2 * Copyright © 2012 Intel Corporation
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:
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
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
24 * Eugeni Dodonov <eugeni.dodonov@intel.com>
28 #include <linux/cpufreq.h>
30 #include "intel_drv.h"
31 #include "../../../platform/x86/intel_ips.h"
32 #include <linux/module.h>
35 * RC6 is a special power stage which allows the GPU to enter an very
36 * low-voltage mode when idle, using down to 0V while at this stage. This
37 * stage is entered automatically when the GPU is idle when RC6 support is
38 * enabled, and as soon as new workload arises GPU wakes up automatically as well.
40 * There are different RC6 modes available in Intel GPU, which differentiate
41 * among each other with the latency required to enter and leave RC6 and
42 * voltage consumed by the GPU in different states.
44 * The combination of the following flags define which states GPU is allowed
45 * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
46 * RC6pp is deepest RC6. Their support by hardware varies according to the
47 * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
48 * which brings the most power savings; deeper states save more power, but
49 * require higher latency to switch to and wake up.
51 #define INTEL_RC6_ENABLE (1<<0)
52 #define INTEL_RC6p_ENABLE (1<<1)
53 #define INTEL_RC6pp_ENABLE (1<<2)
55 static void gen9_init_clock_gating(struct drm_device
*dev
)
57 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
59 /* WaEnableLbsSlaRetryTimerDecrement:skl */
60 I915_WRITE(BDW_SCRATCH1
, I915_READ(BDW_SCRATCH1
) |
61 GEN9_LBS_SLA_RETRY_TIMER_DECREMENT_ENABLE
);
63 /* WaDisableKillLogic:bxt,skl */
64 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) |
68 static void skl_init_clock_gating(struct drm_device
*dev
)
70 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
72 gen9_init_clock_gating(dev
);
74 if (INTEL_REVID(dev
) <= SKL_REVID_B0
) {
76 * WaDisableSDEUnitClockGating:skl
77 * WaSetGAPSunitClckGateDisable:skl
79 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
80 GEN8_GAPSUNIT_CLOCK_GATE_DISABLE
|
81 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
83 /* WaDisableVFUnitClockGating:skl */
84 I915_WRITE(GEN6_UCGCTL2
, I915_READ(GEN6_UCGCTL2
) |
85 GEN6_VFUNIT_CLOCK_GATE_DISABLE
);
88 if (INTEL_REVID(dev
) <= SKL_REVID_D0
) {
89 /* WaDisableHDCInvalidation:skl */
90 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) |
91 BDW_DISABLE_HDC_INVALIDATION
);
93 /* WaDisableChickenBitTSGBarrierAckForFFSliceCS:skl */
94 I915_WRITE(FF_SLICE_CS_CHICKEN2
,
95 _MASKED_BIT_ENABLE(GEN9_TSG_BARRIER_ACK_DISABLE
));
98 /* GEN8_L3SQCREG4 has a dependency with WA batch so any new changes
99 * involving this register should also be added to WA batch as required.
101 if (INTEL_REVID(dev
) <= SKL_REVID_E0
)
102 /* WaDisableLSQCROPERFforOCL:skl */
103 I915_WRITE(GEN8_L3SQCREG4
, I915_READ(GEN8_L3SQCREG4
) |
104 GEN8_LQSC_RO_PERF_DIS
);
106 /* WaEnableGapsTsvCreditFix:skl */
107 if (IS_SKYLAKE(dev
) && (INTEL_REVID(dev
) >= SKL_REVID_C0
)) {
108 I915_WRITE(GEN8_GARBCNTL
, (I915_READ(GEN8_GARBCNTL
) |
109 GEN9_GAPS_TSV_CREDIT_DISABLE
));
113 static void bxt_init_clock_gating(struct drm_device
*dev
)
115 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
117 gen9_init_clock_gating(dev
);
119 /* WaDisableSDEUnitClockGating:bxt */
120 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
121 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
125 * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only.
127 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
128 GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ
);
130 if (INTEL_REVID(dev
) == BXT_REVID_A0
) {
132 * Hardware specification requires this bit to be
135 I915_WRITE(TILECTL
, I915_READ(TILECTL
) | TILECTL_TLBPF
);
138 /* WaSetClckGatingDisableMedia:bxt */
139 if (INTEL_REVID(dev
) == BXT_REVID_A0
) {
140 I915_WRITE(GEN7_MISCCPCTL
, (I915_READ(GEN7_MISCCPCTL
) &
141 ~GEN8_DOP_CLOCK_GATE_MEDIA_ENABLE
));
145 static void i915_pineview_get_mem_freq(struct drm_device
*dev
)
147 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
150 tmp
= I915_READ(CLKCFG
);
152 switch (tmp
& CLKCFG_FSB_MASK
) {
154 dev_priv
->fsb_freq
= 533; /* 133*4 */
157 dev_priv
->fsb_freq
= 800; /* 200*4 */
160 dev_priv
->fsb_freq
= 667; /* 167*4 */
163 dev_priv
->fsb_freq
= 400; /* 100*4 */
167 switch (tmp
& CLKCFG_MEM_MASK
) {
169 dev_priv
->mem_freq
= 533;
172 dev_priv
->mem_freq
= 667;
175 dev_priv
->mem_freq
= 800;
179 /* detect pineview DDR3 setting */
180 tmp
= I915_READ(CSHRDDR3CTL
);
181 dev_priv
->is_ddr3
= (tmp
& CSHRDDR3CTL_DDR3
) ? 1 : 0;
184 static void i915_ironlake_get_mem_freq(struct drm_device
*dev
)
186 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
189 ddrpll
= I915_READ16(DDRMPLL1
);
190 csipll
= I915_READ16(CSIPLL0
);
192 switch (ddrpll
& 0xff) {
194 dev_priv
->mem_freq
= 800;
197 dev_priv
->mem_freq
= 1066;
200 dev_priv
->mem_freq
= 1333;
203 dev_priv
->mem_freq
= 1600;
206 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
208 dev_priv
->mem_freq
= 0;
212 dev_priv
->ips
.r_t
= dev_priv
->mem_freq
;
214 switch (csipll
& 0x3ff) {
216 dev_priv
->fsb_freq
= 3200;
219 dev_priv
->fsb_freq
= 3733;
222 dev_priv
->fsb_freq
= 4266;
225 dev_priv
->fsb_freq
= 4800;
228 dev_priv
->fsb_freq
= 5333;
231 dev_priv
->fsb_freq
= 5866;
234 dev_priv
->fsb_freq
= 6400;
237 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
239 dev_priv
->fsb_freq
= 0;
243 if (dev_priv
->fsb_freq
== 3200) {
244 dev_priv
->ips
.c_m
= 0;
245 } else if (dev_priv
->fsb_freq
> 3200 && dev_priv
->fsb_freq
<= 4800) {
246 dev_priv
->ips
.c_m
= 1;
248 dev_priv
->ips
.c_m
= 2;
252 static const struct cxsr_latency cxsr_latency_table
[] = {
253 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
254 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
255 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
256 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
257 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
259 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
260 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
261 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
262 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
263 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
265 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
266 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
267 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
268 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
269 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
271 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
272 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
273 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
274 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
275 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
277 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
278 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
279 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
280 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
281 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
283 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
284 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
285 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
286 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
287 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
290 static const struct cxsr_latency
*intel_get_cxsr_latency(int is_desktop
,
295 const struct cxsr_latency
*latency
;
298 if (fsb
== 0 || mem
== 0)
301 for (i
= 0; i
< ARRAY_SIZE(cxsr_latency_table
); i
++) {
302 latency
= &cxsr_latency_table
[i
];
303 if (is_desktop
== latency
->is_desktop
&&
304 is_ddr3
== latency
->is_ddr3
&&
305 fsb
== latency
->fsb_freq
&& mem
== latency
->mem_freq
)
309 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
314 static void chv_set_memory_dvfs(struct drm_i915_private
*dev_priv
, bool enable
)
318 mutex_lock(&dev_priv
->rps
.hw_lock
);
320 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
);
322 val
&= ~FORCE_DDR_HIGH_FREQ
;
324 val
|= FORCE_DDR_HIGH_FREQ
;
325 val
&= ~FORCE_DDR_LOW_FREQ
;
326 val
|= FORCE_DDR_FREQ_REQ_ACK
;
327 vlv_punit_write(dev_priv
, PUNIT_REG_DDR_SETUP2
, val
);
329 if (wait_for((vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
) &
330 FORCE_DDR_FREQ_REQ_ACK
) == 0, 3))
331 DRM_ERROR("timed out waiting for Punit DDR DVFS request\n");
333 mutex_unlock(&dev_priv
->rps
.hw_lock
);
336 static void chv_set_memory_pm5(struct drm_i915_private
*dev_priv
, bool enable
)
340 mutex_lock(&dev_priv
->rps
.hw_lock
);
342 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DSPFREQ
);
344 val
|= DSP_MAXFIFO_PM5_ENABLE
;
346 val
&= ~DSP_MAXFIFO_PM5_ENABLE
;
347 vlv_punit_write(dev_priv
, PUNIT_REG_DSPFREQ
, val
);
349 mutex_unlock(&dev_priv
->rps
.hw_lock
);
352 #define FW_WM(value, plane) \
353 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK)
355 void intel_set_memory_cxsr(struct drm_i915_private
*dev_priv
, bool enable
)
357 struct drm_device
*dev
= dev_priv
->dev
;
360 if (IS_VALLEYVIEW(dev
)) {
361 I915_WRITE(FW_BLC_SELF_VLV
, enable
? FW_CSPWRDWNEN
: 0);
362 POSTING_READ(FW_BLC_SELF_VLV
);
363 dev_priv
->wm
.vlv
.cxsr
= enable
;
364 } else if (IS_G4X(dev
) || IS_CRESTLINE(dev
)) {
365 I915_WRITE(FW_BLC_SELF
, enable
? FW_BLC_SELF_EN
: 0);
366 POSTING_READ(FW_BLC_SELF
);
367 } else if (IS_PINEVIEW(dev
)) {
368 val
= I915_READ(DSPFW3
) & ~PINEVIEW_SELF_REFRESH_EN
;
369 val
|= enable
? PINEVIEW_SELF_REFRESH_EN
: 0;
370 I915_WRITE(DSPFW3
, val
);
371 POSTING_READ(DSPFW3
);
372 } else if (IS_I945G(dev
) || IS_I945GM(dev
)) {
373 val
= enable
? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN
) :
374 _MASKED_BIT_DISABLE(FW_BLC_SELF_EN
);
375 I915_WRITE(FW_BLC_SELF
, val
);
376 POSTING_READ(FW_BLC_SELF
);
377 } else if (IS_I915GM(dev
)) {
378 val
= enable
? _MASKED_BIT_ENABLE(INSTPM_SELF_EN
) :
379 _MASKED_BIT_DISABLE(INSTPM_SELF_EN
);
380 I915_WRITE(INSTPM
, val
);
381 POSTING_READ(INSTPM
);
386 DRM_DEBUG_KMS("memory self-refresh is %s\n",
387 enable
? "enabled" : "disabled");
392 * Latency for FIFO fetches is dependent on several factors:
393 * - memory configuration (speed, channels)
395 * - current MCH state
396 * It can be fairly high in some situations, so here we assume a fairly
397 * pessimal value. It's a tradeoff between extra memory fetches (if we
398 * set this value too high, the FIFO will fetch frequently to stay full)
399 * and power consumption (set it too low to save power and we might see
400 * FIFO underruns and display "flicker").
402 * A value of 5us seems to be a good balance; safe for very low end
403 * platforms but not overly aggressive on lower latency configs.
405 static const int pessimal_latency_ns
= 5000;
407 #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \
408 ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8))
410 static int vlv_get_fifo_size(struct drm_device
*dev
,
411 enum pipe pipe
, int plane
)
413 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
414 int sprite0_start
, sprite1_start
, size
;
417 uint32_t dsparb
, dsparb2
, dsparb3
;
419 dsparb
= I915_READ(DSPARB
);
420 dsparb2
= I915_READ(DSPARB2
);
421 sprite0_start
= VLV_FIFO_START(dsparb
, dsparb2
, 0, 0);
422 sprite1_start
= VLV_FIFO_START(dsparb
, dsparb2
, 8, 4);
425 dsparb
= I915_READ(DSPARB
);
426 dsparb2
= I915_READ(DSPARB2
);
427 sprite0_start
= VLV_FIFO_START(dsparb
, dsparb2
, 16, 8);
428 sprite1_start
= VLV_FIFO_START(dsparb
, dsparb2
, 24, 12);
431 dsparb2
= I915_READ(DSPARB2
);
432 dsparb3
= I915_READ(DSPARB3
);
433 sprite0_start
= VLV_FIFO_START(dsparb3
, dsparb2
, 0, 16);
434 sprite1_start
= VLV_FIFO_START(dsparb3
, dsparb2
, 8, 20);
442 size
= sprite0_start
;
445 size
= sprite1_start
- sprite0_start
;
448 size
= 512 - 1 - sprite1_start
;
454 DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n",
455 pipe_name(pipe
), plane
== 0 ? "primary" : "sprite",
456 plane
== 0 ? plane_name(pipe
) : sprite_name(pipe
, plane
- 1),
462 static int i9xx_get_fifo_size(struct drm_device
*dev
, int plane
)
464 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
465 uint32_t dsparb
= I915_READ(DSPARB
);
468 size
= dsparb
& 0x7f;
470 size
= ((dsparb
>> DSPARB_CSTART_SHIFT
) & 0x7f) - size
;
472 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
473 plane
? "B" : "A", size
);
478 static int i830_get_fifo_size(struct drm_device
*dev
, int plane
)
480 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
481 uint32_t dsparb
= I915_READ(DSPARB
);
484 size
= dsparb
& 0x1ff;
486 size
= ((dsparb
>> DSPARB_BEND_SHIFT
) & 0x1ff) - size
;
487 size
>>= 1; /* Convert to cachelines */
489 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
490 plane
? "B" : "A", size
);
495 static int i845_get_fifo_size(struct drm_device
*dev
, int plane
)
497 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
498 uint32_t dsparb
= I915_READ(DSPARB
);
501 size
= dsparb
& 0x7f;
502 size
>>= 2; /* Convert to cachelines */
504 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
511 /* Pineview has different values for various configs */
512 static const struct intel_watermark_params pineview_display_wm
= {
513 .fifo_size
= PINEVIEW_DISPLAY_FIFO
,
514 .max_wm
= PINEVIEW_MAX_WM
,
515 .default_wm
= PINEVIEW_DFT_WM
,
516 .guard_size
= PINEVIEW_GUARD_WM
,
517 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
519 static const struct intel_watermark_params pineview_display_hplloff_wm
= {
520 .fifo_size
= PINEVIEW_DISPLAY_FIFO
,
521 .max_wm
= PINEVIEW_MAX_WM
,
522 .default_wm
= PINEVIEW_DFT_HPLLOFF_WM
,
523 .guard_size
= PINEVIEW_GUARD_WM
,
524 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
526 static const struct intel_watermark_params pineview_cursor_wm
= {
527 .fifo_size
= PINEVIEW_CURSOR_FIFO
,
528 .max_wm
= PINEVIEW_CURSOR_MAX_WM
,
529 .default_wm
= PINEVIEW_CURSOR_DFT_WM
,
530 .guard_size
= PINEVIEW_CURSOR_GUARD_WM
,
531 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
533 static const struct intel_watermark_params pineview_cursor_hplloff_wm
= {
534 .fifo_size
= PINEVIEW_CURSOR_FIFO
,
535 .max_wm
= PINEVIEW_CURSOR_MAX_WM
,
536 .default_wm
= PINEVIEW_CURSOR_DFT_WM
,
537 .guard_size
= PINEVIEW_CURSOR_GUARD_WM
,
538 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
540 static const struct intel_watermark_params g4x_wm_info
= {
541 .fifo_size
= G4X_FIFO_SIZE
,
542 .max_wm
= G4X_MAX_WM
,
543 .default_wm
= G4X_MAX_WM
,
545 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
547 static const struct intel_watermark_params g4x_cursor_wm_info
= {
548 .fifo_size
= I965_CURSOR_FIFO
,
549 .max_wm
= I965_CURSOR_MAX_WM
,
550 .default_wm
= I965_CURSOR_DFT_WM
,
552 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
554 static const struct intel_watermark_params valleyview_wm_info
= {
555 .fifo_size
= VALLEYVIEW_FIFO_SIZE
,
556 .max_wm
= VALLEYVIEW_MAX_WM
,
557 .default_wm
= VALLEYVIEW_MAX_WM
,
559 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
561 static const struct intel_watermark_params valleyview_cursor_wm_info
= {
562 .fifo_size
= I965_CURSOR_FIFO
,
563 .max_wm
= VALLEYVIEW_CURSOR_MAX_WM
,
564 .default_wm
= I965_CURSOR_DFT_WM
,
566 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
568 static const struct intel_watermark_params i965_cursor_wm_info
= {
569 .fifo_size
= I965_CURSOR_FIFO
,
570 .max_wm
= I965_CURSOR_MAX_WM
,
571 .default_wm
= I965_CURSOR_DFT_WM
,
573 .cacheline_size
= I915_FIFO_LINE_SIZE
,
575 static const struct intel_watermark_params i945_wm_info
= {
576 .fifo_size
= I945_FIFO_SIZE
,
577 .max_wm
= I915_MAX_WM
,
580 .cacheline_size
= I915_FIFO_LINE_SIZE
,
582 static const struct intel_watermark_params i915_wm_info
= {
583 .fifo_size
= I915_FIFO_SIZE
,
584 .max_wm
= I915_MAX_WM
,
587 .cacheline_size
= I915_FIFO_LINE_SIZE
,
589 static const struct intel_watermark_params i830_a_wm_info
= {
590 .fifo_size
= I855GM_FIFO_SIZE
,
591 .max_wm
= I915_MAX_WM
,
594 .cacheline_size
= I830_FIFO_LINE_SIZE
,
596 static const struct intel_watermark_params i830_bc_wm_info
= {
597 .fifo_size
= I855GM_FIFO_SIZE
,
598 .max_wm
= I915_MAX_WM
/2,
601 .cacheline_size
= I830_FIFO_LINE_SIZE
,
603 static const struct intel_watermark_params i845_wm_info
= {
604 .fifo_size
= I830_FIFO_SIZE
,
605 .max_wm
= I915_MAX_WM
,
608 .cacheline_size
= I830_FIFO_LINE_SIZE
,
612 * intel_calculate_wm - calculate watermark level
613 * @clock_in_khz: pixel clock
614 * @wm: chip FIFO params
615 * @pixel_size: display pixel size
616 * @latency_ns: memory latency for the platform
618 * Calculate the watermark level (the level at which the display plane will
619 * start fetching from memory again). Each chip has a different display
620 * FIFO size and allocation, so the caller needs to figure that out and pass
621 * in the correct intel_watermark_params structure.
623 * As the pixel clock runs, the FIFO will be drained at a rate that depends
624 * on the pixel size. When it reaches the watermark level, it'll start
625 * fetching FIFO line sized based chunks from memory until the FIFO fills
626 * past the watermark point. If the FIFO drains completely, a FIFO underrun
627 * will occur, and a display engine hang could result.
629 static unsigned long intel_calculate_wm(unsigned long clock_in_khz
,
630 const struct intel_watermark_params
*wm
,
633 unsigned long latency_ns
)
635 long entries_required
, wm_size
;
638 * Note: we need to make sure we don't overflow for various clock &
640 * clocks go from a few thousand to several hundred thousand.
641 * latency is usually a few thousand
643 entries_required
= ((clock_in_khz
/ 1000) * pixel_size
* latency_ns
) /
645 entries_required
= DIV_ROUND_UP(entries_required
, wm
->cacheline_size
);
647 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required
);
649 wm_size
= fifo_size
- (entries_required
+ wm
->guard_size
);
651 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size
);
653 /* Don't promote wm_size to unsigned... */
654 if (wm_size
> (long)wm
->max_wm
)
655 wm_size
= wm
->max_wm
;
657 wm_size
= wm
->default_wm
;
660 * Bspec seems to indicate that the value shouldn't be lower than
661 * 'burst size + 1'. Certainly 830 is quite unhappy with low values.
662 * Lets go for 8 which is the burst size since certain platforms
663 * already use a hardcoded 8 (which is what the spec says should be
672 static struct drm_crtc
*single_enabled_crtc(struct drm_device
*dev
)
674 struct drm_crtc
*crtc
, *enabled
= NULL
;
676 for_each_crtc(dev
, crtc
) {
677 if (intel_crtc_active(crtc
)) {
687 static void pineview_update_wm(struct drm_crtc
*unused_crtc
)
689 struct drm_device
*dev
= unused_crtc
->dev
;
690 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
691 struct drm_crtc
*crtc
;
692 const struct cxsr_latency
*latency
;
696 latency
= intel_get_cxsr_latency(IS_PINEVIEW_G(dev
), dev_priv
->is_ddr3
,
697 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
699 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
700 intel_set_memory_cxsr(dev_priv
, false);
704 crtc
= single_enabled_crtc(dev
);
706 const struct drm_display_mode
*adjusted_mode
;
707 int pixel_size
= crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
710 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
711 clock
= adjusted_mode
->crtc_clock
;
714 wm
= intel_calculate_wm(clock
, &pineview_display_wm
,
715 pineview_display_wm
.fifo_size
,
716 pixel_size
, latency
->display_sr
);
717 reg
= I915_READ(DSPFW1
);
718 reg
&= ~DSPFW_SR_MASK
;
719 reg
|= FW_WM(wm
, SR
);
720 I915_WRITE(DSPFW1
, reg
);
721 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg
);
724 wm
= intel_calculate_wm(clock
, &pineview_cursor_wm
,
725 pineview_display_wm
.fifo_size
,
726 pixel_size
, latency
->cursor_sr
);
727 reg
= I915_READ(DSPFW3
);
728 reg
&= ~DSPFW_CURSOR_SR_MASK
;
729 reg
|= FW_WM(wm
, CURSOR_SR
);
730 I915_WRITE(DSPFW3
, reg
);
732 /* Display HPLL off SR */
733 wm
= intel_calculate_wm(clock
, &pineview_display_hplloff_wm
,
734 pineview_display_hplloff_wm
.fifo_size
,
735 pixel_size
, latency
->display_hpll_disable
);
736 reg
= I915_READ(DSPFW3
);
737 reg
&= ~DSPFW_HPLL_SR_MASK
;
738 reg
|= FW_WM(wm
, HPLL_SR
);
739 I915_WRITE(DSPFW3
, reg
);
741 /* cursor HPLL off SR */
742 wm
= intel_calculate_wm(clock
, &pineview_cursor_hplloff_wm
,
743 pineview_display_hplloff_wm
.fifo_size
,
744 pixel_size
, latency
->cursor_hpll_disable
);
745 reg
= I915_READ(DSPFW3
);
746 reg
&= ~DSPFW_HPLL_CURSOR_MASK
;
747 reg
|= FW_WM(wm
, HPLL_CURSOR
);
748 I915_WRITE(DSPFW3
, reg
);
749 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg
);
751 intel_set_memory_cxsr(dev_priv
, true);
753 intel_set_memory_cxsr(dev_priv
, false);
757 static bool g4x_compute_wm0(struct drm_device
*dev
,
759 const struct intel_watermark_params
*display
,
760 int display_latency_ns
,
761 const struct intel_watermark_params
*cursor
,
762 int cursor_latency_ns
,
766 struct drm_crtc
*crtc
;
767 const struct drm_display_mode
*adjusted_mode
;
768 int htotal
, hdisplay
, clock
, pixel_size
;
769 int line_time_us
, line_count
;
770 int entries
, tlb_miss
;
772 crtc
= intel_get_crtc_for_plane(dev
, plane
);
773 if (!intel_crtc_active(crtc
)) {
774 *cursor_wm
= cursor
->guard_size
;
775 *plane_wm
= display
->guard_size
;
779 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
780 clock
= adjusted_mode
->crtc_clock
;
781 htotal
= adjusted_mode
->crtc_htotal
;
782 hdisplay
= to_intel_crtc(crtc
)->config
->pipe_src_w
;
783 pixel_size
= crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
785 /* Use the small buffer method to calculate plane watermark */
786 entries
= ((clock
* pixel_size
/ 1000) * display_latency_ns
) / 1000;
787 tlb_miss
= display
->fifo_size
*display
->cacheline_size
- hdisplay
* 8;
790 entries
= DIV_ROUND_UP(entries
, display
->cacheline_size
);
791 *plane_wm
= entries
+ display
->guard_size
;
792 if (*plane_wm
> (int)display
->max_wm
)
793 *plane_wm
= display
->max_wm
;
795 /* Use the large buffer method to calculate cursor watermark */
796 line_time_us
= max(htotal
* 1000 / clock
, 1);
797 line_count
= (cursor_latency_ns
/ line_time_us
+ 1000) / 1000;
798 entries
= line_count
* crtc
->cursor
->state
->crtc_w
* pixel_size
;
799 tlb_miss
= cursor
->fifo_size
*cursor
->cacheline_size
- hdisplay
* 8;
802 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
803 *cursor_wm
= entries
+ cursor
->guard_size
;
804 if (*cursor_wm
> (int)cursor
->max_wm
)
805 *cursor_wm
= (int)cursor
->max_wm
;
811 * Check the wm result.
813 * If any calculated watermark values is larger than the maximum value that
814 * can be programmed into the associated watermark register, that watermark
817 static bool g4x_check_srwm(struct drm_device
*dev
,
818 int display_wm
, int cursor_wm
,
819 const struct intel_watermark_params
*display
,
820 const struct intel_watermark_params
*cursor
)
822 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
823 display_wm
, cursor_wm
);
825 if (display_wm
> display
->max_wm
) {
826 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
827 display_wm
, display
->max_wm
);
831 if (cursor_wm
> cursor
->max_wm
) {
832 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
833 cursor_wm
, cursor
->max_wm
);
837 if (!(display_wm
|| cursor_wm
)) {
838 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
845 static bool g4x_compute_srwm(struct drm_device
*dev
,
848 const struct intel_watermark_params
*display
,
849 const struct intel_watermark_params
*cursor
,
850 int *display_wm
, int *cursor_wm
)
852 struct drm_crtc
*crtc
;
853 const struct drm_display_mode
*adjusted_mode
;
854 int hdisplay
, htotal
, pixel_size
, clock
;
855 unsigned long line_time_us
;
856 int line_count
, line_size
;
861 *display_wm
= *cursor_wm
= 0;
865 crtc
= intel_get_crtc_for_plane(dev
, plane
);
866 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
867 clock
= adjusted_mode
->crtc_clock
;
868 htotal
= adjusted_mode
->crtc_htotal
;
869 hdisplay
= to_intel_crtc(crtc
)->config
->pipe_src_w
;
870 pixel_size
= crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
872 line_time_us
= max(htotal
* 1000 / clock
, 1);
873 line_count
= (latency_ns
/ line_time_us
+ 1000) / 1000;
874 line_size
= hdisplay
* pixel_size
;
876 /* Use the minimum of the small and large buffer method for primary */
877 small
= ((clock
* pixel_size
/ 1000) * latency_ns
) / 1000;
878 large
= line_count
* line_size
;
880 entries
= DIV_ROUND_UP(min(small
, large
), display
->cacheline_size
);
881 *display_wm
= entries
+ display
->guard_size
;
883 /* calculate the self-refresh watermark for display cursor */
884 entries
= line_count
* pixel_size
* crtc
->cursor
->state
->crtc_w
;
885 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
886 *cursor_wm
= entries
+ cursor
->guard_size
;
888 return g4x_check_srwm(dev
,
889 *display_wm
, *cursor_wm
,
893 #define FW_WM_VLV(value, plane) \
894 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV)
896 static void vlv_write_wm_values(struct intel_crtc
*crtc
,
897 const struct vlv_wm_values
*wm
)
899 struct drm_i915_private
*dev_priv
= to_i915(crtc
->base
.dev
);
900 enum pipe pipe
= crtc
->pipe
;
902 I915_WRITE(VLV_DDL(pipe
),
903 (wm
->ddl
[pipe
].cursor
<< DDL_CURSOR_SHIFT
) |
904 (wm
->ddl
[pipe
].sprite
[1] << DDL_SPRITE_SHIFT(1)) |
905 (wm
->ddl
[pipe
].sprite
[0] << DDL_SPRITE_SHIFT(0)) |
906 (wm
->ddl
[pipe
].primary
<< DDL_PLANE_SHIFT
));
909 FW_WM(wm
->sr
.plane
, SR
) |
910 FW_WM(wm
->pipe
[PIPE_B
].cursor
, CURSORB
) |
911 FW_WM_VLV(wm
->pipe
[PIPE_B
].primary
, PLANEB
) |
912 FW_WM_VLV(wm
->pipe
[PIPE_A
].primary
, PLANEA
));
914 FW_WM_VLV(wm
->pipe
[PIPE_A
].sprite
[1], SPRITEB
) |
915 FW_WM(wm
->pipe
[PIPE_A
].cursor
, CURSORA
) |
916 FW_WM_VLV(wm
->pipe
[PIPE_A
].sprite
[0], SPRITEA
));
918 FW_WM(wm
->sr
.cursor
, CURSOR_SR
));
920 if (IS_CHERRYVIEW(dev_priv
)) {
921 I915_WRITE(DSPFW7_CHV
,
922 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[1], SPRITED
) |
923 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[0], SPRITEC
));
924 I915_WRITE(DSPFW8_CHV
,
925 FW_WM_VLV(wm
->pipe
[PIPE_C
].sprite
[1], SPRITEF
) |
926 FW_WM_VLV(wm
->pipe
[PIPE_C
].sprite
[0], SPRITEE
));
927 I915_WRITE(DSPFW9_CHV
,
928 FW_WM_VLV(wm
->pipe
[PIPE_C
].primary
, PLANEC
) |
929 FW_WM(wm
->pipe
[PIPE_C
].cursor
, CURSORC
));
931 FW_WM(wm
->sr
.plane
>> 9, SR_HI
) |
932 FW_WM(wm
->pipe
[PIPE_C
].sprite
[1] >> 8, SPRITEF_HI
) |
933 FW_WM(wm
->pipe
[PIPE_C
].sprite
[0] >> 8, SPRITEE_HI
) |
934 FW_WM(wm
->pipe
[PIPE_C
].primary
>> 8, PLANEC_HI
) |
935 FW_WM(wm
->pipe
[PIPE_B
].sprite
[1] >> 8, SPRITED_HI
) |
936 FW_WM(wm
->pipe
[PIPE_B
].sprite
[0] >> 8, SPRITEC_HI
) |
937 FW_WM(wm
->pipe
[PIPE_B
].primary
>> 8, PLANEB_HI
) |
938 FW_WM(wm
->pipe
[PIPE_A
].sprite
[1] >> 8, SPRITEB_HI
) |
939 FW_WM(wm
->pipe
[PIPE_A
].sprite
[0] >> 8, SPRITEA_HI
) |
940 FW_WM(wm
->pipe
[PIPE_A
].primary
>> 8, PLANEA_HI
));
943 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[1], SPRITED
) |
944 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[0], SPRITEC
));
946 FW_WM(wm
->sr
.plane
>> 9, SR_HI
) |
947 FW_WM(wm
->pipe
[PIPE_B
].sprite
[1] >> 8, SPRITED_HI
) |
948 FW_WM(wm
->pipe
[PIPE_B
].sprite
[0] >> 8, SPRITEC_HI
) |
949 FW_WM(wm
->pipe
[PIPE_B
].primary
>> 8, PLANEB_HI
) |
950 FW_WM(wm
->pipe
[PIPE_A
].sprite
[1] >> 8, SPRITEB_HI
) |
951 FW_WM(wm
->pipe
[PIPE_A
].sprite
[0] >> 8, SPRITEA_HI
) |
952 FW_WM(wm
->pipe
[PIPE_A
].primary
>> 8, PLANEA_HI
));
955 /* zero (unused) WM1 watermarks */
956 I915_WRITE(DSPFW4
, 0);
957 I915_WRITE(DSPFW5
, 0);
958 I915_WRITE(DSPFW6
, 0);
959 I915_WRITE(DSPHOWM1
, 0);
961 POSTING_READ(DSPFW1
);
969 VLV_WM_LEVEL_DDR_DVFS
,
971 VLV_WM_NUM_LEVELS
= 1,
974 /* latency must be in 0.1us units. */
975 static unsigned int vlv_wm_method2(unsigned int pixel_rate
,
976 unsigned int pipe_htotal
,
977 unsigned int horiz_pixels
,
978 unsigned int bytes_per_pixel
,
979 unsigned int latency
)
983 ret
= (latency
* pixel_rate
) / (pipe_htotal
* 10000);
984 ret
= (ret
+ 1) * horiz_pixels
* bytes_per_pixel
;
985 ret
= DIV_ROUND_UP(ret
, 64);
990 static void vlv_setup_wm_latency(struct drm_device
*dev
)
992 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
994 /* all latencies in usec */
995 dev_priv
->wm
.pri_latency
[VLV_WM_LEVEL_PM2
] = 3;
997 if (IS_CHERRYVIEW(dev_priv
)) {
998 dev_priv
->wm
.pri_latency
[VLV_WM_LEVEL_PM5
] = 12;
999 dev_priv
->wm
.pri_latency
[VLV_WM_LEVEL_DDR_DVFS
] = 33;
1003 static uint16_t vlv_compute_wm_level(struct intel_plane
*plane
,
1004 struct intel_crtc
*crtc
,
1005 const struct intel_plane_state
*state
,
1008 struct drm_i915_private
*dev_priv
= to_i915(plane
->base
.dev
);
1009 int clock
, htotal
, pixel_size
, width
, wm
;
1011 if (dev_priv
->wm
.pri_latency
[level
] == 0)
1014 if (!state
->visible
)
1017 pixel_size
= drm_format_plane_cpp(state
->base
.fb
->pixel_format
, 0);
1018 clock
= crtc
->config
->base
.adjusted_mode
.crtc_clock
;
1019 htotal
= crtc
->config
->base
.adjusted_mode
.crtc_htotal
;
1020 width
= crtc
->config
->pipe_src_w
;
1021 if (WARN_ON(htotal
== 0))
1024 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
) {
1026 * FIXME the formula gives values that are
1027 * too big for the cursor FIFO, and hence we
1028 * would never be able to use cursors. For
1029 * now just hardcode the watermark.
