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>
37 * RC6 is a special power stage which allows the GPU to enter an very
38 * low-voltage mode when idle, using down to 0V while at this stage. This
39 * stage is entered automatically when the GPU is idle when RC6 support is
40 * enabled, and as soon as new workload arises GPU wakes up automatically as well.
42 * There are different RC6 modes available in Intel GPU, which differentiate
43 * among each other with the latency required to enter and leave RC6 and
44 * voltage consumed by the GPU in different states.
46 * The combination of the following flags define which states GPU is allowed
47 * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
48 * RC6pp is deepest RC6. Their support by hardware varies according to the
49 * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
50 * which brings the most power savings; deeper states save more power, but
51 * require higher latency to switch to and wake up.
53 #define INTEL_RC6_ENABLE (1<<0)
54 #define INTEL_RC6p_ENABLE (1<<1)
55 #define INTEL_RC6pp_ENABLE (1<<2)
57 static void bxt_init_clock_gating(struct drm_device
*dev
)
59 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
61 /* WaDisableSDEUnitClockGating:bxt */
62 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
63 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
67 * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only.
69 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
70 GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ
);
73 * Wa: Backlight PWM may stop in the asserted state, causing backlight
76 if (IS_BXT_REVID(dev_priv
, BXT_REVID_B0
, REVID_FOREVER
))
77 I915_WRITE(GEN9_CLKGATE_DIS_0
, I915_READ(GEN9_CLKGATE_DIS_0
) |
78 PWM1_GATING_DIS
| PWM2_GATING_DIS
);
81 static void i915_pineview_get_mem_freq(struct drm_device
*dev
)
83 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
86 tmp
= I915_READ(CLKCFG
);
88 switch (tmp
& CLKCFG_FSB_MASK
) {
90 dev_priv
->fsb_freq
= 533; /* 133*4 */
93 dev_priv
->fsb_freq
= 800; /* 200*4 */
96 dev_priv
->fsb_freq
= 667; /* 167*4 */
99 dev_priv
->fsb_freq
= 400; /* 100*4 */
103 switch (tmp
& CLKCFG_MEM_MASK
) {
105 dev_priv
->mem_freq
= 533;
108 dev_priv
->mem_freq
= 667;
111 dev_priv
->mem_freq
= 800;
115 /* detect pineview DDR3 setting */
116 tmp
= I915_READ(CSHRDDR3CTL
);
117 dev_priv
->is_ddr3
= (tmp
& CSHRDDR3CTL_DDR3
) ? 1 : 0;
120 static void i915_ironlake_get_mem_freq(struct drm_device
*dev
)
122 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
125 ddrpll
= I915_READ16(DDRMPLL1
);
126 csipll
= I915_READ16(CSIPLL0
);
128 switch (ddrpll
& 0xff) {
130 dev_priv
->mem_freq
= 800;
133 dev_priv
->mem_freq
= 1066;
136 dev_priv
->mem_freq
= 1333;
139 dev_priv
->mem_freq
= 1600;
142 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
144 dev_priv
->mem_freq
= 0;
148 dev_priv
->ips
.r_t
= dev_priv
->mem_freq
;
150 switch (csipll
& 0x3ff) {
152 dev_priv
->fsb_freq
= 3200;
155 dev_priv
->fsb_freq
= 3733;
158 dev_priv
->fsb_freq
= 4266;
161 dev_priv
->fsb_freq
= 4800;
164 dev_priv
->fsb_freq
= 5333;
167 dev_priv
->fsb_freq
= 5866;
170 dev_priv
->fsb_freq
= 6400;
173 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
175 dev_priv
->fsb_freq
= 0;
179 if (dev_priv
->fsb_freq
== 3200) {
180 dev_priv
->ips
.c_m
= 0;
181 } else if (dev_priv
->fsb_freq
> 3200 && dev_priv
->fsb_freq
<= 4800) {
182 dev_priv
->ips
.c_m
= 1;
184 dev_priv
->ips
.c_m
= 2;
188 static const struct cxsr_latency cxsr_latency_table
[] = {
189 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
190 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
191 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
192 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
193 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
195 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
196 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
197 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
198 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
199 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
201 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
202 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
203 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
204 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
205 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
207 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
208 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
209 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
210 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
211 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
213 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
214 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
215 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
216 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
217 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
219 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
220 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
221 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
222 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
223 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
226 static const struct cxsr_latency
*intel_get_cxsr_latency(int is_desktop
,
231 const struct cxsr_latency
*latency
;
234 if (fsb
== 0 || mem
== 0)
237 for (i
= 0; i
< ARRAY_SIZE(cxsr_latency_table
); i
++) {
238 latency
= &cxsr_latency_table
[i
];
239 if (is_desktop
== latency
->is_desktop
&&
240 is_ddr3
== latency
->is_ddr3
&&
241 fsb
== latency
->fsb_freq
&& mem
== latency
->mem_freq
)
245 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
250 static void chv_set_memory_dvfs(struct drm_i915_private
*dev_priv
, bool enable
)
254 mutex_lock(&dev_priv
->rps
.hw_lock
);
256 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
);
258 val
&= ~FORCE_DDR_HIGH_FREQ
;
260 val
|= FORCE_DDR_HIGH_FREQ
;
261 val
&= ~FORCE_DDR_LOW_FREQ
;
262 val
|= FORCE_DDR_FREQ_REQ_ACK
;
263 vlv_punit_write(dev_priv
, PUNIT_REG_DDR_SETUP2
, val
);
265 if (wait_for((vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
) &
266 FORCE_DDR_FREQ_REQ_ACK
) == 0, 3))
267 DRM_ERROR("timed out waiting for Punit DDR DVFS request\n");
269 mutex_unlock(&dev_priv
->rps
.hw_lock
);
272 static void chv_set_memory_pm5(struct drm_i915_private
*dev_priv
, bool enable
)
276 mutex_lock(&dev_priv
->rps
.hw_lock
);
278 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DSPFREQ
);
280 val
|= DSP_MAXFIFO_PM5_ENABLE
;
282 val
&= ~DSP_MAXFIFO_PM5_ENABLE
;
283 vlv_punit_write(dev_priv
, PUNIT_REG_DSPFREQ
, val
);
285 mutex_unlock(&dev_priv
->rps
.hw_lock
);
288 #define FW_WM(value, plane) \
289 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK)
291 void intel_set_memory_cxsr(struct drm_i915_private
*dev_priv
, bool enable
)
293 struct drm_device
*dev
= dev_priv
->dev
;
296 if (IS_VALLEYVIEW(dev
) || IS_CHERRYVIEW(dev
)) {
297 I915_WRITE(FW_BLC_SELF_VLV
, enable
? FW_CSPWRDWNEN
: 0);
298 POSTING_READ(FW_BLC_SELF_VLV
);
299 dev_priv
->wm
.vlv
.cxsr
= enable
;
300 } else if (IS_G4X(dev
) || IS_CRESTLINE(dev
)) {
301 I915_WRITE(FW_BLC_SELF
, enable
? FW_BLC_SELF_EN
: 0);
302 POSTING_READ(FW_BLC_SELF
);
303 } else if (IS_PINEVIEW(dev
)) {
304 val
= I915_READ(DSPFW3
) & ~PINEVIEW_SELF_REFRESH_EN
;
305 val
|= enable
? PINEVIEW_SELF_REFRESH_EN
: 0;
306 I915_WRITE(DSPFW3
, val
);
307 POSTING_READ(DSPFW3
);
308 } else if (IS_I945G(dev
) || IS_I945GM(dev
)) {
309 val
= enable
? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN
) :
310 _MASKED_BIT_DISABLE(FW_BLC_SELF_EN
);
311 I915_WRITE(FW_BLC_SELF
, val
);
312 POSTING_READ(FW_BLC_SELF
);
313 } else if (IS_I915GM(dev
)) {
314 val
= enable
? _MASKED_BIT_ENABLE(INSTPM_SELF_EN
) :
315 _MASKED_BIT_DISABLE(INSTPM_SELF_EN
);
316 I915_WRITE(INSTPM
, val
);
317 POSTING_READ(INSTPM
);
322 DRM_DEBUG_KMS("memory self-refresh is %s\n",
323 enable
? "enabled" : "disabled");
328 * Latency for FIFO fetches is dependent on several factors:
329 * - memory configuration (speed, channels)
331 * - current MCH state
332 * It can be fairly high in some situations, so here we assume a fairly
333 * pessimal value. It's a tradeoff between extra memory fetches (if we
334 * set this value too high, the FIFO will fetch frequently to stay full)
335 * and power consumption (set it too low to save power and we might see
336 * FIFO underruns and display "flicker").
338 * A value of 5us seems to be a good balance; safe for very low end
339 * platforms but not overly aggressive on lower latency configs.
341 static const int pessimal_latency_ns
= 5000;
343 #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \
344 ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8))
346 static int vlv_get_fifo_size(struct drm_device
*dev
,
347 enum pipe pipe
, int plane
)
349 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
350 int sprite0_start
, sprite1_start
, size
;
353 uint32_t dsparb
, dsparb2
, dsparb3
;
355 dsparb
= I915_READ(DSPARB
);
356 dsparb2
= I915_READ(DSPARB2
);
357 sprite0_start
= VLV_FIFO_START(dsparb
, dsparb2
, 0, 0);
358 sprite1_start
= VLV_FIFO_START(dsparb
, dsparb2
, 8, 4);
361 dsparb
= I915_READ(DSPARB
);
362 dsparb2
= I915_READ(DSPARB2
);
363 sprite0_start
= VLV_FIFO_START(dsparb
, dsparb2
, 16, 8);
364 sprite1_start
= VLV_FIFO_START(dsparb
, dsparb2
, 24, 12);
367 dsparb2
= I915_READ(DSPARB2
);
368 dsparb3
= I915_READ(DSPARB3
);
369 sprite0_start
= VLV_FIFO_START(dsparb3
, dsparb2
, 0, 16);
370 sprite1_start
= VLV_FIFO_START(dsparb3
, dsparb2
, 8, 20);
378 size
= sprite0_start
;
381 size
= sprite1_start
- sprite0_start
;
384 size
= 512 - 1 - sprite1_start
;
390 DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n",
391 pipe_name(pipe
), plane
== 0 ? "primary" : "sprite",
392 plane
== 0 ? plane_name(pipe
) : sprite_name(pipe
, plane
- 1),
398 static int i9xx_get_fifo_size(struct drm_device
*dev
, int plane
)
400 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
401 uint32_t dsparb
= I915_READ(DSPARB
);
404 size
= dsparb
& 0x7f;
406 size
= ((dsparb
>> DSPARB_CSTART_SHIFT
) & 0x7f) - size
;
408 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
409 plane
? "B" : "A", size
);
414 static int i830_get_fifo_size(struct drm_device
*dev
, int plane
)
416 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
417 uint32_t dsparb
= I915_READ(DSPARB
);
420 size
= dsparb
& 0x1ff;
422 size
= ((dsparb
>> DSPARB_BEND_SHIFT
) & 0x1ff) - size
;
423 size
>>= 1; /* Convert to cachelines */
425 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
426 plane
? "B" : "A", size
);
431 static int i845_get_fifo_size(struct drm_device
*dev
, int plane
)
433 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
434 uint32_t dsparb
= I915_READ(DSPARB
);
437 size
= dsparb
& 0x7f;
438 size
>>= 2; /* Convert to cachelines */
440 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
447 /* Pineview has different values for various configs */
448 static const struct intel_watermark_params pineview_display_wm
= {
449 .fifo_size
= PINEVIEW_DISPLAY_FIFO
,
450 .max_wm
= PINEVIEW_MAX_WM
,
451 .default_wm
= PINEVIEW_DFT_WM
,
452 .guard_size
= PINEVIEW_GUARD_WM
,
453 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
455 static const struct intel_watermark_params pineview_display_hplloff_wm
= {
456 .fifo_size
= PINEVIEW_DISPLAY_FIFO
,
457 .max_wm
= PINEVIEW_MAX_WM
,
458 .default_wm
= PINEVIEW_DFT_HPLLOFF_WM
,
459 .guard_size
= PINEVIEW_GUARD_WM
,
460 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
462 static const struct intel_watermark_params pineview_cursor_wm
= {
463 .fifo_size
= PINEVIEW_CURSOR_FIFO
,
464 .max_wm
= PINEVIEW_CURSOR_MAX_WM
,
465 .default_wm
= PINEVIEW_CURSOR_DFT_WM
,
466 .guard_size
= PINEVIEW_CURSOR_GUARD_WM
,
467 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
469 static const struct intel_watermark_params pineview_cursor_hplloff_wm
= {
470 .fifo_size
= PINEVIEW_CURSOR_FIFO
,
471 .max_wm
= PINEVIEW_CURSOR_MAX_WM
,
472 .default_wm
= PINEVIEW_CURSOR_DFT_WM
,
473 .guard_size
= PINEVIEW_CURSOR_GUARD_WM
,
474 .cacheline_size
= PINEVIEW_FIFO_LINE_SIZE
,
476 static const struct intel_watermark_params g4x_wm_info
= {
477 .fifo_size
= G4X_FIFO_SIZE
,
478 .max_wm
= G4X_MAX_WM
,
479 .default_wm
= G4X_MAX_WM
,
481 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
483 static const struct intel_watermark_params g4x_cursor_wm_info
= {
484 .fifo_size
= I965_CURSOR_FIFO
,
485 .max_wm
= I965_CURSOR_MAX_WM
,
486 .default_wm
= I965_CURSOR_DFT_WM
,
488 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
490 static const struct intel_watermark_params valleyview_wm_info
= {
491 .fifo_size
= VALLEYVIEW_FIFO_SIZE
,
492 .max_wm
= VALLEYVIEW_MAX_WM
,
493 .default_wm
= VALLEYVIEW_MAX_WM
,
495 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
497 static const struct intel_watermark_params valleyview_cursor_wm_info
= {
498 .fifo_size
= I965_CURSOR_FIFO
,
499 .max_wm
= VALLEYVIEW_CURSOR_MAX_WM
,
500 .default_wm
= I965_CURSOR_DFT_WM
,
502 .cacheline_size
= G4X_FIFO_LINE_SIZE
,
504 static const struct intel_watermark_params i965_cursor_wm_info
= {
505 .fifo_size
= I965_CURSOR_FIFO
,
506 .max_wm
= I965_CURSOR_MAX_WM
,
507 .default_wm
= I965_CURSOR_DFT_WM
,
509 .cacheline_size
= I915_FIFO_LINE_SIZE
,
511 static const struct intel_watermark_params i945_wm_info
= {
512 .fifo_size
= I945_FIFO_SIZE
,
513 .max_wm
= I915_MAX_WM
,
516 .cacheline_size
= I915_FIFO_LINE_SIZE
,
518 static const struct intel_watermark_params i915_wm_info
= {
519 .fifo_size
= I915_FIFO_SIZE
,
520 .max_wm
= I915_MAX_WM
,
523 .cacheline_size
= I915_FIFO_LINE_SIZE
,
525 static const struct intel_watermark_params i830_a_wm_info
= {
526 .fifo_size
= I855GM_FIFO_SIZE
,
527 .max_wm
= I915_MAX_WM
,
530 .cacheline_size
= I830_FIFO_LINE_SIZE
,
532 static const struct intel_watermark_params i830_bc_wm_info
= {
533 .fifo_size
= I855GM_FIFO_SIZE
,
534 .max_wm
= I915_MAX_WM
/2,
537 .cacheline_size
= I830_FIFO_LINE_SIZE
,
539 static const struct intel_watermark_params i845_wm_info
= {
540 .fifo_size
= I830_FIFO_SIZE
,
541 .max_wm
= I915_MAX_WM
,
544 .cacheline_size
= I830_FIFO_LINE_SIZE
,
548 * intel_calculate_wm - calculate watermark level
549 * @clock_in_khz: pixel clock
550 * @wm: chip FIFO params
551 * @cpp: bytes per pixel
552 * @latency_ns: memory latency for the platform
554 * Calculate the watermark level (the level at which the display plane will
555 * start fetching from memory again). Each chip has a different display
556 * FIFO size and allocation, so the caller needs to figure that out and pass
557 * in the correct intel_watermark_params structure.
559 * As the pixel clock runs, the FIFO will be drained at a rate that depends
560 * on the pixel size. When it reaches the watermark level, it'll start
561 * fetching FIFO line sized based chunks from memory until the FIFO fills
562 * past the watermark point. If the FIFO drains completely, a FIFO underrun
563 * will occur, and a display engine hang could result.
565 static unsigned long intel_calculate_wm(unsigned long clock_in_khz
,
566 const struct intel_watermark_params
*wm
,
567 int fifo_size
, int cpp
,
568 unsigned long latency_ns
)
570 long entries_required
, wm_size
;
573 * Note: we need to make sure we don't overflow for various clock &
575 * clocks go from a few thousand to several hundred thousand.
576 * latency is usually a few thousand
578 entries_required
= ((clock_in_khz
/ 1000) * cpp
* latency_ns
) /
580 entries_required
= DIV_ROUND_UP(entries_required
, wm
->cacheline_size
);
582 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required
);
584 wm_size
= fifo_size
- (entries_required
+ wm
->guard_size
);
586 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size
);
588 /* Don't promote wm_size to unsigned... */
589 if (wm_size
> (long)wm
->max_wm
)
590 wm_size
= wm
->max_wm
;
592 wm_size
= wm
->default_wm
;
595 * Bspec seems to indicate that the value shouldn't be lower than
596 * 'burst size + 1'. Certainly 830 is quite unhappy with low values.
597 * Lets go for 8 which is the burst size since certain platforms
598 * already use a hardcoded 8 (which is what the spec says should be
607 static struct drm_crtc
*single_enabled_crtc(struct drm_device
*dev
)
609 struct drm_crtc
*crtc
, *enabled
= NULL
;
611 for_each_crtc(dev
, crtc
) {
612 if (intel_crtc_active(crtc
)) {
622 static void pineview_update_wm(struct drm_crtc
*unused_crtc
)
624 struct drm_device
*dev
= unused_crtc
->dev
;
625 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
626 struct drm_crtc
*crtc
;
627 const struct cxsr_latency
*latency
;
631 latency
= intel_get_cxsr_latency(IS_PINEVIEW_G(dev
), dev_priv
->is_ddr3
,
632 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
634 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
635 intel_set_memory_cxsr(dev_priv
, false);
639 crtc
= single_enabled_crtc(dev
);
641 const struct drm_display_mode
*adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
642 int cpp
= drm_format_plane_cpp(crtc
->primary
->state
->fb
->pixel_format
, 0);
643 int clock
= adjusted_mode
->crtc_clock
;
646 wm
= intel_calculate_wm(clock
, &pineview_display_wm
,
647 pineview_display_wm
.fifo_size
,
648 cpp
, latency
->display_sr
);
649 reg
= I915_READ(DSPFW1
);
650 reg
&= ~DSPFW_SR_MASK
;
651 reg
|= FW_WM(wm
, SR
);
652 I915_WRITE(DSPFW1
, reg
);
653 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg
);
656 wm
= intel_calculate_wm(clock
, &pineview_cursor_wm
,
657 pineview_display_wm
.fifo_size
,
658 cpp
, latency
->cursor_sr
);
659 reg
= I915_READ(DSPFW3
);
660 reg
&= ~DSPFW_CURSOR_SR_MASK
;
661 reg
|= FW_WM(wm
, CURSOR_SR
);
662 I915_WRITE(DSPFW3
, reg
);
664 /* Display HPLL off SR */
665 wm
= intel_calculate_wm(clock
, &pineview_display_hplloff_wm
,
666 pineview_display_hplloff_wm
.fifo_size
,
667 cpp
, latency
->display_hpll_disable
);
668 reg
= I915_READ(DSPFW3
);
669 reg
&= ~DSPFW_HPLL_SR_MASK
;
670 reg
|= FW_WM(wm
, HPLL_SR
);
671 I915_WRITE(DSPFW3
, reg
);
673 /* cursor HPLL off SR */
674 wm
= intel_calculate_wm(clock
, &pineview_cursor_hplloff_wm
,
675 pineview_display_hplloff_wm
.fifo_size
,
676 cpp
, latency
->cursor_hpll_disable
);
677 reg
= I915_READ(DSPFW3
);
678 reg
&= ~DSPFW_HPLL_CURSOR_MASK
;
679 reg
|= FW_WM(wm
, HPLL_CURSOR
);
680 I915_WRITE(DSPFW3
, reg
);
681 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg
);
683 intel_set_memory_cxsr(dev_priv
, true);
685 intel_set_memory_cxsr(dev_priv
, false);
689 static bool g4x_compute_wm0(struct drm_device
*dev
,
691 const struct intel_watermark_params
*display
,
692 int display_latency_ns
,
693 const struct intel_watermark_params
*cursor
,
694 int cursor_latency_ns
,
698 struct drm_crtc
*crtc
;
699 const struct drm_display_mode
*adjusted_mode
;
700 int htotal
, hdisplay
, clock
, cpp
;
701 int line_time_us
, line_count
;
702 int entries
, tlb_miss
;
704 crtc
= intel_get_crtc_for_plane(dev
, plane
);
705 if (!intel_crtc_active(crtc
)) {
706 *cursor_wm
= cursor
->guard_size
;
707 *plane_wm
= display
->guard_size
;
711 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
712 clock
= adjusted_mode
->crtc_clock
;
713 htotal
= adjusted_mode
->crtc_htotal
;
714 hdisplay
= to_intel_crtc(crtc
)->config
->pipe_src_w
;
715 cpp
= drm_format_plane_cpp(crtc
->primary
->state
->fb
->pixel_format
, 0);
717 /* Use the small buffer method to calculate plane watermark */
718 entries
= ((clock
* cpp
/ 1000) * display_latency_ns
) / 1000;
719 tlb_miss
= display
->fifo_size
*display
->cacheline_size
- hdisplay
* 8;
722 entries
= DIV_ROUND_UP(entries
, display
->cacheline_size
);
723 *plane_wm
= entries
+ display
->guard_size
;
724 if (*plane_wm
> (int)display
->max_wm
)
725 *plane_wm
= display
->max_wm
;
727 /* Use the large buffer method to calculate cursor watermark */
728 line_time_us
= max(htotal
* 1000 / clock
, 1);
729 line_count
= (cursor_latency_ns
/ line_time_us
+ 1000) / 1000;
730 entries
= line_count
* crtc
->cursor
->state
->crtc_w
* cpp
;
731 tlb_miss
= cursor
->fifo_size
*cursor
->cacheline_size
- hdisplay
* 8;
734 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
735 *cursor_wm
= entries
+ cursor
->guard_size
;
736 if (*cursor_wm
> (int)cursor
->max_wm
)
737 *cursor_wm
= (int)cursor
->max_wm
;
743 * Check the wm result.
