASK) {
93 case CLKCFG_FSB_533:
94 dev_priv->fsb_freq = 533; /* 133*4 */
95 break;
96 case CLKCFG_FSB_800:
97 dev_priv->fsb_freq = 800; /* 200*4 */
98 break;
99 case CLKCFG_FSB_667:
100 dev_priv->fsb_freq = 667; /* 167*4 */
101 break;
102 case CLKCFG_FSB_400:
103 dev_priv->fsb_freq = 400; /* 100*4 */
104 break;
105 }
106
107 switch (tmp & CLKCFG_MEM_MASK) {
108 case CLKCFG_MEM_533:
109 dev_priv->mem_freq = 533;
110 break;
111 case CLKCFG_MEM_667:
112 dev_priv->mem_freq = 667;
113 break;
114 case CLKCFG_MEM_800:
115 dev_priv->mem_freq = 800;
116 break;
117 }
118
119 /* detect pineview DDR3 setting */
120 tmp = I915_READ(CSHRDDR3CTL);
121 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
122}
123
124static void i915_ironlake_get_mem_freq(struct drm_device *dev)
125{
50227e1c 126 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV		 127 u16 ddrpll, csipll;
128
129 ddrpll = I915_READ16(DDRMPLL1);
130 csipll = I915_READ16(CSIPLL0);
131
132 switch (ddrpll & 0xff) {
133 case 0xc:
134 dev_priv->mem_freq = 800;
135 break;
136 case 0x10:
137 dev_priv->mem_freq = 1066;
138 break;
139 case 0x14:
140 dev_priv->mem_freq = 1333;
141 break;
142 case 0x18:
143 dev_priv->mem_freq = 1600;
144 break;
145 default:
146 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
147 ddrpll & 0xff);
148 dev_priv->mem_freq = 0;
149 break;
150 }
151
20e4d407 152 dev_priv->ips.r_t = dev_priv->mem_freq;
c921aba8
DV		 153
154 switch (csipll & 0x3ff) {
155 case 0x00c:
156 dev_priv->fsb_freq = 3200;
157 break;
158 case 0x00e:
159 dev_priv->fsb_freq = 3733;
160 break;
161 case 0x010:
162 dev_priv->fsb_freq = 4266;
163 break;
164 case 0x012:
165 dev_priv->fsb_freq = 4800;
166 break;
167 case 0x014:
168 dev_priv->fsb_freq = 5333;
169 break;
170 case 0x016:
171 dev_priv->fsb_freq = 5866;
172 break;
173 case 0x018:
174 dev_priv->fsb_freq = 6400;
175 break;
176 default:
177 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
178 csipll & 0x3ff);
179 dev_priv->fsb_freq = 0;
180 break;
181 }
182
183 if (dev_priv->fsb_freq == 3200) {
20e4d407 184 dev_priv->ips.c_m = 0;
c921aba8 185 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
20e4d407 186 dev_priv->ips.c_m = 1;
c921aba8 187 } else {
20e4d407 188 dev_priv->ips.c_m = 2;
c921aba8
DV		 189 }
190}
191
b445e3b0
ED		 192static const struct cxsr_latency cxsr_latency_table[] = {
193 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
194 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
195 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
196 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
197 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
198
199 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
200 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
201 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
202 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
203 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
204
205 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
206 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
207 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
208 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
209 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
210
211 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
212 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
213 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
214 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
215 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
216
217 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
218 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
219 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
220 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
221 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
222
223 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
224 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
225 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
226 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
227 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
228};
229
63c62275 230static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
b445e3b0
ED		 231 int is_ddr3,
232 int fsb,
233 int mem)
234{
235 const struct cxsr_latency *latency;
236 int i;
237
238 if (fsb == 0 || mem == 0)
239 return NULL;
240
241 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
242 latency = &cxsr_latency_table[i];
243 if (is_desktop == latency->is_desktop &&
244 is_ddr3 == latency->is_ddr3 &&
245 fsb == latency->fsb_freq && mem == latency->mem_freq)
246 return latency;
247 }
248
249 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
250
251 return NULL;
252}
253
fc1ac8de
VS		 254static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable)
255{
256 u32 val;
257
258 mutex_lock(&dev_priv->rps.hw_lock);
259
260 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
261 if (enable)
262 val &= ~FORCE_DDR_HIGH_FREQ;
263 else
264 val |= FORCE_DDR_HIGH_FREQ;
265 val &= ~FORCE_DDR_LOW_FREQ;
266 val |= FORCE_DDR_FREQ_REQ_ACK;
267 vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);
268
269 if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
270 FORCE_DDR_FREQ_REQ_ACK) == 0, 3))
271 DRM_ERROR("timed out waiting for Punit DDR DVFS request\n");
272
273 mutex_unlock(&dev_priv->rps.hw_lock);
274}
275
cfb41411
VS		 276static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable)
277{
278 u32 val;
279
280 mutex_lock(&dev_priv->rps.hw_lock);
281
282 val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ);
283 if (enable)
284 val |= DSP_MAXFIFO_PM5_ENABLE;
285 else
286 val &= ~DSP_MAXFIFO_PM5_ENABLE;
287 vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val);
288
289 mutex_unlock(&dev_priv->rps.hw_lock);
290}
291
f4998963
VS		 292#define FW_WM(value, plane) \
293 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK)
294
5209b1f4 295void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable)
b445e3b0 296{
5209b1f4
ID		 297 struct drm_device *dev = dev_priv->dev;
298 u32 val;
b445e3b0 299
666a4537 300 if (IS_VALLEYVIEW(dev) || IS_CHERRYVIEW(dev)) {
5209b1f4 301 I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0);
a7a6c498 302 POSTING_READ(FW_BLC_SELF_VLV);
852eb00d 303 dev_priv->wm.vlv.cxsr = enable;
5209b1f4
ID		 304 } else if (IS_G4X(dev) || IS_CRESTLINE(dev)) {
305 I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0);
a7a6c498 306 POSTING_READ(FW_BLC_SELF);
5209b1f4
ID		 307 } else if (IS_PINEVIEW(dev)) {
308 val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN;
309 val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0;
310 I915_WRITE(DSPFW3, val);
a7a6c498 311 POSTING_READ(DSPFW3);
5209b1f4
ID		 312 } else if (IS_I945G(dev) || IS_I945GM(dev)) {
313 val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) :
314 _MASKED_BIT_DISABLE(FW_BLC_SELF_EN);
315 I915_WRITE(FW_BLC_SELF, val);
a7a6c498 316 POSTING_READ(FW_BLC_SELF);
5209b1f4
ID		 317 } else if (IS_I915GM(dev)) {
318 val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) :
319 _MASKED_BIT_DISABLE(INSTPM_SELF_EN);
320 I915_WRITE(INSTPM, val);
a7a6c498 321 POSTING_READ(INSTPM);
5209b1f4
ID		 322 } else {
323 return;
324 }
b445e3b0 325
5209b1f4
ID		 326 DRM_DEBUG_KMS("memory self-refresh is %s\n",
327 enable ? "enabled" : "disabled");
b445e3b0
ED		 328}
329
fc1ac8de 330
b445e3b0
ED		 331/*
332 * Latency for FIFO fetches is dependent on several factors:
333 * - memory configuration (speed, channels)
334 * - chipset
335 * - current MCH state
336 * It can be fairly high in some situations, so here we assume a fairly
337 * pessimal value. It's a tradeoff between extra memory fetches (if we
338 * set this value too high, the FIFO will fetch frequently to stay full)
339 * and power consumption (set it too low to save power and we might see
340 * FIFO underruns and display "flicker").
341 *
342 * A value of 5us seems to be a good balance; safe for very low end
343 * platforms but not overly aggressive on lower latency configs.
344 */
5aef6003 345static const int pessimal_latency_ns = 5000;
b445e3b0 346
b5004720
VS		 347#define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \
348 ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8))
349
350static int vlv_get_fifo_size(struct drm_device *dev,
351 enum pipe pipe, int plane)
352{
353 struct drm_i915_private *dev_priv = dev->dev_private;
354 int sprite0_start, sprite1_start, size;
355
356 switch (pipe) {
357 uint32_t dsparb, dsparb2, dsparb3;
358 case PIPE_A:
359 dsparb = I915_READ(DSPARB);
360 dsparb2 = I915_READ(DSPARB2);
361 sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0);
362 sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4);
363 break;
364 case PIPE_B:
365 dsparb = I915_READ(DSPARB);
366 dsparb2 = I915_READ(DSPARB2);
367 sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8);
368 sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12);
369 break;
370 case PIPE_C:
371 dsparb2 = I915_READ(DSPARB2);
372 dsparb3 = I915_READ(DSPARB3);
373 sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16);
374 sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20);
375 break;
376 default:
377 return 0;
378 }
379
380 switch (plane) {
381 case 0:
382 size = sprite0_start;
383 break;
384 case 1:
385 size = sprite1_start - sprite0_start;
386 break;
387 case 2:
388 size = 512 - 1 - sprite1_start;
389 break;
390 default:
391 return 0;
392 }
393
394 DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n",
395 pipe_name(pipe), plane == 0 ? "primary" : "sprite",
396 plane == 0 ? plane_name(pipe) : sprite_name(pipe, plane - 1),
397 size);
398
399 return size;
400}
401
1fa61106 402static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 403{
404 struct drm_i915_private *dev_priv = dev->dev_private;
405 uint32_t dsparb = I915_READ(DSPARB);
406 int size;
407
408 size = dsparb & 0x7f;
409 if (plane)
410 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
411
412 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
413 plane ? "B" : "A", size);
414
415 return size;
416}
417
feb56b93 418static int i830_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 419{
420 struct drm_i915_private *dev_priv = dev->dev_private;
421 uint32_t dsparb = I915_READ(DSPARB);
422 int size;
423
424 size = dsparb & 0x1ff;
425 if (plane)
426 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
427 size >>= 1; /* Convert to cachelines */
428
429 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
430 plane ? "B" : "A", size);
431
432 return size;
433}
434
1fa61106 435static int i845_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 436{
437 struct drm_i915_private *dev_priv = dev->dev_private;
438 uint32_t dsparb = I915_READ(DSPARB);
439 int size;
440
441 size = dsparb & 0x7f;
442 size >>= 2; /* Convert to cachelines */
443
444 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
445 plane ? "B" : "A",
446 size);
447
448 return size;
449}
450
b445e3b0
ED		 451/* Pineview has different values for various configs */
452static const struct intel_watermark_params pineview_display_wm = {
e0f0273e
VS		 453 .fifo_size = PINEVIEW_DISPLAY_FIFO,
454 .max_wm = PINEVIEW_MAX_WM,
455 .default_wm = PINEVIEW_DFT_WM,
456 .guard_size = PINEVIEW_GUARD_WM,
457 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 458};
459static const struct intel_watermark_params pineview_display_hplloff_wm = {
e0f0273e
VS		 460 .fifo_size = PINEVIEW_DISPLAY_FIFO,
461 .max_wm = PINEVIEW_MAX_WM,
462 .default_wm = PINEVIEW_DFT_HPLLOFF_WM,
463 .guard_size = PINEVIEW_GUARD_WM,
464 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 465};
466static const struct intel_watermark_params pineview_cursor_wm = {
e0f0273e
VS		 467 .fifo_size = PINEVIEW_CURSOR_FIFO,
468 .max_wm = PINEVIEW_CURSOR_MAX_WM,
469 .default_wm = PINEVIEW_CURSOR_DFT_WM,
470 .guard_size = PINEVIEW_CURSOR_GUARD_WM,
471 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 472};
473static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
e0f0273e
VS		 474 .fifo_size = PINEVIEW_CURSOR_FIFO,
475 .max_wm = PINEVIEW_CURSOR_MAX_WM,
476 .default_wm = PINEVIEW_CURSOR_DFT_WM,
477 .guard_size = PINEVIEW_CURSOR_GUARD_WM,
478 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 479};
480static const struct intel_watermark_params g4x_wm_info = {
e0f0273e
VS		 481 .fifo_size = G4X_FIFO_SIZE,
482 .max_wm = G4X_MAX_WM,
483 .default_wm = G4X_MAX_WM,
484 .guard_size = 2,
485 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED		 486};
487static const struct intel_watermark_params g4x_cursor_wm_info = {
e0f0273e
VS		 488 .fifo_size = I965_CURSOR_FIFO,
489 .max_wm = I965_CURSOR_MAX_WM,
490 .default_wm = I965_CURSOR_DFT_WM,
491 .guard_size = 2,
492 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0 493};
b445e3b0 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
ac484963 541 * @cpp: bytes per pixel
b445e3b0
ED		 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,
ac484963 557 int fifo_size, int cpp,
b445e3b0
ED		 558 unsigned long latency_ns)
559{
560 long entries_required, wm_size;
561
562 /*
563 * Note: we need to make sure we don't overflow for various clock &
564 * latency values.
565 * clocks go from a few thousand to several hundred thousand.
566 * latency is usually a few thousand
567 */
ac484963 568 entries_required = ((clock_in_khz / 1000) * cpp * latency_ns) /
b445e3b0
ED		 569 1000;
570 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
571
572 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
573
574 wm_size = fifo_size - (entries_required + wm->guard_size);
575
576 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
577
578 /* Don't promote wm_size to unsigned... */
579 if (wm_size > (long)wm->max_wm)
580 wm_size = wm->max_wm;
581 if (wm_size <= 0)
582 wm_size = wm->default_wm;
d6feb196
VS		 583
584 /*
585 * Bspec seems to indicate that the value shouldn't be lower than
586 * 'burst size + 1'. Certainly 830 is quite unhappy with low values.
587 * Lets go for 8 which is the burst size since certain platforms
588 * already use a hardcoded 8 (which is what the spec says should be
589 * done).
590 */
591 if (wm_size <= 8)
592 wm_size = 8;
593
b445e3b0
ED		 594 return wm_size;
595}
596
597static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
598{
599 struct drm_crtc *crtc, *enabled = NULL;
600
70e1e0ec 601 for_each_crtc(dev, crtc) {
3490ea5d 602 if (intel_crtc_active(crtc)) {
b445e3b0
ED		 603 if (enabled)
604 return NULL;
605 enabled = crtc;
606 }
607 }
608
609 return enabled;
610}
611
46ba614c 612static void pineview_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 613{
46ba614c 614 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 615 struct drm_i915_private *dev_priv = dev->dev_private;
616 struct drm_crtc *crtc;
617 const struct cxsr_latency *latency;
618 u32 reg;
619 unsigned long wm;
620
621 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
622 dev_priv->fsb_freq, dev_priv->mem_freq);
623 if (!latency) {
624 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
5209b1f4 625 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED		 626 return;
627 }
628
629 crtc = single_enabled_crtc(dev);
630 if (crtc) {
7c5f93b0 631 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
ac484963 632 int cpp = drm_format_plane_cpp(crtc->primary->state->fb->pixel_format, 0);
7c5f93b0 633 int clock = adjusted_mode->crtc_clock;
b445e3b0
ED		 634
635 /* Display SR */
636 wm = intel_calculate_wm(clock, &pineview_display_wm,
637 pineview_display_wm.fifo_size,
ac484963 638 cpp, latency->display_sr);
b445e3b0
ED		 639 reg = I915_READ(DSPFW1);
640 reg &= ~DSPFW_SR_MASK;
f4998963 641 reg |= FW_WM(wm, SR);
b445e3b0
ED		 642 I915_WRITE(DSPFW1, reg);
643 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
644
645 /* cursor SR */
646 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
647 pineview_display_wm.fifo_size,
ac484963 648 cpp, latency->cursor_sr);
b445e3b0
ED		 649 reg = I915_READ(DSPFW3);
650 reg &= ~DSPFW_CURSOR_SR_MASK;
f4998963 651 reg |= FW_WM(wm, CURSOR_SR);
b445e3b0
ED		 652 I915_WRITE(DSPFW3, reg);
653
654 /* Display HPLL off SR */
655 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
656 pineview_display_hplloff_wm.fifo_size,
ac484963 657 cpp, latency->display_hpll_disable);
b445e3b0
ED		 658 reg = I915_READ(DSPFW3);
659 reg &= ~DSPFW_HPLL_SR_MASK;
f4998963 660 reg |= FW_WM(wm, HPLL_SR);
b445e3b0
ED		 661 I915_WRITE(DSPFW3, reg);
662
663 /* cursor HPLL off SR */
664 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
665 pineview_display_hplloff_wm.fifo_size,
ac484963 666 cpp, latency->cursor_hpll_disable);
b445e3b0
ED		 667 reg = I915_READ(DSPFW3);
668 reg &= ~DSPFW_HPLL_CURSOR_MASK;
f4998963 669 reg |= FW_WM(wm, HPLL_CURSOR);
b445e3b0
ED		 670 I915_WRITE(DSPFW3, reg);
671 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
672
5209b1f4 673 intel_set_memory_cxsr(dev_priv, true);
b445e3b0 674 } else {
5209b1f4 675 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED		 676 }
677}
678
679static bool g4x_compute_wm0(struct drm_device *dev,
680 int plane,
681 const struct intel_watermark_params *display,
682 int display_latency_ns,
683 const struct intel_watermark_params *cursor,
684 int cursor_latency_ns,
685 int *plane_wm,
686 int *cursor_wm)
687{
688 struct drm_crtc *crtc;
4fe8590a 689 const struct drm_display_mode *adjusted_mode;
ac484963 690 int htotal, hdisplay, clock, cpp;
b445e3b0
ED		 691 int line_time_us, line_count;
692 int entries, tlb_miss;
693
694 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 695 if (!intel_crtc_active(crtc)) {
b445e3b0
ED		 696 *cursor_wm = cursor->guard_size;
697 *plane_wm = display->guard_size;
698 return false;
699 }
700
6e3c9717 701 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 702 clock = adjusted_mode->crtc_clock;
fec8cba3 703 htotal = adjusted_mode->crtc_htotal;
6e3c9717 704 hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
ac484963 705 cpp = drm_format_plane_cpp(crtc->primary->state->fb->pixel_format, 0);
b445e3b0
ED		 706
707 /* Use the small buffer method to calculate plane watermark */
ac484963 708 entries = ((clock * cpp / 1000) * display_latency_ns) / 1000;
b445e3b0
ED		 709 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
710 if (tlb_miss > 0)
711 entries += tlb_miss;
712 entries = DIV_ROUND_UP(entries, display->cacheline_size);
713 *plane_wm = entries + display->guard_size;
714 if (*plane_wm > (int)display->max_wm)
715 *plane_wm = display->max_wm;
716
717 /* Use the large buffer method to calculate cursor watermark */
922044c9 718 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0 719 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
ac484963 720 entries = line_count * crtc->cursor->state->crtc_w * cpp;
b445e3b0
ED		 721 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
722 if (tlb_miss > 0)
723 entries += tlb_miss;
724 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
725 *cursor_wm = entries + cursor->guard_size;
726 if (*cursor_wm > (int)cursor->max_wm)
727 *cursor_wm = (int)cursor->max_wm;
728
729 return true;
730}
731
732/*
733 * Check the wm result.
734 *
735 * If any calculated watermark values is larger than the maximum value that
736 * can be programmed into the associated watermark register, that watermark
737 * must be disabled.
738 */
739static bool g4x_check_srwm(struct drm_device *dev,
740 int display_wm, int cursor_wm,
741 const struct intel_watermark_params *display,
742 const struct intel_watermark_params *cursor)
743{
744 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
745 display_wm, cursor_wm);
746
747 if (display_wm > display->max_wm) {
748 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
749 display_wm, display->max_wm);
750 return false;
751 }
752
753 if (cursor_wm > cursor->max_wm) {
754 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
755 cursor_wm, cursor->max_wm);
756 return false;
757 }
758
759 if (!(display_wm || cursor_wm)) {
760 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
761 return false;
762 }
763
764 return true;
765}
766
767static bool g4x_compute_srwm(struct drm_device *dev,
768 int plane,
769 int latency_ns,
770 const struct intel_watermark_params *display,
771 const struct intel_watermark_params *cursor,
772 int *display_wm, int *cursor_wm)
773{
774 struct drm_crtc *crtc;
4fe8590a 775 const struct drm_display_mode *adjusted_mode;
ac484963 776 int hdisplay, htotal, cpp, clock;
b445e3b0
ED		 777 unsigned long line_time_us;
778 int line_count, line_size;
779 int small, large;
780 int entries;
781
782 if (!latency_ns) {
783 *display_wm = *cursor_wm = 0;
784 return false;
785 }
786
787 crtc = intel_get_crtc_for_plane(dev, plane);
6e3c9717 788 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 789 clock = adjusted_mode->crtc_clock;
fec8cba3 790 htotal = adjusted_mode->crtc_htotal;
6e3c9717 791 hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
ac484963 792 cpp = drm_format_plane_cpp(crtc->primary->state->fb->pixel_format, 0);
b445e3b0 793
922044c9 794 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0 795 line_count = (latency_ns / line_time_us + 1000) / 1000;
ac484963 796 line_size = hdisplay * cpp;
b445e3b0
ED		 797
798 /* Use the minimum of the small and large buffer method for primary */
ac484963 799 small = ((clock * cpp / 1000) * latency_ns) / 1000;
b445e3b0
ED		 800 large = line_count * line_size;
801
802 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
803 *display_wm = entries + display->guard_size;
804
805 /* calculate the self-refresh watermark for display cursor */
ac484963 806 entries = line_count * cpp * crtc->cursor->state->crtc_w;
b445e3b0
ED		 807 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
808 *cursor_wm = entries + cursor->guard_size;
809
810 return g4x_check_srwm(dev,
811 *display_wm, *cursor_wm,
812 display, cursor);
813}
814
15665979
VS		 815#define FW_WM_VLV(value, plane) \
816 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV)
817
0018fda1
VS		 818static void vlv_write_wm_values(struct intel_crtc *crtc,
819 const struct vlv_wm_values *wm)
820{
821 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
822 enum pipe pipe = crtc->pipe;
823
824 I915_WRITE(VLV_DDL(pipe),
825 (wm->ddl[pipe].cursor << DDL_CURSOR_SHIFT) |
826 (wm->ddl[pipe].sprite[1] << DDL_SPRITE_SHIFT(1)) |
827 (wm->ddl[pipe].sprite[0] << DDL_SPRITE_SHIFT(0)) |
828 (wm->ddl[pipe].primary << DDL_PLANE_SHIFT));
829
ae80152d 830 I915_WRITE(DSPFW1,
15665979
VS		 831 FW_WM(wm->sr.plane, SR) |
832 FW_WM(wm->pipe[PIPE_B].cursor, CURSORB) |
833 FW_WM_VLV(wm->pipe[PIPE_B].primary, PLANEB) |
834 FW_WM_VLV(wm->pipe[PIPE_A].primary, PLANEA));
ae80152d 835 I915_WRITE(DSPFW2,
15665979
VS		 836 FW_WM_VLV(wm->pipe[PIPE_A].sprite[1], SPRITEB) |
837 FW_WM(wm->pipe[PIPE_A].cursor, CURSORA) |
838 FW_WM_VLV(wm->pipe[PIPE_A].sprite[0], SPRITEA));
ae80152d 839 I915_WRITE(DSPFW3,
15665979 840 FW_WM(wm->sr.cursor, CURSOR_SR));
ae80152d
VS		 841
842 if (IS_CHERRYVIEW(dev_priv)) {
843 I915_WRITE(DSPFW7_CHV,
15665979
VS		 844 FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) |
845 FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC));
ae80152d 846 I915_WRITE(DSPFW8_CHV,
15665979
VS		 847 FW_WM_VLV(wm->pipe[PIPE_C].sprite[1], SPRITEF) |
848 FW_WM_VLV(wm->pipe[PIPE_C].sprite[0], SPRITEE));
ae80152d 849 I915_WRITE(DSPFW9_CHV,
15665979
VS		 850 FW_WM_VLV(wm->pipe[PIPE_C].primary, PLANEC) |
851 FW_WM(wm->pipe[PIPE_C].cursor, CURSORC));
ae80152d 852 I915_WRITE(DSPHOWM,
15665979
VS		 853 FW_WM(wm->sr.plane >> 9, SR_HI) |
854 FW_WM(wm->pipe[PIPE_C].sprite[1] >> 8, SPRITEF_HI) |
855 FW_WM(wm->pipe[PIPE_C].sprite[0] >> 8, SPRITEE_HI) |
856 FW_WM(wm->pipe[PIPE_C].primary >> 8, PLANEC_HI) |
857 FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) |
858 FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) |
859 FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) |
860 FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) |
861 FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) |
862 FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI));
ae80152d
VS		 863 } else {
864 I915_WRITE(DSPFW7,
15665979
VS		 865 FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) |
866 FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC));
ae80152d 867 I915_WRITE(DSPHOWM,
15665979
VS		 868 FW_WM(wm->sr.plane >> 9, SR_HI) |
869 FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) |
870 FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) |
871 FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) |
872 FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) |
873 FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) |
874 FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI));
ae80152d
VS		 875 }
876
2cb389b7
VS		 877 /* zero (unused) WM1 watermarks */
878 I915_WRITE(DSPFW4, 0);
879 I915_WRITE(DSPFW5, 0);
880 I915_WRITE(DSPFW6, 0);
881 I915_WRITE(DSPHOWM1, 0);
882
ae80152d 883 POSTING_READ(DSPFW1);
0018fda1
VS		 884}
885
15665979
VS		 886#undef FW_WM_VLV
887
6eb1a681
VS		 888enum vlv_wm_level {
889 VLV_WM_LEVEL_PM2,
890 VLV_WM_LEVEL_PM5,
891 VLV_WM_LEVEL_DDR_DVFS,
6eb1a681
VS		 892};
893
262cd2e1
VS		 894/* latency must be in 0.1us units. */
895static unsigned int vlv_wm_method2(unsigned int pixel_rate,
896 unsigned int pipe_htotal,
897 unsigned int horiz_pixels,
ac484963 898 unsigned int cpp,
262cd2e1
VS		 899 unsigned int latency)
900{
901 unsigned int ret;
902
903 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
ac484963 904 ret = (ret + 1) * horiz_pixels * cpp;
262cd2e1
VS		 905 ret = DIV_ROUND_UP(ret, 64);
906
907 return ret;
908}
909
910static void vlv_setup_wm_latency(struct drm_device *dev)
911{
912 struct drm_i915_private *dev_priv = dev->dev_private;
913
914 /* all latencies in usec */
915 dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3;
916
58590c14
VS		 917 dev_priv->wm.max_level = VLV_WM_LEVEL_PM2;
918
262cd2e1
VS		 919 if (IS_CHERRYVIEW(dev_priv)) {
920 dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12;
921 dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33;
58590c14
VS		 922
923 dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS;
262cd2e1
VS		 924 }
925}
926
927static uint16_t vlv_compute_wm_level(struct intel_plane *plane,
928 struct intel_crtc *crtc,
929 const struct intel_plane_state *state,
930 int level)
931{
932 struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
ac484963 933 int clock, htotal, cpp, width, wm;
262cd2e1
VS		 934
935 if (dev_priv->wm.pri_latency[level] == 0)
936 return USHRT_MAX;
937
938 if (!state->visible)
939 return 0;
940
ac484963 941 cpp = drm_format_plane_cpp(state->base.fb->pixel_format, 0);
262cd2e1
VS		 942 clock = crtc->config->base.adjusted_mode.crtc_clock;
943 htotal = crtc->config->base.adjusted_mode.crtc_htotal;
944 width = crtc->config->pipe_src_w;
945 if (WARN_ON(htotal == 0))
946 htotal = 1;
947
948 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
949 /*
950 * FIXME the formula gives values that are
951 * too big for the cursor FIFO, and hence we
952 * would never be able to use cursors. For
953 * now just hardcode the watermark.