1033 wm
= vlv_wm_method2(clock
, htotal
, width
, pixel_size
,
1034 dev_priv
->wm
.pri_latency
[level
] * 10);
1037 return min_t(int, wm
, USHRT_MAX
);
1040 static void vlv_compute_fifo(struct intel_crtc
*crtc
)
1042 struct drm_device
*dev
= crtc
->base
.dev
;
1043 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1044 struct intel_plane
*plane
;
1045 unsigned int total_rate
= 0;
1046 const int fifo_size
= 512 - 1;
1047 int fifo_extra
, fifo_left
= fifo_size
;
1049 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1050 struct intel_plane_state
*state
=
1051 to_intel_plane_state(plane
->base
.state
);
1053 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
)
1056 if (state
->visible
) {
1057 wm_state
->num_active_planes
++;
1058 total_rate
+= drm_format_plane_cpp(state
->base
.fb
->pixel_format
, 0);
1062 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1063 struct intel_plane_state
*state
=
1064 to_intel_plane_state(plane
->base
.state
);
1067 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
) {
1068 plane
->wm
.fifo_size
= 63;
1072 if (!state
->visible
) {
1073 plane
->wm
.fifo_size
= 0;
1077 rate
= drm_format_plane_cpp(state
->base
.fb
->pixel_format
, 0);
1078 plane
->wm
.fifo_size
= fifo_size
* rate
/ total_rate
;
1079 fifo_left
-= plane
->wm
.fifo_size
;
1082 fifo_extra
= DIV_ROUND_UP(fifo_left
, wm_state
->num_active_planes
?: 1);
1084 /* spread the remainder evenly */
1085 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1091 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
)
1094 /* give it all to the first plane if none are active */
1095 if (plane
->wm
.fifo_size
== 0 &&
1096 wm_state
->num_active_planes
)
1099 plane_extra
= min(fifo_extra
, fifo_left
);
1100 plane
->wm
.fifo_size
+= plane_extra
;
1101 fifo_left
-= plane_extra
;
1104 WARN_ON(fifo_left
!= 0);
1107 static void vlv_invert_wms(struct intel_crtc
*crtc
)
1109 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1112 for (level
= 0; level
< wm_state
->num_levels
; level
++) {
1113 struct drm_device
*dev
= crtc
->base
.dev
;
1114 const int sr_fifo_size
= INTEL_INFO(dev
)->num_pipes
* 512 - 1;
1115 struct intel_plane
*plane
;
1117 wm_state
->sr
[level
].plane
= sr_fifo_size
- wm_state
->sr
[level
].plane
;
1118 wm_state
->sr
[level
].cursor
= 63 - wm_state
->sr
[level
].cursor
;
1120 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1121 switch (plane
->base
.type
) {
1123 case DRM_PLANE_TYPE_CURSOR
:
1124 wm_state
->wm
[level
].cursor
= plane
->wm
.fifo_size
-
1125 wm_state
->wm
[level
].cursor
;
1127 case DRM_PLANE_TYPE_PRIMARY
:
1128 wm_state
->wm
[level
].primary
= plane
->wm
.fifo_size
-
1129 wm_state
->wm
[level
].primary
;
1131 case DRM_PLANE_TYPE_OVERLAY
:
1132 sprite
= plane
->plane
;
1133 wm_state
->wm
[level
].sprite
[sprite
] = plane
->wm
.fifo_size
-
1134 wm_state
->wm
[level
].sprite
[sprite
];
1141 static void vlv_compute_wm(struct intel_crtc
*crtc
)
1143 struct drm_device
*dev
= crtc
->base
.dev
;
1144 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1145 struct intel_plane
*plane
;
1146 int sr_fifo_size
= INTEL_INFO(dev
)->num_pipes
* 512 - 1;
1149 memset(wm_state
, 0, sizeof(*wm_state
));
1151 wm_state
->cxsr
= crtc
->pipe
!= PIPE_C
&& crtc
->wm
.cxsr_allowed
;
1152 if (IS_CHERRYVIEW(dev
))
1153 wm_state
->num_levels
= CHV_WM_NUM_LEVELS
;
1155 wm_state
->num_levels
= VLV_WM_NUM_LEVELS
;
1157 wm_state
->num_active_planes
= 0;
1159 vlv_compute_fifo(crtc
);
1161 if (wm_state
->num_active_planes
!= 1)
1162 wm_state
->cxsr
= false;
1164 if (wm_state
->cxsr
) {
1165 for (level
= 0; level
< wm_state
->num_levels
; level
++) {
1166 wm_state
->sr
[level
].plane
= sr_fifo_size
;
1167 wm_state
->sr
[level
].cursor
= 63;
1171 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1172 struct intel_plane_state
*state
=
1173 to_intel_plane_state(plane
->base
.state
);
1175 if (!state
->visible
)
1178 /* normal watermarks */
1179 for (level
= 0; level
< wm_state
->num_levels
; level
++) {
1180 int wm
= vlv_compute_wm_level(plane
, crtc
, state
, level
);
1181 int max_wm
= plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
? 63 : 511;
1184 if (WARN_ON(level
== 0 && wm
> max_wm
))
1187 if (wm
> plane
->wm
.fifo_size
)
1190 switch (plane
->base
.type
) {
1192 case DRM_PLANE_TYPE_CURSOR
:
1193 wm_state
->wm
[level
].cursor
= wm
;
1195 case DRM_PLANE_TYPE_PRIMARY
:
1196 wm_state
->wm
[level
].primary
= wm
;
1198 case DRM_PLANE_TYPE_OVERLAY
:
1199 sprite
= plane
->plane
;
1200 wm_state
->wm
[level
].sprite
[sprite
] = wm
;
1205 wm_state
->num_levels
= level
;
1207 if (!wm_state
->cxsr
)
1210 /* maxfifo watermarks */
1211 switch (plane
->base
.type
) {
1213 case DRM_PLANE_TYPE_CURSOR
:
1214 for (level
= 0; level
< wm_state
->num_levels
; level
++)
1215 wm_state
->sr
[level
].cursor
=
1216 wm_state
->sr
[level
].cursor
;
1218 case DRM_PLANE_TYPE_PRIMARY
:
1219 for (level
= 0; level
< wm_state
->num_levels
; level
++)
1220 wm_state
->sr
[level
].plane
=
1221 min(wm_state
->sr
[level
].plane
,
1222 wm_state
->wm
[level
].primary
);
1224 case DRM_PLANE_TYPE_OVERLAY
:
1225 sprite
= plane
->plane
;
1226 for (level
= 0; level
< wm_state
->num_levels
; level
++)
1227 wm_state
->sr
[level
].plane
=
1228 min(wm_state
->sr
[level
].plane
,
1229 wm_state
->wm
[level
].sprite
[sprite
]);
1234 /* clear any (partially) filled invalid levels */
1235 for (level
= wm_state
->num_levels
; level
< CHV_WM_NUM_LEVELS
; level
++) {
1236 memset(&wm_state
->wm
[level
], 0, sizeof(wm_state
->wm
[level
]));
1237 memset(&wm_state
->sr
[level
], 0, sizeof(wm_state
->sr
[level
]));
1240 vlv_invert_wms(crtc
);
1243 #define VLV_FIFO(plane, value) \
1244 (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV)
1246 static void vlv_pipe_set_fifo_size(struct intel_crtc
*crtc
)
1248 struct drm_device
*dev
= crtc
->base
.dev
;
1249 struct drm_i915_private
*dev_priv
= to_i915(dev
);
1250 struct intel_plane
*plane
;
1251 int sprite0_start
= 0, sprite1_start
= 0, fifo_size
= 0;
1253 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1254 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
) {
1255 WARN_ON(plane
->wm
.fifo_size
!= 63);
1259 if (plane
->base
.type
== DRM_PLANE_TYPE_PRIMARY
)
1260 sprite0_start
= plane
->wm
.fifo_size
;
1261 else if (plane
->plane
== 0)
1262 sprite1_start
= sprite0_start
+ plane
->wm
.fifo_size
;
1264 fifo_size
= sprite1_start
+ plane
->wm
.fifo_size
;
1267 WARN_ON(fifo_size
!= 512 - 1);
1269 DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n",
1270 pipe_name(crtc
->pipe
), sprite0_start
,
1271 sprite1_start
, fifo_size
);
1273 switch (crtc
->pipe
) {
1274 uint32_t dsparb
, dsparb2
, dsparb3
;
1276 dsparb
= I915_READ(DSPARB
);
1277 dsparb2
= I915_READ(DSPARB2
);
1279 dsparb
&= ~(VLV_FIFO(SPRITEA
, 0xff) |
1280 VLV_FIFO(SPRITEB
, 0xff));
1281 dsparb
|= (VLV_FIFO(SPRITEA
, sprite0_start
) |
1282 VLV_FIFO(SPRITEB
, sprite1_start
));
1284 dsparb2
&= ~(VLV_FIFO(SPRITEA_HI
, 0x1) |
1285 VLV_FIFO(SPRITEB_HI
, 0x1));
1286 dsparb2
|= (VLV_FIFO(SPRITEA_HI
, sprite0_start
>> 8) |
1287 VLV_FIFO(SPRITEB_HI
, sprite1_start
>> 8));
1289 I915_WRITE(DSPARB
, dsparb
);
1290 I915_WRITE(DSPARB2
, dsparb2
);
1293 dsparb
= I915_READ(DSPARB
);
1294 dsparb2
= I915_READ(DSPARB2
);
1296 dsparb
&= ~(VLV_FIFO(SPRITEC
, 0xff) |
1297 VLV_FIFO(SPRITED
, 0xff));
1298 dsparb
|= (VLV_FIFO(SPRITEC
, sprite0_start
) |
1299 VLV_FIFO(SPRITED
, sprite1_start
));
1301 dsparb2
&= ~(VLV_FIFO(SPRITEC_HI
, 0xff) |
1302 VLV_FIFO(SPRITED_HI
, 0xff));
1303 dsparb2
|= (VLV_FIFO(SPRITEC_HI
, sprite0_start
>> 8) |
1304 VLV_FIFO(SPRITED_HI
, sprite1_start
>> 8));
1306 I915_WRITE(DSPARB
, dsparb
);
1307 I915_WRITE(DSPARB2
, dsparb2
);
1310 dsparb3
= I915_READ(DSPARB3
);
1311 dsparb2
= I915_READ(DSPARB2
);
1313 dsparb3
&= ~(VLV_FIFO(SPRITEE
, 0xff) |
1314 VLV_FIFO(SPRITEF
, 0xff));
1315 dsparb3
|= (VLV_FIFO(SPRITEE
, sprite0_start
) |
1316 VLV_FIFO(SPRITEF
, sprite1_start
));
1318 dsparb2
&= ~(VLV_FIFO(SPRITEE_HI
, 0xff) |
1319 VLV_FIFO(SPRITEF_HI
, 0xff));
1320 dsparb2
|= (VLV_FIFO(SPRITEE_HI
, sprite0_start
>> 8) |
1321 VLV_FIFO(SPRITEF_HI
, sprite1_start
>> 8));
1323 I915_WRITE(DSPARB3
, dsparb3
);
1324 I915_WRITE(DSPARB2
, dsparb2
);
1333 static void vlv_merge_wm(struct drm_device
*dev
,
1334 struct vlv_wm_values
*wm
)
1336 struct intel_crtc
*crtc
;
1337 int num_active_crtcs
= 0;
1339 if (IS_CHERRYVIEW(dev
))
1340 wm
->level
= VLV_WM_LEVEL_DDR_DVFS
;
1342 wm
->level
= VLV_WM_LEVEL_PM2
;
1345 for_each_intel_crtc(dev
, crtc
) {
1346 const struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1351 if (!wm_state
->cxsr
)
1355 wm
->level
= min_t(int, wm
->level
, wm_state
->num_levels
- 1);
1358 if (num_active_crtcs
!= 1)
1361 if (num_active_crtcs
> 1)
1362 wm
->level
= VLV_WM_LEVEL_PM2
;
1364 for_each_intel_crtc(dev
, crtc
) {
1365 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1366 enum pipe pipe
= crtc
->pipe
;
1371 wm
->pipe
[pipe
] = wm_state
->wm
[wm
->level
];
1373 wm
->sr
= wm_state
->sr
[wm
->level
];
1375 wm
->ddl
[pipe
].primary
= DDL_PRECISION_HIGH
| 2;
1376 wm
->ddl
[pipe
].sprite
[0] = DDL_PRECISION_HIGH
| 2;
1377 wm
->ddl
[pipe
].sprite
[1] = DDL_PRECISION_HIGH
| 2;
1378 wm
->ddl
[pipe
].cursor
= DDL_PRECISION_HIGH
| 2;
1382 static void vlv_update_wm(struct drm_crtc
*crtc
)
1384 struct drm_device
*dev
= crtc
->dev
;
1385 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1386 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
1387 enum pipe pipe
= intel_crtc
->pipe
;
1388 struct vlv_wm_values wm
= {};
1390 vlv_compute_wm(intel_crtc
);
1391 vlv_merge_wm(dev
, &wm
);
1393 if (memcmp(&dev_priv
->wm
.vlv
, &wm
, sizeof(wm
)) == 0) {
1394 /* FIXME should be part of crtc atomic commit */
1395 vlv_pipe_set_fifo_size(intel_crtc
);
1399 if (wm
.level
< VLV_WM_LEVEL_DDR_DVFS
&&
1400 dev_priv
->wm
.vlv
.level
>= VLV_WM_LEVEL_DDR_DVFS
)
1401 chv_set_memory_dvfs(dev_priv
, false);
1403 if (wm
.level
< VLV_WM_LEVEL_PM5
&&
1404 dev_priv
->wm
.vlv
.level
>= VLV_WM_LEVEL_PM5
)
1405 chv_set_memory_pm5(dev_priv
, false);
1407 if (!wm
.cxsr
&& dev_priv
->wm
.vlv
.cxsr
)
1408 intel_set_memory_cxsr(dev_priv
, false);
1410 /* FIXME should be part of crtc atomic commit */
1411 vlv_pipe_set_fifo_size(intel_crtc
);
1413 vlv_write_wm_values(intel_crtc
, &wm
);
1415 DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, "
1416 "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n",
1417 pipe_name(pipe
), wm
.pipe
[pipe
].primary
, wm
.pipe
[pipe
].cursor
,
1418 wm
.pipe
[pipe
].sprite
[0], wm
.pipe
[pipe
].sprite
[1],
1419 wm
.sr
.plane
, wm
.sr
.cursor
, wm
.level
, wm
.cxsr
);
1421 if (wm
.cxsr
&& !dev_priv
->wm
.vlv
.cxsr
)
1422 intel_set_memory_cxsr(dev_priv
, true);
1424 if (wm
.level
>= VLV_WM_LEVEL_PM5
&&
1425 dev_priv
->wm
.vlv
.level
< VLV_WM_LEVEL_PM5
)
1426 chv_set_memory_pm5(dev_priv
, true);
1428 if (wm
.level
>= VLV_WM_LEVEL_DDR_DVFS
&&
1429 dev_priv
->wm
.vlv
.level
< VLV_WM_LEVEL_DDR_DVFS
)
1430 chv_set_memory_dvfs(dev_priv
, true);
1432 dev_priv
->wm
.vlv
= wm
;
1435 #define single_plane_enabled(mask) is_power_of_2(mask)
1437 static void g4x_update_wm(struct drm_crtc
*crtc
)
1439 struct drm_device
*dev
= crtc
->dev
;
1440 static const int sr_latency_ns
= 12000;
1441 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1442 int planea_wm
, planeb_wm
, cursora_wm
, cursorb_wm
;
1443 int plane_sr
, cursor_sr
;
1444 unsigned int enabled
= 0;
1447 if (g4x_compute_wm0(dev
, PIPE_A
,
1448 &g4x_wm_info
, pessimal_latency_ns
,
1449 &g4x_cursor_wm_info
, pessimal_latency_ns
,
1450 &planea_wm
, &cursora_wm
))
1451 enabled
|= 1 << PIPE_A
;
1453 if (g4x_compute_wm0(dev
, PIPE_B
,
1454 &g4x_wm_info
, pessimal_latency_ns
,
1455 &g4x_cursor_wm_info
, pessimal_latency_ns
,
1456 &planeb_wm
, &cursorb_wm
))
1457 enabled
|= 1 << PIPE_B
;
1459 if (single_plane_enabled(enabled
) &&
1460 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1463 &g4x_cursor_wm_info
,
1464 &plane_sr
, &cursor_sr
)) {
1465 cxsr_enabled
= true;
1467 cxsr_enabled
= false;
1468 intel_set_memory_cxsr(dev_priv
, false);
1469 plane_sr
= cursor_sr
= 0;
1472 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, "
1473 "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1474 planea_wm
, cursora_wm
,
1475 planeb_wm
, cursorb_wm
,
1476 plane_sr
, cursor_sr
);
1479 FW_WM(plane_sr
, SR
) |
1480 FW_WM(cursorb_wm
, CURSORB
) |
1481 FW_WM(planeb_wm
, PLANEB
) |
1482 FW_WM(planea_wm
, PLANEA
));
1484 (I915_READ(DSPFW2
) & ~DSPFW_CURSORA_MASK
) |
1485 FW_WM(cursora_wm
, CURSORA
));
1486 /* HPLL off in SR has some issues on G4x... disable it */
1488 (I915_READ(DSPFW3
) & ~(DSPFW_HPLL_SR_EN
| DSPFW_CURSOR_SR_MASK
)) |
1489 FW_WM(cursor_sr
, CURSOR_SR
));
1492 intel_set_memory_cxsr(dev_priv
, true);
1495 static void i965_update_wm(struct drm_crtc
*unused_crtc
)
1497 struct drm_device
*dev
= unused_crtc
->dev
;
1498 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1499 struct drm_crtc
*crtc
;
1504 /* Calc sr entries for one plane configs */
1505 crtc
= single_enabled_crtc(dev
);
1507 /* self-refresh has much higher latency */
1508 static const int sr_latency_ns
= 12000;
1509 const struct drm_display_mode
*adjusted_mode
=
1510 &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1511 int clock
= adjusted_mode
->crtc_clock
;
1512 int htotal
= adjusted_mode
->crtc_htotal
;
1513 int hdisplay
= to_intel_crtc(crtc
)->config
->pipe_src_w
;
1514 int pixel_size
= crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
1515 unsigned long line_time_us
;
1518 line_time_us
= max(htotal
* 1000 / clock
, 1);
1520 /* Use ns/us then divide to preserve precision */
1521 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1522 pixel_size
* hdisplay
;
1523 entries
= DIV_ROUND_UP(entries
, I915_FIFO_LINE_SIZE
);
1524 srwm
= I965_FIFO_SIZE
- entries
;
1528 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1531 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1532 pixel_size
* crtc
->cursor
->state
->crtc_w
;
1533 entries
= DIV_ROUND_UP(entries
,
1534 i965_cursor_wm_info
.cacheline_size
);
1535 cursor_sr
= i965_cursor_wm_info
.fifo_size
-
1536 (entries
+ i965_cursor_wm_info
.guard_size
);
1538 if (cursor_sr
> i965_cursor_wm_info
.max_wm
)
1539 cursor_sr
= i965_cursor_wm_info
.max_wm
;
1541 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1542 "cursor %d\n", srwm
, cursor_sr
);
1544 cxsr_enabled
= true;
1546 cxsr_enabled
= false;
1547 /* Turn off self refresh if both pipes are enabled */
1548 intel_set_memory_cxsr(dev_priv
, false);
1551 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1554 /* 965 has limitations... */
1555 I915_WRITE(DSPFW1
, FW_WM(srwm
, SR
) |
1559 I915_WRITE(DSPFW2
, FW_WM(8, CURSORA
) |
1560 FW_WM(8, PLANEC_OLD
));
1561 /* update cursor SR watermark */
1562 I915_WRITE(DSPFW3
, FW_WM(cursor_sr
, CURSOR_SR
));
1565 intel_set_memory_cxsr(dev_priv
, true);
1570 static void i9xx_update_wm(struct drm_crtc
*unused_crtc
)
1572 struct drm_device
*dev
= unused_crtc
->dev
;
1573 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1574 const struct intel_watermark_params
*wm_info
;
1579 int planea_wm
, planeb_wm
;
1580 struct drm_crtc
*crtc
, *enabled
= NULL
;
1583 wm_info
= &i945_wm_info
;
1584 else if (!IS_GEN2(dev
))
1585 wm_info
= &i915_wm_info
;
1587 wm_info
= &i830_a_wm_info
;
1589 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 0);
1590 crtc
= intel_get_crtc_for_plane(dev
, 0);
1591 if (intel_crtc_active(crtc
)) {
1592 const struct drm_display_mode
*adjusted_mode
;
1593 int cpp
= crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
1597 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1598 planea_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1599 wm_info
, fifo_size
, cpp
,
1600 pessimal_latency_ns
);
1603 planea_wm
= fifo_size
- wm_info
->guard_size
;
1604 if (planea_wm
> (long)wm_info
->max_wm
)
1605 planea_wm
= wm_info
->max_wm
;
1609 wm_info
= &i830_bc_wm_info
;
1611 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 1);
1612 crtc
= intel_get_crtc_for_plane(dev
, 1);
1613 if (intel_crtc_active(crtc
)) {
1614 const struct drm_display_mode
*adjusted_mode
;
1615 int cpp
= crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
1619 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1620 planeb_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1621 wm_info
, fifo_size
, cpp
,
1622 pessimal_latency_ns
);
1623 if (enabled
== NULL
)
1628 planeb_wm
= fifo_size
- wm_info
->guard_size
;
1629 if (planeb_wm
> (long)wm_info
->max_wm
)
1630 planeb_wm
= wm_info
->max_wm
;
1633 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm
, planeb_wm
);
1635 if (IS_I915GM(dev
) && enabled
) {
1636 struct drm_i915_gem_object
*obj
;
1638 obj
= intel_fb_obj(enabled
->primary
->state
->fb
);
1640 /* self-refresh seems busted with untiled */
1641 if (obj
->tiling_mode
== I915_TILING_NONE
)
1646 * Overlay gets an aggressive default since video jitter is bad.
1650 /* Play safe and disable self-refresh before adjusting watermarks. */
1651 intel_set_memory_cxsr(dev_priv
, false);
1653 /* Calc sr entries for one plane configs */
1654 if (HAS_FW_BLC(dev
) && enabled
) {
1655 /* self-refresh has much higher latency */
1656 static const int sr_latency_ns
= 6000;
1657 const struct drm_display_mode
*adjusted_mode
=
1658 &to_intel_crtc(enabled
)->config
->base
.adjusted_mode
;
1659 int clock
= adjusted_mode
->crtc_clock
;
1660 int htotal
= adjusted_mode
->crtc_htotal
;
1661 int hdisplay
= to_intel_crtc(enabled
)->config
->pipe_src_w
;
1662 int pixel_size
= enabled
->primary
->state
->fb
->bits_per_pixel
/ 8;
1663 unsigned long line_time_us
;
1666 line_time_us
= max(htotal
* 1000 / clock
, 1);
1668 /* Use ns/us then divide to preserve precision */
1669 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1670 pixel_size
* hdisplay
;
1671 entries
= DIV_ROUND_UP(entries
, wm_info
->cacheline_size
);
1672 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries
);
1673 srwm
= wm_info
->fifo_size
- entries
;
1677 if (IS_I945G(dev
) || IS_I945GM(dev
))
1678 I915_WRITE(FW_BLC_SELF
,
1679 FW_BLC_SELF_FIFO_MASK
| (srwm
& 0xff));
1680 else if (IS_I915GM(dev
))
1681 I915_WRITE(FW_BLC_SELF
, srwm
& 0x3f);
1684 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1685 planea_wm
, planeb_wm
, cwm
, srwm
);
1687 fwater_lo
= ((planeb_wm
& 0x3f) << 16) | (planea_wm
& 0x3f);
1688 fwater_hi
= (cwm
& 0x1f);
1690 /* Set request length to 8 cachelines per fetch */
1691 fwater_lo
= fwater_lo
| (1 << 24) | (1 << 8);
1692 fwater_hi
= fwater_hi
| (1 << 8);
1694 I915_WRITE(FW_BLC
, fwater_lo
);
1695 I915_WRITE(FW_BLC2
, fwater_hi
);
1698 intel_set_memory_cxsr(dev_priv
, true);
1701 static void i845_update_wm(struct drm_crtc
*unused_crtc
)
1703 struct drm_device
*dev
= unused_crtc
->dev
;
1704 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1705 struct drm_crtc
*crtc
;
1706 const struct drm_display_mode
*adjusted_mode
;
1710 crtc
= single_enabled_crtc(dev
);
1714 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1715 planea_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1717 dev_priv
->display
.get_fifo_size(dev
, 0),
1718 4, pessimal_latency_ns
);
1719 fwater_lo
= I915_READ(FW_BLC
) & ~0xfff;
1720 fwater_lo
|= (3<<8) | planea_wm
;
1722 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm
);
1724 I915_WRITE(FW_BLC
, fwater_lo
);
1727 uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state
*pipe_config
)
1729 uint32_t pixel_rate
;
1731 pixel_rate
= pipe_config
->base
.adjusted_mode
.crtc_clock
;
1733 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1734 * adjust the pixel_rate here. */
1736 if (pipe_config
->pch_pfit
.enabled
) {
1737 uint64_t pipe_w
, pipe_h
, pfit_w
, pfit_h
;
1738 uint32_t pfit_size
= pipe_config
->pch_pfit
.size
;
1740 pipe_w
= pipe_config
->pipe_src_w
;
1741 pipe_h
= pipe_config
->pipe_src_h
;
1743 pfit_w
= (pfit_size
>> 16) & 0xFFFF;
1744 pfit_h
= pfit_size
& 0xFFFF;
1745 if (pipe_w
< pfit_w
)
1747 if (pipe_h
< pfit_h
)
1750 pixel_rate
= div_u64((uint64_t) pixel_rate
* pipe_w
* pipe_h
,
1757 /* latency must be in 0.1us units. */
1758 static uint32_t ilk_wm_method1(uint32_t pixel_rate
, uint8_t bytes_per_pixel
,
1763 if (WARN(latency
== 0, "Latency value missing\n"))
1766 ret
= (uint64_t) pixel_rate
* bytes_per_pixel
* latency
;
1767 ret
= DIV_ROUND_UP_ULL(ret
, 64 * 10000) + 2;
1772 /* latency must be in 0.1us units. */
1773 static uint32_t ilk_wm_method2(uint32_t pixel_rate
, uint32_t pipe_htotal
,
1774 uint32_t horiz_pixels
, uint8_t bytes_per_pixel
,
1779 if (WARN(latency
== 0, "Latency value missing\n"))
1782 ret
= (latency
* pixel_rate
) / (pipe_htotal
* 10000);
1783 ret
= (ret
+ 1) * horiz_pixels
* bytes_per_pixel
;
1784 ret
= DIV_ROUND_UP(ret
, 64) + 2;
1788 static uint32_t ilk_wm_fbc(uint32_t pri_val
, uint32_t horiz_pixels
,
1789 uint8_t bytes_per_pixel
)
1791 return DIV_ROUND_UP(pri_val
* 64, horiz_pixels
* bytes_per_pixel
) + 2;
1794 struct skl_pipe_wm_parameters
{
1796 uint32_t pipe_htotal
;
1797 uint32_t pixel_rate
; /* in KHz */
1798 struct intel_plane_wm_parameters plane
[I915_MAX_PLANES
];
1799 struct intel_plane_wm_parameters cursor
;
1802 struct ilk_pipe_wm_parameters
{
1804 uint32_t pipe_htotal
;
1805 uint32_t pixel_rate
;
1806 struct intel_plane_wm_parameters pri
;
1807 struct intel_plane_wm_parameters spr
;
1808 struct intel_plane_wm_parameters cur
;
1811 struct ilk_wm_maximums
{
1818 /* used in computing the new watermarks state */
1819 struct intel_wm_config
{
1820 unsigned int num_pipes_active
;
1821 bool sprites_enabled
;
1822 bool sprites_scaled
;
1826 * For both WM_PIPE and WM_LP.
1827 * mem_value must be in 0.1us units.
1829 static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters
*params
,
1833 uint32_t method1
, method2
;
1835 if (!params
->active
|| !params
->pri
.enabled
)
1838 method1
= ilk_wm_method1(params
->pixel_rate
,
1839 params
->pri
.bytes_per_pixel
,
1845 method2
= ilk_wm_method2(params
->pixel_rate
,
1846 params
->pipe_htotal
,
1847 params
->pri
.horiz_pixels
,
1848 params
->pri
.bytes_per_pixel
,
1851 return min(method1
, method2
);
1855 * For both WM_PIPE and WM_LP.
1856 * mem_value must be in 0.1us units.
1858 static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters
*params
,
1861 uint32_t method1
, method2
;
1863 if (!params
->active
|| !params
->spr
.enabled
)
1866 method1
= ilk_wm_method1(params
->pixel_rate
,
1867 params
->spr
.bytes_per_pixel
,
1869 method2
= ilk_wm_method2(params
->pixel_rate
,
1870 params
->pipe_htotal
,
1871 params
->spr
.horiz_pixels
,
1872 params
->spr
.bytes_per_pixel
,
1874 return min(method1
, method2
);
1878 * For both WM_PIPE and WM_LP.
1879 * mem_value must be in 0.1us units.
1881 static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters
*params
,
1884 if (!params
->active
|| !params
->cur
.enabled
)
1887 return ilk_wm_method2(params
->pixel_rate
,
1888 params
->pipe_htotal
,
1889 params
->cur
.horiz_pixels
,
1890 params
->cur
.bytes_per_pixel
,
1894 /* Only for WM_LP. */
1895 static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters
*params
,
1898 if (!params
->active
|| !params
->pri
.enabled
)
1901 return ilk_wm_fbc(pri_val
,
1902 params
->pri
.horiz_pixels
,
1903 params
->pri
.bytes_per_pixel
);
1906 static unsigned int ilk_display_fifo_size(const struct drm_device
*dev
)
1908 if (INTEL_INFO(dev
)->gen
>= 8)
1910 else if (INTEL_INFO(dev
)->gen
>= 7)
1916 static unsigned int ilk_plane_wm_reg_max(const struct drm_device
*dev
,
1917 int level
, bool is_sprite
)
1919 if (INTEL_INFO(dev
)->gen
>= 8)
1920 /* BDW primary/sprite plane watermarks */
1921 return level
== 0 ? 255 : 2047;
1922 else if (INTEL_INFO(dev
)->gen
>= 7)
1923 /* IVB/HSW primary/sprite plane watermarks */
1924 return level
== 0 ? 127 : 1023;
1925 else if (!is_sprite
)
1926 /* ILK/SNB primary plane watermarks */
1927 return level
== 0 ? 127 : 511;
1929 /* ILK/SNB sprite plane watermarks */
1930 return level
== 0 ? 63 : 255;
1933 static unsigned int ilk_cursor_wm_reg_max(const struct drm_device
*dev
,
1936 if (INTEL_INFO(dev
)->gen
>= 7)
1937 return level
== 0 ? 63 : 255;
1939 return level
== 0 ? 31 : 63;
1942 static unsigned int ilk_fbc_wm_reg_max(const struct drm_device
*dev
)
1944 if (INTEL_INFO(dev
)->gen
>= 8)
1950 /* Calculate the maximum primary/sprite plane watermark */
1951 static unsigned int ilk_plane_wm_max(const struct drm_device
*dev
,
1953 const struct intel_wm_config
*config
,
1954 enum intel_ddb_partitioning ddb_partitioning
,
1957 unsigned int fifo_size
= ilk_display_fifo_size(dev
);
1959 /* if sprites aren't enabled, sprites get nothing */
1960 if (is_sprite
&& !config
->sprites_enabled
)
1963 /* HSW allows LP1+ watermarks even with multiple pipes */
1964 if (level
== 0 || config
->num_pipes_active
> 1) {
1965 fifo_size
/= INTEL_INFO(dev
)->num_pipes
;
1968 * For some reason the non self refresh
1969 * FIFO size is only half of the self
1970 * refresh FIFO size on ILK/SNB.