745 * If any calculated watermark values is larger than the maximum value that
746 * can be programmed into the associated watermark register, that watermark
749 static bool g4x_check_srwm(struct drm_device
*dev
,
750 int display_wm
, int cursor_wm
,
751 const struct intel_watermark_params
*display
,
752 const struct intel_watermark_params
*cursor
)
754 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
755 display_wm
, cursor_wm
);
757 if (display_wm
> display
->max_wm
) {
758 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
759 display_wm
, display
->max_wm
);
763 if (cursor_wm
> cursor
->max_wm
) {
764 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
765 cursor_wm
, cursor
->max_wm
);
769 if (!(display_wm
|| cursor_wm
)) {
770 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
777 static bool g4x_compute_srwm(struct drm_device
*dev
,
780 const struct intel_watermark_params
*display
,
781 const struct intel_watermark_params
*cursor
,
782 int *display_wm
, int *cursor_wm
)
784 struct drm_crtc
*crtc
;
785 const struct drm_display_mode
*adjusted_mode
;
786 int hdisplay
, htotal
, cpp
, clock
;
787 unsigned long line_time_us
;
788 int line_count
, line_size
;
793 *display_wm
= *cursor_wm
= 0;
797 crtc
= intel_get_crtc_for_plane(dev
, plane
);
798 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
799 clock
= adjusted_mode
->crtc_clock
;
800 htotal
= adjusted_mode
->crtc_htotal
;
801 hdisplay
= to_intel_crtc(crtc
)->config
->pipe_src_w
;
802 cpp
= drm_format_plane_cpp(crtc
->primary
->state
->fb
->pixel_format
, 0);
804 line_time_us
= max(htotal
* 1000 / clock
, 1);
805 line_count
= (latency_ns
/ line_time_us
+ 1000) / 1000;
806 line_size
= hdisplay
* cpp
;
808 /* Use the minimum of the small and large buffer method for primary */
809 small
= ((clock
* cpp
/ 1000) * latency_ns
) / 1000;
810 large
= line_count
* line_size
;
812 entries
= DIV_ROUND_UP(min(small
, large
), display
->cacheline_size
);
813 *display_wm
= entries
+ display
->guard_size
;
815 /* calculate the self-refresh watermark for display cursor */
816 entries
= line_count
* cpp
* crtc
->cursor
->state
->crtc_w
;
817 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
818 *cursor_wm
= entries
+ cursor
->guard_size
;
820 return g4x_check_srwm(dev
,
821 *display_wm
, *cursor_wm
,
825 #define FW_WM_VLV(value, plane) \
826 (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV)
828 static void vlv_write_wm_values(struct intel_crtc
*crtc
,
829 const struct vlv_wm_values
*wm
)
831 struct drm_i915_private
*dev_priv
= to_i915(crtc
->base
.dev
);
832 enum pipe pipe
= crtc
->pipe
;
834 I915_WRITE(VLV_DDL(pipe
),
835 (wm
->ddl
[pipe
].cursor
<< DDL_CURSOR_SHIFT
) |
836 (wm
->ddl
[pipe
].sprite
[1] << DDL_SPRITE_SHIFT(1)) |
837 (wm
->ddl
[pipe
].sprite
[0] << DDL_SPRITE_SHIFT(0)) |
838 (wm
->ddl
[pipe
].primary
<< DDL_PLANE_SHIFT
));
841 FW_WM(wm
->sr
.plane
, SR
) |
842 FW_WM(wm
->pipe
[PIPE_B
].cursor
, CURSORB
) |
843 FW_WM_VLV(wm
->pipe
[PIPE_B
].primary
, PLANEB
) |
844 FW_WM_VLV(wm
->pipe
[PIPE_A
].primary
, PLANEA
));
846 FW_WM_VLV(wm
->pipe
[PIPE_A
].sprite
[1], SPRITEB
) |
847 FW_WM(wm
->pipe
[PIPE_A
].cursor
, CURSORA
) |
848 FW_WM_VLV(wm
->pipe
[PIPE_A
].sprite
[0], SPRITEA
));
850 FW_WM(wm
->sr
.cursor
, CURSOR_SR
));
852 if (IS_CHERRYVIEW(dev_priv
)) {
853 I915_WRITE(DSPFW7_CHV
,
854 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[1], SPRITED
) |
855 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[0], SPRITEC
));
856 I915_WRITE(DSPFW8_CHV
,
857 FW_WM_VLV(wm
->pipe
[PIPE_C
].sprite
[1], SPRITEF
) |
858 FW_WM_VLV(wm
->pipe
[PIPE_C
].sprite
[0], SPRITEE
));
859 I915_WRITE(DSPFW9_CHV
,
860 FW_WM_VLV(wm
->pipe
[PIPE_C
].primary
, PLANEC
) |
861 FW_WM(wm
->pipe
[PIPE_C
].cursor
, CURSORC
));
863 FW_WM(wm
->sr
.plane
>> 9, SR_HI
) |
864 FW_WM(wm
->pipe
[PIPE_C
].sprite
[1] >> 8, SPRITEF_HI
) |
865 FW_WM(wm
->pipe
[PIPE_C
].sprite
[0] >> 8, SPRITEE_HI
) |
866 FW_WM(wm
->pipe
[PIPE_C
].primary
>> 8, PLANEC_HI
) |
867 FW_WM(wm
->pipe
[PIPE_B
].sprite
[1] >> 8, SPRITED_HI
) |
868 FW_WM(wm
->pipe
[PIPE_B
].sprite
[0] >> 8, SPRITEC_HI
) |
869 FW_WM(wm
->pipe
[PIPE_B
].primary
>> 8, PLANEB_HI
) |
870 FW_WM(wm
->pipe
[PIPE_A
].sprite
[1] >> 8, SPRITEB_HI
) |
871 FW_WM(wm
->pipe
[PIPE_A
].sprite
[0] >> 8, SPRITEA_HI
) |
872 FW_WM(wm
->pipe
[PIPE_A
].primary
>> 8, PLANEA_HI
));
875 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[1], SPRITED
) |
876 FW_WM_VLV(wm
->pipe
[PIPE_B
].sprite
[0], SPRITEC
));
878 FW_WM(wm
->sr
.plane
>> 9, SR_HI
) |
879 FW_WM(wm
->pipe
[PIPE_B
].sprite
[1] >> 8, SPRITED_HI
) |
880 FW_WM(wm
->pipe
[PIPE_B
].sprite
[0] >> 8, SPRITEC_HI
) |
881 FW_WM(wm
->pipe
[PIPE_B
].primary
>> 8, PLANEB_HI
) |
882 FW_WM(wm
->pipe
[PIPE_A
].sprite
[1] >> 8, SPRITEB_HI
) |
883 FW_WM(wm
->pipe
[PIPE_A
].sprite
[0] >> 8, SPRITEA_HI
) |
884 FW_WM(wm
->pipe
[PIPE_A
].primary
>> 8, PLANEA_HI
));
887 /* zero (unused) WM1 watermarks */
888 I915_WRITE(DSPFW4
, 0);
889 I915_WRITE(DSPFW5
, 0);
890 I915_WRITE(DSPFW6
, 0);
891 I915_WRITE(DSPHOWM1
, 0);
893 POSTING_READ(DSPFW1
);
901 VLV_WM_LEVEL_DDR_DVFS
,
904 /* latency must be in 0.1us units. */
905 static unsigned int vlv_wm_method2(unsigned int pixel_rate
,
906 unsigned int pipe_htotal
,
907 unsigned int horiz_pixels
,
909 unsigned int latency
)
913 ret
= (latency
* pixel_rate
) / (pipe_htotal
* 10000);
914 ret
= (ret
+ 1) * horiz_pixels
* cpp
;
915 ret
= DIV_ROUND_UP(ret
, 64);
920 static void vlv_setup_wm_latency(struct drm_device
*dev
)
922 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
924 /* all latencies in usec */
925 dev_priv
->wm
.pri_latency
[VLV_WM_LEVEL_PM2
] = 3;
927 dev_priv
->wm
.max_level
= VLV_WM_LEVEL_PM2
;
929 if (IS_CHERRYVIEW(dev_priv
)) {
930 dev_priv
->wm
.pri_latency
[VLV_WM_LEVEL_PM5
] = 12;
931 dev_priv
->wm
.pri_latency
[VLV_WM_LEVEL_DDR_DVFS
] = 33;
933 dev_priv
->wm
.max_level
= VLV_WM_LEVEL_DDR_DVFS
;
937 static uint16_t vlv_compute_wm_level(struct intel_plane
*plane
,
938 struct intel_crtc
*crtc
,
939 const struct intel_plane_state
*state
,
942 struct drm_i915_private
*dev_priv
= to_i915(plane
->base
.dev
);
943 int clock
, htotal
, cpp
, width
, wm
;
945 if (dev_priv
->wm
.pri_latency
[level
] == 0)
951 cpp
= drm_format_plane_cpp(state
->base
.fb
->pixel_format
, 0);
952 clock
= crtc
->config
->base
.adjusted_mode
.crtc_clock
;
953 htotal
= crtc
->config
->base
.adjusted_mode
.crtc_htotal
;
954 width
= crtc
->config
->pipe_src_w
;
955 if (WARN_ON(htotal
== 0))
958 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
) {
960 * FIXME the formula gives values that are
961 * too big for the cursor FIFO, and hence we
962 * would never be able to use cursors. For
963 * now just hardcode the watermark.
967 wm
= vlv_wm_method2(clock
, htotal
, width
, cpp
,
968 dev_priv
->wm
.pri_latency
[level
] * 10);
971 return min_t(int, wm
, USHRT_MAX
);
974 static void vlv_compute_fifo(struct intel_crtc
*crtc
)
976 struct drm_device
*dev
= crtc
->base
.dev
;
977 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
978 struct intel_plane
*plane
;
979 unsigned int total_rate
= 0;
980 const int fifo_size
= 512 - 1;
981 int fifo_extra
, fifo_left
= fifo_size
;
983 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
984 struct intel_plane_state
*state
=
985 to_intel_plane_state(plane
->base
.state
);
987 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
)
990 if (state
->visible
) {
991 wm_state
->num_active_planes
++;
992 total_rate
+= drm_format_plane_cpp(state
->base
.fb
->pixel_format
, 0);
996 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
997 struct intel_plane_state
*state
=
998 to_intel_plane_state(plane
->base
.state
);
1001 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
) {
1002 plane
->wm
.fifo_size
= 63;
1006 if (!state
->visible
) {
1007 plane
->wm
.fifo_size
= 0;
1011 rate
= drm_format_plane_cpp(state
->base
.fb
->pixel_format
, 0);
1012 plane
->wm
.fifo_size
= fifo_size
* rate
/ total_rate
;
1013 fifo_left
-= plane
->wm
.fifo_size
;
1016 fifo_extra
= DIV_ROUND_UP(fifo_left
, wm_state
->num_active_planes
?: 1);
1018 /* spread the remainder evenly */
1019 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1025 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
)
1028 /* give it all to the first plane if none are active */
1029 if (plane
->wm
.fifo_size
== 0 &&
1030 wm_state
->num_active_planes
)
1033 plane_extra
= min(fifo_extra
, fifo_left
);
1034 plane
->wm
.fifo_size
+= plane_extra
;
1035 fifo_left
-= plane_extra
;
1038 WARN_ON(fifo_left
!= 0);
1041 static void vlv_invert_wms(struct intel_crtc
*crtc
)
1043 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1046 for (level
= 0; level
< wm_state
->num_levels
; level
++) {
1047 struct drm_device
*dev
= crtc
->base
.dev
;
1048 const int sr_fifo_size
= INTEL_INFO(dev
)->num_pipes
* 512 - 1;
1049 struct intel_plane
*plane
;
1051 wm_state
->sr
[level
].plane
= sr_fifo_size
- wm_state
->sr
[level
].plane
;
1052 wm_state
->sr
[level
].cursor
= 63 - wm_state
->sr
[level
].cursor
;
1054 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1055 switch (plane
->base
.type
) {
1057 case DRM_PLANE_TYPE_CURSOR
:
1058 wm_state
->wm
[level
].cursor
= plane
->wm
.fifo_size
-
1059 wm_state
->wm
[level
].cursor
;
1061 case DRM_PLANE_TYPE_PRIMARY
:
1062 wm_state
->wm
[level
].primary
= plane
->wm
.fifo_size
-
1063 wm_state
->wm
[level
].primary
;
1065 case DRM_PLANE_TYPE_OVERLAY
:
1066 sprite
= plane
->plane
;
1067 wm_state
->wm
[level
].sprite
[sprite
] = plane
->wm
.fifo_size
-
1068 wm_state
->wm
[level
].sprite
[sprite
];
1075 static void vlv_compute_wm(struct intel_crtc
*crtc
)
1077 struct drm_device
*dev
= crtc
->base
.dev
;
1078 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1079 struct intel_plane
*plane
;
1080 int sr_fifo_size
= INTEL_INFO(dev
)->num_pipes
* 512 - 1;
1083 memset(wm_state
, 0, sizeof(*wm_state
));
1085 wm_state
->cxsr
= crtc
->pipe
!= PIPE_C
&& crtc
->wm
.cxsr_allowed
;
1086 wm_state
->num_levels
= to_i915(dev
)->wm
.max_level
+ 1;
1088 wm_state
->num_active_planes
= 0;
1090 vlv_compute_fifo(crtc
);
1092 if (wm_state
->num_active_planes
!= 1)
1093 wm_state
->cxsr
= false;
1095 if (wm_state
->cxsr
) {
1096 for (level
= 0; level
< wm_state
->num_levels
; level
++) {
1097 wm_state
->sr
[level
].plane
= sr_fifo_size
;
1098 wm_state
->sr
[level
].cursor
= 63;
1102 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1103 struct intel_plane_state
*state
=
1104 to_intel_plane_state(plane
->base
.state
);
1106 if (!state
->visible
)
1109 /* normal watermarks */
1110 for (level
= 0; level
< wm_state
->num_levels
; level
++) {
1111 int wm
= vlv_compute_wm_level(plane
, crtc
, state
, level
);
1112 int max_wm
= plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
? 63 : 511;
1115 if (WARN_ON(level
== 0 && wm
> max_wm
))
1118 if (wm
> plane
->wm
.fifo_size
)
1121 switch (plane
->base
.type
) {
1123 case DRM_PLANE_TYPE_CURSOR
:
1124 wm_state
->wm
[level
].cursor
= wm
;
1126 case DRM_PLANE_TYPE_PRIMARY
:
1127 wm_state
->wm
[level
].primary
= wm
;
1129 case DRM_PLANE_TYPE_OVERLAY
:
1130 sprite
= plane
->plane
;
1131 wm_state
->wm
[level
].sprite
[sprite
] = wm
;
1136 wm_state
->num_levels
= level
;
1138 if (!wm_state
->cxsr
)
1141 /* maxfifo watermarks */
1142 switch (plane
->base
.type
) {
1144 case DRM_PLANE_TYPE_CURSOR
:
1145 for (level
= 0; level
< wm_state
->num_levels
; level
++)
1146 wm_state
->sr
[level
].cursor
=
1147 wm_state
->wm
[level
].cursor
;
1149 case DRM_PLANE_TYPE_PRIMARY
:
1150 for (level
= 0; level
< wm_state
->num_levels
; level
++)
1151 wm_state
->sr
[level
].plane
=
1152 min(wm_state
->sr
[level
].plane
,
1153 wm_state
->wm
[level
].primary
);
1155 case DRM_PLANE_TYPE_OVERLAY
:
1156 sprite
= plane
->plane
;
1157 for (level
= 0; level
< wm_state
->num_levels
; level
++)
1158 wm_state
->sr
[level
].plane
=
1159 min(wm_state
->sr
[level
].plane
,
1160 wm_state
->wm
[level
].sprite
[sprite
]);
1165 /* clear any (partially) filled invalid levels */
1166 for (level
= wm_state
->num_levels
; level
< to_i915(dev
)->wm
.max_level
+ 1; level
++) {
1167 memset(&wm_state
->wm
[level
], 0, sizeof(wm_state
->wm
[level
]));
1168 memset(&wm_state
->sr
[level
], 0, sizeof(wm_state
->sr
[level
]));
1171 vlv_invert_wms(crtc
);
1174 #define VLV_FIFO(plane, value) \
1175 (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV)
1177 static void vlv_pipe_set_fifo_size(struct intel_crtc
*crtc
)
1179 struct drm_device
*dev
= crtc
->base
.dev
;
1180 struct drm_i915_private
*dev_priv
= to_i915(dev
);
1181 struct intel_plane
*plane
;
1182 int sprite0_start
= 0, sprite1_start
= 0, fifo_size
= 0;
1184 for_each_intel_plane_on_crtc(dev
, crtc
, plane
) {
1185 if (plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
) {
1186 WARN_ON(plane
->wm
.fifo_size
!= 63);
1190 if (plane
->base
.type
== DRM_PLANE_TYPE_PRIMARY
)
1191 sprite0_start
= plane
->wm
.fifo_size
;
1192 else if (plane
->plane
== 0)
1193 sprite1_start
= sprite0_start
+ plane
->wm
.fifo_size
;
1195 fifo_size
= sprite1_start
+ plane
->wm
.fifo_size
;
1198 WARN_ON(fifo_size
!= 512 - 1);
1200 DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n",
1201 pipe_name(crtc
->pipe
), sprite0_start
,
1202 sprite1_start
, fifo_size
);
1204 switch (crtc
->pipe
) {
1205 uint32_t dsparb
, dsparb2
, dsparb3
;
1207 dsparb
= I915_READ(DSPARB
);
1208 dsparb2
= I915_READ(DSPARB2
);
1210 dsparb
&= ~(VLV_FIFO(SPRITEA
, 0xff) |
1211 VLV_FIFO(SPRITEB
, 0xff));
1212 dsparb
|= (VLV_FIFO(SPRITEA
, sprite0_start
) |
1213 VLV_FIFO(SPRITEB
, sprite1_start
));
1215 dsparb2
&= ~(VLV_FIFO(SPRITEA_HI
, 0x1) |
1216 VLV_FIFO(SPRITEB_HI
, 0x1));
1217 dsparb2
|= (VLV_FIFO(SPRITEA_HI
, sprite0_start
>> 8) |
1218 VLV_FIFO(SPRITEB_HI
, sprite1_start
>> 8));
1220 I915_WRITE(DSPARB
, dsparb
);
1221 I915_WRITE(DSPARB2
, dsparb2
);
1224 dsparb
= I915_READ(DSPARB
);
1225 dsparb2
= I915_READ(DSPARB2
);
1227 dsparb
&= ~(VLV_FIFO(SPRITEC
, 0xff) |
1228 VLV_FIFO(SPRITED
, 0xff));
1229 dsparb
|= (VLV_FIFO(SPRITEC
, sprite0_start
) |
1230 VLV_FIFO(SPRITED
, sprite1_start
));
1232 dsparb2
&= ~(VLV_FIFO(SPRITEC_HI
, 0xff) |
1233 VLV_FIFO(SPRITED_HI
, 0xff));
1234 dsparb2
|= (VLV_FIFO(SPRITEC_HI
, sprite0_start
>> 8) |
1235 VLV_FIFO(SPRITED_HI
, sprite1_start
>> 8));
1237 I915_WRITE(DSPARB
, dsparb
);
1238 I915_WRITE(DSPARB2
, dsparb2
);
1241 dsparb3
= I915_READ(DSPARB3
);
1242 dsparb2
= I915_READ(DSPARB2
);
1244 dsparb3
&= ~(VLV_FIFO(SPRITEE
, 0xff) |
1245 VLV_FIFO(SPRITEF
, 0xff));
1246 dsparb3
|= (VLV_FIFO(SPRITEE
, sprite0_start
) |
1247 VLV_FIFO(SPRITEF
, sprite1_start
));
1249 dsparb2
&= ~(VLV_FIFO(SPRITEE_HI
, 0xff) |
1250 VLV_FIFO(SPRITEF_HI
, 0xff));
1251 dsparb2
|= (VLV_FIFO(SPRITEE_HI
, sprite0_start
>> 8) |
1252 VLV_FIFO(SPRITEF_HI
, sprite1_start
>> 8));
1254 I915_WRITE(DSPARB3
, dsparb3
);
1255 I915_WRITE(DSPARB2
, dsparb2
);
1264 static void vlv_merge_wm(struct drm_device
*dev
,
1265 struct vlv_wm_values
*wm
)
1267 struct intel_crtc
*crtc
;
1268 int num_active_crtcs
= 0;
1270 wm
->level
= to_i915(dev
)->wm
.max_level
;
1273 for_each_intel_crtc(dev
, crtc
) {
1274 const struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1279 if (!wm_state
->cxsr
)
1283 wm
->level
= min_t(int, wm
->level
, wm_state
->num_levels
- 1);
1286 if (num_active_crtcs
!= 1)
1289 if (num_active_crtcs
> 1)
1290 wm
->level
= VLV_WM_LEVEL_PM2
;
1292 for_each_intel_crtc(dev
, crtc
) {
1293 struct vlv_wm_state
*wm_state
= &crtc
->wm_state
;
1294 enum pipe pipe
= crtc
->pipe
;
1299 wm
->pipe
[pipe
] = wm_state
->wm
[wm
->level
];
1301 wm
->sr
= wm_state
->sr
[wm
->level
];
1303 wm
->ddl
[pipe
].primary
= DDL_PRECISION_HIGH
| 2;
1304 wm
->ddl
[pipe
].sprite
[0] = DDL_PRECISION_HIGH
| 2;
1305 wm
->ddl
[pipe
].sprite
[1] = DDL_PRECISION_HIGH
| 2;
1306 wm
->ddl
[pipe
].cursor
= DDL_PRECISION_HIGH
| 2;
1310 static void vlv_update_wm(struct drm_crtc
*crtc
)
1312 struct drm_device
*dev
= crtc
->dev
;