954 */
955 wm = 63;
956 } else {
ac484963 957 wm = vlv_wm_method2(clock, htotal, width, cpp,
262cd2e1
VS		 958 dev_priv->wm.pri_latency[level] * 10);
959 }
960
961 return min_t(int, wm, USHRT_MAX);
962}
963
54f1b6e1
VS		 964static void vlv_compute_fifo(struct intel_crtc *crtc)
965{
966 struct drm_device *dev = crtc->base.dev;
967 struct vlv_wm_state *wm_state = &crtc->wm_state;
968 struct intel_plane *plane;
969 unsigned int total_rate = 0;
970 const int fifo_size = 512 - 1;
971 int fifo_extra, fifo_left = fifo_size;
972
973 for_each_intel_plane_on_crtc(dev, crtc, plane) {
974 struct intel_plane_state *state =
975 to_intel_plane_state(plane->base.state);
976
977 if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
978 continue;
979
980 if (state->visible) {
981 wm_state->num_active_planes++;
982 total_rate += drm_format_plane_cpp(state->base.fb->pixel_format, 0);
983 }
984 }
985
986 for_each_intel_plane_on_crtc(dev, crtc, plane) {
987 struct intel_plane_state *state =
988 to_intel_plane_state(plane->base.state);
989 unsigned int rate;
990
991 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
992 plane->wm.fifo_size = 63;
993 continue;
994 }
995
996 if (!state->visible) {
997 plane->wm.fifo_size = 0;
998 continue;
999 }
1000
1001 rate = drm_format_plane_cpp(state->base.fb->pixel_format, 0);
1002 plane->wm.fifo_size = fifo_size * rate / total_rate;
1003 fifo_left -= plane->wm.fifo_size;
1004 }
1005
1006 fifo_extra = DIV_ROUND_UP(fifo_left, wm_state->num_active_planes ?: 1);
1007
1008 /* spread the remainder evenly */
1009 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1010 int plane_extra;
1011
1012 if (fifo_left == 0)
1013 break;
1014
1015 if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
1016 continue;
1017
1018 /* give it all to the first plane if none are active */
1019 if (plane->wm.fifo_size == 0 &&
1020 wm_state->num_active_planes)
1021 continue;
1022
1023 plane_extra = min(fifo_extra, fifo_left);
1024 plane->wm.fifo_size += plane_extra;
1025 fifo_left -= plane_extra;
1026 }
1027
1028 WARN_ON(fifo_left != 0);
1029}
1030
262cd2e1
VS		 1031static void vlv_invert_wms(struct intel_crtc *crtc)
1032{
1033 struct vlv_wm_state *wm_state = &crtc->wm_state;
1034 int level;
1035
1036 for (level = 0; level < wm_state->num_levels; level++) {
1037 struct drm_device *dev = crtc->base.dev;
1038 const int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1;
1039 struct intel_plane *plane;
1040
1041 wm_state->sr[level].plane = sr_fifo_size - wm_state->sr[level].plane;
1042 wm_state->sr[level].cursor = 63 - wm_state->sr[level].cursor;
1043
1044 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1045 switch (plane->base.type) {
1046 int sprite;
1047 case DRM_PLANE_TYPE_CURSOR:
1048 wm_state->wm[level].cursor = plane->wm.fifo_size -
1049 wm_state->wm[level].cursor;
1050 break;
1051 case DRM_PLANE_TYPE_PRIMARY:
1052 wm_state->wm[level].primary = plane->wm.fifo_size -
1053 wm_state->wm[level].primary;
1054 break;
1055 case DRM_PLANE_TYPE_OVERLAY:
1056 sprite = plane->plane;
1057 wm_state->wm[level].sprite[sprite] = plane->wm.fifo_size -
1058 wm_state->wm[level].sprite[sprite];
1059 break;
1060 }
1061 }
1062 }
1063}
1064
26e1fe4f 1065static void vlv_compute_wm(struct intel_crtc *crtc)
262cd2e1
VS		 1066{
1067 struct drm_device *dev = crtc->base.dev;
1068 struct vlv_wm_state *wm_state = &crtc->wm_state;
1069 struct intel_plane *plane;
1070 int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1;
1071 int level;
1072
1073 memset(wm_state, 0, sizeof(*wm_state));
1074
852eb00d 1075 wm_state->cxsr = crtc->pipe != PIPE_C && crtc->wm.cxsr_allowed;
58590c14 1076 wm_state->num_levels = to_i915(dev)->wm.max_level + 1;
262cd2e1
VS		 1077
1078 wm_state->num_active_planes = 0;
262cd2e1 1079
54f1b6e1 1080 vlv_compute_fifo(crtc);
262cd2e1
VS		 1081
1082 if (wm_state->num_active_planes != 1)
1083 wm_state->cxsr = false;
1084
1085 if (wm_state->cxsr) {
1086 for (level = 0; level < wm_state->num_levels; level++) {
1087 wm_state->sr[level].plane = sr_fifo_size;
1088 wm_state->sr[level].cursor = 63;
1089 }
1090 }
1091
1092 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1093 struct intel_plane_state *state =
1094 to_intel_plane_state(plane->base.state);
1095
1096 if (!state->visible)
1097 continue;
1098
1099 /* normal watermarks */
1100 for (level = 0; level < wm_state->num_levels; level++) {
1101 int wm = vlv_compute_wm_level(plane, crtc, state, level);
1102 int max_wm = plane->base.type == DRM_PLANE_TYPE_CURSOR ? 63 : 511;
1103
1104 /* hack */
1105 if (WARN_ON(level == 0 && wm > max_wm))
1106 wm = max_wm;
1107
1108 if (wm > plane->wm.fifo_size)
1109 break;
1110
1111 switch (plane->base.type) {
1112 int sprite;
1113 case DRM_PLANE_TYPE_CURSOR:
1114 wm_state->wm[level].cursor = wm;
1115 break;
1116 case DRM_PLANE_TYPE_PRIMARY:
1117 wm_state->wm[level].primary = wm;
1118 break;
1119 case DRM_PLANE_TYPE_OVERLAY:
1120 sprite = plane->plane;
1121 wm_state->wm[level].sprite[sprite] = wm;
1122 break;
1123 }
1124 }
1125
1126 wm_state->num_levels = level;
1127
1128 if (!wm_state->cxsr)
1129 continue;
1130
1131 /* maxfifo watermarks */
1132 switch (plane->base.type) {
1133 int sprite, level;
1134 case DRM_PLANE_TYPE_CURSOR:
1135 for (level = 0; level < wm_state->num_levels; level++)
1136 wm_state->sr[level].cursor =
5a37ed0a 1137 wm_state->wm[level].cursor;
262cd2e1
VS		 1138 break;
1139 case DRM_PLANE_TYPE_PRIMARY:
1140 for (level = 0; level < wm_state->num_levels; level++)
1141 wm_state->sr[level].plane =
1142 min(wm_state->sr[level].plane,
1143 wm_state->wm[level].primary);
1144 break;
1145 case DRM_PLANE_TYPE_OVERLAY:
1146 sprite = plane->plane;
1147 for (level = 0; level < wm_state->num_levels; level++)
1148 wm_state->sr[level].plane =
1149 min(wm_state->sr[level].plane,
1150 wm_state->wm[level].sprite[sprite]);
1151 break;
1152 }
1153 }
1154
1155 /* clear any (partially) filled invalid levels */
58590c14 1156 for (level = wm_state->num_levels; level < to_i915(dev)->wm.max_level + 1; level++) {
262cd2e1
VS		 1157 memset(&wm_state->wm[level], 0, sizeof(wm_state->wm[level]));
1158 memset(&wm_state->sr[level], 0, sizeof(wm_state->sr[level]));
1159 }
1160
1161 vlv_invert_wms(crtc);
1162}
1163
54f1b6e1
VS		 1164#define VLV_FIFO(plane, value) \
1165 (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV)
1166
1167static void vlv_pipe_set_fifo_size(struct intel_crtc *crtc)
1168{
1169 struct drm_device *dev = crtc->base.dev;
1170 struct drm_i915_private *dev_priv = to_i915(dev);
1171 struct intel_plane *plane;
1172 int sprite0_start = 0, sprite1_start = 0, fifo_size = 0;
1173
1174 for_each_intel_plane_on_crtc(dev, crtc, plane) {
1175 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
1176 WARN_ON(plane->wm.fifo_size != 63);
1177 continue;
1178 }
1179
1180 if (plane->base.type == DRM_PLANE_TYPE_PRIMARY)
1181 sprite0_start = plane->wm.fifo_size;
1182 else if (plane->plane == 0)
1183 sprite1_start = sprite0_start + plane->wm.fifo_size;
1184 else
1185 fifo_size = sprite1_start + plane->wm.fifo_size;
1186 }
1187
1188 WARN_ON(fifo_size != 512 - 1);
1189
1190 DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n",
1191 pipe_name(crtc->pipe), sprite0_start,
1192 sprite1_start, fifo_size);
1193
1194 switch (crtc->pipe) {
1195 uint32_t dsparb, dsparb2, dsparb3;
1196 case PIPE_A:
1197 dsparb = I915_READ(DSPARB);
1198 dsparb2 = I915_READ(DSPARB2);
1199
1200 dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) |
1201 VLV_FIFO(SPRITEB, 0xff));
1202 dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) |
1203 VLV_FIFO(SPRITEB, sprite1_start));
1204
1205 dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) |
1206 VLV_FIFO(SPRITEB_HI, 0x1));
1207 dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) |
1208 VLV_FIFO(SPRITEB_HI, sprite1_start >> 8));
1209
1210 I915_WRITE(DSPARB, dsparb);
1211 I915_WRITE(DSPARB2, dsparb2);
1212 break;
1213 case PIPE_B:
1214 dsparb = I915_READ(DSPARB);
1215 dsparb2 = I915_READ(DSPARB2);
1216
1217 dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) |
1218 VLV_FIFO(SPRITED, 0xff));
1219 dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) |
1220 VLV_FIFO(SPRITED, sprite1_start));
1221
1222 dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) |
1223 VLV_FIFO(SPRITED_HI, 0xff));
1224 dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) |
1225 VLV_FIFO(SPRITED_HI, sprite1_start >> 8));
1226
1227 I915_WRITE(DSPARB, dsparb);
1228 I915_WRITE(DSPARB2, dsparb2);
1229 break;
1230 case PIPE_C:
1231 dsparb3 = I915_READ(DSPARB3);
1232 dsparb2 = I915_READ(DSPARB2);
1233
1234 dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) |
1235 VLV_FIFO(SPRITEF, 0xff));
1236 dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) |
1237 VLV_FIFO(SPRITEF, sprite1_start));
1238
1239 dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) |
1240 VLV_FIFO(SPRITEF_HI, 0xff));
1241 dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) |
1242 VLV_FIFO(SPRITEF_HI, sprite1_start >> 8));
1243
1244 I915_WRITE(DSPARB3, dsparb3);
1245 I915_WRITE(DSPARB2, dsparb2);
1246 break;
1247 default:
1248 break;
1249 }
1250}
1251
1252#undef VLV_FIFO
1253
262cd2e1
VS		 1254static void vlv_merge_wm(struct drm_device *dev,
1255 struct vlv_wm_values *wm)
1256{
1257 struct intel_crtc *crtc;
1258 int num_active_crtcs = 0;
1259
58590c14 1260 wm->level = to_i915(dev)->wm.max_level;
262cd2e1
VS		 1261 wm->cxsr = true;
1262
1263 for_each_intel_crtc(dev, crtc) {
1264 const struct vlv_wm_state *wm_state = &crtc->wm_state;
1265
1266 if (!crtc->active)
1267 continue;
1268
1269 if (!wm_state->cxsr)
1270 wm->cxsr = false;
1271
1272 num_active_crtcs++;
1273 wm->level = min_t(int, wm->level, wm_state->num_levels - 1);
1274 }
1275
1276 if (num_active_crtcs != 1)
1277 wm->cxsr = false;
1278
6f9c784b
VS		 1279 if (num_active_crtcs > 1)
1280 wm->level = VLV_WM_LEVEL_PM2;
1281
262cd2e1
VS		 1282 for_each_intel_crtc(dev, crtc) {
1283 struct vlv_wm_state *wm_state = &crtc->wm_state;
1284 enum pipe pipe = crtc->pipe;
1285
1286 if (!crtc->active)
1287 continue;
1288
1289 wm->pipe[pipe] = wm_state->wm[wm->level];
1290 if (wm->cxsr)
1291 wm->sr = wm_state->sr[wm->level];
1292
1293 wm->ddl[pipe].primary = DDL_PRECISION_HIGH | 2;
1294 wm->ddl[pipe].sprite[0] = DDL_PRECISION_HIGH | 2;
1295 wm->ddl[pipe].sprite[1] = DDL_PRECISION_HIGH | 2;
1296 wm->ddl[pipe].cursor = DDL_PRECISION_HIGH | 2;
1297 }
1298}
1299
1300static void vlv_update_wm(struct drm_crtc *crtc)
1301{
1302 struct drm_device *dev = crtc->dev;
1303 struct drm_i915_private *dev_priv = dev->dev_private;
1304 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
1305 enum pipe pipe = intel_crtc->pipe;
1306 struct vlv_wm_values wm = {};
1307
26e1fe4f 1308 vlv_compute_wm(intel_crtc);
262cd2e1
VS		 1309 vlv_merge_wm(dev, &wm);
1310
54f1b6e1
VS		 1311 if (memcmp(&dev_priv->wm.vlv, &wm, sizeof(wm)) == 0) {
1312 /* FIXME should be part of crtc atomic commit */
1313 vlv_pipe_set_fifo_size(intel_crtc);
262cd2e1 1314 return;
54f1b6e1 1315 }
262cd2e1
VS		 1316
1317 if (wm.level < VLV_WM_LEVEL_DDR_DVFS &&
1318 dev_priv->wm.vlv.level >= VLV_WM_LEVEL_DDR_DVFS)
1319 chv_set_memory_dvfs(dev_priv, false);
1320
1321 if (wm.level < VLV_WM_LEVEL_PM5 &&
1322 dev_priv->wm.vlv.level >= VLV_WM_LEVEL_PM5)
1323 chv_set_memory_pm5(dev_priv, false);
1324
852eb00d 1325 if (!wm.cxsr && dev_priv->wm.vlv.cxsr)
262cd2e1 1326 intel_set_memory_cxsr(dev_priv, false);
262cd2e1 1327
54f1b6e1
VS		 1328 /* FIXME should be part of crtc atomic commit */
1329 vlv_pipe_set_fifo_size(intel_crtc);
1330
262cd2e1
VS		 1331 vlv_write_wm_values(intel_crtc, &wm);
1332
1333 DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, "
1334 "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n",
1335 pipe_name(pipe), wm.pipe[pipe].primary, wm.pipe[pipe].cursor,
1336 wm.pipe[pipe].sprite[0], wm.pipe[pipe].sprite[1],
1337 wm.sr.plane, wm.sr.cursor, wm.level, wm.cxsr);
1338
852eb00d 1339 if (wm.cxsr && !dev_priv->wm.vlv.cxsr)
262cd2e1 1340 intel_set_memory_cxsr(dev_priv, true);
262cd2e1
VS		 1341
1342 if (wm.level >= VLV_WM_LEVEL_PM5 &&
1343 dev_priv->wm.vlv.level < VLV_WM_LEVEL_PM5)
1344 chv_set_memory_pm5(dev_priv, true);
1345
1346 if (wm.level >= VLV_WM_LEVEL_DDR_DVFS &&
1347 dev_priv->wm.vlv.level < VLV_WM_LEVEL_DDR_DVFS)
1348 chv_set_memory_dvfs(dev_priv, true);
1349
1350 dev_priv->wm.vlv = wm;
3c2777fd
VS		 1351}
1352
ae80152d
VS		 1353#define single_plane_enabled(mask) is_power_of_2(mask)
1354
46ba614c 1355static void g4x_update_wm(struct drm_crtc *crtc)
b445e3b0 1356{
46ba614c 1357 struct drm_device *dev = crtc->dev;
b445e3b0
ED		 1358 static const int sr_latency_ns = 12000;
1359 struct drm_i915_private *dev_priv = dev->dev_private;
1360 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1361 int plane_sr, cursor_sr;
1362 unsigned int enabled = 0;
9858425c 1363 bool cxsr_enabled;
b445e3b0 1364
51cea1f4 1365 if (g4x_compute_wm0(dev, PIPE_A,
5aef6003
CW		 1366 &g4x_wm_info, pessimal_latency_ns,
1367 &g4x_cursor_wm_info, pessimal_latency_ns,
b445e3b0 1368 &planea_wm, &cursora_wm))
51cea1f4 1369 enabled |= 1 << PIPE_A;
b445e3b0 1370
51cea1f4 1371 if (g4x_compute_wm0(dev, PIPE_B,
5aef6003
CW		 1372 &g4x_wm_info, pessimal_latency_ns,
1373 &g4x_cursor_wm_info, pessimal_latency_ns,
b445e3b0 1374 &planeb_wm, &cursorb_wm))
51cea1f4 1375 enabled |= 1 << PIPE_B;
b445e3b0 1376
b445e3b0
ED		 1377 if (single_plane_enabled(enabled) &&
1378 g4x_compute_srwm(dev, ffs(enabled) - 1,
1379 sr_latency_ns,
1380 &g4x_wm_info,
1381 &g4x_cursor_wm_info,
52bd02d8 1382 &plane_sr, &cursor_sr)) {
9858425c 1383 cxsr_enabled = true;
52bd02d8 1384 } else {
9858425c 1385 cxsr_enabled = false;
5209b1f4 1386 intel_set_memory_cxsr(dev_priv, false);
52bd02d8
CW		 1387 plane_sr = cursor_sr = 0;
1388 }
b445e3b0 1389
a5043453
VS		 1390 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, "
1391 "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
b445e3b0
ED		 1392 planea_wm, cursora_wm,
1393 planeb_wm, cursorb_wm,
1394 plane_sr, cursor_sr);
1395
1396 I915_WRITE(DSPFW1,
f4998963
VS		 1397 FW_WM(plane_sr, SR) |
1398 FW_WM(cursorb_wm, CURSORB) |
1399 FW_WM(planeb_wm, PLANEB) |
1400 FW_WM(planea_wm, PLANEA));
b445e3b0 1401 I915_WRITE(DSPFW2,
8c919b28 1402 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
f4998963 1403 FW_WM(cursora_wm, CURSORA));
b445e3b0
ED		 1404 /* HPLL off in SR has some issues on G4x... disable it */
1405 I915_WRITE(DSPFW3,
8c919b28 1406 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
f4998963 1407 FW_WM(cursor_sr, CURSOR_SR));
9858425c
ID		 1408
1409 if (cxsr_enabled)
1410 intel_set_memory_cxsr(dev_priv, true);
b445e3b0
ED		 1411}
1412
46ba614c 1413static void i965_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1414{
46ba614c 1415 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1416 struct drm_i915_private *dev_priv = dev->dev_private;
1417 struct drm_crtc *crtc;
1418 int srwm = 1;
1419 int cursor_sr = 16;
9858425c 1420 bool cxsr_enabled;
b445e3b0
ED		 1421
1422 /* Calc sr entries for one plane configs */
1423 crtc = single_enabled_crtc(dev);
1424 if (crtc) {
1425 /* self-refresh has much higher latency */
1426 static const int sr_latency_ns = 12000;
124abe07 1427 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1428 int clock = adjusted_mode->crtc_clock;
fec8cba3 1429 int htotal = adjusted_mode->crtc_htotal;
6e3c9717 1430 int hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
ac484963 1431 int cpp = drm_format_plane_cpp(crtc->primary->state->fb->pixel_format, 0);
b445e3b0
ED		 1432 unsigned long line_time_us;
1433 int entries;
1434
922044c9 1435 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1436
1437 /* Use ns/us then divide to preserve precision */
1438 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
ac484963 1439 cpp * hdisplay;
b445e3b0
ED		 1440 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1441 srwm = I965_FIFO_SIZE - entries;
1442 if (srwm < 0)
1443 srwm = 1;
1444 srwm &= 0x1ff;
1445 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1446 entries, srwm);
1447
1448 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
ac484963 1449 cpp * crtc->cursor->state->crtc_w;
b445e3b0
ED		 1450 entries = DIV_ROUND_UP(entries,
1451 i965_cursor_wm_info.cacheline_size);
1452 cursor_sr = i965_cursor_wm_info.fifo_size -
1453 (entries + i965_cursor_wm_info.guard_size);
1454
1455 if (cursor_sr > i965_cursor_wm_info.max_wm)
1456 cursor_sr = i965_cursor_wm_info.max_wm;
1457
1458 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1459 "cursor %d\n", srwm, cursor_sr);
1460
9858425c 1461 cxsr_enabled = true;
b445e3b0 1462 } else {
9858425c 1463 cxsr_enabled = false;
b445e3b0 1464 /* Turn off self refresh if both pipes are enabled */
5209b1f4 1465 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED		 1466 }
1467
1468 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1469 srwm);
1470
1471 /* 965 has limitations... */
f4998963
VS		 1472 I915_WRITE(DSPFW1, FW_WM(srwm, SR) |
1473 FW_WM(8, CURSORB) |
1474 FW_WM(8, PLANEB) |
1475 FW_WM(8, PLANEA));
1476 I915_WRITE(DSPFW2, FW_WM(8, CURSORA) |
1477 FW_WM(8, PLANEC_OLD));
b445e3b0 1478 /* update cursor SR watermark */
f4998963 1479 I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR));
9858425c
ID		 1480
1481 if (cxsr_enabled)
1482 intel_set_memory_cxsr(dev_priv, true);
b445e3b0
ED		 1483}
1484
f4998963
VS		 1485#undef FW_WM
1486
46ba614c 1487static void i9xx_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1488{
46ba614c 1489 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1490 struct drm_i915_private *dev_priv = dev->dev_private;
1491 const struct intel_watermark_params *wm_info;
1492 uint32_t fwater_lo;
1493 uint32_t fwater_hi;
1494 int cwm, srwm = 1;
1495 int fifo_size;
1496 int planea_wm, planeb_wm;
1497 struct drm_crtc *crtc, *enabled = NULL;
1498
1499 if (IS_I945GM(dev))
1500 wm_info = &i945_wm_info;
1501 else if (!IS_GEN2(dev))
1502 wm_info = &i915_wm_info;
1503 else
9d539105 1504 wm_info = &i830_a_wm_info;
b445e3b0
ED		 1505
1506 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1507 crtc = intel_get_crtc_for_plane(dev, 0);
3490ea5d 1508 if (intel_crtc_active(crtc)) {
241bfc38 1509 const struct drm_display_mode *adjusted_mode;
ac484963 1510 int cpp = drm_format_plane_cpp(crtc->primary->state->fb->pixel_format, 0);
b9e0bda3
CW		 1511 if (IS_GEN2(dev))
1512 cpp = 4;
1513
6e3c9717 1514 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1515 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1516 wm_info, fifo_size, cpp,
5aef6003 1517 pessimal_latency_ns);
b445e3b0 1518 enabled = crtc;
9d539105 1519 } else {
b445e3b0 1520 planea_wm = fifo_size - wm_info->guard_size;
9d539105
VS		 1521 if (planea_wm > (long)wm_info->max_wm)
1522 planea_wm = wm_info->max_wm;
1523 }
1524
1525 if (IS_GEN2(dev))
1526 wm_info = &i830_bc_wm_info;
b445e3b0
ED		 1527
1528 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1529 crtc = intel_get_crtc_for_plane(dev, 1);
3490ea5d 1530 if (intel_crtc_active(crtc)) {
241bfc38 1531 const struct drm_display_mode *adjusted_mode;
ac484963 1532 int cpp = drm_format_plane_cpp(crtc->primary->state->fb->pixel_format, 0);
b9e0bda3
CW		 1533 if (IS_GEN2(dev))
1534 cpp = 4;
1535
6e3c9717 1536 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1537 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1538 wm_info, fifo_size, cpp,
5aef6003 1539 pessimal_latency_ns);
b445e3b0
ED		 1540 if (enabled == NULL)
1541 enabled = crtc;
1542 else
1543 enabled = NULL;
9d539105 1544 } else {
b445e3b0 1545 planeb_wm = fifo_size - wm_info->guard_size;
9d539105
VS		 1546 if (planeb_wm > (long)wm_info->max_wm)
1547 planeb_wm = wm_info->max_wm;
1548 }
b445e3b0
ED		 1549
1550 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1551
2ab1bc9d 1552 if (IS_I915GM(dev) && enabled) {
2ff8fde1 1553 struct drm_i915_gem_object *obj;
2ab1bc9d 1554
59bea882 1555 obj = intel_fb_obj(enabled->primary->state->fb);
2ab1bc9d
DV		 1556
1557 /* self-refresh seems busted with untiled */
2ff8fde1 1558 if (obj->tiling_mode == I915_TILING_NONE)
2ab1bc9d
DV		 1559 enabled = NULL;
1560 }
1561
b445e3b0
ED		 1562 /*
1563 * Overlay gets an aggressive default since video jitter is bad.
1564 */
1565 cwm = 2;
1566
1567 /* Play safe and disable self-refresh before adjusting watermarks. */
5209b1f4 1568 intel_set_memory_cxsr(dev_priv, false);
b445e3b0
ED		 1569
1570 /* Calc sr entries for one plane configs */
1571 if (HAS_FW_BLC(dev) && enabled) {
1572 /* self-refresh has much higher latency */
1573 static const int sr_latency_ns = 6000;
124abe07 1574 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(enabled)->config->base.adjusted_mode;
241bfc38 1575 int clock = adjusted_mode->crtc_clock;
fec8cba3 1576 int htotal = adjusted_mode->crtc_htotal;
6e3c9717 1577 int hdisplay = to_intel_crtc(enabled)->config->pipe_src_w;
ac484963 1578 int cpp = drm_format_plane_cpp(enabled->primary->state->fb->pixel_format, 0);
b445e3b0
ED		 1579 unsigned long line_time_us;
1580 int entries;
1581
922044c9 1582 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1583
1584 /* Use ns/us then divide to preserve precision */
1585 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
ac484963 1586 cpp * hdisplay;
b445e3b0
ED		 1587 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1588 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1589 srwm = wm_info->fifo_size - entries;
1590 if (srwm < 0)
1591 srwm = 1;
1592
1593 if (IS_I945G(dev) || IS_I945GM(dev))
1594 I915_WRITE(FW_BLC_SELF,
1595 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1596 else if (IS_I915GM(dev))
1597 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1598 }
1599
1600 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1601 planea_wm, planeb_wm, cwm, srwm);
1602
1603 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1604 fwater_hi = (cwm & 0x1f);
1605
1606 /* Set request length to 8 cachelines per fetch */
1607 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1608 fwater_hi = fwater_hi | (1 << 8);
1609
1610 I915_WRITE(FW_BLC, fwater_lo);
1611 I915_WRITE(FW_BLC2, fwater_hi);
1612
5209b1f4
ID		 1613 if (enabled)
1614 intel_set_memory_cxsr(dev_priv, true);
b445e3b0
ED		 1615}
1616
feb56b93 1617static void i845_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1618{
46ba614c 1619 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1620 struct drm_i915_private *dev_priv = dev->dev_private;
1621 struct drm_crtc *crtc;
241bfc38 1622 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1623 uint32_t fwater_lo;
1624 int planea_wm;
1625
1626 crtc = single_enabled_crtc(dev);
1627 if (crtc == NULL)
1628 return;
1629
6e3c9717 1630 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
241bfc38 1631 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
feb56b93 1632 &i845_wm_info,
b445e3b0 1633 dev_priv->display.get_fifo_size(dev, 0),
5aef6003 1634 4, pessimal_latency_ns);
b445e3b0
ED		 1635 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1636 fwater_lo |= (3<<8) | planea_wm;
1637
1638 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1639
1640 I915_WRITE(FW_BLC, fwater_lo);
1641}
1642
8cfb3407 1643uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state *pipe_config)
801bcfff 1644{
fd4daa9c 1645 uint32_t pixel_rate;
801bcfff 1646
8cfb3407 1647 pixel_rate = pipe_config->base.adjusted_mode.crtc_clock;
801bcfff
PZ		 1648
1649 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1650 * adjust the pixel_rate here. */
1651
8cfb3407 1652 if (pipe_config->pch_pfit.enabled) {
801bcfff 1653 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
8cfb3407
VS		 1654 uint32_t pfit_size = pipe_config->pch_pfit.size;
1655
1656 pipe_w = pipe_config->pipe_src_w;
1657 pipe_h = pipe_config->pipe_src_h;
801bcfff 1658
801bcfff
PZ		 1659 pfit_w = (pfit_size >> 16) & 0xFFFF;
1660 pfit_h = pfit_size & 0xFFFF;
1661 if (pipe_w < pfit_w)
1662 pipe_w = pfit_w;
1663 if (pipe_h < pfit_h)
1664 pipe_h = pfit_h;
1665
15126882
MR		 1666 if (WARN_ON(!pfit_w || !pfit_h))
1667 return pixel_rate;
1668
801bcfff
PZ		 1669 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
1670 pfit_w * pfit_h);
1671 }
1672
1673 return pixel_rate;
1674}
1675
37126462 1676/* latency must be in 0.1us units. */
ac484963 1677static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t cpp, uint32_t latency)
801bcfff
PZ		 1678{
1679 uint64_t ret;
1680
3312ba65
VS		 1681 if (WARN(latency == 0, "Latency value missing\n"))
1682 return UINT_MAX;
1683
ac484963 1684 ret = (uint64_t) pixel_rate * cpp * latency;
801bcfff
PZ		 1685 ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
1686
1687 return ret;
1688}
1689
37126462 1690/* latency must be in 0.1us units. */
23297044 1691static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
ac484963 1692 uint32_t horiz_pixels, uint8_t cpp,
801bcfff
PZ		 1693 uint32_t latency)
1694{
1695 uint32_t ret;
1696
3312ba65
VS		 1697 if (WARN(latency == 0, "Latency value missing\n"))
1698 return UINT_MAX;
15126882
MR		 1699 if (WARN_ON(!pipe_htotal))
1700 return UINT_MAX;
3312ba65 1701
801bcfff 1702 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
ac484963 1703 ret = (ret + 1) * horiz_pixels * cpp;
801bcfff
PZ		 1704 ret = DIV_ROUND_UP(ret, 64) + 2;
1705 return ret;
1706}
1707
23297044 1708static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
ac484963 1709 uint8_t cpp)
cca32e9a 1710{
15126882
MR		 1711 /*
1712 * Neither of these should be possible since this function shouldn't be
1713 * called if the CRTC is off or the plane is invisible. But let's be
1714 * extra paranoid to avoid a potential divide-by-zero if we screw up
1715 * elsewhere in the driver.
1716 */
ac484963 1717 if (WARN_ON(!cpp))
15126882
MR		 1718 return 0;
1719 if (WARN_ON(!horiz_pixels))
1720 return 0;
1721
ac484963 1722 return DIV_ROUND_UP(pri_val * 64, horiz_pixels * cpp) + 2;
cca32e9a
PZ		 1723}
1724
820c1980 1725struct ilk_wm_maximums {
cca32e9a
PZ		 1726 uint16_t pri;
1727 uint16_t spr;
1728 uint16_t cur;
1729 uint16_t fbc;
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{
ac484963
VS		 1741 int cpp = pstate->base.fb ?
1742 drm_format_plane_cpp(pstate->base.fb->pixel_format, 0) : 0;
cca32e9a
PZ		 1743 uint32_t method1, method2;
1744
7221fc33 1745 if (!cstate->base.active || !pstate->visible)
801bcfff
PZ		 1746 return 0;
1747
ac484963 1748 method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), cpp, mem_value);
cca32e9a
PZ		 1749
1750 if (!is_lp)
1751 return method1;
1752
7221fc33
MR		 1753 method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
1754 cstate->base.adjusted_mode.crtc_htotal,
43d59eda 1755 drm_rect_width(&pstate->dst),
ac484963 1756 cpp, mem_value);
cca32e9a
PZ		 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{
ac484963
VS		 1769 int cpp = pstate->base.fb ?
1770 drm_format_plane_cpp(pstate->base.fb->pixel_format, 0) : 0;
801bcfff
PZ		 1771 uint32_t method1, method2;
1772
7221fc33 1773 if (!cstate->base.active || !pstate->visible)
801bcfff
PZ		 1774 return 0;
1775
ac484963 1776 method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), cpp, mem_value);
7221fc33
MR		 1777 method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
1778 cstate->base.adjusted_mode.crtc_htotal,
43d59eda 1779 drm_rect_width(&pstate->dst),
ac484963 1780 cpp, mem_value);
801bcfff
PZ		 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{
b2435692
MR		 1792 /*
1793 * We treat the cursor plane as always-on for the purposes of watermark
1794 * calculation. Until we have two-stage watermark programming merged,
1795 * this is necessary to avoid flickering.
1796 */
1797 int cpp = 4;
1798 int width = pstate->visible ? pstate->base.crtc_w : 64;
43d59eda 1799
b2435692 1800 if (!cstate->base.active)
801bcfff
PZ		 1801 return 0;
1802
7221fc33
MR		 1803 return ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
1804 cstate->base.adjusted_mode.crtc_htotal,
b2435692 1805 width, cpp, mem_value);
801bcfff
PZ		 1806}
1807
cca32e9a 1808/* Only for WM_LP. */
7221fc33 1809static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate,
43d59eda 1810 const struct intel_plane_state *pstate,
1fda9882 1811 uint32_t pri_val)
cca32e9a 1812{
ac484963
VS		 1813 int cpp = pstate->base.fb ?
1814 drm_format_plane_cpp(pstate->base.fb->pixel_format, 0) : 0;
43d59eda 1815
7221fc33 1816 if (!cstate->base.active || !pstate->visible)
cca32e9a
PZ		 1817 return 0;
1818
ac484963 1819 return ilk_wm_fbc(pri_val, drm_rect_width(&pstate->dst), cpp);
cca32e9a
PZ		 1820}
1821
158ae64f
VS		 1822static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
1823{
416f4727
VS		 1824 if (INTEL_INFO(dev)->gen >= 8)
1825 return 3072;
1826 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS		 1827 return 768;
1828 else
1829 return 512;
1830}
1831
4e975081
VS		 1832static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev,
1833 int level, bool is_sprite)
1834{
1835 if (INTEL_INFO(dev)->gen >= 8)
1836 /* BDW primary/sprite plane watermarks */
1837 return level == 0 ? 255 : 2047;
1838 else if (INTEL_INFO(dev)->gen >= 7)
1839 /* IVB/HSW primary/sprite plane watermarks */
1840 return level == 0 ? 127 : 1023;
1841 else if (!is_sprite)
1842 /* ILK/SNB primary plane watermarks */
1843 return level == 0 ? 127 : 511;
1844 else
1845 /* ILK/SNB sprite plane watermarks */
1846 return level == 0 ? 63 : 255;
1847}
1848
1849static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev,
1850 int level)
1851{
1852 if (INTEL_INFO(dev)->gen >= 7)
1853 return level == 0 ? 63 : 255;
1854 else
1855 return level == 0 ? 31 : 63;
1856}
1857
1858static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev)
1859{
1860 if (INTEL_INFO(dev)->gen >= 8)
1861 return 31;
1862 else
1863 return 15;
1864}
1865
158ae64f
VS		 1866/* Calculate the maximum primary/sprite plane watermark */
1867static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
1868 int level,
240264f4 1869 const struct intel_wm_config *config,
158ae64f
VS		 1870 enum intel_ddb_partitioning ddb_partitioning,
1871 bool is_sprite)
1872{
1873 unsigned int fifo_size = ilk_display_fifo_size(dev);
158ae64f
VS		 1874
1875 /* if sprites aren't enabled, sprites get nothing */
240264f4 1876 if (is_sprite && !config->sprites_enabled)
158ae64f
VS		 1877 return 0;
1878
1879 /* HSW allows LP1+ watermarks even with multiple pipes */
240264f4 1880 if (level == 0 || config->num_pipes_active > 1) {
158ae64f
VS		 1881 fifo_size /= INTEL_INFO(dev)->num_pipes;
1882
1883 /*
1884 * For some reason the non self refresh
1885 * FIFO size is only half of the self
1886 * refresh FIFO size on ILK/SNB.
1887 */
1888 if (INTEL_INFO(dev)->gen <= 6)
1889 fifo_size /= 2;
1890 }
1891
240264f4 1892 if (config->sprites_enabled) {
158ae64f
VS		 1893 /* level 0 is always calculated with 1:1 split */
1894 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
1895 if (is_sprite)
1896 fifo_size *= 5;
1897 fifo_size /= 6;
1898 } else {
1899 fifo_size /= 2;
1900 }
1901 }
1902
1903 /* clamp to max that the registers can hold */
4e975081 1904 return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite));
158ae64f
VS		 1905}
1906
1907/* Calculate the maximum cursor plane watermark */
1908static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
240264f4
VS		 1909 int level,
1910 const struct intel_wm_config *config)
158ae64f
VS		 1911{
1912 /* HSW LP1+ watermarks w/ multiple pipes */
240264f4 1913 if (level > 0 && config->num_pipes_active > 1)
158ae64f
VS		 1914 return 64;
1915
1916 /* otherwise just report max that registers can hold */
4e975081 1917 return ilk_cursor_wm_reg_max(dev, level);
158ae64f
VS		 1918}
1919
d34ff9c6 1920static void ilk_compute_wm_maximums(const struct drm_device *dev,
34982fe1
VS		 1921 int level,
1922 const struct intel_wm_config *config,
1923 enum intel_ddb_partitioning ddb_partitioning,
820c1980 1924 struct ilk_wm_maximums *max)
158ae64f 1925{
240264f4
VS		 1926 max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
1927 max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
1928 max->cur = ilk_cursor_wm_max(dev, level, config);
4e975081 1929 max->fbc = ilk_fbc_wm_reg_max(dev);
158ae64f
VS		 1930}
1931
a3cb4048
VS		 1932static void ilk_compute_wm_reg_maximums(struct drm_device *dev,
1933 int level,
1934 struct ilk_wm_maximums *max)
1935{
1936 max->pri = ilk_plane_wm_reg_max(dev, level, false);
1937 max->spr = ilk_plane_wm_reg_max(dev, level, true);
1938 max->cur = ilk_cursor_wm_reg_max(dev, level);
1939 max->fbc = ilk_fbc_wm_reg_max(dev);
1940}
1941
d9395655 1942static bool ilk_validate_wm_level(int level,
820c1980 1943 const struct ilk_wm_maximums *max,
d9395655 1944 struct intel_wm_level *result)
a9786a11
VS		 1945{
1946 bool ret;
1947
1948 /* already determined to be invalid? */
1949 if (!result->enable)
1950 return false;
1951
1952 result->enable = result->pri_val <= max->pri &&
1953 result->spr_val <= max->spr &&
1954 result->cur_val <= max->cur;
1955
1956 ret = result->enable;
1957
1958 /*
1959 * HACK until we can pre-compute everything,
1960 * and thus fail gracefully if LP0 watermarks
1961 * are exceeded...
1962 */
1963 if (level == 0 && !result->enable) {
1964 if (result->pri_val > max->pri)
1965 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1966 level, result->pri_val, max->pri);
1967 if (result->spr_val > max->spr)
1968 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1969 level, result->spr_val, max->spr);
1970 if (result->cur_val > max->cur)
1971 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
1972 level, result->cur_val, max->cur);
1973
1974 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
1975 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
1976 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
1977 result->enable = true;
1978 }
1979
a9786a11
VS		 1980 return ret;
1981}
1982
d34ff9c6 1983static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
43d59eda 1984 const struct intel_crtc *intel_crtc,
6f5ddd17 1985 int level,
7221fc33 1986 struct intel_crtc_state *cstate,
86c8bbbe
MR		 1987 struct intel_plane_state *pristate,
1988 struct intel_plane_state *sprstate,
1989 struct intel_plane_state *curstate,
1fd527cc 1990 struct intel_wm_level *result)
6f5ddd17
VS		 1991{
1992 uint16_t pri_latency = dev_priv->wm.pri_latency[level];
1993 uint16_t spr_latency = dev_priv->wm.spr_latency[level];
1994 uint16_t cur_latency = dev_priv->wm.cur_latency[level];
1995
1996 /* WM1+ latency values stored in 0.5us units */
1997 if (level > 0) {
1998 pri_latency *= 5;
1999 spr_latency *= 5;
2000 cur_latency *= 5;
2001 }
2002
e3bddded
ML		 2003 if (pristate) {
2004 result->pri_val = ilk_compute_pri_wm(cstate, pristate,
2005 pri_latency, level);
2006 result->fbc_val = ilk_compute_fbc_wm(cstate, pristate, result->pri_val);
2007 }
2008
2009 if (sprstate)
2010 result->spr_val = ilk_compute_spr_wm(cstate, sprstate, spr_latency);
2011
2012 if (curstate)
2013 result->cur_val = ilk_compute_cur_wm(cstate, curstate, cur_latency);
2014
6f5ddd17
VS		 2015 result->enable = true;
2016}
2017
801bcfff 2018static uint32_t
532f7a7f 2019hsw_compute_linetime_wm(const struct intel_crtc_state *cstate)
1f8eeabf 2020{
532f7a7f
VS		 2021 const struct intel_atomic_state *intel_state =
2022 to_intel_atomic_state(cstate->base.state);
ee91a159
MR		 2023 const struct drm_display_mode *adjusted_mode =
2024 &cstate->base.adjusted_mode;
85a02deb 2025 u32 linetime, ips_linetime;
1f8eeabf 2026
ee91a159
MR		 2027 if (!cstate->base.active)
2028 return 0;
2029 if (WARN_ON(adjusted_mode->crtc_clock == 0))
2030 return 0;
532f7a7f 2031 if (WARN_ON(intel_state->cdclk == 0))
801bcfff 2032 return 0;
1011d8c4 2033
1f8eeabf
ED		 2034 /* The WM are computed with base on how long it takes to fill a single
2035 * row at the given clock rate, multiplied by 8.
2036 * */
124abe07
VS		 2037 linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
2038 adjusted_mode->crtc_clock);
2039 ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
532f7a7f 2040 intel_state->cdclk);
1f8eeabf 2041
801bcfff
PZ		 2042 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
2043 PIPE_WM_LINETIME_TIME(linetime);
1f8eeabf
ED		 2044}
2045
2af30a5c 2046static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[8])
12b134df
VS		 2047{
2048 struct drm_i915_private *dev_priv = dev->dev_private;
2049
2af30a5c
PB		 2050 if (IS_GEN9(dev)) {
2051 uint32_t val;
4f947386 2052 int ret, i;
367294be 2053 int level, max_level = ilk_wm_max_level(dev);
2af30a5c
PB		 2054
2055 /* read the first set of memory latencies[0:3] */
2056 val = 0; /* data0 to be programmed to 0 for first set */
2057 mutex_lock(&dev_priv->rps.hw_lock);
2058 ret = sandybridge_pcode_read(dev_priv,
2059 GEN9_PCODE_READ_MEM_LATENCY,
2060 &val);
2061 mutex_unlock(&dev_priv->rps.hw_lock);
2062
2063 if (ret) {
2064 DRM_ERROR("SKL Mailbox read error = %d\n", ret);
2065 return;
2066 }
2067
2068 wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
2069 wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
2070 GEN9_MEM_LATENCY_LEVEL_MASK;
2071 wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
2072 GEN9_MEM_LATENCY_LEVEL_MASK;
2073 wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
2074 GEN9_MEM_LATENCY_LEVEL_MASK;
2075
2076 /* read the second set of memory latencies[4:7] */
2077 val = 1; /* data0 to be programmed to 1 for second set */
2078 mutex_lock(&dev_priv->rps.hw_lock);
2079 ret = sandybridge_pcode_read(dev_priv,
2080 GEN9_PCODE_READ_MEM_LATENCY,
2081 &val);
2082 mutex_unlock(&dev_priv->rps.hw_lock);
2083 if (ret) {
2084 DRM_ERROR("SKL Mailbox read error = %d\n", ret);
2085 return;
2086 }
2087
2088 wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
2089 wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
2090 GEN9_MEM_LATENCY_LEVEL_MASK;
2091 wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
2092 GEN9_MEM_LATENCY_LEVEL_MASK;
2093 wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
2094 GEN9_MEM_LATENCY_LEVEL_MASK;
2095
367294be 2096 /*
6f97235b
DL		 2097 * WaWmMemoryReadLatency:skl
2098 *
367294be
VK		 2099 * punit doesn't take into account the read latency so we need
2100 * to add 2us to the various latency levels we retrieve from
2101 * the punit.
2102 * - W0 is a bit special in that it's the only level that
2103 * can't be disabled if we want to have display working, so
2104 * we always add 2us there.
2105 * - For levels >=1, punit returns 0us latency when they are
2106 * disabled, so we respect that and don't add 2us then
4f947386
VK		 2107 *
2108 * Additionally, if a level n (n > 1) has a 0us latency, all
2109 * levels m (m >= n) need to be disabled. We make sure to
2110 * sanitize the values out of the punit to satisfy this
2111 * requirement.
367294be
VK		 2112 */
2113 wm[0] += 2;
2114 for (level = 1; level <= max_level; level++)
2115 if (wm[level] != 0)
2116 wm[level] += 2;
4f947386
VK		 2117 else {
2118 for (i = level + 1; i <= max_level; i++)
2119 wm[i] = 0;
367294be 2120
4f947386
VK		 2121 break;
2122 }
2af30a5c 2123 } else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
12b134df
VS		 2124 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2125
2126 wm[0] = (sskpd >> 56) & 0xFF;
2127 if (wm[0] == 0)
2128 wm[0] = sskpd & 0xF;
e5d5019e
VS		 2129 wm[1] = (sskpd >> 4) & 0xFF;
2130 wm[2] = (sskpd >> 12) & 0xFF;
2131 wm[3] = (sskpd >> 20) & 0x1FF;
2132 wm[4] = (sskpd >> 32) & 0x1FF;
63cf9a13
VS		 2133 } else if (INTEL_INFO(dev)->gen >= 6) {
2134 uint32_t sskpd = I915_READ(MCH_SSKPD);
2135
2136 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2137 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2138 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2139 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
3a88d0ac
VS		 2140 } else if (INTEL_INFO(dev)->gen >= 5) {
2141 uint32_t mltr = I915_READ(MLTR_ILK);
2142
2143 /* ILK primary LP0 latency is 700 ns */
2144 wm[0] = 7;
2145 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2146 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
12b134df
VS		 2147 }
2148}
2149
53615a5e
VS		 2150static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2151{
2152 /* ILK sprite LP0 latency is 1300 ns */
7e22dbbb 2153 if (IS_GEN5(dev))
53615a5e
VS		 2154 wm[0] = 13;
2155}
2156
2157static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2158{
2159 /* ILK cursor LP0 latency is 1300 ns */
7e22dbbb 2160 if (IS_GEN5(dev))
53615a5e
VS		 2161 wm[0] = 13;
2162
2163 /* WaDoubleCursorLP3Latency:ivb */
2164 if (IS_IVYBRIDGE(dev))
2165 wm[3] *= 2;
2166}
2167
546c81fd 2168int ilk_wm_max_level(const struct drm_device *dev)
26ec971e 2169{
26ec971e 2170 /* how many WM levels are we expecting */
b6e742f6 2171 if (INTEL_INFO(dev)->gen >= 9)
2af30a5c
PB		 2172 return 7;
2173 else if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ad0d6dc4 2174 return 4;
26ec971e 2175 else if (INTEL_INFO(dev)->gen >= 6)
ad0d6dc4 2176 return 3;
26ec971e 2177 else
ad0d6dc4
VS		 2178 return 2;
2179}
7526ed79 2180
ad0d6dc4
VS		 2181static void intel_print_wm_latency(struct drm_device *dev,
2182 const char *name,
2af30a5c 2183 const uint16_t wm[8])
ad0d6dc4
VS		 2184{
2185 int level, max_level = ilk_wm_max_level(dev);
26ec971e
VS		 2186
2187 for (level = 0; level <= max_level; level++) {
2188 unsigned int latency = wm[level];
2189
2190 if (latency == 0) {
2191 DRM_ERROR("%s WM%d latency not provided\n",
2192 name, level);
2193 continue;
2194 }
2195
2af30a5c
PB		 2196 /*
2197 * - latencies are in us on gen9.