1972 if (INTEL_INFO(dev
)->gen
<= 6)
1976 if (config
->sprites_enabled
) {
1977 /* level 0 is always calculated with 1:1 split */
1978 if (level
> 0 && ddb_partitioning
== INTEL_DDB_PART_5_6
) {
1987 /* clamp to max that the registers can hold */
1988 return min(fifo_size
, ilk_plane_wm_reg_max(dev
, level
, is_sprite
));
1991 /* Calculate the maximum cursor plane watermark */
1992 static unsigned int ilk_cursor_wm_max(const struct drm_device
*dev
,
1994 const struct intel_wm_config
*config
)
1996 /* HSW LP1+ watermarks w/ multiple pipes */
1997 if (level
> 0 && config
->num_pipes_active
> 1)
2000 /* otherwise just report max that registers can hold */
2001 return ilk_cursor_wm_reg_max(dev
, level
);
2004 static void ilk_compute_wm_maximums(const struct drm_device
*dev
,
2006 const struct intel_wm_config
*config
,
2007 enum intel_ddb_partitioning ddb_partitioning
,
2008 struct ilk_wm_maximums
*max
)
2010 max
->pri
= ilk_plane_wm_max(dev
, level
, config
, ddb_partitioning
, false);
2011 max
->spr
= ilk_plane_wm_max(dev
, level
, config
, ddb_partitioning
, true);
2012 max
->cur
= ilk_cursor_wm_max(dev
, level
, config
);
2013 max
->fbc
= ilk_fbc_wm_reg_max(dev
);
2016 static void ilk_compute_wm_reg_maximums(struct drm_device
*dev
,
2018 struct ilk_wm_maximums
*max
)
2020 max
->pri
= ilk_plane_wm_reg_max(dev
, level
, false);
2021 max
->spr
= ilk_plane_wm_reg_max(dev
, level
, true);
2022 max
->cur
= ilk_cursor_wm_reg_max(dev
, level
);
2023 max
->fbc
= ilk_fbc_wm_reg_max(dev
);
2026 static bool ilk_validate_wm_level(int level
,
2027 const struct ilk_wm_maximums
*max
,
2028 struct intel_wm_level
*result
)
2032 /* already determined to be invalid? */
2033 if (!result
->enable
)
2036 result
->enable
= result
->pri_val
<= max
->pri
&&
2037 result
->spr_val
<= max
->spr
&&
2038 result
->cur_val
<= max
->cur
;
2040 ret
= result
->enable
;
2043 * HACK until we can pre-compute everything,
2044 * and thus fail gracefully if LP0 watermarks
2047 if (level
== 0 && !result
->enable
) {
2048 if (result
->pri_val
> max
->pri
)
2049 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
2050 level
, result
->pri_val
, max
->pri
);
2051 if (result
->spr_val
> max
->spr
)
2052 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
2053 level
, result
->spr_val
, max
->spr
);
2054 if (result
->cur_val
> max
->cur
)
2055 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
2056 level
, result
->cur_val
, max
->cur
);
2058 result
->pri_val
= min_t(uint32_t, result
->pri_val
, max
->pri
);
2059 result
->spr_val
= min_t(uint32_t, result
->spr_val
, max
->spr
);
2060 result
->cur_val
= min_t(uint32_t, result
->cur_val
, max
->cur
);
2061 result
->enable
= true;
2067 static void ilk_compute_wm_level(const struct drm_i915_private
*dev_priv
,
2069 const struct ilk_pipe_wm_parameters
*p
,
2070 struct intel_wm_level
*result
)
2072 uint16_t pri_latency
= dev_priv
->wm
.pri_latency
[level
];
2073 uint16_t spr_latency
= dev_priv
->wm
.spr_latency
[level
];
2074 uint16_t cur_latency
= dev_priv
->wm
.cur_latency
[level
];
2076 /* WM1+ latency values stored in 0.5us units */
2083 result
->pri_val
= ilk_compute_pri_wm(p
, pri_latency
, level
);
2084 result
->spr_val
= ilk_compute_spr_wm(p
, spr_latency
);
2085 result
->cur_val
= ilk_compute_cur_wm(p
, cur_latency
);
2086 result
->fbc_val
= ilk_compute_fbc_wm(p
, result
->pri_val
);
2087 result
->enable
= true;
2091 hsw_compute_linetime_wm(struct drm_device
*dev
, struct drm_crtc
*crtc
)
2093 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2094 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2095 struct drm_display_mode
*mode
= &intel_crtc
->config
->base
.adjusted_mode
;
2096 u32 linetime
, ips_linetime
;
2098 if (!intel_crtc
->active
)
2101 /* The WM are computed with base on how long it takes to fill a single
2102 * row at the given clock rate, multiplied by 8.
2104 linetime
= DIV_ROUND_CLOSEST(mode
->crtc_htotal
* 1000 * 8,
2106 ips_linetime
= DIV_ROUND_CLOSEST(mode
->crtc_htotal
* 1000 * 8,
2107 dev_priv
->cdclk_freq
);
2109 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime
) |
2110 PIPE_WM_LINETIME_TIME(linetime
);
2113 static void intel_read_wm_latency(struct drm_device
*dev
, uint16_t wm
[8])
2115 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2120 int level
, max_level
= ilk_wm_max_level(dev
);
2122 /* read the first set of memory latencies[0:3] */
2123 val
= 0; /* data0 to be programmed to 0 for first set */
2124 mutex_lock(&dev_priv
->rps
.hw_lock
);
2125 ret
= sandybridge_pcode_read(dev_priv
,
2126 GEN9_PCODE_READ_MEM_LATENCY
,
2128 mutex_unlock(&dev_priv
->rps
.hw_lock
);
2131 DRM_ERROR("SKL Mailbox read error = %d\n", ret
);
2135 wm
[0] = val
& GEN9_MEM_LATENCY_LEVEL_MASK
;
2136 wm
[1] = (val
>> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT
) &
2137 GEN9_MEM_LATENCY_LEVEL_MASK
;
2138 wm
[2] = (val
>> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT
) &
2139 GEN9_MEM_LATENCY_LEVEL_MASK
;
2140 wm
[3] = (val
>> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT
) &
2141 GEN9_MEM_LATENCY_LEVEL_MASK
;
2143 /* read the second set of memory latencies[4:7] */
2144 val
= 1; /* data0 to be programmed to 1 for second set */
2145 mutex_lock(&dev_priv
->rps
.hw_lock
);
2146 ret
= sandybridge_pcode_read(dev_priv
,
2147 GEN9_PCODE_READ_MEM_LATENCY
,
2149 mutex_unlock(&dev_priv
->rps
.hw_lock
);
2151 DRM_ERROR("SKL Mailbox read error = %d\n", ret
);
2155 wm
[4] = val
& GEN9_MEM_LATENCY_LEVEL_MASK
;
2156 wm
[5] = (val
>> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT
) &
2157 GEN9_MEM_LATENCY_LEVEL_MASK
;
2158 wm
[6] = (val
>> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT
) &
2159 GEN9_MEM_LATENCY_LEVEL_MASK
;
2160 wm
[7] = (val
>> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT
) &
2161 GEN9_MEM_LATENCY_LEVEL_MASK
;
2164 * WaWmMemoryReadLatency:skl
2166 * punit doesn't take into account the read latency so we need
2167 * to add 2us to the various latency levels we retrieve from
2169 * - W0 is a bit special in that it's the only level that
2170 * can't be disabled if we want to have display working, so
2171 * we always add 2us there.
2172 * - For levels >=1, punit returns 0us latency when they are
2173 * disabled, so we respect that and don't add 2us then
2175 * Additionally, if a level n (n > 1) has a 0us latency, all
2176 * levels m (m >= n) need to be disabled. We make sure to
2177 * sanitize the values out of the punit to satisfy this
2181 for (level
= 1; level
<= max_level
; level
++)
2185 for (i
= level
+ 1; i
<= max_level
; i
++)
2190 } else if (IS_HASWELL(dev
) || IS_BROADWELL(dev
)) {
2191 uint64_t sskpd
= I915_READ64(MCH_SSKPD
);
2193 wm
[0] = (sskpd
>> 56) & 0xFF;
2195 wm
[0] = sskpd
& 0xF;
2196 wm
[1] = (sskpd
>> 4) & 0xFF;
2197 wm
[2] = (sskpd
>> 12) & 0xFF;
2198 wm
[3] = (sskpd
>> 20) & 0x1FF;
2199 wm
[4] = (sskpd
>> 32) & 0x1FF;
2200 } else if (INTEL_INFO(dev
)->gen
>= 6) {
2201 uint32_t sskpd
= I915_READ(MCH_SSKPD
);
2203 wm
[0] = (sskpd
>> SSKPD_WM0_SHIFT
) & SSKPD_WM_MASK
;
2204 wm
[1] = (sskpd
>> SSKPD_WM1_SHIFT
) & SSKPD_WM_MASK
;
2205 wm
[2] = (sskpd
>> SSKPD_WM2_SHIFT
) & SSKPD_WM_MASK
;
2206 wm
[3] = (sskpd
>> SSKPD_WM3_SHIFT
) & SSKPD_WM_MASK
;
2207 } else if (INTEL_INFO(dev
)->gen
>= 5) {
2208 uint32_t mltr
= I915_READ(MLTR_ILK
);
2210 /* ILK primary LP0 latency is 700 ns */
2212 wm
[1] = (mltr
>> MLTR_WM1_SHIFT
) & ILK_SRLT_MASK
;
2213 wm
[2] = (mltr
>> MLTR_WM2_SHIFT
) & ILK_SRLT_MASK
;
2217 static void intel_fixup_spr_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2219 /* ILK sprite LP0 latency is 1300 ns */
2220 if (INTEL_INFO(dev
)->gen
== 5)
2224 static void intel_fixup_cur_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2226 /* ILK cursor LP0 latency is 1300 ns */
2227 if (INTEL_INFO(dev
)->gen
== 5)
2230 /* WaDoubleCursorLP3Latency:ivb */
2231 if (IS_IVYBRIDGE(dev
))
2235 int ilk_wm_max_level(const struct drm_device
*dev
)
2237 /* how many WM levels are we expecting */
2238 if (INTEL_INFO(dev
)->gen
>= 9)
2240 else if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2242 else if (INTEL_INFO(dev
)->gen
>= 6)
2248 static void intel_print_wm_latency(struct drm_device
*dev
,
2250 const uint16_t wm
[8])
2252 int level
, max_level
= ilk_wm_max_level(dev
);
2254 for (level
= 0; level
<= max_level
; level
++) {
2255 unsigned int latency
= wm
[level
];
2258 DRM_ERROR("%s WM%d latency not provided\n",
2264 * - latencies are in us on gen9.
2265 * - before then, WM1+ latency values are in 0.5us units
2272 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2273 name
, level
, wm
[level
],
2274 latency
/ 10, latency
% 10);
2278 static bool ilk_increase_wm_latency(struct drm_i915_private
*dev_priv
,
2279 uint16_t wm
[5], uint16_t min
)
2281 int level
, max_level
= ilk_wm_max_level(dev_priv
->dev
);
2286 wm
[0] = max(wm
[0], min
);
2287 for (level
= 1; level
<= max_level
; level
++)
2288 wm
[level
] = max_t(uint16_t, wm
[level
], DIV_ROUND_UP(min
, 5));
2293 static void snb_wm_latency_quirk(struct drm_device
*dev
)
2295 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2299 * The BIOS provided WM memory latency values are often
2300 * inadequate for high resolution displays. Adjust them.
2302 changed
= ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.pri_latency
, 12) |
2303 ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.spr_latency
, 12) |
2304 ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.cur_latency
, 12);
2309 DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
2310 intel_print_wm_latency(dev
, "Primary", dev_priv
->wm
.pri_latency
);
2311 intel_print_wm_latency(dev
, "Sprite", dev_priv
->wm
.spr_latency
);
2312 intel_print_wm_latency(dev
, "Cursor", dev_priv
->wm
.cur_latency
);
2315 static void ilk_setup_wm_latency(struct drm_device
*dev
)
2317 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2319 intel_read_wm_latency(dev
, dev_priv
->wm
.pri_latency
);
2321 memcpy(dev_priv
->wm
.spr_latency
, dev_priv
->wm
.pri_latency
,
2322 sizeof(dev_priv
->wm
.pri_latency
));
2323 memcpy(dev_priv
->wm
.cur_latency
, dev_priv
->wm
.pri_latency
,
2324 sizeof(dev_priv
->wm
.pri_latency
));
2326 intel_fixup_spr_wm_latency(dev
, dev_priv
->wm
.spr_latency
);
2327 intel_fixup_cur_wm_latency(dev
, dev_priv
->wm
.cur_latency
);
2329 intel_print_wm_latency(dev
, "Primary", dev_priv
->wm
.pri_latency
);
2330 intel_print_wm_latency(dev
, "Sprite", dev_priv
->wm
.spr_latency
);
2331 intel_print_wm_latency(dev
, "Cursor", dev_priv
->wm
.cur_latency
);
2334 snb_wm_latency_quirk(dev
);
2337 static void skl_setup_wm_latency(struct drm_device
*dev
)
2339 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2341 intel_read_wm_latency(dev
, dev_priv
->wm
.skl_latency
);
2342 intel_print_wm_latency(dev
, "Gen9 Plane", dev_priv
->wm
.skl_latency
);
2345 static void ilk_compute_wm_parameters(struct drm_crtc
*crtc
,
2346 struct ilk_pipe_wm_parameters
*p
)
2348 struct drm_device
*dev
= crtc
->dev
;
2349 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2350 enum pipe pipe
= intel_crtc
->pipe
;
2351 struct drm_plane
*plane
;
2353 if (!intel_crtc
->active
)
2357 p
->pipe_htotal
= intel_crtc
->config
->base
.adjusted_mode
.crtc_htotal
;
2358 p
->pixel_rate
= ilk_pipe_pixel_rate(intel_crtc
->config
);
2360 if (crtc
->primary
->state
->fb
)
2361 p
->pri
.bytes_per_pixel
=
2362 crtc
->primary
->state
->fb
->bits_per_pixel
/ 8;
2364 p
->pri
.bytes_per_pixel
= 4;
2366 p
->cur
.bytes_per_pixel
= 4;
2368 * TODO: for now, assume primary and cursor planes are always enabled.
2369 * Setting them to false makes the screen flicker.
2371 p
->pri
.enabled
= true;
2372 p
->cur
.enabled
= true;
2374 p
->pri
.horiz_pixels
= intel_crtc
->config
->pipe_src_w
;
2375 p
->cur
.horiz_pixels
= intel_crtc
->base
.cursor
->state
->crtc_w
;
2377 drm_for_each_legacy_plane(plane
, dev
) {
2378 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
2380 if (intel_plane
->pipe
== pipe
) {
2381 p
->spr
= intel_plane
->wm
;
2387 static void ilk_compute_wm_config(struct drm_device
*dev
,
2388 struct intel_wm_config
*config
)
2390 struct intel_crtc
*intel_crtc
;
2392 /* Compute the currently _active_ config */
2393 for_each_intel_crtc(dev
, intel_crtc
) {
2394 const struct intel_pipe_wm
*wm
= &intel_crtc
->wm
.active
;
2396 if (!wm
->pipe_enabled
)
2399 config
->sprites_enabled
|= wm
->sprites_enabled
;
2400 config
->sprites_scaled
|= wm
->sprites_scaled
;
2401 config
->num_pipes_active
++;
2405 /* Compute new watermarks for the pipe */
2406 static bool intel_compute_pipe_wm(struct drm_crtc
*crtc
,
2407 const struct ilk_pipe_wm_parameters
*params
,
2408 struct intel_pipe_wm
*pipe_wm
)
2410 struct drm_device
*dev
= crtc
->dev
;
2411 const struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2412 int level
, max_level
= ilk_wm_max_level(dev
);
2413 /* LP0 watermark maximums depend on this pipe alone */
2414 struct intel_wm_config config
= {
2415 .num_pipes_active
= 1,
2416 .sprites_enabled
= params
->spr
.enabled
,
2417 .sprites_scaled
= params
->spr
.scaled
,
2419 struct ilk_wm_maximums max
;
2421 pipe_wm
->pipe_enabled
= params
->active
;
2422 pipe_wm
->sprites_enabled
= params
->spr
.enabled
;
2423 pipe_wm
->sprites_scaled
= params
->spr
.scaled
;
2425 /* ILK/SNB: LP2+ watermarks only w/o sprites */
2426 if (INTEL_INFO(dev
)->gen
<= 6 && params
->spr
.enabled
)
2429 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
2430 if (params
->spr
.scaled
)
2433 ilk_compute_wm_level(dev_priv
, 0, params
, &pipe_wm
->wm
[0]);
2435 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2436 pipe_wm
->linetime
= hsw_compute_linetime_wm(dev
, crtc
);
2438 /* LP0 watermarks always use 1/2 DDB partitioning */
2439 ilk_compute_wm_maximums(dev
, 0, &config
, INTEL_DDB_PART_1_2
, &max
);
2441 /* At least LP0 must be valid */
2442 if (!ilk_validate_wm_level(0, &max
, &pipe_wm
->wm
[0]))
2445 ilk_compute_wm_reg_maximums(dev
, 1, &max
);
2447 for (level
= 1; level
<= max_level
; level
++) {
2448 struct intel_wm_level wm
= {};
2450 ilk_compute_wm_level(dev_priv
, level
, params
, &wm
);
2453 * Disable any watermark level that exceeds the
2454 * register maximums since such watermarks are
2457 if (!ilk_validate_wm_level(level
, &max
, &wm
))
2460 pipe_wm
->wm
[level
] = wm
;
2467 * Merge the watermarks from all active pipes for a specific level.
2469 static void ilk_merge_wm_level(struct drm_device
*dev
,
2471 struct intel_wm_level
*ret_wm
)
2473 const struct intel_crtc
*intel_crtc
;
2475 ret_wm
->enable
= true;
2477 for_each_intel_crtc(dev
, intel_crtc
) {
2478 const struct intel_pipe_wm
*active
= &intel_crtc
->wm
.active
;
2479 const struct intel_wm_level
*wm
= &active
->wm
[level
];
2481 if (!active
->pipe_enabled
)
2485 * The watermark values may have been used in the past,
2486 * so we must maintain them in the registers for some
2487 * time even if the level is now disabled.
2490 ret_wm
->enable
= false;
2492 ret_wm
->pri_val
= max(ret_wm
->pri_val
, wm
->pri_val
);
2493 ret_wm
->spr_val
= max(ret_wm
->spr_val
, wm
->spr_val
);
2494 ret_wm
->cur_val
= max(ret_wm
->cur_val
, wm
->cur_val
);
2495 ret_wm
->fbc_val
= max(ret_wm
->fbc_val
, wm
->fbc_val
);
2500 * Merge all low power watermarks for all active pipes.
2502 static void ilk_wm_merge(struct drm_device
*dev
,
2503 const struct intel_wm_config
*config
,
2504 const struct ilk_wm_maximums
*max
,
2505 struct intel_pipe_wm
*merged
)
2507 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2508 int level
, max_level
= ilk_wm_max_level(dev
);
2509 int last_enabled_level
= max_level
;
2511 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2512 if ((INTEL_INFO(dev
)->gen
<= 6 || IS_IVYBRIDGE(dev
)) &&
2513 config
->num_pipes_active
> 1)
2516 /* ILK: FBC WM must be disabled always */
2517 merged
->fbc_wm_enabled
= INTEL_INFO(dev
)->gen
>= 6;
2519 /* merge each WM1+ level */
2520 for (level
= 1; level
<= max_level
; level
++) {
2521 struct intel_wm_level
*wm
= &merged
->wm
[level
];
2523 ilk_merge_wm_level(dev
, level
, wm
);
2525 if (level
> last_enabled_level
)
2527 else if (!ilk_validate_wm_level(level
, max
, wm
))
2528 /* make sure all following levels get disabled */
2529 last_enabled_level
= level
- 1;
2532 * The spec says it is preferred to disable
2533 * FBC WMs instead of disabling a WM level.
2535 if (wm
->fbc_val
> max
->fbc
) {
2537 merged
->fbc_wm_enabled
= false;
2542 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2544 * FIXME this is racy. FBC might get enabled later.
2545 * What we should check here is whether FBC can be
2546 * enabled sometime later.
2548 if (IS_GEN5(dev
) && !merged
->fbc_wm_enabled
&&
2549 intel_fbc_enabled(dev_priv
)) {
2550 for (level
= 2; level
<= max_level
; level
++) {
2551 struct intel_wm_level
*wm
= &merged
->wm
[level
];
2558 static int ilk_wm_lp_to_level(int wm_lp
, const struct intel_pipe_wm
*pipe_wm
)
2560 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2561 return wm_lp
+ (wm_lp
>= 2 && pipe_wm
->wm
[4].enable
);
2564 /* The value we need to program into the WM_LPx latency field */
2565 static unsigned int ilk_wm_lp_latency(struct drm_device
*dev
, int level
)
2567 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2569 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2572 return dev_priv
->wm
.pri_latency
[level
];
2575 static void ilk_compute_wm_results(struct drm_device
*dev
,
2576 const struct intel_pipe_wm
*merged
,
2577 enum intel_ddb_partitioning partitioning
,
2578 struct ilk_wm_values
*results
)
2580 struct intel_crtc
*intel_crtc
;
2583 results
->enable_fbc_wm
= merged
->fbc_wm_enabled
;
2584 results
->partitioning
= partitioning
;
2586 /* LP1+ register values */
2587 for (wm_lp
= 1; wm_lp
<= 3; wm_lp
++) {
2588 const struct intel_wm_level
*r
;
2590 level
= ilk_wm_lp_to_level(wm_lp
, merged
);
2592 r
= &merged
->wm
[level
];
2595 * Maintain the watermark values even if the level is
2596 * disabled. Doing otherwise could cause underruns.
2598 results
->wm_lp
[wm_lp
- 1] =
2599 (ilk_wm_lp_latency(dev
, level
) << WM1_LP_LATENCY_SHIFT
) |
2600 (r
->pri_val
<< WM1_LP_SR_SHIFT
) |
2604 results
->wm_lp
[wm_lp
- 1] |= WM1_LP_SR_EN
;
2606 if (INTEL_INFO(dev
)->gen
>= 8)
2607 results
->wm_lp
[wm_lp
- 1] |=
2608 r
->fbc_val
<< WM1_LP_FBC_SHIFT_BDW
;
2610 results
->wm_lp
[wm_lp
- 1] |=
2611 r
->fbc_val
<< WM1_LP_FBC_SHIFT
;
2614 * Always set WM1S_LP_EN when spr_val != 0, even if the
2615 * level is disabled. Doing otherwise could cause underruns.
2617 if (INTEL_INFO(dev
)->gen
<= 6 && r
->spr_val
) {
2618 WARN_ON(wm_lp
!= 1);
2619 results
->wm_lp_spr
[wm_lp
- 1] = WM1S_LP_EN
| r
->spr_val
;
2621 results
->wm_lp_spr
[wm_lp
- 1] = r
->spr_val
;
2624 /* LP0 register values */
2625 for_each_intel_crtc(dev
, intel_crtc
) {
2626 enum pipe pipe
= intel_crtc
->pipe
;
2627 const struct intel_wm_level
*r
=
2628 &intel_crtc
->wm
.active
.wm
[0];
2630 if (WARN_ON(!r
->enable
))
2633 results
->wm_linetime
[pipe
] = intel_crtc
->wm
.active
.linetime
;
2635 results
->wm_pipe
[pipe
] =
2636 (r
->pri_val
<< WM0_PIPE_PLANE_SHIFT
) |
2637 (r
->spr_val
<< WM0_PIPE_SPRITE_SHIFT
) |
2642 /* Find the result with the highest level enabled. Check for enable_fbc_wm in
2643 * case both are at the same level. Prefer r1 in case they're the same. */
2644 static struct intel_pipe_wm
*ilk_find_best_result(struct drm_device
*dev
,
2645 struct intel_pipe_wm
*r1
,
2646 struct intel_pipe_wm
*r2
)
2648 int level
, max_level
= ilk_wm_max_level(dev
);
2649 int level1
= 0, level2
= 0;
2651 for (level
= 1; level
<= max_level
; level
++) {
2652 if (r1
->wm
[level
].enable
)
2654 if (r2
->wm
[level
].enable
)
2658 if (level1
== level2
) {
2659 if (r2
->fbc_wm_enabled
&& !r1
->fbc_wm_enabled
)
2663 } else if (level1
> level2
) {
2670 /* dirty bits used to track which watermarks need changes */
2671 #define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2672 #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2673 #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2674 #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2675 #define WM_DIRTY_FBC (1 << 24)
2676 #define WM_DIRTY_DDB (1 << 25)
2678 static unsigned int ilk_compute_wm_dirty(struct drm_i915_private
*dev_priv
,
2679 const struct ilk_wm_values
*old
,
2680 const struct ilk_wm_values
*new)
2682 unsigned int dirty
= 0;
2686 for_each_pipe(dev_priv
, pipe
) {
2687 if (old
->wm_linetime
[pipe
] != new->wm_linetime
[pipe
]) {
2688 dirty
|= WM_DIRTY_LINETIME(pipe
);
2689 /* Must disable LP1+ watermarks too */
2690 dirty
|= WM_DIRTY_LP_ALL
;
2693 if (old
->wm_pipe
[pipe
] != new->wm_pipe
[pipe
]) {
2694 dirty
|= WM_DIRTY_PIPE(pipe
);
2695 /* Must disable LP1+ watermarks too */
2696 dirty
|= WM_DIRTY_LP_ALL
;
2700 if (old
->enable_fbc_wm
!= new->enable_fbc_wm
) {
2701 dirty
|= WM_DIRTY_FBC
;
2702 /* Must disable LP1+ watermarks too */
2703 dirty
|= WM_DIRTY_LP_ALL
;
2706 if (old
->partitioning
!= new->partitioning
) {
2707 dirty
|= WM_DIRTY_DDB
;
2708 /* Must disable LP1+ watermarks too */
2709 dirty
|= WM_DIRTY_LP_ALL
;
2712 /* LP1+ watermarks already deemed dirty, no need to continue */
2713 if (dirty
& WM_DIRTY_LP_ALL
)
2716 /* Find the lowest numbered LP1+ watermark in need of an update... */
2717 for (wm_lp
= 1; wm_lp
<= 3; wm_lp
++) {
2718 if (old
->wm_lp
[wm_lp
- 1] != new->wm_lp
[wm_lp
- 1] ||
2719 old
->wm_lp_spr
[wm_lp
- 1] != new->wm_lp_spr
[wm_lp
- 1])
2723 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2724 for (; wm_lp
<= 3; wm_lp
++)
2725 dirty
|= WM_DIRTY_LP(wm_lp
);
2730 static bool _ilk_disable_lp_wm(struct drm_i915_private
*dev_priv
,
2733 struct ilk_wm_values
*previous
= &dev_priv
->wm
.hw
;
2734 bool changed
= false;
2736 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp
[2] & WM1_LP_SR_EN
) {
2737 previous
->wm_lp
[2] &= ~WM1_LP_SR_EN
;
2738 I915_WRITE(WM3_LP_ILK
, previous
->wm_lp
[2]);
2741 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp
[1] & WM1_LP_SR_EN
) {
2742 previous
->wm_lp
[1] &= ~WM1_LP_SR_EN
;
2743 I915_WRITE(WM2_LP_ILK
, previous
->wm_lp
[1]);
2746 if (dirty
& WM_DIRTY_LP(1) && previous
->wm_lp
[0] & WM1_LP_SR_EN
) {
2747 previous
->wm_lp
[0] &= ~WM1_LP_SR_EN
;
2748 I915_WRITE(WM1_LP_ILK
, previous
->wm_lp
[0]);
2753 * Don't touch WM1S_LP_EN here.
2754 * Doing so could cause underruns.
2761 * The spec says we shouldn't write when we don't need, because every write
2762 * causes WMs to be re-evaluated, expending some power.
2764 static void ilk_write_wm_values(struct drm_i915_private
*dev_priv
,
2765 struct ilk_wm_values
*results
)
2767 struct drm_device
*dev
= dev_priv
->dev
;
2768 struct ilk_wm_values
*previous
= &dev_priv
->wm
.hw
;
2772 dirty
= ilk_compute_wm_dirty(dev_priv
, previous
, results
);
2776 _ilk_disable_lp_wm(dev_priv
, dirty
);
2778 if (dirty
& WM_DIRTY_PIPE(PIPE_A
))
2779 I915_WRITE(WM0_PIPEA_ILK
, results
->wm_pipe
[0]);
2780 if (dirty
& WM_DIRTY_PIPE(PIPE_B
))
2781 I915_WRITE(WM0_PIPEB_ILK
, results
->wm_pipe
[1]);
2782 if (dirty
& WM_DIRTY_PIPE(PIPE_C
))
2783 I915_WRITE(WM0_PIPEC_IVB
, results
->wm_pipe
[2]);
2785 if (dirty
& WM_DIRTY_LINETIME(PIPE_A
))
2786 I915_WRITE(PIPE_WM_LINETIME(PIPE_A
), results
->wm_linetime
[0]);
2787 if (dirty
& WM_DIRTY_LINETIME(PIPE_B
))
2788 I915_WRITE(PIPE_WM_LINETIME(PIPE_B
), results
->wm_linetime
[1]);
2789 if (dirty
& WM_DIRTY_LINETIME(PIPE_C
))
2790 I915_WRITE(PIPE_WM_LINETIME(PIPE_C
), results
->wm_linetime
[2]);
2792 if (dirty
& WM_DIRTY_DDB
) {
2793 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
)) {
2794 val
= I915_READ(WM_MISC
);
2795 if (results
->partitioning
== INTEL_DDB_PART_1_2
)
2796 val
&= ~WM_MISC_DATA_PARTITION_5_6
;
2798 val
|= WM_MISC_DATA_PARTITION_5_6
;
2799 I915_WRITE(WM_MISC
, val
);
2801 val
= I915_READ(DISP_ARB_CTL2
);
2802 if (results
->partitioning
== INTEL_DDB_PART_1_2
)
2803 val
&= ~DISP_DATA_PARTITION_5_6
;
2805 val
|= DISP_DATA_PARTITION_5_6
;
2806 I915_WRITE(DISP_ARB_CTL2
, val
);
2810 if (dirty
& WM_DIRTY_FBC
) {
2811 val
= I915_READ(DISP_ARB_CTL
);
2812 if (results
->enable_fbc_wm
)
2813 val
&= ~DISP_FBC_WM_DIS
;
2815 val
|= DISP_FBC_WM_DIS
;
2816 I915_WRITE(DISP_ARB_CTL
, val
);
2819 if (dirty
& WM_DIRTY_LP(1) &&
2820 previous
->wm_lp_spr
[0] != results
->wm_lp_spr
[0])
2821 I915_WRITE(WM1S_LP_ILK
, results
->wm_lp_spr
[0]);
2823 if (INTEL_INFO(dev
)->gen
>= 7) {
2824 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp_spr
[1] != results
->wm_lp_spr
[1])
2825 I915_WRITE(WM2S_LP_IVB
, results
->wm_lp_spr
[1]);
2826 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp_spr
[2] != results
->wm_lp_spr
[2])
2827 I915_WRITE(WM3S_LP_IVB
, results
->wm_lp_spr
[2]);
2830 if (dirty
& WM_DIRTY_LP(1) && previous
->wm_lp
[0] != results
->wm_lp
[0])
2831 I915_WRITE(WM1_LP_ILK
, results
->wm_lp
[0]);
2832 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp
[1] != results
->wm_lp
[1])
2833 I915_WRITE(WM2_LP_ILK
, results
->wm_lp
[1]);
2834 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp
[2] != results
->wm_lp
[2])
2835 I915_WRITE(WM3_LP_ILK
, results
->wm_lp
[2]);
2837 dev_priv
->wm
.hw
= *results
;
2840 static bool ilk_disable_lp_wm(struct drm_device
*dev
)
2842 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2844 return _ilk_disable_lp_wm(dev_priv
, WM_DIRTY_LP_ALL
);
2848 * On gen9, we need to allocate Display Data Buffer (DDB) portions to the
2849 * different active planes.