1313 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1314 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
1315 enum pipe pipe
= intel_crtc
->pipe
;
1316 struct vlv_wm_values wm
= {};
1318 vlv_compute_wm(intel_crtc
);
1319 vlv_merge_wm(dev
, &wm
);
1321 if (memcmp(&dev_priv
->wm
.vlv
, &wm
, sizeof(wm
)) == 0) {
1322 /* FIXME should be part of crtc atomic commit */
1323 vlv_pipe_set_fifo_size(intel_crtc
);
1327 if (wm
.level
< VLV_WM_LEVEL_DDR_DVFS
&&
1328 dev_priv
->wm
.vlv
.level
>= VLV_WM_LEVEL_DDR_DVFS
)
1329 chv_set_memory_dvfs(dev_priv
, false);
1331 if (wm
.level
< VLV_WM_LEVEL_PM5
&&
1332 dev_priv
->wm
.vlv
.level
>= VLV_WM_LEVEL_PM5
)
1333 chv_set_memory_pm5(dev_priv
, false);
1335 if (!wm
.cxsr
&& dev_priv
->wm
.vlv
.cxsr
)
1336 intel_set_memory_cxsr(dev_priv
, false);
1338 /* FIXME should be part of crtc atomic commit */
1339 vlv_pipe_set_fifo_size(intel_crtc
);
1341 vlv_write_wm_values(intel_crtc
, &wm
);
1343 DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, "
1344 "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n",
1345 pipe_name(pipe
), wm
.pipe
[pipe
].primary
, wm
.pipe
[pipe
].cursor
,
1346 wm
.pipe
[pipe
].sprite
[0], wm
.pipe
[pipe
].sprite
[1],
1347 wm
.sr
.plane
, wm
.sr
.cursor
, wm
.level
, wm
.cxsr
);
1349 if (wm
.cxsr
&& !dev_priv
->wm
.vlv
.cxsr
)
1350 intel_set_memory_cxsr(dev_priv
, true);
1352 if (wm
.level
>= VLV_WM_LEVEL_PM5
&&
1353 dev_priv
->wm
.vlv
.level
< VLV_WM_LEVEL_PM5
)
1354 chv_set_memory_pm5(dev_priv
, true);
1356 if (wm
.level
>= VLV_WM_LEVEL_DDR_DVFS
&&
1357 dev_priv
->wm
.vlv
.level
< VLV_WM_LEVEL_DDR_DVFS
)
1358 chv_set_memory_dvfs(dev_priv
, true);
1360 dev_priv
->wm
.vlv
= wm
;
1363 #define single_plane_enabled(mask) is_power_of_2(mask)
1365 static void g4x_update_wm(struct drm_crtc
*crtc
)
1367 struct drm_device
*dev
= crtc
->dev
;
1368 static const int sr_latency_ns
= 12000;
1369 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1370 int planea_wm
, planeb_wm
, cursora_wm
, cursorb_wm
;
1371 int plane_sr
, cursor_sr
;
1372 unsigned int enabled
= 0;
1375 if (g4x_compute_wm0(dev
, PIPE_A
,
1376 &g4x_wm_info
, pessimal_latency_ns
,
1377 &g4x_cursor_wm_info
, pessimal_latency_ns
,
1378 &planea_wm
, &cursora_wm
))
1379 enabled
|= 1 << PIPE_A
;
1381 if (g4x_compute_wm0(dev
, PIPE_B
,
1382 &g4x_wm_info
, pessimal_latency_ns
,
1383 &g4x_cursor_wm_info
, pessimal_latency_ns
,
1384 &planeb_wm
, &cursorb_wm
))
1385 enabled
|= 1 << PIPE_B
;
1387 if (single_plane_enabled(enabled
) &&
1388 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1391 &g4x_cursor_wm_info
,
1392 &plane_sr
, &cursor_sr
)) {
1393 cxsr_enabled
= true;
1395 cxsr_enabled
= false;
1396 intel_set_memory_cxsr(dev_priv
, false);
1397 plane_sr
= cursor_sr
= 0;
1400 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, "
1401 "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1402 planea_wm
, cursora_wm
,
1403 planeb_wm
, cursorb_wm
,
1404 plane_sr
, cursor_sr
);
1407 FW_WM(plane_sr
, SR
) |
1408 FW_WM(cursorb_wm
, CURSORB
) |
1409 FW_WM(planeb_wm
, PLANEB
) |
1410 FW_WM(planea_wm
, PLANEA
));
1412 (I915_READ(DSPFW2
) & ~DSPFW_CURSORA_MASK
) |
1413 FW_WM(cursora_wm
, CURSORA
));
1414 /* HPLL off in SR has some issues on G4x... disable it */
1416 (I915_READ(DSPFW3
) & ~(DSPFW_HPLL_SR_EN
| DSPFW_CURSOR_SR_MASK
)) |
1417 FW_WM(cursor_sr
, CURSOR_SR
));
1420 intel_set_memory_cxsr(dev_priv
, true);
1423 static void i965_update_wm(struct drm_crtc
*unused_crtc
)
1425 struct drm_device
*dev
= unused_crtc
->dev
;
1426 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1427 struct drm_crtc
*crtc
;
1432 /* Calc sr entries for one plane configs */
1433 crtc
= single_enabled_crtc(dev
);
1435 /* self-refresh has much higher latency */
1436 static const int sr_latency_ns
= 12000;
1437 const struct drm_display_mode
*adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1438 int clock
= adjusted_mode
->crtc_clock
;
1439 int htotal
= adjusted_mode
->crtc_htotal
;
1440 int hdisplay
= to_intel_crtc(crtc
)->config
->pipe_src_w
;
1441 int cpp
= drm_format_plane_cpp(crtc
->primary
->state
->fb
->pixel_format
, 0);
1442 unsigned long line_time_us
;
1445 line_time_us
= max(htotal
* 1000 / clock
, 1);
1447 /* Use ns/us then divide to preserve precision */
1448 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1450 entries
= DIV_ROUND_UP(entries
, I915_FIFO_LINE_SIZE
);
1451 srwm
= I965_FIFO_SIZE
- entries
;
1455 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1458 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1459 cpp
* crtc
->cursor
->state
->crtc_w
;
1460 entries
= DIV_ROUND_UP(entries
,
1461 i965_cursor_wm_info
.cacheline_size
);
1462 cursor_sr
= i965_cursor_wm_info
.fifo_size
-
1463 (entries
+ i965_cursor_wm_info
.guard_size
);
1465 if (cursor_sr
> i965_cursor_wm_info
.max_wm
)
1466 cursor_sr
= i965_cursor_wm_info
.max_wm
;
1468 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1469 "cursor %d\n", srwm
, cursor_sr
);
1471 cxsr_enabled
= true;
1473 cxsr_enabled
= false;
1474 /* Turn off self refresh if both pipes are enabled */
1475 intel_set_memory_cxsr(dev_priv
, false);
1478 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1481 /* 965 has limitations... */
1482 I915_WRITE(DSPFW1
, FW_WM(srwm
, SR
) |
1486 I915_WRITE(DSPFW2
, FW_WM(8, CURSORA
) |
1487 FW_WM(8, PLANEC_OLD
));
1488 /* update cursor SR watermark */
1489 I915_WRITE(DSPFW3
, FW_WM(cursor_sr
, CURSOR_SR
));
1492 intel_set_memory_cxsr(dev_priv
, true);
1497 static void i9xx_update_wm(struct drm_crtc
*unused_crtc
)
1499 struct drm_device
*dev
= unused_crtc
->dev
;
1500 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1501 const struct intel_watermark_params
*wm_info
;
1506 int planea_wm
, planeb_wm
;
1507 struct drm_crtc
*crtc
, *enabled
= NULL
;
1510 wm_info
= &i945_wm_info
;
1511 else if (!IS_GEN2(dev
))
1512 wm_info
= &i915_wm_info
;
1514 wm_info
= &i830_a_wm_info
;
1516 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 0);
1517 crtc
= intel_get_crtc_for_plane(dev
, 0);
1518 if (intel_crtc_active(crtc
)) {
1519 const struct drm_display_mode
*adjusted_mode
;
1520 int cpp
= drm_format_plane_cpp(crtc
->primary
->state
->fb
->pixel_format
, 0);
1524 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1525 planea_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1526 wm_info
, fifo_size
, cpp
,
1527 pessimal_latency_ns
);
1530 planea_wm
= fifo_size
- wm_info
->guard_size
;
1531 if (planea_wm
> (long)wm_info
->max_wm
)
1532 planea_wm
= wm_info
->max_wm
;
1536 wm_info
= &i830_bc_wm_info
;
1538 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 1);
1539 crtc
= intel_get_crtc_for_plane(dev
, 1);
1540 if (intel_crtc_active(crtc
)) {
1541 const struct drm_display_mode
*adjusted_mode
;
1542 int cpp
= drm_format_plane_cpp(crtc
->primary
->state
->fb
->pixel_format
, 0);
1546 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1547 planeb_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1548 wm_info
, fifo_size
, cpp
,
1549 pessimal_latency_ns
);
1550 if (enabled
== NULL
)
1555 planeb_wm
= fifo_size
- wm_info
->guard_size
;
1556 if (planeb_wm
> (long)wm_info
->max_wm
)
1557 planeb_wm
= wm_info
->max_wm
;
1560 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm
, planeb_wm
);
1562 if (IS_I915GM(dev
) && enabled
) {
1563 struct drm_i915_gem_object
*obj
;
1565 obj
= intel_fb_obj(enabled
->primary
->state
->fb
);
1567 /* self-refresh seems busted with untiled */
1568 if (obj
->tiling_mode
== I915_TILING_NONE
)
1573 * Overlay gets an aggressive default since video jitter is bad.
1577 /* Play safe and disable self-refresh before adjusting watermarks. */
1578 intel_set_memory_cxsr(dev_priv
, false);
1580 /* Calc sr entries for one plane configs */
1581 if (HAS_FW_BLC(dev
) && enabled
) {
1582 /* self-refresh has much higher latency */
1583 static const int sr_latency_ns
= 6000;
1584 const struct drm_display_mode
*adjusted_mode
= &to_intel_crtc(enabled
)->config
->base
.adjusted_mode
;
1585 int clock
= adjusted_mode
->crtc_clock
;
1586 int htotal
= adjusted_mode
->crtc_htotal
;
1587 int hdisplay
= to_intel_crtc(enabled
)->config
->pipe_src_w
;
1588 int cpp
= drm_format_plane_cpp(enabled
->primary
->state
->fb
->pixel_format
, 0);
1589 unsigned long line_time_us
;
1592 line_time_us
= max(htotal
* 1000 / clock
, 1);
1594 /* Use ns/us then divide to preserve precision */
1595 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1597 entries
= DIV_ROUND_UP(entries
, wm_info
->cacheline_size
);
1598 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries
);
1599 srwm
= wm_info
->fifo_size
- entries
;
1603 if (IS_I945G(dev
) || IS_I945GM(dev
))
1604 I915_WRITE(FW_BLC_SELF
,
1605 FW_BLC_SELF_FIFO_MASK
| (srwm
& 0xff));
1606 else if (IS_I915GM(dev
))
1607 I915_WRITE(FW_BLC_SELF
, srwm
& 0x3f);
1610 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1611 planea_wm
, planeb_wm
, cwm
, srwm
);
1613 fwater_lo
= ((planeb_wm
& 0x3f) << 16) | (planea_wm
& 0x3f);
1614 fwater_hi
= (cwm
& 0x1f);
1616 /* Set request length to 8 cachelines per fetch */
1617 fwater_lo
= fwater_lo
| (1 << 24) | (1 << 8);
1618 fwater_hi
= fwater_hi
| (1 << 8);
1620 I915_WRITE(FW_BLC
, fwater_lo
);
1621 I915_WRITE(FW_BLC2
, fwater_hi
);
1624 intel_set_memory_cxsr(dev_priv
, true);
1627 static void i845_update_wm(struct drm_crtc
*unused_crtc
)
1629 struct drm_device
*dev
= unused_crtc
->dev
;
1630 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1631 struct drm_crtc
*crtc
;
1632 const struct drm_display_mode
*adjusted_mode
;
1636 crtc
= single_enabled_crtc(dev
);
1640 adjusted_mode
= &to_intel_crtc(crtc
)->config
->base
.adjusted_mode
;
1641 planea_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1643 dev_priv
->display
.get_fifo_size(dev
, 0),
1644 4, pessimal_latency_ns
);
1645 fwater_lo
= I915_READ(FW_BLC
) & ~0xfff;
1646 fwater_lo
|= (3<<8) | planea_wm
;
1648 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm
);
1650 I915_WRITE(FW_BLC
, fwater_lo
);
1653 uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state
*pipe_config
)
1655 uint32_t pixel_rate
;
1657 pixel_rate
= pipe_config
->base
.adjusted_mode
.crtc_clock
;
1659 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1660 * adjust the pixel_rate here. */
1662 if (pipe_config
->pch_pfit
.enabled
) {
1663 uint64_t pipe_w
, pipe_h
, pfit_w
, pfit_h
;
1664 uint32_t pfit_size
= pipe_config
->pch_pfit
.size
;
1666 pipe_w
= pipe_config
->pipe_src_w
;
1667 pipe_h
= pipe_config
->pipe_src_h
;
1669 pfit_w
= (pfit_size
>> 16) & 0xFFFF;
1670 pfit_h
= pfit_size
& 0xFFFF;
1671 if (pipe_w
< pfit_w
)
1673 if (pipe_h
< pfit_h
)
1676 if (WARN_ON(!pfit_w
|| !pfit_h
))
1679 pixel_rate
= div_u64((uint64_t) pixel_rate
* pipe_w
* pipe_h
,
1686 /* latency must be in 0.1us units. */
1687 static uint32_t ilk_wm_method1(uint32_t pixel_rate
, uint8_t cpp
, uint32_t latency
)
1691 if (WARN(latency
== 0, "Latency value missing\n"))
1694 ret
= (uint64_t) pixel_rate
* cpp
* latency
;
1695 ret
= DIV_ROUND_UP_ULL(ret
, 64 * 10000) + 2;
1700 /* latency must be in 0.1us units. */
1701 static uint32_t ilk_wm_method2(uint32_t pixel_rate
, uint32_t pipe_htotal
,
1702 uint32_t horiz_pixels
, uint8_t cpp
,
1707 if (WARN(latency
== 0, "Latency value missing\n"))
1709 if (WARN_ON(!pipe_htotal
))
1712 ret
= (latency
* pixel_rate
) / (pipe_htotal
* 10000);
1713 ret
= (ret
+ 1) * horiz_pixels
* cpp
;
1714 ret
= DIV_ROUND_UP(ret
, 64) + 2;
1718 static uint32_t ilk_wm_fbc(uint32_t pri_val
, uint32_t horiz_pixels
,
1722 * Neither of these should be possible since this function shouldn't be
1723 * called if the CRTC is off or the plane is invisible. But let's be
1724 * extra paranoid to avoid a potential divide-by-zero if we screw up
1725 * elsewhere in the driver.
1729 if (WARN_ON(!horiz_pixels
))
1732 return DIV_ROUND_UP(pri_val
* 64, horiz_pixels
* cpp
) + 2;
1735 struct ilk_wm_maximums
{
1743 * For both WM_PIPE and WM_LP.
1744 * mem_value must be in 0.1us units.
1746 static uint32_t ilk_compute_pri_wm(const struct intel_crtc_state
*cstate
,
1747 const struct intel_plane_state
*pstate
,
1751 int cpp
= pstate
->base
.fb
?
1752 drm_format_plane_cpp(pstate
->base
.fb
->pixel_format
, 0) : 0;
1753 uint32_t method1
, method2
;
1755 if (!cstate
->base
.active
|| !pstate
->visible
)
1758 method1
= ilk_wm_method1(ilk_pipe_pixel_rate(cstate
), cpp
, mem_value
);
1763 method2
= ilk_wm_method2(ilk_pipe_pixel_rate(cstate
),
1764 cstate
->base
.adjusted_mode
.crtc_htotal
,
1765 drm_rect_width(&pstate
->dst
),
1768 return min(method1
, method2
);
1772 * For both WM_PIPE and WM_LP.
1773 * mem_value must be in 0.1us units.
1775 static uint32_t ilk_compute_spr_wm(const struct intel_crtc_state
*cstate
,
1776 const struct intel_plane_state
*pstate
,
1779 int cpp
= pstate
->base
.fb
?
1780 drm_format_plane_cpp(pstate
->base
.fb
->pixel_format
, 0) : 0;
1781 uint32_t method1
, method2
;
1783 if (!cstate
->base
.active
|| !pstate
->visible
)
1786 method1
= ilk_wm_method1(ilk_pipe_pixel_rate(cstate
), cpp
, mem_value
);
1787 method2
= ilk_wm_method2(ilk_pipe_pixel_rate(cstate
),
1788 cstate
->base
.adjusted_mode
.crtc_htotal
,
1789 drm_rect_width(&pstate
->dst
),
1791 return min(method1
, method2
);
1795 * For both WM_PIPE and WM_LP.
1796 * mem_value must be in 0.1us units.
1798 static uint32_t ilk_compute_cur_wm(const struct intel_crtc_state
*cstate
,
1799 const struct intel_plane_state
*pstate
,
1803 * We treat the cursor plane as always-on for the purposes of watermark
1804 * calculation. Until we have two-stage watermark programming merged,
1805 * this is necessary to avoid flickering.
1808 int width
= pstate
->visible
? pstate
->base
.crtc_w
: 64;
1810 if (!cstate
->base
.active
)
1813 return ilk_wm_method2(ilk_pipe_pixel_rate(cstate
),
1814 cstate
->base
.adjusted_mode
.crtc_htotal
,
1815 width
, cpp
, mem_value
);
1818 /* Only for WM_LP. */
1819 static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state
*cstate
,
1820 const struct intel_plane_state
*pstate
,
1823 int cpp
= pstate
->base
.fb
?
1824 drm_format_plane_cpp(pstate
->base
.fb
->pixel_format
, 0) : 0;
1826 if (!cstate
->base
.active
|| !pstate
->visible
)
1829 return ilk_wm_fbc(pri_val
, drm_rect_width(&pstate
->dst
), cpp
);
1832 static unsigned int ilk_display_fifo_size(const struct drm_device
*dev
)
1834 if (INTEL_INFO(dev
)->gen
>= 8)
1836 else if (INTEL_INFO(dev
)->gen
>= 7)
1842 static unsigned int ilk_plane_wm_reg_max(const struct drm_device
*dev
,
1843 int level
, bool is_sprite
)
1845 if (INTEL_INFO(dev
)->gen
>= 8)
1846 /* BDW primary/sprite plane watermarks */
1847 return level
== 0 ? 255 : 2047;
1848 else if (INTEL_INFO(dev
)->gen
>= 7)
1849 /* IVB/HSW primary/sprite plane watermarks */
1850 return level
== 0 ? 127 : 1023;
1851 else if (!is_sprite
)
1852 /* ILK/SNB primary plane watermarks */
1853 return level
== 0 ? 127 : 511;
1855 /* ILK/SNB sprite plane watermarks */
1856 return level
== 0 ? 63 : 255;
1859 static unsigned int ilk_cursor_wm_reg_max(const struct drm_device
*dev
,
1862 if (INTEL_INFO(dev
)->gen
>= 7)
1863 return level
== 0 ? 63 : 255;
1865 return level
== 0 ? 31 : 63;
1868 static unsigned int ilk_fbc_wm_reg_max(const struct drm_device
*dev
)
1870 if (INTEL_INFO(dev
)->gen
>= 8)
1876 /* Calculate the maximum primary/sprite plane watermark */
1877 static unsigned int ilk_plane_wm_max(const struct drm_device
*dev
,
1879 const struct intel_wm_config
*config
,
1880 enum intel_ddb_partitioning ddb_partitioning
,
1883 unsigned int fifo_size
= ilk_display_fifo_size(dev
);
1885 /* if sprites aren't enabled, sprites get nothing */
1886 if (is_sprite
&& !config
->sprites_enabled
)
1889 /* HSW allows LP1+ watermarks even with multiple pipes */
1890 if (level
== 0 || config
->num_pipes_active
> 1) {
1891 fifo_size
/= INTEL_INFO(dev
)->num_pipes
;
1894 * For some reason the non self refresh
1895 * FIFO size is only half of the self
1896 * refresh FIFO size on ILK/SNB.