2198 * - before then, WM1+ latency values are in 0.5us units
2199 */
2200 if (IS_GEN9(dev))
2201 latency *= 10;
2202 else if (level > 0)
26ec971e
VS		 2203 latency *= 5;
2204
2205 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2206 name, level, wm[level],
2207 latency / 10, latency % 10);
2208 }
2209}
2210
e95a2f75
VS		 2211static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
2212 uint16_t wm[5], uint16_t min)
2213{
2214 int level, max_level = ilk_wm_max_level(dev_priv->dev);
2215
2216 if (wm[0] >= min)
2217 return false;
2218
2219 wm[0] = max(wm[0], min);
2220 for (level = 1; level <= max_level; level++)
2221 wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
2222
2223 return true;
2224}
2225
2226static void snb_wm_latency_quirk(struct drm_device *dev)
2227{
2228 struct drm_i915_private *dev_priv = dev->dev_private;
2229 bool changed;
2230
2231 /*
2232 * The BIOS provided WM memory latency values are often
2233 * inadequate for high resolution displays. Adjust them.
2234 */
2235 changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
2236 ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
2237 ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
2238
2239 if (!changed)
2240 return;
2241
2242 DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
2243 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2244 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2245 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
2246}
2247
fa50ad61 2248static void ilk_setup_wm_latency(struct drm_device *dev)
53615a5e
VS		 2249{
2250 struct drm_i915_private *dev_priv = dev->dev_private;
2251
2252 intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2253
2254 memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2255 sizeof(dev_priv->wm.pri_latency));
2256 memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2257 sizeof(dev_priv->wm.pri_latency));
2258
2259 intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2260 intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
26ec971e
VS		 2261
2262 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2263 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2264 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
e95a2f75
VS		 2265
2266 if (IS_GEN6(dev))
2267 snb_wm_latency_quirk(dev);
53615a5e
VS		 2268}
2269
2af30a5c
PB		 2270static void skl_setup_wm_latency(struct drm_device *dev)
2271{
2272 struct drm_i915_private *dev_priv = dev->dev_private;
2273
2274 intel_read_wm_latency(dev, dev_priv->wm.skl_latency);
2275 intel_print_wm_latency(dev, "Gen9 Plane", dev_priv->wm.skl_latency);
2276}
2277
ed4a6a7c
MR		 2278static bool ilk_validate_pipe_wm(struct drm_device *dev,
2279 struct intel_pipe_wm *pipe_wm)
2280{
2281 /* LP0 watermark maximums depend on this pipe alone */
2282 const struct intel_wm_config config = {
2283 .num_pipes_active = 1,
2284 .sprites_enabled = pipe_wm->sprites_enabled,
2285 .sprites_scaled = pipe_wm->sprites_scaled,
2286 };
2287 struct ilk_wm_maximums max;
2288
2289 /* LP0 watermarks always use 1/2 DDB partitioning */
2290 ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
2291
2292 /* At least LP0 must be valid */
2293 if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) {
2294 DRM_DEBUG_KMS("LP0 watermark invalid\n");
2295 return false;
2296 }
2297
2298 return true;
2299}
2300
0b2ae6d7 2301/* Compute new watermarks for the pipe */
e3bddded 2302static int ilk_compute_pipe_wm(struct intel_crtc_state *cstate)
0b2ae6d7 2303{
e3bddded
ML		 2304 struct drm_atomic_state *state = cstate->base.state;
2305 struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc);
86c8bbbe 2306 struct intel_pipe_wm *pipe_wm;
e3bddded 2307 struct drm_device *dev = state->dev;
d34ff9c6 2308 const struct drm_i915_private *dev_priv = dev->dev_private;
43d59eda 2309 struct intel_plane *intel_plane;
86c8bbbe 2310 struct intel_plane_state *pristate = NULL;
43d59eda 2311 struct intel_plane_state *sprstate = NULL;
86c8bbbe 2312 struct intel_plane_state *curstate = NULL;
d81f04c5 2313 int level, max_level = ilk_wm_max_level(dev), usable_level;
820c1980 2314 struct ilk_wm_maximums max;
0b2ae6d7 2315
e8f1f02e 2316 pipe_wm = &cstate->wm.ilk.optimal;
86c8bbbe 2317
43d59eda 2318 for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) {
e3bddded
ML		 2319 struct intel_plane_state *ps;
2320
2321 ps = intel_atomic_get_existing_plane_state(state,
2322 intel_plane);
2323 if (!ps)
2324 continue;
86c8bbbe
MR		 2325
2326 if (intel_plane->base.type == DRM_PLANE_TYPE_PRIMARY)
e3bddded 2327 pristate = ps;
86c8bbbe 2328 else if (intel_plane->base.type == DRM_PLANE_TYPE_OVERLAY)
e3bddded 2329 sprstate = ps;
86c8bbbe 2330 else if (intel_plane->base.type == DRM_PLANE_TYPE_CURSOR)
e3bddded 2331 curstate = ps;
43d59eda
MR		 2332 }
2333
ed4a6a7c 2334 pipe_wm->pipe_enabled = cstate->base.active;
e3bddded
ML		 2335 if (sprstate) {
2336 pipe_wm->sprites_enabled = sprstate->visible;
2337 pipe_wm->sprites_scaled = sprstate->visible &&
2338 (drm_rect_width(&sprstate->dst) != drm_rect_width(&sprstate->src) >> 16 ||
2339 drm_rect_height(&sprstate->dst) != drm_rect_height(&sprstate->src) >> 16);
2340 }
2341
d81f04c5
ML		 2342 usable_level = max_level;
2343
7b39a0b7 2344 /* ILK/SNB: LP2+ watermarks only w/o sprites */
e3bddded 2345 if (INTEL_INFO(dev)->gen <= 6 && pipe_wm->sprites_enabled)
d81f04c5 2346 usable_level = 1;
7b39a0b7
VS		 2347
2348 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
ed4a6a7c 2349 if (pipe_wm->sprites_scaled)
d81f04c5 2350 usable_level = 0;
7b39a0b7 2351
86c8bbbe 2352 ilk_compute_wm_level(dev_priv, intel_crtc, 0, cstate,
71f0a626
ML		 2353 pristate, sprstate, curstate, &pipe_wm->raw_wm[0]);
2354
2355 memset(&pipe_wm->wm, 0, sizeof(pipe_wm->wm));
2356 pipe_wm->wm[0] = pipe_wm->raw_wm[0];
0b2ae6d7 2357
a42a5719 2358 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
532f7a7f 2359 pipe_wm->linetime = hsw_compute_linetime_wm(cstate);
0b2ae6d7 2360
ed4a6a7c 2361 if (!ilk_validate_pipe_wm(dev, pipe_wm))
1a426d61 2362 return -EINVAL;
a3cb4048
VS		 2363
2364 ilk_compute_wm_reg_maximums(dev, 1, &max);
2365
2366 for (level = 1; level <= max_level; level++) {
71f0a626 2367 struct intel_wm_level *wm = &pipe_wm->raw_wm[level];
a3cb4048 2368
86c8bbbe 2369 ilk_compute_wm_level(dev_priv, intel_crtc, level, cstate,
d81f04c5 2370 pristate, sprstate, curstate, wm);
a3cb4048
VS		 2371
2372 /*
2373 * Disable any watermark level that exceeds the
2374 * register maximums since such watermarks are
2375 * always invalid.
2376 */
71f0a626
ML		 2377 if (level > usable_level)
2378 continue;
2379
2380 if (ilk_validate_wm_level(level, &max, wm))
2381 pipe_wm->wm[level] = *wm;
2382 else
d81f04c5 2383 usable_level = level;
a3cb4048
VS		 2384 }
2385
86c8bbbe 2386 return 0;
0b2ae6d7
VS		 2387}
2388
ed4a6a7c
MR		 2389/*
2390 * Build a set of 'intermediate' watermark values that satisfy both the old
2391 * state and the new state. These can be programmed to the hardware
2392 * immediately.
2393 */
2394static int ilk_compute_intermediate_wm(struct drm_device *dev,
2395 struct intel_crtc *intel_crtc,
2396 struct intel_crtc_state *newstate)
2397{
e8f1f02e 2398 struct intel_pipe_wm *a = &newstate->wm.ilk.intermediate;
ed4a6a7c
MR		 2399 struct intel_pipe_wm *b = &intel_crtc->wm.active.ilk;
2400 int level, max_level = ilk_wm_max_level(dev);
2401
2402 /*
2403 * Start with the final, target watermarks, then combine with the
2404 * currently active watermarks to get values that are safe both before
2405 * and after the vblank.
2406 */
e8f1f02e 2407 *a = newstate->wm.ilk.optimal;
ed4a6a7c
MR		 2408 a->pipe_enabled |= b->pipe_enabled;
2409 a->sprites_enabled |= b->sprites_enabled;
2410 a->sprites_scaled |= b->sprites_scaled;
2411
2412 for (level = 0; level <= max_level; level++) {
2413 struct intel_wm_level *a_wm = &a->wm[level];
2414 const struct intel_wm_level *b_wm = &b->wm[level];
2415
2416 a_wm->enable &= b_wm->enable;
2417 a_wm->pri_val = max(a_wm->pri_val, b_wm->pri_val);
2418 a_wm->spr_val = max(a_wm->spr_val, b_wm->spr_val);
2419 a_wm->cur_val = max(a_wm->cur_val, b_wm->cur_val);
2420 a_wm->fbc_val = max(a_wm->fbc_val, b_wm->fbc_val);
2421 }
2422
2423 /*
2424 * We need to make sure that these merged watermark values are
2425 * actually a valid configuration themselves. If they're not,
2426 * there's no safe way to transition from the old state to
2427 * the new state, so we need to fail the atomic transaction.
2428 */
2429 if (!ilk_validate_pipe_wm(dev, a))
2430 return -EINVAL;
2431
2432 /*
2433 * If our intermediate WM are identical to the final WM, then we can
2434 * omit the post-vblank programming; only update if it's different.
2435 */
e8f1f02e 2436 if (memcmp(a, &newstate->wm.ilk.optimal, sizeof(*a)) == 0)
ed4a6a7c
MR		 2437 newstate->wm.need_postvbl_update = false;
2438
2439 return 0;
2440}
2441
0b2ae6d7
VS		 2442/*
2443 * Merge the watermarks from all active pipes for a specific level.
2444 */
2445static void ilk_merge_wm_level(struct drm_device *dev,
2446 int level,
2447 struct intel_wm_level *ret_wm)
2448{
2449 const struct intel_crtc *intel_crtc;
2450
d52fea5b
VS		 2451 ret_wm->enable = true;
2452
d3fcc808 2453 for_each_intel_crtc(dev, intel_crtc) {
ed4a6a7c 2454 const struct intel_pipe_wm *active = &intel_crtc->wm.active.ilk;
fe392efd
VS		 2455 const struct intel_wm_level *wm = &active->wm[level];
2456
2457 if (!active->pipe_enabled)
2458 continue;
0b2ae6d7 2459
d52fea5b
VS		 2460 /*
2461 * The watermark values may have been used in the past,
2462 * so we must maintain them in the registers for some
2463 * time even if the level is now disabled.
2464 */
0b2ae6d7 2465 if (!wm->enable)
d52fea5b 2466 ret_wm->enable = false;
0b2ae6d7
VS		 2467
2468 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2469 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2470 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2471 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2472 }
0b2ae6d7
VS		 2473}
2474
2475/*
2476 * Merge all low power watermarks for all active pipes.
2477 */
2478static void ilk_wm_merge(struct drm_device *dev,
0ba22e26 2479 const struct intel_wm_config *config,
820c1980 2480 const struct ilk_wm_maximums *max,
0b2ae6d7
VS		 2481 struct intel_pipe_wm *merged)
2482{
7733b49b 2483 struct drm_i915_private *dev_priv = dev->dev_private;
0b2ae6d7 2484 int level, max_level = ilk_wm_max_level(dev);
d52fea5b 2485 int last_enabled_level = max_level;
0b2ae6d7 2486
0ba22e26
VS		 2487 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2488 if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) &&
2489 config->num_pipes_active > 1)
1204d5ba 2490 last_enabled_level = 0;
0ba22e26 2491
6c8b6c28
VS		 2492 /* ILK: FBC WM must be disabled always */
2493 merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6;
0b2ae6d7
VS		 2494
2495 /* merge each WM1+ level */
2496 for (level = 1; level <= max_level; level++) {
2497 struct intel_wm_level *wm = &merged->wm[level];
2498
2499 ilk_merge_wm_level(dev, level, wm);
2500
d52fea5b
VS		 2501 if (level > last_enabled_level)
2502 wm->enable = false;
2503 else if (!ilk_validate_wm_level(level, max, wm))
2504 /* make sure all following levels get disabled */
2505 last_enabled_level = level - 1;
0b2ae6d7
VS		 2506
2507 /*
2508 * The spec says it is preferred to disable
2509 * FBC WMs instead of disabling a WM level.
2510 */
2511 if (wm->fbc_val > max->fbc) {
d52fea5b
VS		 2512 if (wm->enable)
2513 merged->fbc_wm_enabled = false;
0b2ae6d7
VS		 2514 wm->fbc_val = 0;
2515 }
2516 }
6c8b6c28
VS		 2517
2518 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2519 /*
2520 * FIXME this is racy. FBC might get enabled later.
2521 * What we should check here is whether FBC can be
2522 * enabled sometime later.
2523 */
7733b49b 2524 if (IS_GEN5(dev) && !merged->fbc_wm_enabled &&
0e631adc 2525 intel_fbc_is_active(dev_priv)) {
6c8b6c28
VS		 2526 for (level = 2; level <= max_level; level++) {
2527 struct intel_wm_level *wm = &merged->wm[level];
2528
2529 wm->enable = false;
2530 }
2531 }
0b2ae6d7
VS		 2532}
2533
b380ca3c
VS		 2534static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2535{
2536 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2537 return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2538}
2539
a68d68ee
VS		 2540/* The value we need to program into the WM_LPx latency field */
2541static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level)
2542{
2543 struct drm_i915_private *dev_priv = dev->dev_private;
2544
a42a5719 2545 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
a68d68ee
VS		 2546 return 2 * level;
2547 else
2548 return dev_priv->wm.pri_latency[level];
2549}
2550
820c1980 2551static void ilk_compute_wm_results(struct drm_device *dev,
0362c781 2552 const struct intel_pipe_wm *merged,
609cedef 2553 enum intel_ddb_partitioning partitioning,
820c1980 2554 struct ilk_wm_values *results)
801bcfff 2555{
0b2ae6d7
VS		 2556 struct intel_crtc *intel_crtc;
2557 int level, wm_lp;
cca32e9a 2558
0362c781 2559 results->enable_fbc_wm = merged->fbc_wm_enabled;
609cedef 2560 results->partitioning = partitioning;
cca32e9a 2561
0b2ae6d7 2562 /* LP1+ register values */
cca32e9a 2563 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
1fd527cc 2564 const struct intel_wm_level *r;
801bcfff 2565
b380ca3c 2566 level = ilk_wm_lp_to_level(wm_lp, merged);
0b2ae6d7 2567
0362c781 2568 r = &merged->wm[level];
cca32e9a 2569
d52fea5b
VS		 2570 /*
2571 * Maintain the watermark values even if the level is
2572 * disabled. Doing otherwise could cause underruns.
2573 */
2574 results->wm_lp[wm_lp - 1] =
a68d68ee 2575 (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) |
416f4727
VS		 2576 (r->pri_val << WM1_LP_SR_SHIFT) |
2577 r->cur_val;
2578
d52fea5b
VS		 2579 if (r->enable)
2580 results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN;
2581
416f4727
VS		 2582 if (INTEL_INFO(dev)->gen >= 8)
2583 results->wm_lp[wm_lp - 1] |=
2584 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
2585 else
2586 results->wm_lp[wm_lp - 1] |=
2587 r->fbc_val << WM1_LP_FBC_SHIFT;
2588
d52fea5b
VS		 2589 /*
2590 * Always set WM1S_LP_EN when spr_val != 0, even if the
2591 * level is disabled. Doing otherwise could cause underruns.
2592 */
6cef2b8a
VS		 2593 if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) {
2594 WARN_ON(wm_lp != 1);
2595 results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
2596 } else
2597 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
cca32e9a 2598 }
801bcfff 2599
0b2ae6d7 2600 /* LP0 register values */
d3fcc808 2601 for_each_intel_crtc(dev, intel_crtc) {
0b2ae6d7 2602 enum pipe pipe = intel_crtc->pipe;
ed4a6a7c
MR		 2603 const struct intel_wm_level *r =
2604 &intel_crtc->wm.active.ilk.wm[0];
0b2ae6d7
VS		 2605
2606 if (WARN_ON(!r->enable))
2607 continue;
2608
ed4a6a7c 2609 results->wm_linetime[pipe] = intel_crtc->wm.active.ilk.linetime;
1011d8c4 2610
0b2ae6d7
VS		 2611 results->wm_pipe[pipe] =
2612 (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2613 (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2614 r->cur_val;
801bcfff
PZ		 2615 }
2616}
2617
861f3389
PZ		 2618/* Find the result with the highest level enabled. Check for enable_fbc_wm in
2619 * case both are at the same level. Prefer r1 in case they're the same. */
820c1980 2620static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev,
198a1e9b
VS		 2621 struct intel_pipe_wm *r1,
2622 struct intel_pipe_wm *r2)
861f3389 2623{
198a1e9b
VS		 2624 int level, max_level = ilk_wm_max_level(dev);
2625 int level1 = 0, level2 = 0;
861f3389 2626
198a1e9b
VS		 2627 for (level = 1; level <= max_level; level++) {
2628 if (r1->wm[level].enable)
2629 level1 = level;
2630 if (r2->wm[level].enable)
2631 level2 = level;
861f3389
PZ		 2632 }
2633
198a1e9b
VS		 2634 if (level1 == level2) {
2635 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
861f3389
PZ		 2636 return r2;
2637 else
2638 return r1;
198a1e9b 2639 } else if (level1 > level2) {
861f3389
PZ		 2640 return r1;
2641 } else {
2642 return r2;
2643 }
2644}
2645
49a687c4
VS		 2646/* dirty bits used to track which watermarks need changes */
2647#define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2648#define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2649#define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2650#define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2651#define WM_DIRTY_FBC (1 << 24)
2652#define WM_DIRTY_DDB (1 << 25)
2653
055e393f 2654static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv,
820c1980
ID		 2655 const struct ilk_wm_values *old,
2656 const struct ilk_wm_values *new)
49a687c4
VS		 2657{
2658 unsigned int dirty = 0;
2659 enum pipe pipe;
2660 int wm_lp;
2661
055e393f 2662 for_each_pipe(dev_priv, pipe) {
49a687c4
VS		 2663 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
2664 dirty |= WM_DIRTY_LINETIME(pipe);
2665 /* Must disable LP1+ watermarks too */
2666 dirty |= WM_DIRTY_LP_ALL;
2667 }
2668
2669 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
2670 dirty |= WM_DIRTY_PIPE(pipe);
2671 /* Must disable LP1+ watermarks too */
2672 dirty |= WM_DIRTY_LP_ALL;
2673 }
2674 }
2675
2676 if (old->enable_fbc_wm != new->enable_fbc_wm) {
2677 dirty |= WM_DIRTY_FBC;
2678 /* Must disable LP1+ watermarks too */
2679 dirty |= WM_DIRTY_LP_ALL;
2680 }
2681
2682 if (old->partitioning != new->partitioning) {
2683 dirty |= WM_DIRTY_DDB;
2684 /* Must disable LP1+ watermarks too */
2685 dirty |= WM_DIRTY_LP_ALL;
2686 }
2687
2688 /* LP1+ watermarks already deemed dirty, no need to continue */
2689 if (dirty & WM_DIRTY_LP_ALL)
2690 return dirty;
2691
2692 /* Find the lowest numbered LP1+ watermark in need of an update... */
2693 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2694 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
2695 old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
2696 break;
2697 }
2698
2699 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2700 for (; wm_lp <= 3; wm_lp++)
2701 dirty |= WM_DIRTY_LP(wm_lp);
2702
2703 return dirty;
2704}
2705
8553c18e
VS		 2706static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
2707 unsigned int dirty)
801bcfff 2708{
820c1980 2709 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e 2710 bool changed = false;
801bcfff 2711
facd619b
VS		 2712 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
2713 previous->wm_lp[2] &= ~WM1_LP_SR_EN;
2714 I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
8553c18e 2715 changed = true;
facd619b
VS		 2716 }
2717 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
2718 previous->wm_lp[1] &= ~WM1_LP_SR_EN;
2719 I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
8553c18e 2720 changed = true;
facd619b
VS		 2721 }
2722 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
2723 previous->wm_lp[0] &= ~WM1_LP_SR_EN;
2724 I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
8553c18e 2725 changed = true;
facd619b 2726 }
801bcfff 2727
facd619b
VS		 2728 /*
2729 * Don't touch WM1S_LP_EN here.
2730 * Doing so could cause underruns.
2731 */
6cef2b8a 2732
8553c18e
VS		 2733 return changed;
2734}
2735
2736/*
2737 * The spec says we shouldn't write when we don't need, because every write
2738 * causes WMs to be re-evaluated, expending some power.
2739 */
820c1980
ID		 2740static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
2741 struct ilk_wm_values *results)
8553c18e
VS		 2742{
2743 struct drm_device *dev = dev_priv->dev;
820c1980 2744 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e
VS		 2745 unsigned int dirty;
2746 uint32_t val;
2747
055e393f 2748 dirty = ilk_compute_wm_dirty(dev_priv, previous, results);
8553c18e
VS		 2749 if (!dirty)
2750 return;
2751
2752 _ilk_disable_lp_wm(dev_priv, dirty);
2753
49a687c4 2754 if (dirty & WM_DIRTY_PIPE(PIPE_A))
801bcfff 2755 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
49a687c4 2756 if (dirty & WM_DIRTY_PIPE(PIPE_B))
801bcfff 2757 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
49a687c4 2758 if (dirty & WM_DIRTY_PIPE(PIPE_C))
801bcfff
PZ		 2759 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2760
49a687c4 2761 if (dirty & WM_DIRTY_LINETIME(PIPE_A))
801bcfff 2762 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
49a687c4 2763 if (dirty & WM_DIRTY_LINETIME(PIPE_B))
801bcfff 2764 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
49a687c4 2765 if (dirty & WM_DIRTY_LINETIME(PIPE_C))
801bcfff
PZ		 2766 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2767
49a687c4 2768 if (dirty & WM_DIRTY_DDB) {
a42a5719 2769 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
ac9545fd
VS		 2770 val = I915_READ(WM_MISC);
2771 if (results->partitioning == INTEL_DDB_PART_1_2)
2772 val &= ~WM_MISC_DATA_PARTITION_5_6;
2773 else
2774 val |= WM_MISC_DATA_PARTITION_5_6;
2775 I915_WRITE(WM_MISC, val);
2776 } else {
2777 val = I915_READ(DISP_ARB_CTL2);
2778 if (results->partitioning == INTEL_DDB_PART_1_2)
2779 val &= ~DISP_DATA_PARTITION_5_6;
2780 else
2781 val |= DISP_DATA_PARTITION_5_6;
2782 I915_WRITE(DISP_ARB_CTL2, val);
2783 }
1011d8c4
PZ		 2784 }
2785
49a687c4 2786 if (dirty & WM_DIRTY_FBC) {
cca32e9a
PZ		 2787 val = I915_READ(DISP_ARB_CTL);
2788 if (results->enable_fbc_wm)
2789 val &= ~DISP_FBC_WM_DIS;
2790 else
2791 val |= DISP_FBC_WM_DIS;
2792 I915_WRITE(DISP_ARB_CTL, val);
2793 }
2794
954911eb
ID		 2795 if (dirty & WM_DIRTY_LP(1) &&
2796 previous->wm_lp_spr[0] != results->wm_lp_spr[0])
2797 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2798
2799 if (INTEL_INFO(dev)->gen >= 7) {
6cef2b8a
VS		 2800 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
2801 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2802 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
2803 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2804 }
801bcfff 2805
facd619b 2806 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
801bcfff 2807 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
facd619b 2808 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
801bcfff 2809 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
facd619b 2810 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
801bcfff 2811 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
609cedef
VS		 2812
2813 dev_priv->wm.hw = *results;
801bcfff
PZ		 2814}
2815
ed4a6a7c 2816bool ilk_disable_lp_wm(struct drm_device *dev)
8553c18e
VS		 2817{
2818 struct drm_i915_private *dev_priv = dev->dev_private;
2819
2820 return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
2821}
2822
b9cec075
DL		 2823/*
2824 * On gen9, we need to allocate Display Data Buffer (DDB) portions to the
2825 * different active planes.
2826 */
2827
2828#define SKL_DDB_SIZE 896 /* in blocks */
43d735a6 2829#define BXT_DDB_SIZE 512
b9cec075 2830
024c9045
MR		 2831/*
2832 * Return the index of a plane in the SKL DDB and wm result arrays. Primary
2833 * plane is always in slot 0, cursor is always in slot I915_MAX_PLANES-1, and
2834 * other universal planes are in indices 1..n. Note that this may leave unused
2835 * indices between the top "sprite" plane and the cursor.
2836 */
2837static int
2838skl_wm_plane_id(const struct intel_plane *plane)
2839{
2840 switch (plane->base.type) {
2841 case DRM_PLANE_TYPE_PRIMARY:
2842 return 0;
2843 case DRM_PLANE_TYPE_CURSOR:
2844 return PLANE_CURSOR;
2845 case DRM_PLANE_TYPE_OVERLAY:
2846 return plane->plane + 1;
2847 default:
2848 MISSING_CASE(plane->base.type);
2849 return plane->plane;
2850 }
2851}
2852
b9cec075
DL		 2853static void
2854skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
024c9045 2855 const struct intel_crtc_state *cstate,
c107acfe
MR		 2856 struct skl_ddb_entry *alloc, /* out */
2857 int *num_active /* out */)
b9cec075 2858{
c107acfe
MR		 2859 struct drm_atomic_state *state = cstate->base.state;
2860 struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
2861 struct drm_i915_private *dev_priv = to_i915(dev);
024c9045 2862 struct drm_crtc *for_crtc = cstate->base.crtc;
b9cec075
DL		 2863 unsigned int pipe_size, ddb_size;
2864 int nth_active_pipe;
c107acfe
MR		 2865 int pipe = to_intel_crtc(for_crtc)->pipe;
2866
a6d3460e 2867 if (WARN_ON(!state) || !cstate->base.active) {
b9cec075
DL		 2868 alloc->start = 0;
2869 alloc->end = 0;
a6d3460e 2870 *num_active = hweight32(dev_priv->active_crtcs);
b9cec075
DL		 2871 return;
2872 }
2873
a6d3460e
MR		 2874 if (intel_state->active_pipe_changes)
2875 *num_active = hweight32(intel_state->active_crtcs);
2876 else
2877 *num_active = hweight32(dev_priv->active_crtcs);
2878
43d735a6
DL		 2879 if (IS_BROXTON(dev))
2880 ddb_size = BXT_DDB_SIZE;
2881 else
2882 ddb_size = SKL_DDB_SIZE;
b9cec075
DL		 2883
2884 ddb_size -= 4; /* 4 blocks for bypass path allocation */
2885
c107acfe 2886 /*
a6d3460e
MR		 2887 * If the state doesn't change the active CRTC's, then there's
2888 * no need to recalculate; the existing pipe allocation limits
2889 * should remain unchanged. Note that we're safe from racing
2890 * commits since any racing commit that changes the active CRTC
2891 * list would need to grab _all_ crtc locks, including the one
2892 * we currently hold.
c107acfe 2893 */
a6d3460e
MR		 2894 if (!intel_state->active_pipe_changes) {
2895 *alloc = dev_priv->wm.skl_hw.ddb.pipe[pipe];
2896 return;
c107acfe 2897 }
a6d3460e
MR		 2898
2899 nth_active_pipe = hweight32(intel_state->active_crtcs &
2900 (drm_crtc_mask(for_crtc) - 1));
2901 pipe_size = ddb_size / hweight32(intel_state->active_crtcs);
2902 alloc->start = nth_active_pipe * ddb_size / *num_active;
2903 alloc->end = alloc->start + pipe_size;
b9cec075
DL		 2904}
2905
c107acfe 2906static unsigned int skl_cursor_allocation(int num_active)
b9cec075 2907{
c107acfe 2908 if (num_active == 1)
b9cec075
DL		 2909 return 32;
2910
2911 return 8;
2912}
2913
a269c583
DL		 2914static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg)
2915{
2916 entry->start = reg & 0x3ff;
2917 entry->end = (reg >> 16) & 0x3ff;
16160e3d
DL		 2918 if (entry->end)
2919 entry->end += 1;
a269c583
DL		 2920}
2921
08db6652
DL		 2922void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv,
2923 struct skl_ddb_allocation *ddb /* out */)
a269c583 2924{
a269c583
DL		 2925 enum pipe pipe;
2926 int plane;
2927 u32 val;
2928
b10f1b20
ML		 2929 memset(ddb, 0, sizeof(*ddb));
2930
a269c583 2931 for_each_pipe(dev_priv, pipe) {
4d800030
ID		 2932 enum intel_display_power_domain power_domain;
2933
2934 power_domain = POWER_DOMAIN_PIPE(pipe);
2935 if (!intel_display_power_get_if_enabled(dev_priv, power_domain))
b10f1b20
ML		 2936 continue;
2937
dd740780 2938 for_each_plane(dev_priv, pipe, plane) {
a269c583
DL		 2939 val = I915_READ(PLANE_BUF_CFG(pipe, plane));
2940 skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane],
2941 val);
2942 }
2943
2944 val = I915_READ(CUR_BUF_CFG(pipe));
4969d33e
MR		 2945 skl_ddb_entry_init_from_hw(&ddb->plane[pipe][PLANE_CURSOR],
2946 val);
4d800030
ID		 2947
2948 intel_display_power_put(dev_priv, power_domain);
a269c583
DL		 2949 }
2950}
2951
b9cec075 2952static unsigned int
024c9045
MR		 2953skl_plane_relative_data_rate(const struct intel_crtc_state *cstate,
2954 const struct drm_plane_state *pstate,
2955 int y)
b9cec075 2956{
a280f7dd 2957 struct intel_plane_state *intel_pstate = to_intel_plane_state(pstate);
024c9045 2958 struct drm_framebuffer *fb = pstate->fb;
a280f7dd 2959 uint32_t width = 0, height = 0;
a1de91e5
MR		 2960 unsigned format = fb ? fb->pixel_format : DRM_FORMAT_XRGB8888;
2961
2962 if (!intel_pstate->visible)
2963 return 0;
2964 if (pstate->plane->type == DRM_PLANE_TYPE_CURSOR)
2965 return 0;
2966 if (y && format != DRM_FORMAT_NV12)
2967 return 0;
a280f7dd
KM		 2968
2969 width = drm_rect_width(&intel_pstate->src) >> 16;
2970 height = drm_rect_height(&intel_pstate->src) >> 16;
2971
2972 if (intel_rotation_90_or_270(pstate->rotation))
2973 swap(width, height);
2cd601c6
CK		 2974
2975 /* for planar format */
a1de91e5 2976 if (format == DRM_FORMAT_NV12) {
2cd601c6 2977 if (y) /* y-plane data rate */
a280f7dd 2978 return width * height *
a1de91e5 2979 drm_format_plane_cpp(format, 0);
2cd601c6 2980 else /* uv-plane data rate */
a280f7dd 2981 return (width / 2) * (height / 2) *
a1de91e5 2982 drm_format_plane_cpp(format, 1);
2cd601c6
CK		 2983 }
2984
2985 /* for packed formats */
a1de91e5 2986 return width * height * drm_format_plane_cpp(format, 0);
b9cec075
DL		 2987}
2988
2989/*
2990 * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching
2991 * a 8192x4096@32bpp framebuffer:
2992 * 3 * 4096 * 8192 * 4 < 2^32
2993 */
2994static unsigned int
9c74d826 2995skl_get_total_relative_data_rate(struct intel_crtc_state *intel_cstate)
b9cec075 2996{
9c74d826
MR		 2997 struct drm_crtc_state *cstate = &intel_cstate->base;
2998 struct drm_atomic_state *state = cstate->state;
2999 struct drm_crtc *crtc = cstate->crtc;
3000 struct drm_device *dev = crtc->dev;
3001 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
a6d3460e 3002 const struct drm_plane *plane;
024c9045 3003 const struct intel_plane *intel_plane;
a6d3460e 3004 struct drm_plane_state *pstate;
a1de91e5 3005 unsigned int rate, total_data_rate = 0;
9c74d826 3006 int id;
a6d3460e
MR		 3007 int i;
3008
3009 if (WARN_ON(!state))
3010 return 0;
b9cec075 3011
a1de91e5 3012 /* Calculate and cache data rate for each plane */
a6d3460e
MR		 3013 for_each_plane_in_state(state, plane, pstate, i) {
3014 id = skl_wm_plane_id(to_intel_plane(plane));
3015 intel_plane = to_intel_plane(plane);
3016
3017 if (intel_plane->pipe != intel_crtc->pipe)
3018 continue;
3019
3020 /* packed/uv */
3021 rate = skl_plane_relative_data_rate(intel_cstate,
3022 pstate, 0);
3023 intel_cstate->wm.skl.plane_data_rate[id] = rate;
3024
3025 /* y-plane */
3026 rate = skl_plane_relative_data_rate(intel_cstate,
3027 pstate, 1);
3028 intel_cstate->wm.skl.plane_y_data_rate[id] = rate;
a1de91e5 3029 }
024c9045 3030
a1de91e5
MR		 3031 /* Calculate CRTC's total data rate from cached values */
3032 for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) {
3033 int id = skl_wm_plane_id(intel_plane);
024c9045 3034
a1de91e5 3035 /* packed/uv */
9c74d826
MR		 3036 total_data_rate += intel_cstate->wm.skl.plane_data_rate[id];
3037 total_data_rate += intel_cstate->wm.skl.plane_y_data_rate[id];
b9cec075
DL		 3038 }
3039
9c74d826
MR		 3040 WARN_ON(cstate->plane_mask && total_data_rate == 0);
3041
b9cec075
DL		 3042 return total_data_rate;
3043}
3044
c107acfe 3045static int
024c9045 3046skl_allocate_pipe_ddb(struct intel_crtc_state *cstate,
b9cec075
DL		 3047 struct skl_ddb_allocation *ddb /* out */)
3048{
c107acfe 3049 struct drm_atomic_state *state = cstate->base.state;
024c9045 3050 struct drm_crtc *crtc = cstate->base.crtc;
b9cec075
DL		 3051 struct drm_device *dev = crtc->dev;
3052 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
024c9045 3053 struct intel_plane *intel_plane;
c107acfe
MR		 3054 struct drm_plane *plane;
3055 struct drm_plane_state *pstate;
b9cec075 3056 enum pipe pipe = intel_crtc->pipe;
34bb56af 3057 struct skl_ddb_entry *alloc = &ddb->pipe[pipe];
b9cec075 3058 uint16_t alloc_size, start, cursor_blocks;
86a2100a
MR		 3059 uint16_t *minimum = cstate->wm.skl.minimum_blocks;
3060 uint16_t *y_minimum = cstate->wm.skl.minimum_y_blocks;
b9cec075 3061 unsigned int total_data_rate;
c107acfe
MR		 3062 int num_active;
3063 int id, i;
b9cec075 3064
a6d3460e
MR		 3065 if (WARN_ON(!state))
3066 return 0;
3067
c107acfe
MR		 3068 if (!cstate->base.active) {
3069 ddb->pipe[pipe].start = ddb->pipe[pipe].end = 0;
3070 memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe]));
3071 memset(ddb->y_plane[pipe], 0, sizeof(ddb->y_plane[pipe]));
3072 return 0;
3073 }
3074
a6d3460e 3075 skl_ddb_get_pipe_allocation_limits(dev, cstate, alloc, &num_active);
34bb56af 3076 alloc_size = skl_ddb_entry_size(alloc);
b9cec075
DL		 3077 if (alloc_size == 0) {
3078 memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe]));
c107acfe 3079 return 0;
b9cec075
DL		 3080 }
3081
c107acfe 3082 cursor_blocks = skl_cursor_allocation(num_active);
4969d33e
MR		 3083 ddb->plane[pipe][PLANE_CURSOR].start = alloc->end - cursor_blocks;
3084 ddb->plane[pipe][PLANE_CURSOR].end = alloc->end;
b9cec075
DL		 3085
3086 alloc_size -= cursor_blocks;
b9cec075 3087
80958155 3088 /* 1. Allocate the mininum required blocks for each active plane */
a6d3460e
MR		 3089 for_each_plane_in_state(state, plane, pstate, i) {
3090 intel_plane = to_intel_plane(plane);
3091 id = skl_wm_plane_id(intel_plane);
c107acfe 3092
a6d3460e
MR		 3093 if (intel_plane->pipe != pipe)
3094 continue;
c107acfe 3095
a6d3460e
MR		 3096 if (!to_intel_plane_state(pstate)->visible) {
3097 minimum[id] = 0;
3098 y_minimum[id] = 0;
3099 continue;
3100 }
3101 if (plane->type == DRM_PLANE_TYPE_CURSOR) {
3102 minimum[id] = 0;
3103 y_minimum[id] = 0;
3104 continue;
c107acfe 3105 }
a6d3460e
MR		 3106
3107 minimum[id] = 8;
3108 if (pstate->fb->pixel_format == DRM_FORMAT_NV12)
3109 y_minimum[id] = 8;
3110 else
3111 y_minimum[id] = 0;
c107acfe 3112 }
80958155 3113
c107acfe
MR		 3114 for (i = 0; i < PLANE_CURSOR; i++) {
3115 alloc_size -= minimum[i];
3116 alloc_size -= y_minimum[i];
80958155
DL		 3117 }
3118
b9cec075 3119 /*
80958155
DL		 3120 * 2. Distribute the remaining space in proportion to the amount of
3121 * data each plane needs to fetch from memory.
b9cec075
DL		 3122 *
3123 * FIXME: we may not allocate every single block here.