2852 #define SKL_DDB_SIZE 896 /* in blocks */
2853 #define BXT_DDB_SIZE 512
2856 skl_ddb_get_pipe_allocation_limits(struct drm_device
*dev
,
2857 struct drm_crtc
*for_crtc
,
2858 const struct intel_wm_config
*config
,
2859 const struct skl_pipe_wm_parameters
*params
,
2860 struct skl_ddb_entry
*alloc
/* out */)
2862 struct drm_crtc
*crtc
;
2863 unsigned int pipe_size
, ddb_size
;
2864 int nth_active_pipe
;
2866 if (!params
->active
) {
2872 if (IS_BROXTON(dev
))
2873 ddb_size
= BXT_DDB_SIZE
;
2875 ddb_size
= SKL_DDB_SIZE
;
2877 ddb_size
-= 4; /* 4 blocks for bypass path allocation */
2879 nth_active_pipe
= 0;
2880 for_each_crtc(dev
, crtc
) {
2881 if (!to_intel_crtc(crtc
)->active
)
2884 if (crtc
== for_crtc
)
2890 pipe_size
= ddb_size
/ config
->num_pipes_active
;
2891 alloc
->start
= nth_active_pipe
* ddb_size
/ config
->num_pipes_active
;
2892 alloc
->end
= alloc
->start
+ pipe_size
;
2895 static unsigned int skl_cursor_allocation(const struct intel_wm_config
*config
)
2897 if (config
->num_pipes_active
== 1)
2903 static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry
*entry
, u32 reg
)
2905 entry
->start
= reg
& 0x3ff;
2906 entry
->end
= (reg
>> 16) & 0x3ff;
2911 void skl_ddb_get_hw_state(struct drm_i915_private
*dev_priv
,
2912 struct skl_ddb_allocation
*ddb
/* out */)
2918 for_each_pipe(dev_priv
, pipe
) {
2919 for_each_plane(dev_priv
, pipe
, plane
) {
2920 val
= I915_READ(PLANE_BUF_CFG(pipe
, plane
));
2921 skl_ddb_entry_init_from_hw(&ddb
->plane
[pipe
][plane
],
2925 val
= I915_READ(CUR_BUF_CFG(pipe
));
2926 skl_ddb_entry_init_from_hw(&ddb
->cursor
[pipe
], val
);
2931 skl_plane_relative_data_rate(const struct intel_plane_wm_parameters
*p
, int y
)
2934 /* for planar format */
2935 if (p
->y_bytes_per_pixel
) {
2936 if (y
) /* y-plane data rate */
2937 return p
->horiz_pixels
* p
->vert_pixels
* p
->y_bytes_per_pixel
;
2938 else /* uv-plane data rate */
2939 return (p
->horiz_pixels
/2) * (p
->vert_pixels
/2) * p
->bytes_per_pixel
;
2942 /* for packed formats */
2943 return p
->horiz_pixels
* p
->vert_pixels
* p
->bytes_per_pixel
;
2947 * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching
2948 * a 8192x4096@32bpp framebuffer:
2949 * 3 * 4096 * 8192 * 4 < 2^32
2952 skl_get_total_relative_data_rate(struct intel_crtc
*intel_crtc
,
2953 const struct skl_pipe_wm_parameters
*params
)
2955 unsigned int total_data_rate
= 0;
2958 for (plane
= 0; plane
< intel_num_planes(intel_crtc
); plane
++) {
2959 const struct intel_plane_wm_parameters
*p
;
2961 p
= ¶ms
->plane
[plane
];
2965 total_data_rate
+= skl_plane_relative_data_rate(p
, 0); /* packed/uv */
2966 if (p
->y_bytes_per_pixel
) {
2967 total_data_rate
+= skl_plane_relative_data_rate(p
, 1); /* y-plane */
2971 return total_data_rate
;
2975 skl_allocate_pipe_ddb(struct drm_crtc
*crtc
,
2976 const struct intel_wm_config
*config
,
2977 const struct skl_pipe_wm_parameters
*params
,
2978 struct skl_ddb_allocation
*ddb
/* out */)
2980 struct drm_device
*dev
= crtc
->dev
;
2981 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2982 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2983 enum pipe pipe
= intel_crtc
->pipe
;
2984 struct skl_ddb_entry
*alloc
= &ddb
->pipe
[pipe
];
2985 uint16_t alloc_size
, start
, cursor_blocks
;
2986 uint16_t minimum
[I915_MAX_PLANES
];
2987 uint16_t y_minimum
[I915_MAX_PLANES
];
2988 unsigned int total_data_rate
;
2991 skl_ddb_get_pipe_allocation_limits(dev
, crtc
, config
, params
, alloc
);
2992 alloc_size
= skl_ddb_entry_size(alloc
);
2993 if (alloc_size
== 0) {
2994 memset(ddb
->plane
[pipe
], 0, sizeof(ddb
->plane
[pipe
]));
2995 memset(&ddb
->cursor
[pipe
], 0, sizeof(ddb
->cursor
[pipe
]));
2999 cursor_blocks
= skl_cursor_allocation(config
);
3000 ddb
->cursor
[pipe
].start
= alloc
->end
- cursor_blocks
;
3001 ddb
->cursor
[pipe
].end
= alloc
->end
;
3003 alloc_size
-= cursor_blocks
;
3004 alloc
->end
-= cursor_blocks
;
3006 /* 1. Allocate the mininum required blocks for each active plane */
3007 for_each_plane(dev_priv
, pipe
, plane
) {
3008 const struct intel_plane_wm_parameters
*p
;
3010 p
= ¶ms
->plane
[plane
];
3015 alloc_size
-= minimum
[plane
];
3016 y_minimum
[plane
] = p
->y_bytes_per_pixel
? 8 : 0;
3017 alloc_size
-= y_minimum
[plane
];
3021 * 2. Distribute the remaining space in proportion to the amount of
3022 * data each plane needs to fetch from memory.
3024 * FIXME: we may not allocate every single block here.
3026 total_data_rate
= skl_get_total_relative_data_rate(intel_crtc
, params
);
3028 start
= alloc
->start
;
3029 for (plane
= 0; plane
< intel_num_planes(intel_crtc
); plane
++) {
3030 const struct intel_plane_wm_parameters
*p
;
3031 unsigned int data_rate
, y_data_rate
;
3032 uint16_t plane_blocks
, y_plane_blocks
= 0;
3034 p
= ¶ms
->plane
[plane
];
3038 data_rate
= skl_plane_relative_data_rate(p
, 0);
3041 * allocation for (packed formats) or (uv-plane part of planar format):
3042 * promote the expression to 64 bits to avoid overflowing, the
3043 * result is < available as data_rate / total_data_rate < 1
3045 plane_blocks
= minimum
[plane
];
3046 plane_blocks
+= div_u64((uint64_t)alloc_size
* data_rate
,
3049 ddb
->plane
[pipe
][plane
].start
= start
;
3050 ddb
->plane
[pipe
][plane
].end
= start
+ plane_blocks
;
3052 start
+= plane_blocks
;
3055 * allocation for y_plane part of planar format:
3057 if (p
->y_bytes_per_pixel
) {
3058 y_data_rate
= skl_plane_relative_data_rate(p
, 1);
3059 y_plane_blocks
= y_minimum
[plane
];
3060 y_plane_blocks
+= div_u64((uint64_t)alloc_size
* y_data_rate
,
3063 ddb
->y_plane
[pipe
][plane
].start
= start
;
3064 ddb
->y_plane
[pipe
][plane
].end
= start
+ y_plane_blocks
;
3066 start
+= y_plane_blocks
;
3073 static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state
*config
)
3075 /* TODO: Take into account the scalers once we support them */
3076 return config
->base
.adjusted_mode
.crtc_clock
;
3080 * The max latency should be 257 (max the punit can code is 255 and we add 2us
3081 * for the read latency) and bytes_per_pixel should always be <= 8, so that
3082 * should allow pixel_rate up to ~2 GHz which seems sufficient since max
3083 * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
3085 static uint32_t skl_wm_method1(uint32_t pixel_rate
, uint8_t bytes_per_pixel
,
3088 uint32_t wm_intermediate_val
, ret
;
3093 wm_intermediate_val
= latency
* pixel_rate
* bytes_per_pixel
/ 512;
3094 ret
= DIV_ROUND_UP(wm_intermediate_val
, 1000);
3099 static uint32_t skl_wm_method2(uint32_t pixel_rate
, uint32_t pipe_htotal
,
3100 uint32_t horiz_pixels
, uint8_t bytes_per_pixel
,
3101 uint64_t tiling
, uint32_t latency
)
3104 uint32_t plane_bytes_per_line
, plane_blocks_per_line
;
3105 uint32_t wm_intermediate_val
;
3110 plane_bytes_per_line
= horiz_pixels
* bytes_per_pixel
;
3112 if (tiling
== I915_FORMAT_MOD_Y_TILED
||
3113 tiling
== I915_FORMAT_MOD_Yf_TILED
) {
3114 plane_bytes_per_line
*= 4;
3115 plane_blocks_per_line
= DIV_ROUND_UP(plane_bytes_per_line
, 512);
3116 plane_blocks_per_line
/= 4;
3118 plane_blocks_per_line
= DIV_ROUND_UP(plane_bytes_per_line
, 512);
3121 wm_intermediate_val
= latency
* pixel_rate
;
3122 ret
= DIV_ROUND_UP(wm_intermediate_val
, pipe_htotal
* 1000) *
3123 plane_blocks_per_line
;
3128 static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation
*new_ddb
,
3129 const struct intel_crtc
*intel_crtc
)
3131 struct drm_device
*dev
= intel_crtc
->base
.dev
;
3132 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3133 const struct skl_ddb_allocation
*cur_ddb
= &dev_priv
->wm
.skl_hw
.ddb
;
3134 enum pipe pipe
= intel_crtc
->pipe
;
3136 if (memcmp(new_ddb
->plane
[pipe
], cur_ddb
->plane
[pipe
],
3137 sizeof(new_ddb
->plane
[pipe
])))
3140 if (memcmp(&new_ddb
->cursor
[pipe
], &cur_ddb
->cursor
[pipe
],
3141 sizeof(new_ddb
->cursor
[pipe
])))
3147 static void skl_compute_wm_global_parameters(struct drm_device
*dev
,
3148 struct intel_wm_config
*config
)
3150 struct drm_crtc
*crtc
;
3151 struct drm_plane
*plane
;
3153 list_for_each_entry(crtc
, &dev
->mode_config
.crtc_list
, head
)
3154 config
->num_pipes_active
+= to_intel_crtc(crtc
)->active
;
3156 /* FIXME: I don't think we need those two global parameters on SKL */
3157 list_for_each_entry(plane
, &dev
->mode_config
.plane_list
, head
) {
3158 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
3160 config
->sprites_enabled
|= intel_plane
->wm
.enabled
;
3161 config
->sprites_scaled
|= intel_plane
->wm
.scaled
;
3165 static void skl_compute_wm_pipe_parameters(struct drm_crtc
*crtc
,
3166 struct skl_pipe_wm_parameters
*p
)
3168 struct drm_device
*dev
= crtc
->dev
;
3169 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3170 enum pipe pipe
= intel_crtc
->pipe
;
3171 struct drm_plane
*plane
;
3172 struct drm_framebuffer
*fb
;
3173 int i
= 1; /* Index for sprite planes start */
3175 p
->active
= intel_crtc
->active
;
3177 p
->pipe_htotal
= intel_crtc
->config
->base
.adjusted_mode
.crtc_htotal
;
3178 p
->pixel_rate
= skl_pipe_pixel_rate(intel_crtc
->config
);
3180 fb
= crtc
->primary
->state
->fb
;
3181 /* For planar: Bpp is for uv plane, y_Bpp is for y plane */
3183 p
->plane
[0].enabled
= true;
3184 p
->plane
[0].bytes_per_pixel
= fb
->pixel_format
== DRM_FORMAT_NV12
?
3185 drm_format_plane_cpp(fb
->pixel_format
, 1) :
3186 drm_format_plane_cpp(fb
->pixel_format
, 0);
3187 p
->plane
[0].y_bytes_per_pixel
= fb
->pixel_format
== DRM_FORMAT_NV12
?
3188 drm_format_plane_cpp(fb
->pixel_format
, 0) : 0;
3189 p
->plane
[0].tiling
= fb
->modifier
[0];
3191 p
->plane
[0].enabled
= false;
3192 p
->plane
[0].bytes_per_pixel
= 0;
3193 p
->plane
[0].y_bytes_per_pixel
= 0;
3194 p
->plane
[0].tiling
= DRM_FORMAT_MOD_NONE
;
3196 p
->plane
[0].horiz_pixels
= intel_crtc
->config
->pipe_src_w
;
3197 p
->plane
[0].vert_pixels
= intel_crtc
->config
->pipe_src_h
;
3198 p
->plane
[0].rotation
= crtc
->primary
->state
->rotation
;
3200 fb
= crtc
->cursor
->state
->fb
;
3201 p
->cursor
.y_bytes_per_pixel
= 0;
3203 p
->cursor
.enabled
= true;
3204 p
->cursor
.bytes_per_pixel
= fb
->bits_per_pixel
/ 8;
3205 p
->cursor
.horiz_pixels
= crtc
->cursor
->state
->crtc_w
;
3206 p
->cursor
.vert_pixels
= crtc
->cursor
->state
->crtc_h
;
3208 p
->cursor
.enabled
= false;
3209 p
->cursor
.bytes_per_pixel
= 0;
3210 p
->cursor
.horiz_pixels
= 64;
3211 p
->cursor
.vert_pixels
= 64;
3215 list_for_each_entry(plane
, &dev
->mode_config
.plane_list
, head
) {
3216 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
3218 if (intel_plane
->pipe
== pipe
&&
3219 plane
->type
== DRM_PLANE_TYPE_OVERLAY
)
3220 p
->plane
[i
++] = intel_plane
->wm
;
3224 static bool skl_compute_plane_wm(const struct drm_i915_private
*dev_priv
,
3225 struct skl_pipe_wm_parameters
*p
,
3226 struct intel_plane_wm_parameters
*p_params
,
3227 uint16_t ddb_allocation
,
3229 uint16_t *out_blocks
, /* out */
3230 uint8_t *out_lines
/* out */)
3232 uint32_t latency
= dev_priv
->wm
.skl_latency
[level
];
3233 uint32_t method1
, method2
;
3234 uint32_t plane_bytes_per_line
, plane_blocks_per_line
;
3235 uint32_t res_blocks
, res_lines
;
3236 uint32_t selected_result
;
3237 uint8_t bytes_per_pixel
;
3239 if (latency
== 0 || !p
->active
|| !p_params
->enabled
)
3242 bytes_per_pixel
= p_params
->y_bytes_per_pixel
?
3243 p_params
->y_bytes_per_pixel
:
3244 p_params
->bytes_per_pixel
;
3245 method1
= skl_wm_method1(p
->pixel_rate
,
3248 method2
= skl_wm_method2(p
->pixel_rate
,
3250 p_params
->horiz_pixels
,
3255 plane_bytes_per_line
= p_params
->horiz_pixels
* bytes_per_pixel
;
3256 plane_blocks_per_line
= DIV_ROUND_UP(plane_bytes_per_line
, 512);
3258 if (p_params
->tiling
== I915_FORMAT_MOD_Y_TILED
||
3259 p_params
->tiling
== I915_FORMAT_MOD_Yf_TILED
) {
3260 uint32_t min_scanlines
= 4;
3261 uint32_t y_tile_minimum
;
3262 if (intel_rotation_90_or_270(p_params
->rotation
)) {
3263 switch (p_params
->bytes_per_pixel
) {
3271 WARN(1, "Unsupported pixel depth for rotation");
3274 y_tile_minimum
= plane_blocks_per_line
* min_scanlines
;
3275 selected_result
= max(method2
, y_tile_minimum
);
3277 if ((ddb_allocation
/ plane_blocks_per_line
) >= 1)
3278 selected_result
= min(method1
, method2
);
3280 selected_result
= method1
;
3283 res_blocks
= selected_result
+ 1;
3284 res_lines
= DIV_ROUND_UP(selected_result
, plane_blocks_per_line
);
3286 if (level
>= 1 && level
<= 7) {
3287 if (p_params
->tiling
== I915_FORMAT_MOD_Y_TILED
||
3288 p_params
->tiling
== I915_FORMAT_MOD_Yf_TILED
)
3294 if (res_blocks
>= ddb_allocation
|| res_lines
> 31)
3297 *out_blocks
= res_blocks
;
3298 *out_lines
= res_lines
;
3303 static void skl_compute_wm_level(const struct drm_i915_private
*dev_priv
,
3304 struct skl_ddb_allocation
*ddb
,
3305 struct skl_pipe_wm_parameters
*p
,
3309 struct skl_wm_level
*result
)
3311 uint16_t ddb_blocks
;
3314 for (i
= 0; i
< num_planes
; i
++) {
3315 ddb_blocks
= skl_ddb_entry_size(&ddb
->plane
[pipe
][i
]);
3317 result
->plane_en
[i
] = skl_compute_plane_wm(dev_priv
,
3321 &result
->plane_res_b
[i
],
3322 &result
->plane_res_l
[i
]);
3325 ddb_blocks
= skl_ddb_entry_size(&ddb
->cursor
[pipe
]);
3326 result
->cursor_en
= skl_compute_plane_wm(dev_priv
, p
, &p
->cursor
,
3328 &result
->cursor_res_b
,
3329 &result
->cursor_res_l
);
3333 skl_compute_linetime_wm(struct drm_crtc
*crtc
, struct skl_pipe_wm_parameters
*p
)
3335 if (!to_intel_crtc(crtc
)->active
)
3338 if (WARN_ON(p
->pixel_rate
== 0))
3341 return DIV_ROUND_UP(8 * p
->pipe_htotal
* 1000, p
->pixel_rate
);
3344 static void skl_compute_transition_wm(struct drm_crtc
*crtc
,
3345 struct skl_pipe_wm_parameters
*params
,
3346 struct skl_wm_level
*trans_wm
/* out */)
3348 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3351 if (!params
->active
)
3354 /* Until we know more, just disable transition WMs */
3355 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++)
3356 trans_wm
->plane_en
[i
] = false;
3357 trans_wm
->cursor_en
= false;
3360 static void skl_compute_pipe_wm(struct drm_crtc
*crtc
,
3361 struct skl_ddb_allocation
*ddb
,
3362 struct skl_pipe_wm_parameters
*params
,
3363 struct skl_pipe_wm
*pipe_wm
)
3365 struct drm_device
*dev
= crtc
->dev
;
3366 const struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3367 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3368 int level
, max_level
= ilk_wm_max_level(dev
);
3370 for (level
= 0; level
<= max_level
; level
++) {
3371 skl_compute_wm_level(dev_priv
, ddb
, params
, intel_crtc
->pipe
,
3372 level
, intel_num_planes(intel_crtc
),
3373 &pipe_wm
->wm
[level
]);
3375 pipe_wm
->linetime
= skl_compute_linetime_wm(crtc
, params
);
3377 skl_compute_transition_wm(crtc
, params
, &pipe_wm
->trans_wm
);
3380 static void skl_compute_wm_results(struct drm_device
*dev
,
3381 struct skl_pipe_wm_parameters
*p
,
3382 struct skl_pipe_wm
*p_wm
,
3383 struct skl_wm_values
*r
,
3384 struct intel_crtc
*intel_crtc
)
3386 int level
, max_level
= ilk_wm_max_level(dev
);
3387 enum pipe pipe
= intel_crtc
->pipe
;
3391 for (level
= 0; level
<= max_level
; level
++) {
3392 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3395 temp
|= p_wm
->wm
[level
].plane_res_l
[i
] <<
3396 PLANE_WM_LINES_SHIFT
;
3397 temp
|= p_wm
->wm
[level
].plane_res_b
[i
];
3398 if (p_wm
->wm
[level
].plane_en
[i
])
3399 temp
|= PLANE_WM_EN
;
3401 r
->plane
[pipe
][i
][level
] = temp
;
3406 temp
|= p_wm
->wm
[level
].cursor_res_l
<< PLANE_WM_LINES_SHIFT
;
3407 temp
|= p_wm
->wm
[level
].cursor_res_b
;
3409 if (p_wm
->wm
[level
].cursor_en
)
3410 temp
|= PLANE_WM_EN
;
3412 r
->cursor
[pipe
][level
] = temp
;
3416 /* transition WMs */
3417 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3419 temp
|= p_wm
->trans_wm
.plane_res_l
[i
] << PLANE_WM_LINES_SHIFT
;
3420 temp
|= p_wm
->trans_wm
.plane_res_b
[i
];
3421 if (p_wm
->trans_wm
.plane_en
[i
])
3422 temp
|= PLANE_WM_EN
;
3424 r
->plane_trans
[pipe
][i
] = temp
;
3428 temp
|= p_wm
->trans_wm
.cursor_res_l
<< PLANE_WM_LINES_SHIFT
;
3429 temp
|= p_wm
->trans_wm
.cursor_res_b
;
3430 if (p_wm
->trans_wm
.cursor_en
)
3431 temp
|= PLANE_WM_EN
;
3433 r
->cursor_trans
[pipe
] = temp
;
3435 r
->wm_linetime
[pipe
] = p_wm
->linetime
;
3438 static void skl_ddb_entry_write(struct drm_i915_private
*dev_priv
, uint32_t reg
,
3439 const struct skl_ddb_entry
*entry
)
3442 I915_WRITE(reg
, (entry
->end
- 1) << 16 | entry
->start
);
3447 static void skl_write_wm_values(struct drm_i915_private
*dev_priv
,
3448 const struct skl_wm_values
*new)
3450 struct drm_device
*dev
= dev_priv
->dev
;
3451 struct intel_crtc
*crtc
;
3453 list_for_each_entry(crtc
, &dev
->mode_config
.crtc_list
, base
.head
) {
3454 int i
, level
, max_level
= ilk_wm_max_level(dev
);
3455 enum pipe pipe
= crtc
->pipe
;
3457 if (!new->dirty
[pipe
])
3460 I915_WRITE(PIPE_WM_LINETIME(pipe
), new->wm_linetime
[pipe
]);
3462 for (level
= 0; level
<= max_level
; level
++) {
3463 for (i
= 0; i
< intel_num_planes(crtc
); i
++)
3464 I915_WRITE(PLANE_WM(pipe
, i
, level
),
3465 new->plane
[pipe
][i
][level
]);
3466 I915_WRITE(CUR_WM(pipe
, level
),
3467 new->cursor
[pipe
][level
]);
3469 for (i
= 0; i
< intel_num_planes(crtc
); i
++)
3470 I915_WRITE(PLANE_WM_TRANS(pipe
, i
),
3471 new->plane_trans
[pipe
][i
]);
3472 I915_WRITE(CUR_WM_TRANS(pipe
), new->cursor_trans
[pipe
]);
3474 for (i
= 0; i
< intel_num_planes(crtc
); i
++) {
3475 skl_ddb_entry_write(dev_priv
,
3476 PLANE_BUF_CFG(pipe
, i
),
3477 &new->ddb
.plane
[pipe
][i
]);
3478 skl_ddb_entry_write(dev_priv
,
3479 PLANE_NV12_BUF_CFG(pipe
, i
),
3480 &new->ddb
.y_plane
[pipe
][i
]);
3483 skl_ddb_entry_write(dev_priv
, CUR_BUF_CFG(pipe
),
3484 &new->ddb
.cursor
[pipe
]);
3489 * When setting up a new DDB allocation arrangement, we need to correctly
3490 * sequence the times at which the new allocations for the pipes are taken into
3491 * account or we'll have pipes fetching from space previously allocated to
3494 * Roughly the sequence looks like:
3495 * 1. re-allocate the pipe(s) with the allocation being reduced and not
3496 * overlapping with a previous light-up pipe (another way to put it is:
3497 * pipes with their new allocation strickly included into their old ones).
3498 * 2. re-allocate the other pipes that get their allocation reduced
3499 * 3. allocate the pipes having their allocation increased
3501 * Steps 1. and 2. are here to take care of the following case:
3502 * - Initially DDB looks like this:
3505 * - pipe B has a reduced DDB allocation that overlaps with the old pipe C
3509 * We need to sequence the re-allocation: C, B, A (and not B, C, A).
3513 skl_wm_flush_pipe(struct drm_i915_private
*dev_priv
, enum pipe pipe
, int pass
)
3517 DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe
), pass
);
3519 for_each_plane(dev_priv
, pipe
, plane
) {
3520 I915_WRITE(PLANE_SURF(pipe
, plane
),
3521 I915_READ(PLANE_SURF(pipe
, plane
)));
3523 I915_WRITE(CURBASE(pipe
), I915_READ(CURBASE(pipe
)));
3527 skl_ddb_allocation_included(const struct skl_ddb_allocation
*old
,
3528 const struct skl_ddb_allocation
*new,
3531 uint16_t old_size
, new_size
;
3533 old_size
= skl_ddb_entry_size(&old
->pipe
[pipe
]);
3534 new_size
= skl_ddb_entry_size(&new->pipe
[pipe
]);
3536 return old_size
!= new_size
&&
3537 new->pipe
[pipe
].start
>= old
->pipe
[pipe
].start
&&
3538 new->pipe
[pipe
].end
<= old
->pipe
[pipe
].end
;
3541 static void skl_flush_wm_values(struct drm_i915_private
*dev_priv
,
3542 struct skl_wm_values
*new_values
)
3544 struct drm_device
*dev
= dev_priv
->dev
;
3545 struct skl_ddb_allocation
*cur_ddb
, *new_ddb
;
3546 bool reallocated
[I915_MAX_PIPES
] = {};
3547 struct intel_crtc
*crtc
;
3550 new_ddb
= &new_values
->ddb
;
3551 cur_ddb
= &dev_priv
->wm
.skl_hw
.ddb
;
3554 * First pass: flush the pipes with the new allocation contained into
3557 * We'll wait for the vblank on those pipes to ensure we can safely
3558 * re-allocate the freed space without this pipe fetching from it.
3560 for_each_intel_crtc(dev
, crtc
) {
3566 if (!skl_ddb_allocation_included(cur_ddb
, new_ddb
, pipe
))
3569 skl_wm_flush_pipe(dev_priv
, pipe
, 1);
3570 intel_wait_for_vblank(dev
, pipe
);
3572 reallocated
[pipe
] = true;
3577 * Second pass: flush the pipes that are having their allocation
3578 * reduced, but overlapping with a previous allocation.
3580 * Here as well we need to wait for the vblank to make sure the freed
3581 * space is not used anymore.
3583 for_each_intel_crtc(dev
, crtc
) {
3589 if (reallocated
[pipe
])
3592 if (skl_ddb_entry_size(&new_ddb
->pipe
[pipe
]) <
3593 skl_ddb_entry_size(&cur_ddb
->pipe
[pipe
])) {
3594 skl_wm_flush_pipe(dev_priv
, pipe
, 2);
3595 intel_wait_for_vblank(dev
, pipe
);
3596 reallocated
[pipe
] = true;
3601 * Third pass: flush the pipes that got more space allocated.
3603 * We don't need to actively wait for the update here, next vblank
3604 * will just get more DDB space with the correct WM values.
3606 for_each_intel_crtc(dev
, crtc
) {
3613 * At this point, only the pipes more space than before are
3614 * left to re-allocate.
3616 if (reallocated
[pipe
])
3619 skl_wm_flush_pipe(dev_priv
, pipe
, 3);
3623 static bool skl_update_pipe_wm(struct drm_crtc
*crtc
,
3624 struct skl_pipe_wm_parameters
*params
,
3625 struct intel_wm_config
*config
,
3626 struct skl_ddb_allocation
*ddb
, /* out */
3627 struct skl_pipe_wm
*pipe_wm
/* out */)
3629 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3631 skl_compute_wm_pipe_parameters(crtc
, params
);
3632 skl_allocate_pipe_ddb(crtc
, config
, params
, ddb
);
3633 skl_compute_pipe_wm(crtc
, ddb
, params
, pipe_wm
);
3635 if (!memcmp(&intel_crtc
->wm
.skl_active
, pipe_wm
, sizeof(*pipe_wm
)))
3638 intel_crtc
->wm
.skl_active
= *pipe_wm
;
3643 static void skl_update_other_pipe_wm(struct drm_device
*dev
,
3644 struct drm_crtc
*crtc
,
3645 struct intel_wm_config
*config
,
3646 struct skl_wm_values
*r
)
3648 struct intel_crtc
*intel_crtc
;
3649 struct intel_crtc
*this_crtc
= to_intel_crtc(crtc
);
3652 * If the WM update hasn't changed the allocation for this_crtc (the
3653 * crtc we are currently computing the new WM values for), other
3654 * enabled crtcs will keep the same allocation and we don't need to
3655 * recompute anything for them.
3657 if (!skl_ddb_allocation_changed(&r
->ddb
, this_crtc
))
3661 * Otherwise, because of this_crtc being freshly enabled/disabled, the
3662 * other active pipes need new DDB allocation and WM values.
3664 list_for_each_entry(intel_crtc
, &dev
->mode_config
.crtc_list
,
3666 struct skl_pipe_wm_parameters params
= {};
3667 struct skl_pipe_wm pipe_wm
= {};
3670 if (this_crtc
->pipe
== intel_crtc
->pipe
)
3673 if (!intel_crtc
->active
)
3676 wm_changed
= skl_update_pipe_wm(&intel_crtc
->base
,
3681 * If we end up re-computing the other pipe WM values, it's
3682 * because it was really needed, so we expect the WM values to
3685 WARN_ON(!wm_changed
);
3687 skl_compute_wm_results(dev
, ¶ms
, &pipe_wm
, r
, intel_crtc
);
3688 r
->dirty
[intel_crtc
->pipe
] = true;
3692 static void skl_clear_wm(struct skl_wm_values
*watermarks
, enum pipe pipe
)
3694 watermarks
->wm_linetime
[pipe
] = 0;
3695 memset(watermarks
->plane
[pipe
], 0,
3696 sizeof(uint32_t) * 8 * I915_MAX_PLANES
);
3697 memset(watermarks
->cursor
[pipe
], 0, sizeof(uint32_t) * 8);
3698 memset(watermarks
->plane_trans
[pipe
],
3699 0, sizeof(uint32_t) * I915_MAX_PLANES
);
3700 watermarks
->cursor_trans
[pipe
] = 0;
3702 /* Clear ddb entries for pipe */
3703 memset(&watermarks
->ddb
.pipe
[pipe
], 0, sizeof(struct skl_ddb_entry
));
3704 memset(&watermarks
->ddb
.plane
[pipe
], 0,
3705 sizeof(struct skl_ddb_entry
) * I915_MAX_PLANES
);
3706 memset(&watermarks
->ddb
.y_plane
[pipe
], 0,
3707 sizeof(struct skl_ddb_entry
) * I915_MAX_PLANES
);
3708 memset(&watermarks
->ddb
.cursor
[pipe
], 0, sizeof(struct skl_ddb_entry
));
3712 static void skl_update_wm(struct drm_crtc
*crtc
)
3714 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3715 struct drm_device
*dev
= crtc
->dev
;
3716 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3717 struct skl_pipe_wm_parameters params
= {};
3718 struct skl_wm_values
*results
= &dev_priv
->wm
.skl_results
;
3719 struct skl_pipe_wm pipe_wm
= {};
3720 struct intel_wm_config config
= {};
3723 /* Clear all dirty flags */
3724 memset(results
->dirty
, 0, sizeof(bool) * I915_MAX_PIPES
);
3726 skl_clear_wm(results
, intel_crtc
->pipe
);
3728 skl_compute_wm_global_parameters(dev
, &config
);
3730 if (!skl_update_pipe_wm(crtc
, ¶ms
, &config
,
3731 &results
->ddb
, &pipe_wm
))
3734 skl_compute_wm_results(dev
, ¶ms
, &pipe_wm
, results
, intel_crtc
);
3735 results
->dirty
[intel_crtc
->pipe
] = true;
3737 skl_update_other_pipe_wm(dev
, crtc
, &config
, results
);
3738 skl_write_wm_values(dev_priv
, results
);
3739 skl_flush_wm_values(dev_priv
, results
);
3741 /* store the new configuration */
3742 dev_priv
->wm
.skl_hw
= *results
;
3746 skl_update_sprite_wm(struct drm_plane
*plane
, struct drm_crtc
*crtc
,
3747 uint32_t sprite_width
, uint32_t sprite_height
,
3748 int pixel_size
, bool enabled
, bool scaled
)
3750 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
3751 struct drm_framebuffer
*fb
= plane
->state
->fb
;
3753 intel_plane
->wm
.enabled
= enabled
;
3754 intel_plane
->wm
.scaled
= scaled
;
3755 intel_plane
->wm
.horiz_pixels
= sprite_width
;
3756 intel_plane
->wm
.vert_pixels
= sprite_height
;
3757 intel_plane
->wm
.tiling
= DRM_FORMAT_MOD_NONE
;
3759 /* For planar: Bpp is for UV plane, y_Bpp is for Y plane */
3760 intel_plane
->wm
.bytes_per_pixel
=
3761 (fb
&& fb
->pixel_format
== DRM_FORMAT_NV12
) ?