1898 if (INTEL_INFO(dev
)->gen
<= 6)
1902 if (config
->sprites_enabled
) {
1903 /* level 0 is always calculated with 1:1 split */
1904 if (level
> 0 && ddb_partitioning
== INTEL_DDB_PART_5_6
) {
1913 /* clamp to max that the registers can hold */
1914 return min(fifo_size
, ilk_plane_wm_reg_max(dev
, level
, is_sprite
));
1917 /* Calculate the maximum cursor plane watermark */
1918 static unsigned int ilk_cursor_wm_max(const struct drm_device
*dev
,
1920 const struct intel_wm_config
*config
)
1922 /* HSW LP1+ watermarks w/ multiple pipes */
1923 if (level
> 0 && config
->num_pipes_active
> 1)
1926 /* otherwise just report max that registers can hold */
1927 return ilk_cursor_wm_reg_max(dev
, level
);
1930 static void ilk_compute_wm_maximums(const struct drm_device
*dev
,
1932 const struct intel_wm_config
*config
,
1933 enum intel_ddb_partitioning ddb_partitioning
,
1934 struct ilk_wm_maximums
*max
)
1936 max
->pri
= ilk_plane_wm_max(dev
, level
, config
, ddb_partitioning
, false);
1937 max
->spr
= ilk_plane_wm_max(dev
, level
, config
, ddb_partitioning
, true);
1938 max
->cur
= ilk_cursor_wm_max(dev
, level
, config
);
1939 max
->fbc
= ilk_fbc_wm_reg_max(dev
);
1942 static void ilk_compute_wm_reg_maximums(struct drm_device
*dev
,
1944 struct ilk_wm_maximums
*max
)
1946 max
->pri
= ilk_plane_wm_reg_max(dev
, level
, false);
1947 max
->spr
= ilk_plane_wm_reg_max(dev
, level
, true);
1948 max
->cur
= ilk_cursor_wm_reg_max(dev
, level
);
1949 max
->fbc
= ilk_fbc_wm_reg_max(dev
);
1952 static bool ilk_validate_wm_level(int level
,
1953 const struct ilk_wm_maximums
*max
,
1954 struct intel_wm_level
*result
)
1958 /* already determined to be invalid? */
1959 if (!result
->enable
)
1962 result
->enable
= result
->pri_val
<= max
->pri
&&
1963 result
->spr_val
<= max
->spr
&&
1964 result
->cur_val
<= max
->cur
;
1966 ret
= result
->enable
;
1969 * HACK until we can pre-compute everything,
1970 * and thus fail gracefully if LP0 watermarks
1973 if (level
== 0 && !result
->enable
) {
1974 if (result
->pri_val
> max
->pri
)
1975 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1976 level
, result
->pri_val
, max
->pri
);
1977 if (result
->spr_val
> max
->spr
)
1978 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1979 level
, result
->spr_val
, max
->spr
);
1980 if (result
->cur_val
> max
->cur
)
1981 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
1982 level
, result
->cur_val
, max
->cur
);
1984 result
->pri_val
= min_t(uint32_t, result
->pri_val
, max
->pri
);
1985 result
->spr_val
= min_t(uint32_t, result
->spr_val
, max
->spr
);
1986 result
->cur_val
= min_t(uint32_t, result
->cur_val
, max
->cur
);
1987 result
->enable
= true;
1993 static void ilk_compute_wm_level(const struct drm_i915_private
*dev_priv
,
1994 const struct intel_crtc
*intel_crtc
,
1996 struct intel_crtc_state
*cstate
,
1997 struct intel_plane_state
*pristate
,
1998 struct intel_plane_state
*sprstate
,
1999 struct intel_plane_state
*curstate
,
2000 struct intel_wm_level
*result
)
2002 uint16_t pri_latency
= dev_priv
->wm
.pri_latency
[level
];
2003 uint16_t spr_latency
= dev_priv
->wm
.spr_latency
[level
];
2004 uint16_t cur_latency
= dev_priv
->wm
.cur_latency
[level
];
2006 /* WM1+ latency values stored in 0.5us units */
2013 result
->pri_val
= ilk_compute_pri_wm(cstate
, pristate
,
2014 pri_latency
, level
);
2015 result
->spr_val
= ilk_compute_spr_wm(cstate
, sprstate
, spr_latency
);
2016 result
->cur_val
= ilk_compute_cur_wm(cstate
, curstate
, cur_latency
);
2017 result
->fbc_val
= ilk_compute_fbc_wm(cstate
, pristate
, result
->pri_val
);
2018 result
->enable
= true;
2022 hsw_compute_linetime_wm(struct drm_device
*dev
,
2023 struct intel_crtc_state
*cstate
)
2025 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2026 const struct drm_display_mode
*adjusted_mode
=
2027 &cstate
->base
.adjusted_mode
;
2028 u32 linetime
, ips_linetime
;
2030 if (!cstate
->base
.active
)
2032 if (WARN_ON(adjusted_mode
->crtc_clock
== 0))
2034 if (WARN_ON(dev_priv
->cdclk_freq
== 0))
2037 /* The WM are computed with base on how long it takes to fill a single
2038 * row at the given clock rate, multiplied by 8.
2040 linetime
= DIV_ROUND_CLOSEST(adjusted_mode
->crtc_htotal
* 1000 * 8,
2041 adjusted_mode
->crtc_clock
);
2042 ips_linetime
= DIV_ROUND_CLOSEST(adjusted_mode
->crtc_htotal
* 1000 * 8,
2043 dev_priv
->cdclk_freq
);
2045 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime
) |
2046 PIPE_WM_LINETIME_TIME(linetime
);
2049 static void intel_read_wm_latency(struct drm_device
*dev
, uint16_t wm
[8])
2051 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2056 int level
, max_level
= ilk_wm_max_level(dev
);
2058 /* read the first set of memory latencies[0:3] */
2059 val
= 0; /* data0 to be programmed to 0 for first set */
2060 mutex_lock(&dev_priv
->rps
.hw_lock
);
2061 ret
= sandybridge_pcode_read(dev_priv
,
2062 GEN9_PCODE_READ_MEM_LATENCY
,
2064 mutex_unlock(&dev_priv
->rps
.hw_lock
);
2067 DRM_ERROR("SKL Mailbox read error = %d\n", ret
);
2071 wm
[0] = val
& GEN9_MEM_LATENCY_LEVEL_MASK
;
2072 wm
[1] = (val
>> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT
) &
2073 GEN9_MEM_LATENCY_LEVEL_MASK
;
2074 wm
[2] = (val
>> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT
) &
2075 GEN9_MEM_LATENCY_LEVEL_MASK
;
2076 wm
[3] = (val
>> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT
) &
2077 GEN9_MEM_LATENCY_LEVEL_MASK
;
2079 /* read the second set of memory latencies[4:7] */
2080 val
= 1; /* data0 to be programmed to 1 for second set */
2081 mutex_lock(&dev_priv
->rps
.hw_lock
);
2082 ret
= sandybridge_pcode_read(dev_priv
,
2083 GEN9_PCODE_READ_MEM_LATENCY
,
2085 mutex_unlock(&dev_priv
->rps
.hw_lock
);
2087 DRM_ERROR("SKL Mailbox read error = %d\n", ret
);
2091 wm
[4] = val
& GEN9_MEM_LATENCY_LEVEL_MASK
;
2092 wm
[5] = (val
>> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT
) &
2093 GEN9_MEM_LATENCY_LEVEL_MASK
;
2094 wm
[6] = (val
>> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT
) &
2095 GEN9_MEM_LATENCY_LEVEL_MASK
;
2096 wm
[7] = (val
>> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT
) &
2097 GEN9_MEM_LATENCY_LEVEL_MASK
;
2100 * WaWmMemoryReadLatency:skl
2102 * punit doesn't take into account the read latency so we need
2103 * to add 2us to the various latency levels we retrieve from
2105 * - W0 is a bit special in that it's the only level that
2106 * can't be disabled if we want to have display working, so
2107 * we always add 2us there.
2108 * - For levels >=1, punit returns 0us latency when they are
2109 * disabled, so we respect that and don't add 2us then
2111 * Additionally, if a level n (n > 1) has a 0us latency, all
2112 * levels m (m >= n) need to be disabled. We make sure to
2113 * sanitize the values out of the punit to satisfy this
2117 for (level
= 1; level
<= max_level
; level
++)
2121 for (i
= level
+ 1; i
<= max_level
; i
++)
2126 } else if (IS_HASWELL(dev
) || IS_BROADWELL(dev
)) {
2127 uint64_t sskpd
= I915_READ64(MCH_SSKPD
);
2129 wm
[0] = (sskpd
>> 56) & 0xFF;
2131 wm
[0] = sskpd
& 0xF;
2132 wm
[1] = (sskpd
>> 4) & 0xFF;
2133 wm
[2] = (sskpd
>> 12) & 0xFF;
2134 wm
[3] = (sskpd
>> 20) & 0x1FF;
2135 wm
[4] = (sskpd
>> 32) & 0x1FF;
2136 } else if (INTEL_INFO(dev
)->gen
>= 6) {
2137 uint32_t sskpd
= I915_READ(MCH_SSKPD
);
2139 wm
[0] = (sskpd
>> SSKPD_WM0_SHIFT
) & SSKPD_WM_MASK
;
2140 wm
[1] = (sskpd
>> SSKPD_WM1_SHIFT
) & SSKPD_WM_MASK
;
2141 wm
[2] = (sskpd
>> SSKPD_WM2_SHIFT
) & SSKPD_WM_MASK
;
2142 wm
[3] = (sskpd
>> SSKPD_WM3_SHIFT
) & SSKPD_WM_MASK
;
2143 } else if (INTEL_INFO(dev
)->gen
>= 5) {
2144 uint32_t mltr
= I915_READ(MLTR_ILK
);
2146 /* ILK primary LP0 latency is 700 ns */
2148 wm
[1] = (mltr
>> MLTR_WM1_SHIFT
) & ILK_SRLT_MASK
;
2149 wm
[2] = (mltr
>> MLTR_WM2_SHIFT
) & ILK_SRLT_MASK
;
2153 static void intel_fixup_spr_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2155 /* ILK sprite LP0 latency is 1300 ns */
2156 if (INTEL_INFO(dev
)->gen
== 5)
2160 static void intel_fixup_cur_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2162 /* ILK cursor LP0 latency is 1300 ns */
2163 if (INTEL_INFO(dev
)->gen
== 5)
2166 /* WaDoubleCursorLP3Latency:ivb */
2167 if (IS_IVYBRIDGE(dev
))
2171 int ilk_wm_max_level(const struct drm_device
*dev
)
2173 /* how many WM levels are we expecting */
2174 if (INTEL_INFO(dev
)->gen
>= 9)
2176 else if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2178 else if (INTEL_INFO(dev
)->gen
>= 6)
2184 static void intel_print_wm_latency(struct drm_device
*dev
,
2186 const uint16_t wm
[8])
2188 int level
, max_level
= ilk_wm_max_level(dev
);
2190 for (level
= 0; level
<= max_level
; level
++) {
2191 unsigned int latency
= wm
[level
];
2194 DRM_ERROR("%s WM%d latency not provided\n",
2200 * - latencies are in us on gen9.
2201 * - before then, WM1+ latency values are in 0.5us units
2208 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2209 name
, level
, wm
[level
],
2210 latency
/ 10, latency
% 10);
2214 static bool ilk_increase_wm_latency(struct drm_i915_private
*dev_priv
,
2215 uint16_t wm
[5], uint16_t min
)
2217 int level
, max_level
= ilk_wm_max_level(dev_priv
->dev
);
2222 wm
[0] = max(wm
[0], min
);
2223 for (level
= 1; level
<= max_level
; level
++)
2224 wm
[level
] = max_t(uint16_t, wm
[level
], DIV_ROUND_UP(min
, 5));
2229 static void snb_wm_latency_quirk(struct drm_device
*dev
)
2231 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2235 * The BIOS provided WM memory latency values are often
2236 * inadequate for high resolution displays. Adjust them.
2238 changed
= ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.pri_latency
, 12) |
2239 ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.spr_latency
, 12) |
2240 ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.cur_latency
, 12);
2245 DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
2246 intel_print_wm_latency(dev
, "Primary", dev_priv
->wm
.pri_latency
);
2247 intel_print_wm_latency(dev
, "Sprite", dev_priv
->wm
.spr_latency
);
2248 intel_print_wm_latency(dev
, "Cursor", dev_priv
->wm
.cur_latency
);
2251 static void ilk_setup_wm_latency(struct drm_device
*dev
)
2253 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2255 intel_read_wm_latency(dev
, dev_priv
->wm
.pri_latency
);
2257 memcpy(dev_priv
->wm
.spr_latency
, dev_priv
->wm
.pri_latency
,
2258 sizeof(dev_priv
->wm
.pri_latency
));
2259 memcpy(dev_priv
->wm
.cur_latency
, dev_priv
->wm
.pri_latency
,
2260 sizeof(dev_priv
->wm
.pri_latency
));
2262 intel_fixup_spr_wm_latency(dev
, dev_priv
->wm
.spr_latency
);
2263 intel_fixup_cur_wm_latency(dev
, dev_priv
->wm
.cur_latency
);
2265 intel_print_wm_latency(dev
, "Primary", dev_priv
->wm
.pri_latency
);
2266 intel_print_wm_latency(dev
, "Sprite", dev_priv
->wm
.spr_latency
);
2267 intel_print_wm_latency(dev
, "Cursor", dev_priv
->wm
.cur_latency
);
2270 snb_wm_latency_quirk(dev
);
2273 static void skl_setup_wm_latency(struct drm_device
*dev
)
2275 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2277 intel_read_wm_latency(dev
, dev_priv
->wm
.skl_latency
);
2278 intel_print_wm_latency(dev
, "Gen9 Plane", dev_priv
->wm
.skl_latency
);
2281 /* Compute new watermarks for the pipe */
2282 static int ilk_compute_pipe_wm(struct intel_crtc
*intel_crtc
,
2283 struct drm_atomic_state
*state
)
2285 struct intel_pipe_wm
*pipe_wm
;
2286 struct drm_device
*dev
= intel_crtc
->base
.dev
;
2287 const struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2288 struct intel_crtc_state
*cstate
= NULL
;
2289 struct intel_plane
*intel_plane
;
2290 struct drm_plane_state
*ps
;
2291 struct intel_plane_state
*pristate
= NULL
;
2292 struct intel_plane_state
*sprstate
= NULL
;
2293 struct intel_plane_state
*curstate
= NULL
;
2294 int level
, max_level
= ilk_wm_max_level(dev
);
2295 /* LP0 watermark maximums depend on this pipe alone */
2296 struct intel_wm_config config
= {
2297 .num_pipes_active
= 1,
2299 struct ilk_wm_maximums max
;
2301 cstate
= intel_atomic_get_crtc_state(state
, intel_crtc
);
2303 return PTR_ERR(cstate
);
2305 pipe_wm
= &cstate
->wm
.optimal
.ilk
;
2306 memset(pipe_wm
, 0, sizeof(*pipe_wm
));
2308 for_each_intel_plane_on_crtc(dev
, intel_crtc
, intel_plane
) {
2309 ps
= drm_atomic_get_plane_state(state
,
2310 &intel_plane
->base
);
2314 if (intel_plane
->base
.type
== DRM_PLANE_TYPE_PRIMARY
)
2315 pristate
= to_intel_plane_state(ps
);
2316 else if (intel_plane
->base
.type
== DRM_PLANE_TYPE_OVERLAY
)
2317 sprstate
= to_intel_plane_state(ps
);
2318 else if (intel_plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
)
2319 curstate
= to_intel_plane_state(ps
);
2322 config
.sprites_enabled
= sprstate
->visible
;
2323 config
.sprites_scaled
= sprstate
->visible
&&
2324 (drm_rect_width(&sprstate
->dst
) != drm_rect_width(&sprstate
->src
) >> 16 ||
2325 drm_rect_height(&sprstate
->dst
) != drm_rect_height(&sprstate
->src
) >> 16);
2327 pipe_wm
->pipe_enabled
= cstate
->base
.active
;
2328 pipe_wm
->sprites_enabled
= config
.sprites_enabled
;
2329 pipe_wm
->sprites_scaled
= config
.sprites_scaled
;
2331 /* ILK/SNB: LP2+ watermarks only w/o sprites */
2332 if (INTEL_INFO(dev
)->gen
<= 6 && sprstate
->visible
)
2335 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
2336 if (config
.sprites_scaled
)
2339 ilk_compute_wm_level(dev_priv
, intel_crtc
, 0, cstate
,
2340 pristate
, sprstate
, curstate
, &pipe_wm
->wm
[0]);
2342 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2343 pipe_wm
->linetime
= hsw_compute_linetime_wm(dev
, cstate
);
2345 /* LP0 watermarks always use 1/2 DDB partitioning */
2346 ilk_compute_wm_maximums(dev
, 0, &config
, INTEL_DDB_PART_1_2
, &max
);
2348 /* At least LP0 must be valid */
2349 if (!ilk_validate_wm_level(0, &max
, &pipe_wm
->wm
[0]))
2352 ilk_compute_wm_reg_maximums(dev
, 1, &max
);
2354 for (level
= 1; level
<= max_level
; level
++) {
2355 struct intel_wm_level wm
= {};
2357 ilk_compute_wm_level(dev_priv
, intel_crtc
, level
, cstate
,
2358 pristate
, sprstate
, curstate
, &wm
);
2361 * Disable any watermark level that exceeds the
2362 * register maximums since such watermarks are
2365 if (!ilk_validate_wm_level(level
, &max
, &wm
))
2368 pipe_wm
->wm
[level
] = wm
;
2375 * Merge the watermarks from all active pipes for a specific level.
2377 static void ilk_merge_wm_level(struct drm_device
*dev
,
2379 struct intel_wm_level
*ret_wm
)
2381 const struct intel_crtc
*intel_crtc
;
2383 ret_wm
->enable
= true;
2385 for_each_intel_crtc(dev
, intel_crtc
) {
2386 const struct intel_crtc_state
*cstate
=
2387 to_intel_crtc_state(intel_crtc
->base
.state
);
2388 const struct intel_pipe_wm
*active
= &cstate
->wm
.optimal
.ilk
;
2389 const struct intel_wm_level
*wm
= &active
->wm
[level
];
2391 if (!active
->pipe_enabled
)
2395 * The watermark values may have been used in the past,
2396 * so we must maintain them in the registers for some
2397 * time even if the level is now disabled.
2400 ret_wm
->enable
= false;
2402 ret_wm
->pri_val
= max(ret_wm
->pri_val
, wm
->pri_val
);
2403 ret_wm
->spr_val
= max(ret_wm
->spr_val
, wm
->spr_val
);
2404 ret_wm
->cur_val
= max(ret_wm
->cur_val
, wm
->cur_val
);
2405 ret_wm
->fbc_val
= max(ret_wm
->fbc_val
, wm
->fbc_val
);
2410 * Merge all low power watermarks for all active pipes.
2412 static void ilk_wm_merge(struct drm_device
*dev
,
2413 const struct intel_wm_config
*config
,
2414 const struct ilk_wm_maximums
*max
,
2415 struct intel_pipe_wm
*merged
)
2417 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2418 int level
, max_level
= ilk_wm_max_level(dev
);
2419 int last_enabled_level
= max_level
;
2421 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2422 if ((INTEL_INFO(dev
)->gen
<= 6 || IS_IVYBRIDGE(dev
)) &&
2423 config
->num_pipes_active
> 1)
2426 /* ILK: FBC WM must be disabled always */
2427 merged
->fbc_wm_enabled
= INTEL_INFO(dev
)->gen
>= 6;
2429 /* merge each WM1+ level */
2430 for (level
= 1; level
<= max_level
; level
++) {
2431 struct intel_wm_level
*wm
= &merged
->wm
[level
];
2433 ilk_merge_wm_level(dev
, level
, wm
);
2435 if (level
> last_enabled_level
)
2437 else if (!ilk_validate_wm_level(level
, max
, wm
))
2438 /* make sure all following levels get disabled */
2439 last_enabled_level
= level
- 1;
2442 * The spec says it is preferred to disable
2443 * FBC WMs instead of disabling a WM level.
2445 if (wm
->fbc_val
> max
->fbc
) {
2447 merged
->fbc_wm_enabled
= false;
2452 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2454 * FIXME this is racy. FBC might get enabled later.
2455 * What we should check here is whether FBC can be
2456 * enabled sometime later.
2458 if (IS_GEN5(dev
) && !merged
->fbc_wm_enabled
&&
2459 intel_fbc_is_active(dev_priv
)) {
2460 for (level
= 2; level
<= max_level
; level
++) {
2461 struct intel_wm_level
*wm
= &merged
->wm
[level
];
2468 static int ilk_wm_lp_to_level(int wm_lp
, const struct intel_pipe_wm
*pipe_wm
)
2470 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2471 return wm_lp
+ (wm_lp
>= 2 && pipe_wm
->wm
[4].enable
);
2474 /* The value we need to program into the WM_LPx latency field */
2475 static unsigned int ilk_wm_lp_latency(struct drm_device
*dev
, int level
)
2477 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2479 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2482 return dev_priv
->wm
.pri_latency
[level
];
2485 static void ilk_compute_wm_results(struct drm_device
*dev
,
2486 const struct intel_pipe_wm
*merged
,
2487 enum intel_ddb_partitioning partitioning
,
2488 struct ilk_wm_values
*results
)
2490 struct intel_crtc
*intel_crtc
;
2493 results
->enable_fbc_wm
= merged
->fbc_wm_enabled
;
2494 results
->partitioning
= partitioning
;
2496 /* LP1+ register values */
2497 for (wm_lp
= 1; wm_lp
<= 3; wm_lp
++) {
2498 const struct intel_wm_level
*r
;
2500 level
= ilk_wm_lp_to_level(wm_lp
, merged
);
2502 r
= &merged
->wm
[level
];
2505 * Maintain the watermark values even if the level is
2506 * disabled. Doing otherwise could cause underruns.