3124 */
024c9045 3125 total_data_rate = skl_get_total_relative_data_rate(cstate);
a1de91e5 3126 if (total_data_rate == 0)
c107acfe 3127 return 0;
b9cec075 3128
34bb56af 3129 start = alloc->start;
024c9045 3130 for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) {
2cd601c6
CK		 3131 unsigned int data_rate, y_data_rate;
3132 uint16_t plane_blocks, y_plane_blocks = 0;
024c9045 3133 int id = skl_wm_plane_id(intel_plane);
b9cec075 3134
a1de91e5 3135 data_rate = cstate->wm.skl.plane_data_rate[id];
b9cec075
DL		 3136
3137 /*
2cd601c6 3138 * allocation for (packed formats) or (uv-plane part of planar format):
b9cec075
DL		 3139 * promote the expression to 64 bits to avoid overflowing, the
3140 * result is < available as data_rate / total_data_rate < 1
3141 */
024c9045 3142 plane_blocks = minimum[id];
80958155
DL		 3143 plane_blocks += div_u64((uint64_t)alloc_size * data_rate,
3144 total_data_rate);
b9cec075 3145
c107acfe
MR		 3146 /* Leave disabled planes at (0,0) */
3147 if (data_rate) {
3148 ddb->plane[pipe][id].start = start;
3149 ddb->plane[pipe][id].end = start + plane_blocks;
3150 }
b9cec075
DL		 3151
3152 start += plane_blocks;
2cd601c6
CK		 3153
3154 /*
3155 * allocation for y_plane part of planar format:
3156 */
a1de91e5
MR		 3157 y_data_rate = cstate->wm.skl.plane_y_data_rate[id];
3158
3159 y_plane_blocks = y_minimum[id];
3160 y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate,
3161 total_data_rate);
2cd601c6 3162
c107acfe
MR		 3163 if (y_data_rate) {
3164 ddb->y_plane[pipe][id].start = start;
3165 ddb->y_plane[pipe][id].end = start + y_plane_blocks;
3166 }
a1de91e5
MR		 3167
3168 start += y_plane_blocks;
b9cec075
DL		 3169 }
3170
c107acfe 3171 return 0;
b9cec075
DL		 3172}
3173
5cec258b 3174static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state *config)
2d41c0b5
PB		 3175{
3176 /* TODO: Take into account the scalers once we support them */
2d112de7 3177 return config->base.adjusted_mode.crtc_clock;
2d41c0b5
PB		 3178}
3179
3180/*
3181 * The max latency should be 257 (max the punit can code is 255 and we add 2us
ac484963 3182 * for the read latency) and cpp should always be <= 8, so that
2d41c0b5
PB		 3183 * should allow pixel_rate up to ~2 GHz which seems sufficient since max
3184 * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
3185*/
ac484963 3186static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t cpp, uint32_t latency)
2d41c0b5
PB		 3187{
3188 uint32_t wm_intermediate_val, ret;
3189
3190 if (latency == 0)
3191 return UINT_MAX;
3192
ac484963 3193 wm_intermediate_val = latency * pixel_rate * cpp / 512;
2d41c0b5
PB		 3194 ret = DIV_ROUND_UP(wm_intermediate_val, 1000);
3195
3196 return ret;
3197}
3198
3199static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
ac484963 3200 uint32_t horiz_pixels, uint8_t cpp,
0fda6568 3201 uint64_t tiling, uint32_t latency)
2d41c0b5 3202{
d4c2aa60
TU		 3203 uint32_t ret;
3204 uint32_t plane_bytes_per_line, plane_blocks_per_line;
3205 uint32_t wm_intermediate_val;
2d41c0b5
PB		 3206
3207 if (latency == 0)
3208 return UINT_MAX;
3209
ac484963 3210 plane_bytes_per_line = horiz_pixels * cpp;
0fda6568
TU		 3211
3212 if (tiling == I915_FORMAT_MOD_Y_TILED ||
3213 tiling == I915_FORMAT_MOD_Yf_TILED) {
3214 plane_bytes_per_line *= 4;
3215 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
3216 plane_blocks_per_line /= 4;
3217 } else {
3218 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
3219 }
3220
2d41c0b5
PB		 3221 wm_intermediate_val = latency * pixel_rate;
3222 ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) *
d4c2aa60 3223 plane_blocks_per_line;
2d41c0b5
PB		 3224
3225 return ret;
3226}
3227
55994c2c
MR		 3228static int skl_compute_plane_wm(const struct drm_i915_private *dev_priv,
3229 struct intel_crtc_state *cstate,
3230 struct intel_plane_state *intel_pstate,
3231 uint16_t ddb_allocation,
3232 int level,
3233 uint16_t *out_blocks, /* out */
3234 uint8_t *out_lines, /* out */
3235 bool *enabled /* out */)
2d41c0b5 3236{
33815fa5
MR		 3237 struct drm_plane_state *pstate = &intel_pstate->base;
3238 struct drm_framebuffer *fb = pstate->fb;
d4c2aa60
TU		 3239 uint32_t latency = dev_priv->wm.skl_latency[level];
3240 uint32_t method1, method2;
3241 uint32_t plane_bytes_per_line, plane_blocks_per_line;
3242 uint32_t res_blocks, res_lines;
3243 uint32_t selected_result;
ac484963 3244 uint8_t cpp;
a280f7dd 3245 uint32_t width = 0, height = 0;
2d41c0b5 3246
55994c2c
MR		 3247 if (latency == 0 || !cstate->base.active || !intel_pstate->visible) {
3248 *enabled = false;
3249 return 0;
3250 }
2d41c0b5 3251
a280f7dd
KM		 3252 width = drm_rect_width(&intel_pstate->src) >> 16;
3253 height = drm_rect_height(&intel_pstate->src) >> 16;
3254
33815fa5 3255 if (intel_rotation_90_or_270(pstate->rotation))
a280f7dd
KM		 3256 swap(width, height);
3257
ac484963 3258 cpp = drm_format_plane_cpp(fb->pixel_format, 0);
024c9045 3259 method1 = skl_wm_method1(skl_pipe_pixel_rate(cstate),
ac484963 3260 cpp, latency);
024c9045
MR		 3261 method2 = skl_wm_method2(skl_pipe_pixel_rate(cstate),
3262 cstate->base.adjusted_mode.crtc_htotal,
a280f7dd
KM		 3263 width,
3264 cpp,
3265 fb->modifier[0],
d4c2aa60 3266 latency);
2d41c0b5 3267
a280f7dd 3268 plane_bytes_per_line = width * cpp;
d4c2aa60 3269 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
2d41c0b5 3270
024c9045
MR		 3271 if (fb->modifier[0] == I915_FORMAT_MOD_Y_TILED ||
3272 fb->modifier[0] == I915_FORMAT_MOD_Yf_TILED) {
1fc0a8f7
TU		 3273 uint32_t min_scanlines = 4;
3274 uint32_t y_tile_minimum;
33815fa5 3275 if (intel_rotation_90_or_270(pstate->rotation)) {
ac484963 3276 int cpp = (fb->pixel_format == DRM_FORMAT_NV12) ?
024c9045
MR		 3277 drm_format_plane_cpp(fb->pixel_format, 1) :
3278 drm_format_plane_cpp(fb->pixel_format, 0);
3279
ac484963 3280 switch (cpp) {
1fc0a8f7
TU		 3281 case 1:
3282 min_scanlines = 16;
3283 break;
3284 case 2:
3285 min_scanlines = 8;
3286 break;
3287 case 8:
3288 WARN(1, "Unsupported pixel depth for rotation");
2f0b5790 3289 }
1fc0a8f7
TU		 3290 }
3291 y_tile_minimum = plane_blocks_per_line * min_scanlines;
0fda6568
TU		 3292 selected_result = max(method2, y_tile_minimum);
3293 } else {
3294 if ((ddb_allocation / plane_blocks_per_line) >= 1)
3295 selected_result = min(method1, method2);
3296 else
3297 selected_result = method1;
3298 }
2d41c0b5 3299
d4c2aa60
TU		 3300 res_blocks = selected_result + 1;
3301 res_lines = DIV_ROUND_UP(selected_result, plane_blocks_per_line);
e6d66171 3302
0fda6568 3303 if (level >= 1 && level <= 7) {
024c9045
MR		 3304 if (fb->modifier[0] == I915_FORMAT_MOD_Y_TILED ||
3305 fb->modifier[0] == I915_FORMAT_MOD_Yf_TILED)
0fda6568
TU		 3306 res_lines += 4;
3307 else
3308 res_blocks++;
3309 }
e6d66171 3310
55994c2c
MR		 3311 if (res_blocks >= ddb_allocation || res_lines > 31) {
3312 *enabled = false;
6b6bada7
MR		 3313
3314 /*
3315 * If there are no valid level 0 watermarks, then we can't
3316 * support this display configuration.
3317 */
3318 if (level) {
3319 return 0;
3320 } else {
3321 DRM_DEBUG_KMS("Requested display configuration exceeds system watermark limitations\n");
3322 DRM_DEBUG_KMS("Plane %d.%d: blocks required = %u/%u, lines required = %u/31\n",
3323 to_intel_crtc(cstate->base.crtc)->pipe,
3324 skl_wm_plane_id(to_intel_plane(pstate->plane)),
3325 res_blocks, ddb_allocation, res_lines);
3326
3327 return -EINVAL;
3328 }
55994c2c 3329 }
e6d66171
DL		 3330
3331 *out_blocks = res_blocks;
3332 *out_lines = res_lines;
55994c2c 3333 *enabled = true;
2d41c0b5 3334
55994c2c 3335 return 0;
2d41c0b5
PB		 3336}
3337
f4a96752
MR		 3338static int
3339skl_compute_wm_level(const struct drm_i915_private *dev_priv,
3340 struct skl_ddb_allocation *ddb,
3341 struct intel_crtc_state *cstate,
3342 int level,
3343 struct skl_wm_level *result)
2d41c0b5 3344{
024c9045 3345 struct drm_device *dev = dev_priv->dev;
f4a96752 3346 struct drm_atomic_state *state = cstate->base.state;
024c9045 3347 struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc);
f4a96752 3348 struct drm_plane *plane;
024c9045 3349 struct intel_plane *intel_plane;
33815fa5 3350 struct intel_plane_state *intel_pstate;
2d41c0b5 3351 uint16_t ddb_blocks;
024c9045 3352 enum pipe pipe = intel_crtc->pipe;
55994c2c 3353 int ret;
024c9045 3354
f4a96752
MR		 3355 /*
3356 * We'll only calculate watermarks for planes that are actually
3357 * enabled, so make sure all other planes are set as disabled.
3358 */
3359 memset(result, 0, sizeof(*result));
3360
3361 for_each_intel_plane_mask(dev, intel_plane, cstate->base.plane_mask) {
024c9045 3362 int i = skl_wm_plane_id(intel_plane);
2d41c0b5 3363
f4a96752
MR		 3364 plane = &intel_plane->base;
3365 intel_pstate = NULL;
3366 if (state)
3367 intel_pstate =
3368 intel_atomic_get_existing_plane_state(state,
3369 intel_plane);
3370
3371 /*
3372 * Note: If we start supporting multiple pending atomic commits
3373 * against the same planes/CRTC's in the future, plane->state
3374 * will no longer be the correct pre-state to use for the
3375 * calculations here and we'll need to change where we get the
3376 * 'unchanged' plane data from.
3377 *
3378 * For now this is fine because we only allow one queued commit
3379 * against a CRTC. Even if the plane isn't modified by this
3380 * transaction and we don't have a plane lock, we still have
3381 * the CRTC's lock, so we know that no other transactions are
3382 * racing with us to update it.
3383 */
3384 if (!intel_pstate)
3385 intel_pstate = to_intel_plane_state(plane->state);
3386
3387 WARN_ON(!intel_pstate->base.fb);
3388
2d41c0b5
PB		 3389 ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]);
3390
55994c2c
MR		 3391 ret = skl_compute_plane_wm(dev_priv,
3392 cstate,
3393 intel_pstate,
3394 ddb_blocks,
3395 level,
3396 &result->plane_res_b[i],
3397 &result->plane_res_l[i],
3398 &result->plane_en[i]);
3399 if (ret)
3400 return ret;
2d41c0b5 3401 }
f4a96752
MR		 3402
3403 return 0;
2d41c0b5
PB		 3404}
3405
407b50f3 3406static uint32_t
024c9045 3407skl_compute_linetime_wm(struct intel_crtc_state *cstate)
407b50f3 3408{
024c9045 3409 if (!cstate->base.active)
407b50f3
DL		 3410 return 0;
3411
024c9045 3412 if (WARN_ON(skl_pipe_pixel_rate(cstate) == 0))
661abfc0 3413 return 0;
407b50f3 3414
024c9045
MR		 3415 return DIV_ROUND_UP(8 * cstate->base.adjusted_mode.crtc_htotal * 1000,
3416 skl_pipe_pixel_rate(cstate));
407b50f3
DL		 3417}
3418
024c9045 3419static void skl_compute_transition_wm(struct intel_crtc_state *cstate,
9414f563 3420 struct skl_wm_level *trans_wm /* out */)
407b50f3 3421{
024c9045 3422 struct drm_crtc *crtc = cstate->base.crtc;
9414f563 3423 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
024c9045 3424 struct intel_plane *intel_plane;
9414f563 3425
024c9045 3426 if (!cstate->base.active)
407b50f3 3427 return;
9414f563
DL		 3428
3429 /* Until we know more, just disable transition WMs */
024c9045
MR		 3430 for_each_intel_plane_on_crtc(crtc->dev, intel_crtc, intel_plane) {
3431 int i = skl_wm_plane_id(intel_plane);
3432
9414f563 3433 trans_wm->plane_en[i] = false;
024c9045 3434 }
407b50f3
DL		 3435}
3436
55994c2c
MR		 3437static int skl_build_pipe_wm(struct intel_crtc_state *cstate,
3438 struct skl_ddb_allocation *ddb,
3439 struct skl_pipe_wm *pipe_wm)
2d41c0b5 3440{
024c9045 3441 struct drm_device *dev = cstate->base.crtc->dev;
2d41c0b5 3442 const struct drm_i915_private *dev_priv = dev->dev_private;
2d41c0b5 3443 int level, max_level = ilk_wm_max_level(dev);
55994c2c 3444 int ret;
2d41c0b5
PB		 3445
3446 for (level = 0; level <= max_level; level++) {
55994c2c
MR		 3447 ret = skl_compute_wm_level(dev_priv, ddb, cstate,
3448 level, &pipe_wm->wm[level]);
3449 if (ret)
3450 return ret;
2d41c0b5 3451 }
024c9045 3452 pipe_wm->linetime = skl_compute_linetime_wm(cstate);
2d41c0b5 3453
024c9045 3454 skl_compute_transition_wm(cstate, &pipe_wm->trans_wm);
55994c2c
MR		 3455
3456 return 0;
2d41c0b5
PB		 3457}
3458
3459static void skl_compute_wm_results(struct drm_device *dev,
2d41c0b5
PB		 3460 struct skl_pipe_wm *p_wm,
3461 struct skl_wm_values *r,
3462 struct intel_crtc *intel_crtc)
3463{
3464 int level, max_level = ilk_wm_max_level(dev);
3465 enum pipe pipe = intel_crtc->pipe;
9414f563
DL		 3466 uint32_t temp;
3467 int i;
2d41c0b5
PB		 3468
3469 for (level = 0; level <= max_level; level++) {
2d41c0b5
PB		 3470 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
3471 temp = 0;
2d41c0b5
PB		 3472
3473 temp |= p_wm->wm[level].plane_res_l[i] <<
3474 PLANE_WM_LINES_SHIFT;
3475 temp |= p_wm->wm[level].plane_res_b[i];
3476 if (p_wm->wm[level].plane_en[i])
3477 temp |= PLANE_WM_EN;
3478
3479 r->plane[pipe][i][level] = temp;
2d41c0b5
PB		 3480 }
3481
3482 temp = 0;
2d41c0b5 3483
4969d33e
MR		 3484 temp |= p_wm->wm[level].plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT;
3485 temp |= p_wm->wm[level].plane_res_b[PLANE_CURSOR];
2d41c0b5 3486
4969d33e 3487 if (p_wm->wm[level].plane_en[PLANE_CURSOR])
2d41c0b5
PB		 3488 temp |= PLANE_WM_EN;
3489
4969d33e 3490 r->plane[pipe][PLANE_CURSOR][level] = temp;
2d41c0b5
PB		 3491
3492 }
3493
9414f563
DL		 3494 /* transition WMs */
3495 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
3496 temp = 0;
3497 temp |= p_wm->trans_wm.plane_res_l[i] << PLANE_WM_LINES_SHIFT;
3498 temp |= p_wm->trans_wm.plane_res_b[i];
3499 if (p_wm->trans_wm.plane_en[i])
3500 temp |= PLANE_WM_EN;
3501
3502 r->plane_trans[pipe][i] = temp;
3503 }
3504
3505 temp = 0;
4969d33e
MR		 3506 temp |= p_wm->trans_wm.plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT;
3507 temp |= p_wm->trans_wm.plane_res_b[PLANE_CURSOR];
3508 if (p_wm->trans_wm.plane_en[PLANE_CURSOR])
9414f563
DL		 3509 temp |= PLANE_WM_EN;
3510
4969d33e 3511 r->plane_trans[pipe][PLANE_CURSOR] = temp;
9414f563 3512
2d41c0b5
PB		 3513 r->wm_linetime[pipe] = p_wm->linetime;
3514}
3515
f0f59a00
VS		 3516static void skl_ddb_entry_write(struct drm_i915_private *dev_priv,
3517 i915_reg_t reg,
16160e3d
DL		 3518 const struct skl_ddb_entry *entry)
3519{
3520 if (entry->end)
3521 I915_WRITE(reg, (entry->end - 1) << 16 | entry->start);
3522 else
3523 I915_WRITE(reg, 0);
3524}
3525
2d41c0b5
PB		 3526static void skl_write_wm_values(struct drm_i915_private *dev_priv,
3527 const struct skl_wm_values *new)
3528{
3529 struct drm_device *dev = dev_priv->dev;
3530 struct intel_crtc *crtc;
3531
19c8054c 3532 for_each_intel_crtc(dev, crtc) {
2d41c0b5
PB		 3533 int i, level, max_level = ilk_wm_max_level(dev);
3534 enum pipe pipe = crtc->pipe;
3535
2b4b9f35 3536 if ((new->dirty_pipes & drm_crtc_mask(&crtc->base)) == 0)
5d374d96 3537 continue;
734fa01f
MR		 3538 if (!crtc->active)
3539 continue;
8211bd5b 3540
5d374d96 3541 I915_WRITE(PIPE_WM_LINETIME(pipe), new->wm_linetime[pipe]);
8211bd5b 3542
5d374d96
DL		 3543 for (level = 0; level <= max_level; level++) {
3544 for (i = 0; i < intel_num_planes(crtc); i++)
3545 I915_WRITE(PLANE_WM(pipe, i, level),
3546 new->plane[pipe][i][level]);
3547 I915_WRITE(CUR_WM(pipe, level),
4969d33e 3548 new->plane[pipe][PLANE_CURSOR][level]);
2d41c0b5 3549 }
5d374d96
DL		 3550 for (i = 0; i < intel_num_planes(crtc); i++)
3551 I915_WRITE(PLANE_WM_TRANS(pipe, i),
3552 new->plane_trans[pipe][i]);
4969d33e
MR		 3553 I915_WRITE(CUR_WM_TRANS(pipe),
3554 new->plane_trans[pipe][PLANE_CURSOR]);
5d374d96 3555
2cd601c6 3556 for (i = 0; i < intel_num_planes(crtc); i++) {
5d374d96
DL		 3557 skl_ddb_entry_write(dev_priv,
3558 PLANE_BUF_CFG(pipe, i),
3559 &new->ddb.plane[pipe][i]);
2cd601c6
CK		 3560 skl_ddb_entry_write(dev_priv,
3561 PLANE_NV12_BUF_CFG(pipe, i),
3562 &new->ddb.y_plane[pipe][i]);
3563 }
5d374d96
DL		 3564
3565 skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe),
4969d33e 3566 &new->ddb.plane[pipe][PLANE_CURSOR]);
2d41c0b5 3567 }
2d41c0b5
PB		 3568}
3569
0e8fb7ba
DL		 3570/*
3571 * When setting up a new DDB allocation arrangement, we need to correctly
3572 * sequence the times at which the new allocations for the pipes are taken into
3573 * account or we'll have pipes fetching from space previously allocated to
3574 * another pipe.
3575 *
3576 * Roughly the sequence looks like:
3577 * 1. re-allocate the pipe(s) with the allocation being reduced and not
3578 * overlapping with a previous light-up pipe (another way to put it is:
3579 * pipes with their new allocation strickly included into their old ones).
3580 * 2. re-allocate the other pipes that get their allocation reduced
3581 * 3. allocate the pipes having their allocation increased
3582 *
3583 * Steps 1. and 2. are here to take care of the following case:
3584 * - Initially DDB looks like this:
3585 * | B | C |
3586 * - enable pipe A.
3587 * - pipe B has a reduced DDB allocation that overlaps with the old pipe C
3588 * allocation
3589 * | A | B | C |
3590 *
3591 * We need to sequence the re-allocation: C, B, A (and not B, C, A).
3592 */
3593
d21b795c
DL		 3594static void
3595skl_wm_flush_pipe(struct drm_i915_private *dev_priv, enum pipe pipe, int pass)
0e8fb7ba 3596{
0e8fb7ba
DL		 3597 int plane;
3598
d21b795c
DL		 3599 DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe), pass);
3600
dd740780 3601 for_each_plane(dev_priv, pipe, plane) {
0e8fb7ba
DL		 3602 I915_WRITE(PLANE_SURF(pipe, plane),
3603 I915_READ(PLANE_SURF(pipe, plane)));
3604 }
3605 I915_WRITE(CURBASE(pipe), I915_READ(CURBASE(pipe)));
3606}
3607
3608static bool
3609skl_ddb_allocation_included(const struct skl_ddb_allocation *old,
3610 const struct skl_ddb_allocation *new,
3611 enum pipe pipe)
3612{
3613 uint16_t old_size, new_size;
3614
3615 old_size = skl_ddb_entry_size(&old->pipe[pipe]);
3616 new_size = skl_ddb_entry_size(&new->pipe[pipe]);
3617
3618 return old_size != new_size &&
3619 new->pipe[pipe].start >= old->pipe[pipe].start &&
3620 new->pipe[pipe].end <= old->pipe[pipe].end;
3621}
3622
3623static void skl_flush_wm_values(struct drm_i915_private *dev_priv,
3624 struct skl_wm_values *new_values)
3625{
3626 struct drm_device *dev = dev_priv->dev;
3627 struct skl_ddb_allocation *cur_ddb, *new_ddb;
c929cb45 3628 bool reallocated[I915_MAX_PIPES] = {};
0e8fb7ba
DL		 3629 struct intel_crtc *crtc;
3630 enum pipe pipe;
3631
3632 new_ddb = &new_values->ddb;
3633 cur_ddb = &dev_priv->wm.skl_hw.ddb;
3634
3635 /*
3636 * First pass: flush the pipes with the new allocation contained into
3637 * the old space.
3638 *
3639 * We'll wait for the vblank on those pipes to ensure we can safely
3640 * re-allocate the freed space without this pipe fetching from it.
3641 */
3642 for_each_intel_crtc(dev, crtc) {
3643 if (!crtc->active)
3644 continue;
3645
3646 pipe = crtc->pipe;
3647
3648 if (!skl_ddb_allocation_included(cur_ddb, new_ddb, pipe))
3649 continue;
3650
d21b795c 3651 skl_wm_flush_pipe(dev_priv, pipe, 1);
0e8fb7ba
DL		 3652 intel_wait_for_vblank(dev, pipe);
3653
3654 reallocated[pipe] = true;
3655 }
3656
3657
3658 /*
3659 * Second pass: flush the pipes that are having their allocation
3660 * reduced, but overlapping with a previous allocation.
3661 *
3662 * Here as well we need to wait for the vblank to make sure the freed
3663 * space is not used anymore.
3664 */
3665 for_each_intel_crtc(dev, crtc) {
3666 if (!crtc->active)
3667 continue;
3668
3669 pipe = crtc->pipe;
3670
3671 if (reallocated[pipe])
3672 continue;
3673
3674 if (skl_ddb_entry_size(&new_ddb->pipe[pipe]) <
3675 skl_ddb_entry_size(&cur_ddb->pipe[pipe])) {
d21b795c 3676 skl_wm_flush_pipe(dev_priv, pipe, 2);
0e8fb7ba 3677 intel_wait_for_vblank(dev, pipe);
d9d8e6b3 3678 reallocated[pipe] = true;
0e8fb7ba 3679 }
0e8fb7ba
DL		 3680 }
3681
3682 /*
3683 * Third pass: flush the pipes that got more space allocated.
3684 *
3685 * We don't need to actively wait for the update here, next vblank
3686 * will just get more DDB space with the correct WM values.
3687 */
3688 for_each_intel_crtc(dev, crtc) {
3689 if (!crtc->active)
3690 continue;
3691
3692 pipe = crtc->pipe;
3693
3694 /*
3695 * At this point, only the pipes more space than before are
3696 * left to re-allocate.
3697 */
3698 if (reallocated[pipe])
3699 continue;
3700
d21b795c 3701 skl_wm_flush_pipe(dev_priv, pipe, 3);
0e8fb7ba
DL		 3702 }
3703}
3704
55994c2c
MR		 3705static int skl_update_pipe_wm(struct drm_crtc_state *cstate,
3706 struct skl_ddb_allocation *ddb, /* out */
3707 struct skl_pipe_wm *pipe_wm, /* out */
3708 bool *changed /* out */)
2d41c0b5 3709{
f4a96752
MR		 3710 struct intel_crtc *intel_crtc = to_intel_crtc(cstate->crtc);
3711 struct intel_crtc_state *intel_cstate = to_intel_crtc_state(cstate);
55994c2c 3712 int ret;
2d41c0b5 3713
55994c2c
MR		 3714 ret = skl_build_pipe_wm(intel_cstate, ddb, pipe_wm);
3715 if (ret)
3716 return ret;
2d41c0b5 3717
4e0963c7 3718 if (!memcmp(&intel_crtc->wm.active.skl, pipe_wm, sizeof(*pipe_wm)))
55994c2c
MR		 3719 *changed = false;
3720 else
3721 *changed = true;
2d41c0b5 3722
55994c2c 3723 return 0;
2d41c0b5
PB		 3724}
3725
98d39494
MR		 3726static int
3727skl_compute_ddb(struct drm_atomic_state *state)
3728{
3729 struct drm_device *dev = state->dev;
3730 struct drm_i915_private *dev_priv = to_i915(dev);
3731 struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
3732 struct intel_crtc *intel_crtc;
734fa01f 3733 struct skl_ddb_allocation *ddb = &intel_state->wm_results.ddb;
98d39494
MR		 3734 unsigned realloc_pipes = dev_priv->active_crtcs;
3735 int ret;
3736
3737 /*
3738 * If this is our first atomic update following hardware readout,
3739 * we can't trust the DDB that the BIOS programmed for us. Let's
3740 * pretend that all pipes switched active status so that we'll
3741 * ensure a full DDB recompute.
3742 */
3743 if (dev_priv->wm.distrust_bios_wm)
3744 intel_state->active_pipe_changes = ~0;
3745
3746 /*
3747 * If the modeset changes which CRTC's are active, we need to
3748 * recompute the DDB allocation for *all* active pipes, even
3749 * those that weren't otherwise being modified in any way by this
3750 * atomic commit. Due to the shrinking of the per-pipe allocations
3751 * when new active CRTC's are added, it's possible for a pipe that
3752 * we were already using and aren't changing at all here to suddenly
3753 * become invalid if its DDB needs exceeds its new allocation.
3754 *
3755 * Note that if we wind up doing a full DDB recompute, we can't let
3756 * any other display updates race with this transaction, so we need
3757 * to grab the lock on *all* CRTC's.
3758 */
734fa01f 3759 if (intel_state->active_pipe_changes) {
98d39494 3760 realloc_pipes = ~0;
734fa01f
MR		 3761 intel_state->wm_results.dirty_pipes = ~0;
3762 }
98d39494
MR		 3763
3764 for_each_intel_crtc_mask(dev, intel_crtc, realloc_pipes) {
3765 struct intel_crtc_state *cstate;
3766
3767 cstate = intel_atomic_get_crtc_state(state, intel_crtc);
3768 if (IS_ERR(cstate))
3769 return PTR_ERR(cstate);
3770
734fa01f 3771 ret = skl_allocate_pipe_ddb(cstate, ddb);
98d39494
MR		 3772 if (ret)
3773 return ret;
3774 }
3775
3776 return 0;
3777}
3778
3779static int
3780skl_compute_wm(struct drm_atomic_state *state)
3781{
3782 struct drm_crtc *crtc;
3783 struct drm_crtc_state *cstate;
734fa01f
MR		 3784 struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
3785 struct skl_wm_values *results = &intel_state->wm_results;
3786 struct skl_pipe_wm *pipe_wm;
98d39494 3787 bool changed = false;
734fa01f 3788 int ret, i;
98d39494
MR		 3789
3790 /*
3791 * If this transaction isn't actually touching any CRTC's, don't
3792 * bother with watermark calculation. Note that if we pass this
3793 * test, we're guaranteed to hold at least one CRTC state mutex,
3794 * which means we can safely use values like dev_priv->active_crtcs
3795 * since any racing commits that want to update them would need to
3796 * hold _all_ CRTC state mutexes.
3797 */
3798 for_each_crtc_in_state(state, crtc, cstate, i)
3799 changed = true;
3800 if (!changed)
3801 return 0;
3802
734fa01f
MR		 3803 /* Clear all dirty flags */
3804 results->dirty_pipes = 0;
3805
98d39494
MR		 3806 ret = skl_compute_ddb(state);
3807 if (ret)
3808 return ret;
3809
734fa01f
MR		 3810 /*
3811 * Calculate WM's for all pipes that are part of this transaction.
3812 * Note that the DDB allocation above may have added more CRTC's that
3813 * weren't otherwise being modified (and set bits in dirty_pipes) if
3814 * pipe allocations had to change.
3815 *
3816 * FIXME: Now that we're doing this in the atomic check phase, we
3817 * should allow skl_update_pipe_wm() to return failure in cases where
3818 * no suitable watermark values can be found.
3819 */
3820 for_each_crtc_in_state(state, crtc, cstate, i) {
3821 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3822 struct intel_crtc_state *intel_cstate =
3823 to_intel_crtc_state(cstate);
3824
3825 pipe_wm = &intel_cstate->wm.skl.optimal;
3826 ret = skl_update_pipe_wm(cstate, &results->ddb, pipe_wm,
3827 &changed);
3828 if (ret)
3829 return ret;
3830
3831 if (changed)
3832 results->dirty_pipes |= drm_crtc_mask(crtc);
3833
3834 if ((results->dirty_pipes & drm_crtc_mask(crtc)) == 0)
3835 /* This pipe's WM's did not change */
3836 continue;
3837
3838 intel_cstate->update_wm_pre = true;
3839 skl_compute_wm_results(crtc->dev, pipe_wm, results, intel_crtc);
3840 }
3841
98d39494
MR		 3842 return 0;
3843}
3844
2d41c0b5
PB		 3845static void skl_update_wm(struct drm_crtc *crtc)
3846{
3847 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3848 struct drm_device *dev = crtc->dev;
3849 struct drm_i915_private *dev_priv = dev->dev_private;
2d41c0b5 3850 struct skl_wm_values *results = &dev_priv->wm.skl_results;
4e0963c7 3851 struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state);
e8f1f02e 3852 struct skl_pipe_wm *pipe_wm = &cstate->wm.skl.optimal;
adda50b8 3853
734fa01f 3854 if ((results->dirty_pipes & drm_crtc_mask(crtc)) == 0)
2d41c0b5
PB		 3855 return;
3856
734fa01f
MR		 3857 intel_crtc->wm.active.skl = *pipe_wm;
3858
3859 mutex_lock(&dev_priv->wm.wm_mutex);
2d41c0b5 3860
2d41c0b5 3861 skl_write_wm_values(dev_priv, results);
0e8fb7ba 3862 skl_flush_wm_values(dev_priv, results);
53b0deb4
DL		 3863
3864 /* store the new configuration */
3865 dev_priv->wm.skl_hw = *results;
734fa01f
MR		 3866
3867 mutex_unlock(&dev_priv->wm.wm_mutex);
2d41c0b5
PB		 3868}
3869
d890565c
VS		 3870static void ilk_compute_wm_config(struct drm_device *dev,
3871 struct intel_wm_config *config)
3872{
3873 struct intel_crtc *crtc;
3874
3875 /* Compute the currently _active_ config */
3876 for_each_intel_crtc(dev, crtc) {
3877 const struct intel_pipe_wm *wm = &crtc->wm.active.ilk;
3878
3879 if (!wm->pipe_enabled)
3880 continue;
3881
3882 config->sprites_enabled |= wm->sprites_enabled;
3883 config->sprites_scaled |= wm->sprites_scaled;
3884 config->num_pipes_active++;
3885 }
3886}
3887
ed4a6a7c 3888static void ilk_program_watermarks(struct drm_i915_private *dev_priv)
801bcfff 3889{
ed4a6a7c 3890 struct drm_device *dev = dev_priv->dev;
b9d5c839 3891 struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
820c1980 3892 struct ilk_wm_maximums max;
d890565c 3893 struct intel_wm_config config = {};
820c1980 3894 struct ilk_wm_values results = {};
77c122bc 3895 enum intel_ddb_partitioning partitioning;
261a27d1 3896
d890565c
VS		 3897 ilk_compute_wm_config(dev, &config);
3898
3899 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
3900 ilk_wm_merge(dev, &config, &max, &lp_wm_1_2);
a485bfb8
VS		 3901
3902 /* 5/6 split only in single pipe config on IVB+ */
ec98c8d1 3903 if (INTEL_INFO(dev)->gen >= 7 &&
d890565c
VS		 3904 config.num_pipes_active == 1 && config.sprites_enabled) {
3905 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
3906 ilk_wm_merge(dev, &config, &max, &lp_wm_5_6);
0362c781 3907
820c1980 3908 best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
861f3389 3909 } else {
198a1e9b 3910 best_lp_wm = &lp_wm_1_2;
861f3389
PZ		 3911 }
3912
198a1e9b 3913 partitioning = (best_lp_wm == &lp_wm_1_2) ?