3762 drm_format_plane_cpp(plane
->state
->fb
->pixel_format
, 1) : pixel_size
;
3763 intel_plane
->wm
.y_bytes_per_pixel
=
3764 (fb
&& fb
->pixel_format
== DRM_FORMAT_NV12
) ?
3765 drm_format_plane_cpp(plane
->state
->fb
->pixel_format
, 0) : 0;
3768 * Framebuffer can be NULL on plane disable, but it does not
3769 * matter for watermarks if we assume no tiling in that case.
3772 intel_plane
->wm
.tiling
= fb
->modifier
[0];
3773 intel_plane
->wm
.rotation
= plane
->state
->rotation
;
3775 skl_update_wm(crtc
);
3778 static void ilk_update_wm(struct drm_crtc
*crtc
)
3780 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3781 struct drm_device
*dev
= crtc
->dev
;
3782 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3783 struct ilk_wm_maximums max
;
3784 struct ilk_pipe_wm_parameters params
= {};
3785 struct ilk_wm_values results
= {};
3786 enum intel_ddb_partitioning partitioning
;
3787 struct intel_pipe_wm pipe_wm
= {};
3788 struct intel_pipe_wm lp_wm_1_2
= {}, lp_wm_5_6
= {}, *best_lp_wm
;
3789 struct intel_wm_config config
= {};
3791 ilk_compute_wm_parameters(crtc
, ¶ms
);
3793 intel_compute_pipe_wm(crtc
, ¶ms
, &pipe_wm
);
3795 if (!memcmp(&intel_crtc
->wm
.active
, &pipe_wm
, sizeof(pipe_wm
)))
3798 intel_crtc
->wm
.active
= pipe_wm
;
3800 ilk_compute_wm_config(dev
, &config
);
3802 ilk_compute_wm_maximums(dev
, 1, &config
, INTEL_DDB_PART_1_2
, &max
);
3803 ilk_wm_merge(dev
, &config
, &max
, &lp_wm_1_2
);
3805 /* 5/6 split only in single pipe config on IVB+ */
3806 if (INTEL_INFO(dev
)->gen
>= 7 &&
3807 config
.num_pipes_active
== 1 && config
.sprites_enabled
) {
3808 ilk_compute_wm_maximums(dev
, 1, &config
, INTEL_DDB_PART_5_6
, &max
);
3809 ilk_wm_merge(dev
, &config
, &max
, &lp_wm_5_6
);
3811 best_lp_wm
= ilk_find_best_result(dev
, &lp_wm_1_2
, &lp_wm_5_6
);
3813 best_lp_wm
= &lp_wm_1_2
;
3816 partitioning
= (best_lp_wm
== &lp_wm_1_2
) ?
3817 INTEL_DDB_PART_1_2
: INTEL_DDB_PART_5_6
;
3819 ilk_compute_wm_results(dev
, best_lp_wm
, partitioning
, &results
);
3821 ilk_write_wm_values(dev_priv
, &results
);
3825 ilk_update_sprite_wm(struct drm_plane
*plane
,
3826 struct drm_crtc
*crtc
,
3827 uint32_t sprite_width
, uint32_t sprite_height
,
3828 int pixel_size
, bool enabled
, bool scaled
)
3830 struct drm_device
*dev
= plane
->dev
;
3831 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
3833 intel_plane
->wm
.enabled
= enabled
;
3834 intel_plane
->wm
.scaled
= scaled
;
3835 intel_plane
->wm
.horiz_pixels
= sprite_width
;
3836 intel_plane
->wm
.vert_pixels
= sprite_width
;
3837 intel_plane
->wm
.bytes_per_pixel
= pixel_size
;
3840 * IVB workaround: must disable low power watermarks for at least
3841 * one frame before enabling scaling. LP watermarks can be re-enabled
3842 * when scaling is disabled.
3844 * WaCxSRDisabledForSpriteScaling:ivb
3846 if (IS_IVYBRIDGE(dev
) && scaled
&& ilk_disable_lp_wm(dev
))
3847 intel_wait_for_vblank(dev
, intel_plane
->pipe
);
3849 ilk_update_wm(crtc
);
3852 static void skl_pipe_wm_active_state(uint32_t val
,
3853 struct skl_pipe_wm
*active
,
3859 bool is_enabled
= (val
& PLANE_WM_EN
) != 0;
3863 active
->wm
[level
].plane_en
[i
] = is_enabled
;
3864 active
->wm
[level
].plane_res_b
[i
] =
3865 val
& PLANE_WM_BLOCKS_MASK
;
3866 active
->wm
[level
].plane_res_l
[i
] =
3867 (val
>> PLANE_WM_LINES_SHIFT
) &
3868 PLANE_WM_LINES_MASK
;
3870 active
->wm
[level
].cursor_en
= is_enabled
;
3871 active
->wm
[level
].cursor_res_b
=
3872 val
& PLANE_WM_BLOCKS_MASK
;
3873 active
->wm
[level
].cursor_res_l
=
3874 (val
>> PLANE_WM_LINES_SHIFT
) &
3875 PLANE_WM_LINES_MASK
;
3879 active
->trans_wm
.plane_en
[i
] = is_enabled
;
3880 active
->trans_wm
.plane_res_b
[i
] =
3881 val
& PLANE_WM_BLOCKS_MASK
;
3882 active
->trans_wm
.plane_res_l
[i
] =
3883 (val
>> PLANE_WM_LINES_SHIFT
) &
3884 PLANE_WM_LINES_MASK
;
3886 active
->trans_wm
.cursor_en
= is_enabled
;
3887 active
->trans_wm
.cursor_res_b
=
3888 val
& PLANE_WM_BLOCKS_MASK
;
3889 active
->trans_wm
.cursor_res_l
=
3890 (val
>> PLANE_WM_LINES_SHIFT
) &
3891 PLANE_WM_LINES_MASK
;
3896 static void skl_pipe_wm_get_hw_state(struct drm_crtc
*crtc
)
3898 struct drm_device
*dev
= crtc
->dev
;
3899 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3900 struct skl_wm_values
*hw
= &dev_priv
->wm
.skl_hw
;
3901 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3902 struct skl_pipe_wm
*active
= &intel_crtc
->wm
.skl_active
;
3903 enum pipe pipe
= intel_crtc
->pipe
;
3904 int level
, i
, max_level
;
3907 max_level
= ilk_wm_max_level(dev
);
3909 hw
->wm_linetime
[pipe
] = I915_READ(PIPE_WM_LINETIME(pipe
));
3911 for (level
= 0; level
<= max_level
; level
++) {
3912 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++)
3913 hw
->plane
[pipe
][i
][level
] =
3914 I915_READ(PLANE_WM(pipe
, i
, level
));
3915 hw
->cursor
[pipe
][level
] = I915_READ(CUR_WM(pipe
, level
));
3918 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++)
3919 hw
->plane_trans
[pipe
][i
] = I915_READ(PLANE_WM_TRANS(pipe
, i
));
3920 hw
->cursor_trans
[pipe
] = I915_READ(CUR_WM_TRANS(pipe
));
3922 if (!intel_crtc
->active
)
3925 hw
->dirty
[pipe
] = true;
3927 active
->linetime
= hw
->wm_linetime
[pipe
];
3929 for (level
= 0; level
<= max_level
; level
++) {
3930 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3931 temp
= hw
->plane
[pipe
][i
][level
];
3932 skl_pipe_wm_active_state(temp
, active
, false,
3935 temp
= hw
->cursor
[pipe
][level
];
3936 skl_pipe_wm_active_state(temp
, active
, false, true, i
, level
);
3939 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3940 temp
= hw
->plane_trans
[pipe
][i
];
3941 skl_pipe_wm_active_state(temp
, active
, true, false, i
, 0);
3944 temp
= hw
->cursor_trans
[pipe
];
3945 skl_pipe_wm_active_state(temp
, active
, true, true, i
, 0);
3948 void skl_wm_get_hw_state(struct drm_device
*dev
)
3950 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3951 struct skl_ddb_allocation
*ddb
= &dev_priv
->wm
.skl_hw
.ddb
;
3952 struct drm_crtc
*crtc
;
3954 skl_ddb_get_hw_state(dev_priv
, ddb
);
3955 list_for_each_entry(crtc
, &dev
->mode_config
.crtc_list
, head
)
3956 skl_pipe_wm_get_hw_state(crtc
);
3959 static void ilk_pipe_wm_get_hw_state(struct drm_crtc
*crtc
)
3961 struct drm_device
*dev
= crtc
->dev
;
3962 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3963 struct ilk_wm_values
*hw
= &dev_priv
->wm
.hw
;
3964 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3965 struct intel_pipe_wm
*active
= &intel_crtc
->wm
.active
;
3966 enum pipe pipe
= intel_crtc
->pipe
;
3967 static const unsigned int wm0_pipe_reg
[] = {
3968 [PIPE_A
] = WM0_PIPEA_ILK
,
3969 [PIPE_B
] = WM0_PIPEB_ILK
,
3970 [PIPE_C
] = WM0_PIPEC_IVB
,
3973 hw
->wm_pipe
[pipe
] = I915_READ(wm0_pipe_reg
[pipe
]);
3974 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
3975 hw
->wm_linetime
[pipe
] = I915_READ(PIPE_WM_LINETIME(pipe
));
3977 active
->pipe_enabled
= intel_crtc
->active
;
3979 if (active
->pipe_enabled
) {
3980 u32 tmp
= hw
->wm_pipe
[pipe
];
3983 * For active pipes LP0 watermark is marked as
3984 * enabled, and LP1+ watermaks as disabled since
3985 * we can't really reverse compute them in case
3986 * multiple pipes are active.
3988 active
->wm
[0].enable
= true;
3989 active
->wm
[0].pri_val
= (tmp
& WM0_PIPE_PLANE_MASK
) >> WM0_PIPE_PLANE_SHIFT
;
3990 active
->wm
[0].spr_val
= (tmp
& WM0_PIPE_SPRITE_MASK
) >> WM0_PIPE_SPRITE_SHIFT
;
3991 active
->wm
[0].cur_val
= tmp
& WM0_PIPE_CURSOR_MASK
;
3992 active
->linetime
= hw
->wm_linetime
[pipe
];
3994 int level
, max_level
= ilk_wm_max_level(dev
);
3997 * For inactive pipes, all watermark levels
3998 * should be marked as enabled but zeroed,
3999 * which is what we'd compute them to.
4001 for (level
= 0; level
<= max_level
; level
++)
4002 active
->wm
[level
].enable
= true;
4006 #define _FW_WM(value, plane) \
4007 (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT)
4008 #define _FW_WM_VLV(value, plane) \
4009 (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT)
4011 static void vlv_read_wm_values(struct drm_i915_private
*dev_priv
,
4012 struct vlv_wm_values
*wm
)
4017 for_each_pipe(dev_priv
, pipe
) {
4018 tmp
= I915_READ(VLV_DDL(pipe
));
4020 wm
->ddl
[pipe
].primary
=
4021 (tmp
>> DDL_PLANE_SHIFT
) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
4022 wm
->ddl
[pipe
].cursor
=
4023 (tmp
>> DDL_CURSOR_SHIFT
) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
4024 wm
->ddl
[pipe
].sprite
[0] =
4025 (tmp
>> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
4026 wm
->ddl
[pipe
].sprite
[1] =
4027 (tmp
>> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
4030 tmp
= I915_READ(DSPFW1
);
4031 wm
->sr
.plane
= _FW_WM(tmp
, SR
);
4032 wm
->pipe
[PIPE_B
].cursor
= _FW_WM(tmp
, CURSORB
);
4033 wm
->pipe
[PIPE_B
].primary
= _FW_WM_VLV(tmp
, PLANEB
);
4034 wm
->pipe
[PIPE_A
].primary
= _FW_WM_VLV(tmp
, PLANEA
);
4036 tmp
= I915_READ(DSPFW2
);
4037 wm
->pipe
[PIPE_A
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITEB
);
4038 wm
->pipe
[PIPE_A
].cursor
= _FW_WM(tmp
, CURSORA
);
4039 wm
->pipe
[PIPE_A
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEA
);
4041 tmp
= I915_READ(DSPFW3
);
4042 wm
->sr
.cursor
= _FW_WM(tmp
, CURSOR_SR
);
4044 if (IS_CHERRYVIEW(dev_priv
)) {
4045 tmp
= I915_READ(DSPFW7_CHV
);
4046 wm
->pipe
[PIPE_B
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITED
);
4047 wm
->pipe
[PIPE_B
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEC
);
4049 tmp
= I915_READ(DSPFW8_CHV
);
4050 wm
->pipe
[PIPE_C
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITEF
);
4051 wm
->pipe
[PIPE_C
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEE
);
4053 tmp
= I915_READ(DSPFW9_CHV
);
4054 wm
->pipe
[PIPE_C
].primary
= _FW_WM_VLV(tmp
, PLANEC
);
4055 wm
->pipe
[PIPE_C
].cursor
= _FW_WM(tmp
, CURSORC
);
4057 tmp
= I915_READ(DSPHOWM
);
4058 wm
->sr
.plane
|= _FW_WM(tmp
, SR_HI
) << 9;
4059 wm
->pipe
[PIPE_C
].sprite
[1] |= _FW_WM(tmp
, SPRITEF_HI
) << 8;
4060 wm
->pipe
[PIPE_C
].sprite
[0] |= _FW_WM(tmp
, SPRITEE_HI
) << 8;
4061 wm
->pipe
[PIPE_C
].primary
|= _FW_WM(tmp
, PLANEC_HI
) << 8;
4062 wm
->pipe
[PIPE_B
].sprite
[1] |= _FW_WM(tmp
, SPRITED_HI
) << 8;
4063 wm
->pipe
[PIPE_B
].sprite
[0] |= _FW_WM(tmp
, SPRITEC_HI
) << 8;
4064 wm
->pipe
[PIPE_B
].primary
|= _FW_WM(tmp
, PLANEB_HI
) << 8;
4065 wm
->pipe
[PIPE_A
].sprite
[1] |= _FW_WM(tmp
, SPRITEB_HI
) << 8;
4066 wm
->pipe
[PIPE_A
].sprite
[0] |= _FW_WM(tmp
, SPRITEA_HI
) << 8;
4067 wm
->pipe
[PIPE_A
].primary
|= _FW_WM(tmp
, PLANEA_HI
) << 8;
4069 tmp
= I915_READ(DSPFW7
);
4070 wm
->pipe
[PIPE_B
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITED
);
4071 wm
->pipe
[PIPE_B
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEC
);
4073 tmp
= I915_READ(DSPHOWM
);
4074 wm
->sr
.plane
|= _FW_WM(tmp
, SR_HI
) << 9;
4075 wm
->pipe
[PIPE_B
].sprite
[1] |= _FW_WM(tmp
, SPRITED_HI
) << 8;
4076 wm
->pipe
[PIPE_B
].sprite
[0] |= _FW_WM(tmp
, SPRITEC_HI
) << 8;
4077 wm
->pipe
[PIPE_B
].primary
|= _FW_WM(tmp
, PLANEB_HI
) << 8;
4078 wm
->pipe
[PIPE_A
].sprite
[1] |= _FW_WM(tmp
, SPRITEB_HI
) << 8;
4079 wm
->pipe
[PIPE_A
].sprite
[0] |= _FW_WM(tmp
, SPRITEA_HI
) << 8;
4080 wm
->pipe
[PIPE_A
].primary
|= _FW_WM(tmp
, PLANEA_HI
) << 8;
4087 void vlv_wm_get_hw_state(struct drm_device
*dev
)
4089 struct drm_i915_private
*dev_priv
= to_i915(dev
);
4090 struct vlv_wm_values
*wm
= &dev_priv
->wm
.vlv
;
4091 struct intel_plane
*plane
;
4095 vlv_read_wm_values(dev_priv
, wm
);
4097 for_each_intel_plane(dev
, plane
) {
4098 switch (plane
->base
.type
) {
4100 case DRM_PLANE_TYPE_CURSOR
:
4101 plane
->wm
.fifo_size
= 63;
4103 case DRM_PLANE_TYPE_PRIMARY
:
4104 plane
->wm
.fifo_size
= vlv_get_fifo_size(dev
, plane
->pipe
, 0);
4106 case DRM_PLANE_TYPE_OVERLAY
:
4107 sprite
= plane
->plane
;
4108 plane
->wm
.fifo_size
= vlv_get_fifo_size(dev
, plane
->pipe
, sprite
+ 1);
4113 wm
->cxsr
= I915_READ(FW_BLC_SELF_VLV
) & FW_CSPWRDWNEN
;
4114 wm
->level
= VLV_WM_LEVEL_PM2
;
4116 if (IS_CHERRYVIEW(dev_priv
)) {
4117 mutex_lock(&dev_priv
->rps
.hw_lock
);
4119 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DSPFREQ
);
4120 if (val
& DSP_MAXFIFO_PM5_ENABLE
)
4121 wm
->level
= VLV_WM_LEVEL_PM5
;
4123 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
);
4124 if ((val
& FORCE_DDR_HIGH_FREQ
) == 0)
4125 wm
->level
= VLV_WM_LEVEL_DDR_DVFS
;
4127 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4130 for_each_pipe(dev_priv
, pipe
)
4131 DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n",
4132 pipe_name(pipe
), wm
->pipe
[pipe
].primary
, wm
->pipe
[pipe
].cursor
,
4133 wm
->pipe
[pipe
].sprite
[0], wm
->pipe
[pipe
].sprite
[1]);
4135 DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n",
4136 wm
->sr
.plane
, wm
->sr
.cursor
, wm
->level
, wm
->cxsr
);
4139 void ilk_wm_get_hw_state(struct drm_device
*dev
)
4141 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4142 struct ilk_wm_values
*hw
= &dev_priv
->wm
.hw
;
4143 struct drm_crtc
*crtc
;
4145 for_each_crtc(dev
, crtc
)
4146 ilk_pipe_wm_get_hw_state(crtc
);
4148 hw
->wm_lp
[0] = I915_READ(WM1_LP_ILK
);
4149 hw
->wm_lp
[1] = I915_READ(WM2_LP_ILK
);
4150 hw
->wm_lp
[2] = I915_READ(WM3_LP_ILK
);
4152 hw
->wm_lp_spr
[0] = I915_READ(WM1S_LP_ILK
);
4153 if (INTEL_INFO(dev
)->gen
>= 7) {
4154 hw
->wm_lp_spr
[1] = I915_READ(WM2S_LP_IVB
);
4155 hw
->wm_lp_spr
[2] = I915_READ(WM3S_LP_IVB
);
4158 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
4159 hw
->partitioning
= (I915_READ(WM_MISC
) & WM_MISC_DATA_PARTITION_5_6
) ?
4160 INTEL_DDB_PART_5_6
: INTEL_DDB_PART_1_2
;
4161 else if (IS_IVYBRIDGE(dev
))
4162 hw
->partitioning
= (I915_READ(DISP_ARB_CTL2
) & DISP_DATA_PARTITION_5_6
) ?
4163 INTEL_DDB_PART_5_6
: INTEL_DDB_PART_1_2
;
4166 !(I915_READ(DISP_ARB_CTL
) & DISP_FBC_WM_DIS
);
4170 * intel_update_watermarks - update FIFO watermark values based on current modes
4172 * Calculate watermark values for the various WM regs based on current mode
4173 * and plane configuration.
4175 * There are several cases to deal with here:
4176 * - normal (i.e. non-self-refresh)
4177 * - self-refresh (SR) mode
4178 * - lines are large relative to FIFO size (buffer can hold up to 2)
4179 * - lines are small relative to FIFO size (buffer can hold more than 2
4180 * lines), so need to account for TLB latency
4182 * The normal calculation is:
4183 * watermark = dotclock * bytes per pixel * latency
4184 * where latency is platform & configuration dependent (we assume pessimal
4187 * The SR calculation is:
4188 * watermark = (trunc(latency/line time)+1) * surface width *
4191 * line time = htotal / dotclock
4192 * surface width = hdisplay for normal plane and 64 for cursor
4193 * and latency is assumed to be high, as above.
4195 * The final value programmed to the register should always be rounded up,
4196 * and include an extra 2 entries to account for clock crossings.
4198 * We don't use the sprite, so we can ignore that. And on Crestline we have
4199 * to set the non-SR watermarks to 8.
4201 void intel_update_watermarks(struct drm_crtc
*crtc
)
4203 struct drm_i915_private
*dev_priv
= crtc
->dev
->dev_private
;
4205 if (dev_priv
->display
.update_wm
)
4206 dev_priv
->display
.update_wm(crtc
);
4209 void intel_update_sprite_watermarks(struct drm_plane
*plane
,
4210 struct drm_crtc
*crtc
,
4211 uint32_t sprite_width
,
4212 uint32_t sprite_height
,
4214 bool enabled
, bool scaled
)
4216 struct drm_i915_private
*dev_priv
= plane
->dev
->dev_private
;
4218 if (dev_priv
->display
.update_sprite_wm
)
4219 dev_priv
->display
.update_sprite_wm(plane
, crtc
,
4220 sprite_width
, sprite_height
,
4221 pixel_size
, enabled
, scaled
);
4225 * Lock protecting IPS related data structures
4227 DEFINE_SPINLOCK(mchdev_lock
);
4229 /* Global for IPS driver to get at the current i915 device. Protected by
4231 static struct drm_i915_private
*i915_mch_dev
;
4233 bool ironlake_set_drps(struct drm_device
*dev
, u8 val
)
4235 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4238 assert_spin_locked(&mchdev_lock
);
4240 rgvswctl
= I915_READ16(MEMSWCTL
);
4241 if (rgvswctl
& MEMCTL_CMD_STS
) {
4242 DRM_DEBUG("gpu busy, RCS change rejected\n");
4243 return false; /* still busy with another command */
4246 rgvswctl
= (MEMCTL_CMD_CHFREQ
<< MEMCTL_CMD_SHIFT
) |
4247 (val
<< MEMCTL_FREQ_SHIFT
) | MEMCTL_SFCAVM
;
4248 I915_WRITE16(MEMSWCTL
, rgvswctl
);
4249 POSTING_READ16(MEMSWCTL
);
4251 rgvswctl
|= MEMCTL_CMD_STS
;
4252 I915_WRITE16(MEMSWCTL
, rgvswctl
);
4257 static void ironlake_enable_drps(struct drm_device
*dev
)
4259 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4260 u32 rgvmodectl
= I915_READ(MEMMODECTL
);
4261 u8 fmax
, fmin
, fstart
, vstart
;
4263 spin_lock_irq(&mchdev_lock
);
4265 /* Enable temp reporting */
4266 I915_WRITE16(PMMISC
, I915_READ(PMMISC
) | MCPPCE_EN
);
4267 I915_WRITE16(TSC1
, I915_READ(TSC1
) | TSE
);
4269 /* 100ms RC evaluation intervals */
4270 I915_WRITE(RCUPEI
, 100000);
4271 I915_WRITE(RCDNEI
, 100000);
4273 /* Set max/min thresholds to 90ms and 80ms respectively */
4274 I915_WRITE(RCBMAXAVG
, 90000);
4275 I915_WRITE(RCBMINAVG
, 80000);
4277 I915_WRITE(MEMIHYST
, 1);
4279 /* Set up min, max, and cur for interrupt handling */
4280 fmax
= (rgvmodectl
& MEMMODE_FMAX_MASK
) >> MEMMODE_FMAX_SHIFT
;
4281 fmin
= (rgvmodectl
& MEMMODE_FMIN_MASK
);
4282 fstart
= (rgvmodectl
& MEMMODE_FSTART_MASK
) >>
4283 MEMMODE_FSTART_SHIFT
;
4285 vstart
= (I915_READ(PXVFREQ(fstart
)) & PXVFREQ_PX_MASK
) >>
4288 dev_priv
->ips
.fmax
= fmax
; /* IPS callback will increase this */
4289 dev_priv
->ips
.fstart
= fstart
;
4291 dev_priv
->ips
.max_delay
= fstart
;
4292 dev_priv
->ips
.min_delay
= fmin
;
4293 dev_priv
->ips
.cur_delay
= fstart
;
4295 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
4296 fmax
, fmin
, fstart
);
4298 I915_WRITE(MEMINTREN
, MEMINT_CX_SUPR_EN
| MEMINT_EVAL_CHG_EN
);
4301 * Interrupts will be enabled in ironlake_irq_postinstall
4304 I915_WRITE(VIDSTART
, vstart
);
4305 POSTING_READ(VIDSTART
);
4307 rgvmodectl
|= MEMMODE_SWMODE_EN
;
4308 I915_WRITE(MEMMODECTL
, rgvmodectl
);
4310 if (wait_for_atomic((I915_READ(MEMSWCTL
) & MEMCTL_CMD_STS
) == 0, 10))
4311 DRM_ERROR("stuck trying to change perf mode\n");
4314 ironlake_set_drps(dev
, fstart
);
4316 dev_priv
->ips
.last_count1
= I915_READ(DMIEC
) +
4317 I915_READ(DDREC
) + I915_READ(CSIEC
);
4318 dev_priv
->ips
.last_time1
= jiffies_to_msecs(jiffies
);
4319 dev_priv
->ips
.last_count2
= I915_READ(GFXEC
);
4320 dev_priv
->ips
.last_time2
= ktime_get_raw_ns();
4322 spin_unlock_irq(&mchdev_lock
);
4325 static void ironlake_disable_drps(struct drm_device
*dev
)
4327 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4330 spin_lock_irq(&mchdev_lock
);
4332 rgvswctl
= I915_READ16(MEMSWCTL
);
4334 /* Ack interrupts, disable EFC interrupt */
4335 I915_WRITE(MEMINTREN
, I915_READ(MEMINTREN
) & ~MEMINT_EVAL_CHG_EN
);
4336 I915_WRITE(MEMINTRSTS
, MEMINT_EVAL_CHG
);
4337 I915_WRITE(DEIER
, I915_READ(DEIER
) & ~DE_PCU_EVENT
);
4338 I915_WRITE(DEIIR
, DE_PCU_EVENT
);
4339 I915_WRITE(DEIMR
, I915_READ(DEIMR
) | DE_PCU_EVENT
);
4341 /* Go back to the starting frequency */
4342 ironlake_set_drps(dev
, dev_priv
->ips
.fstart
);
4344 rgvswctl
|= MEMCTL_CMD_STS
;
4345 I915_WRITE(MEMSWCTL
, rgvswctl
);
4348 spin_unlock_irq(&mchdev_lock
);
4351 /* There's a funny hw issue where the hw returns all 0 when reading from
4352 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
4353 * ourselves, instead of doing a rmw cycle (which might result in us clearing
4354 * all limits and the gpu stuck at whatever frequency it is at atm).
4356 static u32
intel_rps_limits(struct drm_i915_private
*dev_priv
, u8 val
)
4360 /* Only set the down limit when we've reached the lowest level to avoid
4361 * getting more interrupts, otherwise leave this clear. This prevents a
4362 * race in the hw when coming out of rc6: There's a tiny window where
4363 * the hw runs at the minimal clock before selecting the desired
4364 * frequency, if the down threshold expires in that window we will not
4365 * receive a down interrupt. */
4366 if (IS_GEN9(dev_priv
->dev
)) {
4367 limits
= (dev_priv
->rps
.max_freq_softlimit
) << 23;
4368 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
4369 limits
|= (dev_priv
->rps
.min_freq_softlimit
) << 14;
4371 limits
= dev_priv
->rps
.max_freq_softlimit
<< 24;
4372 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
4373 limits
|= dev_priv
->rps
.min_freq_softlimit
<< 16;
4379 static void gen6_set_rps_thresholds(struct drm_i915_private
*dev_priv
, u8 val
)
4382 u32 threshold_up
= 0, threshold_down
= 0; /* in % */
4383 u32 ei_up
= 0, ei_down
= 0;
4385 new_power
= dev_priv
->rps
.power
;
4386 switch (dev_priv
->rps
.power
) {
4388 if (val
> dev_priv
->rps
.efficient_freq
+ 1 && val
> dev_priv
->rps
.cur_freq
)
4389 new_power
= BETWEEN
;
4393 if (val
<= dev_priv
->rps
.efficient_freq
&& val
< dev_priv
->rps
.cur_freq
)
4394 new_power
= LOW_POWER
;
4395 else if (val
>= dev_priv
->rps
.rp0_freq
&& val
> dev_priv
->rps
.cur_freq
)
4396 new_power
= HIGH_POWER
;
4400 if (val
< (dev_priv
->rps
.rp1_freq
+ dev_priv
->rps
.rp0_freq
) >> 1 && val
< dev_priv
->rps
.cur_freq
)
4401 new_power
= BETWEEN
;
4404 /* Max/min bins are special */
4405 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
4406 new_power
= LOW_POWER
;
4407 if (val
>= dev_priv
->rps
.max_freq_softlimit
)
4408 new_power
= HIGH_POWER
;
4409 if (new_power
== dev_priv
->rps
.power
)
4412 /* Note the units here are not exactly 1us, but 1280ns. */
4413 switch (new_power
) {
4415 /* Upclock if more than 95% busy over 16ms */
4419 /* Downclock if less than 85% busy over 32ms */
4421 threshold_down
= 85;
4425 /* Upclock if more than 90% busy over 13ms */
4429 /* Downclock if less than 75% busy over 32ms */
4431 threshold_down
= 75;
4435 /* Upclock if more than 85% busy over 10ms */
4439 /* Downclock if less than 60% busy over 32ms */
4441 threshold_down
= 60;
4445 I915_WRITE(GEN6_RP_UP_EI
,
4446 GT_INTERVAL_FROM_US(dev_priv
, ei_up
));
4447 I915_WRITE(GEN6_RP_UP_THRESHOLD
,
4448 GT_INTERVAL_FROM_US(dev_priv
, (ei_up
* threshold_up
/ 100)));
4450 I915_WRITE(GEN6_RP_DOWN_EI
,
4451 GT_INTERVAL_FROM_US(dev_priv
, ei_down
));
4452 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
,
4453 GT_INTERVAL_FROM_US(dev_priv
, (ei_down
* threshold_down
/ 100)));
4455 I915_WRITE(GEN6_RP_CONTROL
,
4456 GEN6_RP_MEDIA_TURBO
|
4457 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
4458 GEN6_RP_MEDIA_IS_GFX
|
4460 GEN6_RP_UP_BUSY_AVG
|
4461 GEN6_RP_DOWN_IDLE_AVG
);
4463 dev_priv
->rps
.power
= new_power
;
4464 dev_priv
->rps
.up_threshold
= threshold_up
;
4465 dev_priv
->rps
.down_threshold
= threshold_down
;
4466 dev_priv
->rps
.last_adj
= 0;
4469 static u32
gen6_rps_pm_mask(struct drm_i915_private
*dev_priv
, u8 val
)
4473 if (val
> dev_priv
->rps
.min_freq_softlimit
)
4474 mask
|= GEN6_PM_RP_DOWN_EI_EXPIRED
| GEN6_PM_RP_DOWN_THRESHOLD
| GEN6_PM_RP_DOWN_TIMEOUT
;
4475 if (val
< dev_priv
->rps
.max_freq_softlimit
)
4476 mask
|= GEN6_PM_RP_UP_EI_EXPIRED
| GEN6_PM_RP_UP_THRESHOLD
;
4478 mask
&= dev_priv
->pm_rps_events
;
4480 return gen6_sanitize_rps_pm_mask(dev_priv
, ~mask
);
4483 /* gen6_set_rps is called to update the frequency request, but should also be
4484 * called when the range (min_delay and max_delay) is modified so that we can
4485 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
4486 static void gen6_set_rps(struct drm_device
*dev
, u8 val
)
4488 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4490 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
4491 if (IS_BROXTON(dev
) && (INTEL_REVID(dev
) < BXT_REVID_B0
))
4494 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
4495 WARN_ON(val
> dev_priv
->rps
.max_freq
);
4496 WARN_ON(val
< dev_priv
->rps
.min_freq
);
4498 /* min/max delay may still have been modified so be sure to
4499 * write the limits value.
4501 if (val
!= dev_priv
->rps
.cur_freq
) {
4502 gen6_set_rps_thresholds(dev_priv
, val
);
4505 I915_WRITE(GEN6_RPNSWREQ
,
4506 GEN9_FREQUENCY(val
));
4507 else if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
4508 I915_WRITE(GEN6_RPNSWREQ
,
4509 HSW_FREQUENCY(val
));
4511 I915_WRITE(GEN6_RPNSWREQ
,
4512 GEN6_FREQUENCY(val
) |
4514 GEN6_AGGRESSIVE_TURBO
);
4517 /* Make sure we continue to get interrupts
4518 * until we hit the minimum or maximum frequencies.