2508 results
->wm_lp
[wm_lp
- 1] =
2509 (ilk_wm_lp_latency(dev
, level
) << WM1_LP_LATENCY_SHIFT
) |
2510 (r
->pri_val
<< WM1_LP_SR_SHIFT
) |
2514 results
->wm_lp
[wm_lp
- 1] |= WM1_LP_SR_EN
;
2516 if (INTEL_INFO(dev
)->gen
>= 8)
2517 results
->wm_lp
[wm_lp
- 1] |=
2518 r
->fbc_val
<< WM1_LP_FBC_SHIFT_BDW
;
2520 results
->wm_lp
[wm_lp
- 1] |=
2521 r
->fbc_val
<< WM1_LP_FBC_SHIFT
;
2524 * Always set WM1S_LP_EN when spr_val != 0, even if the
2525 * level is disabled. Doing otherwise could cause underruns.
2527 if (INTEL_INFO(dev
)->gen
<= 6 && r
->spr_val
) {
2528 WARN_ON(wm_lp
!= 1);
2529 results
->wm_lp_spr
[wm_lp
- 1] = WM1S_LP_EN
| r
->spr_val
;
2531 results
->wm_lp_spr
[wm_lp
- 1] = r
->spr_val
;
2534 /* LP0 register values */
2535 for_each_intel_crtc(dev
, intel_crtc
) {
2536 const struct intel_crtc_state
*cstate
=
2537 to_intel_crtc_state(intel_crtc
->base
.state
);
2538 enum pipe pipe
= intel_crtc
->pipe
;
2539 const struct intel_wm_level
*r
= &cstate
->wm
.optimal
.ilk
.wm
[0];
2541 if (WARN_ON(!r
->enable
))
2544 results
->wm_linetime
[pipe
] = cstate
->wm
.optimal
.ilk
.linetime
;
2546 results
->wm_pipe
[pipe
] =
2547 (r
->pri_val
<< WM0_PIPE_PLANE_SHIFT
) |
2548 (r
->spr_val
<< WM0_PIPE_SPRITE_SHIFT
) |
2553 /* Find the result with the highest level enabled. Check for enable_fbc_wm in
2554 * case both are at the same level. Prefer r1 in case they're the same. */
2555 static struct intel_pipe_wm
*ilk_find_best_result(struct drm_device
*dev
,
2556 struct intel_pipe_wm
*r1
,
2557 struct intel_pipe_wm
*r2
)
2559 int level
, max_level
= ilk_wm_max_level(dev
);
2560 int level1
= 0, level2
= 0;
2562 for (level
= 1; level
<= max_level
; level
++) {
2563 if (r1
->wm
[level
].enable
)
2565 if (r2
->wm
[level
].enable
)
2569 if (level1
== level2
) {
2570 if (r2
->fbc_wm_enabled
&& !r1
->fbc_wm_enabled
)
2574 } else if (level1
> level2
) {
2581 /* dirty bits used to track which watermarks need changes */
2582 #define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2583 #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2584 #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2585 #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2586 #define WM_DIRTY_FBC (1 << 24)
2587 #define WM_DIRTY_DDB (1 << 25)
2589 static unsigned int ilk_compute_wm_dirty(struct drm_i915_private
*dev_priv
,
2590 const struct ilk_wm_values
*old
,
2591 const struct ilk_wm_values
*new)
2593 unsigned int dirty
= 0;
2597 for_each_pipe(dev_priv
, pipe
) {
2598 if (old
->wm_linetime
[pipe
] != new->wm_linetime
[pipe
]) {
2599 dirty
|= WM_DIRTY_LINETIME(pipe
);
2600 /* Must disable LP1+ watermarks too */
2601 dirty
|= WM_DIRTY_LP_ALL
;
2604 if (old
->wm_pipe
[pipe
] != new->wm_pipe
[pipe
]) {
2605 dirty
|= WM_DIRTY_PIPE(pipe
);
2606 /* Must disable LP1+ watermarks too */
2607 dirty
|= WM_DIRTY_LP_ALL
;
2611 if (old
->enable_fbc_wm
!= new->enable_fbc_wm
) {
2612 dirty
|= WM_DIRTY_FBC
;
2613 /* Must disable LP1+ watermarks too */
2614 dirty
|= WM_DIRTY_LP_ALL
;
2617 if (old
->partitioning
!= new->partitioning
) {
2618 dirty
|= WM_DIRTY_DDB
;
2619 /* Must disable LP1+ watermarks too */
2620 dirty
|= WM_DIRTY_LP_ALL
;
2623 /* LP1+ watermarks already deemed dirty, no need to continue */
2624 if (dirty
& WM_DIRTY_LP_ALL
)
2627 /* Find the lowest numbered LP1+ watermark in need of an update... */
2628 for (wm_lp
= 1; wm_lp
<= 3; wm_lp
++) {
2629 if (old
->wm_lp
[wm_lp
- 1] != new->wm_lp
[wm_lp
- 1] ||
2630 old
->wm_lp_spr
[wm_lp
- 1] != new->wm_lp_spr
[wm_lp
- 1])
2634 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2635 for (; wm_lp
<= 3; wm_lp
++)
2636 dirty
|= WM_DIRTY_LP(wm_lp
);
2641 static bool _ilk_disable_lp_wm(struct drm_i915_private
*dev_priv
,
2644 struct ilk_wm_values
*previous
= &dev_priv
->wm
.hw
;
2645 bool changed
= false;
2647 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp
[2] & WM1_LP_SR_EN
) {
2648 previous
->wm_lp
[2] &= ~WM1_LP_SR_EN
;
2649 I915_WRITE(WM3_LP_ILK
, previous
->wm_lp
[2]);
2652 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp
[1] & WM1_LP_SR_EN
) {
2653 previous
->wm_lp
[1] &= ~WM1_LP_SR_EN
;
2654 I915_WRITE(WM2_LP_ILK
, previous
->wm_lp
[1]);
2657 if (dirty
& WM_DIRTY_LP(1) && previous
->wm_lp
[0] & WM1_LP_SR_EN
) {
2658 previous
->wm_lp
[0] &= ~WM1_LP_SR_EN
;
2659 I915_WRITE(WM1_LP_ILK
, previous
->wm_lp
[0]);
2664 * Don't touch WM1S_LP_EN here.
2665 * Doing so could cause underruns.
2672 * The spec says we shouldn't write when we don't need, because every write
2673 * causes WMs to be re-evaluated, expending some power.
2675 static void ilk_write_wm_values(struct drm_i915_private
*dev_priv
,
2676 struct ilk_wm_values
*results
)
2678 struct drm_device
*dev
= dev_priv
->dev
;
2679 struct ilk_wm_values
*previous
= &dev_priv
->wm
.hw
;
2683 dirty
= ilk_compute_wm_dirty(dev_priv
, previous
, results
);
2687 _ilk_disable_lp_wm(dev_priv
, dirty
);
2689 if (dirty
& WM_DIRTY_PIPE(PIPE_A
))
2690 I915_WRITE(WM0_PIPEA_ILK
, results
->wm_pipe
[0]);
2691 if (dirty
& WM_DIRTY_PIPE(PIPE_B
))
2692 I915_WRITE(WM0_PIPEB_ILK
, results
->wm_pipe
[1]);
2693 if (dirty
& WM_DIRTY_PIPE(PIPE_C
))
2694 I915_WRITE(WM0_PIPEC_IVB
, results
->wm_pipe
[2]);
2696 if (dirty
& WM_DIRTY_LINETIME(PIPE_A
))
2697 I915_WRITE(PIPE_WM_LINETIME(PIPE_A
), results
->wm_linetime
[0]);
2698 if (dirty
& WM_DIRTY_LINETIME(PIPE_B
))
2699 I915_WRITE(PIPE_WM_LINETIME(PIPE_B
), results
->wm_linetime
[1]);
2700 if (dirty
& WM_DIRTY_LINETIME(PIPE_C
))
2701 I915_WRITE(PIPE_WM_LINETIME(PIPE_C
), results
->wm_linetime
[2]);
2703 if (dirty
& WM_DIRTY_DDB
) {
2704 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
)) {
2705 val
= I915_READ(WM_MISC
);
2706 if (results
->partitioning
== INTEL_DDB_PART_1_2
)
2707 val
&= ~WM_MISC_DATA_PARTITION_5_6
;
2709 val
|= WM_MISC_DATA_PARTITION_5_6
;
2710 I915_WRITE(WM_MISC
, val
);
2712 val
= I915_READ(DISP_ARB_CTL2
);
2713 if (results
->partitioning
== INTEL_DDB_PART_1_2
)
2714 val
&= ~DISP_DATA_PARTITION_5_6
;
2716 val
|= DISP_DATA_PARTITION_5_6
;
2717 I915_WRITE(DISP_ARB_CTL2
, val
);
2721 if (dirty
& WM_DIRTY_FBC
) {
2722 val
= I915_READ(DISP_ARB_CTL
);
2723 if (results
->enable_fbc_wm
)
2724 val
&= ~DISP_FBC_WM_DIS
;
2726 val
|= DISP_FBC_WM_DIS
;
2727 I915_WRITE(DISP_ARB_CTL
, val
);
2730 if (dirty
& WM_DIRTY_LP(1) &&
2731 previous
->wm_lp_spr
[0] != results
->wm_lp_spr
[0])
2732 I915_WRITE(WM1S_LP_ILK
, results
->wm_lp_spr
[0]);
2734 if (INTEL_INFO(dev
)->gen
>= 7) {
2735 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp_spr
[1] != results
->wm_lp_spr
[1])
2736 I915_WRITE(WM2S_LP_IVB
, results
->wm_lp_spr
[1]);
2737 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp_spr
[2] != results
->wm_lp_spr
[2])
2738 I915_WRITE(WM3S_LP_IVB
, results
->wm_lp_spr
[2]);
2741 if (dirty
& WM_DIRTY_LP(1) && previous
->wm_lp
[0] != results
->wm_lp
[0])
2742 I915_WRITE(WM1_LP_ILK
, results
->wm_lp
[0]);
2743 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp
[1] != results
->wm_lp
[1])
2744 I915_WRITE(WM2_LP_ILK
, results
->wm_lp
[1]);
2745 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp
[2] != results
->wm_lp
[2])
2746 I915_WRITE(WM3_LP_ILK
, results
->wm_lp
[2]);
2748 dev_priv
->wm
.hw
= *results
;
2751 static bool ilk_disable_lp_wm(struct drm_device
*dev
)
2753 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2755 return _ilk_disable_lp_wm(dev_priv
, WM_DIRTY_LP_ALL
);
2759 * On gen9, we need to allocate Display Data Buffer (DDB) portions to the
2760 * different active planes.
2763 #define SKL_DDB_SIZE 896 /* in blocks */
2764 #define BXT_DDB_SIZE 512
2767 * Return the index of a plane in the SKL DDB and wm result arrays. Primary
2768 * plane is always in slot 0, cursor is always in slot I915_MAX_PLANES-1, and
2769 * other universal planes are in indices 1..n. Note that this may leave unused
2770 * indices between the top "sprite" plane and the cursor.
2773 skl_wm_plane_id(const struct intel_plane
*plane
)
2775 switch (plane
->base
.type
) {
2776 case DRM_PLANE_TYPE_PRIMARY
:
2778 case DRM_PLANE_TYPE_CURSOR
:
2779 return PLANE_CURSOR
;
2780 case DRM_PLANE_TYPE_OVERLAY
:
2781 return plane
->plane
+ 1;
2783 MISSING_CASE(plane
->base
.type
);
2784 return plane
->plane
;
2789 skl_ddb_get_pipe_allocation_limits(struct drm_device
*dev
,
2790 const struct intel_crtc_state
*cstate
,
2791 const struct intel_wm_config
*config
,
2792 struct skl_ddb_entry
*alloc
/* out */)
2794 struct drm_crtc
*for_crtc
= cstate
->base
.crtc
;
2795 struct drm_crtc
*crtc
;
2796 unsigned int pipe_size
, ddb_size
;
2797 int nth_active_pipe
;
2799 if (!cstate
->base
.active
) {
2805 if (IS_BROXTON(dev
))
2806 ddb_size
= BXT_DDB_SIZE
;
2808 ddb_size
= SKL_DDB_SIZE
;
2810 ddb_size
-= 4; /* 4 blocks for bypass path allocation */
2812 nth_active_pipe
= 0;
2813 for_each_crtc(dev
, crtc
) {
2814 if (!to_intel_crtc(crtc
)->active
)
2817 if (crtc
== for_crtc
)
2823 pipe_size
= ddb_size
/ config
->num_pipes_active
;
2824 alloc
->start
= nth_active_pipe
* ddb_size
/ config
->num_pipes_active
;
2825 alloc
->end
= alloc
->start
+ pipe_size
;
2828 static unsigned int skl_cursor_allocation(const struct intel_wm_config
*config
)
2830 if (config
->num_pipes_active
== 1)
2836 static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry
*entry
, u32 reg
)
2838 entry
->start
= reg
& 0x3ff;
2839 entry
->end
= (reg
>> 16) & 0x3ff;
2844 void skl_ddb_get_hw_state(struct drm_i915_private
*dev_priv
,
2845 struct skl_ddb_allocation
*ddb
/* out */)
2851 memset(ddb
, 0, sizeof(*ddb
));
2853 for_each_pipe(dev_priv
, pipe
) {
2854 enum intel_display_power_domain power_domain
;
2856 power_domain
= POWER_DOMAIN_PIPE(pipe
);
2857 if (!intel_display_power_get_if_enabled(dev_priv
, power_domain
))
2860 for_each_plane(dev_priv
, pipe
, plane
) {
2861 val
= I915_READ(PLANE_BUF_CFG(pipe
, plane
));
2862 skl_ddb_entry_init_from_hw(&ddb
->plane
[pipe
][plane
],
2866 val
= I915_READ(CUR_BUF_CFG(pipe
));
2867 skl_ddb_entry_init_from_hw(&ddb
->plane
[pipe
][PLANE_CURSOR
],
2870 intel_display_power_put(dev_priv
, power_domain
);
2875 skl_plane_relative_data_rate(const struct intel_crtc_state
*cstate
,
2876 const struct drm_plane_state
*pstate
,
2879 struct intel_crtc
*intel_crtc
= to_intel_crtc(cstate
->base
.crtc
);
2880 struct drm_framebuffer
*fb
= pstate
->fb
;
2882 /* for planar format */
2883 if (fb
->pixel_format
== DRM_FORMAT_NV12
) {
2884 if (y
) /* y-plane data rate */
2885 return intel_crtc
->config
->pipe_src_w
*
2886 intel_crtc
->config
->pipe_src_h
*
2887 drm_format_plane_cpp(fb
->pixel_format
, 0);
2888 else /* uv-plane data rate */
2889 return (intel_crtc
->config
->pipe_src_w
/2) *
2890 (intel_crtc
->config
->pipe_src_h
/2) *
2891 drm_format_plane_cpp(fb
->pixel_format
, 1);
2894 /* for packed formats */
2895 return intel_crtc
->config
->pipe_src_w
*
2896 intel_crtc
->config
->pipe_src_h
*
2897 drm_format_plane_cpp(fb
->pixel_format
, 0);
2901 * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching
2902 * a 8192x4096@32bpp framebuffer:
2903 * 3 * 4096 * 8192 * 4 < 2^32
2906 skl_get_total_relative_data_rate(const struct intel_crtc_state
*cstate
)
2908 struct intel_crtc
*intel_crtc
= to_intel_crtc(cstate
->base
.crtc
);
2909 struct drm_device
*dev
= intel_crtc
->base
.dev
;
2910 const struct intel_plane
*intel_plane
;
2911 unsigned int total_data_rate
= 0;
2913 for_each_intel_plane_on_crtc(dev
, intel_crtc
, intel_plane
) {
2914 const struct drm_plane_state
*pstate
= intel_plane
->base
.state
;
2916 if (pstate
->fb
== NULL
)
2919 if (intel_plane
->base
.type
== DRM_PLANE_TYPE_CURSOR
)
2923 total_data_rate
+= skl_plane_relative_data_rate(cstate
,
2927 if (pstate
->fb
->pixel_format
== DRM_FORMAT_NV12
)
2929 total_data_rate
+= skl_plane_relative_data_rate(cstate
,
2934 return total_data_rate
;
2938 skl_allocate_pipe_ddb(struct intel_crtc_state
*cstate
,
2939 struct skl_ddb_allocation
*ddb
/* out */)
2941 struct drm_crtc
*crtc
= cstate
->base
.crtc
;
2942 struct drm_device
*dev
= crtc
->dev
;
2943 struct drm_i915_private
*dev_priv
= to_i915(dev
);
2944 struct intel_wm_config
*config
= &dev_priv
->wm
.config
;
2945 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2946 struct intel_plane
*intel_plane
;
2947 enum pipe pipe
= intel_crtc
->pipe
;
2948 struct skl_ddb_entry
*alloc
= &ddb
->pipe
[pipe
];
2949 uint16_t alloc_size
, start
, cursor_blocks
;
2950 uint16_t minimum
[I915_MAX_PLANES
];
2951 uint16_t y_minimum
[I915_MAX_PLANES
];
2952 unsigned int total_data_rate
;
2954 skl_ddb_get_pipe_allocation_limits(dev
, cstate
, config
, alloc
);
2955 alloc_size
= skl_ddb_entry_size(alloc
);
2956 if (alloc_size
== 0) {
2957 memset(ddb
->plane
[pipe
], 0, sizeof(ddb
->plane
[pipe
]));
2958 memset(&ddb
->plane
[pipe
][PLANE_CURSOR
], 0,
2959 sizeof(ddb
->plane
[pipe
][PLANE_CURSOR
]));
2963 cursor_blocks
= skl_cursor_allocation(config
);
2964 ddb
->plane
[pipe
][PLANE_CURSOR
].start
= alloc
->end
- cursor_blocks
;
2965 ddb
->plane
[pipe
][PLANE_CURSOR
].end
= alloc
->end
;
2967 alloc_size
-= cursor_blocks
;
2968 alloc
->end
-= cursor_blocks
;
2970 /* 1. Allocate the mininum required blocks for each active plane */
2971 for_each_intel_plane_on_crtc(dev
, intel_crtc
, intel_plane
) {
2972 struct drm_plane
*plane
= &intel_plane
->base
;
2973 struct drm_framebuffer
*fb
= plane
->state
->fb
;
2974 int id
= skl_wm_plane_id(intel_plane
);
2978 if (plane
->type
== DRM_PLANE_TYPE_CURSOR
)
2982 alloc_size
-= minimum
[id
];
2983 y_minimum
[id
] = (fb
->pixel_format
== DRM_FORMAT_NV12
) ? 8 : 0;
2984 alloc_size
-= y_minimum
[id
];
2988 * 2. Distribute the remaining space in proportion to the amount of
2989 * data each plane needs to fetch from memory.
2991 * FIXME: we may not allocate every single block here.