77c122bc 3914 INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
801bcfff 3915
820c1980 3916 ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results);
609cedef 3917
820c1980 3918 ilk_write_wm_values(dev_priv, &results);
1011d8c4
PZ		 3919}
3920
ed4a6a7c 3921static void ilk_initial_watermarks(struct intel_crtc_state *cstate)
b9d5c839 3922{
ed4a6a7c
MR		 3923 struct drm_i915_private *dev_priv = to_i915(cstate->base.crtc->dev);
3924 struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc);
b9d5c839 3925
ed4a6a7c 3926 mutex_lock(&dev_priv->wm.wm_mutex);
e8f1f02e 3927 intel_crtc->wm.active.ilk = cstate->wm.ilk.intermediate;
ed4a6a7c
MR		 3928 ilk_program_watermarks(dev_priv);
3929 mutex_unlock(&dev_priv->wm.wm_mutex);
3930}
bf220452 3931
ed4a6a7c
MR		 3932static void ilk_optimize_watermarks(struct intel_crtc_state *cstate)
3933{
3934 struct drm_i915_private *dev_priv = to_i915(cstate->base.crtc->dev);
3935 struct intel_crtc *intel_crtc = to_intel_crtc(cstate->base.crtc);
bf220452 3936
ed4a6a7c
MR		 3937 mutex_lock(&dev_priv->wm.wm_mutex);
3938 if (cstate->wm.need_postvbl_update) {
e8f1f02e 3939 intel_crtc->wm.active.ilk = cstate->wm.ilk.optimal;
ed4a6a7c
MR		 3940 ilk_program_watermarks(dev_priv);
3941 }
3942 mutex_unlock(&dev_priv->wm.wm_mutex);
b9d5c839
VS		 3943}
3944
3078999f
PB		 3945static void skl_pipe_wm_active_state(uint32_t val,
3946 struct skl_pipe_wm *active,
3947 bool is_transwm,
3948 bool is_cursor,
3949 int i,
3950 int level)
3951{
3952 bool is_enabled = (val & PLANE_WM_EN) != 0;
3953
3954 if (!is_transwm) {
3955 if (!is_cursor) {
3956 active->wm[level].plane_en[i] = is_enabled;
3957 active->wm[level].plane_res_b[i] =
3958 val & PLANE_WM_BLOCKS_MASK;
3959 active->wm[level].plane_res_l[i] =
3960 (val >> PLANE_WM_LINES_SHIFT) &
3961 PLANE_WM_LINES_MASK;
3962 } else {
4969d33e
MR		 3963 active->wm[level].plane_en[PLANE_CURSOR] = is_enabled;
3964 active->wm[level].plane_res_b[PLANE_CURSOR] =
3078999f 3965 val & PLANE_WM_BLOCKS_MASK;
4969d33e 3966 active->wm[level].plane_res_l[PLANE_CURSOR] =
3078999f
PB		 3967 (val >> PLANE_WM_LINES_SHIFT) &
3968 PLANE_WM_LINES_MASK;
3969 }
3970 } else {
3971 if (!is_cursor) {
3972 active->trans_wm.plane_en[i] = is_enabled;
3973 active->trans_wm.plane_res_b[i] =
3974 val & PLANE_WM_BLOCKS_MASK;
3975 active->trans_wm.plane_res_l[i] =
3976 (val >> PLANE_WM_LINES_SHIFT) &
3977 PLANE_WM_LINES_MASK;
3978 } else {
4969d33e
MR		 3979 active->trans_wm.plane_en[PLANE_CURSOR] = is_enabled;
3980 active->trans_wm.plane_res_b[PLANE_CURSOR] =
3078999f 3981 val & PLANE_WM_BLOCKS_MASK;
4969d33e 3982 active->trans_wm.plane_res_l[PLANE_CURSOR] =
3078999f
PB		 3983 (val >> PLANE_WM_LINES_SHIFT) &
3984 PLANE_WM_LINES_MASK;
3985 }
3986 }
3987}
3988
3989static void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc)
3990{
3991 struct drm_device *dev = crtc->dev;
3992 struct drm_i915_private *dev_priv = dev->dev_private;
3993 struct skl_wm_values *hw = &dev_priv->wm.skl_hw;
3994 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
4e0963c7 3995 struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state);
e8f1f02e 3996 struct skl_pipe_wm *active = &cstate->wm.skl.optimal;
3078999f
PB		 3997 enum pipe pipe = intel_crtc->pipe;
3998 int level, i, max_level;
3999 uint32_t temp;
4000
4001 max_level = ilk_wm_max_level(dev);
4002
4003 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
4004
4005 for (level = 0; level <= max_level; level++) {
4006 for (i = 0; i < intel_num_planes(intel_crtc); i++)
4007 hw->plane[pipe][i][level] =
4008 I915_READ(PLANE_WM(pipe, i, level));
4969d33e 4009 hw->plane[pipe][PLANE_CURSOR][level] = I915_READ(CUR_WM(pipe, level));
3078999f
PB		 4010 }
4011
4012 for (i = 0; i < intel_num_planes(intel_crtc); i++)
4013 hw->plane_trans[pipe][i] = I915_READ(PLANE_WM_TRANS(pipe, i));
4969d33e 4014 hw->plane_trans[pipe][PLANE_CURSOR] = I915_READ(CUR_WM_TRANS(pipe));
3078999f 4015
3ef00284 4016 if (!intel_crtc->active)
3078999f
PB		 4017 return;
4018
2b4b9f35 4019 hw->dirty_pipes |= drm_crtc_mask(crtc);
3078999f
PB		 4020
4021 active->linetime = hw->wm_linetime[pipe];
4022
4023 for (level = 0; level <= max_level; level++) {
4024 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
4025 temp = hw->plane[pipe][i][level];
4026 skl_pipe_wm_active_state(temp, active, false,
4027 false, i, level);
4028 }
4969d33e 4029 temp = hw->plane[pipe][PLANE_CURSOR][level];
3078999f
PB		 4030 skl_pipe_wm_active_state(temp, active, false, true, i, level);
4031 }
4032
4033 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
4034 temp = hw->plane_trans[pipe][i];
4035 skl_pipe_wm_active_state(temp, active, true, false, i, 0);
4036 }
4037
4969d33e 4038 temp = hw->plane_trans[pipe][PLANE_CURSOR];
3078999f 4039 skl_pipe_wm_active_state(temp, active, true, true, i, 0);
4e0963c7
MR		 4040
4041 intel_crtc->wm.active.skl = *active;
3078999f
PB		 4042}
4043
4044void skl_wm_get_hw_state(struct drm_device *dev)
4045{
a269c583
DL		 4046 struct drm_i915_private *dev_priv = dev->dev_private;
4047 struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb;
3078999f 4048 struct drm_crtc *crtc;
a1de91e5 4049 struct intel_crtc *intel_crtc;
3078999f 4050
a269c583 4051 skl_ddb_get_hw_state(dev_priv, ddb);
3078999f
PB		 4052 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
4053 skl_pipe_wm_get_hw_state(crtc);
a1de91e5 4054
279e99d7
MR		 4055 if (dev_priv->active_crtcs) {
4056 /* Fully recompute DDB on first atomic commit */
4057 dev_priv->wm.distrust_bios_wm = true;
4058 } else {
4059 /* Easy/common case; just sanitize DDB now if everything off */
4060 memset(ddb, 0, sizeof(*ddb));
4061 }
4062
a1de91e5
MR		 4063 /* Calculate plane data rates */
4064 for_each_intel_crtc(dev, intel_crtc) {
4065 struct intel_crtc_state *cstate = intel_crtc->config;
4066 struct intel_plane *intel_plane;
4067
4068 for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) {
4069 const struct drm_plane_state *pstate =
4070 intel_plane->base.state;
4071 int id = skl_wm_plane_id(intel_plane);
4072
4073 cstate->wm.skl.plane_data_rate[id] =
4074 skl_plane_relative_data_rate(cstate, pstate, 0);
4075 cstate->wm.skl.plane_y_data_rate[id] =
4076 skl_plane_relative_data_rate(cstate, pstate, 1);
4077 }
4078 }
3078999f
PB		 4079}
4080
243e6a44
VS		 4081static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
4082{
4083 struct drm_device *dev = crtc->dev;
4084 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 4085 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44 4086 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
4e0963c7 4087 struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state);
e8f1f02e 4088 struct intel_pipe_wm *active = &cstate->wm.ilk.optimal;
243e6a44 4089 enum pipe pipe = intel_crtc->pipe;
f0f59a00 4090 static const i915_reg_t wm0_pipe_reg[] = {
243e6a44
VS		 4091 [PIPE_A] = WM0_PIPEA_ILK,
4092 [PIPE_B] = WM0_PIPEB_ILK,
4093 [PIPE_C] = WM0_PIPEC_IVB,
4094 };
4095
4096 hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
a42a5719 4097 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 4098 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
243e6a44 4099
15606534
VS		 4100 memset(active, 0, sizeof(*active));
4101
3ef00284 4102 active->pipe_enabled = intel_crtc->active;
2a44b76b
VS		 4103
4104 if (active->pipe_enabled) {
243e6a44
VS		 4105 u32 tmp = hw->wm_pipe[pipe];
4106
4107 /*
4108 * For active pipes LP0 watermark is marked as
4109 * enabled, and LP1+ watermaks as disabled since
4110 * we can't really reverse compute them in case
4111 * multiple pipes are active.
4112 */
4113 active->wm[0].enable = true;
4114 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
4115 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
4116 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
4117 active->linetime = hw->wm_linetime[pipe];
4118 } else {
4119 int level, max_level = ilk_wm_max_level(dev);
4120
4121 /*
4122 * For inactive pipes, all watermark levels
4123 * should be marked as enabled but zeroed,
4124 * which is what we'd compute them to.
4125 */
4126 for (level = 0; level <= max_level; level++)
4127 active->wm[level].enable = true;
4128 }
4e0963c7
MR		 4129
4130 intel_crtc->wm.active.ilk = *active;
243e6a44
VS		 4131}
4132
6eb1a681
VS		 4133#define _FW_WM(value, plane) \
4134 (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT)
4135#define _FW_WM_VLV(value, plane) \
4136 (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT)
4137
4138static void vlv_read_wm_values(struct drm_i915_private *dev_priv,
4139 struct vlv_wm_values *wm)
4140{
4141 enum pipe pipe;
4142 uint32_t tmp;
4143
4144 for_each_pipe(dev_priv, pipe) {
4145 tmp = I915_READ(VLV_DDL(pipe));
4146
4147 wm->ddl[pipe].primary =
4148 (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
4149 wm->ddl[pipe].cursor =
4150 (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
4151 wm->ddl[pipe].sprite[0] =
4152 (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
4153 wm->ddl[pipe].sprite[1] =
4154 (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
4155 }
4156
4157 tmp = I915_READ(DSPFW1);
4158 wm->sr.plane = _FW_WM(tmp, SR);
4159 wm->pipe[PIPE_B].cursor = _FW_WM(tmp, CURSORB);
4160 wm->pipe[PIPE_B].primary = _FW_WM_VLV(tmp, PLANEB);
4161 wm->pipe[PIPE_A].primary = _FW_WM_VLV(tmp, PLANEA);
4162
4163 tmp = I915_READ(DSPFW2);
4164 wm->pipe[PIPE_A].sprite[1] = _FW_WM_VLV(tmp, SPRITEB);
4165 wm->pipe[PIPE_A].cursor = _FW_WM(tmp, CURSORA);
4166 wm->pipe[PIPE_A].sprite[0] = _FW_WM_VLV(tmp, SPRITEA);
4167
4168 tmp = I915_READ(DSPFW3);
4169 wm->sr.cursor = _FW_WM(tmp, CURSOR_SR);
4170
4171 if (IS_CHERRYVIEW(dev_priv)) {
4172 tmp = I915_READ(DSPFW7_CHV);
4173 wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED);
4174 wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC);
4175
4176 tmp = I915_READ(DSPFW8_CHV);
4177 wm->pipe[PIPE_C].sprite[1] = _FW_WM_VLV(tmp, SPRITEF);
4178 wm->pipe[PIPE_C].sprite[0] = _FW_WM_VLV(tmp, SPRITEE);
4179
4180 tmp = I915_READ(DSPFW9_CHV);
4181 wm->pipe[PIPE_C].primary = _FW_WM_VLV(tmp, PLANEC);
4182 wm->pipe[PIPE_C].cursor = _FW_WM(tmp, CURSORC);
4183
4184 tmp = I915_READ(DSPHOWM);
4185 wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
4186 wm->pipe[PIPE_C].sprite[1] |= _FW_WM(tmp, SPRITEF_HI) << 8;
4187 wm->pipe[PIPE_C].sprite[0] |= _FW_WM(tmp, SPRITEE_HI) << 8;
4188 wm->pipe[PIPE_C].primary |= _FW_WM(tmp, PLANEC_HI) << 8;
4189 wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8;
4190 wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
4191 wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8;
4192 wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
4193 wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
4194 wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8;
4195 } else {
4196 tmp = I915_READ(DSPFW7);
4197 wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED);
4198 wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC);
4199
4200 tmp = I915_READ(DSPHOWM);
4201 wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
4202 wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8;
4203 wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
4204 wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8;
4205 wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
4206 wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
4207 wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8;
4208 }
4209}
4210
4211#undef _FW_WM
4212#undef _FW_WM_VLV
4213
4214void vlv_wm_get_hw_state(struct drm_device *dev)
4215{
4216 struct drm_i915_private *dev_priv = to_i915(dev);
4217 struct vlv_wm_values *wm = &dev_priv->wm.vlv;
4218 struct intel_plane *plane;
4219 enum pipe pipe;
4220 u32 val;
4221
4222 vlv_read_wm_values(dev_priv, wm);
4223
4224 for_each_intel_plane(dev, plane) {
4225 switch (plane->base.type) {
4226 int sprite;
4227 case DRM_PLANE_TYPE_CURSOR:
4228 plane->wm.fifo_size = 63;
4229 break;
4230 case DRM_PLANE_TYPE_PRIMARY:
4231 plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, 0);
4232 break;
4233 case DRM_PLANE_TYPE_OVERLAY:
4234 sprite = plane->plane;
4235 plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, sprite + 1);
4236 break;
4237 }
4238 }
4239
4240 wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN;
4241 wm->level = VLV_WM_LEVEL_PM2;
4242
4243 if (IS_CHERRYVIEW(dev_priv)) {
4244 mutex_lock(&dev_priv->rps.hw_lock);
4245
4246 val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ);
4247 if (val & DSP_MAXFIFO_PM5_ENABLE)
4248 wm->level = VLV_WM_LEVEL_PM5;
4249
58590c14
VS		 4250 /*
4251 * If DDR DVFS is disabled in the BIOS, Punit
4252 * will never ack the request. So if that happens
4253 * assume we don't have to enable/disable DDR DVFS
4254 * dynamically. To test that just set the REQ_ACK
4255 * bit to poke the Punit, but don't change the
4256 * HIGH/LOW bits so that we don't actually change
4257 * the current state.
4258 */
6eb1a681 4259 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
58590c14
VS		 4260 val |= FORCE_DDR_FREQ_REQ_ACK;
4261 vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);
4262
4263 if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
4264 FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) {
4265 DRM_DEBUG_KMS("Punit not acking DDR DVFS request, "
4266 "assuming DDR DVFS is disabled\n");
4267 dev_priv->wm.max_level = VLV_WM_LEVEL_PM5;
4268 } else {
4269 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
4270 if ((val & FORCE_DDR_HIGH_FREQ) == 0)
4271 wm->level = VLV_WM_LEVEL_DDR_DVFS;
4272 }
6eb1a681
VS		 4273
4274 mutex_unlock(&dev_priv->rps.hw_lock);
4275 }
4276
4277 for_each_pipe(dev_priv, pipe)
4278 DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n",
4279 pipe_name(pipe), wm->pipe[pipe].primary, wm->pipe[pipe].cursor,
4280 wm->pipe[pipe].sprite[0], wm->pipe[pipe].sprite[1]);
4281
4282 DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n",
4283 wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr);
4284}
4285
243e6a44
VS		 4286void ilk_wm_get_hw_state(struct drm_device *dev)
4287{
4288 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 4289 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS		 4290 struct drm_crtc *crtc;
4291
70e1e0ec 4292 for_each_crtc(dev, crtc)
243e6a44
VS		 4293 ilk_pipe_wm_get_hw_state(crtc);
4294
4295 hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
4296 hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
4297 hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
4298
4299 hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
cfa7698b
VS		 4300 if (INTEL_INFO(dev)->gen >= 7) {
4301 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
4302 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
4303 }
243e6a44 4304
a42a5719 4305 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ac9545fd
VS		 4306 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
4307 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
4308 else if (IS_IVYBRIDGE(dev))
4309 hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
4310 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
243e6a44
VS		 4311
4312 hw->enable_fbc_wm =
4313 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
4314}
4315
b445e3b0
ED		 4316/**
4317 * intel_update_watermarks - update FIFO watermark values based on current modes
4318 *
4319 * Calculate watermark values for the various WM regs based on current mode
4320 * and plane configuration.
4321 *
4322 * There are several cases to deal with here:
4323 * - normal (i.e. non-self-refresh)
4324 * - self-refresh (SR) mode
4325 * - lines are large relative to FIFO size (buffer can hold up to 2)
4326 * - lines are small relative to FIFO size (buffer can hold more than 2
4327 * lines), so need to account for TLB latency
4328 *
4329 * The normal calculation is:
4330 * watermark = dotclock * bytes per pixel * latency
4331 * where latency is platform & configuration dependent (we assume pessimal
4332 * values here).
4333 *
4334 * The SR calculation is:
4335 * watermark = (trunc(latency/line time)+1) * surface width *
4336 * bytes per pixel
4337 * where
4338 * line time = htotal / dotclock
4339 * surface width = hdisplay for normal plane and 64 for cursor
4340 * and latency is assumed to be high, as above.
4341 *
4342 * The final value programmed to the register should always be rounded up,
4343 * and include an extra 2 entries to account for clock crossings.
4344 *
4345 * We don't use the sprite, so we can ignore that. And on Crestline we have
4346 * to set the non-SR watermarks to 8.
4347 */
46ba614c 4348void intel_update_watermarks(struct drm_crtc *crtc)
b445e3b0 4349{
46ba614c 4350 struct drm_i915_private *dev_priv = crtc->dev->dev_private;
b445e3b0
ED		 4351
4352 if (dev_priv->display.update_wm)
46ba614c 4353 dev_priv->display.update_wm(crtc);
b445e3b0
ED		 4354}
4355
e2828914 4356/*
9270388e 4357 * Lock protecting IPS related data structures
9270388e
DV		 4358 */
4359DEFINE_SPINLOCK(mchdev_lock);
4360
4361/* Global for IPS driver to get at the current i915 device. Protected by
4362 * mchdev_lock. */
4363static struct drm_i915_private *i915_mch_dev;
4364
91d14251 4365bool ironlake_set_drps(struct drm_i915_private *dev_priv, u8 val)
2b4e57bd 4366{
2b4e57bd
ED		 4367 u16 rgvswctl;
4368
9270388e
DV		 4369 assert_spin_locked(&mchdev_lock);
4370
2b4e57bd
ED		 4371 rgvswctl = I915_READ16(MEMSWCTL);
4372 if (rgvswctl & MEMCTL_CMD_STS) {
4373 DRM_DEBUG("gpu busy, RCS change rejected\n");
4374 return false; /* still busy with another command */
4375 }
4376
4377 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
4378 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
4379 I915_WRITE16(MEMSWCTL, rgvswctl);
4380 POSTING_READ16(MEMSWCTL);
4381
4382 rgvswctl |= MEMCTL_CMD_STS;
4383 I915_WRITE16(MEMSWCTL, rgvswctl);
4384
4385 return true;
4386}
4387
91d14251 4388static void ironlake_enable_drps(struct drm_i915_private *dev_priv)
2b4e57bd 4389{
84f1b20f 4390 u32 rgvmodectl;
2b4e57bd
ED		 4391 u8 fmax, fmin, fstart, vstart;
4392
9270388e
DV		 4393 spin_lock_irq(&mchdev_lock);
4394
84f1b20f
TU		 4395 rgvmodectl = I915_READ(MEMMODECTL);
4396
2b4e57bd
ED		 4397 /* Enable temp reporting */
4398 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
4399 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
4400
4401 /* 100ms RC evaluation intervals */
4402 I915_WRITE(RCUPEI, 100000);
4403 I915_WRITE(RCDNEI, 100000);
4404
4405 /* Set max/min thresholds to 90ms and 80ms respectively */
4406 I915_WRITE(RCBMAXAVG, 90000);
4407 I915_WRITE(RCBMINAVG, 80000);
4408
4409 I915_WRITE(MEMIHYST, 1);
4410
4411 /* Set up min, max, and cur for interrupt handling */
4412 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
4413 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
4414 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
4415 MEMMODE_FSTART_SHIFT;
4416
616847e7 4417 vstart = (I915_READ(PXVFREQ(fstart)) & PXVFREQ_PX_MASK) >>
2b4e57bd
ED		 4418 PXVFREQ_PX_SHIFT;
4419
20e4d407
DV		 4420 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
4421 dev_priv->ips.fstart = fstart;
2b4e57bd 4422
20e4d407
DV		 4423 dev_priv->ips.max_delay = fstart;
4424 dev_priv->ips.min_delay = fmin;
4425 dev_priv->ips.cur_delay = fstart;
2b4e57bd
ED		 4426
4427 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
4428 fmax, fmin, fstart);
4429
4430 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
4431
4432 /*
4433 * Interrupts will be enabled in ironlake_irq_postinstall
4434 */
4435
4436 I915_WRITE(VIDSTART, vstart);
4437 POSTING_READ(VIDSTART);
4438
4439 rgvmodectl |= MEMMODE_SWMODE_EN;
4440 I915_WRITE(MEMMODECTL, rgvmodectl);
4441
9270388e 4442 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2b4e57bd 4443 DRM_ERROR("stuck trying to change perf mode\n");
dd92d8de 4444 mdelay(1);
2b4e57bd 4445
91d14251 4446 ironlake_set_drps(dev_priv, fstart);
2b4e57bd 4447
7d81c3e0
VS		 4448 dev_priv->ips.last_count1 = I915_READ(DMIEC) +
4449 I915_READ(DDREC) + I915_READ(CSIEC);
20e4d407 4450 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
7d81c3e0 4451 dev_priv->ips.last_count2 = I915_READ(GFXEC);
5ed0bdf2 4452 dev_priv->ips.last_time2 = ktime_get_raw_ns();
9270388e
DV		 4453
4454 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED		 4455}
4456
91d14251 4457static void ironlake_disable_drps(struct drm_i915_private *dev_priv)
2b4e57bd 4458{
9270388e
DV		 4459 u16 rgvswctl;
4460
4461 spin_lock_irq(&mchdev_lock);
4462
4463 rgvswctl = I915_READ16(MEMSWCTL);
2b4e57bd
ED		 4464
4465 /* Ack interrupts, disable EFC interrupt */
4466 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
4467 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
4468 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
4469 I915_WRITE(DEIIR, DE_PCU_EVENT);
4470 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
4471
4472 /* Go back to the starting frequency */
91d14251 4473 ironlake_set_drps(dev_priv, dev_priv->ips.fstart);
dd92d8de 4474 mdelay(1);
2b4e57bd
ED		 4475 rgvswctl |= MEMCTL_CMD_STS;
4476 I915_WRITE(MEMSWCTL, rgvswctl);
dd92d8de 4477 mdelay(1);
2b4e57bd 4478
9270388e 4479 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED		 4480}
4481
acbe9475
DV		 4482/* There's a funny hw issue where the hw returns all 0 when reading from
4483 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
4484 * ourselves, instead of doing a rmw cycle (which might result in us clearing
4485 * all limits and the gpu stuck at whatever frequency it is at atm).
4486 */
74ef1173 4487static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val)
2b4e57bd 4488{
7b9e0ae6 4489 u32 limits;
2b4e57bd 4490
20b46e59
DV		 4491 /* Only set the down limit when we've reached the lowest level to avoid
4492 * getting more interrupts, otherwise leave this clear. This prevents a
4493 * race in the hw when coming out of rc6: There's a tiny window where
4494 * the hw runs at the minimal clock before selecting the desired
4495 * frequency, if the down threshold expires in that window we will not
4496 * receive a down interrupt. */
2d1fe073 4497 if (IS_GEN9(dev_priv)) {
74ef1173
AG		 4498 limits = (dev_priv->rps.max_freq_softlimit) << 23;
4499 if (val <= dev_priv->rps.min_freq_softlimit)
4500 limits |= (dev_priv->rps.min_freq_softlimit) << 14;
4501 } else {
4502 limits = dev_priv->rps.max_freq_softlimit << 24;
4503 if (val <= dev_priv->rps.min_freq_softlimit)
4504 limits |= dev_priv->rps.min_freq_softlimit << 16;
4505 }
20b46e59
DV		 4506
4507 return limits;
4508}
4509
dd75fdc8
CW		 4510static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
4511{
4512 int new_power;
8a586437
AG		 4513 u32 threshold_up = 0, threshold_down = 0; /* in % */
4514 u32 ei_up = 0, ei_down = 0;
dd75fdc8
CW		 4515
4516 new_power = dev_priv->rps.power;
4517 switch (dev_priv->rps.power) {
4518 case LOW_POWER:
b39fb297 4519 if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq)
dd75fdc8
CW		 4520 new_power = BETWEEN;
4521 break;
4522
4523 case BETWEEN:
b39fb297 4524 if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq)
dd75fdc8 4525 new_power = LOW_POWER;
b39fb297 4526 else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq)
dd75fdc8
CW		 4527 new_power = HIGH_POWER;
4528 break;
4529
4530 case HIGH_POWER:
b39fb297 4531 if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq)
dd75fdc8
CW		 4532 new_power = BETWEEN;
4533 break;
4534 }
4535 /* Max/min bins are special */
aed242ff 4536 if (val <= dev_priv->rps.min_freq_softlimit)
dd75fdc8 4537 new_power = LOW_POWER;
aed242ff 4538 if (val >= dev_priv->rps.max_freq_softlimit)
dd75fdc8
CW		 4539 new_power = HIGH_POWER;
4540 if (new_power == dev_priv->rps.power)
4541 return;
4542
4543 /* Note the units here are not exactly 1us, but 1280ns. */
4544 switch (new_power) {
4545 case LOW_POWER:
4546 /* Upclock if more than 95% busy over 16ms */
8a586437
AG		 4547 ei_up = 16000;
4548 threshold_up = 95;
dd75fdc8
CW		 4549
4550 /* Downclock if less than 85% busy over 32ms */
8a586437
AG		 4551 ei_down = 32000;
4552 threshold_down = 85;
dd75fdc8
CW		 4553 break;
4554
4555 case BETWEEN:
4556 /* Upclock if more than 90% busy over 13ms */
8a586437
AG		 4557 ei_up = 13000;
4558 threshold_up = 90;
dd75fdc8
CW		 4559
4560 /* Downclock if less than 75% busy over 32ms */
8a586437
AG		 4561 ei_down = 32000;
4562 threshold_down = 75;
dd75fdc8
CW		 4563 break;
4564
4565 case HIGH_POWER:
4566 /* Upclock if more than 85% busy over 10ms */
8a586437
AG		 4567 ei_up = 10000;
4568 threshold_up = 85;
dd75fdc8
CW		 4569
4570 /* Downclock if less than 60% busy over 32ms */
8a586437
AG		 4571 ei_down = 32000;
4572 threshold_down = 60;
dd75fdc8
CW		 4573 break;
4574 }
4575
8a586437
AG		 4576 I915_WRITE(GEN6_RP_UP_EI,
4577 GT_INTERVAL_FROM_US(dev_priv, ei_up));
4578 I915_WRITE(GEN6_RP_UP_THRESHOLD,
4579 GT_INTERVAL_FROM_US(dev_priv, (ei_up * threshold_up / 100)));
4580
4581 I915_WRITE(GEN6_RP_DOWN_EI,
4582 GT_INTERVAL_FROM_US(dev_priv, ei_down));
4583 I915_WRITE(GEN6_RP_DOWN_THRESHOLD,
4584 GT_INTERVAL_FROM_US(dev_priv, (ei_down * threshold_down / 100)));
4585
4586 I915_WRITE(GEN6_RP_CONTROL,
4587 GEN6_RP_MEDIA_TURBO |
4588 GEN6_RP_MEDIA_HW_NORMAL_MODE |
4589 GEN6_RP_MEDIA_IS_GFX |
4590 GEN6_RP_ENABLE |
4591 GEN6_RP_UP_BUSY_AVG |
4592 GEN6_RP_DOWN_IDLE_AVG);
4593
dd75fdc8 4594 dev_priv->rps.power = new_power;
8fb55197
CW		 4595 dev_priv->rps.up_threshold = threshold_up;
4596 dev_priv->rps.down_threshold = threshold_down;
dd75fdc8
CW		 4597 dev_priv->rps.last_adj = 0;
4598}
4599
2876ce73
CW		 4600static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val)
4601{
4602 u32 mask = 0;
4603
4604 if (val > dev_priv->rps.min_freq_softlimit)
6f4b12f8 4605 mask |= GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT;
2876ce73 4606 if (val < dev_priv->rps.max_freq_softlimit)
6f4b12f8 4607 mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD;
2876ce73 4608
7b3c29f6
CW		 4609 mask &= dev_priv->pm_rps_events;
4610
59d02a1f 4611 return gen6_sanitize_rps_pm_mask(dev_priv, ~mask);
2876ce73
CW		 4612}
4613
b8a5ff8d
JM		 4614/* gen6_set_rps is called to update the frequency request, but should also be
4615 * called when the range (min_delay and max_delay) is modified so that we can
4616 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
dc97997a 4617static void gen6_set_rps(struct drm_i915_private *dev_priv, u8 val)
20b46e59 4618{
23eafea6 4619 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
dc97997a 4620 if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1))
23eafea6
SAK		 4621 return;
4622
4fc688ce 4623 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
aed242ff
CW		 4624 WARN_ON(val > dev_priv->rps.max_freq);
4625 WARN_ON(val < dev_priv->rps.min_freq);
004777cb 4626
eb64cad1
CW		 4627 /* min/max delay may still have been modified so be sure to
4628 * write the limits value.
4629 */
4630 if (val != dev_priv->rps.cur_freq) {
4631 gen6_set_rps_thresholds(dev_priv, val);
b8a5ff8d 4632
dc97997a 4633 if (IS_GEN9(dev_priv))
5704195c
AG		 4634 I915_WRITE(GEN6_RPNSWREQ,
4635 GEN9_FREQUENCY(val));
dc97997a 4636 else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
eb64cad1
CW		 4637 I915_WRITE(GEN6_RPNSWREQ,
4638 HSW_FREQUENCY(val));
4639 else
4640 I915_WRITE(GEN6_RPNSWREQ,
4641 GEN6_FREQUENCY(val) |
4642 GEN6_OFFSET(0) |
4643 GEN6_AGGRESSIVE_TURBO);
b8a5ff8d 4644 }
7b9e0ae6 4645
7b9e0ae6
CW		 4646 /* Make sure we continue to get interrupts
4647 * until we hit the minimum or maximum frequencies.
4648 */
74ef1173 4649 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val));
2876ce73 4650 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
7b9e0ae6 4651
d5570a72
BW		 4652 POSTING_READ(GEN6_RPNSWREQ);
4653
b39fb297 4654 dev_priv->rps.cur_freq = val;
0f94592e 4655 trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val));
2b4e57bd
ED		 4656}
4657
dc97997a 4658static void valleyview_set_rps(struct drm_i915_private *dev_priv, u8 val)
ffe02b40 4659{
ffe02b40 4660 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
aed242ff
CW		 4661 WARN_ON(val > dev_priv->rps.max_freq);
4662 WARN_ON(val < dev_priv->rps.min_freq);
ffe02b40 4663
dc97997a 4664 if (WARN_ONCE(IS_CHERRYVIEW(dev_priv) && (val & 1),
ffe02b40
VS		 4665 "Odd GPU freq value\n"))
4666 val &= ~1;
4667
cd25dd5b
D		 4668 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
4669
8fb55197 4670 if (val != dev_priv->rps.cur_freq) {
ffe02b40 4671 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
8fb55197
CW		 4672 if (!IS_CHERRYVIEW(dev_priv))
4673 gen6_set_rps_thresholds(dev_priv, val);
4674 }
ffe02b40 4675
ffe02b40
VS		 4676 dev_priv->rps.cur_freq = val;
4677 trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val));
4678}
4679
a7f6e231 4680/* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down
76c3552f
D		 4681 *
4682 * * If Gfx is Idle, then
a7f6e231
D		 4683 * 1. Forcewake Media well.
4684 * 2. Request idle freq.
4685 * 3. Release Forcewake of Media well.
76c3552f
D		 4686*/
4687static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
4688{
aed242ff 4689 u32 val = dev_priv->rps.idle_freq;
5549d25f 4690
aed242ff 4691 if (dev_priv->rps.cur_freq <= val)
76c3552f
D		 4692 return;
4693
a7f6e231
D		 4694 /* Wake up the media well, as that takes a lot less
4695 * power than the Render well. */
4696 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA);
dc97997a 4697 valleyview_set_rps(dev_priv, val);
a7f6e231 4698 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA);
76c3552f
D		 4699}
4700
43cf3bf0
CW		 4701void gen6_rps_busy(struct drm_i915_private *dev_priv)
4702{
4703 mutex_lock(&dev_priv->rps.hw_lock);
4704 if (dev_priv->rps.enabled) {
4705 if (dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED))
4706 gen6_rps_reset_ei(dev_priv);
4707 I915_WRITE(GEN6_PMINTRMSK,
4708 gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq));
4709 }
4710 mutex_unlock(&dev_priv->rps.hw_lock);
4711}
4712
b29c19b6
CW		 4713void gen6_rps_idle(struct drm_i915_private *dev_priv)
4714{
4715 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 4716 if (dev_priv->rps.enabled) {
dc97997a 4717 if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv))
76c3552f 4718 vlv_set_rps_idle(dev_priv);
7526ed79 4719 else
dc97997a 4720 gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
c0951f0c 4721 dev_priv->rps.last_adj = 0;
43cf3bf0 4722 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
c0951f0c 4723 }
8d3afd7d 4724 mutex_unlock(&dev_priv->rps.hw_lock);
1854d5ca 4725
8d3afd7d 4726 spin_lock(&dev_priv->rps.client_lock);
1854d5ca
CW		 4727 while (!list_empty(&dev_priv->rps.clients))
4728 list_del_init(dev_priv->rps.clients.next);
8d3afd7d 4729 spin_unlock(&dev_priv->rps.client_lock);
b29c19b6
CW		 4730}
4731
1854d5ca 4732void gen6_rps_boost(struct drm_i915_private *dev_priv,
e61b9958
CW		 4733 struct intel_rps_client *rps,
4734 unsigned long submitted)
b29c19b6 4735{
8d3afd7d
CW		 4736 /* This is intentionally racy! We peek at the state here, then
4737 * validate inside the RPS worker.
4738 */
4739 if (!(dev_priv->mm.busy &&
4740 dev_priv->rps.enabled &&
4741 dev_priv->rps.cur_freq < dev_priv->rps.max_freq_softlimit))
4742 return;
43cf3bf0 4743
e61b9958
CW		 4744 /* Force a RPS boost (and don't count it against the client) if
4745 * the GPU is severely congested.
4746 */
d0bc54f2 4747 if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES))
e61b9958
CW		 4748 rps = NULL;
4749
8d3afd7d
CW		 4750 spin_lock(&dev_priv->rps.client_lock);
4751 if (rps == NULL || list_empty(&rps->link)) {
4752 spin_lock_irq(&dev_priv->irq_lock);
4753 if (dev_priv->rps.interrupts_enabled) {
4754 dev_priv->rps.client_boost = true;
4755 queue_work(dev_priv->wq, &dev_priv->rps.work);
4756 }
4757 spin_unlock_irq(&dev_priv->irq_lock);
1854d5ca 4758
2e1b8730
CW		 4759 if (rps != NULL) {
4760 list_add(&rps->link, &dev_priv->rps.clients);
4761 rps->boosts++;
1854d5ca
CW		 4762 } else
4763 dev_priv->rps.boosts++;
c0951f0c 4764 }
8d3afd7d 4765 spin_unlock(&dev_priv->rps.client_lock);
b29c19b6
CW		 4766}
4767
dc97997a 4768void intel_set_rps(struct drm_i915_private *dev_priv, u8 val)
0a073b84 4769{
dc97997a
CW		 4770 if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv))
4771 valleyview_set_rps(dev_priv, val);
ffe02b40 4772 else
dc97997a 4773 gen6_set_rps(dev_priv, val);
0a073b84
JB		 4774}
4775
dc97997a 4776static void gen9_disable_rc6(struct drm_i915_private *dev_priv)
20e49366 4777{
20e49366 4778 I915_WRITE(GEN6_RC_CONTROL, 0);
38c23527 4779 I915_WRITE(GEN9_PG_ENABLE, 0);
20e49366
ZW		 4780}
4781
dc97997a 4782static void gen9_disable_rps(struct drm_i915_private *dev_priv)
2030d684 4783{
2030d684
AG		 4784 I915_WRITE(GEN6_RP_CONTROL, 0);
4785}
4786
dc97997a 4787static void gen6_disable_rps(struct drm_i915_private *dev_priv)
d20d4f0c 4788{
d20d4f0c 4789 I915_WRITE(GEN6_RC_CONTROL, 0);
44fc7d5c 4790 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
2030d684 4791 I915_WRITE(GEN6_RP_CONTROL, 0);
44fc7d5c
DV		 4792}
4793
dc97997a 4794static void cherryview_disable_rps(struct drm_i915_private *dev_priv)
38807746 4795{
38807746
D		 4796 I915_WRITE(GEN6_RC_CONTROL, 0);
4797}
4798
dc97997a 4799static void valleyview_disable_rps(struct drm_i915_private *dev_priv)
44fc7d5c 4800{
98a2e5f9
D		 4801 /* we're doing forcewake before Disabling RC6,
4802 * This what the BIOS expects when going into suspend */
59bad947 4803 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
98a2e5f9 4804
44fc7d5c 4805 I915_WRITE(GEN6_RC_CONTROL, 0);
d20d4f0c 4806
59bad947 4807 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
d20d4f0c
JB		 4808}
4809
dc97997a 4810static void intel_print_rc6_info(struct drm_i915_private *dev_priv, u32 mode)
dc39fff7 4811{
dc97997a 4812 if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
91ca689a
ID		 4813 if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1)))
4814 mode = GEN6_RC_CTL_RC6_ENABLE;
4815 else
4816 mode = 0;
4817 }
dc97997a 4818 if (HAS_RC6p(dev_priv))
58abf1da 4819 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n",
87ad3212
JN		 4820 onoff(mode & GEN6_RC_CTL_RC6_ENABLE),
4821 onoff(mode & GEN6_RC_CTL_RC6p_ENABLE),
4822 onoff(mode & GEN6_RC_CTL_RC6pp_ENABLE));
58abf1da
RV		 4823
4824 else
4825 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n",
87ad3212 4826 onoff(mode & GEN6_RC_CTL_RC6_ENABLE));
dc39fff7
BW		 4827}
4828
dc97997a 4829static bool bxt_check_bios_rc6_setup(struct drm_i915_private *dev_priv)
274008e8 4830{
72e96d64 4831 struct i915_ggtt *ggtt = &dev_priv->ggtt;
274008e8
SAK		 4832 bool enable_rc6 = true;
4833 unsigned long rc6_ctx_base;
4834
4835 if (!(I915_READ(RC6_LOCATION) & RC6_CTX_IN_DRAM)) {
4836 DRM_DEBUG_KMS("RC6 Base location not set properly.\n");
4837 enable_rc6 = false;
4838 }
4839
4840 /*
4841 * The exact context size is not known for BXT, so assume a page size
4842 * for this check.