4520 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
, intel_rps_limits(dev_priv
, val
));
4521 I915_WRITE(GEN6_PMINTRMSK
, gen6_rps_pm_mask(dev_priv
, val
));
4523 POSTING_READ(GEN6_RPNSWREQ
);
4525 dev_priv
->rps
.cur_freq
= val
;
4526 trace_intel_gpu_freq_change(val
* 50);
4529 static void valleyview_set_rps(struct drm_device
*dev
, u8 val
)
4531 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4533 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
4534 WARN_ON(val
> dev_priv
->rps
.max_freq
);
4535 WARN_ON(val
< dev_priv
->rps
.min_freq
);
4537 if (WARN_ONCE(IS_CHERRYVIEW(dev
) && (val
& 1),
4538 "Odd GPU freq value\n"))
4541 I915_WRITE(GEN6_PMINTRMSK
, gen6_rps_pm_mask(dev_priv
, val
));
4543 if (val
!= dev_priv
->rps
.cur_freq
) {
4544 vlv_punit_write(dev_priv
, PUNIT_REG_GPU_FREQ_REQ
, val
);
4545 if (!IS_CHERRYVIEW(dev_priv
))
4546 gen6_set_rps_thresholds(dev_priv
, val
);
4549 dev_priv
->rps
.cur_freq
= val
;
4550 trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv
, val
));
4553 /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down
4555 * * If Gfx is Idle, then
4556 * 1. Forcewake Media well.
4557 * 2. Request idle freq.
4558 * 3. Release Forcewake of Media well.
4560 static void vlv_set_rps_idle(struct drm_i915_private
*dev_priv
)
4562 u32 val
= dev_priv
->rps
.idle_freq
;
4564 if (dev_priv
->rps
.cur_freq
<= val
)
4567 /* Wake up the media well, as that takes a lot less
4568 * power than the Render well. */
4569 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_MEDIA
);
4570 valleyview_set_rps(dev_priv
->dev
, val
);
4571 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_MEDIA
);
4574 void gen6_rps_busy(struct drm_i915_private
*dev_priv
)
4576 mutex_lock(&dev_priv
->rps
.hw_lock
);
4577 if (dev_priv
->rps
.enabled
) {
4578 if (dev_priv
->pm_rps_events
& (GEN6_PM_RP_DOWN_EI_EXPIRED
| GEN6_PM_RP_UP_EI_EXPIRED
))
4579 gen6_rps_reset_ei(dev_priv
);
4580 I915_WRITE(GEN6_PMINTRMSK
,
4581 gen6_rps_pm_mask(dev_priv
, dev_priv
->rps
.cur_freq
));
4583 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4586 void gen6_rps_idle(struct drm_i915_private
*dev_priv
)
4588 struct drm_device
*dev
= dev_priv
->dev
;
4590 mutex_lock(&dev_priv
->rps
.hw_lock
);
4591 if (dev_priv
->rps
.enabled
) {
4592 if (IS_VALLEYVIEW(dev
))
4593 vlv_set_rps_idle(dev_priv
);
4595 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.idle_freq
);
4596 dev_priv
->rps
.last_adj
= 0;
4597 I915_WRITE(GEN6_PMINTRMSK
, 0xffffffff);
4599 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4601 spin_lock(&dev_priv
->rps
.client_lock
);
4602 while (!list_empty(&dev_priv
->rps
.clients
))
4603 list_del_init(dev_priv
->rps
.clients
.next
);
4604 spin_unlock(&dev_priv
->rps
.client_lock
);
4607 void gen6_rps_boost(struct drm_i915_private
*dev_priv
,
4608 struct intel_rps_client
*rps
,
4609 unsigned long submitted
)
4611 /* This is intentionally racy! We peek at the state here, then
4612 * validate inside the RPS worker.
4614 if (!(dev_priv
->mm
.busy
&&
4615 dev_priv
->rps
.enabled
&&
4616 dev_priv
->rps
.cur_freq
< dev_priv
->rps
.max_freq_softlimit
))
4619 /* Force a RPS boost (and don't count it against the client) if
4620 * the GPU is severely congested.
4622 if (rps
&& time_after(jiffies
, submitted
+ DRM_I915_THROTTLE_JIFFIES
))
4625 spin_lock(&dev_priv
->rps
.client_lock
);
4626 if (rps
== NULL
|| list_empty(&rps
->link
)) {
4627 spin_lock_irq(&dev_priv
->irq_lock
);
4628 if (dev_priv
->rps
.interrupts_enabled
) {
4629 dev_priv
->rps
.client_boost
= true;
4630 queue_work(dev_priv
->wq
, &dev_priv
->rps
.work
);
4632 spin_unlock_irq(&dev_priv
->irq_lock
);
4635 list_add(&rps
->link
, &dev_priv
->rps
.clients
);
4638 dev_priv
->rps
.boosts
++;
4640 spin_unlock(&dev_priv
->rps
.client_lock
);
4643 void intel_set_rps(struct drm_device
*dev
, u8 val
)
4645 if (IS_VALLEYVIEW(dev
))
4646 valleyview_set_rps(dev
, val
);
4648 gen6_set_rps(dev
, val
);
4651 static void gen9_disable_rps(struct drm_device
*dev
)
4653 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4655 I915_WRITE(GEN6_RC_CONTROL
, 0);
4656 I915_WRITE(GEN9_PG_ENABLE
, 0);
4659 static void gen6_disable_rps(struct drm_device
*dev
)
4661 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4663 I915_WRITE(GEN6_RC_CONTROL
, 0);
4664 I915_WRITE(GEN6_RPNSWREQ
, 1 << 31);
4667 static void cherryview_disable_rps(struct drm_device
*dev
)
4669 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4671 I915_WRITE(GEN6_RC_CONTROL
, 0);
4674 static void valleyview_disable_rps(struct drm_device
*dev
)
4676 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4678 /* we're doing forcewake before Disabling RC6,
4679 * This what the BIOS expects when going into suspend */
4680 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4682 I915_WRITE(GEN6_RC_CONTROL
, 0);
4684 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4687 static void intel_print_rc6_info(struct drm_device
*dev
, u32 mode
)
4689 if (IS_VALLEYVIEW(dev
)) {
4690 if (mode
& (GEN7_RC_CTL_TO_MODE
| GEN6_RC_CTL_EI_MODE(1)))
4691 mode
= GEN6_RC_CTL_RC6_ENABLE
;
4696 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n",
4697 (mode
& GEN6_RC_CTL_RC6_ENABLE
) ? "on" : "off",
4698 (mode
& GEN6_RC_CTL_RC6p_ENABLE
) ? "on" : "off",
4699 (mode
& GEN6_RC_CTL_RC6pp_ENABLE
) ? "on" : "off");
4702 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n",
4703 (mode
& GEN6_RC_CTL_RC6_ENABLE
) ? "on" : "off");
4706 static int sanitize_rc6_option(const struct drm_device
*dev
, int enable_rc6
)
4708 /* No RC6 before Ironlake and code is gone for ilk. */
4709 if (INTEL_INFO(dev
)->gen
< 6)
4712 /* Respect the kernel parameter if it is set */
4713 if (enable_rc6
>= 0) {
4717 mask
= INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
|
4720 mask
= INTEL_RC6_ENABLE
;
4722 if ((enable_rc6
& mask
) != enable_rc6
)
4723 DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
4724 enable_rc6
& mask
, enable_rc6
, mask
);
4726 return enable_rc6
& mask
;
4729 if (IS_IVYBRIDGE(dev
))
4730 return (INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
);
4732 return INTEL_RC6_ENABLE
;
4735 int intel_enable_rc6(const struct drm_device
*dev
)
4737 return i915
.enable_rc6
;
4740 static void gen6_init_rps_frequencies(struct drm_device
*dev
)
4742 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4743 uint32_t rp_state_cap
;
4744 u32 ddcc_status
= 0;
4747 /* All of these values are in units of 50MHz */
4748 dev_priv
->rps
.cur_freq
= 0;
4749 /* static values from HW: RP0 > RP1 > RPn (min_freq) */
4750 if (IS_BROXTON(dev
)) {
4751 rp_state_cap
= I915_READ(BXT_RP_STATE_CAP
);
4752 dev_priv
->rps
.rp0_freq
= (rp_state_cap
>> 16) & 0xff;
4753 dev_priv
->rps
.rp1_freq
= (rp_state_cap
>> 8) & 0xff;
4754 dev_priv
->rps
.min_freq
= (rp_state_cap
>> 0) & 0xff;
4756 rp_state_cap
= I915_READ(GEN6_RP_STATE_CAP
);
4757 dev_priv
->rps
.rp0_freq
= (rp_state_cap
>> 0) & 0xff;
4758 dev_priv
->rps
.rp1_freq
= (rp_state_cap
>> 8) & 0xff;
4759 dev_priv
->rps
.min_freq
= (rp_state_cap
>> 16) & 0xff;
4762 /* hw_max = RP0 until we check for overclocking */
4763 dev_priv
->rps
.max_freq
= dev_priv
->rps
.rp0_freq
;
4765 dev_priv
->rps
.efficient_freq
= dev_priv
->rps
.rp1_freq
;
4766 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
) || IS_SKYLAKE(dev
)) {
4767 ret
= sandybridge_pcode_read(dev_priv
,
4768 HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL
,
4771 dev_priv
->rps
.efficient_freq
=
4773 ((ddcc_status
>> 8) & 0xff),
4774 dev_priv
->rps
.min_freq
,
4775 dev_priv
->rps
.max_freq
);
4778 if (IS_SKYLAKE(dev
)) {
4779 /* Store the frequency values in 16.66 MHZ units, which is
4780 the natural hardware unit for SKL */
4781 dev_priv
->rps
.rp0_freq
*= GEN9_FREQ_SCALER
;
4782 dev_priv
->rps
.rp1_freq
*= GEN9_FREQ_SCALER
;
4783 dev_priv
->rps
.min_freq
*= GEN9_FREQ_SCALER
;
4784 dev_priv
->rps
.max_freq
*= GEN9_FREQ_SCALER
;
4785 dev_priv
->rps
.efficient_freq
*= GEN9_FREQ_SCALER
;
4788 dev_priv
->rps
.idle_freq
= dev_priv
->rps
.min_freq
;
4790 /* Preserve min/max settings in case of re-init */
4791 if (dev_priv
->rps
.max_freq_softlimit
== 0)
4792 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
4794 if (dev_priv
->rps
.min_freq_softlimit
== 0) {
4795 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
4796 dev_priv
->rps
.min_freq_softlimit
=
4797 max_t(int, dev_priv
->rps
.efficient_freq
,
4798 intel_freq_opcode(dev_priv
, 450));
4800 dev_priv
->rps
.min_freq_softlimit
=
4801 dev_priv
->rps
.min_freq
;
4805 /* See the Gen9_GT_PM_Programming_Guide doc for the below */
4806 static void gen9_enable_rps(struct drm_device
*dev
)
4808 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4810 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4812 gen6_init_rps_frequencies(dev
);
4814 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
4815 if (IS_BROXTON(dev
) && (INTEL_REVID(dev
) < BXT_REVID_B0
)) {
4816 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4820 /* Program defaults and thresholds for RPS*/
4821 I915_WRITE(GEN6_RC_VIDEO_FREQ
,
4822 GEN9_FREQUENCY(dev_priv
->rps
.rp1_freq
));
4824 /* 1 second timeout*/
4825 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
,
4826 GT_INTERVAL_FROM_US(dev_priv
, 1000000));
4828 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 0xa);
4830 /* Leaning on the below call to gen6_set_rps to program/setup the
4831 * Up/Down EI & threshold registers, as well as the RP_CONTROL,
4832 * RP_INTERRUPT_LIMITS & RPNSWREQ registers */
4833 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
4834 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.min_freq_softlimit
);
4836 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4839 static void gen9_enable_rc6(struct drm_device
*dev
)
4841 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4842 struct intel_engine_cs
*ring
;
4843 uint32_t rc6_mask
= 0;
4846 /* 1a: Software RC state - RC0 */
4847 I915_WRITE(GEN6_RC_STATE
, 0);
4849 /* 1b: Get forcewake during program sequence. Although the driver
4850 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
4851 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4853 /* 2a: Disable RC states. */
4854 I915_WRITE(GEN6_RC_CONTROL
, 0);
4856 /* 2b: Program RC6 thresholds.*/
4858 /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */
4859 if (IS_SKYLAKE(dev
) && !((IS_SKL_GT3(dev
) || IS_SKL_GT4(dev
)) &&
4860 (INTEL_REVID(dev
) <= SKL_REVID_E0
)))
4861 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 108 << 16);
4863 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 54 << 16);
4864 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
4865 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
4866 for_each_ring(ring
, dev_priv
, unused
)
4867 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
4869 if (HAS_GUC_UCODE(dev
))
4870 I915_WRITE(GUC_MAX_IDLE_COUNT
, 0xA);
4872 I915_WRITE(GEN6_RC_SLEEP
, 0);
4873 I915_WRITE(GEN6_RC6_THRESHOLD
, 37500); /* 37.5/125ms per EI */
4875 /* 2c: Program Coarse Power Gating Policies. */
4876 I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS
, 25);
4877 I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS
, 25);
4879 /* 3a: Enable RC6 */
4880 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
4881 rc6_mask
= GEN6_RC_CTL_RC6_ENABLE
;
4882 DRM_INFO("RC6 %s\n", (rc6_mask
& GEN6_RC_CTL_RC6_ENABLE
) ?
4885 if ((IS_SKYLAKE(dev
) && INTEL_REVID(dev
) <= SKL_REVID_D0
) ||
4886 (IS_BROXTON(dev
) && INTEL_REVID(dev
) <= BXT_REVID_A0
))
4887 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4888 GEN7_RC_CTL_TO_MODE
|
4891 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4892 GEN6_RC_CTL_EI_MODE(1) |
4896 * 3b: Enable Coarse Power Gating only when RC6 is enabled.
4897 * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6.
4899 if ((IS_BROXTON(dev
) && (INTEL_REVID(dev
) < BXT_REVID_B0
)) ||
4900 ((IS_SKL_GT3(dev
) || IS_SKL_GT4(dev
)) && (INTEL_REVID(dev
) <= SKL_REVID_E0
)))
4901 I915_WRITE(GEN9_PG_ENABLE
, 0);
4903 I915_WRITE(GEN9_PG_ENABLE
, (rc6_mask
& GEN6_RC_CTL_RC6_ENABLE
) ?
4904 (GEN9_RENDER_PG_ENABLE
| GEN9_MEDIA_PG_ENABLE
) : 0);
4906 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4910 static void gen8_enable_rps(struct drm_device
*dev
)
4912 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4913 struct intel_engine_cs
*ring
;
4914 uint32_t rc6_mask
= 0;
4917 /* 1a: Software RC state - RC0 */
4918 I915_WRITE(GEN6_RC_STATE
, 0);
4920 /* 1c & 1d: Get forcewake during program sequence. Although the driver
4921 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
4922 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4924 /* 2a: Disable RC states. */
4925 I915_WRITE(GEN6_RC_CONTROL
, 0);
4927 /* Initialize rps frequencies */
4928 gen6_init_rps_frequencies(dev
);
4930 /* 2b: Program RC6 thresholds.*/
4931 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16);
4932 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
4933 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
4934 for_each_ring(ring
, dev_priv
, unused
)
4935 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
4936 I915_WRITE(GEN6_RC_SLEEP
, 0);
4937 if (IS_BROADWELL(dev
))
4938 I915_WRITE(GEN6_RC6_THRESHOLD
, 625); /* 800us/1.28 for TO */
4940 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000); /* 50/125ms per EI */
4943 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
4944 rc6_mask
= GEN6_RC_CTL_RC6_ENABLE
;
4945 intel_print_rc6_info(dev
, rc6_mask
);
4946 if (IS_BROADWELL(dev
))
4947 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4948 GEN7_RC_CTL_TO_MODE
|
4951 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4952 GEN6_RC_CTL_EI_MODE(1) |
4955 /* 4 Program defaults and thresholds for RPS*/
4956 I915_WRITE(GEN6_RPNSWREQ
,
4957 HSW_FREQUENCY(dev_priv
->rps
.rp1_freq
));
4958 I915_WRITE(GEN6_RC_VIDEO_FREQ
,
4959 HSW_FREQUENCY(dev_priv
->rps
.rp1_freq
));
4960 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
4961 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 100000000 / 128); /* 1 second timeout */
4963 /* Docs recommend 900MHz, and 300 MHz respectively */
4964 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
,
4965 dev_priv
->rps
.max_freq_softlimit
<< 24 |
4966 dev_priv
->rps
.min_freq_softlimit
<< 16);
4968 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 7600000 / 128); /* 76ms busyness per EI, 90% */
4969 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 31300000 / 128); /* 313ms busyness per EI, 70%*/
4970 I915_WRITE(GEN6_RP_UP_EI
, 66000); /* 84.48ms, XXX: random? */
4971 I915_WRITE(GEN6_RP_DOWN_EI
, 350000); /* 448ms, XXX: random? */
4973 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
4976 I915_WRITE(GEN6_RP_CONTROL
,
4977 GEN6_RP_MEDIA_TURBO
|
4978 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
4979 GEN6_RP_MEDIA_IS_GFX
|
4981 GEN6_RP_UP_BUSY_AVG
|
4982 GEN6_RP_DOWN_IDLE_AVG
);
4984 /* 6: Ring frequency + overclocking (our driver does this later */
4986 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
4987 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.idle_freq
);
4989 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4992 static void gen6_enable_rps(struct drm_device
*dev
)
4994 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4995 struct intel_engine_cs
*ring
;
4996 u32 rc6vids
, pcu_mbox
= 0, rc6_mask
= 0;
5001 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5003 /* Here begins a magic sequence of register writes to enable
5004 * auto-downclocking.
5006 * Perhaps there might be some value in exposing these to
5009 I915_WRITE(GEN6_RC_STATE
, 0);
5011 /* Clear the DBG now so we don't confuse earlier errors */
5012 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
5013 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg
);
5014 I915_WRITE(GTFIFODBG
, gtfifodbg
);
5017 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
5019 /* Initialize rps frequencies */
5020 gen6_init_rps_frequencies(dev
);
5022 /* disable the counters and set deterministic thresholds */
5023 I915_WRITE(GEN6_RC_CONTROL
, 0);
5025 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT
, 1000 << 16);
5026 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16 | 30);
5027 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT
, 30);
5028 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
5029 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
5031 for_each_ring(ring
, dev_priv
, i
)
5032 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
5034 I915_WRITE(GEN6_RC_SLEEP
, 0);
5035 I915_WRITE(GEN6_RC1e_THRESHOLD
, 1000);
5036 if (IS_IVYBRIDGE(dev
))
5037 I915_WRITE(GEN6_RC6_THRESHOLD
, 125000);
5039 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000);
5040 I915_WRITE(GEN6_RC6p_THRESHOLD
, 150000);
5041 I915_WRITE(GEN6_RC6pp_THRESHOLD
, 64000); /* unused */
5043 /* Check if we are enabling RC6 */
5044 rc6_mode
= intel_enable_rc6(dev_priv
->dev
);
5045 if (rc6_mode
& INTEL_RC6_ENABLE
)
5046 rc6_mask
|= GEN6_RC_CTL_RC6_ENABLE
;
5048 /* We don't use those on Haswell */
5049 if (!IS_HASWELL(dev
)) {
5050 if (rc6_mode
& INTEL_RC6p_ENABLE
)
5051 rc6_mask
|= GEN6_RC_CTL_RC6p_ENABLE
;
5053 if (rc6_mode
& INTEL_RC6pp_ENABLE
)
5054 rc6_mask
|= GEN6_RC_CTL_RC6pp_ENABLE
;
5057 intel_print_rc6_info(dev
, rc6_mask
);
5059 I915_WRITE(GEN6_RC_CONTROL
,
5061 GEN6_RC_CTL_EI_MODE(1) |
5062 GEN6_RC_CTL_HW_ENABLE
);
5064 /* Power down if completely idle for over 50ms */
5065 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 50000);
5066 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
5068 ret
= sandybridge_pcode_write(dev_priv
, GEN6_PCODE_WRITE_MIN_FREQ_TABLE
, 0);
5070 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
5072 ret
= sandybridge_pcode_read(dev_priv
, GEN6_READ_OC_PARAMS
, &pcu_mbox
);
5073 if (!ret
&& (pcu_mbox
& (1<<31))) { /* OC supported */
5074 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
5075 (dev_priv
->rps
.max_freq_softlimit
& 0xff) * 50,
5076 (pcu_mbox
& 0xff) * 50);
5077 dev_priv
->rps
.max_freq
= pcu_mbox
& 0xff;
5080 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
5081 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.idle_freq
);
5084 ret
= sandybridge_pcode_read(dev_priv
, GEN6_PCODE_READ_RC6VIDS
, &rc6vids
);
5085 if (IS_GEN6(dev
) && ret
) {
5086 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
5087 } else if (IS_GEN6(dev
) && (GEN6_DECODE_RC6_VID(rc6vids
& 0xff) < 450)) {
5088 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
5089 GEN6_DECODE_RC6_VID(rc6vids
& 0xff), 450);
5090 rc6vids
&= 0xffff00;
5091 rc6vids
|= GEN6_ENCODE_RC6_VID(450);
5092 ret
= sandybridge_pcode_write(dev_priv
, GEN6_PCODE_WRITE_RC6VIDS
, rc6vids
);
5094 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
5097 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
5100 static void __gen6_update_ring_freq(struct drm_device
*dev
)
5102 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5104 unsigned int gpu_freq
;
5105 unsigned int max_ia_freq
, min_ring_freq
;
5106 unsigned int max_gpu_freq
, min_gpu_freq
;
5107 int scaling_factor
= 180;
5108 struct cpufreq_policy
*policy
;
5110 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5112 policy
= cpufreq_cpu_get(0);
5114 max_ia_freq
= policy
->cpuinfo
.max_freq
;
5115 cpufreq_cpu_put(policy
);
5118 * Default to measured freq if none found, PCU will ensure we
5121 max_ia_freq
= tsc_khz
;
5124 /* Convert from kHz to MHz */
5125 max_ia_freq
/= 1000;
5127 min_ring_freq
= I915_READ(DCLK
) & 0xf;
5128 /* convert DDR frequency from units of 266.6MHz to bandwidth */
5129 min_ring_freq
= mult_frac(min_ring_freq
, 8, 3);
5131 if (IS_SKYLAKE(dev
)) {
5132 /* Convert GT frequency to 50 HZ units */
5133 min_gpu_freq
= dev_priv
->rps
.min_freq
/ GEN9_FREQ_SCALER
;
5134 max_gpu_freq
= dev_priv
->rps
.max_freq
/ GEN9_FREQ_SCALER
;
5136 min_gpu_freq
= dev_priv
->rps
.min_freq
;
5137 max_gpu_freq
= dev_priv
->rps
.max_freq
;
5141 * For each potential GPU frequency, load a ring frequency we'd like
5142 * to use for memory access. We do this by specifying the IA frequency
5143 * the PCU should use as a reference to determine the ring frequency.
5145 for (gpu_freq
= max_gpu_freq
; gpu_freq
>= min_gpu_freq
; gpu_freq
--) {
5146 int diff
= max_gpu_freq
- gpu_freq
;
5147 unsigned int ia_freq
= 0, ring_freq
= 0;
5149 if (IS_SKYLAKE(dev
)) {
5151 * ring_freq = 2 * GT. ring_freq is in 100MHz units
5152 * No floor required for ring frequency on SKL.
5154 ring_freq
= gpu_freq
;
5155 } else if (INTEL_INFO(dev
)->gen
>= 8) {
5156 /* max(2 * GT, DDR). NB: GT is 50MHz units */
5157 ring_freq
= max(min_ring_freq
, gpu_freq
);
5158 } else if (IS_HASWELL(dev
)) {
5159 ring_freq
= mult_frac(gpu_freq
, 5, 4);
5160 ring_freq
= max(min_ring_freq
, ring_freq
);
5161 /* leave ia_freq as the default, chosen by cpufreq */
5163 /* On older processors, there is no separate ring
5164 * clock domain, so in order to boost the bandwidth
5165 * of the ring, we need to upclock the CPU (ia_freq).
5167 * For GPU frequencies less than 750MHz,
5168 * just use the lowest ring freq.
5170 if (gpu_freq
< min_freq
)
5173 ia_freq
= max_ia_freq
- ((diff
* scaling_factor
) / 2);
5174 ia_freq
= DIV_ROUND_CLOSEST(ia_freq
, 100);
5177 sandybridge_pcode_write(dev_priv
,
5178 GEN6_PCODE_WRITE_MIN_FREQ_TABLE
,
5179 ia_freq
<< GEN6_PCODE_FREQ_IA_RATIO_SHIFT
|
5180 ring_freq
<< GEN6_PCODE_FREQ_RING_RATIO_SHIFT
|
5185 void gen6_update_ring_freq(struct drm_device
*dev
)
5187 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5189 if (!HAS_CORE_RING_FREQ(dev
))
5192 mutex_lock(&dev_priv
->rps
.hw_lock
);
5193 __gen6_update_ring_freq(dev
);
5194 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5197 static int cherryview_rps_max_freq(struct drm_i915_private
*dev_priv
)
5199 struct drm_device
*dev
= dev_priv
->dev
;
5202 if (dev
->pdev
->revision
>= 0x20) {
5203 val
= vlv_punit_read(dev_priv
, FB_GFX_FMAX_AT_VMAX_FUSE
);
5205 switch (INTEL_INFO(dev
)->eu_total
) {
5207 /* (2 * 4) config */
5208 rp0
= (val
>> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT
);
5211 /* (2 * 6) config */
5212 rp0
= (val
>> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT
);
5215 /* (2 * 8) config */
5217 /* Setting (2 * 8) Min RP0 for any other combination */
5218 rp0
= (val
>> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT
);
5221 rp0
= (rp0
& FB_GFX_FREQ_FUSE_MASK
);
5223 /* For pre-production hardware */
5224 val
= vlv_punit_read(dev_priv
, PUNIT_GPU_STATUS_REG
);
5225 rp0
= (val
>> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT
) &
5226 PUNIT_GPU_STATUS_MAX_FREQ_MASK
;
5231 static int cherryview_rps_rpe_freq(struct drm_i915_private
*dev_priv
)
5235 val
= vlv_punit_read(dev_priv
, PUNIT_GPU_DUTYCYCLE_REG
);
5236 rpe
= (val
>> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT
) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK
;
5241 static int cherryview_rps_guar_freq(struct drm_i915_private
*dev_priv
)
5243 struct drm_device
*dev
= dev_priv
->dev
;
5246 if (dev
->pdev
->revision
>= 0x20) {
5247 val
= vlv_punit_read(dev_priv
, FB_GFX_FMAX_AT_VMAX_FUSE
);
5248 rp1
= (val
& FB_GFX_FREQ_FUSE_MASK
);
5250 /* For pre-production hardware */
5251 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5252 rp1
= ((val
>> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT
) &
5253 PUNIT_GPU_STATUS_MAX_FREQ_MASK
);
5258 static int valleyview_rps_guar_freq(struct drm_i915_private
*dev_priv
)
5262 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FREQ_FUSE
);
5264 rp1
= (val
& FB_GFX_FGUARANTEED_FREQ_FUSE_MASK
) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT
;
5269 static int valleyview_rps_max_freq(struct drm_i915_private
*dev_priv
)
5273 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FREQ_FUSE
);
5275 rp0
= (val
& FB_GFX_MAX_FREQ_FUSE_MASK
) >> FB_GFX_MAX_FREQ_FUSE_SHIFT
;
5277 rp0
= min_t(u32
, rp0
, 0xea);
5282 static int valleyview_rps_rpe_freq(struct drm_i915_private
*dev_priv
)
5286 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FMAX_FUSE_LO
);
5287 rpe
= (val
& FB_FMAX_VMIN_FREQ_LO_MASK
) >> FB_FMAX_VMIN_FREQ_LO_SHIFT
;
5288 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FMAX_FUSE_HI
);
5289 rpe
|= (val
& FB_FMAX_VMIN_FREQ_HI_MASK
) << 5;
5294 static int valleyview_rps_min_freq(struct drm_i915_private
*dev_priv
)
5296 return vlv_punit_read(dev_priv
, PUNIT_REG_GPU_LFM
) & 0xff;
5299 /* Check that the pctx buffer wasn't move under us. */
5300 static void valleyview_check_pctx(struct drm_i915_private
*dev_priv
)
5302 unsigned long pctx_addr
= I915_READ(VLV_PCBR
) & ~4095;
5304 WARN_ON(pctx_addr
!= dev_priv
->mm
.stolen_base
+
5305 dev_priv
->vlv_pctx
->stolen
->start
);
5309 /* Check that the pcbr address is not empty. */
5310 static void cherryview_check_pctx(struct drm_i915_private
*dev_priv
)
5312 unsigned long pctx_addr
= I915_READ(VLV_PCBR
) & ~4095;
5314 WARN_ON((pctx_addr
>> VLV_PCBR_ADDR_SHIFT
) == 0);
5317 static void cherryview_setup_pctx(struct drm_device
*dev
)
5319 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5320 unsigned long pctx_paddr
, paddr
;
5321 struct i915_gtt
*gtt
= &dev_priv
->gtt
;
5323 int pctx_size
= 32*1024;
5325 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
5327 pcbr
= I915_READ(VLV_PCBR
);
5328 if ((pcbr
>> VLV_PCBR_ADDR_SHIFT
) == 0) {
5329 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
5330 paddr
= (dev_priv
->mm
.stolen_base
+
5331 (gtt
->stolen_size
- pctx_size
));
5333 pctx_paddr
= (paddr
& (~4095));
5334 I915_WRITE(VLV_PCBR
, pctx_paddr
);
5337 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR
));
5340 static void valleyview_setup_pctx(struct drm_device
*dev
)
5342 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5343 struct drm_i915_gem_object
*pctx
;
5344 unsigned long pctx_paddr
;
5346 int pctx_size
= 24*1024;
5348 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
5350 pcbr
= I915_READ(VLV_PCBR
);
5352 /* BIOS set it up already, grab the pre-alloc'd space */
5355 pcbr_offset
= (pcbr
& (~4095)) - dev_priv
->mm
.stolen_base
;
5356 pctx
= i915_gem_object_create_stolen_for_preallocated(dev_priv
->dev
,
5358 I915_GTT_OFFSET_NONE
,
5363 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
5366 * From the Gunit register HAS:
5367 * The Gfx driver is expected to program this register and ensure
5368 * proper allocation within Gfx stolen memory. For example, this
5369 * register should be programmed such than the PCBR range does not
5370 * overlap with other ranges, such as the frame buffer, protected
5371 * memory, or any other relevant ranges.