2993 total_data_rate
= skl_get_total_relative_data_rate(cstate
);
2995 start
= alloc
->start
;
2996 for_each_intel_plane_on_crtc(dev
, intel_crtc
, intel_plane
) {
2997 struct drm_plane
*plane
= &intel_plane
->base
;
2998 struct drm_plane_state
*pstate
= intel_plane
->base
.state
;
2999 unsigned int data_rate
, y_data_rate
;
3000 uint16_t plane_blocks
, y_plane_blocks
= 0;
3001 int id
= skl_wm_plane_id(intel_plane
);
3003 if (pstate
->fb
== NULL
)
3005 if (plane
->type
== DRM_PLANE_TYPE_CURSOR
)
3008 data_rate
= skl_plane_relative_data_rate(cstate
, pstate
, 0);
3011 * allocation for (packed formats) or (uv-plane part of planar format):
3012 * promote the expression to 64 bits to avoid overflowing, the
3013 * result is < available as data_rate / total_data_rate < 1
3015 plane_blocks
= minimum
[id
];
3016 plane_blocks
+= div_u64((uint64_t)alloc_size
* data_rate
,
3019 ddb
->plane
[pipe
][id
].start
= start
;
3020 ddb
->plane
[pipe
][id
].end
= start
+ plane_blocks
;
3022 start
+= plane_blocks
;
3025 * allocation for y_plane part of planar format:
3027 if (pstate
->fb
->pixel_format
== DRM_FORMAT_NV12
) {
3028 y_data_rate
= skl_plane_relative_data_rate(cstate
,
3031 y_plane_blocks
= y_minimum
[id
];
3032 y_plane_blocks
+= div_u64((uint64_t)alloc_size
* y_data_rate
,
3035 ddb
->y_plane
[pipe
][id
].start
= start
;
3036 ddb
->y_plane
[pipe
][id
].end
= start
+ y_plane_blocks
;
3038 start
+= y_plane_blocks
;
3045 static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state
*config
)
3047 /* TODO: Take into account the scalers once we support them */
3048 return config
->base
.adjusted_mode
.crtc_clock
;
3052 * The max latency should be 257 (max the punit can code is 255 and we add 2us
3053 * for the read latency) and cpp should always be <= 8, so that
3054 * should allow pixel_rate up to ~2 GHz which seems sufficient since max
3055 * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
3057 static uint32_t skl_wm_method1(uint32_t pixel_rate
, uint8_t cpp
, uint32_t latency
)
3059 uint32_t wm_intermediate_val
, ret
;
3064 wm_intermediate_val
= latency
* pixel_rate
* cpp
/ 512;
3065 ret
= DIV_ROUND_UP(wm_intermediate_val
, 1000);
3070 static uint32_t skl_wm_method2(uint32_t pixel_rate
, uint32_t pipe_htotal
,
3071 uint32_t horiz_pixels
, uint8_t cpp
,
3072 uint64_t tiling
, uint32_t latency
)
3075 uint32_t plane_bytes_per_line
, plane_blocks_per_line
;
3076 uint32_t wm_intermediate_val
;
3081 plane_bytes_per_line
= horiz_pixels
* cpp
;
3083 if (tiling
== I915_FORMAT_MOD_Y_TILED
||
3084 tiling
== I915_FORMAT_MOD_Yf_TILED
) {
3085 plane_bytes_per_line
*= 4;
3086 plane_blocks_per_line
= DIV_ROUND_UP(plane_bytes_per_line
, 512);
3087 plane_blocks_per_line
/= 4;
3089 plane_blocks_per_line
= DIV_ROUND_UP(plane_bytes_per_line
, 512);
3092 wm_intermediate_val
= latency
* pixel_rate
;
3093 ret
= DIV_ROUND_UP(wm_intermediate_val
, pipe_htotal
* 1000) *
3094 plane_blocks_per_line
;
3099 static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation
*new_ddb
,
3100 const struct intel_crtc
*intel_crtc
)
3102 struct drm_device
*dev
= intel_crtc
->base
.dev
;
3103 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3104 const struct skl_ddb_allocation
*cur_ddb
= &dev_priv
->wm
.skl_hw
.ddb
;
3107 * If ddb allocation of pipes changed, it may require recalculation of
3110 if (memcmp(new_ddb
->pipe
, cur_ddb
->pipe
, sizeof(new_ddb
->pipe
)))
3116 static bool skl_compute_plane_wm(const struct drm_i915_private
*dev_priv
,
3117 struct intel_crtc_state
*cstate
,
3118 struct intel_plane
*intel_plane
,
3119 uint16_t ddb_allocation
,
3121 uint16_t *out_blocks
, /* out */
3122 uint8_t *out_lines
/* out */)
3124 struct drm_plane
*plane
= &intel_plane
->base
;
3125 struct drm_framebuffer
*fb
= plane
->state
->fb
;
3126 uint32_t latency
= dev_priv
->wm
.skl_latency
[level
];
3127 uint32_t method1
, method2
;
3128 uint32_t plane_bytes_per_line
, plane_blocks_per_line
;
3129 uint32_t res_blocks
, res_lines
;
3130 uint32_t selected_result
;
3133 if (latency
== 0 || !cstate
->base
.active
|| !fb
)
3136 cpp
= drm_format_plane_cpp(fb
->pixel_format
, 0);
3137 method1
= skl_wm_method1(skl_pipe_pixel_rate(cstate
),
3139 method2
= skl_wm_method2(skl_pipe_pixel_rate(cstate
),
3140 cstate
->base
.adjusted_mode
.crtc_htotal
,
3142 cpp
, fb
->modifier
[0],
3145 plane_bytes_per_line
= cstate
->pipe_src_w
* cpp
;
3146 plane_blocks_per_line
= DIV_ROUND_UP(plane_bytes_per_line
, 512);
3148 if (fb
->modifier
[0] == I915_FORMAT_MOD_Y_TILED
||
3149 fb
->modifier
[0] == I915_FORMAT_MOD_Yf_TILED
) {
3150 uint32_t min_scanlines
= 4;
3151 uint32_t y_tile_minimum
;
3152 if (intel_rotation_90_or_270(plane
->state
->rotation
)) {
3153 int cpp
= (fb
->pixel_format
== DRM_FORMAT_NV12
) ?
3154 drm_format_plane_cpp(fb
->pixel_format
, 1) :
3155 drm_format_plane_cpp(fb
->pixel_format
, 0);
3165 WARN(1, "Unsupported pixel depth for rotation");
3168 y_tile_minimum
= plane_blocks_per_line
* min_scanlines
;
3169 selected_result
= max(method2
, y_tile_minimum
);
3171 if ((ddb_allocation
/ plane_blocks_per_line
) >= 1)
3172 selected_result
= min(method1
, method2
);
3174 selected_result
= method1
;
3177 res_blocks
= selected_result
+ 1;
3178 res_lines
= DIV_ROUND_UP(selected_result
, plane_blocks_per_line
);
3180 if (level
>= 1 && level
<= 7) {
3181 if (fb
->modifier
[0] == I915_FORMAT_MOD_Y_TILED
||
3182 fb
->modifier
[0] == I915_FORMAT_MOD_Yf_TILED
)
3188 if (res_blocks
>= ddb_allocation
|| res_lines
> 31)
3191 *out_blocks
= res_blocks
;
3192 *out_lines
= res_lines
;
3197 static void skl_compute_wm_level(const struct drm_i915_private
*dev_priv
,
3198 struct skl_ddb_allocation
*ddb
,
3199 struct intel_crtc_state
*cstate
,
3201 struct skl_wm_level
*result
)
3203 struct drm_device
*dev
= dev_priv
->dev
;
3204 struct intel_crtc
*intel_crtc
= to_intel_crtc(cstate
->base
.crtc
);
3205 struct intel_plane
*intel_plane
;
3206 uint16_t ddb_blocks
;
3207 enum pipe pipe
= intel_crtc
->pipe
;
3209 for_each_intel_plane_on_crtc(dev
, intel_crtc
, intel_plane
) {
3210 int i
= skl_wm_plane_id(intel_plane
);
3212 ddb_blocks
= skl_ddb_entry_size(&ddb
->plane
[pipe
][i
]);
3214 result
->plane_en
[i
] = skl_compute_plane_wm(dev_priv
,
3219 &result
->plane_res_b
[i
],
3220 &result
->plane_res_l
[i
]);
3225 skl_compute_linetime_wm(struct intel_crtc_state
*cstate
)
3227 if (!cstate
->base
.active
)
3230 if (WARN_ON(skl_pipe_pixel_rate(cstate
) == 0))
3233 return DIV_ROUND_UP(8 * cstate
->base
.adjusted_mode
.crtc_htotal
* 1000,
3234 skl_pipe_pixel_rate(cstate
));
3237 static void skl_compute_transition_wm(struct intel_crtc_state
*cstate
,
3238 struct skl_wm_level
*trans_wm
/* out */)
3240 struct drm_crtc
*crtc
= cstate
->base
.crtc
;
3241 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3242 struct intel_plane
*intel_plane
;
3244 if (!cstate
->base
.active
)
3247 /* Until we know more, just disable transition WMs */
3248 for_each_intel_plane_on_crtc(crtc
->dev
, intel_crtc
, intel_plane
) {
3249 int i
= skl_wm_plane_id(intel_plane
);
3251 trans_wm
->plane_en
[i
] = false;
3255 static void skl_compute_pipe_wm(struct intel_crtc_state
*cstate
,
3256 struct skl_ddb_allocation
*ddb
,
3257 struct skl_pipe_wm
*pipe_wm
)
3259 struct drm_device
*dev
= cstate
->base
.crtc
->dev
;
3260 const struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3261 int level
, max_level
= ilk_wm_max_level(dev
);
3263 for (level
= 0; level
<= max_level
; level
++) {
3264 skl_compute_wm_level(dev_priv
, ddb
, cstate
,
3265 level
, &pipe_wm
->wm
[level
]);
3267 pipe_wm
->linetime
= skl_compute_linetime_wm(cstate
);
3269 skl_compute_transition_wm(cstate
, &pipe_wm
->trans_wm
);
3272 static void skl_compute_wm_results(struct drm_device
*dev
,
3273 struct skl_pipe_wm
*p_wm
,
3274 struct skl_wm_values
*r
,
3275 struct intel_crtc
*intel_crtc
)
3277 int level
, max_level
= ilk_wm_max_level(dev
);
3278 enum pipe pipe
= intel_crtc
->pipe
;
3282 for (level
= 0; level
<= max_level
; level
++) {
3283 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3286 temp
|= p_wm
->wm
[level
].plane_res_l
[i
] <<
3287 PLANE_WM_LINES_SHIFT
;
3288 temp
|= p_wm
->wm
[level
].plane_res_b
[i
];
3289 if (p_wm
->wm
[level
].plane_en
[i
])
3290 temp
|= PLANE_WM_EN
;
3292 r
->plane
[pipe
][i
][level
] = temp
;
3297 temp
|= p_wm
->wm
[level
].plane_res_l
[PLANE_CURSOR
] << PLANE_WM_LINES_SHIFT
;
3298 temp
|= p_wm
->wm
[level
].plane_res_b
[PLANE_CURSOR
];
3300 if (p_wm
->wm
[level
].plane_en
[PLANE_CURSOR
])
3301 temp
|= PLANE_WM_EN
;
3303 r
->plane
[pipe
][PLANE_CURSOR
][level
] = temp
;
3307 /* transition WMs */
3308 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3310 temp
|= p_wm
->trans_wm
.plane_res_l
[i
] << PLANE_WM_LINES_SHIFT
;
3311 temp
|= p_wm
->trans_wm
.plane_res_b
[i
];
3312 if (p_wm
->trans_wm
.plane_en
[i
])
3313 temp
|= PLANE_WM_EN
;
3315 r
->plane_trans
[pipe
][i
] = temp
;
3319 temp
|= p_wm
->trans_wm
.plane_res_l
[PLANE_CURSOR
] << PLANE_WM_LINES_SHIFT
;
3320 temp
|= p_wm
->trans_wm
.plane_res_b
[PLANE_CURSOR
];
3321 if (p_wm
->trans_wm
.plane_en
[PLANE_CURSOR
])
3322 temp
|= PLANE_WM_EN
;
3324 r
->plane_trans
[pipe
][PLANE_CURSOR
] = temp
;
3326 r
->wm_linetime
[pipe
] = p_wm
->linetime
;
3329 static void skl_ddb_entry_write(struct drm_i915_private
*dev_priv
,
3331 const struct skl_ddb_entry
*entry
)
3334 I915_WRITE(reg
, (entry
->end
- 1) << 16 | entry
->start
);
3339 static void skl_write_wm_values(struct drm_i915_private
*dev_priv
,
3340 const struct skl_wm_values
*new)
3342 struct drm_device
*dev
= dev_priv
->dev
;
3343 struct intel_crtc
*crtc
;
3345 for_each_intel_crtc(dev
, crtc
) {
3346 int i
, level
, max_level
= ilk_wm_max_level(dev
);
3347 enum pipe pipe
= crtc
->pipe
;
3349 if (!new->dirty
[pipe
])
3352 I915_WRITE(PIPE_WM_LINETIME(pipe
), new->wm_linetime
[pipe
]);
3354 for (level
= 0; level
<= max_level
; level
++) {
3355 for (i
= 0; i
< intel_num_planes(crtc
); i
++)
3356 I915_WRITE(PLANE_WM(pipe
, i
, level
),
3357 new->plane
[pipe
][i
][level
]);
3358 I915_WRITE(CUR_WM(pipe
, level
),
3359 new->plane
[pipe
][PLANE_CURSOR
][level
]);
3361 for (i
= 0; i
< intel_num_planes(crtc
); i
++)
3362 I915_WRITE(PLANE_WM_TRANS(pipe
, i
),
3363 new->plane_trans
[pipe
][i
]);
3364 I915_WRITE(CUR_WM_TRANS(pipe
),
3365 new->plane_trans
[pipe
][PLANE_CURSOR
]);
3367 for (i
= 0; i
< intel_num_planes(crtc
); i
++) {
3368 skl_ddb_entry_write(dev_priv
,
3369 PLANE_BUF_CFG(pipe
, i
),
3370 &new->ddb
.plane
[pipe
][i
]);
3371 skl_ddb_entry_write(dev_priv
,
3372 PLANE_NV12_BUF_CFG(pipe
, i
),
3373 &new->ddb
.y_plane
[pipe
][i
]);
3376 skl_ddb_entry_write(dev_priv
, CUR_BUF_CFG(pipe
),
3377 &new->ddb
.plane
[pipe
][PLANE_CURSOR
]);
3382 * When setting up a new DDB allocation arrangement, we need to correctly
3383 * sequence the times at which the new allocations for the pipes are taken into
3384 * account or we'll have pipes fetching from space previously allocated to
3387 * Roughly the sequence looks like:
3388 * 1. re-allocate the pipe(s) with the allocation being reduced and not
3389 * overlapping with a previous light-up pipe (another way to put it is:
3390 * pipes with their new allocation strickly included into their old ones).
3391 * 2. re-allocate the other pipes that get their allocation reduced
3392 * 3. allocate the pipes having their allocation increased
3394 * Steps 1. and 2. are here to take care of the following case:
3395 * - Initially DDB looks like this:
3398 * - pipe B has a reduced DDB allocation that overlaps with the old pipe C
3402 * We need to sequence the re-allocation: C, B, A (and not B, C, A).
3406 skl_wm_flush_pipe(struct drm_i915_private
*dev_priv
, enum pipe pipe
, int pass
)
3410 DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe
), pass
);
3412 for_each_plane(dev_priv
, pipe
, plane
) {
3413 I915_WRITE(PLANE_SURF(pipe
, plane
),
3414 I915_READ(PLANE_SURF(pipe
, plane
)));
3416 I915_WRITE(CURBASE(pipe
), I915_READ(CURBASE(pipe
)));
3420 skl_ddb_allocation_included(const struct skl_ddb_allocation
*old
,
3421 const struct skl_ddb_allocation
*new,
3424 uint16_t old_size
, new_size
;
3426 old_size
= skl_ddb_entry_size(&old
->pipe
[pipe
]);
3427 new_size
= skl_ddb_entry_size(&new->pipe
[pipe
]);
3429 return old_size
!= new_size
&&
3430 new->pipe
[pipe
].start
>= old
->pipe
[pipe
].start
&&
3431 new->pipe
[pipe
].end
<= old
->pipe
[pipe
].end
;
3434 static void skl_flush_wm_values(struct drm_i915_private
*dev_priv
,
3435 struct skl_wm_values
*new_values
)
3437 struct drm_device
*dev
= dev_priv
->dev
;
3438 struct skl_ddb_allocation
*cur_ddb
, *new_ddb
;
3439 bool reallocated
[I915_MAX_PIPES
] = {};
3440 struct intel_crtc
*crtc
;
3443 new_ddb
= &new_values
->ddb
;
3444 cur_ddb
= &dev_priv
->wm
.skl_hw
.ddb
;
3447 * First pass: flush the pipes with the new allocation contained into
3450 * We'll wait for the vblank on those pipes to ensure we can safely
3451 * re-allocate the freed space without this pipe fetching from it.
3453 for_each_intel_crtc(dev
, crtc
) {
3459 if (!skl_ddb_allocation_included(cur_ddb
, new_ddb
, pipe
))
3462 skl_wm_flush_pipe(dev_priv
, pipe
, 1);
3463 intel_wait_for_vblank(dev
, pipe
);
3465 reallocated
[pipe
] = true;
3470 * Second pass: flush the pipes that are having their allocation
3471 * reduced, but overlapping with a previous allocation.
3473 * Here as well we need to wait for the vblank to make sure the freed
3474 * space is not used anymore.
3476 for_each_intel_crtc(dev
, crtc
) {
3482 if (reallocated
[pipe
])
3485 if (skl_ddb_entry_size(&new_ddb
->pipe
[pipe
]) <
3486 skl_ddb_entry_size(&cur_ddb
->pipe
[pipe
])) {
3487 skl_wm_flush_pipe(dev_priv
, pipe
, 2);
3488 intel_wait_for_vblank(dev
, pipe
);
3489 reallocated
[pipe
] = true;
3494 * Third pass: flush the pipes that got more space allocated.
3496 * We don't need to actively wait for the update here, next vblank
3497 * will just get more DDB space with the correct WM values.
3499 for_each_intel_crtc(dev
, crtc
) {
3506 * At this point, only the pipes more space than before are
3507 * left to re-allocate.
3509 if (reallocated
[pipe
])
3512 skl_wm_flush_pipe(dev_priv
, pipe
, 3);
3516 static bool skl_update_pipe_wm(struct drm_crtc
*crtc
,
3517 struct skl_ddb_allocation
*ddb
, /* out */
3518 struct skl_pipe_wm
*pipe_wm
/* out */)
3520 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3521 struct intel_crtc_state
*cstate
= to_intel_crtc_state(crtc
->state
);
3523 skl_allocate_pipe_ddb(cstate
, ddb
);
3524 skl_compute_pipe_wm(cstate
, ddb
, pipe_wm
);
3526 if (!memcmp(&intel_crtc
->wm
.active
.skl
, pipe_wm
, sizeof(*pipe_wm
)))
3529 intel_crtc
->wm
.active
.skl
= *pipe_wm
;
3534 static void skl_update_other_pipe_wm(struct drm_device
*dev
,
3535 struct drm_crtc
*crtc
,
3536 struct skl_wm_values
*r
)
3538 struct intel_crtc
*intel_crtc
;
3539 struct intel_crtc
*this_crtc
= to_intel_crtc(crtc
);
3542 * If the WM update hasn't changed the allocation for this_crtc (the
3543 * crtc we are currently computing the new WM values for), other
3544 * enabled crtcs will keep the same allocation and we don't need to
3545 * recompute anything for them.
3547 if (!skl_ddb_allocation_changed(&r
->ddb
, this_crtc
))
3551 * Otherwise, because of this_crtc being freshly enabled/disabled, the
3552 * other active pipes need new DDB allocation and WM values.
3554 for_each_intel_crtc(dev
, intel_crtc
) {
3555 struct skl_pipe_wm pipe_wm
= {};
3558 if (this_crtc
->pipe
== intel_crtc
->pipe
)
3561 if (!intel_crtc
->active
)
3564 wm_changed
= skl_update_pipe_wm(&intel_crtc
->base
,
3568 * If we end up re-computing the other pipe WM values, it's
3569 * because it was really needed, so we expect the WM values to
3572 WARN_ON(!wm_changed
);
3574 skl_compute_wm_results(dev
, &pipe_wm
, r
, intel_crtc
);
3575 r
->dirty
[intel_crtc
->pipe
] = true;
3579 static void skl_clear_wm(struct skl_wm_values
*watermarks
, enum pipe pipe
)
3581 watermarks
->wm_linetime
[pipe
] = 0;
3582 memset(watermarks
->plane
[pipe
], 0,
3583 sizeof(uint32_t) * 8 * I915_MAX_PLANES
);
3584 memset(watermarks
->plane_trans
[pipe
],
3585 0, sizeof(uint32_t) * I915_MAX_PLANES
);
3586 watermarks
->plane_trans
[pipe
][PLANE_CURSOR
] = 0;
3588 /* Clear ddb entries for pipe */
3589 memset(&watermarks
->ddb
.pipe
[pipe
], 0, sizeof(struct skl_ddb_entry
));
3590 memset(&watermarks
->ddb
.plane
[pipe
], 0,
3591 sizeof(struct skl_ddb_entry
) * I915_MAX_PLANES
);
3592 memset(&watermarks
->ddb
.y_plane
[pipe
], 0,
3593 sizeof(struct skl_ddb_entry
) * I915_MAX_PLANES
);
3594 memset(&watermarks
->ddb
.plane
[pipe
][PLANE_CURSOR
], 0,
3595 sizeof(struct skl_ddb_entry
));
3599 static void skl_update_wm(struct drm_crtc
*crtc
)
3601 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3602 struct drm_device
*dev
= crtc
->dev
;
3603 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3604 struct skl_wm_values
*results
= &dev_priv
->wm
.skl_results
;
3605 struct intel_crtc_state
*cstate
= to_intel_crtc_state(crtc
->state
);
3606 struct skl_pipe_wm
*pipe_wm
= &cstate
->wm
.optimal
.skl
;
3609 /* Clear all dirty flags */
3610 memset(results
->dirty
, 0, sizeof(bool) * I915_MAX_PIPES
);
3612 skl_clear_wm(results
, intel_crtc
->pipe
);
3614 if (!skl_update_pipe_wm(crtc
, &results
->ddb
, pipe_wm
))
3617 skl_compute_wm_results(dev
, pipe_wm
, results
, intel_crtc
);
3618 results
->dirty
[intel_crtc
->pipe
] = true;
3620 skl_update_other_pipe_wm(dev
, crtc
, results
);
3621 skl_write_wm_values(dev_priv
, results
);
3622 skl_flush_wm_values(dev_priv
, results
);
3624 /* store the new configuration */
3625 dev_priv
->wm
.skl_hw
= *results
;
3628 static void ilk_compute_wm_config(struct drm_device
*dev
,
3629 struct intel_wm_config
*config
)
3631 struct intel_crtc
*crtc
;
3633 /* Compute the currently _active_ config */
3634 for_each_intel_crtc(dev
, crtc
) {
3635 const struct intel_pipe_wm
*wm
= &crtc
->wm
.active
.ilk
;
3637 if (!wm
->pipe_enabled
)
3640 config
->sprites_enabled
|= wm
->sprites_enabled
;
3641 config
->sprites_scaled
|= wm
->sprites_scaled
;
3642 config
->num_pipes_active
++;
3646 static void ilk_program_watermarks(struct intel_crtc_state
*cstate
)
3648 struct drm_crtc
*crtc
= cstate
->base
.crtc
;
3649 struct drm_device
*dev
= crtc
->dev
;
3650 struct drm_i915_private
*dev_priv
= to_i915(dev
);
3651 struct intel_pipe_wm lp_wm_1_2
= {}, lp_wm_5_6
= {}, *best_lp_wm
;
3652 struct ilk_wm_maximums max
;
3653 struct intel_wm_config config
= {};
3654 struct ilk_wm_values results
= {};
3655 enum intel_ddb_partitioning partitioning
;
3657 ilk_compute_wm_config(dev
, &config
);
3659 ilk_compute_wm_maximums(dev
, 1, &config
, INTEL_DDB_PART_1_2
, &max
);
3660 ilk_wm_merge(dev
, &config
, &max
, &lp_wm_1_2
);
3662 /* 5/6 split only in single pipe config on IVB+ */
3663 if (INTEL_INFO(dev
)->gen
>= 7 &&
3664 config
.num_pipes_active
== 1 && config
.sprites_enabled
) {
3665 ilk_compute_wm_maximums(dev
, 1, &config
, INTEL_DDB_PART_5_6
, &max
);
3666 ilk_wm_merge(dev
, &config
, &max
, &lp_wm_5_6
);
3668 best_lp_wm
= ilk_find_best_result(dev
, &lp_wm_1_2
, &lp_wm_5_6
);
3670 best_lp_wm
= &lp_wm_1_2
;
3673 partitioning
= (best_lp_wm
== &lp_wm_1_2
) ?