4843 */
4844 rc6_ctx_base = I915_READ(RC6_CTX_BASE) & RC6_CTX_BASE_MASK;
72e96d64
JL		 4845 if (!((rc6_ctx_base >= ggtt->stolen_reserved_base) &&
4846 (rc6_ctx_base + PAGE_SIZE <= ggtt->stolen_reserved_base +
4847 ggtt->stolen_reserved_size))) {
274008e8
SAK		 4848 DRM_DEBUG_KMS("RC6 Base address not as expected.\n");
4849 enable_rc6 = false;
4850 }
4851
4852 if (!(((I915_READ(PWRCTX_MAXCNT_RCSUNIT) & IDLE_TIME_MASK) > 1) &&
4853 ((I915_READ(PWRCTX_MAXCNT_VCSUNIT0) & IDLE_TIME_MASK) > 1) &&
4854 ((I915_READ(PWRCTX_MAXCNT_BCSUNIT) & IDLE_TIME_MASK) > 1) &&
4855 ((I915_READ(PWRCTX_MAXCNT_VECSUNIT) & IDLE_TIME_MASK) > 1))) {
4856 DRM_DEBUG_KMS("Engine Idle wait time not set properly.\n");
4857 enable_rc6 = false;
4858 }
4859
4860 if (!(I915_READ(GEN6_RC_CONTROL) & (GEN6_RC_CTL_RC6_ENABLE |
4861 GEN6_RC_CTL_HW_ENABLE)) &&
4862 ((I915_READ(GEN6_RC_CONTROL) & GEN6_RC_CTL_HW_ENABLE) ||
4863 !(I915_READ(GEN6_RC_STATE) & RC6_STATE))) {
4864 DRM_DEBUG_KMS("HW/SW RC6 is not enabled by BIOS.\n");
4865 enable_rc6 = false;
4866 }
4867
4868 return enable_rc6;
4869}
4870
dc97997a 4871int sanitize_rc6_option(struct drm_i915_private *dev_priv, int enable_rc6)
2b4e57bd 4872{
e7d66d89 4873 /* No RC6 before Ironlake and code is gone for ilk. */
dc97997a 4874 if (INTEL_INFO(dev_priv)->gen < 6)
e6069ca8
ID		 4875 return 0;
4876
274008e8
SAK		 4877 if (!enable_rc6)
4878 return 0;
4879
dc97997a 4880 if (IS_BROXTON(dev_priv) && !bxt_check_bios_rc6_setup(dev_priv)) {
274008e8
SAK		 4881 DRM_INFO("RC6 disabled by BIOS\n");
4882 return 0;
4883 }
4884
456470eb 4885 /* Respect the kernel parameter if it is set */
e6069ca8
ID		 4886 if (enable_rc6 >= 0) {
4887 int mask;
4888
dc97997a 4889 if (HAS_RC6p(dev_priv))
e6069ca8
ID		 4890 mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE |
4891 INTEL_RC6pp_ENABLE;
4892 else
4893 mask = INTEL_RC6_ENABLE;
4894
4895 if ((enable_rc6 & mask) != enable_rc6)
8dfd1f04
DV		 4896 DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
4897 enable_rc6 & mask, enable_rc6, mask);
e6069ca8
ID		 4898
4899 return enable_rc6 & mask;
4900 }
2b4e57bd 4901
dc97997a 4902 if (IS_IVYBRIDGE(dev_priv))
cca84a1f 4903 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
8bade1ad
BW		 4904
4905 return INTEL_RC6_ENABLE;
2b4e57bd
ED		 4906}
4907
dc97997a 4908static void gen6_init_rps_frequencies(struct drm_i915_private *dev_priv)
3280e8b0 4909{
93ee2920
TR		 4910 uint32_t rp_state_cap;
4911 u32 ddcc_status = 0;
4912 int ret;
4913
3280e8b0
BW		 4914 /* All of these values are in units of 50MHz */
4915 dev_priv->rps.cur_freq = 0;
93ee2920 4916 /* static values from HW: RP0 > RP1 > RPn (min_freq) */
dc97997a 4917 if (IS_BROXTON(dev_priv)) {
35040562
BP		 4918 rp_state_cap = I915_READ(BXT_RP_STATE_CAP);
4919 dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff;
4920 dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff;
4921 dev_priv->rps.min_freq = (rp_state_cap >> 0) & 0xff;
4922 } else {
4923 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
4924 dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff;
4925 dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff;
4926 dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff;
4927 }
4928
3280e8b0
BW		 4929 /* hw_max = RP0 until we check for overclocking */
4930 dev_priv->rps.max_freq = dev_priv->rps.rp0_freq;
4931
93ee2920 4932 dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq;
dc97997a
CW		 4933 if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv) ||
4934 IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
93ee2920
TR		 4935 ret = sandybridge_pcode_read(dev_priv,
4936 HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL,
4937 &ddcc_status);
4938 if (0 == ret)
4939 dev_priv->rps.efficient_freq =
46efa4ab
TR		 4940 clamp_t(u8,
4941 ((ddcc_status >> 8) & 0xff),
4942 dev_priv->rps.min_freq,
4943 dev_priv->rps.max_freq);
93ee2920
TR		 4944 }
4945
dc97997a 4946 if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
c5e0688c
AG		 4947 /* Store the frequency values in 16.66 MHZ units, which is
4948 the natural hardware unit for SKL */
4949 dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER;
4950 dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER;
4951 dev_priv->rps.min_freq *= GEN9_FREQ_SCALER;
4952 dev_priv->rps.max_freq *= GEN9_FREQ_SCALER;
4953 dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER;
4954 }
4955
aed242ff
CW		 4956 dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
4957
3280e8b0
BW		 4958 /* Preserve min/max settings in case of re-init */
4959 if (dev_priv->rps.max_freq_softlimit == 0)
4960 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
4961
93ee2920 4962 if (dev_priv->rps.min_freq_softlimit == 0) {
dc97997a 4963 if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
93ee2920 4964 dev_priv->rps.min_freq_softlimit =
813b5e69
VS		 4965 max_t(int, dev_priv->rps.efficient_freq,
4966 intel_freq_opcode(dev_priv, 450));
93ee2920
TR		 4967 else
4968 dev_priv->rps.min_freq_softlimit =
4969 dev_priv->rps.min_freq;
4970 }
3280e8b0
BW		 4971}
4972
b6fef0ef 4973/* See the Gen9_GT_PM_Programming_Guide doc for the below */
dc97997a 4974static void gen9_enable_rps(struct drm_i915_private *dev_priv)
b6fef0ef 4975{
b6fef0ef
JB		 4976 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
4977
dc97997a 4978 gen6_init_rps_frequencies(dev_priv);
ba1c554c 4979
23eafea6 4980 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
dc97997a 4981 if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) {
2030d684
AG		 4982 /*
4983 * BIOS could leave the Hw Turbo enabled, so need to explicitly
4984 * clear out the Control register just to avoid inconsitency
4985 * with debugfs interface, which will show Turbo as enabled
4986 * only and that is not expected by the User after adding the
4987 * WaGsvDisableTurbo. Apart from this there is no problem even
4988 * if the Turbo is left enabled in the Control register, as the
4989 * Up/Down interrupts would remain masked.
4990 */
dc97997a 4991 gen9_disable_rps(dev_priv);
23eafea6
SAK		 4992 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
4993 return;
4994 }
4995
0beb059a
AG		 4996 /* Program defaults and thresholds for RPS*/
4997 I915_WRITE(GEN6_RC_VIDEO_FREQ,
4998 GEN9_FREQUENCY(dev_priv->rps.rp1_freq));
4999
5000 /* 1 second timeout*/
5001 I915_WRITE(GEN6_RP_DOWN_TIMEOUT,
5002 GT_INTERVAL_FROM_US(dev_priv, 1000000));
5003
b6fef0ef 5004 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa);
b6fef0ef 5005
0beb059a
AG		 5006 /* Leaning on the below call to gen6_set_rps to program/setup the
5007 * Up/Down EI & threshold registers, as well as the RP_CONTROL,
5008 * RP_INTERRUPT_LIMITS & RPNSWREQ registers */
5009 dev_priv->rps.power = HIGH_POWER; /* force a reset */
dc97997a 5010 gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
b6fef0ef
JB		 5011
5012 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
5013}
5014
dc97997a 5015static void gen9_enable_rc6(struct drm_i915_private *dev_priv)
20e49366 5016{
e2f80391 5017 struct intel_engine_cs *engine;
20e49366 5018 uint32_t rc6_mask = 0;
20e49366
ZW		 5019
5020 /* 1a: Software RC state - RC0 */
5021 I915_WRITE(GEN6_RC_STATE, 0);
5022
5023 /* 1b: Get forcewake during program sequence. Although the driver
5024 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
59bad947 5025 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
20e49366
ZW		 5026
5027 /* 2a: Disable RC states. */
5028 I915_WRITE(GEN6_RC_CONTROL, 0);
5029
5030 /* 2b: Program RC6 thresholds.*/
63a4dec2
SAK		 5031
5032 /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */
dc97997a 5033 if (IS_SKYLAKE(dev_priv))
63a4dec2
SAK		 5034 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16);
5035 else
5036 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16);
20e49366
ZW		 5037 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
5038 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
b4ac5afc 5039 for_each_engine(engine, dev_priv)
e2f80391 5040 I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10);
97c322e7 5041
dc97997a 5042 if (HAS_GUC_UCODE(dev_priv))
97c322e7
SAK		 5043 I915_WRITE(GUC_MAX_IDLE_COUNT, 0xA);
5044
20e49366 5045 I915_WRITE(GEN6_RC_SLEEP, 0);
20e49366 5046
38c23527
ZW		 5047 /* 2c: Program Coarse Power Gating Policies. */
5048 I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25);
5049 I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25);
5050
20e49366 5051 /* 3a: Enable RC6 */
dc97997a 5052 if (intel_enable_rc6() & INTEL_RC6_ENABLE)
20e49366 5053 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
87ad3212 5054 DRM_INFO("RC6 %s\n", onoff(rc6_mask & GEN6_RC_CTL_RC6_ENABLE));
3e7732a0 5055 /* WaRsUseTimeoutMode */
dc97997a
CW		 5056 if (IS_SKL_REVID(dev_priv, 0, SKL_REVID_D0) ||
5057 IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) {
3e7732a0 5058 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us */
e3429cd2
SAK		 5059 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
5060 GEN7_RC_CTL_TO_MODE |
5061 rc6_mask);
3e7732a0
SAK		 5062 } else {
5063 I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */
e3429cd2
SAK		 5064 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
5065 GEN6_RC_CTL_EI_MODE(1) |
5066 rc6_mask);
3e7732a0 5067 }
20e49366 5068
cb07bae0
SK		 5069 /*
5070 * 3b: Enable Coarse Power Gating only when RC6 is enabled.
f2d2fe95 5071 * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6.
cb07bae0 5072 */
dc97997a 5073 if (NEEDS_WaRsDisableCoarsePowerGating(dev_priv))
f2d2fe95
SAK		 5074 I915_WRITE(GEN9_PG_ENABLE, 0);
5075 else
5076 I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ?
5077 (GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE) : 0);
38c23527 5078
59bad947 5079 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
20e49366
ZW		 5080}
5081
dc97997a 5082static void gen8_enable_rps(struct drm_i915_private *dev_priv)
6edee7f3 5083{
e2f80391 5084 struct intel_engine_cs *engine;
93ee2920 5085 uint32_t rc6_mask = 0;
6edee7f3
BW		 5086
5087 /* 1a: Software RC state - RC0 */
5088 I915_WRITE(GEN6_RC_STATE, 0);
5089
5090 /* 1c & 1d: Get forcewake during program sequence. Although the driver
5091 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
59bad947 5092 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW		 5093
5094 /* 2a: Disable RC states. */
5095 I915_WRITE(GEN6_RC_CONTROL, 0);
5096
93ee2920 5097 /* Initialize rps frequencies */
dc97997a 5098 gen6_init_rps_frequencies(dev_priv);
6edee7f3
BW		 5099
5100 /* 2b: Program RC6 thresholds.*/
5101 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
5102 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
5103 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
b4ac5afc 5104 for_each_engine(engine, dev_priv)
e2f80391 5105 I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10);
6edee7f3 5106 I915_WRITE(GEN6_RC_SLEEP, 0);
dc97997a 5107 if (IS_BROADWELL(dev_priv))
0d68b25e
TR		 5108 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */
5109 else
5110 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
6edee7f3
BW		 5111
5112 /* 3: Enable RC6 */
dc97997a 5113 if (intel_enable_rc6() & INTEL_RC6_ENABLE)
6edee7f3 5114 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
dc97997a
CW		 5115 intel_print_rc6_info(dev_priv, rc6_mask);
5116 if (IS_BROADWELL(dev_priv))
0d68b25e
TR		 5117 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
5118 GEN7_RC_CTL_TO_MODE |
5119 rc6_mask);
5120 else
5121 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
5122 GEN6_RC_CTL_EI_MODE(1) |
5123 rc6_mask);
6edee7f3
BW		 5124
5125 /* 4 Program defaults and thresholds for RPS*/
f9bdc585
BW		 5126 I915_WRITE(GEN6_RPNSWREQ,
5127 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
5128 I915_WRITE(GEN6_RC_VIDEO_FREQ,
5129 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
7526ed79
DV		 5130 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
5131 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
5132
5133 /* Docs recommend 900MHz, and 300 MHz respectively */
5134 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
5135 dev_priv->rps.max_freq_softlimit << 24 |
5136 dev_priv->rps.min_freq_softlimit << 16);
5137
5138 I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
5139 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
5140 I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
5141 I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
5142
5143 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
6edee7f3
BW		 5144
5145 /* 5: Enable RPS */
7526ed79
DV		 5146 I915_WRITE(GEN6_RP_CONTROL,
5147 GEN6_RP_MEDIA_TURBO |
5148 GEN6_RP_MEDIA_HW_NORMAL_MODE |
5149 GEN6_RP_MEDIA_IS_GFX |
5150 GEN6_RP_ENABLE |
5151 GEN6_RP_UP_BUSY_AVG |
5152 GEN6_RP_DOWN_IDLE_AVG);
5153
5154 /* 6: Ring frequency + overclocking (our driver does this later */
5155
c7f3153a 5156 dev_priv->rps.power = HIGH_POWER; /* force a reset */
dc97997a 5157 gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
7526ed79 5158
59bad947 5159 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW		 5160}
5161
dc97997a 5162static void gen6_enable_rps(struct drm_i915_private *dev_priv)
2b4e57bd 5163{
e2f80391 5164 struct intel_engine_cs *engine;
d060c169 5165 u32 rc6vids, pcu_mbox = 0, rc6_mask = 0;
2b4e57bd 5166 u32 gtfifodbg;
2b4e57bd 5167 int rc6_mode;
b4ac5afc 5168 int ret;
2b4e57bd 5169
4fc688ce 5170 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 5171
2b4e57bd
ED		 5172 /* Here begins a magic sequence of register writes to enable
5173 * auto-downclocking.
5174 *
5175 * Perhaps there might be some value in exposing these to
5176 * userspace...
5177 */
5178 I915_WRITE(GEN6_RC_STATE, 0);
2b4e57bd
ED		 5179
5180 /* Clear the DBG now so we don't confuse earlier errors */
297b32ec
VS		 5181 gtfifodbg = I915_READ(GTFIFODBG);
5182 if (gtfifodbg) {
2b4e57bd
ED		 5183 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
5184 I915_WRITE(GTFIFODBG, gtfifodbg);
5185 }
5186
59bad947 5187 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
2b4e57bd 5188
93ee2920 5189 /* Initialize rps frequencies */
dc97997a 5190 gen6_init_rps_frequencies(dev_priv);
dd0a1aa1 5191
2b4e57bd
ED		 5192 /* disable the counters and set deterministic thresholds */
5193 I915_WRITE(GEN6_RC_CONTROL, 0);
5194
5195 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
5196 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
5197 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
5198 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
5199 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
5200
b4ac5afc 5201 for_each_engine(engine, dev_priv)
e2f80391 5202 I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10);
2b4e57bd
ED		 5203
5204 I915_WRITE(GEN6_RC_SLEEP, 0);
5205 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
dc97997a 5206 if (IS_IVYBRIDGE(dev_priv))
351aa566
SM		 5207 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
5208 else
5209 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
0920a487 5210 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2b4e57bd
ED		 5211 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
5212
5a7dc92a 5213 /* Check if we are enabling RC6 */
dc97997a 5214 rc6_mode = intel_enable_rc6();
2b4e57bd
ED		 5215 if (rc6_mode & INTEL_RC6_ENABLE)
5216 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
5217
5a7dc92a 5218 /* We don't use those on Haswell */
dc97997a 5219 if (!IS_HASWELL(dev_priv)) {
5a7dc92a
ED		 5220 if (rc6_mode & INTEL_RC6p_ENABLE)
5221 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2b4e57bd 5222
5a7dc92a
ED		 5223 if (rc6_mode & INTEL_RC6pp_ENABLE)
5224 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
5225 }
2b4e57bd 5226
dc97997a 5227 intel_print_rc6_info(dev_priv, rc6_mask);
2b4e57bd
ED		 5228
5229 I915_WRITE(GEN6_RC_CONTROL,
5230 rc6_mask |
5231 GEN6_RC_CTL_EI_MODE(1) |
5232 GEN6_RC_CTL_HW_ENABLE);
5233
dd75fdc8
CW		 5234 /* Power down if completely idle for over 50ms */
5235 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
2b4e57bd 5236 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2b4e57bd 5237
42c0526c 5238 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
d060c169 5239 if (ret)
42c0526c 5240 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
d060c169
BW		 5241
5242 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
5243 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
5244 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
b39fb297 5245 (dev_priv->rps.max_freq_softlimit & 0xff) * 50,
d060c169 5246 (pcu_mbox & 0xff) * 50);
b39fb297 5247 dev_priv->rps.max_freq = pcu_mbox & 0xff;
2b4e57bd
ED		 5248 }
5249
dd75fdc8 5250 dev_priv->rps.power = HIGH_POWER; /* force a reset */
dc97997a 5251 gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
2b4e57bd 5252
31643d54
BW		 5253 rc6vids = 0;
5254 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
dc97997a 5255 if (IS_GEN6(dev_priv) && ret) {
31643d54 5256 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
dc97997a 5257 } else if (IS_GEN6(dev_priv) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
31643d54
BW		 5258 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
5259 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
5260 rc6vids &= 0xffff00;
5261 rc6vids |= GEN6_ENCODE_RC6_VID(450);
5262 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
5263 if (ret)
5264 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
5265 }
5266
59bad947 5267 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
2b4e57bd
ED		 5268}
5269
dc97997a 5270static void __gen6_update_ring_freq(struct drm_i915_private *dev_priv)
2b4e57bd
ED		 5271{
5272 int min_freq = 15;
3ebecd07
CW		 5273 unsigned int gpu_freq;
5274 unsigned int max_ia_freq, min_ring_freq;
4c8c7743 5275 unsigned int max_gpu_freq, min_gpu_freq;
2b4e57bd 5276 int scaling_factor = 180;
eda79642 5277 struct cpufreq_policy *policy;
2b4e57bd 5278
4fc688ce 5279 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 5280
eda79642
BW		 5281 policy = cpufreq_cpu_get(0);
5282 if (policy) {
5283 max_ia_freq = policy->cpuinfo.max_freq;
5284 cpufreq_cpu_put(policy);
5285 } else {
5286 /*
5287 * Default to measured freq if none found, PCU will ensure we
5288 * don't go over
5289 */
2b4e57bd 5290 max_ia_freq = tsc_khz;
eda79642 5291 }
2b4e57bd
ED		 5292
5293 /* Convert from kHz to MHz */
5294 max_ia_freq /= 1000;
5295
153b4b95 5296 min_ring_freq = I915_READ(DCLK) & 0xf;
f6aca45c
BW		 5297 /* convert DDR frequency from units of 266.6MHz to bandwidth */
5298 min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3ebecd07 5299
dc97997a 5300 if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
4c8c7743
AG		 5301 /* Convert GT frequency to 50 HZ units */
5302 min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER;
5303 max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER;
5304 } else {
5305 min_gpu_freq = dev_priv->rps.min_freq;
5306 max_gpu_freq = dev_priv->rps.max_freq;
5307 }
5308
2b4e57bd
ED		 5309 /*
5310 * For each potential GPU frequency, load a ring frequency we'd like
5311 * to use for memory access. We do this by specifying the IA frequency
5312 * the PCU should use as a reference to determine the ring frequency.
5313 */
4c8c7743
AG		 5314 for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) {
5315 int diff = max_gpu_freq - gpu_freq;
3ebecd07
CW		 5316 unsigned int ia_freq = 0, ring_freq = 0;
5317
dc97997a 5318 if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
4c8c7743
AG		 5319 /*
5320 * ring_freq = 2 * GT. ring_freq is in 100MHz units
5321 * No floor required for ring frequency on SKL.
5322 */
5323 ring_freq = gpu_freq;
dc97997a 5324 } else if (INTEL_INFO(dev_priv)->gen >= 8) {
46c764d4
BW		 5325 /* max(2 * GT, DDR). NB: GT is 50MHz units */
5326 ring_freq = max(min_ring_freq, gpu_freq);
dc97997a 5327 } else if (IS_HASWELL(dev_priv)) {
f6aca45c 5328 ring_freq = mult_frac(gpu_freq, 5, 4);
3ebecd07
CW		 5329 ring_freq = max(min_ring_freq, ring_freq);
5330 /* leave ia_freq as the default, chosen by cpufreq */
5331 } else {
5332 /* On older processors, there is no separate ring
5333 * clock domain, so in order to boost the bandwidth
5334 * of the ring, we need to upclock the CPU (ia_freq).
5335 *
5336 * For GPU frequencies less than 750MHz,
5337 * just use the lowest ring freq.
5338 */
5339 if (gpu_freq < min_freq)
5340 ia_freq = 800;
5341 else
5342 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
5343 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
5344 }
2b4e57bd 5345
42c0526c
BW		 5346 sandybridge_pcode_write(dev_priv,
5347 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3ebecd07
CW		 5348 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
5349 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
5350 gpu_freq);
2b4e57bd 5351 }
2b4e57bd
ED		 5352}
5353
dc97997a 5354void gen6_update_ring_freq(struct drm_i915_private *dev_priv)
c2bc2fc5 5355{
dc97997a 5356 if (!HAS_CORE_RING_FREQ(dev_priv))
c2bc2fc5
ID		 5357 return;
5358
5359 mutex_lock(&dev_priv->rps.hw_lock);
dc97997a 5360 __gen6_update_ring_freq(dev_priv);
c2bc2fc5
ID		 5361 mutex_unlock(&dev_priv->rps.hw_lock);
5362}
5363
03af2045 5364static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv)
2b6b3a09
D		 5365{
5366 u32 val, rp0;
5367
5b5929cb 5368 val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE);
2b6b3a09 5369
dc97997a 5370 switch (INTEL_INFO(dev_priv)->eu_total) {
5b5929cb
JN		 5371 case 8:
5372 /* (2 * 4) config */
5373 rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT);
5374 break;
5375 case 12:
5376 /* (2 * 6) config */
5377 rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT);
5378 break;
5379 case 16:
5380 /* (2 * 8) config */
5381 default:
5382 /* Setting (2 * 8) Min RP0 for any other combination */
5383 rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT);
5384 break;
095acd5f 5385 }
5b5929cb
JN		 5386
5387 rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK);
5388
2b6b3a09
D		 5389 return rp0;
5390}
5391
5392static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv)
5393{
5394 u32 val, rpe;
5395
5396 val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG);
5397 rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK;
5398
5399 return rpe;
5400}
5401
7707df4a
D		 5402static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv)
5403{
5404 u32 val, rp1;
5405
5b5929cb
JN		 5406 val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE);
5407 rp1 = (val & FB_GFX_FREQ_FUSE_MASK);
5408
7707df4a
D		 5409 return rp1;
5410}
5411
f8f2b001
D		 5412static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv)
5413{
5414 u32 val, rp1;
5415
5416 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
5417
5418 rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT;
5419
5420 return rp1;
5421}
5422
03af2045 5423static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
0a073b84
JB		 5424{
5425 u32 val, rp0;
5426
64936258 5427 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
0a073b84
JB		 5428
5429 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
5430 /* Clamp to max */
5431 rp0 = min_t(u32, rp0, 0xea);
5432
5433 return rp0;
5434}
5435
5436static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
5437{
5438 u32 val, rpe;
5439
64936258 5440 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
0a073b84 5441 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
64936258 5442 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
0a073b84
JB		 5443 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
5444
5445 return rpe;
5446}
5447
03af2045 5448static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
0a073b84 5449{
36146035
ID		 5450 u32 val;
5451
5452 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
5453 /*
5454 * According to the BYT Punit GPU turbo HAS 1.1.6.3 the minimum value
5455 * for the minimum frequency in GPLL mode is 0xc1. Contrary to this on
5456 * a BYT-M B0 the above register contains 0xbf. Moreover when setting
5457 * a frequency Punit will not allow values below 0xc0. Clamp it 0xc0
5458 * to make sure it matches what Punit accepts.
5459 */
5460 return max_t(u32, val, 0xc0);
0a073b84
JB		 5461}
5462
ae48434c
ID		 5463/* Check that the pctx buffer wasn't move under us. */
5464static void valleyview_check_pctx(struct drm_i915_private *dev_priv)
5465{
5466 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
5467
5468 WARN_ON(pctx_addr != dev_priv->mm.stolen_base +
5469 dev_priv->vlv_pctx->stolen->start);
5470}
5471
38807746
D		 5472
5473/* Check that the pcbr address is not empty. */
5474static void cherryview_check_pctx(struct drm_i915_private *dev_priv)
5475{
5476 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
5477
5478 WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0);
5479}
5480
dc97997a 5481static void cherryview_setup_pctx(struct drm_i915_private *dev_priv)
38807746 5482{
62106b4f 5483 struct i915_ggtt *ggtt = &dev_priv->ggtt;
72e96d64 5484 unsigned long pctx_paddr, paddr;
38807746
D		 5485 u32 pcbr;
5486 int pctx_size = 32*1024;
5487
38807746
D		 5488 pcbr = I915_READ(VLV_PCBR);
5489 if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) {
ce611ef8 5490 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
38807746 5491 paddr = (dev_priv->mm.stolen_base +
62106b4f 5492 (ggtt->stolen_size - pctx_size));
38807746
D		 5493
5494 pctx_paddr = (paddr & (~4095));
5495 I915_WRITE(VLV_PCBR, pctx_paddr);
5496 }
ce611ef8
VS		 5497
5498 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR));
38807746
D		 5499}
5500
dc97997a 5501static void valleyview_setup_pctx(struct drm_i915_private *dev_priv)
c9cddffc 5502{
c9cddffc
JB		 5503 struct drm_i915_gem_object *pctx;
5504 unsigned long pctx_paddr;
5505 u32 pcbr;
5506 int pctx_size = 24*1024;
5507
dc97997a 5508 mutex_lock(&dev_priv->dev->struct_mutex);
17b0c1f7 5509
c9cddffc
JB		 5510 pcbr = I915_READ(VLV_PCBR);
5511 if (pcbr) {
5512 /* BIOS set it up already, grab the pre-alloc'd space */
5513 int pcbr_offset;
5514
5515 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
5516 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
5517 pcbr_offset,
190d6cd5 5518 I915_GTT_OFFSET_NONE,
c9cddffc
JB		 5519 pctx_size);
5520 goto out;
5521 }
5522
ce611ef8
VS		 5523 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
5524
c9cddffc
JB		 5525 /*
5526 * From the Gunit register HAS:
5527 * The Gfx driver is expected to program this register and ensure
5528 * proper allocation within Gfx stolen memory. For example, this
5529 * register should be programmed such than the PCBR range does not
5530 * overlap with other ranges, such as the frame buffer, protected
5531 * memory, or any other relevant ranges.
5532 */
dc97997a 5533 pctx = i915_gem_object_create_stolen(dev_priv->dev, pctx_size);
c9cddffc
JB		 5534 if (!pctx) {
5535 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
ee504898 5536 goto out;
c9cddffc
JB		 5537 }
5538
5539 pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
5540 I915_WRITE(VLV_PCBR, pctx_paddr);
5541
5542out:
ce611ef8 5543 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR));
c9cddffc 5544 dev_priv->vlv_pctx = pctx;
dc97997a 5545 mutex_unlock(&dev_priv->dev->struct_mutex);
c9cddffc
JB		 5546}
5547
dc97997a 5548static void valleyview_cleanup_pctx(struct drm_i915_private *dev_priv)
ae48434c 5549{
ae48434c
ID		 5550 if (WARN_ON(!dev_priv->vlv_pctx))
5551 return;
5552
ee504898 5553 drm_gem_object_unreference_unlocked(&dev_priv->vlv_pctx->base);
ae48434c
ID		 5554 dev_priv->vlv_pctx = NULL;
5555}
5556
c30fec65
VS		 5557static void vlv_init_gpll_ref_freq(struct drm_i915_private *dev_priv)
5558{
5559 dev_priv->rps.gpll_ref_freq =
5560 vlv_get_cck_clock(dev_priv, "GPLL ref",
5561 CCK_GPLL_CLOCK_CONTROL,
5562 dev_priv->czclk_freq);
5563
5564 DRM_DEBUG_DRIVER("GPLL reference freq: %d kHz\n",
5565 dev_priv->rps.gpll_ref_freq);
5566}
5567
dc97997a 5568static void valleyview_init_gt_powersave(struct drm_i915_private *dev_priv)
4e80519e 5569{
2bb25c17 5570 u32 val;
4e80519e 5571
dc97997a 5572 valleyview_setup_pctx(dev_priv);
4e80519e 5573
c30fec65
VS		 5574 vlv_init_gpll_ref_freq(dev_priv);
5575
4e80519e
ID		 5576 mutex_lock(&dev_priv->rps.hw_lock);
5577
2bb25c17
VS		 5578 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5579 switch ((val >> 6) & 3) {
5580 case 0:
5581 case 1:
5582 dev_priv->mem_freq = 800;
5583 break;
5584 case 2:
5585 dev_priv->mem_freq = 1066;
5586 break;
5587 case 3:
5588 dev_priv->mem_freq = 1333;
5589 break;
5590 }
80b83b62 5591 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq);
2bb25c17 5592
4e80519e
ID		 5593 dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv);
5594 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
5595 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
7c59a9c1 5596 intel_gpu_freq(dev_priv, dev_priv->rps.max_freq),
4e80519e
ID		 5597 dev_priv->rps.max_freq);
5598
5599 dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv);
5600 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
7c59a9c1 5601 intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
4e80519e
ID		 5602 dev_priv->rps.efficient_freq);
5603
f8f2b001
D		 5604 dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv);
5605 DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n",
7c59a9c1 5606 intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq),
f8f2b001
D		 5607 dev_priv->rps.rp1_freq);
5608
4e80519e
ID		 5609 dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv);
5610 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
7c59a9c1 5611 intel_gpu_freq(dev_priv, dev_priv->rps.min_freq),
4e80519e
ID		 5612 dev_priv->rps.min_freq);
5613
aed242ff
CW		 5614 dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
5615
4e80519e
ID		 5616 /* Preserve min/max settings in case of re-init */
5617 if (dev_priv->rps.max_freq_softlimit == 0)
5618 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
5619
5620 if (dev_priv->rps.min_freq_softlimit == 0)
5621 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
5622
5623 mutex_unlock(&dev_priv->rps.hw_lock);
5624}
5625
dc97997a 5626static void cherryview_init_gt_powersave(struct drm_i915_private *dev_priv)
38807746 5627{
2bb25c17 5628 u32 val;
2b6b3a09 5629
dc97997a 5630 cherryview_setup_pctx(dev_priv);
2b6b3a09 5631
c30fec65
VS		 5632 vlv_init_gpll_ref_freq(dev_priv);
5633
2b6b3a09
D		 5634 mutex_lock(&dev_priv->rps.hw_lock);
5635
a580516d 5636 mutex_lock(&dev_priv->sb_lock);
c6e8f39d 5637 val = vlv_cck_read(dev_priv, CCK_FUSE_REG);
a580516d 5638 mutex_unlock(&dev_priv->sb_lock);
c6e8f39d 5639
2bb25c17 5640 switch ((val >> 2) & 0x7) {
2bb25c17 5641 case 3:
2bb25c17
VS		 5642 dev_priv->mem_freq = 2000;
5643 break;
bfa7df01 5644 default:
2bb25c17
VS		 5645 dev_priv->mem_freq = 1600;
5646 break;
5647 }
80b83b62 5648 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq);
2bb25c17 5649
2b6b3a09
D		 5650 dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv);
5651 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
5652 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
7c59a9c1 5653 intel_gpu_freq(dev_priv, dev_priv->rps.max_freq),
2b6b3a09
D		 5654 dev_priv->rps.max_freq);
5655
5656 dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv);
5657 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
7c59a9c1 5658 intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
2b6b3a09
D		 5659 dev_priv->rps.efficient_freq);
5660
7707df4a
D		 5661 dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv);
5662 DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n",
7c59a9c1 5663 intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq),
7707df4a
D		 5664 dev_priv->rps.rp1_freq);
5665
5b7c91b7
D		 5666 /* PUnit validated range is only [RPe, RP0] */
5667 dev_priv->rps.min_freq = dev_priv->rps.efficient_freq;
2b6b3a09 5668 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
7c59a9c1 5669 intel_gpu_freq(dev_priv, dev_priv->rps.min_freq),
2b6b3a09
D		 5670 dev_priv->rps.min_freq);
5671
1c14762d
VS		 5672 WARN_ONCE((dev_priv->rps.max_freq |
5673 dev_priv->rps.efficient_freq |
5674 dev_priv->rps.rp1_freq |
5675 dev_priv->rps.min_freq) & 1,
5676 "Odd GPU freq values\n");
5677
aed242ff
CW		 5678 dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
5679
2b6b3a09
D		 5680 /* Preserve min/max settings in case of re-init */
5681 if (dev_priv->rps.max_freq_softlimit == 0)
5682 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
5683
5684 if (dev_priv->rps.min_freq_softlimit == 0)
5685 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
5686
5687 mutex_unlock(&dev_priv->rps.hw_lock);
38807746
D		 5688}
5689
dc97997a 5690static void valleyview_cleanup_gt_powersave(struct drm_i915_private *dev_priv)
4e80519e 5691{
dc97997a 5692 valleyview_cleanup_pctx(dev_priv);
4e80519e
ID		 5693}
5694
dc97997a 5695static void cherryview_enable_rps(struct drm_i915_private *dev_priv)
38807746 5696{
e2f80391 5697 struct intel_engine_cs *engine;
2b6b3a09 5698 u32 gtfifodbg, val, rc6_mode = 0, pcbr;
38807746
D		 5699
5700 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
5701
297b32ec
VS		 5702 gtfifodbg = I915_READ(GTFIFODBG) & ~(GT_FIFO_SBDEDICATE_FREE_ENTRY_CHV |
5703 GT_FIFO_FREE_ENTRIES_CHV);
38807746
D		 5704 if (gtfifodbg) {
5705 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5706 gtfifodbg);
5707 I915_WRITE(GTFIFODBG, gtfifodbg);
5708 }
5709
5710 cherryview_check_pctx(dev_priv);
5711
5712 /* 1a & 1b: Get forcewake during program sequence. Although the driver
5713 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
59bad947 5714 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
38807746 5715
160614a2
VS		 5716 /* Disable RC states. */
5717 I915_WRITE(GEN6_RC_CONTROL, 0);
5718
38807746
D		 5719 /* 2a: Program RC6 thresholds.*/
5720 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
5721 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
5722 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
5723
b4ac5afc 5724 for_each_engine(engine, dev_priv)
e2f80391 5725 I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10);
38807746
D		 5726 I915_WRITE(GEN6_RC_SLEEP, 0);
5727
f4f71c7d
D		 5728 /* TO threshold set to 500 us ( 0x186 * 1.28 us) */
5729 I915_WRITE(GEN6_RC6_THRESHOLD, 0x186);
38807746
D		 5730
5731 /* allows RC6 residency counter to work */
5732 I915_WRITE(VLV_COUNTER_CONTROL,
5733 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
5734 VLV_MEDIA_RC6_COUNT_EN |
5735 VLV_RENDER_RC6_COUNT_EN));
5736
5737 /* For now we assume BIOS is allocating and populating the PCBR */
5738 pcbr = I915_READ(VLV_PCBR);
5739
38807746 5740 /* 3: Enable RC6 */
dc97997a
CW		 5741 if ((intel_enable_rc6() & INTEL_RC6_ENABLE) &&
5742 (pcbr >> VLV_PCBR_ADDR_SHIFT))
af5a75a3 5743 rc6_mode = GEN7_RC_CTL_TO_MODE;
38807746
D		 5744
5745 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
5746
2b6b3a09 5747 /* 4 Program defaults and thresholds for RPS*/
3cbdb48f 5748 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
2b6b3a09
D		 5749 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
5750 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
5751 I915_WRITE(GEN6_RP_UP_EI, 66000);
5752 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
5753
5754 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
5755
5756 /* 5: Enable RPS */
5757 I915_WRITE(GEN6_RP_CONTROL,
5758 GEN6_RP_MEDIA_HW_NORMAL_MODE |
eb973a5e 5759 GEN6_RP_MEDIA_IS_GFX |
2b6b3a09
D		 5760 GEN6_RP_ENABLE |
5761 GEN6_RP_UP_BUSY_AVG |
5762 GEN6_RP_DOWN_IDLE_AVG);
5763
3ef62342
D		 5764 /* Setting Fixed Bias */
5765 val = VLV_OVERRIDE_EN |
5766 VLV_SOC_TDP_EN |
5767 CHV_BIAS_CPU_50_SOC_50;
5768 vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val);
5769
2b6b3a09
D		 5770 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5771
8d40c3ae
VS		 5772 /* RPS code assumes GPLL is used */
5773 WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n");
5774
742f491d 5775 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
2b6b3a09
D		 5776 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
5777
5778 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
5779 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
7c59a9c1 5780 intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
2b6b3a09
D		 5781 dev_priv->rps.cur_freq);
5782
5783 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
5fd9f523
VS		 5784 intel_gpu_freq(dev_priv, dev_priv->rps.idle_freq),
5785 dev_priv->rps.idle_freq);
2b6b3a09 5786
dc97997a 5787 valleyview_set_rps(dev_priv, dev_priv->rps.idle_freq);
2b6b3a09 5788
59bad947 5789 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
38807746
D		 5790}
5791
dc97997a 5792static void valleyview_enable_rps(struct drm_i915_private *dev_priv)
0a073b84 5793{
e2f80391 5794 struct intel_engine_cs *engine;
2a5913a8 5795 u32 gtfifodbg, val, rc6_mode = 0;
0a073b84
JB		 5796
5797 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
5798
ae48434c
ID		 5799 valleyview_check_pctx(dev_priv);
5800
297b32ec
VS		 5801 gtfifodbg = I915_READ(GTFIFODBG);
5802 if (gtfifodbg) {
f7d85c1e
JB		 5803 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5804 gtfifodbg);
0a073b84
JB		 5805 I915_WRITE(GTFIFODBG, gtfifodbg);
5806 }
5807
c8d9a590 5808 /* If VLV, Forcewake all wells, else re-direct to regular path */
59bad947 5809 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
0a073b84 5810
160614a2
VS		 5811 /* Disable RC states. */
5812 I915_WRITE(GEN6_RC_CONTROL, 0);
5813
cad725fe 5814 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
0a073b84
JB		 5815 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
5816 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
5817 I915_WRITE(GEN6_RP_UP_EI, 66000);
5818 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
5819
5820 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
5821
5822 I915_WRITE(GEN6_RP_CONTROL,
5823 GEN6_RP_MEDIA_TURBO |
5824 GEN6_RP_MEDIA_HW_NORMAL_MODE |
5825 GEN6_RP_MEDIA_IS_GFX |
5826 GEN6_RP_ENABLE |
5827 GEN6_RP_UP_BUSY_AVG |
5828 GEN6_RP_DOWN_IDLE_CONT);
5829
5830 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
5831 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
5832 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
5833
b4ac5afc 5834 for_each_engine(engine, dev_priv)
e2f80391 5835 I915_WRITE(RING_MAX_IDLE(engine->mmio_base), 10);
0a073b84 5836
2f0aa304 5837 I915_WRITE(GEN6_RC6_THRESHOLD, 0x557);
0a073b84
JB		 5838
5839 /* allows RC6 residency counter to work */
49798eb2 5840 I915_WRITE(VLV_COUNTER_CONTROL,
31685c25
D		 5841 _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN |
5842 VLV_RENDER_RC0_COUNT_EN |
49798eb2
JB		 5843 VLV_MEDIA_RC6_COUNT_EN |
5844 VLV_RENDER_RC6_COUNT_EN));
31685c25 5845
dc97997a 5846 if (intel_enable_rc6() & INTEL_RC6_ENABLE)
6b88f295 5847 rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
dc39fff7 5848
dc97997a 5849 intel_print_rc6_info(dev_priv, rc6_mode);
dc39fff7 5850
a2b23fe0 5851 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
0a073b84 5852
3ef62342
D		 5853 /* Setting Fixed Bias */
5854 val = VLV_OVERRIDE_EN |
5855 VLV_SOC_TDP_EN |
5856 VLV_BIAS_CPU_125_SOC_875;
5857 vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val);
5858
64936258 5859 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
0a073b84 5860
8d40c3ae
VS		 5861 /* RPS code assumes GPLL is used */
5862 WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n");
5863
742f491d 5864 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
0a073b84
JB		 5865 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
5866
b39fb297 5867 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
73008b98 5868 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
7c59a9c1 5869 intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
b39fb297 5870 dev_priv->rps.cur_freq);
0a073b84 5871
73008b98 5872 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
5fd9f523
VS		 5873 intel_gpu_freq(dev_priv, dev_priv->rps.idle_freq),
5874 dev_priv->rps.idle_freq);
0a073b84 5875
dc97997a 5876 valleyview_set_rps(dev_priv, dev_priv->rps.idle_freq);
0a073b84 5877
59bad947 5878 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
0a073b84
JB		 5879}
5880
dde18883
ED		 5881static unsigned long intel_pxfreq(u32 vidfreq)
5882{
5883 unsigned long freq;
5884 int div = (vidfreq & 0x3f0000) >> 16;
5885 int post = (vidfreq & 0x3000) >> 12;
5886 int pre = (vidfreq & 0x7);
5887
5888 if (!pre)
5889 return 0;
5890
5891 freq = ((div * 133333) / ((1<<post) * pre));
5892
5893 return freq;
5894}
5895
eb48eb00
DV		 5896static const struct cparams {
5897 u16 i;
5898 u16 t;
5899 u16 m;
5900 u16 c;
5901} cparams[] = {
5902 { 1, 1333, 301, 28664 },
5903 { 1, 1066, 294, 24460 },
5904 { 1, 800, 294, 25192 },
5905 { 0, 1333, 276, 27605 },
5906 { 0, 1066, 276, 27605 },
5907 { 0, 800, 231, 23784 },
5908};
5909
f531dcb2 5910static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 5911{
5912 u64 total_count, diff, ret;
5913 u32 count1, count2, count3, m = 0, c = 0;
5914 unsigned long now = jiffies_to_msecs(jiffies), diff1;
5915 int i;
5916
02d71956
DV		 5917 assert_spin_locked(&mchdev_lock);
5918
20e4d407 5919 diff1 = now - dev_priv->ips.last_time1;
eb48eb00
DV		 5920
5921 /* Prevent division-by-zero if we are asking too fast.