5373 pctx
= i915_gem_object_create_stolen(dev
, pctx_size
);
5375 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
5379 pctx_paddr
= dev_priv
->mm
.stolen_base
+ pctx
->stolen
->start
;
5380 I915_WRITE(VLV_PCBR
, pctx_paddr
);
5383 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR
));
5384 dev_priv
->vlv_pctx
= pctx
;
5387 static void valleyview_cleanup_pctx(struct drm_device
*dev
)
5389 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5391 if (WARN_ON(!dev_priv
->vlv_pctx
))
5394 drm_gem_object_unreference(&dev_priv
->vlv_pctx
->base
);
5395 dev_priv
->vlv_pctx
= NULL
;
5398 static void valleyview_init_gt_powersave(struct drm_device
*dev
)
5400 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5403 valleyview_setup_pctx(dev
);
5405 mutex_lock(&dev_priv
->rps
.hw_lock
);
5407 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5408 switch ((val
>> 6) & 3) {
5411 dev_priv
->mem_freq
= 800;
5414 dev_priv
->mem_freq
= 1066;
5417 dev_priv
->mem_freq
= 1333;
5420 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv
->mem_freq
);
5422 dev_priv
->rps
.max_freq
= valleyview_rps_max_freq(dev_priv
);
5423 dev_priv
->rps
.rp0_freq
= dev_priv
->rps
.max_freq
;
5424 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
5425 intel_gpu_freq(dev_priv
, dev_priv
->rps
.max_freq
),
5426 dev_priv
->rps
.max_freq
);
5428 dev_priv
->rps
.efficient_freq
= valleyview_rps_rpe_freq(dev_priv
);
5429 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
5430 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5431 dev_priv
->rps
.efficient_freq
);
5433 dev_priv
->rps
.rp1_freq
= valleyview_rps_guar_freq(dev_priv
);
5434 DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n",
5435 intel_gpu_freq(dev_priv
, dev_priv
->rps
.rp1_freq
),
5436 dev_priv
->rps
.rp1_freq
);
5438 dev_priv
->rps
.min_freq
= valleyview_rps_min_freq(dev_priv
);
5439 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
5440 intel_gpu_freq(dev_priv
, dev_priv
->rps
.min_freq
),
5441 dev_priv
->rps
.min_freq
);
5443 dev_priv
->rps
.idle_freq
= dev_priv
->rps
.min_freq
;
5445 /* Preserve min/max settings in case of re-init */
5446 if (dev_priv
->rps
.max_freq_softlimit
== 0)
5447 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
5449 if (dev_priv
->rps
.min_freq_softlimit
== 0)
5450 dev_priv
->rps
.min_freq_softlimit
= dev_priv
->rps
.min_freq
;
5452 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5455 static void cherryview_init_gt_powersave(struct drm_device
*dev
)
5457 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5460 cherryview_setup_pctx(dev
);
5462 mutex_lock(&dev_priv
->rps
.hw_lock
);
5464 mutex_lock(&dev_priv
->sb_lock
);
5465 val
= vlv_cck_read(dev_priv
, CCK_FUSE_REG
);
5466 mutex_unlock(&dev_priv
->sb_lock
);
5468 switch ((val
>> 2) & 0x7) {
5471 dev_priv
->rps
.cz_freq
= 200;
5472 dev_priv
->mem_freq
= 1600;
5475 dev_priv
->rps
.cz_freq
= 267;
5476 dev_priv
->mem_freq
= 1600;
5479 dev_priv
->rps
.cz_freq
= 333;
5480 dev_priv
->mem_freq
= 2000;
5483 dev_priv
->rps
.cz_freq
= 320;
5484 dev_priv
->mem_freq
= 1600;
5487 dev_priv
->rps
.cz_freq
= 400;
5488 dev_priv
->mem_freq
= 1600;
5491 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv
->mem_freq
);
5493 dev_priv
->rps
.max_freq
= cherryview_rps_max_freq(dev_priv
);
5494 dev_priv
->rps
.rp0_freq
= dev_priv
->rps
.max_freq
;
5495 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
5496 intel_gpu_freq(dev_priv
, dev_priv
->rps
.max_freq
),
5497 dev_priv
->rps
.max_freq
);
5499 dev_priv
->rps
.efficient_freq
= cherryview_rps_rpe_freq(dev_priv
);
5500 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
5501 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5502 dev_priv
->rps
.efficient_freq
);
5504 dev_priv
->rps
.rp1_freq
= cherryview_rps_guar_freq(dev_priv
);
5505 DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n",
5506 intel_gpu_freq(dev_priv
, dev_priv
->rps
.rp1_freq
),
5507 dev_priv
->rps
.rp1_freq
);
5509 /* PUnit validated range is only [RPe, RP0] */
5510 dev_priv
->rps
.min_freq
= dev_priv
->rps
.efficient_freq
;
5511 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
5512 intel_gpu_freq(dev_priv
, dev_priv
->rps
.min_freq
),
5513 dev_priv
->rps
.min_freq
);
5515 WARN_ONCE((dev_priv
->rps
.max_freq
|
5516 dev_priv
->rps
.efficient_freq
|
5517 dev_priv
->rps
.rp1_freq
|
5518 dev_priv
->rps
.min_freq
) & 1,
5519 "Odd GPU freq values\n");
5521 dev_priv
->rps
.idle_freq
= dev_priv
->rps
.min_freq
;
5523 /* Preserve min/max settings in case of re-init */
5524 if (dev_priv
->rps
.max_freq_softlimit
== 0)
5525 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
5527 if (dev_priv
->rps
.min_freq_softlimit
== 0)
5528 dev_priv
->rps
.min_freq_softlimit
= dev_priv
->rps
.min_freq
;
5530 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5533 static void valleyview_cleanup_gt_powersave(struct drm_device
*dev
)
5535 valleyview_cleanup_pctx(dev
);
5538 static void cherryview_enable_rps(struct drm_device
*dev
)
5540 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5541 struct intel_engine_cs
*ring
;
5542 u32 gtfifodbg
, val
, rc6_mode
= 0, pcbr
;
5545 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5547 gtfifodbg
= I915_READ(GTFIFODBG
);
5549 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5551 I915_WRITE(GTFIFODBG
, gtfifodbg
);
5554 cherryview_check_pctx(dev_priv
);
5556 /* 1a & 1b: Get forcewake during program sequence. Although the driver
5557 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
5558 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
5560 /* Disable RC states. */
5561 I915_WRITE(GEN6_RC_CONTROL
, 0);
5563 /* 2a: Program RC6 thresholds.*/
5564 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16);
5565 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
5566 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
5568 for_each_ring(ring
, dev_priv
, i
)
5569 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
5570 I915_WRITE(GEN6_RC_SLEEP
, 0);
5572 /* TO threshold set to 500 us ( 0x186 * 1.28 us) */
5573 I915_WRITE(GEN6_RC6_THRESHOLD
, 0x186);
5575 /* allows RC6 residency counter to work */
5576 I915_WRITE(VLV_COUNTER_CONTROL
,
5577 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH
|
5578 VLV_MEDIA_RC6_COUNT_EN
|
5579 VLV_RENDER_RC6_COUNT_EN
));
5581 /* For now we assume BIOS is allocating and populating the PCBR */
5582 pcbr
= I915_READ(VLV_PCBR
);
5585 if ((intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
) &&
5586 (pcbr
>> VLV_PCBR_ADDR_SHIFT
))
5587 rc6_mode
= GEN7_RC_CTL_TO_MODE
;
5589 I915_WRITE(GEN6_RC_CONTROL
, rc6_mode
);
5591 /* 4 Program defaults and thresholds for RPS*/
5592 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 1000000);
5593 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 59400);
5594 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 245000);
5595 I915_WRITE(GEN6_RP_UP_EI
, 66000);
5596 I915_WRITE(GEN6_RP_DOWN_EI
, 350000);
5598 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
5601 I915_WRITE(GEN6_RP_CONTROL
,
5602 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
5603 GEN6_RP_MEDIA_IS_GFX
|
5605 GEN6_RP_UP_BUSY_AVG
|
5606 GEN6_RP_DOWN_IDLE_AVG
);
5608 /* Setting Fixed Bias */
5609 val
= VLV_OVERRIDE_EN
|
5611 CHV_BIAS_CPU_50_SOC_50
;
5612 vlv_punit_write(dev_priv
, VLV_TURBO_SOC_OVERRIDE
, val
);
5614 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5616 /* RPS code assumes GPLL is used */
5617 WARN_ONCE((val
& GPLLENABLE
) == 0, "GPLL not enabled\n");
5619 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val
& GPLLENABLE
));
5620 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val
);
5622 dev_priv
->rps
.cur_freq
= (val
>> 8) & 0xff;
5623 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
5624 intel_gpu_freq(dev_priv
, dev_priv
->rps
.cur_freq
),
5625 dev_priv
->rps
.cur_freq
);
5627 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
5628 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5629 dev_priv
->rps
.efficient_freq
);
5631 valleyview_set_rps(dev_priv
->dev
, dev_priv
->rps
.efficient_freq
);
5633 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
5636 static void valleyview_enable_rps(struct drm_device
*dev
)
5638 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5639 struct intel_engine_cs
*ring
;
5640 u32 gtfifodbg
, val
, rc6_mode
= 0;
5643 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5645 valleyview_check_pctx(dev_priv
);
5647 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
5648 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5650 I915_WRITE(GTFIFODBG
, gtfifodbg
);
5653 /* If VLV, Forcewake all wells, else re-direct to regular path */
5654 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
5656 /* Disable RC states. */
5657 I915_WRITE(GEN6_RC_CONTROL
, 0);
5659 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 1000000);
5660 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 59400);
5661 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 245000);
5662 I915_WRITE(GEN6_RP_UP_EI
, 66000);
5663 I915_WRITE(GEN6_RP_DOWN_EI
, 350000);
5665 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
5667 I915_WRITE(GEN6_RP_CONTROL
,
5668 GEN6_RP_MEDIA_TURBO
|
5669 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
5670 GEN6_RP_MEDIA_IS_GFX
|
5672 GEN6_RP_UP_BUSY_AVG
|
5673 GEN6_RP_DOWN_IDLE_CONT
);
5675 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 0x00280000);
5676 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
5677 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
5679 for_each_ring(ring
, dev_priv
, i
)
5680 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
5682 I915_WRITE(GEN6_RC6_THRESHOLD
, 0x557);
5684 /* allows RC6 residency counter to work */
5685 I915_WRITE(VLV_COUNTER_CONTROL
,
5686 _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN
|
5687 VLV_RENDER_RC0_COUNT_EN
|
5688 VLV_MEDIA_RC6_COUNT_EN
|
5689 VLV_RENDER_RC6_COUNT_EN
));
5691 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
5692 rc6_mode
= GEN7_RC_CTL_TO_MODE
| VLV_RC_CTL_CTX_RST_PARALLEL
;
5694 intel_print_rc6_info(dev
, rc6_mode
);
5696 I915_WRITE(GEN6_RC_CONTROL
, rc6_mode
);
5698 /* Setting Fixed Bias */
5699 val
= VLV_OVERRIDE_EN
|
5701 VLV_BIAS_CPU_125_SOC_875
;
5702 vlv_punit_write(dev_priv
, VLV_TURBO_SOC_OVERRIDE
, val
);
5704 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5706 /* RPS code assumes GPLL is used */
5707 WARN_ONCE((val
& GPLLENABLE
) == 0, "GPLL not enabled\n");
5709 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val
& GPLLENABLE
));
5710 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val
);
5712 dev_priv
->rps
.cur_freq
= (val
>> 8) & 0xff;
5713 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
5714 intel_gpu_freq(dev_priv
, dev_priv
->rps
.cur_freq
),
5715 dev_priv
->rps
.cur_freq
);
5717 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
5718 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5719 dev_priv
->rps
.efficient_freq
);
5721 valleyview_set_rps(dev_priv
->dev
, dev_priv
->rps
.efficient_freq
);
5723 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
5726 static unsigned long intel_pxfreq(u32 vidfreq
)
5729 int div
= (vidfreq
& 0x3f0000) >> 16;
5730 int post
= (vidfreq
& 0x3000) >> 12;
5731 int pre
= (vidfreq
& 0x7);
5736 freq
= ((div
* 133333) / ((1<<post
) * pre
));
5741 static const struct cparams
{
5747 { 1, 1333, 301, 28664 },
5748 { 1, 1066, 294, 24460 },
5749 { 1, 800, 294, 25192 },
5750 { 0, 1333, 276, 27605 },
5751 { 0, 1066, 276, 27605 },
5752 { 0, 800, 231, 23784 },
5755 static unsigned long __i915_chipset_val(struct drm_i915_private
*dev_priv
)
5757 u64 total_count
, diff
, ret
;
5758 u32 count1
, count2
, count3
, m
= 0, c
= 0;
5759 unsigned long now
= jiffies_to_msecs(jiffies
), diff1
;
5762 assert_spin_locked(&mchdev_lock
);
5764 diff1
= now
- dev_priv
->ips
.last_time1
;
5766 /* Prevent division-by-zero if we are asking too fast.
5767 * Also, we don't get interesting results if we are polling
5768 * faster than once in 10ms, so just return the saved value
5772 return dev_priv
->ips
.chipset_power
;
5774 count1
= I915_READ(DMIEC
);
5775 count2
= I915_READ(DDREC
);
5776 count3
= I915_READ(CSIEC
);
5778 total_count
= count1
+ count2
+ count3
;
5780 /* FIXME: handle per-counter overflow */
5781 if (total_count
< dev_priv
->ips
.last_count1
) {
5782 diff
= ~0UL - dev_priv
->ips
.last_count1
;
5783 diff
+= total_count
;
5785 diff
= total_count
- dev_priv
->ips
.last_count1
;
5788 for (i
= 0; i
< ARRAY_SIZE(cparams
); i
++) {
5789 if (cparams
[i
].i
== dev_priv
->ips
.c_m
&&
5790 cparams
[i
].t
== dev_priv
->ips
.r_t
) {
5797 diff
= div_u64(diff
, diff1
);
5798 ret
= ((m
* diff
) + c
);
5799 ret
= div_u64(ret
, 10);
5801 dev_priv
->ips
.last_count1
= total_count
;
5802 dev_priv
->ips
.last_time1
= now
;
5804 dev_priv
->ips
.chipset_power
= ret
;
5809 unsigned long i915_chipset_val(struct drm_i915_private
*dev_priv
)
5811 struct drm_device
*dev
= dev_priv
->dev
;
5814 if (INTEL_INFO(dev
)->gen
!= 5)
5817 spin_lock_irq(&mchdev_lock
);
5819 val
= __i915_chipset_val(dev_priv
);
5821 spin_unlock_irq(&mchdev_lock
);
5826 unsigned long i915_mch_val(struct drm_i915_private
*dev_priv
)
5828 unsigned long m
, x
, b
;
5831 tsfs
= I915_READ(TSFS
);
5833 m
= ((tsfs
& TSFS_SLOPE_MASK
) >> TSFS_SLOPE_SHIFT
);
5834 x
= I915_READ8(TR1
);
5836 b
= tsfs
& TSFS_INTR_MASK
;
5838 return ((m
* x
) / 127) - b
;
5841 static int _pxvid_to_vd(u8 pxvid
)
5846 if (pxvid
>= 8 && pxvid
< 31)
5849 return (pxvid
+ 2) * 125;
5852 static u32
pvid_to_extvid(struct drm_i915_private
*dev_priv
, u8 pxvid
)
5854 struct drm_device
*dev
= dev_priv
->dev
;
5855 const int vd
= _pxvid_to_vd(pxvid
);
5856 const int vm
= vd
- 1125;
5858 if (INTEL_INFO(dev
)->is_mobile
)
5859 return vm
> 0 ? vm
: 0;
5864 static void __i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
5866 u64 now
, diff
, diffms
;
5869 assert_spin_locked(&mchdev_lock
);
5871 now
= ktime_get_raw_ns();
5872 diffms
= now
- dev_priv
->ips
.last_time2
;
5873 do_div(diffms
, NSEC_PER_MSEC
);
5875 /* Don't divide by 0 */
5879 count
= I915_READ(GFXEC
);
5881 if (count
< dev_priv
->ips
.last_count2
) {
5882 diff
= ~0UL - dev_priv
->ips
.last_count2
;
5885 diff
= count
- dev_priv
->ips
.last_count2
;
5888 dev_priv
->ips
.last_count2
= count
;
5889 dev_priv
->ips
.last_time2
= now
;
5891 /* More magic constants... */
5893 diff
= div_u64(diff
, diffms
* 10);
5894 dev_priv
->ips
.gfx_power
= diff
;
5897 void i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
5899 struct drm_device
*dev
= dev_priv
->dev
;
5901 if (INTEL_INFO(dev
)->gen
!= 5)
5904 spin_lock_irq(&mchdev_lock
);
5906 __i915_update_gfx_val(dev_priv
);
5908 spin_unlock_irq(&mchdev_lock
);
5911 static unsigned long __i915_gfx_val(struct drm_i915_private
*dev_priv
)
5913 unsigned long t
, corr
, state1
, corr2
, state2
;
5916 assert_spin_locked(&mchdev_lock
);
5918 pxvid
= I915_READ(PXVFREQ(dev_priv
->rps
.cur_freq
));
5919 pxvid
= (pxvid
>> 24) & 0x7f;
5920 ext_v
= pvid_to_extvid(dev_priv
, pxvid
);
5924 t
= i915_mch_val(dev_priv
);
5926 /* Revel in the empirically derived constants */
5928 /* Correction factor in 1/100000 units */
5930 corr
= ((t
* 2349) + 135940);
5932 corr
= ((t
* 964) + 29317);
5934 corr
= ((t
* 301) + 1004);
5936 corr
= corr
* ((150142 * state1
) / 10000 - 78642);
5938 corr2
= (corr
* dev_priv
->ips
.corr
);
5940 state2
= (corr2
* state1
) / 10000;
5941 state2
/= 100; /* convert to mW */
5943 __i915_update_gfx_val(dev_priv
);
5945 return dev_priv
->ips
.gfx_power
+ state2
;
5948 unsigned long i915_gfx_val(struct drm_i915_private
*dev_priv
)
5950 struct drm_device
*dev
= dev_priv
->dev
;
5953 if (INTEL_INFO(dev
)->gen
!= 5)
5956 spin_lock_irq(&mchdev_lock
);
5958 val
= __i915_gfx_val(dev_priv
);
5960 spin_unlock_irq(&mchdev_lock
);
5966 * i915_read_mch_val - return value for IPS use
5968 * Calculate and return a value for the IPS driver to use when deciding whether
5969 * we have thermal and power headroom to increase CPU or GPU power budget.
5971 unsigned long i915_read_mch_val(void)
5973 struct drm_i915_private
*dev_priv
;
5974 unsigned long chipset_val
, graphics_val
, ret
= 0;
5976 spin_lock_irq(&mchdev_lock
);
5979 dev_priv
= i915_mch_dev
;
5981 chipset_val
= __i915_chipset_val(dev_priv
);
5982 graphics_val
= __i915_gfx_val(dev_priv
);
5984 ret
= chipset_val
+ graphics_val
;
5987 spin_unlock_irq(&mchdev_lock
);
5991 EXPORT_SYMBOL_GPL(i915_read_mch_val
);
5994 * i915_gpu_raise - raise GPU frequency limit
5996 * Raise the limit; IPS indicates we have thermal headroom.
5998 bool i915_gpu_raise(void)
6000 struct drm_i915_private
*dev_priv
;
6003 spin_lock_irq(&mchdev_lock
);
6004 if (!i915_mch_dev
) {
6008 dev_priv
= i915_mch_dev
;
6010 if (dev_priv
->ips
.max_delay
> dev_priv
->ips
.fmax
)
6011 dev_priv
->ips
.max_delay
--;
6014 spin_unlock_irq(&mchdev_lock
);
6018 EXPORT_SYMBOL_GPL(i915_gpu_raise
);
6021 * i915_gpu_lower - lower GPU frequency limit
6023 * IPS indicates we're close to a thermal limit, so throttle back the GPU
6024 * frequency maximum.
6026 bool i915_gpu_lower(void)
6028 struct drm_i915_private
*dev_priv
;
6031 spin_lock_irq(&mchdev_lock
);
6032 if (!i915_mch_dev
) {
6036 dev_priv
= i915_mch_dev
;
6038 if (dev_priv
->ips
.max_delay
< dev_priv
->ips
.min_delay
)
6039 dev_priv
->ips
.max_delay
++;
6042 spin_unlock_irq(&mchdev_lock
);
6046 EXPORT_SYMBOL_GPL(i915_gpu_lower
);
6049 * i915_gpu_busy - indicate GPU business to IPS
6051 * Tell the IPS driver whether or not the GPU is busy.
6053 bool i915_gpu_busy(void)
6055 struct drm_i915_private
*dev_priv
;
6056 struct intel_engine_cs
*ring
;
6060 spin_lock_irq(&mchdev_lock
);
6063 dev_priv
= i915_mch_dev
;
6065 for_each_ring(ring
, dev_priv
, i
)
6066 ret
|= !list_empty(&ring
->request_list
);
6069 spin_unlock_irq(&mchdev_lock
);
6073 EXPORT_SYMBOL_GPL(i915_gpu_busy
);
6076 * i915_gpu_turbo_disable - disable graphics turbo
6078 * Disable graphics turbo by resetting the max frequency and setting the
6079 * current frequency to the default.
6081 bool i915_gpu_turbo_disable(void)
6083 struct drm_i915_private
*dev_priv
;
6086 spin_lock_irq(&mchdev_lock
);
6087 if (!i915_mch_dev
) {
6091 dev_priv
= i915_mch_dev
;
6093 dev_priv
->ips
.max_delay
= dev_priv
->ips
.fstart
;
6095 if (!ironlake_set_drps(dev_priv
->dev
, dev_priv
->ips
.fstart
))
6099 spin_unlock_irq(&mchdev_lock
);
6103 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable
);
6106 * Tells the intel_ips driver that the i915 driver is now loaded, if
6107 * IPS got loaded first.
6109 * This awkward dance is so that neither module has to depend on the
6110 * other in order for IPS to do the appropriate communication of
6111 * GPU turbo limits to i915.
6114 ips_ping_for_i915_load(void)
6118 link
= symbol_get(ips_link_to_i915_driver
);
6121 symbol_put(ips_link_to_i915_driver
);
6125 void intel_gpu_ips_init(struct drm_i915_private
*dev_priv
)
6127 /* We only register the i915 ips part with intel-ips once everything is
6128 * set up, to avoid intel-ips sneaking in and reading bogus values. */
6129 spin_lock_irq(&mchdev_lock
);
6130 i915_mch_dev
= dev_priv
;
6131 spin_unlock_irq(&mchdev_lock
);
6133 ips_ping_for_i915_load();
6136 void intel_gpu_ips_teardown(void)
6138 spin_lock_irq(&mchdev_lock
);
6139 i915_mch_dev
= NULL
;
6140 spin_unlock_irq(&mchdev_lock
);
6143 static void intel_init_emon(struct drm_device
*dev
)
6145 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6150 /* Disable to program */
6154 /* Program energy weights for various events */
6155 I915_WRITE(SDEW
, 0x15040d00);
6156 I915_WRITE(CSIEW0
, 0x007f0000);
6157 I915_WRITE(CSIEW1
, 0x1e220004);
6158 I915_WRITE(CSIEW2
, 0x04000004);
6160 for (i
= 0; i
< 5; i
++)
6161 I915_WRITE(PEW(i
), 0);
6162 for (i
= 0; i
< 3; i
++)
6163 I915_WRITE(DEW(i
), 0);
6165 /* Program P-state weights to account for frequency power adjustment */
6166 for (i
= 0; i
< 16; i
++) {
6167 u32 pxvidfreq
= I915_READ(PXVFREQ(i
));
6168 unsigned long freq
= intel_pxfreq(pxvidfreq
);
6169 unsigned long vid
= (pxvidfreq
& PXVFREQ_PX_MASK
) >>
6174 val
*= (freq
/ 1000);
6176 val
/= (127*127*900);
6178 DRM_ERROR("bad pxval: %ld\n", val
);
6181 /* Render standby states get 0 weight */
6185 for (i
= 0; i
< 4; i
++) {
6186 u32 val
= (pxw
[i
*4] << 24) | (pxw
[(i
*4)+1] << 16) |
6187 (pxw
[(i
*4)+2] << 8) | (pxw
[(i
*4)+3]);
6188 I915_WRITE(PXW(i
), val
);
6191 /* Adjust magic regs to magic values (more experimental results) */
6192 I915_WRITE(OGW0
, 0);
6193 I915_WRITE(OGW1
, 0);
6194 I915_WRITE(EG0
, 0x00007f00);
6195 I915_WRITE(EG1
, 0x0000000e);
6196 I915_WRITE(EG2
, 0x000e0000);
6197 I915_WRITE(EG3
, 0x68000300);
6198 I915_WRITE(EG4
, 0x42000000);
6199 I915_WRITE(EG5
, 0x00140031);
6203 for (i
= 0; i
< 8; i
++)
6204 I915_WRITE(PXWL(i
), 0);
6206 /* Enable PMON + select events */
6207 I915_WRITE(ECR
, 0x80000019);
6209 lcfuse
= I915_READ(LCFUSE02
);
6211 dev_priv
->ips
.corr
= (lcfuse
& LCFUSE_HIV_MASK
);
6214 void intel_init_gt_powersave(struct drm_device
*dev
)
6216 i915
.enable_rc6
= sanitize_rc6_option(dev
, i915
.enable_rc6
);
6218 if (IS_CHERRYVIEW(dev
))
6219 cherryview_init_gt_powersave(dev
);
6220 else if (IS_VALLEYVIEW(dev
))
6221 valleyview_init_gt_powersave(dev
);
6224 void intel_cleanup_gt_powersave(struct drm_device
*dev
)
6226 if (IS_CHERRYVIEW(dev
))
6228 else if (IS_VALLEYVIEW(dev
))
6229 valleyview_cleanup_gt_powersave(dev
);
6232 static void gen6_suspend_rps(struct drm_device
*dev
)
6234 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6236 flush_delayed_work(&dev_priv
->rps
.delayed_resume_work
);
6238 gen6_disable_rps_interrupts(dev
);
6242 * intel_suspend_gt_powersave - suspend PM work and helper threads
6245 * We don't want to disable RC6 or other features here, we just want
6246 * to make sure any work we've queued has finished and won't bother
6247 * us while we're suspended.
6249 void intel_suspend_gt_powersave(struct drm_device
*dev
)
6251 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6253 if (INTEL_INFO(dev
)->gen
< 6)
6256 gen6_suspend_rps(dev
);
6258 /* Force GPU to min freq during suspend */
6259 gen6_rps_idle(dev_priv
);
6262 void intel_disable_gt_powersave(struct drm_device
*dev
)
6264 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6266 if (IS_IRONLAKE_M(dev
)) {
6267 ironlake_disable_drps(dev
);
6268 } else if (INTEL_INFO(dev
)->gen
>= 6) {
6269 intel_suspend_gt_powersave(dev
);
6271 mutex_lock(&dev_priv
->rps
.hw_lock
);
6272 if (INTEL_INFO(dev
)->gen
>= 9)
6273 gen9_disable_rps(dev
);
6274 else if (IS_CHERRYVIEW(dev
))
6275 cherryview_disable_rps(dev
);
6276 else if (IS_VALLEYVIEW(dev
))
6277 valleyview_disable_rps(dev
);
6279 gen6_disable_rps(dev
);
6281 dev_priv
->rps
.enabled
= false;
6282 mutex_unlock(&dev_priv
->rps
.hw_lock
);
6286 static void intel_gen6_powersave_work(struct work_struct
*work
)
6288 struct drm_i915_private
*dev_priv
=
6289 container_of(work
, struct drm_i915_private
,
6290 rps
.delayed_resume_work
.work
);
6291 struct drm_device
*dev
= dev_priv
->dev
;
6293 mutex_lock(&dev_priv
->rps
.hw_lock
);
6295 gen6_reset_rps_interrupts(dev
);
6297 if (IS_CHERRYVIEW(dev
)) {
6298 cherryview_enable_rps(dev
);
6299 } else if (IS_VALLEYVIEW(dev
)) {
6300 valleyview_enable_rps(dev
);
6301 } else if (INTEL_INFO(dev
)->gen
>= 9) {
6302 gen9_enable_rc6(dev
);
6303 gen9_enable_rps(dev
);
6304 if (IS_SKYLAKE(dev
))
6305 __gen6_update_ring_freq(dev
);
6306 } else if (IS_BROADWELL(dev
)) {
6307 gen8_enable_rps(dev
);
6308 __gen6_update_ring_freq(dev
);
6310 gen6_enable_rps(dev
);
6311 __gen6_update_ring_freq(dev
);
6314 WARN_ON(dev_priv
->rps
.max_freq
< dev_priv
->rps
.min_freq
);
6315 WARN_ON(dev_priv
->rps
.idle_freq
> dev_priv
->rps
.max_freq
);
6317 WARN_ON(dev_priv
->rps
.efficient_freq
< dev_priv
->rps
.min_freq
);
6318 WARN_ON(dev_priv
->rps
.efficient_freq
> dev_priv
->rps
.max_freq
);
6320 dev_priv
->rps
.enabled
= true;
6322 gen6_enable_rps_interrupts(dev
);
6324 mutex_unlock(&dev_priv
->rps
.hw_lock
);
6326 intel_runtime_pm_put(dev_priv
);
6329 void intel_enable_gt_powersave(struct drm_device
*dev
)
6331 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6333 /* Powersaving is controlled by the host when inside a VM */
6334 if (intel_vgpu_active(dev
))
6337 if (IS_IRONLAKE_M(dev
)) {
6338 mutex_lock(&dev
->struct_mutex
);
6339 ironlake_enable_drps(dev
);
6340 intel_init_emon(dev
);
6341 mutex_unlock(&dev
->struct_mutex
);
6342 } else if (INTEL_INFO(dev
)->gen
>= 6) {
6344 * PCU communication is slow and this doesn't need to be
6345 * done at any specific time, so do this out of our fast path
6346 * to make resume and init faster.
6348 * We depend on the HW RC6 power context save/restore
6349 * mechanism when entering D3 through runtime PM suspend. So
6350 * disable RPM until RPS/RC6 is properly setup. We can only
6351 * get here via the driver load/system resume/runtime resume
6352 * paths, so the _noresume version is enough (and in case of
6353 * runtime resume it's necessary).
6355 if (schedule_delayed_work(&dev_priv
->rps
.delayed_resume_work
,
6356 round_jiffies_up_relative(HZ
)))
6357 intel_runtime_pm_get_noresume(dev_priv
);
6361 void intel_reset_gt_powersave(struct drm_device
*dev
)
6363 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6365 if (INTEL_INFO(dev
)->gen
< 6)
6368 gen6_suspend_rps(dev
);
6369 dev_priv
->rps
.enabled
= false;
6372 static void ibx_init_clock_gating(struct drm_device
*dev
)
6374 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6377 * On Ibex Peak and Cougar Point, we need to disable clock
6378 * gating for the panel power sequencer or it will fail to
6379 * start up when no ports are active.
6381 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
);
6384 static void g4x_disable_trickle_feed(struct drm_device
*dev
)
6386 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6389 for_each_pipe(dev_priv
, pipe
) {
6390 I915_WRITE(DSPCNTR(pipe
),
6391 I915_READ(DSPCNTR(pipe
)) |
6392 DISPPLANE_TRICKLE_FEED_DISABLE
);
6394 I915_WRITE(DSPSURF(pipe
), I915_READ(DSPSURF(pipe
)));
6395 POSTING_READ(DSPSURF(pipe
));
6399 static void ilk_init_lp_watermarks(struct drm_device
*dev
)
6401 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6403 I915_WRITE(WM3_LP_ILK
, I915_READ(WM3_LP_ILK
) & ~WM1_LP_SR_EN
);
6404 I915_WRITE(WM2_LP_ILK
, I915_READ(WM2_LP_ILK
) & ~WM1_LP_SR_EN
);
6405 I915_WRITE(WM1_LP_ILK
, I915_READ(WM1_LP_ILK
) & ~WM1_LP_SR_EN
);
6408 * Don't touch WM1S_LP_EN here.
6409 * Doing so could cause underruns.
6413 static void ironlake_init_clock_gating(struct drm_device
*dev
)
6415 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6416 uint32_t dspclk_gate
= ILK_VRHUNIT_CLOCK_GATE_DISABLE
;
6420 * WaFbcDisableDpfcClockGating:ilk
6422 dspclk_gate
|= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE
|
6423 ILK_DPFCUNIT_CLOCK_GATE_DISABLE
|
6424 ILK_DPFDUNIT_CLOCK_GATE_ENABLE
;
6426 I915_WRITE(PCH_3DCGDIS0
,
6427 MARIUNIT_CLOCK_GATE_DISABLE
|
6428 SVSMUNIT_CLOCK_GATE_DISABLE
);
6429 I915_WRITE(PCH_3DCGDIS1
,
6430 VFMUNIT_CLOCK_GATE_DISABLE
);
6433 * According to the spec the following bits should be set in
6434 * order to enable memory self-refresh
6435 * The bit 22/21 of 0x42004
6436 * The bit 5 of 0x42020
6437 * The bit 15 of 0x45000
6439 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6440 (I915_READ(ILK_DISPLAY_CHICKEN2
) |
6441 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
));
6442 dspclk_gate
|= ILK_DPARBUNIT_CLOCK_GATE_ENABLE
;
6443 I915_WRITE(DISP_ARB_CTL
,
6444 (I915_READ(DISP_ARB_CTL
) |
6447 ilk_init_lp_watermarks(dev
);
6450 * Based on the document from hardware guys the following bits
6451 * should be set unconditionally in order to enable FBC.
6452 * The bit 22 of 0x42000
6453 * The bit 22 of 0x42004
6454 * The bit 7,8,9 of 0x42020.
6456 if (IS_IRONLAKE_M(dev
)) {
6457 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6458 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
6459 I915_READ(ILK_DISPLAY_CHICKEN1
) |
6461 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6462 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6466 I915_WRITE(ILK_DSPCLK_GATE_D
, dspclk_gate
);
6468 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6469 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6470 ILK_ELPIN_409_SELECT
);
6471 I915_WRITE(_3D_CHICKEN2
,
6472 _3D_CHICKEN2_WM_READ_PIPELINED
<< 16 |
6473 _3D_CHICKEN2_WM_READ_PIPELINED
);
6475 /* WaDisableRenderCachePipelinedFlush:ilk */
6476 I915_WRITE(CACHE_MODE_0
,
6477 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE
));
6479 /* WaDisable_RenderCache_OperationalFlush:ilk */
6480 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6482 g4x_disable_trickle_feed(dev
);
6484 ibx_init_clock_gating(dev
);
6487 static void cpt_init_clock_gating(struct drm_device
*dev
)
6489 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6494 * On Ibex Peak and Cougar Point, we need to disable clock
6495 * gating for the panel power sequencer or it will fail to
6496 * start up when no ports are active.