3674 INTEL_DDB_PART_1_2
: INTEL_DDB_PART_5_6
;
3676 ilk_compute_wm_results(dev
, best_lp_wm
, partitioning
, &results
);
3678 ilk_write_wm_values(dev_priv
, &results
);
3681 static void ilk_update_wm(struct drm_crtc
*crtc
)
3683 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3684 struct intel_crtc_state
*cstate
= to_intel_crtc_state(crtc
->state
);
3686 WARN_ON(cstate
->base
.active
!= intel_crtc
->active
);
3689 * IVB workaround: must disable low power watermarks for at least
3690 * one frame before enabling scaling. LP watermarks can be re-enabled
3691 * when scaling is disabled.
3693 * WaCxSRDisabledForSpriteScaling:ivb
3695 if (cstate
->disable_lp_wm
) {
3696 ilk_disable_lp_wm(crtc
->dev
);
3697 intel_wait_for_vblank(crtc
->dev
, intel_crtc
->pipe
);
3700 intel_crtc
->wm
.active
.ilk
= cstate
->wm
.optimal
.ilk
;
3702 ilk_program_watermarks(cstate
);
3705 static void skl_pipe_wm_active_state(uint32_t val
,
3706 struct skl_pipe_wm
*active
,
3712 bool is_enabled
= (val
& PLANE_WM_EN
) != 0;
3716 active
->wm
[level
].plane_en
[i
] = is_enabled
;
3717 active
->wm
[level
].plane_res_b
[i
] =
3718 val
& PLANE_WM_BLOCKS_MASK
;
3719 active
->wm
[level
].plane_res_l
[i
] =
3720 (val
>> PLANE_WM_LINES_SHIFT
) &
3721 PLANE_WM_LINES_MASK
;
3723 active
->wm
[level
].plane_en
[PLANE_CURSOR
] = is_enabled
;
3724 active
->wm
[level
].plane_res_b
[PLANE_CURSOR
] =
3725 val
& PLANE_WM_BLOCKS_MASK
;
3726 active
->wm
[level
].plane_res_l
[PLANE_CURSOR
] =
3727 (val
>> PLANE_WM_LINES_SHIFT
) &
3728 PLANE_WM_LINES_MASK
;
3732 active
->trans_wm
.plane_en
[i
] = is_enabled
;
3733 active
->trans_wm
.plane_res_b
[i
] =
3734 val
& PLANE_WM_BLOCKS_MASK
;
3735 active
->trans_wm
.plane_res_l
[i
] =
3736 (val
>> PLANE_WM_LINES_SHIFT
) &
3737 PLANE_WM_LINES_MASK
;
3739 active
->trans_wm
.plane_en
[PLANE_CURSOR
] = is_enabled
;
3740 active
->trans_wm
.plane_res_b
[PLANE_CURSOR
] =
3741 val
& PLANE_WM_BLOCKS_MASK
;
3742 active
->trans_wm
.plane_res_l
[PLANE_CURSOR
] =
3743 (val
>> PLANE_WM_LINES_SHIFT
) &
3744 PLANE_WM_LINES_MASK
;
3749 static void skl_pipe_wm_get_hw_state(struct drm_crtc
*crtc
)
3751 struct drm_device
*dev
= crtc
->dev
;
3752 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3753 struct skl_wm_values
*hw
= &dev_priv
->wm
.skl_hw
;
3754 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3755 struct intel_crtc_state
*cstate
= to_intel_crtc_state(crtc
->state
);
3756 struct skl_pipe_wm
*active
= &cstate
->wm
.optimal
.skl
;
3757 enum pipe pipe
= intel_crtc
->pipe
;
3758 int level
, i
, max_level
;
3761 max_level
= ilk_wm_max_level(dev
);
3763 hw
->wm_linetime
[pipe
] = I915_READ(PIPE_WM_LINETIME(pipe
));
3765 for (level
= 0; level
<= max_level
; level
++) {
3766 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++)
3767 hw
->plane
[pipe
][i
][level
] =
3768 I915_READ(PLANE_WM(pipe
, i
, level
));
3769 hw
->plane
[pipe
][PLANE_CURSOR
][level
] = I915_READ(CUR_WM(pipe
, level
));
3772 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++)
3773 hw
->plane_trans
[pipe
][i
] = I915_READ(PLANE_WM_TRANS(pipe
, i
));
3774 hw
->plane_trans
[pipe
][PLANE_CURSOR
] = I915_READ(CUR_WM_TRANS(pipe
));
3776 if (!intel_crtc
->active
)
3779 hw
->dirty
[pipe
] = true;
3781 active
->linetime
= hw
->wm_linetime
[pipe
];
3783 for (level
= 0; level
<= max_level
; level
++) {
3784 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3785 temp
= hw
->plane
[pipe
][i
][level
];
3786 skl_pipe_wm_active_state(temp
, active
, false,
3789 temp
= hw
->plane
[pipe
][PLANE_CURSOR
][level
];
3790 skl_pipe_wm_active_state(temp
, active
, false, true, i
, level
);
3793 for (i
= 0; i
< intel_num_planes(intel_crtc
); i
++) {
3794 temp
= hw
->plane_trans
[pipe
][i
];
3795 skl_pipe_wm_active_state(temp
, active
, true, false, i
, 0);
3798 temp
= hw
->plane_trans
[pipe
][PLANE_CURSOR
];
3799 skl_pipe_wm_active_state(temp
, active
, true, true, i
, 0);
3801 intel_crtc
->wm
.active
.skl
= *active
;
3804 void skl_wm_get_hw_state(struct drm_device
*dev
)
3806 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3807 struct skl_ddb_allocation
*ddb
= &dev_priv
->wm
.skl_hw
.ddb
;
3808 struct drm_crtc
*crtc
;
3810 skl_ddb_get_hw_state(dev_priv
, ddb
);
3811 list_for_each_entry(crtc
, &dev
->mode_config
.crtc_list
, head
)
3812 skl_pipe_wm_get_hw_state(crtc
);
3815 static void ilk_pipe_wm_get_hw_state(struct drm_crtc
*crtc
)
3817 struct drm_device
*dev
= crtc
->dev
;
3818 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3819 struct ilk_wm_values
*hw
= &dev_priv
->wm
.hw
;
3820 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
3821 struct intel_crtc_state
*cstate
= to_intel_crtc_state(crtc
->state
);
3822 struct intel_pipe_wm
*active
= &cstate
->wm
.optimal
.ilk
;
3823 enum pipe pipe
= intel_crtc
->pipe
;
3824 static const i915_reg_t wm0_pipe_reg
[] = {
3825 [PIPE_A
] = WM0_PIPEA_ILK
,
3826 [PIPE_B
] = WM0_PIPEB_ILK
,
3827 [PIPE_C
] = WM0_PIPEC_IVB
,
3830 hw
->wm_pipe
[pipe
] = I915_READ(wm0_pipe_reg
[pipe
]);
3831 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
3832 hw
->wm_linetime
[pipe
] = I915_READ(PIPE_WM_LINETIME(pipe
));
3834 active
->pipe_enabled
= intel_crtc
->active
;
3836 if (active
->pipe_enabled
) {
3837 u32 tmp
= hw
->wm_pipe
[pipe
];
3840 * For active pipes LP0 watermark is marked as
3841 * enabled, and LP1+ watermaks as disabled since
3842 * we can't really reverse compute them in case
3843 * multiple pipes are active.
3845 active
->wm
[0].enable
= true;
3846 active
->wm
[0].pri_val
= (tmp
& WM0_PIPE_PLANE_MASK
) >> WM0_PIPE_PLANE_SHIFT
;
3847 active
->wm
[0].spr_val
= (tmp
& WM0_PIPE_SPRITE_MASK
) >> WM0_PIPE_SPRITE_SHIFT
;
3848 active
->wm
[0].cur_val
= tmp
& WM0_PIPE_CURSOR_MASK
;
3849 active
->linetime
= hw
->wm_linetime
[pipe
];
3851 int level
, max_level
= ilk_wm_max_level(dev
);
3854 * For inactive pipes, all watermark levels
3855 * should be marked as enabled but zeroed,
3856 * which is what we'd compute them to.
3858 for (level
= 0; level
<= max_level
; level
++)
3859 active
->wm
[level
].enable
= true;
3862 intel_crtc
->wm
.active
.ilk
= *active
;
3865 #define _FW_WM(value, plane) \
3866 (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT)
3867 #define _FW_WM_VLV(value, plane) \
3868 (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT)
3870 static void vlv_read_wm_values(struct drm_i915_private
*dev_priv
,
3871 struct vlv_wm_values
*wm
)
3876 for_each_pipe(dev_priv
, pipe
) {
3877 tmp
= I915_READ(VLV_DDL(pipe
));
3879 wm
->ddl
[pipe
].primary
=
3880 (tmp
>> DDL_PLANE_SHIFT
) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
3881 wm
->ddl
[pipe
].cursor
=
3882 (tmp
>> DDL_CURSOR_SHIFT
) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
3883 wm
->ddl
[pipe
].sprite
[0] =
3884 (tmp
>> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
3885 wm
->ddl
[pipe
].sprite
[1] =
3886 (tmp
>> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH
| DRAIN_LATENCY_MASK
);
3889 tmp
= I915_READ(DSPFW1
);
3890 wm
->sr
.plane
= _FW_WM(tmp
, SR
);
3891 wm
->pipe
[PIPE_B
].cursor
= _FW_WM(tmp
, CURSORB
);
3892 wm
->pipe
[PIPE_B
].primary
= _FW_WM_VLV(tmp
, PLANEB
);
3893 wm
->pipe
[PIPE_A
].primary
= _FW_WM_VLV(tmp
, PLANEA
);
3895 tmp
= I915_READ(DSPFW2
);
3896 wm
->pipe
[PIPE_A
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITEB
);
3897 wm
->pipe
[PIPE_A
].cursor
= _FW_WM(tmp
, CURSORA
);
3898 wm
->pipe
[PIPE_A
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEA
);
3900 tmp
= I915_READ(DSPFW3
);
3901 wm
->sr
.cursor
= _FW_WM(tmp
, CURSOR_SR
);
3903 if (IS_CHERRYVIEW(dev_priv
)) {
3904 tmp
= I915_READ(DSPFW7_CHV
);
3905 wm
->pipe
[PIPE_B
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITED
);
3906 wm
->pipe
[PIPE_B
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEC
);
3908 tmp
= I915_READ(DSPFW8_CHV
);
3909 wm
->pipe
[PIPE_C
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITEF
);
3910 wm
->pipe
[PIPE_C
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEE
);
3912 tmp
= I915_READ(DSPFW9_CHV
);
3913 wm
->pipe
[PIPE_C
].primary
= _FW_WM_VLV(tmp
, PLANEC
);
3914 wm
->pipe
[PIPE_C
].cursor
= _FW_WM(tmp
, CURSORC
);
3916 tmp
= I915_READ(DSPHOWM
);
3917 wm
->sr
.plane
|= _FW_WM(tmp
, SR_HI
) << 9;
3918 wm
->pipe
[PIPE_C
].sprite
[1] |= _FW_WM(tmp
, SPRITEF_HI
) << 8;
3919 wm
->pipe
[PIPE_C
].sprite
[0] |= _FW_WM(tmp
, SPRITEE_HI
) << 8;
3920 wm
->pipe
[PIPE_C
].primary
|= _FW_WM(tmp
, PLANEC_HI
) << 8;
3921 wm
->pipe
[PIPE_B
].sprite
[1] |= _FW_WM(tmp
, SPRITED_HI
) << 8;
3922 wm
->pipe
[PIPE_B
].sprite
[0] |= _FW_WM(tmp
, SPRITEC_HI
) << 8;
3923 wm
->pipe
[PIPE_B
].primary
|= _FW_WM(tmp
, PLANEB_HI
) << 8;
3924 wm
->pipe
[PIPE_A
].sprite
[1] |= _FW_WM(tmp
, SPRITEB_HI
) << 8;
3925 wm
->pipe
[PIPE_A
].sprite
[0] |= _FW_WM(tmp
, SPRITEA_HI
) << 8;
3926 wm
->pipe
[PIPE_A
].primary
|= _FW_WM(tmp
, PLANEA_HI
) << 8;
3928 tmp
= I915_READ(DSPFW7
);
3929 wm
->pipe
[PIPE_B
].sprite
[1] = _FW_WM_VLV(tmp
, SPRITED
);
3930 wm
->pipe
[PIPE_B
].sprite
[0] = _FW_WM_VLV(tmp
, SPRITEC
);
3932 tmp
= I915_READ(DSPHOWM
);
3933 wm
->sr
.plane
|= _FW_WM(tmp
, SR_HI
) << 9;
3934 wm
->pipe
[PIPE_B
].sprite
[1] |= _FW_WM(tmp
, SPRITED_HI
) << 8;
3935 wm
->pipe
[PIPE_B
].sprite
[0] |= _FW_WM(tmp
, SPRITEC_HI
) << 8;
3936 wm
->pipe
[PIPE_B
].primary
|= _FW_WM(tmp
, PLANEB_HI
) << 8;
3937 wm
->pipe
[PIPE_A
].sprite
[1] |= _FW_WM(tmp
, SPRITEB_HI
) << 8;
3938 wm
->pipe
[PIPE_A
].sprite
[0] |= _FW_WM(tmp
, SPRITEA_HI
) << 8;
3939 wm
->pipe
[PIPE_A
].primary
|= _FW_WM(tmp
, PLANEA_HI
) << 8;
3946 void vlv_wm_get_hw_state(struct drm_device
*dev
)
3948 struct drm_i915_private
*dev_priv
= to_i915(dev
);
3949 struct vlv_wm_values
*wm
= &dev_priv
->wm
.vlv
;
3950 struct intel_plane
*plane
;
3954 vlv_read_wm_values(dev_priv
, wm
);
3956 for_each_intel_plane(dev
, plane
) {
3957 switch (plane
->base
.type
) {
3959 case DRM_PLANE_TYPE_CURSOR
:
3960 plane
->wm
.fifo_size
= 63;
3962 case DRM_PLANE_TYPE_PRIMARY
:
3963 plane
->wm
.fifo_size
= vlv_get_fifo_size(dev
, plane
->pipe
, 0);
3965 case DRM_PLANE_TYPE_OVERLAY
:
3966 sprite
= plane
->plane
;
3967 plane
->wm
.fifo_size
= vlv_get_fifo_size(dev
, plane
->pipe
, sprite
+ 1);
3972 wm
->cxsr
= I915_READ(FW_BLC_SELF_VLV
) & FW_CSPWRDWNEN
;
3973 wm
->level
= VLV_WM_LEVEL_PM2
;
3975 if (IS_CHERRYVIEW(dev_priv
)) {
3976 mutex_lock(&dev_priv
->rps
.hw_lock
);
3978 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DSPFREQ
);
3979 if (val
& DSP_MAXFIFO_PM5_ENABLE
)
3980 wm
->level
= VLV_WM_LEVEL_PM5
;
3983 * If DDR DVFS is disabled in the BIOS, Punit
3984 * will never ack the request. So if that happens
3985 * assume we don't have to enable/disable DDR DVFS
3986 * dynamically. To test that just set the REQ_ACK
3987 * bit to poke the Punit, but don't change the
3988 * HIGH/LOW bits so that we don't actually change
3989 * the current state.
3991 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
);
3992 val
|= FORCE_DDR_FREQ_REQ_ACK
;
3993 vlv_punit_write(dev_priv
, PUNIT_REG_DDR_SETUP2
, val
);
3995 if (wait_for((vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
) &
3996 FORCE_DDR_FREQ_REQ_ACK
) == 0, 3)) {
3997 DRM_DEBUG_KMS("Punit not acking DDR DVFS request, "
3998 "assuming DDR DVFS is disabled\n");
3999 dev_priv
->wm
.max_level
= VLV_WM_LEVEL_PM5
;
4001 val
= vlv_punit_read(dev_priv
, PUNIT_REG_DDR_SETUP2
);
4002 if ((val
& FORCE_DDR_HIGH_FREQ
) == 0)
4003 wm
->level
= VLV_WM_LEVEL_DDR_DVFS
;
4006 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4009 for_each_pipe(dev_priv
, pipe
)
4010 DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n",
4011 pipe_name(pipe
), wm
->pipe
[pipe
].primary
, wm
->pipe
[pipe
].cursor
,
4012 wm
->pipe
[pipe
].sprite
[0], wm
->pipe
[pipe
].sprite
[1]);
4014 DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n",
4015 wm
->sr
.plane
, wm
->sr
.cursor
, wm
->level
, wm
->cxsr
);
4018 void ilk_wm_get_hw_state(struct drm_device
*dev
)
4020 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4021 struct ilk_wm_values
*hw
= &dev_priv
->wm
.hw
;
4022 struct drm_crtc
*crtc
;
4024 for_each_crtc(dev
, crtc
)
4025 ilk_pipe_wm_get_hw_state(crtc
);
4027 hw
->wm_lp
[0] = I915_READ(WM1_LP_ILK
);
4028 hw
->wm_lp
[1] = I915_READ(WM2_LP_ILK
);
4029 hw
->wm_lp
[2] = I915_READ(WM3_LP_ILK
);
4031 hw
->wm_lp_spr
[0] = I915_READ(WM1S_LP_ILK
);
4032 if (INTEL_INFO(dev
)->gen
>= 7) {
4033 hw
->wm_lp_spr
[1] = I915_READ(WM2S_LP_IVB
);
4034 hw
->wm_lp_spr
[2] = I915_READ(WM3S_LP_IVB
);
4037 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
4038 hw
->partitioning
= (I915_READ(WM_MISC
) & WM_MISC_DATA_PARTITION_5_6
) ?
4039 INTEL_DDB_PART_5_6
: INTEL_DDB_PART_1_2
;
4040 else if (IS_IVYBRIDGE(dev
))
4041 hw
->partitioning
= (I915_READ(DISP_ARB_CTL2
) & DISP_DATA_PARTITION_5_6
) ?
4042 INTEL_DDB_PART_5_6
: INTEL_DDB_PART_1_2
;
4045 !(I915_READ(DISP_ARB_CTL
) & DISP_FBC_WM_DIS
);
4049 * intel_update_watermarks - update FIFO watermark values based on current modes
4051 * Calculate watermark values for the various WM regs based on current mode
4052 * and plane configuration.
4054 * There are several cases to deal with here:
4055 * - normal (i.e. non-self-refresh)
4056 * - self-refresh (SR) mode
4057 * - lines are large relative to FIFO size (buffer can hold up to 2)
4058 * - lines are small relative to FIFO size (buffer can hold more than 2
4059 * lines), so need to account for TLB latency
4061 * The normal calculation is:
4062 * watermark = dotclock * bytes per pixel * latency
4063 * where latency is platform & configuration dependent (we assume pessimal
4066 * The SR calculation is:
4067 * watermark = (trunc(latency/line time)+1) * surface width *
4070 * line time = htotal / dotclock
4071 * surface width = hdisplay for normal plane and 64 for cursor
4072 * and latency is assumed to be high, as above.
4074 * The final value programmed to the register should always be rounded up,
4075 * and include an extra 2 entries to account for clock crossings.
4077 * We don't use the sprite, so we can ignore that. And on Crestline we have
4078 * to set the non-SR watermarks to 8.