5922 * Also, we don't get interesting results if we are polling
5923 * faster than once in 10ms, so just return the saved value
5924 * in such cases.
5925 */
5926 if (diff1 <= 10)
20e4d407 5927 return dev_priv->ips.chipset_power;
eb48eb00
DV		 5928
5929 count1 = I915_READ(DMIEC);
5930 count2 = I915_READ(DDREC);
5931 count3 = I915_READ(CSIEC);
5932
5933 total_count = count1 + count2 + count3;
5934
5935 /* FIXME: handle per-counter overflow */
20e4d407
DV		 5936 if (total_count < dev_priv->ips.last_count1) {
5937 diff = ~0UL - dev_priv->ips.last_count1;
eb48eb00
DV		 5938 diff += total_count;
5939 } else {
20e4d407 5940 diff = total_count - dev_priv->ips.last_count1;
eb48eb00
DV		 5941 }
5942
5943 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
20e4d407
DV		 5944 if (cparams[i].i == dev_priv->ips.c_m &&
5945 cparams[i].t == dev_priv->ips.r_t) {
eb48eb00
DV		 5946 m = cparams[i].m;
5947 c = cparams[i].c;
5948 break;
5949 }
5950 }
5951
5952 diff = div_u64(diff, diff1);
5953 ret = ((m * diff) + c);
5954 ret = div_u64(ret, 10);
5955
20e4d407
DV		 5956 dev_priv->ips.last_count1 = total_count;
5957 dev_priv->ips.last_time1 = now;
eb48eb00 5958
20e4d407 5959 dev_priv->ips.chipset_power = ret;
eb48eb00
DV		 5960
5961 return ret;
5962}
5963
f531dcb2
CW		 5964unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
5965{
5966 unsigned long val;
5967
dc97997a 5968 if (INTEL_INFO(dev_priv)->gen != 5)
f531dcb2
CW		 5969 return 0;
5970
5971 spin_lock_irq(&mchdev_lock);
5972
5973 val = __i915_chipset_val(dev_priv);
5974
5975 spin_unlock_irq(&mchdev_lock);
5976
5977 return val;
5978}
5979
eb48eb00
DV		 5980unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
5981{
5982 unsigned long m, x, b;
5983 u32 tsfs;
5984
5985 tsfs = I915_READ(TSFS);
5986
5987 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
5988 x = I915_READ8(TR1);
5989
5990 b = tsfs & TSFS_INTR_MASK;
5991
5992 return ((m * x) / 127) - b;
5993}
5994
d972d6ee
MK		 5995static int _pxvid_to_vd(u8 pxvid)
5996{
5997 if (pxvid == 0)
5998 return 0;
5999
6000 if (pxvid >= 8 && pxvid < 31)
6001 pxvid = 31;
6002
6003 return (pxvid + 2) * 125;
6004}
6005
6006static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
eb48eb00 6007{
d972d6ee
MK		 6008 const int vd = _pxvid_to_vd(pxvid);
6009 const int vm = vd - 1125;
6010
dc97997a 6011 if (INTEL_INFO(dev_priv)->is_mobile)
d972d6ee
MK		 6012 return vm > 0 ? vm : 0;
6013
6014 return vd;
eb48eb00
DV		 6015}
6016
02d71956 6017static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00 6018{
5ed0bdf2 6019 u64 now, diff, diffms;
eb48eb00
DV		 6020 u32 count;
6021
02d71956 6022 assert_spin_locked(&mchdev_lock);
eb48eb00 6023
5ed0bdf2
TG		 6024 now = ktime_get_raw_ns();
6025 diffms = now - dev_priv->ips.last_time2;
6026 do_div(diffms, NSEC_PER_MSEC);
eb48eb00
DV		 6027
6028 /* Don't divide by 0 */
eb48eb00
DV		 6029 if (!diffms)
6030 return;
6031
6032 count = I915_READ(GFXEC);
6033
20e4d407
DV		 6034 if (count < dev_priv->ips.last_count2) {
6035 diff = ~0UL - dev_priv->ips.last_count2;
eb48eb00
DV		 6036 diff += count;
6037 } else {
20e4d407 6038 diff = count - dev_priv->ips.last_count2;
eb48eb00
DV		 6039 }
6040
20e4d407
DV		 6041 dev_priv->ips.last_count2 = count;
6042 dev_priv->ips.last_time2 = now;
eb48eb00
DV		 6043
6044 /* More magic constants... */
6045 diff = diff * 1181;
6046 diff = div_u64(diff, diffms * 10);
20e4d407 6047 dev_priv->ips.gfx_power = diff;
eb48eb00
DV		 6048}
6049
02d71956
DV		 6050void i915_update_gfx_val(struct drm_i915_private *dev_priv)
6051{
dc97997a 6052 if (INTEL_INFO(dev_priv)->gen != 5)
02d71956
DV		 6053 return;
6054
9270388e 6055 spin_lock_irq(&mchdev_lock);
02d71956
DV		 6056
6057 __i915_update_gfx_val(dev_priv);
6058
9270388e 6059 spin_unlock_irq(&mchdev_lock);
02d71956
DV		 6060}
6061
f531dcb2 6062static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 6063{
6064 unsigned long t, corr, state1, corr2, state2;
6065 u32 pxvid, ext_v;
6066
02d71956
DV		 6067 assert_spin_locked(&mchdev_lock);
6068
616847e7 6069 pxvid = I915_READ(PXVFREQ(dev_priv->rps.cur_freq));
eb48eb00
DV		 6070 pxvid = (pxvid >> 24) & 0x7f;
6071 ext_v = pvid_to_extvid(dev_priv, pxvid);
6072
6073 state1 = ext_v;
6074
6075 t = i915_mch_val(dev_priv);
6076
6077 /* Revel in the empirically derived constants */
6078
6079 /* Correction factor in 1/100000 units */
6080 if (t > 80)
6081 corr = ((t * 2349) + 135940);
6082 else if (t >= 50)
6083 corr = ((t * 964) + 29317);
6084 else /* < 50 */
6085 corr = ((t * 301) + 1004);
6086
6087 corr = corr * ((150142 * state1) / 10000 - 78642);
6088 corr /= 100000;
20e4d407 6089 corr2 = (corr * dev_priv->ips.corr);
eb48eb00
DV		 6090
6091 state2 = (corr2 * state1) / 10000;
6092 state2 /= 100; /* convert to mW */
6093
02d71956 6094 __i915_update_gfx_val(dev_priv);
eb48eb00 6095
20e4d407 6096 return dev_priv->ips.gfx_power + state2;
eb48eb00
DV		 6097}
6098
f531dcb2
CW		 6099unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
6100{
6101 unsigned long val;
6102
dc97997a 6103 if (INTEL_INFO(dev_priv)->gen != 5)
f531dcb2
CW		 6104 return 0;
6105
6106 spin_lock_irq(&mchdev_lock);
6107
6108 val = __i915_gfx_val(dev_priv);
6109
6110 spin_unlock_irq(&mchdev_lock);
6111
6112 return val;
6113}
6114
eb48eb00
DV		 6115/**
6116 * i915_read_mch_val - return value for IPS use
6117 *
6118 * Calculate and return a value for the IPS driver to use when deciding whether
6119 * we have thermal and power headroom to increase CPU or GPU power budget.
6120 */
6121unsigned long i915_read_mch_val(void)
6122{
6123 struct drm_i915_private *dev_priv;
6124 unsigned long chipset_val, graphics_val, ret = 0;
6125
9270388e 6126 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 6127 if (!i915_mch_dev)
6128 goto out_unlock;
6129 dev_priv = i915_mch_dev;
6130
f531dcb2
CW		 6131 chipset_val = __i915_chipset_val(dev_priv);
6132 graphics_val = __i915_gfx_val(dev_priv);
eb48eb00
DV		 6133
6134 ret = chipset_val + graphics_val;
6135
6136out_unlock:
9270388e 6137 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 6138
6139 return ret;
6140}
6141EXPORT_SYMBOL_GPL(i915_read_mch_val);
6142
6143/**
6144 * i915_gpu_raise - raise GPU frequency limit
6145 *
6146 * Raise the limit; IPS indicates we have thermal headroom.
6147 */
6148bool i915_gpu_raise(void)
6149{
6150 struct drm_i915_private *dev_priv;
6151 bool ret = true;
6152
9270388e 6153 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 6154 if (!i915_mch_dev) {
6155 ret = false;
6156 goto out_unlock;
6157 }
6158 dev_priv = i915_mch_dev;
6159
20e4d407
DV		 6160 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
6161 dev_priv->ips.max_delay--;
eb48eb00
DV		 6162
6163out_unlock:
9270388e 6164 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 6165
6166 return ret;
6167}
6168EXPORT_SYMBOL_GPL(i915_gpu_raise);
6169
6170/**
6171 * i915_gpu_lower - lower GPU frequency limit
6172 *
6173 * IPS indicates we're close to a thermal limit, so throttle back the GPU
6174 * frequency maximum.
6175 */
6176bool i915_gpu_lower(void)
6177{
6178 struct drm_i915_private *dev_priv;
6179 bool ret = true;
6180
9270388e 6181 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 6182 if (!i915_mch_dev) {
6183 ret = false;
6184 goto out_unlock;
6185 }
6186 dev_priv = i915_mch_dev;
6187
20e4d407
DV		 6188 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
6189 dev_priv->ips.max_delay++;
eb48eb00
DV		 6190
6191out_unlock:
9270388e 6192 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 6193
6194 return ret;
6195}
6196EXPORT_SYMBOL_GPL(i915_gpu_lower);
6197
6198/**
6199 * i915_gpu_busy - indicate GPU business to IPS
6200 *
6201 * Tell the IPS driver whether or not the GPU is busy.
6202 */
6203bool i915_gpu_busy(void)
6204{
6205 struct drm_i915_private *dev_priv;
e2f80391 6206 struct intel_engine_cs *engine;
eb48eb00
DV		 6207 bool ret = false;
6208
9270388e 6209 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 6210 if (!i915_mch_dev)
6211 goto out_unlock;
6212 dev_priv = i915_mch_dev;
6213
b4ac5afc 6214 for_each_engine(engine, dev_priv)
e2f80391 6215 ret |= !list_empty(&engine->request_list);
eb48eb00
DV		 6216
6217out_unlock:
9270388e 6218 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 6219
6220 return ret;
6221}
6222EXPORT_SYMBOL_GPL(i915_gpu_busy);
6223
6224/**
6225 * i915_gpu_turbo_disable - disable graphics turbo
6226 *
6227 * Disable graphics turbo by resetting the max frequency and setting the
6228 * current frequency to the default.
6229 */
6230bool i915_gpu_turbo_disable(void)
6231{
6232 struct drm_i915_private *dev_priv;
6233 bool ret = true;
6234
9270388e 6235 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 6236 if (!i915_mch_dev) {
6237 ret = false;
6238 goto out_unlock;
6239 }
6240 dev_priv = i915_mch_dev;
6241
20e4d407 6242 dev_priv->ips.max_delay = dev_priv->ips.fstart;
eb48eb00 6243
91d14251 6244 if (!ironlake_set_drps(dev_priv, dev_priv->ips.fstart))
eb48eb00
DV		 6245 ret = false;
6246
6247out_unlock:
9270388e 6248 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 6249
6250 return ret;
6251}
6252EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
6253
6254/**
6255 * Tells the intel_ips driver that the i915 driver is now loaded, if
6256 * IPS got loaded first.
6257 *
6258 * This awkward dance is so that neither module has to depend on the
6259 * other in order for IPS to do the appropriate communication of
6260 * GPU turbo limits to i915.
6261 */
6262static void
6263ips_ping_for_i915_load(void)
6264{
6265 void (*link)(void);
6266
6267 link = symbol_get(ips_link_to_i915_driver);
6268 if (link) {
6269 link();
6270 symbol_put(ips_link_to_i915_driver);
6271 }
6272}
6273
6274void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
6275{
02d71956
DV		 6276 /* We only register the i915 ips part with intel-ips once everything is
6277 * set up, to avoid intel-ips sneaking in and reading bogus values. */
9270388e 6278 spin_lock_irq(&mchdev_lock);
eb48eb00 6279 i915_mch_dev = dev_priv;
9270388e 6280 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 6281
6282 ips_ping_for_i915_load();
6283}
6284
6285void intel_gpu_ips_teardown(void)
6286{
9270388e 6287 spin_lock_irq(&mchdev_lock);
eb48eb00 6288 i915_mch_dev = NULL;
9270388e 6289 spin_unlock_irq(&mchdev_lock);
eb48eb00 6290}
76c3552f 6291
dc97997a 6292static void intel_init_emon(struct drm_i915_private *dev_priv)
dde18883 6293{
dde18883
ED		 6294 u32 lcfuse;
6295 u8 pxw[16];
6296 int i;
6297
6298 /* Disable to program */
6299 I915_WRITE(ECR, 0);
6300 POSTING_READ(ECR);
6301
6302 /* Program energy weights for various events */
6303 I915_WRITE(SDEW, 0x15040d00);
6304 I915_WRITE(CSIEW0, 0x007f0000);
6305 I915_WRITE(CSIEW1, 0x1e220004);
6306 I915_WRITE(CSIEW2, 0x04000004);
6307
6308 for (i = 0; i < 5; i++)
616847e7 6309 I915_WRITE(PEW(i), 0);
dde18883 6310 for (i = 0; i < 3; i++)
616847e7 6311 I915_WRITE(DEW(i), 0);
dde18883
ED		 6312
6313 /* Program P-state weights to account for frequency power adjustment */
6314 for (i = 0; i < 16; i++) {
616847e7 6315 u32 pxvidfreq = I915_READ(PXVFREQ(i));
dde18883
ED		 6316 unsigned long freq = intel_pxfreq(pxvidfreq);
6317 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
6318 PXVFREQ_PX_SHIFT;
6319 unsigned long val;
6320
6321 val = vid * vid;
6322 val *= (freq / 1000);
6323 val *= 255;
6324 val /= (127*127*900);
6325 if (val > 0xff)
6326 DRM_ERROR("bad pxval: %ld\n", val);
6327 pxw[i] = val;
6328 }
6329 /* Render standby states get 0 weight */
6330 pxw[14] = 0;
6331 pxw[15] = 0;
6332
6333 for (i = 0; i < 4; i++) {
6334 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
6335 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
616847e7 6336 I915_WRITE(PXW(i), val);
dde18883
ED		 6337 }
6338
6339 /* Adjust magic regs to magic values (more experimental results) */
6340 I915_WRITE(OGW0, 0);
6341 I915_WRITE(OGW1, 0);
6342 I915_WRITE(EG0, 0x00007f00);
6343 I915_WRITE(EG1, 0x0000000e);
6344 I915_WRITE(EG2, 0x000e0000);
6345 I915_WRITE(EG3, 0x68000300);
6346 I915_WRITE(EG4, 0x42000000);
6347 I915_WRITE(EG5, 0x00140031);
6348 I915_WRITE(EG6, 0);
6349 I915_WRITE(EG7, 0);
6350
6351 for (i = 0; i < 8; i++)
616847e7 6352 I915_WRITE(PXWL(i), 0);
dde18883
ED		 6353
6354 /* Enable PMON + select events */
6355 I915_WRITE(ECR, 0x80000019);
6356
6357 lcfuse = I915_READ(LCFUSE02);
6358
20e4d407 6359 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
dde18883
ED		 6360}
6361
dc97997a 6362void intel_init_gt_powersave(struct drm_i915_private *dev_priv)
ae48434c 6363{
b268c699
ID		 6364 /*
6365 * RPM depends on RC6 to save restore the GT HW context, so make RC6 a
6366 * requirement.
6367 */
6368 if (!i915.enable_rc6) {
6369 DRM_INFO("RC6 disabled, disabling runtime PM support\n");
6370 intel_runtime_pm_get(dev_priv);
6371 }
e6069ca8 6372
dc97997a
CW		 6373 if (IS_CHERRYVIEW(dev_priv))
6374 cherryview_init_gt_powersave(dev_priv);
6375 else if (IS_VALLEYVIEW(dev_priv))
6376 valleyview_init_gt_powersave(dev_priv);
ae48434c
ID		 6377}
6378
dc97997a 6379void intel_cleanup_gt_powersave(struct drm_i915_private *dev_priv)
ae48434c 6380{
dc97997a 6381 if (IS_CHERRYVIEW(dev_priv))
38807746 6382 return;
dc97997a
CW		 6383 else if (IS_VALLEYVIEW(dev_priv))
6384 valleyview_cleanup_gt_powersave(dev_priv);
b268c699
ID		 6385
6386 if (!i915.enable_rc6)
6387 intel_runtime_pm_put(dev_priv);
ae48434c
ID		 6388}
6389
91d14251 6390static void gen6_suspend_rps(struct drm_i915_private *dev_priv)
dbea3cea 6391{
dbea3cea
ID		 6392 flush_delayed_work(&dev_priv->rps.delayed_resume_work);
6393
91d14251 6394 gen6_disable_rps_interrupts(dev_priv);
dbea3cea
ID		 6395}
6396
156c7ca0
JB		 6397/**
6398 * intel_suspend_gt_powersave - suspend PM work and helper threads
dc97997a 6399 * @dev_priv: i915 device
156c7ca0
JB		 6400 *
6401 * We don't want to disable RC6 or other features here, we just want
6402 * to make sure any work we've queued has finished and won't bother
6403 * us while we're suspended.
6404 */
dc97997a 6405void intel_suspend_gt_powersave(struct drm_i915_private *dev_priv)
156c7ca0 6406{
91d14251 6407 if (INTEL_GEN(dev_priv) < 6)
d4d70aa5
ID		 6408 return;
6409
91d14251 6410 gen6_suspend_rps(dev_priv);
b47adc17
D		 6411
6412 /* Force GPU to min freq during suspend */
6413 gen6_rps_idle(dev_priv);
156c7ca0
JB		 6414}
6415
dc97997a 6416void intel_disable_gt_powersave(struct drm_i915_private *dev_priv)
8090c6b9 6417{
dc97997a 6418 if (IS_IRONLAKE_M(dev_priv)) {
91d14251 6419 ironlake_disable_drps(dev_priv);
dc97997a
CW		 6420 } else if (INTEL_INFO(dev_priv)->gen >= 6) {
6421 intel_suspend_gt_powersave(dev_priv);
e494837a 6422
4fc688ce 6423 mutex_lock(&dev_priv->rps.hw_lock);
dc97997a
CW		 6424 if (INTEL_INFO(dev_priv)->gen >= 9) {
6425 gen9_disable_rc6(dev_priv);
6426 gen9_disable_rps(dev_priv);
6427 } else if (IS_CHERRYVIEW(dev_priv))
6428 cherryview_disable_rps(dev_priv);
6429 else if (IS_VALLEYVIEW(dev_priv))
6430 valleyview_disable_rps(dev_priv);
d20d4f0c 6431 else
dc97997a 6432 gen6_disable_rps(dev_priv);
e534770a 6433
c0951f0c 6434 dev_priv->rps.enabled = false;
4fc688ce 6435 mutex_unlock(&dev_priv->rps.hw_lock);
930ebb46 6436 }
8090c6b9
DV		 6437}
6438
1a01ab3b
JB		 6439static void intel_gen6_powersave_work(struct work_struct *work)
6440{
6441 struct drm_i915_private *dev_priv =
6442 container_of(work, struct drm_i915_private,
6443 rps.delayed_resume_work.work);
1a01ab3b 6444
4fc688ce 6445 mutex_lock(&dev_priv->rps.hw_lock);
0a073b84 6446
dc97997a
CW		 6447 gen6_reset_rps_interrupts(dev_priv);
6448
6449 if (IS_CHERRYVIEW(dev_priv)) {
6450 cherryview_enable_rps(dev_priv);
6451 } else if (IS_VALLEYVIEW(dev_priv)) {
6452 valleyview_enable_rps(dev_priv);
6453 } else if (INTEL_INFO(dev_priv)->gen >= 9) {
6454 gen9_enable_rc6(dev_priv);
6455 gen9_enable_rps(dev_priv);
6456 if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
6457 __gen6_update_ring_freq(dev_priv);
6458 } else if (IS_BROADWELL(dev_priv)) {
6459 gen8_enable_rps(dev_priv);
6460 __gen6_update_ring_freq(dev_priv);
0a073b84 6461 } else {
dc97997a
CW		 6462 gen6_enable_rps(dev_priv);
6463 __gen6_update_ring_freq(dev_priv);
0a073b84 6464 }
aed242ff
CW		 6465
6466 WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq);
6467 WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq);
6468
6469 WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq);
6470 WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq);
6471
c0951f0c 6472 dev_priv->rps.enabled = true;
3cc134e3 6473
91d14251 6474 gen6_enable_rps_interrupts(dev_priv);
3cc134e3 6475
4fc688ce 6476 mutex_unlock(&dev_priv->rps.hw_lock);
c6df39b5
ID		 6477
6478 intel_runtime_pm_put(dev_priv);
1a01ab3b
JB		 6479}
6480
dc97997a 6481void intel_enable_gt_powersave(struct drm_i915_private *dev_priv)
8090c6b9 6482{
f61018b1 6483 /* Powersaving is controlled by the host when inside a VM */
c033666a 6484 if (intel_vgpu_active(dev_priv))
f61018b1
YZ		 6485 return;
6486
dc97997a 6487 if (IS_IRONLAKE_M(dev_priv)) {
91d14251 6488 ironlake_enable_drps(dev_priv);
dc97997a
CW		 6489 mutex_lock(&dev_priv->dev->struct_mutex);
6490 intel_init_emon(dev_priv);
6491 mutex_unlock(&dev_priv->dev->struct_mutex);
6492 } else if (INTEL_INFO(dev_priv)->gen >= 6) {
1a01ab3b
JB		 6493 /*
6494 * PCU communication is slow and this doesn't need to be
6495 * done at any specific time, so do this out of our fast path
6496 * to make resume and init faster.
c6df39b5
ID		 6497 *
6498 * We depend on the HW RC6 power context save/restore
6499 * mechanism when entering D3 through runtime PM suspend. So
6500 * disable RPM until RPS/RC6 is properly setup. We can only
6501 * get here via the driver load/system resume/runtime resume
6502 * paths, so the _noresume version is enough (and in case of
6503 * runtime resume it's necessary).
1a01ab3b 6504 */
c6df39b5
ID		 6505 if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
6506 round_jiffies_up_relative(HZ)))
6507 intel_runtime_pm_get_noresume(dev_priv);
8090c6b9
DV		 6508 }
6509}
6510
dc97997a 6511void intel_reset_gt_powersave(struct drm_i915_private *dev_priv)
c6df39b5 6512{
dc97997a 6513 if (INTEL_INFO(dev_priv)->gen < 6)
dbea3cea
ID		 6514 return;
6515
91d14251 6516 gen6_suspend_rps(dev_priv);
c6df39b5 6517 dev_priv->rps.enabled = false;
c6df39b5
ID		 6518}
6519
3107bd48
DV		 6520static void ibx_init_clock_gating(struct drm_device *dev)
6521{
6522 struct drm_i915_private *dev_priv = dev->dev_private;
6523
6524 /*
6525 * On Ibex Peak and Cougar Point, we need to disable clock
6526 * gating for the panel power sequencer or it will fail to
6527 * start up when no ports are active.
6528 */
6529 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
6530}
6531
0e088b8f
VS		 6532static void g4x_disable_trickle_feed(struct drm_device *dev)
6533{
6534 struct drm_i915_private *dev_priv = dev->dev_private;
b12ce1d8 6535 enum pipe pipe;
0e088b8f 6536
055e393f 6537 for_each_pipe(dev_priv, pipe) {
0e088b8f
VS		 6538 I915_WRITE(DSPCNTR(pipe),
6539 I915_READ(DSPCNTR(pipe)) |
6540 DISPPLANE_TRICKLE_FEED_DISABLE);
b12ce1d8
VS		 6541
6542 I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe)));
6543 POSTING_READ(DSPSURF(pipe));
0e088b8f
VS		 6544 }
6545}
6546
017636cc
VS		 6547static void ilk_init_lp_watermarks(struct drm_device *dev)
6548{
6549 struct drm_i915_private *dev_priv = dev->dev_private;
6550
6551 I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
6552 I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
6553 I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
6554
6555 /*
6556 * Don't touch WM1S_LP_EN here.
6557 * Doing so could cause underruns.
6558 */
6559}
6560
1fa61106 6561static void ironlake_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 6562{
6563 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 6564 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 6565
f1e8fa56
DL		 6566 /*
6567 * Required for FBC
6568 * WaFbcDisableDpfcClockGating:ilk
6569 */
4d47e4f5
DL		 6570 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
6571 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
6572 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED		 6573
6574 I915_WRITE(PCH_3DCGDIS0,
6575 MARIUNIT_CLOCK_GATE_DISABLE |
6576 SVSMUNIT_CLOCK_GATE_DISABLE);
6577 I915_WRITE(PCH_3DCGDIS1,
6578 VFMUNIT_CLOCK_GATE_DISABLE);
6579
6f1d69b0
ED		 6580 /*
6581 * According to the spec the following bits should be set in
6582 * order to enable memory self-refresh
6583 * The bit 22/21 of 0x42004
6584 * The bit 5 of 0x42020
6585 * The bit 15 of 0x45000
6586 */
6587 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6588 (I915_READ(ILK_DISPLAY_CHICKEN2) |
6589 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4d47e4f5 6590 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED		 6591 I915_WRITE(DISP_ARB_CTL,
6592 (I915_READ(DISP_ARB_CTL) |
6593 DISP_FBC_WM_DIS));
017636cc
VS		 6594
6595 ilk_init_lp_watermarks(dev);
6f1d69b0
ED		 6596
6597 /*
6598 * Based on the document from hardware guys the following bits
6599 * should be set unconditionally in order to enable FBC.
6600 * The bit 22 of 0x42000
6601 * The bit 22 of 0x42004
6602 * The bit 7,8,9 of 0x42020.
6603 */
6604 if (IS_IRONLAKE_M(dev)) {
4bb35334 6605 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6f1d69b0
ED		 6606 I915_WRITE(ILK_DISPLAY_CHICKEN1,
6607 I915_READ(ILK_DISPLAY_CHICKEN1) |
6608 ILK_FBCQ_DIS);
6609 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6610 I915_READ(ILK_DISPLAY_CHICKEN2) |
6611 ILK_DPARB_GATE);
6f1d69b0
ED		 6612 }
6613
4d47e4f5
DL		 6614 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6615
6f1d69b0
ED		 6616 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6617 I915_READ(ILK_DISPLAY_CHICKEN2) |
6618 ILK_ELPIN_409_SELECT);
6619 I915_WRITE(_3D_CHICKEN2,
6620 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
6621 _3D_CHICKEN2_WM_READ_PIPELINED);
4358a374 6622
ecdb4eb7 6623 /* WaDisableRenderCachePipelinedFlush:ilk */
4358a374
DV		 6624 I915_WRITE(CACHE_MODE_0,
6625 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3107bd48 6626
4e04632e
AG		 6627 /* WaDisable_RenderCache_OperationalFlush:ilk */
6628 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6629
0e088b8f 6630 g4x_disable_trickle_feed(dev);
bdad2b2f 6631
3107bd48
DV		 6632 ibx_init_clock_gating(dev);
6633}
6634
6635static void cpt_init_clock_gating(struct drm_device *dev)
6636{
6637 struct drm_i915_private *dev_priv = dev->dev_private;
6638 int pipe;
3f704fa2 6639 uint32_t val;
3107bd48
DV		 6640
6641 /*
6642 * On Ibex Peak and Cougar Point, we need to disable clock
6643 * gating for the panel power sequencer or it will fail to
6644 * start up when no ports are active.
6645 */
cd664078
JB		 6646 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
6647 PCH_DPLUNIT_CLOCK_GATE_DISABLE |
6648 PCH_CPUNIT_CLOCK_GATE_DISABLE);
3107bd48
DV		 6649 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
6650 DPLS_EDP_PPS_FIX_DIS);
335c07b7
TI		 6651 /* The below fixes the weird display corruption, a few pixels shifted
6652 * downward, on (only) LVDS of some HP laptops with IVY.
6653 */
055e393f 6654 for_each_pipe(dev_priv, pipe) {
dc4bd2d1
PZ		 6655 val = I915_READ(TRANS_CHICKEN2(pipe));
6656 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
6657 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
41aa3448 6658 if (dev_priv->vbt.fdi_rx_polarity_inverted)
3f704fa2 6659 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
dc4bd2d1
PZ		 6660 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
6661 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
6662 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3f704fa2
PZ		 6663 I915_WRITE(TRANS_CHICKEN2(pipe), val);
6664 }
3107bd48 6665 /* WADP0ClockGatingDisable */
055e393f 6666 for_each_pipe(dev_priv, pipe) {
3107bd48
DV		 6667 I915_WRITE(TRANS_CHICKEN1(pipe),
6668 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
6669 }
6f1d69b0
ED		 6670}
6671
1d7aaa0c
DV		 6672static void gen6_check_mch_setup(struct drm_device *dev)
6673{
6674 struct drm_i915_private *dev_priv = dev->dev_private;
6675 uint32_t tmp;
6676
6677 tmp = I915_READ(MCH_SSKPD);
df662a28
DV		 6678 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL)
6679 DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n",
6680 tmp);
1d7aaa0c
DV		 6681}
6682
1fa61106 6683static void gen6_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 6684{
6685 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 6686 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 6687
231e54f6 6688 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6f1d69b0
ED		 6689
6690 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6691 I915_READ(ILK_DISPLAY_CHICKEN2) |
6692 ILK_ELPIN_409_SELECT);
6693
ecdb4eb7 6694 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4283908e
DV		 6695 I915_WRITE(_3D_CHICKEN,
6696 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
6697
4e04632e
AG		 6698 /* WaDisable_RenderCache_OperationalFlush:snb */
6699 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6700
8d85d272
VS		 6701 /*
6702 * BSpec recoomends 8x4 when MSAA is used,
6703 * however in practice 16x4 seems fastest.
c5c98a58
VS		 6704 *
6705 * Note that PS/WM thread counts depend on the WIZ hashing
6706 * disable bit, which we don't touch here, but it's good
6707 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
8d85d272
VS		 6708 */
6709 I915_WRITE(GEN6_GT_MODE,
98533251 6710 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
8d85d272 6711
017636cc 6712 ilk_init_lp_watermarks(dev);
6f1d69b0 6713
6f1d69b0 6714 I915_WRITE(CACHE_MODE_0,
50743298 6715 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
6f1d69b0
ED		 6716
6717 I915_WRITE(GEN6_UCGCTL1,
6718 I915_READ(GEN6_UCGCTL1) |
6719 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
6720 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
6721
6722 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
6723 * gating disable must be set. Failure to set it results in
6724 * flickering pixels due to Z write ordering failures after
6725 * some amount of runtime in the Mesa "fire" demo, and Unigine
6726 * Sanctuary and Tropics, and apparently anything else with
6727 * alpha test or pixel discard.