6498 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
|
6499 PCH_DPLUNIT_CLOCK_GATE_DISABLE
|
6500 PCH_CPUNIT_CLOCK_GATE_DISABLE
);
6501 I915_WRITE(SOUTH_CHICKEN2
, I915_READ(SOUTH_CHICKEN2
) |
6502 DPLS_EDP_PPS_FIX_DIS
);
6503 /* The below fixes the weird display corruption, a few pixels shifted
6504 * downward, on (only) LVDS of some HP laptops with IVY.
6506 for_each_pipe(dev_priv
, pipe
) {
6507 val
= I915_READ(TRANS_CHICKEN2(pipe
));
6508 val
|= TRANS_CHICKEN2_TIMING_OVERRIDE
;
6509 val
&= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED
;
6510 if (dev_priv
->vbt
.fdi_rx_polarity_inverted
)
6511 val
|= TRANS_CHICKEN2_FDI_POLARITY_REVERSED
;
6512 val
&= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK
;
6513 val
&= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER
;
6514 val
&= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH
;
6515 I915_WRITE(TRANS_CHICKEN2(pipe
), val
);
6517 /* WADP0ClockGatingDisable */
6518 for_each_pipe(dev_priv
, pipe
) {
6519 I915_WRITE(TRANS_CHICKEN1(pipe
),
6520 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE
);
6524 static void gen6_check_mch_setup(struct drm_device
*dev
)
6526 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6529 tmp
= I915_READ(MCH_SSKPD
);
6530 if ((tmp
& MCH_SSKPD_WM0_MASK
) != MCH_SSKPD_WM0_VAL
)
6531 DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n",
6535 static void gen6_init_clock_gating(struct drm_device
*dev
)
6537 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6538 uint32_t dspclk_gate
= ILK_VRHUNIT_CLOCK_GATE_DISABLE
;
6540 I915_WRITE(ILK_DSPCLK_GATE_D
, dspclk_gate
);
6542 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6543 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6544 ILK_ELPIN_409_SELECT
);
6546 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
6547 I915_WRITE(_3D_CHICKEN
,
6548 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB
));
6550 /* WaDisable_RenderCache_OperationalFlush:snb */
6551 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6554 * BSpec recoomends 8x4 when MSAA is used,
6555 * however in practice 16x4 seems fastest.
6557 * Note that PS/WM thread counts depend on the WIZ hashing
6558 * disable bit, which we don't touch here, but it's good
6559 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6561 I915_WRITE(GEN6_GT_MODE
,
6562 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6564 ilk_init_lp_watermarks(dev
);
6566 I915_WRITE(CACHE_MODE_0
,
6567 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB
));
6569 I915_WRITE(GEN6_UCGCTL1
,
6570 I915_READ(GEN6_UCGCTL1
) |
6571 GEN6_BLBUNIT_CLOCK_GATE_DISABLE
|
6572 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
6574 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
6575 * gating disable must be set. Failure to set it results in
6576 * flickering pixels due to Z write ordering failures after
6577 * some amount of runtime in the Mesa "fire" demo, and Unigine
6578 * Sanctuary and Tropics, and apparently anything else with
6579 * alpha test or pixel discard.
6581 * According to the spec, bit 11 (RCCUNIT) must also be set,
6582 * but we didn't debug actual testcases to find it out.
6584 * WaDisableRCCUnitClockGating:snb
6585 * WaDisableRCPBUnitClockGating:snb
6587 I915_WRITE(GEN6_UCGCTL2
,
6588 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE
|
6589 GEN6_RCCUNIT_CLOCK_GATE_DISABLE
);
6591 /* WaStripsFansDisableFastClipPerformanceFix:snb */
6592 I915_WRITE(_3D_CHICKEN3
,
6593 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL
));
6597 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
6598 * 3DSTATE_SF number of SF output attributes is more than 16."
6600 I915_WRITE(_3D_CHICKEN3
,
6601 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH
));
6604 * According to the spec the following bits should be
6605 * set in order to enable memory self-refresh and fbc:
6606 * The bit21 and bit22 of 0x42000
6607 * The bit21 and bit22 of 0x42004
6608 * The bit5 and bit7 of 0x42020
6609 * The bit14 of 0x70180
6610 * The bit14 of 0x71180
6612 * WaFbcAsynchFlipDisableFbcQueue:snb
6614 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
6615 I915_READ(ILK_DISPLAY_CHICKEN1
) |
6616 ILK_FBCQ_DIS
| ILK_PABSTRETCH_DIS
);
6617 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6618 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6619 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
);
6620 I915_WRITE(ILK_DSPCLK_GATE_D
,
6621 I915_READ(ILK_DSPCLK_GATE_D
) |
6622 ILK_DPARBUNIT_CLOCK_GATE_ENABLE
|
6623 ILK_DPFDUNIT_CLOCK_GATE_ENABLE
);
6625 g4x_disable_trickle_feed(dev
);
6627 cpt_init_clock_gating(dev
);
6629 gen6_check_mch_setup(dev
);
6632 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private
*dev_priv
)
6634 uint32_t reg
= I915_READ(GEN7_FF_THREAD_MODE
);
6637 * WaVSThreadDispatchOverride:ivb,vlv
6639 * This actually overrides the dispatch
6640 * mode for all thread types.
6642 reg
&= ~GEN7_FF_SCHED_MASK
;
6643 reg
|= GEN7_FF_TS_SCHED_HW
;
6644 reg
|= GEN7_FF_VS_SCHED_HW
;
6645 reg
|= GEN7_FF_DS_SCHED_HW
;
6647 I915_WRITE(GEN7_FF_THREAD_MODE
, reg
);
6650 static void lpt_init_clock_gating(struct drm_device
*dev
)
6652 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6655 * TODO: this bit should only be enabled when really needed, then
6656 * disabled when not needed anymore in order to save power.
6658 if (HAS_PCH_LPT_LP(dev
))
6659 I915_WRITE(SOUTH_DSPCLK_GATE_D
,
6660 I915_READ(SOUTH_DSPCLK_GATE_D
) |
6661 PCH_LP_PARTITION_LEVEL_DISABLE
);
6663 /* WADPOClockGatingDisable:hsw */
6664 I915_WRITE(_TRANSA_CHICKEN1
,
6665 I915_READ(_TRANSA_CHICKEN1
) |
6666 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE
);
6669 static void lpt_suspend_hw(struct drm_device
*dev
)
6671 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6673 if (HAS_PCH_LPT_LP(dev
)) {
6674 uint32_t val
= I915_READ(SOUTH_DSPCLK_GATE_D
);
6676 val
&= ~PCH_LP_PARTITION_LEVEL_DISABLE
;
6677 I915_WRITE(SOUTH_DSPCLK_GATE_D
, val
);
6681 static void broadwell_init_clock_gating(struct drm_device
*dev
)
6683 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6687 ilk_init_lp_watermarks(dev
);
6689 /* WaSwitchSolVfFArbitrationPriority:bdw */
6690 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) | HSW_ECOCHK_ARB_PRIO_SOL
);
6692 /* WaPsrDPAMaskVBlankInSRD:bdw */
6693 I915_WRITE(CHICKEN_PAR1_1
,
6694 I915_READ(CHICKEN_PAR1_1
) | DPA_MASK_VBLANK_SRD
);
6696 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
6697 for_each_pipe(dev_priv
, pipe
) {
6698 I915_WRITE(CHICKEN_PIPESL_1(pipe
),
6699 I915_READ(CHICKEN_PIPESL_1(pipe
)) |
6700 BDW_DPRS_MASK_VBLANK_SRD
);
6703 /* WaVSRefCountFullforceMissDisable:bdw */
6704 /* WaDSRefCountFullforceMissDisable:bdw */
6705 I915_WRITE(GEN7_FF_THREAD_MODE
,
6706 I915_READ(GEN7_FF_THREAD_MODE
) &
6707 ~(GEN8_FF_DS_REF_CNT_FFME
| GEN7_FF_VS_REF_CNT_FFME
));
6709 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
6710 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE
));
6712 /* WaDisableSDEUnitClockGating:bdw */
6713 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
6714 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
6717 * WaProgramL3SqcReg1Default:bdw
6718 * WaTempDisableDOPClkGating:bdw
6720 misccpctl
= I915_READ(GEN7_MISCCPCTL
);
6721 I915_WRITE(GEN7_MISCCPCTL
, misccpctl
& ~GEN7_DOP_CLOCK_GATE_ENABLE
);
6722 I915_WRITE(GEN8_L3SQCREG1
, BDW_WA_L3SQCREG1_DEFAULT
);
6723 I915_WRITE(GEN7_MISCCPCTL
, misccpctl
);
6726 * WaGttCachingOffByDefault:bdw
6727 * GTT cache may not work with big pages, so if those
6728 * are ever enabled GTT cache may need to be disabled.
6730 I915_WRITE(HSW_GTT_CACHE_EN
, GTT_CACHE_EN_ALL
);
6732 lpt_init_clock_gating(dev
);
6735 static void haswell_init_clock_gating(struct drm_device
*dev
)
6737 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6739 ilk_init_lp_watermarks(dev
);
6741 /* L3 caching of data atomics doesn't work -- disable it. */
6742 I915_WRITE(HSW_SCRATCH1
, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE
);
6743 I915_WRITE(HSW_ROW_CHICKEN3
,
6744 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE
));
6746 /* This is required by WaCatErrorRejectionIssue:hsw */
6747 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
6748 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
6749 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
6751 /* WaVSRefCountFullforceMissDisable:hsw */
6752 I915_WRITE(GEN7_FF_THREAD_MODE
,
6753 I915_READ(GEN7_FF_THREAD_MODE
) & ~GEN7_FF_VS_REF_CNT_FFME
);
6755 /* WaDisable_RenderCache_OperationalFlush:hsw */
6756 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6758 /* enable HiZ Raw Stall Optimization */
6759 I915_WRITE(CACHE_MODE_0_GEN7
,
6760 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE
));
6762 /* WaDisable4x2SubspanOptimization:hsw */
6763 I915_WRITE(CACHE_MODE_1
,
6764 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
6767 * BSpec recommends 8x4 when MSAA is used,
6768 * however in practice 16x4 seems fastest.
6770 * Note that PS/WM thread counts depend on the WIZ hashing
6771 * disable bit, which we don't touch here, but it's good
6772 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6774 I915_WRITE(GEN7_GT_MODE
,
6775 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6777 /* WaSampleCChickenBitEnable:hsw */
6778 I915_WRITE(HALF_SLICE_CHICKEN3
,
6779 _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE
));
6781 /* WaSwitchSolVfFArbitrationPriority:hsw */
6782 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) | HSW_ECOCHK_ARB_PRIO_SOL
);
6784 /* WaRsPkgCStateDisplayPMReq:hsw */
6785 I915_WRITE(CHICKEN_PAR1_1
,
6786 I915_READ(CHICKEN_PAR1_1
) | FORCE_ARB_IDLE_PLANES
);
6788 lpt_init_clock_gating(dev
);
6791 static void ivybridge_init_clock_gating(struct drm_device
*dev
)
6793 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6796 ilk_init_lp_watermarks(dev
);
6798 I915_WRITE(ILK_DSPCLK_GATE_D
, ILK_VRHUNIT_CLOCK_GATE_DISABLE
);
6800 /* WaDisableEarlyCull:ivb */
6801 I915_WRITE(_3D_CHICKEN3
,
6802 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL
));
6804 /* WaDisableBackToBackFlipFix:ivb */
6805 I915_WRITE(IVB_CHICKEN3
,
6806 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
6807 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
6809 /* WaDisablePSDDualDispatchEnable:ivb */
6810 if (IS_IVB_GT1(dev
))
6811 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
6812 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
));
6814 /* WaDisable_RenderCache_OperationalFlush:ivb */
6815 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6817 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6818 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1
,
6819 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
6821 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6822 I915_WRITE(GEN7_L3CNTLREG1
,
6823 GEN7_WA_FOR_GEN7_L3_CONTROL
);
6824 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER
,
6825 GEN7_WA_L3_CHICKEN_MODE
);
6826 if (IS_IVB_GT1(dev
))
6827 I915_WRITE(GEN7_ROW_CHICKEN2
,
6828 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6830 /* must write both registers */
6831 I915_WRITE(GEN7_ROW_CHICKEN2
,
6832 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6833 I915_WRITE(GEN7_ROW_CHICKEN2_GT2
,
6834 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6837 /* WaForceL3Serialization:ivb */
6838 I915_WRITE(GEN7_L3SQCREG4
, I915_READ(GEN7_L3SQCREG4
) &
6839 ~L3SQ_URB_READ_CAM_MATCH_DISABLE
);
6842 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
6843 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
6845 I915_WRITE(GEN6_UCGCTL2
,
6846 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
6848 /* This is required by WaCatErrorRejectionIssue:ivb */
6849 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
6850 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
6851 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
6853 g4x_disable_trickle_feed(dev
);
6855 gen7_setup_fixed_func_scheduler(dev_priv
);
6857 if (0) { /* causes HiZ corruption on ivb:gt1 */
6858 /* enable HiZ Raw Stall Optimization */
6859 I915_WRITE(CACHE_MODE_0_GEN7
,
6860 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE
));
6863 /* WaDisable4x2SubspanOptimization:ivb */
6864 I915_WRITE(CACHE_MODE_1
,
6865 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
6868 * BSpec recommends 8x4 when MSAA is used,
6869 * however in practice 16x4 seems fastest.
6871 * Note that PS/WM thread counts depend on the WIZ hashing
6872 * disable bit, which we don't touch here, but it's good
6873 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6875 I915_WRITE(GEN7_GT_MODE
,
6876 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6878 snpcr
= I915_READ(GEN6_MBCUNIT_SNPCR
);
6879 snpcr
&= ~GEN6_MBC_SNPCR_MASK
;
6880 snpcr
|= GEN6_MBC_SNPCR_MED
;
6881 I915_WRITE(GEN6_MBCUNIT_SNPCR
, snpcr
);
6883 if (!HAS_PCH_NOP(dev
))
6884 cpt_init_clock_gating(dev
);
6886 gen6_check_mch_setup(dev
);
6889 static void vlv_init_display_clock_gating(struct drm_i915_private
*dev_priv
)
6891 I915_WRITE(DSPCLK_GATE_D
, VRHUNIT_CLOCK_GATE_DISABLE
);
6894 * Disable trickle feed and enable pnd deadline calculation
6896 I915_WRITE(MI_ARB_VLV
, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
);
6897 I915_WRITE(CBR1_VLV
, 0);
6900 static void valleyview_init_clock_gating(struct drm_device
*dev
)
6902 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6904 vlv_init_display_clock_gating(dev_priv
);
6906 /* WaDisableEarlyCull:vlv */
6907 I915_WRITE(_3D_CHICKEN3
,
6908 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL
));
6910 /* WaDisableBackToBackFlipFix:vlv */
6911 I915_WRITE(IVB_CHICKEN3
,
6912 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
6913 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
6915 /* WaPsdDispatchEnable:vlv */
6916 /* WaDisablePSDDualDispatchEnable:vlv */
6917 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
6918 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP
|
6919 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
));
6921 /* WaDisable_RenderCache_OperationalFlush:vlv */
6922 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6924 /* WaForceL3Serialization:vlv */
6925 I915_WRITE(GEN7_L3SQCREG4
, I915_READ(GEN7_L3SQCREG4
) &
6926 ~L3SQ_URB_READ_CAM_MATCH_DISABLE
);
6928 /* WaDisableDopClockGating:vlv */
6929 I915_WRITE(GEN7_ROW_CHICKEN2
,
6930 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6932 /* This is required by WaCatErrorRejectionIssue:vlv */
6933 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
6934 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
6935 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
6937 gen7_setup_fixed_func_scheduler(dev_priv
);
6940 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
6941 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
6943 I915_WRITE(GEN6_UCGCTL2
,
6944 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
6946 /* WaDisableL3Bank2xClockGate:vlv
6947 * Disabling L3 clock gating- MMIO 940c[25] = 1
6948 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
6949 I915_WRITE(GEN7_UCGCTL4
,
6950 I915_READ(GEN7_UCGCTL4
) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE
);
6953 * BSpec says this must be set, even though
6954 * WaDisable4x2SubspanOptimization isn't listed for VLV.
6956 I915_WRITE(CACHE_MODE_1
,
6957 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
6960 * BSpec recommends 8x4 when MSAA is used,
6961 * however in practice 16x4 seems fastest.
6963 * Note that PS/WM thread counts depend on the WIZ hashing
6964 * disable bit, which we don't touch here, but it's good
6965 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6967 I915_WRITE(GEN7_GT_MODE
,
6968 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6971 * WaIncreaseL3CreditsForVLVB0:vlv
6972 * This is the hardware default actually.
6974 I915_WRITE(GEN7_L3SQCREG1
, VLV_B0_WA_L3SQCREG1_VALUE
);
6977 * WaDisableVLVClockGating_VBIIssue:vlv
6978 * Disable clock gating on th GCFG unit to prevent a delay
6979 * in the reporting of vblank events.
6981 I915_WRITE(VLV_GUNIT_CLOCK_GATE
, GCFG_DIS
);
6984 static void cherryview_init_clock_gating(struct drm_device
*dev
)
6986 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6988 vlv_init_display_clock_gating(dev_priv
);
6990 /* WaVSRefCountFullforceMissDisable:chv */
6991 /* WaDSRefCountFullforceMissDisable:chv */
6992 I915_WRITE(GEN7_FF_THREAD_MODE
,
6993 I915_READ(GEN7_FF_THREAD_MODE
) &
6994 ~(GEN8_FF_DS_REF_CNT_FFME
| GEN7_FF_VS_REF_CNT_FFME
));
6996 /* WaDisableSemaphoreAndSyncFlipWait:chv */
6997 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
6998 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE
));
7000 /* WaDisableCSUnitClockGating:chv */
7001 I915_WRITE(GEN6_UCGCTL1
, I915_READ(GEN6_UCGCTL1
) |
7002 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
7004 /* WaDisableSDEUnitClockGating:chv */
7005 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
7006 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
7009 * GTT cache may not work with big pages, so if those
7010 * are ever enabled GTT cache may need to be disabled.
7012 I915_WRITE(HSW_GTT_CACHE_EN
, GTT_CACHE_EN_ALL
);
7015 static void g4x_init_clock_gating(struct drm_device
*dev
)
7017 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7018 uint32_t dspclk_gate
;
7020 I915_WRITE(RENCLK_GATE_D1
, 0);
7021 I915_WRITE(RENCLK_GATE_D2
, VF_UNIT_CLOCK_GATE_DISABLE
|
7022 GS_UNIT_CLOCK_GATE_DISABLE
|
7023 CL_UNIT_CLOCK_GATE_DISABLE
);
7024 I915_WRITE(RAMCLK_GATE_D
, 0);
7025 dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
|
7026 OVRUNIT_CLOCK_GATE_DISABLE
|
7027 OVCUNIT_CLOCK_GATE_DISABLE
;
7029 dspclk_gate
|= DSSUNIT_CLOCK_GATE_DISABLE
;
7030 I915_WRITE(DSPCLK_GATE_D
, dspclk_gate
);
7032 /* WaDisableRenderCachePipelinedFlush */
7033 I915_WRITE(CACHE_MODE_0
,
7034 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE
));
7036 /* WaDisable_RenderCache_OperationalFlush:g4x */
7037 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
7039 g4x_disable_trickle_feed(dev
);
7042 static void crestline_init_clock_gating(struct drm_device
*dev
)
7044 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7046 I915_WRITE(RENCLK_GATE_D1
, I965_RCC_CLOCK_GATE_DISABLE
);
7047 I915_WRITE(RENCLK_GATE_D2
, 0);
7048 I915_WRITE(DSPCLK_GATE_D
, 0);
7049 I915_WRITE(RAMCLK_GATE_D
, 0);
7050 I915_WRITE16(DEUC
, 0);
7051 I915_WRITE(MI_ARB_STATE
,
7052 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
7054 /* WaDisable_RenderCache_OperationalFlush:gen4 */
7055 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
7058 static void broadwater_init_clock_gating(struct drm_device
*dev
)
7060 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7062 I915_WRITE(RENCLK_GATE_D1
, I965_RCZ_CLOCK_GATE_DISABLE
|
7063 I965_RCC_CLOCK_GATE_DISABLE
|
7064 I965_RCPB_CLOCK_GATE_DISABLE
|
7065 I965_ISC_CLOCK_GATE_DISABLE
|
7066 I965_FBC_CLOCK_GATE_DISABLE
);
7067 I915_WRITE(RENCLK_GATE_D2
, 0);
7068 I915_WRITE(MI_ARB_STATE
,
7069 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
7071 /* WaDisable_RenderCache_OperationalFlush:gen4 */
7072 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
7075 static void gen3_init_clock_gating(struct drm_device
*dev
)
7077 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7078 u32 dstate
= I915_READ(D_STATE
);
7080 dstate
|= DSTATE_PLL_D3_OFF
| DSTATE_GFX_CLOCK_GATING
|
7081 DSTATE_DOT_CLOCK_GATING
;
7082 I915_WRITE(D_STATE
, dstate
);
7084 if (IS_PINEVIEW(dev
))
7085 I915_WRITE(ECOSKPD
, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY
));
7087 /* IIR "flip pending" means done if this bit is set */
7088 I915_WRITE(ECOSKPD
, _MASKED_BIT_DISABLE(ECO_FLIP_DONE
));
7090 /* interrupts should cause a wake up from C3 */
7091 I915_WRITE(INSTPM
, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN
));
7093 /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
7094 I915_WRITE(MI_ARB_STATE
, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE
));
7096 I915_WRITE(MI_ARB_STATE
,
7097 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
7100 static void i85x_init_clock_gating(struct drm_device
*dev
)
7102 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7104 I915_WRITE(RENCLK_GATE_D1
, SV_CLOCK_GATE_DISABLE
);
7106 /* interrupts should cause a wake up from C3 */
7107 I915_WRITE(MI_STATE
, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN
) |
7108 _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE
));
7110 I915_WRITE(MEM_MODE
,
7111 _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE
));
7114 static void i830_init_clock_gating(struct drm_device
*dev
)
7116 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7118 I915_WRITE(DSPCLK_GATE_D
, OVRUNIT_CLOCK_GATE_DISABLE
);
7120 I915_WRITE(MEM_MODE
,
7121 _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE
) |
7122 _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE
));
7125 void intel_init_clock_gating(struct drm_device
*dev
)
7127 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7129 if (dev_priv
->display
.init_clock_gating
)
7130 dev_priv
->display
.init_clock_gating(dev
);
7133 void intel_suspend_hw(struct drm_device
*dev
)
7135 if (HAS_PCH_LPT(dev
))
7136 lpt_suspend_hw(dev
);
7139 /* Set up chip specific power management-related functions */
7140 void intel_init_pm(struct drm_device
*dev
)
7142 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7144 intel_fbc_init(dev_priv
);
7147 if (IS_PINEVIEW(dev
))
7148 i915_pineview_get_mem_freq(dev
);
7149 else if (IS_GEN5(dev
))
7150 i915_ironlake_get_mem_freq(dev
);
7152 /* For FIFO watermark updates */
7153 if (INTEL_INFO(dev
)->gen
>= 9) {
7154 skl_setup_wm_latency(dev
);
7156 if (IS_BROXTON(dev
))
7157 dev_priv
->display
.init_clock_gating
=
7158 bxt_init_clock_gating
;
7159 else if (IS_SKYLAKE(dev
))
7160 dev_priv
->display
.init_clock_gating
=
7161 skl_init_clock_gating
;
7162 dev_priv
->display
.update_wm
= skl_update_wm
;
7163 dev_priv
->display
.update_sprite_wm
= skl_update_sprite_wm
;
7164 } else if (HAS_PCH_SPLIT(dev
)) {
7165 ilk_setup_wm_latency(dev
);
7167 if ((IS_GEN5(dev
) && dev_priv
->wm
.pri_latency
[1] &&
7168 dev_priv
->wm
.spr_latency
[1] && dev_priv
->wm
.cur_latency
[1]) ||
7169 (!IS_GEN5(dev
) && dev_priv
->wm
.pri_latency
[0] &&
7170 dev_priv
->wm
.spr_latency
[0] && dev_priv
->wm
.cur_latency
[0])) {
7171 dev_priv
->display
.update_wm
= ilk_update_wm
;
7172 dev_priv
->display
.update_sprite_wm
= ilk_update_sprite_wm
;
7174 DRM_DEBUG_KMS("Failed to read display plane latency. "
7179 dev_priv
->display
.init_clock_gating
= ironlake_init_clock_gating
;
7180 else if (IS_GEN6(dev
))
7181 dev_priv
->display
.init_clock_gating
= gen6_init_clock_gating
;
7182 else if (IS_IVYBRIDGE(dev
))
7183 dev_priv
->display
.init_clock_gating
= ivybridge_init_clock_gating
;
7184 else if (IS_HASWELL(dev
))
7185 dev_priv
->display
.init_clock_gating
= haswell_init_clock_gating
;
7186 else if (INTEL_INFO(dev
)->gen
== 8)
7187 dev_priv
->display
.init_clock_gating
= broadwell_init_clock_gating
;
7188 } else if (IS_CHERRYVIEW(dev
)) {
7189 vlv_setup_wm_latency(dev
);
7191 dev_priv
->display
.update_wm
= vlv_update_wm
;
7192 dev_priv
->display
.init_clock_gating
=
7193 cherryview_init_clock_gating
;
7194 } else if (IS_VALLEYVIEW(dev
)) {
7195 vlv_setup_wm_latency(dev
);
7197 dev_priv
->display
.update_wm
= vlv_update_wm
;
7198 dev_priv
->display
.init_clock_gating
=
7199 valleyview_init_clock_gating
;
7200 } else if (IS_PINEVIEW(dev
)) {
7201 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev
),
7204 dev_priv
->mem_freq
)) {
7205 DRM_INFO("failed to find known CxSR latency "
7206 "(found ddr%s fsb freq %d, mem freq %d), "
7208 (dev_priv
->is_ddr3
== 1) ? "3" : "2",
7209 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
7210 /* Disable CxSR and never update its watermark again */
7211 intel_set_memory_cxsr(dev_priv
, false);
7212 dev_priv
->display
.update_wm
= NULL
;
7214 dev_priv
->display
.update_wm
= pineview_update_wm
;
7215 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
7216 } else if (IS_G4X(dev
)) {
7217 dev_priv
->display
.update_wm
= g4x_update_wm
;
7218 dev_priv
->display
.init_clock_gating
= g4x_init_clock_gating
;
7219 } else if (IS_GEN4(dev
)) {
7220 dev_priv
->display
.update_wm
= i965_update_wm
;
7221 if (IS_CRESTLINE(dev
))
7222 dev_priv
->display
.init_clock_gating
= crestline_init_clock_gating
;
7223 else if (IS_BROADWATER(dev
))
7224 dev_priv
->display
.init_clock_gating
= broadwater_init_clock_gating
;
7225 } else if (IS_GEN3(dev
)) {
7226 dev_priv
->display
.update_wm
= i9xx_update_wm
;
7227 dev_priv
->display
.get_fifo_size
= i9xx_get_fifo_size
;
7228 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
7229 } else if (IS_GEN2(dev
)) {
7230 if (INTEL_INFO(dev
)->num_pipes
== 1) {
7231 dev_priv
->display
.update_wm
= i845_update_wm
;
7232 dev_priv
->display
.get_fifo_size
= i845_get_fifo_size
;
7234 dev_priv
->display
.update_wm
= i9xx_update_wm
;
7235 dev_priv
->display
.get_fifo_size
= i830_get_fifo_size
;
7238 if (IS_I85X(dev
) || IS_I865G(dev
))
7239 dev_priv
->display
.init_clock_gating
= i85x_init_clock_gating
;
7241 dev_priv
->display
.init_clock_gating
= i830_init_clock_gating
;
7243 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
7247 int sandybridge_pcode_read(struct drm_i915_private
*dev_priv
, u32 mbox
, u32
*val
)
7249 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
7251 if (I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) {
7252 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
7256 I915_WRITE(GEN6_PCODE_DATA
, *val
);
7257 I915_WRITE(GEN6_PCODE_DATA1
, 0);
7258 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
| mbox
);
7260 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
7262 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox
);
7266 *val
= I915_READ(GEN6_PCODE_DATA
);
7267 I915_WRITE(GEN6_PCODE_DATA
, 0);
7272 int sandybridge_pcode_write(struct drm_i915_private
*dev_priv
, u32 mbox
, u32 val
)
7274 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
7276 if (I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) {
7277 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
7281 I915_WRITE(GEN6_PCODE_DATA
, val
);
7282 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
| mbox
);
7284 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
7286 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox
);
7290 I915_WRITE(GEN6_PCODE_DATA
, 0);
7295 static int vlv_gpu_freq_div(unsigned int czclk_freq
)
7297 switch (czclk_freq
) {
7312 static int byt_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
7314 int div
, czclk_freq
= DIV_ROUND_CLOSEST(dev_priv
->mem_freq
, 4);
7316 div
= vlv_gpu_freq_div(czclk_freq
);
7320 return DIV_ROUND_CLOSEST(czclk_freq
* (val
+ 6 - 0xbd), div
);
7323 static int byt_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
7325 int mul
, czclk_freq
= DIV_ROUND_CLOSEST(dev_priv
->mem_freq
, 4);
7327 mul
= vlv_gpu_freq_div(czclk_freq
);
7331 return DIV_ROUND_CLOSEST(mul
* val
, czclk_freq
) + 0xbd - 6;
7334 static int chv_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
7336 int div
, czclk_freq
= dev_priv
->rps
.cz_freq
;
7338 div
= vlv_gpu_freq_div(czclk_freq
) / 2;
7342 return DIV_ROUND_CLOSEST(czclk_freq
* val
, 2 * div
) / 2;
7345 static int chv_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
7347 int mul
, czclk_freq
= dev_priv
->rps
.cz_freq
;
7349 mul
= vlv_gpu_freq_div(czclk_freq
) / 2;
7353 /* CHV needs even values */
7354 return DIV_ROUND_CLOSEST(val
* 2 * mul
, czclk_freq
) * 2;
7357 int intel_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
7359 if (IS_GEN9(dev_priv
->dev
))
7360 return (val
* GT_FREQUENCY_MULTIPLIER
) / GEN9_FREQ_SCALER
;
7361 else if (IS_CHERRYVIEW(dev_priv
->dev
))
7362 return chv_gpu_freq(dev_priv
, val
);
7363 else if (IS_VALLEYVIEW(dev_priv
->dev
))
7364 return byt_gpu_freq(dev_priv
, val
);
7366 return val
* GT_FREQUENCY_MULTIPLIER
;
7369 int intel_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
7371 if (IS_GEN9(dev_priv
->dev
))
7372 return (val
* GEN9_FREQ_SCALER
) / GT_FREQUENCY_MULTIPLIER
;
7373 else if (IS_CHERRYVIEW(dev_priv
->dev
))
7374 return chv_freq_opcode(dev_priv
, val
);
7375 else if (IS_VALLEYVIEW(dev_priv
->dev
))
7376 return byt_freq_opcode(dev_priv
, val
);
7378 return val
/ GT_FREQUENCY_MULTIPLIER
;
7381 struct request_boost
{
7382 struct work_struct work
;
7383 struct drm_i915_gem_request
*req
;
7386 static void __intel_rps_boost_work(struct work_struct
*work
)
7388 struct request_boost
*boost
= container_of(work
, struct request_boost
, work
);
7389 struct drm_i915_gem_request
*req
= boost
->req
;
7391 if (!i915_gem_request_completed(req
, true))
7392 gen6_rps_boost(to_i915(req
->ring
->dev
), NULL
,
7393 req
->emitted_jiffies
);
7395 i915_gem_request_unreference__unlocked(req
);
7399 void intel_queue_rps_boost_for_request(struct drm_device
*dev
,
7400 struct drm_i915_gem_request
*req
)
7402 struct request_boost
*boost
;
7404 if (req
== NULL
|| INTEL_INFO(dev
)->gen
< 6)
7407 if (i915_gem_request_completed(req
, true))
7410 boost
= kmalloc(sizeof(*boost
), GFP_ATOMIC
);
7414 i915_gem_request_reference(req
);
7417 INIT_WORK(&boost
->work
, __intel_rps_boost_work
);
7418 queue_work(to_i915(dev
)->wq
, &boost
->work
);
7421 void intel_pm_setup(struct drm_device
*dev
)
7423 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7425 mutex_init(&dev_priv
->rps
.hw_lock
);
7426 spin_lock_init(&dev_priv
->rps
.client_lock
);
7428 INIT_DELAYED_WORK(&dev_priv
->rps
.delayed_resume_work
,
7429 intel_gen6_powersave_work
);
7430 INIT_LIST_HEAD(&dev_priv
->rps
.clients
);
7431 INIT_LIST_HEAD(&dev_priv
->rps
.semaphores
.link
);
7432 INIT_LIST_HEAD(&dev_priv
->rps
.mmioflips
.link
);
7434 dev_priv
->pm
.suspended
= false;