6728 *
6729 * According to the spec, bit 11 (RCCUNIT) must also be set,
6730 * but we didn't debug actual testcases to find it out.
0f846f81 6731 *
ef59318c
VS		 6732 * WaDisableRCCUnitClockGating:snb
6733 * WaDisableRCPBUnitClockGating:snb
6f1d69b0
ED		 6734 */
6735 I915_WRITE(GEN6_UCGCTL2,
6736 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
6737 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
6738
5eb146dd 6739 /* WaStripsFansDisableFastClipPerformanceFix:snb */
743b57d8
VS		 6740 I915_WRITE(_3D_CHICKEN3,
6741 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL));
6f1d69b0 6742
e927ecde
VS		 6743 /*
6744 * Bspec says:
6745 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
6746 * 3DSTATE_SF number of SF output attributes is more than 16."
6747 */
6748 I915_WRITE(_3D_CHICKEN3,
6749 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH));
6750
6f1d69b0
ED		 6751 /*
6752 * According to the spec the following bits should be
6753 * set in order to enable memory self-refresh and fbc:
6754 * The bit21 and bit22 of 0x42000
6755 * The bit21 and bit22 of 0x42004
6756 * The bit5 and bit7 of 0x42020
6757 * The bit14 of 0x70180
6758 * The bit14 of 0x71180
4bb35334
DL		 6759 *
6760 * WaFbcAsynchFlipDisableFbcQueue:snb
6f1d69b0
ED		 6761 */
6762 I915_WRITE(ILK_DISPLAY_CHICKEN1,
6763 I915_READ(ILK_DISPLAY_CHICKEN1) |
6764 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
6765 I915_WRITE(ILK_DISPLAY_CHICKEN2,
6766 I915_READ(ILK_DISPLAY_CHICKEN2) |
6767 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
231e54f6
DL		 6768 I915_WRITE(ILK_DSPCLK_GATE_D,
6769 I915_READ(ILK_DSPCLK_GATE_D) |
6770 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
6771 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
6f1d69b0 6772
0e088b8f 6773 g4x_disable_trickle_feed(dev);
f8f2ac9a 6774
3107bd48 6775 cpt_init_clock_gating(dev);
1d7aaa0c
DV		 6776
6777 gen6_check_mch_setup(dev);
6f1d69b0
ED		 6778}
6779
6780static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
6781{
6782 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
6783
3aad9059 6784 /*
46680e0a 6785 * WaVSThreadDispatchOverride:ivb,vlv
3aad9059
VS		 6786 *
6787 * This actually overrides the dispatch
6788 * mode for all thread types.
6789 */
6f1d69b0
ED		 6790 reg &= ~GEN7_FF_SCHED_MASK;
6791 reg |= GEN7_FF_TS_SCHED_HW;
6792 reg |= GEN7_FF_VS_SCHED_HW;
6793 reg |= GEN7_FF_DS_SCHED_HW;
6794
6795 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
6796}
6797
17a303ec
PZ		 6798static void lpt_init_clock_gating(struct drm_device *dev)
6799{
6800 struct drm_i915_private *dev_priv = dev->dev_private;
6801
6802 /*
6803 * TODO: this bit should only be enabled when really needed, then
6804 * disabled when not needed anymore in order to save power.
6805 */
c2699524 6806 if (HAS_PCH_LPT_LP(dev))
17a303ec
PZ		 6807 I915_WRITE(SOUTH_DSPCLK_GATE_D,
6808 I915_READ(SOUTH_DSPCLK_GATE_D) |
6809 PCH_LP_PARTITION_LEVEL_DISABLE);
0a790cdb
PZ		 6810
6811 /* WADPOClockGatingDisable:hsw */
36c0d0cf
VS		 6812 I915_WRITE(TRANS_CHICKEN1(PIPE_A),
6813 I915_READ(TRANS_CHICKEN1(PIPE_A)) |
0a790cdb 6814 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
17a303ec
PZ		 6815}
6816
7d708ee4
ID		 6817static void lpt_suspend_hw(struct drm_device *dev)
6818{
6819 struct drm_i915_private *dev_priv = dev->dev_private;
6820
c2699524 6821 if (HAS_PCH_LPT_LP(dev)) {
7d708ee4
ID		 6822 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
6823
6824 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
6825 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
6826 }
6827}
6828
450174fe
ID		 6829static void gen8_set_l3sqc_credits(struct drm_i915_private *dev_priv,
6830 int general_prio_credits,
6831 int high_prio_credits)
6832{
6833 u32 misccpctl;
6834
6835 /* WaTempDisableDOPClkGating:bdw */
6836 misccpctl = I915_READ(GEN7_MISCCPCTL);
6837 I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE);
6838
6839 I915_WRITE(GEN8_L3SQCREG1,
6840 L3_GENERAL_PRIO_CREDITS(general_prio_credits) |
6841 L3_HIGH_PRIO_CREDITS(high_prio_credits));
6842
6843 /*
6844 * Wait at least 100 clocks before re-enabling clock gating.
6845 * See the definition of L3SQCREG1 in BSpec.
6846 */
6847 POSTING_READ(GEN8_L3SQCREG1);
6848 udelay(1);
6849 I915_WRITE(GEN7_MISCCPCTL, misccpctl);
6850}
6851
dc00b6a0
DV		 6852static void skylake_init_clock_gating(struct drm_device *dev)
6853{
6854 struct drm_i915_private *dev_priv = dev->dev_private;
6855
6856 /* See Bspec note for PSR2_CTL bit 31, Wa#828:skl,kbl */
6857 I915_WRITE(CHICKEN_PAR1_1,
6858 I915_READ(CHICKEN_PAR1_1) | SKL_EDP_PSR_FIX_RDWRAP);
6859}
6860
47c2bd97 6861static void broadwell_init_clock_gating(struct drm_device *dev)
1020a5c2
BW		 6862{
6863 struct drm_i915_private *dev_priv = dev->dev_private;
07d27e20 6864 enum pipe pipe;
1020a5c2 6865
7ad0dbab 6866 ilk_init_lp_watermarks(dev);
50ed5fbd 6867
ab57fff1 6868 /* WaSwitchSolVfFArbitrationPriority:bdw */
50ed5fbd 6869 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
fe4ab3ce 6870
ab57fff1 6871 /* WaPsrDPAMaskVBlankInSRD:bdw */
fe4ab3ce
BW		 6872 I915_WRITE(CHICKEN_PAR1_1,
6873 I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
6874
ab57fff1 6875 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
055e393f 6876 for_each_pipe(dev_priv, pipe) {
07d27e20 6877 I915_WRITE(CHICKEN_PIPESL_1(pipe),
c7c65622 6878 I915_READ(CHICKEN_PIPESL_1(pipe)) |
8f670bb1 6879 BDW_DPRS_MASK_VBLANK_SRD);
fe4ab3ce 6880 }
63801f21 6881
ab57fff1
BW		 6882 /* WaVSRefCountFullforceMissDisable:bdw */
6883 /* WaDSRefCountFullforceMissDisable:bdw */
6884 I915_WRITE(GEN7_FF_THREAD_MODE,
6885 I915_READ(GEN7_FF_THREAD_MODE) &
6886 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
36075a4c 6887
295e8bb7
VS		 6888 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
6889 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
4f1ca9e9
VS		 6890
6891 /* WaDisableSDEUnitClockGating:bdw */
6892 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
6893 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
5d708680 6894
450174fe
ID		 6895 /* WaProgramL3SqcReg1Default:bdw */
6896 gen8_set_l3sqc_credits(dev_priv, 30, 2);
4d487cff 6897
6d50b065
VS		 6898 /*
6899 * WaGttCachingOffByDefault:bdw
6900 * GTT cache may not work with big pages, so if those
6901 * are ever enabled GTT cache may need to be disabled.
6902 */
6903 I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL);
6904
89d6b2b8 6905 lpt_init_clock_gating(dev);
1020a5c2
BW		 6906}
6907
cad2a2d7
ED		 6908static void haswell_init_clock_gating(struct drm_device *dev)
6909{
6910 struct drm_i915_private *dev_priv = dev->dev_private;
cad2a2d7 6911
017636cc 6912 ilk_init_lp_watermarks(dev);
cad2a2d7 6913
f3fc4884
FJ		 6914 /* L3 caching of data atomics doesn't work -- disable it. */
6915 I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
6916 I915_WRITE(HSW_ROW_CHICKEN3,
6917 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
6918
ecdb4eb7 6919 /* This is required by WaCatErrorRejectionIssue:hsw */
cad2a2d7
ED		 6920 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
6921 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
6922 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
6923
e36ea7ff
VS		 6924 /* WaVSRefCountFullforceMissDisable:hsw */
6925 I915_WRITE(GEN7_FF_THREAD_MODE,
6926 I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);
cad2a2d7 6927
4e04632e
AG		 6928 /* WaDisable_RenderCache_OperationalFlush:hsw */
6929 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6930
fe27c606
CW		 6931 /* enable HiZ Raw Stall Optimization */
6932 I915_WRITE(CACHE_MODE_0_GEN7,
6933 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
6934
ecdb4eb7 6935 /* WaDisable4x2SubspanOptimization:hsw */
cad2a2d7
ED		 6936 I915_WRITE(CACHE_MODE_1,
6937 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
1544d9d5 6938
a12c4967
VS		 6939 /*
6940 * BSpec recommends 8x4 when MSAA is used,
6941 * however in practice 16x4 seems fastest.
c5c98a58
VS		 6942 *
6943 * Note that PS/WM thread counts depend on the WIZ hashing
6944 * disable bit, which we don't touch here, but it's good
6945 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a12c4967
VS		 6946 */
6947 I915_WRITE(GEN7_GT_MODE,
98533251 6948 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
a12c4967 6949
94411593
KG		 6950 /* WaSampleCChickenBitEnable:hsw */
6951 I915_WRITE(HALF_SLICE_CHICKEN3,
6952 _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE));
6953
ecdb4eb7 6954 /* WaSwitchSolVfFArbitrationPriority:hsw */
e3dff585
BW		 6955 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
6956
90a88643
PZ		 6957 /* WaRsPkgCStateDisplayPMReq:hsw */
6958 I915_WRITE(CHICKEN_PAR1_1,
6959 I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
1544d9d5 6960
17a303ec 6961 lpt_init_clock_gating(dev);
cad2a2d7
ED		 6962}
6963
1fa61106 6964static void ivybridge_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 6965{
6966 struct drm_i915_private *dev_priv = dev->dev_private;
20848223 6967 uint32_t snpcr;
6f1d69b0 6968
017636cc 6969 ilk_init_lp_watermarks(dev);
6f1d69b0 6970
231e54f6 6971 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 6972
ecdb4eb7 6973 /* WaDisableEarlyCull:ivb */
87f8020e
JB		 6974 I915_WRITE(_3D_CHICKEN3,
6975 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
6976
ecdb4eb7 6977 /* WaDisableBackToBackFlipFix:ivb */
6f1d69b0
ED		 6978 I915_WRITE(IVB_CHICKEN3,
6979 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
6980 CHICKEN3_DGMG_DONE_FIX_DISABLE);
6981
ecdb4eb7 6982 /* WaDisablePSDDualDispatchEnable:ivb */
12f3382b
JB		 6983 if (IS_IVB_GT1(dev))
6984 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
6985 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 6986
4e04632e
AG		 6987 /* WaDisable_RenderCache_OperationalFlush:ivb */
6988 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6989
ecdb4eb7 6990 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6f1d69b0
ED		 6991 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
6992 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
6993
ecdb4eb7 6994 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6f1d69b0
ED		 6995 I915_WRITE(GEN7_L3CNTLREG1,
6996 GEN7_WA_FOR_GEN7_L3_CONTROL);
6997 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
8ab43976
JB		 6998 GEN7_WA_L3_CHICKEN_MODE);
6999 if (IS_IVB_GT1(dev))
7000 I915_WRITE(GEN7_ROW_CHICKEN2,
7001 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2
VS		 7002 else {
7003 /* must write both registers */
7004 I915_WRITE(GEN7_ROW_CHICKEN2,
7005 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
8ab43976
JB		 7006 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
7007 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2 7008 }
6f1d69b0 7009
ecdb4eb7 7010 /* WaForceL3Serialization:ivb */
61939d97
JB		 7011 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
7012 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
7013
1b80a19a 7014 /*
0f846f81 7015 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 7016 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
0f846f81
JB		 7017 */
7018 I915_WRITE(GEN6_UCGCTL2,
28acf3b2 7019 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 7020
ecdb4eb7 7021 /* This is required by WaCatErrorRejectionIssue:ivb */
6f1d69b0
ED		 7022 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
7023 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
7024 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
7025
0e088b8f 7026 g4x_disable_trickle_feed(dev);
6f1d69b0
ED		 7027
7028 gen7_setup_fixed_func_scheduler(dev_priv);
97e1930f 7029
22721343
CW		 7030 if (0) { /* causes HiZ corruption on ivb:gt1 */
7031 /* enable HiZ Raw Stall Optimization */
7032 I915_WRITE(CACHE_MODE_0_GEN7,
7033 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
7034 }
116f2b6d 7035
ecdb4eb7 7036 /* WaDisable4x2SubspanOptimization:ivb */
97e1930f
DV		 7037 I915_WRITE(CACHE_MODE_1,
7038 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
20848223 7039
a607c1a4
VS		 7040 /*
7041 * BSpec recommends 8x4 when MSAA is used,
7042 * however in practice 16x4 seems fastest.
c5c98a58
VS		 7043 *
7044 * Note that PS/WM thread counts depend on the WIZ hashing
7045 * disable bit, which we don't touch here, but it's good
7046 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a607c1a4
VS		 7047 */
7048 I915_WRITE(GEN7_GT_MODE,
98533251 7049 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
a607c1a4 7050
20848223
BW		 7051 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
7052 snpcr &= ~GEN6_MBC_SNPCR_MASK;
7053 snpcr |= GEN6_MBC_SNPCR_MED;
7054 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
3107bd48 7055
ab5c608b
BW		 7056 if (!HAS_PCH_NOP(dev))
7057 cpt_init_clock_gating(dev);
1d7aaa0c
DV		 7058
7059 gen6_check_mch_setup(dev);
6f1d69b0
ED		 7060}
7061
1fa61106 7062static void valleyview_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7063{
7064 struct drm_i915_private *dev_priv = dev->dev_private;
6f1d69b0 7065
ecdb4eb7 7066 /* WaDisableEarlyCull:vlv */
87f8020e
JB		 7067 I915_WRITE(_3D_CHICKEN3,
7068 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
7069
ecdb4eb7 7070 /* WaDisableBackToBackFlipFix:vlv */
6f1d69b0
ED		 7071 I915_WRITE(IVB_CHICKEN3,
7072 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
7073 CHICKEN3_DGMG_DONE_FIX_DISABLE);
7074
fad7d36e 7075 /* WaPsdDispatchEnable:vlv */
ecdb4eb7 7076 /* WaDisablePSDDualDispatchEnable:vlv */
12f3382b 7077 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
d3bc0303
JB		 7078 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
7079 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 7080
4e04632e
AG		 7081 /* WaDisable_RenderCache_OperationalFlush:vlv */
7082 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
7083
ecdb4eb7 7084 /* WaForceL3Serialization:vlv */
61939d97
JB		 7085 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
7086 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
7087
ecdb4eb7 7088 /* WaDisableDopClockGating:vlv */
8ab43976
JB		 7089 I915_WRITE(GEN7_ROW_CHICKEN2,
7090 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
7091
ecdb4eb7 7092 /* This is required by WaCatErrorRejectionIssue:vlv */
6f1d69b0
ED		 7093 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
7094 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
7095 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
7096
46680e0a
VS		 7097 gen7_setup_fixed_func_scheduler(dev_priv);
7098
3c0edaeb 7099 /*
0f846f81 7100 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 7101 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
0f846f81
JB		 7102 */
7103 I915_WRITE(GEN6_UCGCTL2,
3c0edaeb 7104 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 7105
c98f5062
AG		 7106 /* WaDisableL3Bank2xClockGate:vlv
7107 * Disabling L3 clock gating- MMIO 940c[25] = 1
7108 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
7109 I915_WRITE(GEN7_UCGCTL4,
7110 I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
e3f33d46 7111
afd58e79
VS		 7112 /*
7113 * BSpec says this must be set, even though
7114 * WaDisable4x2SubspanOptimization isn't listed for VLV.
7115 */
6b26c86d
DV		 7116 I915_WRITE(CACHE_MODE_1,
7117 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
7983117f 7118
da2518f9
VS		 7119 /*
7120 * BSpec recommends 8x4 when MSAA is used,
7121 * however in practice 16x4 seems fastest.
7122 *
7123 * Note that PS/WM thread counts depend on the WIZ hashing
7124 * disable bit, which we don't touch here, but it's good
7125 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
7126 */
7127 I915_WRITE(GEN7_GT_MODE,
7128 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
7129
031994ee
VS		 7130 /*
7131 * WaIncreaseL3CreditsForVLVB0:vlv
7132 * This is the hardware default actually.
7133 */
7134 I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);
7135
2d809570 7136 /*
ecdb4eb7 7137 * WaDisableVLVClockGating_VBIIssue:vlv
2d809570
JB		 7138 * Disable clock gating on th GCFG unit to prevent a delay
7139 * in the reporting of vblank events.
7140 */
7a0d1eed 7141 I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
6f1d69b0
ED		 7142}
7143
a4565da8
VS		 7144static void cherryview_init_clock_gating(struct drm_device *dev)
7145{
7146 struct drm_i915_private *dev_priv = dev->dev_private;
7147
232ce337
VS		 7148 /* WaVSRefCountFullforceMissDisable:chv */
7149 /* WaDSRefCountFullforceMissDisable:chv */
7150 I915_WRITE(GEN7_FF_THREAD_MODE,
7151 I915_READ(GEN7_FF_THREAD_MODE) &
7152 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
acea6f95
VS		 7153
7154 /* WaDisableSemaphoreAndSyncFlipWait:chv */
7155 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
7156 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
0846697c
VS		 7157
7158 /* WaDisableCSUnitClockGating:chv */
7159 I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
7160 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
c631780f
VS		 7161
7162 /* WaDisableSDEUnitClockGating:chv */
7163 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
7164 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
6d50b065 7165
450174fe
ID		 7166 /*
7167 * WaProgramL3SqcReg1Default:chv
7168 * See gfxspecs/Related Documents/Performance Guide/
7169 * LSQC Setting Recommendations.
7170 */
7171 gen8_set_l3sqc_credits(dev_priv, 38, 2);
7172
6d50b065
VS		 7173 /*
7174 * GTT cache may not work with big pages, so if those
7175 * are ever enabled GTT cache may need to be disabled.
7176 */
7177 I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL);
a4565da8
VS		 7178}
7179
1fa61106 7180static void g4x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7181{
7182 struct drm_i915_private *dev_priv = dev->dev_private;
7183 uint32_t dspclk_gate;
7184
7185 I915_WRITE(RENCLK_GATE_D1, 0);
7186 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
7187 GS_UNIT_CLOCK_GATE_DISABLE |
7188 CL_UNIT_CLOCK_GATE_DISABLE);
7189 I915_WRITE(RAMCLK_GATE_D, 0);
7190 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
7191 OVRUNIT_CLOCK_GATE_DISABLE |
7192 OVCUNIT_CLOCK_GATE_DISABLE;
7193 if (IS_GM45(dev))
7194 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
7195 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4358a374
DV		 7196
7197 /* WaDisableRenderCachePipelinedFlush */
7198 I915_WRITE(CACHE_MODE_0,
7199 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
de1aa629 7200
4e04632e
AG		 7201 /* WaDisable_RenderCache_OperationalFlush:g4x */
7202 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
7203
0e088b8f 7204 g4x_disable_trickle_feed(dev);
6f1d69b0
ED		 7205}
7206
1fa61106 7207static void crestline_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7208{
7209 struct drm_i915_private *dev_priv = dev->dev_private;
7210
7211 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
7212 I915_WRITE(RENCLK_GATE_D2, 0);
7213 I915_WRITE(DSPCLK_GATE_D, 0);
7214 I915_WRITE(RAMCLK_GATE_D, 0);
7215 I915_WRITE16(DEUC, 0);
20f94967
VS		 7216 I915_WRITE(MI_ARB_STATE,
7217 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
4e04632e
AG		 7218
7219 /* WaDisable_RenderCache_OperationalFlush:gen4 */
7220 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6f1d69b0
ED		 7221}
7222
1fa61106 7223static void broadwater_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7224{
7225 struct drm_i915_private *dev_priv = dev->dev_private;
7226
7227 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
7228 I965_RCC_CLOCK_GATE_DISABLE |
7229 I965_RCPB_CLOCK_GATE_DISABLE |
7230 I965_ISC_CLOCK_GATE_DISABLE |
7231 I965_FBC_CLOCK_GATE_DISABLE);
7232 I915_WRITE(RENCLK_GATE_D2, 0);
20f94967
VS		 7233 I915_WRITE(MI_ARB_STATE,
7234 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
4e04632e
AG		 7235
7236 /* WaDisable_RenderCache_OperationalFlush:gen4 */
7237 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6f1d69b0
ED		 7238}
7239
1fa61106 7240static void gen3_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7241{
7242 struct drm_i915_private *dev_priv = dev->dev_private;
7243 u32 dstate = I915_READ(D_STATE);
7244
7245 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
7246 DSTATE_DOT_CLOCK_GATING;
7247 I915_WRITE(D_STATE, dstate);
13a86b85
CW		 7248
7249 if (IS_PINEVIEW(dev))
7250 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
974a3b0f
DV		 7251
7252 /* IIR "flip pending" means done if this bit is set */
7253 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
12fabbcb
VS		 7254
7255 /* interrupts should cause a wake up from C3 */
3299254f 7256 I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN));
dbb42748
VS		 7257
7258 /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
7259 I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE));
1038392b
VS		 7260
7261 I915_WRITE(MI_ARB_STATE,
7262 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED		 7263}
7264
1fa61106 7265static void i85x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7266{
7267 struct drm_i915_private *dev_priv = dev->dev_private;
7268
7269 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
54e472ae
VS		 7270
7271 /* interrupts should cause a wake up from C3 */
7272 I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) |
7273 _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE));
1038392b
VS		 7274
7275 I915_WRITE(MEM_MODE,
7276 _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED		 7277}
7278
1fa61106 7279static void i830_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 7280{
7281 struct drm_i915_private *dev_priv = dev->dev_private;
7282
7283 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
1038392b
VS		 7284
7285 I915_WRITE(MEM_MODE,
7286 _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) |
7287 _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE));
6f1d69b0
ED		 7288}
7289
6f1d69b0
ED		 7290void intel_init_clock_gating(struct drm_device *dev)
7291{
7292 struct drm_i915_private *dev_priv = dev->dev_private;
7293
bb400da9 7294 dev_priv->display.init_clock_gating(dev);
6f1d69b0
ED		 7295}
7296
7d708ee4
ID		 7297void intel_suspend_hw(struct drm_device *dev)
7298{
7299 if (HAS_PCH_LPT(dev))
7300 lpt_suspend_hw(dev);
7301}
7302
bb400da9
ID		 7303static void nop_init_clock_gating(struct drm_device *dev)
7304{
7305 DRM_DEBUG_KMS("No clock gating settings or workarounds applied.\n");
7306}
7307
7308/**
7309 * intel_init_clock_gating_hooks - setup the clock gating hooks
7310 * @dev_priv: device private
7311 *
7312 * Setup the hooks that configure which clocks of a given platform can be
7313 * gated and also apply various GT and display specific workarounds for these
7314 * platforms. Note that some GT specific workarounds are applied separately
7315 * when GPU contexts or batchbuffers start their execution.
7316 */
7317void intel_init_clock_gating_hooks(struct drm_i915_private *dev_priv)
7318{
7319 if (IS_SKYLAKE(dev_priv))
dc00b6a0 7320 dev_priv->display.init_clock_gating = skylake_init_clock_gating;
bb400da9 7321 else if (IS_KABYLAKE(dev_priv))
dc00b6a0 7322 dev_priv->display.init_clock_gating = skylake_init_clock_gating;
bb400da9
ID		 7323 else if (IS_BROXTON(dev_priv))
7324 dev_priv->display.init_clock_gating = bxt_init_clock_gating;
7325 else if (IS_BROADWELL(dev_priv))
7326 dev_priv->display.init_clock_gating = broadwell_init_clock_gating;
7327 else if (IS_CHERRYVIEW(dev_priv))
7328 dev_priv->display.init_clock_gating = cherryview_init_clock_gating;
7329 else if (IS_HASWELL(dev_priv))
7330 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
7331 else if (IS_IVYBRIDGE(dev_priv))
7332 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
7333 else if (IS_VALLEYVIEW(dev_priv))
7334 dev_priv->display.init_clock_gating = valleyview_init_clock_gating;
7335 else if (IS_GEN6(dev_priv))
7336 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
7337 else if (IS_GEN5(dev_priv))
7338 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
7339 else if (IS_G4X(dev_priv))
7340 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
7341 else if (IS_CRESTLINE(dev_priv))
7342 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
7343 else if (IS_BROADWATER(dev_priv))
7344 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
7345 else if (IS_GEN3(dev_priv))
7346 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
7347 else if (IS_I85X(dev_priv) || IS_I865G(dev_priv))
7348 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
7349 else if (IS_GEN2(dev_priv))
7350 dev_priv->display.init_clock_gating = i830_init_clock_gating;
7351 else {
7352 MISSING_CASE(INTEL_DEVID(dev_priv));
7353 dev_priv->display.init_clock_gating = nop_init_clock_gating;
7354 }
7355}
7356
1fa61106
ED		 7357/* Set up chip specific power management-related functions */
7358void intel_init_pm(struct drm_device *dev)
7359{
7360 struct drm_i915_private *dev_priv = dev->dev_private;
7361
7ff0ebcc 7362 intel_fbc_init(dev_priv);
1fa61106 7363
c921aba8
DV		 7364 /* For cxsr */
7365 if (IS_PINEVIEW(dev))
7366 i915_pineview_get_mem_freq(dev);
7367 else if (IS_GEN5(dev))
7368 i915_ironlake_get_mem_freq(dev);
7369
1fa61106 7370 /* For FIFO watermark updates */
f5ed50cb 7371 if (INTEL_INFO(dev)->gen >= 9) {
2af30a5c 7372 skl_setup_wm_latency(dev);
2d41c0b5 7373 dev_priv->display.update_wm = skl_update_wm;
98d39494 7374 dev_priv->display.compute_global_watermarks = skl_compute_wm;
c83155a6 7375 } else if (HAS_PCH_SPLIT(dev)) {
fa50ad61 7376 ilk_setup_wm_latency(dev);
53615a5e 7377
bd602544
VS		 7378 if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] &&
7379 dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
7380 (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] &&
7381 dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
86c8bbbe 7382 dev_priv->display.compute_pipe_wm = ilk_compute_pipe_wm;
ed4a6a7c
MR		 7383 dev_priv->display.compute_intermediate_wm =
7384 ilk_compute_intermediate_wm;
7385 dev_priv->display.initial_watermarks =
7386 ilk_initial_watermarks;
7387 dev_priv->display.optimize_watermarks =
7388 ilk_optimize_watermarks;
bd602544
VS		 7389 } else {
7390 DRM_DEBUG_KMS("Failed to read display plane latency. "
7391 "Disable CxSR\n");
7392 }
a4565da8 7393 } else if (IS_CHERRYVIEW(dev)) {
262cd2e1 7394 vlv_setup_wm_latency(dev);
262cd2e1 7395 dev_priv->display.update_wm = vlv_update_wm;
1fa61106 7396 } else if (IS_VALLEYVIEW(dev)) {
26e1fe4f 7397 vlv_setup_wm_latency(dev);
26e1fe4f 7398 dev_priv->display.update_wm = vlv_update_wm;
1fa61106
ED		 7399 } else if (IS_PINEVIEW(dev)) {
7400 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
7401 dev_priv->is_ddr3,
7402 dev_priv->fsb_freq,
7403 dev_priv->mem_freq)) {
7404 DRM_INFO("failed to find known CxSR latency "
7405 "(found ddr%s fsb freq %d, mem freq %d), "
7406 "disabling CxSR\n",
7407 (dev_priv->is_ddr3 == 1) ? "3" : "2",
7408 dev_priv->fsb_freq, dev_priv->mem_freq);
7409 /* Disable CxSR and never update its watermark again */
5209b1f4 7410 intel_set_memory_cxsr(dev_priv, false);
1fa61106
ED		 7411 dev_priv->display.update_wm = NULL;
7412 } else
7413 dev_priv->display.update_wm = pineview_update_wm;
1fa61106
ED		 7414 } else if (IS_G4X(dev)) {
7415 dev_priv->display.update_wm = g4x_update_wm;
1fa61106
ED		 7416 } else if (IS_GEN4(dev)) {
7417 dev_priv->display.update_wm = i965_update_wm;
1fa61106
ED		 7418 } else if (IS_GEN3(dev)) {
7419 dev_priv->display.update_wm = i9xx_update_wm;
7420 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
feb56b93
DV		 7421 } else if (IS_GEN2(dev)) {
7422 if (INTEL_INFO(dev)->num_pipes == 1) {
7423 dev_priv->display.update_wm = i845_update_wm;
1fa61106 7424 dev_priv->display.get_fifo_size = i845_get_fifo_size;
feb56b93
DV		 7425 } else {
7426 dev_priv->display.update_wm = i9xx_update_wm;
1fa61106 7427 dev_priv->display.get_fifo_size = i830_get_fifo_size;
feb56b93 7428 }
feb56b93
DV		 7429 } else {
7430 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
1fa61106
ED		 7431 }
7432}
7433
151a49d0 7434int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val)
42c0526c 7435{
4fc688ce 7436 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW		 7437
7438 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
7439 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
7440 return -EAGAIN;
7441 }
7442
7443 I915_WRITE(GEN6_PCODE_DATA, *val);
dddab346 7444 I915_WRITE(GEN6_PCODE_DATA1, 0);
42c0526c
BW		 7445 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
7446
7447 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
7448 500)) {
7449 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
7450 return -ETIMEDOUT;
7451 }
7452
7453 *val = I915_READ(GEN6_PCODE_DATA);
7454 I915_WRITE(GEN6_PCODE_DATA, 0);
7455
7456 return 0;
7457}
7458
151a49d0 7459int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u32 mbox, u32 val)
42c0526c 7460{
4fc688ce 7461 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW		 7462
7463 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
7464 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
7465 return -EAGAIN;
7466 }
7467
7468 I915_WRITE(GEN6_PCODE_DATA, val);
7469 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
7470
7471 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
7472 500)) {
7473 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
7474 return -ETIMEDOUT;
7475 }
7476
7477 I915_WRITE(GEN6_PCODE_DATA, 0);
7478
7479 return 0;
7480}
a0e4e199 7481
dd06f88c
VS		 7482static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val)
7483{
c30fec65
VS		 7484 /*
7485 * N = val - 0xb7
7486 * Slow = Fast = GPLL ref * N
7487 */
7488 return DIV_ROUND_CLOSEST(dev_priv->rps.gpll_ref_freq * (val - 0xb7), 1000);
855ba3be
JB		 7489}
7490
b55dd647 7491static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val)
855ba3be 7492{
c30fec65 7493 return DIV_ROUND_CLOSEST(1000 * val, dev_priv->rps.gpll_ref_freq) + 0xb7;
855ba3be
JB		 7494}
7495
b55dd647 7496static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val)
22b1b2f8 7497{
c30fec65
VS		 7498 /*
7499 * N = val / 2
7500 * CU (slow) = CU2x (fast) / 2 = GPLL ref * N / 2
7501 */
7502 return DIV_ROUND_CLOSEST(dev_priv->rps.gpll_ref_freq * val, 2 * 2 * 1000);
22b1b2f8
D		 7503}
7504
b55dd647 7505static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val)
22b1b2f8 7506{
1c14762d 7507 /* CHV needs even values */
c30fec65 7508 return DIV_ROUND_CLOSEST(2 * 1000 * val, dev_priv->rps.gpll_ref_freq) * 2;
22b1b2f8
D		 7509}
7510
616bc820 7511int intel_gpu_freq(struct drm_i915_private *dev_priv, int val)
22b1b2f8 7512{
2d1fe073 7513 if (IS_GEN9(dev_priv))
500a3d2e
MK		 7514 return DIV_ROUND_CLOSEST(val * GT_FREQUENCY_MULTIPLIER,
7515 GEN9_FREQ_SCALER);
2d1fe073 7516 else if (IS_CHERRYVIEW(dev_priv))
616bc820 7517 return chv_gpu_freq(dev_priv, val);
2d1fe073 7518 else if (IS_VALLEYVIEW(dev_priv))
616bc820
VS		 7519 return byt_gpu_freq(dev_priv, val);
7520 else
7521 return val * GT_FREQUENCY_MULTIPLIER;
22b1b2f8
D		 7522}
7523
616bc820
VS		 7524int intel_freq_opcode(struct drm_i915_private *dev_priv, int val)
7525{
2d1fe073 7526 if (IS_GEN9(dev_priv))
500a3d2e
MK		 7527 return DIV_ROUND_CLOSEST(val * GEN9_FREQ_SCALER,
7528 GT_FREQUENCY_MULTIPLIER);
2d1fe073 7529 else if (IS_CHERRYVIEW(dev_priv))
616bc820 7530 return chv_freq_opcode(dev_priv, val);
2d1fe073 7531 else if (IS_VALLEYVIEW(dev_priv))
616bc820
VS		 7532 return byt_freq_opcode(dev_priv, val);
7533 else
500a3d2e 7534 return DIV_ROUND_CLOSEST(val, GT_FREQUENCY_MULTIPLIER);
616bc820 7535}
22b1b2f8 7536
6ad790c0
CW		 7537struct request_boost {
7538 struct work_struct work;
eed29a5b 7539 struct drm_i915_gem_request *req;
6ad790c0
CW		 7540};
7541
7542static void __intel_rps_boost_work(struct work_struct *work)
7543{
7544 struct request_boost *boost = container_of(work, struct request_boost, work);
e61b9958 7545 struct drm_i915_gem_request *req = boost->req;
6ad790c0 7546
e61b9958 7547 if (!i915_gem_request_completed(req, true))
c033666a 7548 gen6_rps_boost(req->i915, NULL, req->emitted_jiffies);
6ad790c0 7549
73db04cf 7550 i915_gem_request_unreference(req);
6ad790c0
CW		 7551 kfree(boost);
7552}
7553
91d14251 7554void intel_queue_rps_boost_for_request(struct drm_i915_gem_request *req)
6ad790c0
CW		 7555{
7556 struct request_boost *boost;
7557
91d14251 7558 if (req == NULL || INTEL_GEN(req->i915) < 6)
6ad790c0
CW		 7559 return;
7560
e61b9958
CW		 7561 if (i915_gem_request_completed(req, true))
7562 return;
7563
6ad790c0
CW		 7564 boost = kmalloc(sizeof(*boost), GFP_ATOMIC);
7565 if (boost == NULL)
7566 return;
7567
eed29a5b
DV		 7568 i915_gem_request_reference(req);
7569 boost->req = req;
6ad790c0
CW		 7570
7571 INIT_WORK(&boost->work, __intel_rps_boost_work);
91d14251 7572 queue_work(req->i915->wq, &boost->work);
6ad790c0
CW		 7573}
7574
f742a552 7575void intel_pm_setup(struct drm_device *dev)
907b28c5
CW		 7576{
7577 struct drm_i915_private *dev_priv = dev->dev_private;
7578
f742a552 7579 mutex_init(&dev_priv->rps.hw_lock);
8d3afd7d 7580 spin_lock_init(&dev_priv->rps.client_lock);
f742a552 7581
907b28c5
CW		 7582 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
7583 intel_gen6_powersave_work);
1854d5ca 7584 INIT_LIST_HEAD(&dev_priv->rps.clients);
2e1b8730
CW		 7585 INIT_LIST_HEAD(&dev_priv->rps.semaphores.link);
7586 INIT_LIST_HEAD(&dev_priv->rps.mmioflips.link);
5d584b2e 7587
33688d95 7588 dev_priv->pm.suspended = false;
1f814dac 7589 atomic_set(&dev_priv->pm.wakeref_count, 0);
2b19efeb 7590 atomic_set(&dev_priv->pm.atomic_seq, 0);
907b28c5 7591}
Ubuntu Jammy Linux kernel mirror