4080 void intel_update_watermarks(struct drm_crtc
*crtc
)
4082 struct drm_i915_private
*dev_priv
= crtc
->dev
->dev_private
;
4084 if (dev_priv
->display
.update_wm
)
4085 dev_priv
->display
.update_wm(crtc
);
4089 * Lock protecting IPS related data structures
4091 DEFINE_SPINLOCK(mchdev_lock
);
4093 /* Global for IPS driver to get at the current i915 device. Protected by
4095 static struct drm_i915_private
*i915_mch_dev
;
4097 bool ironlake_set_drps(struct drm_device
*dev
, u8 val
)
4099 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4102 assert_spin_locked(&mchdev_lock
);
4104 rgvswctl
= I915_READ16(MEMSWCTL
);
4105 if (rgvswctl
& MEMCTL_CMD_STS
) {
4106 DRM_DEBUG("gpu busy, RCS change rejected\n");
4107 return false; /* still busy with another command */
4110 rgvswctl
= (MEMCTL_CMD_CHFREQ
<< MEMCTL_CMD_SHIFT
) |
4111 (val
<< MEMCTL_FREQ_SHIFT
) | MEMCTL_SFCAVM
;
4112 I915_WRITE16(MEMSWCTL
, rgvswctl
);
4113 POSTING_READ16(MEMSWCTL
);
4115 rgvswctl
|= MEMCTL_CMD_STS
;
4116 I915_WRITE16(MEMSWCTL
, rgvswctl
);
4121 static void ironlake_enable_drps(struct drm_device
*dev
)
4123 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4125 u8 fmax
, fmin
, fstart
, vstart
;
4127 spin_lock_irq(&mchdev_lock
);
4129 rgvmodectl
= I915_READ(MEMMODECTL
);
4131 /* Enable temp reporting */
4132 I915_WRITE16(PMMISC
, I915_READ(PMMISC
) | MCPPCE_EN
);
4133 I915_WRITE16(TSC1
, I915_READ(TSC1
) | TSE
);
4135 /* 100ms RC evaluation intervals */
4136 I915_WRITE(RCUPEI
, 100000);
4137 I915_WRITE(RCDNEI
, 100000);
4139 /* Set max/min thresholds to 90ms and 80ms respectively */
4140 I915_WRITE(RCBMAXAVG
, 90000);
4141 I915_WRITE(RCBMINAVG
, 80000);
4143 I915_WRITE(MEMIHYST
, 1);
4145 /* Set up min, max, and cur for interrupt handling */
4146 fmax
= (rgvmodectl
& MEMMODE_FMAX_MASK
) >> MEMMODE_FMAX_SHIFT
;
4147 fmin
= (rgvmodectl
& MEMMODE_FMIN_MASK
);
4148 fstart
= (rgvmodectl
& MEMMODE_FSTART_MASK
) >>
4149 MEMMODE_FSTART_SHIFT
;
4151 vstart
= (I915_READ(PXVFREQ(fstart
)) & PXVFREQ_PX_MASK
) >>
4154 dev_priv
->ips
.fmax
= fmax
; /* IPS callback will increase this */
4155 dev_priv
->ips
.fstart
= fstart
;
4157 dev_priv
->ips
.max_delay
= fstart
;
4158 dev_priv
->ips
.min_delay
= fmin
;
4159 dev_priv
->ips
.cur_delay
= fstart
;
4161 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
4162 fmax
, fmin
, fstart
);
4164 I915_WRITE(MEMINTREN
, MEMINT_CX_SUPR_EN
| MEMINT_EVAL_CHG_EN
);
4167 * Interrupts will be enabled in ironlake_irq_postinstall
4170 I915_WRITE(VIDSTART
, vstart
);
4171 POSTING_READ(VIDSTART
);
4173 rgvmodectl
|= MEMMODE_SWMODE_EN
;
4174 I915_WRITE(MEMMODECTL
, rgvmodectl
);
4176 if (wait_for_atomic((I915_READ(MEMSWCTL
) & MEMCTL_CMD_STS
) == 0, 10))
4177 DRM_ERROR("stuck trying to change perf mode\n");
4180 ironlake_set_drps(dev
, fstart
);
4182 dev_priv
->ips
.last_count1
= I915_READ(DMIEC
) +
4183 I915_READ(DDREC
) + I915_READ(CSIEC
);
4184 dev_priv
->ips
.last_time1
= jiffies_to_msecs(jiffies
);
4185 dev_priv
->ips
.last_count2
= I915_READ(GFXEC
);
4186 dev_priv
->ips
.last_time2
= ktime_get_raw_ns();
4188 spin_unlock_irq(&mchdev_lock
);
4191 static void ironlake_disable_drps(struct drm_device
*dev
)
4193 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4196 spin_lock_irq(&mchdev_lock
);
4198 rgvswctl
= I915_READ16(MEMSWCTL
);
4200 /* Ack interrupts, disable EFC interrupt */
4201 I915_WRITE(MEMINTREN
, I915_READ(MEMINTREN
) & ~MEMINT_EVAL_CHG_EN
);
4202 I915_WRITE(MEMINTRSTS
, MEMINT_EVAL_CHG
);
4203 I915_WRITE(DEIER
, I915_READ(DEIER
) & ~DE_PCU_EVENT
);
4204 I915_WRITE(DEIIR
, DE_PCU_EVENT
);
4205 I915_WRITE(DEIMR
, I915_READ(DEIMR
) | DE_PCU_EVENT
);
4207 /* Go back to the starting frequency */
4208 ironlake_set_drps(dev
, dev_priv
->ips
.fstart
);
4210 rgvswctl
|= MEMCTL_CMD_STS
;
4211 I915_WRITE(MEMSWCTL
, rgvswctl
);
4214 spin_unlock_irq(&mchdev_lock
);
4217 /* There's a funny hw issue where the hw returns all 0 when reading from
4218 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
4219 * ourselves, instead of doing a rmw cycle (which might result in us clearing
4220 * all limits and the gpu stuck at whatever frequency it is at atm).
4222 static u32
intel_rps_limits(struct drm_i915_private
*dev_priv
, u8 val
)
4226 /* Only set the down limit when we've reached the lowest level to avoid
4227 * getting more interrupts, otherwise leave this clear. This prevents a
4228 * race in the hw when coming out of rc6: There's a tiny window where
4229 * the hw runs at the minimal clock before selecting the desired
4230 * frequency, if the down threshold expires in that window we will not
4231 * receive a down interrupt. */
4232 if (IS_GEN9(dev_priv
->dev
)) {
4233 limits
= (dev_priv
->rps
.max_freq_softlimit
) << 23;
4234 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
4235 limits
|= (dev_priv
->rps
.min_freq_softlimit
) << 14;
4237 limits
= dev_priv
->rps
.max_freq_softlimit
<< 24;
4238 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
4239 limits
|= dev_priv
->rps
.min_freq_softlimit
<< 16;
4245 static void gen6_set_rps_thresholds(struct drm_i915_private
*dev_priv
, u8 val
)
4248 u32 threshold_up
= 0, threshold_down
= 0; /* in % */
4249 u32 ei_up
= 0, ei_down
= 0;
4251 new_power
= dev_priv
->rps
.power
;
4252 switch (dev_priv
->rps
.power
) {
4254 if (val
> dev_priv
->rps
.efficient_freq
+ 1 && val
> dev_priv
->rps
.cur_freq
)
4255 new_power
= BETWEEN
;
4259 if (val
<= dev_priv
->rps
.efficient_freq
&& val
< dev_priv
->rps
.cur_freq
)
4260 new_power
= LOW_POWER
;
4261 else if (val
>= dev_priv
->rps
.rp0_freq
&& val
> dev_priv
->rps
.cur_freq
)
4262 new_power
= HIGH_POWER
;
4266 if (val
< (dev_priv
->rps
.rp1_freq
+ dev_priv
->rps
.rp0_freq
) >> 1 && val
< dev_priv
->rps
.cur_freq
)
4267 new_power
= BETWEEN
;
4270 /* Max/min bins are special */
4271 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
4272 new_power
= LOW_POWER
;
4273 if (val
>= dev_priv
->rps
.max_freq_softlimit
)
4274 new_power
= HIGH_POWER
;
4275 if (new_power
== dev_priv
->rps
.power
)
4278 /* Note the units here are not exactly 1us, but 1280ns. */
4279 switch (new_power
) {
4281 /* Upclock if more than 95% busy over 16ms */
4285 /* Downclock if less than 85% busy over 32ms */
4287 threshold_down
= 85;
4291 /* Upclock if more than 90% busy over 13ms */
4295 /* Downclock if less than 75% busy over 32ms */
4297 threshold_down
= 75;
4301 /* Upclock if more than 85% busy over 10ms */
4305 /* Downclock if less than 60% busy over 32ms */
4307 threshold_down
= 60;
4311 I915_WRITE(GEN6_RP_UP_EI
,
4312 GT_INTERVAL_FROM_US(dev_priv
, ei_up
));
4313 I915_WRITE(GEN6_RP_UP_THRESHOLD
,
4314 GT_INTERVAL_FROM_US(dev_priv
, (ei_up
* threshold_up
/ 100)));
4316 I915_WRITE(GEN6_RP_DOWN_EI
,
4317 GT_INTERVAL_FROM_US(dev_priv
, ei_down
));
4318 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
,
4319 GT_INTERVAL_FROM_US(dev_priv
, (ei_down
* threshold_down
/ 100)));
4321 I915_WRITE(GEN6_RP_CONTROL
,
4322 GEN6_RP_MEDIA_TURBO
|
4323 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
4324 GEN6_RP_MEDIA_IS_GFX
|
4326 GEN6_RP_UP_BUSY_AVG
|
4327 GEN6_RP_DOWN_IDLE_AVG
);
4329 dev_priv
->rps
.power
= new_power
;
4330 dev_priv
->rps
.up_threshold
= threshold_up
;
4331 dev_priv
->rps
.down_threshold
= threshold_down
;
4332 dev_priv
->rps
.last_adj
= 0;
4335 static u32
gen6_rps_pm_mask(struct drm_i915_private
*dev_priv
, u8 val
)
4339 if (val
> dev_priv
->rps
.min_freq_softlimit
)
4340 mask
|= GEN6_PM_RP_DOWN_EI_EXPIRED
| GEN6_PM_RP_DOWN_THRESHOLD
| GEN6_PM_RP_DOWN_TIMEOUT
;
4341 if (val
< dev_priv
->rps
.max_freq_softlimit
)
4342 mask
|= GEN6_PM_RP_UP_EI_EXPIRED
| GEN6_PM_RP_UP_THRESHOLD
;
4344 mask
&= dev_priv
->pm_rps_events
;
4346 return gen6_sanitize_rps_pm_mask(dev_priv
, ~mask
);
4349 /* gen6_set_rps is called to update the frequency request, but should also be
4350 * called when the range (min_delay and max_delay) is modified so that we can
4351 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
4352 static void gen6_set_rps(struct drm_device
*dev
, u8 val
)
4354 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4356 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
4357 if (IS_BXT_REVID(dev
, 0, BXT_REVID_A1
))
4360 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
4361 WARN_ON(val
> dev_priv
->rps
.max_freq
);
4362 WARN_ON(val
< dev_priv
->rps
.min_freq
);
4364 /* min/max delay may still have been modified so be sure to
4365 * write the limits value.
4367 if (val
!= dev_priv
->rps
.cur_freq
) {
4368 gen6_set_rps_thresholds(dev_priv
, val
);
4371 I915_WRITE(GEN6_RPNSWREQ
,
4372 GEN9_FREQUENCY(val
));
4373 else if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
4374 I915_WRITE(GEN6_RPNSWREQ
,
4375 HSW_FREQUENCY(val
));
4377 I915_WRITE(GEN6_RPNSWREQ
,
4378 GEN6_FREQUENCY(val
) |
4380 GEN6_AGGRESSIVE_TURBO
);
4383 /* Make sure we continue to get interrupts
4384 * until we hit the minimum or maximum frequencies.
4386 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
, intel_rps_limits(dev_priv
, val
));
4387 I915_WRITE(GEN6_PMINTRMSK
, gen6_rps_pm_mask(dev_priv
, val
));
4389 POSTING_READ(GEN6_RPNSWREQ
);
4391 dev_priv
->rps
.cur_freq
= val
;
4392 trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv
, val
));
4395 static void valleyview_set_rps(struct drm_device
*dev
, u8 val
)
4397 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4399 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
4400 WARN_ON(val
> dev_priv
->rps
.max_freq
);
4401 WARN_ON(val
< dev_priv
->rps
.min_freq
);
4403 if (WARN_ONCE(IS_CHERRYVIEW(dev
) && (val
& 1),
4404 "Odd GPU freq value\n"))
4407 I915_WRITE(GEN6_PMINTRMSK
, gen6_rps_pm_mask(dev_priv
, val
));
4409 if (val
!= dev_priv
->rps
.cur_freq
) {
4410 vlv_punit_write(dev_priv
, PUNIT_REG_GPU_FREQ_REQ
, val
);
4411 if (!IS_CHERRYVIEW(dev_priv
))
4412 gen6_set_rps_thresholds(dev_priv
, val
);
4415 dev_priv
->rps
.cur_freq
= val
;
4416 trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv
, val
));
4419 /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down
4421 * * If Gfx is Idle, then
4422 * 1. Forcewake Media well.
4423 * 2. Request idle freq.
4424 * 3. Release Forcewake of Media well.
4426 static void vlv_set_rps_idle(struct drm_i915_private
*dev_priv
)
4428 u32 val
= dev_priv
->rps
.idle_freq
;
4430 if (dev_priv
->rps
.cur_freq
<= val
)
4433 /* Wake up the media well, as that takes a lot less
4434 * power than the Render well. */
4435 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_MEDIA
);
4436 valleyview_set_rps(dev_priv
->dev
, val
);
4437 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_MEDIA
);
4440 void gen6_rps_busy(struct drm_i915_private
*dev_priv
)
4442 mutex_lock(&dev_priv
->rps
.hw_lock
);
4443 if (dev_priv
->rps
.enabled
) {
4444 if (dev_priv
->pm_rps_events
& (GEN6_PM_RP_DOWN_EI_EXPIRED
| GEN6_PM_RP_UP_EI_EXPIRED
))
4445 gen6_rps_reset_ei(dev_priv
);
4446 I915_WRITE(GEN6_PMINTRMSK
,
4447 gen6_rps_pm_mask(dev_priv
, dev_priv
->rps
.cur_freq
));
4449 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4452 void gen6_rps_idle(struct drm_i915_private
*dev_priv
)
4454 struct drm_device
*dev
= dev_priv
->dev
;
4456 mutex_lock(&dev_priv
->rps
.hw_lock
);
4457 if (dev_priv
->rps
.enabled
) {
4458 if (IS_VALLEYVIEW(dev
) || IS_CHERRYVIEW(dev
))
4459 vlv_set_rps_idle(dev_priv
);
4461 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.idle_freq
);
4462 dev_priv
->rps
.last_adj
= 0;
4463 I915_WRITE(GEN6_PMINTRMSK
, 0xffffffff);
4465 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4467 spin_lock(&dev_priv
->rps
.client_lock
);
4468 while (!list_empty(&dev_priv
->rps
.clients
))
4469 list_del_init(dev_priv
->rps
.clients
.next
);
4470 spin_unlock(&dev_priv
->rps
.client_lock
);
4473 void gen6_rps_boost(struct drm_i915_private
*dev_priv
,
4474 struct intel_rps_client
*rps
,
4475 unsigned long submitted
)
4477 /* This is intentionally racy! We peek at the state here, then
4478 * validate inside the RPS worker.
4480 if (!(dev_priv
->mm
.busy
&&
4481 dev_priv
->rps
.enabled
&&
4482 dev_priv
->rps
.cur_freq
< dev_priv
->rps
.max_freq_softlimit
))
4485 /* Force a RPS boost (and don't count it against the client) if
4486 * the GPU is severely congested.
4488 if (rps
&& time_after(jiffies
, submitted
+ DRM_I915_THROTTLE_JIFFIES
))
4491 spin_lock(&dev_priv
->rps
.client_lock
);
4492 if (rps
== NULL
|| list_empty(&rps
->link
)) {
4493 spin_lock_irq(&dev_priv
->irq_lock
);
4494 if (dev_priv
->rps
.interrupts_enabled
) {
4495 dev_priv
->rps
.client_boost
= true;
4496 queue_work(dev_priv
->wq
, &dev_priv
->rps
.work
);
4498 spin_unlock_irq(&dev_priv
->irq_lock
);
4501 list_add(&rps
->link
, &dev_priv
->rps
.clients
);
4504 dev_priv
->rps
.boosts
++;
4506 spin_unlock(&dev_priv
->rps
.client_lock
);
4509 void intel_set_rps(struct drm_device
*dev
, u8 val
)
4511 if (IS_VALLEYVIEW(dev
) || IS_CHERRYVIEW(dev
))
4512 valleyview_set_rps(dev
, val
);
4514 gen6_set_rps(dev
, val
);
4517 static void gen9_disable_rps(struct drm_device
*dev
)
4519 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4521 I915_WRITE(GEN6_RC_CONTROL
, 0);
4522 I915_WRITE(GEN9_PG_ENABLE
, 0);
4525 static void gen6_disable_rps(struct drm_device
*dev
)
4527 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4529 I915_WRITE(GEN6_RC_CONTROL
, 0);
4530 I915_WRITE(GEN6_RPNSWREQ
, 1 << 31);
4533 static void cherryview_disable_rps(struct drm_device
*dev
)
4535 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4537 I915_WRITE(GEN6_RC_CONTROL
, 0);
4540 static void valleyview_disable_rps(struct drm_device
*dev
)
4542 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4544 /* we're doing forcewake before Disabling RC6,
4545 * This what the BIOS expects when going into suspend */
4546 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4548 I915_WRITE(GEN6_RC_CONTROL
, 0);
4550 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4553 static void intel_print_rc6_info(struct drm_device
*dev
, u32 mode
)
4555 if (IS_VALLEYVIEW(dev
) || IS_CHERRYVIEW(dev
)) {
4556 if (mode
& (GEN7_RC_CTL_TO_MODE
| GEN6_RC_CTL_EI_MODE(1)))
4557 mode
= GEN6_RC_CTL_RC6_ENABLE
;
4562 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n",
4563 onoff(mode
& GEN6_RC_CTL_RC6_ENABLE
),
4564 onoff(mode
& GEN6_RC_CTL_RC6p_ENABLE
),
4565 onoff(mode
& GEN6_RC_CTL_RC6pp_ENABLE
));
4568 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n",
4569 onoff(mode
& GEN6_RC_CTL_RC6_ENABLE
));
4572 static bool bxt_check_bios_rc6_setup(const struct drm_device
*dev
)
4574 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4575 bool enable_rc6
= true;
4576 unsigned long rc6_ctx_base
;
4578 if (!(I915_READ(RC6_LOCATION
) & RC6_CTX_IN_DRAM
)) {
4579 DRM_DEBUG_KMS("RC6 Base location not set properly.\n");
4584 * The exact context size is not known for BXT, so assume a page size
4587 rc6_ctx_base
= I915_READ(RC6_CTX_BASE
) & RC6_CTX_BASE_MASK
;
4588 if (!((rc6_ctx_base
>= dev_priv
->gtt
.stolen_reserved_base
) &&
4589 (rc6_ctx_base
+ PAGE_SIZE
<= dev_priv
->gtt
.stolen_reserved_base
+
4590 dev_priv
->gtt
.stolen_reserved_size
))) {
4591 DRM_DEBUG_KMS("RC6 Base address not as expected.\n");
4595 if (!(((I915_READ(PWRCTX_MAXCNT_RCSUNIT
) & IDLE_TIME_MASK
) > 1) &&
4596 ((I915_READ(PWRCTX_MAXCNT_VCSUNIT0
) & IDLE_TIME_MASK
) > 1) &&
4597 ((I915_READ(PWRCTX_MAXCNT_BCSUNIT
) & IDLE_TIME_MASK
) > 1) &&
4598 ((I915_READ(PWRCTX_MAXCNT_VECSUNIT
) & IDLE_TIME_MASK
) > 1))) {
4599 DRM_DEBUG_KMS("Engine Idle wait time not set properly.\n");
4603 if (!(I915_READ(GEN6_RC_CONTROL
) & (GEN6_RC_CTL_RC6_ENABLE
|
4604 GEN6_RC_CTL_HW_ENABLE
)) &&
4605 ((I915_READ(GEN6_RC_CONTROL
) & GEN6_RC_CTL_HW_ENABLE
) ||
4606 !(I915_READ(GEN6_RC_STATE
) & RC6_STATE
))) {
4607 DRM_DEBUG_KMS("HW/SW RC6 is not enabled by BIOS.\n");
4614 int sanitize_rc6_option(const struct drm_device
*dev
, int enable_rc6
)
4616 /* No RC6 before Ironlake and code is gone for ilk. */
4617 if (INTEL_INFO(dev
)->gen
< 6)
4623 if (IS_BROXTON(dev
) && !bxt_check_bios_rc6_setup(dev
)) {
4624 DRM_INFO("RC6 disabled by BIOS\n");
4628 /* Respect the kernel parameter if it is set */
4629 if (enable_rc6
>= 0) {
4633 mask
= INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
|
4636 mask
= INTEL_RC6_ENABLE
;
4638 if ((enable_rc6
& mask
) != enable_rc6
)
4639 DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
4640 enable_rc6
& mask
, enable_rc6
, mask
);
4642 return enable_rc6
& mask
;
4645 if (IS_IVYBRIDGE(dev
))
4646 return (INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
);
4648 return INTEL_RC6_ENABLE
;
4651 int intel_enable_rc6(const struct drm_device
*dev
)
4653 return i915
.enable_rc6
;
4656 static void gen6_init_rps_frequencies(struct drm_device
*dev
)
4658 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4659 uint32_t rp_state_cap
;
4660 u32 ddcc_status
= 0;
4663 /* All of these values are in units of 50MHz */
4664 dev_priv
->rps
.cur_freq
= 0;
4665 /* static values from HW: RP0 > RP1 > RPn (min_freq) */
4666 if (IS_BROXTON(dev
)) {
4667 rp_state_cap
= I915_READ(BXT_RP_STATE_CAP
);
4668 dev_priv
->rps
.rp0_freq
= (rp_state_cap
>> 16) & 0xff;
4669 dev_priv
->rps
.rp1_freq
= (rp_state_cap
>> 8) & 0xff;
4670 dev_priv
->rps
.min_freq
= (rp_state_cap
>> 0) & 0xff;
4672 rp_state_cap
= I915_READ(GEN6_RP_STATE_CAP
);
4673 dev_priv
->rps
.rp0_freq
= (rp_state_cap
>> 0) & 0xff;
4674 dev_priv
->rps
.rp1_freq
= (rp_state_cap
>> 8) & 0xff;
4675 dev_priv
->rps
.min_freq
= (rp_state_cap
>> 16) & 0xff;
4678 /* hw_max = RP0 until we check for overclocking */
4679 dev_priv
->rps
.max_freq
= dev_priv
->rps
.rp0_freq
;
4681 dev_priv
->rps
.efficient_freq
= dev_priv
->rps
.rp1_freq
;
4682 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
) ||
4683 IS_SKYLAKE(dev
) || IS_KABYLAKE(dev
)) {
4684 ret
= sandybridge_pcode_read(dev_priv
,
4685 HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL
,
4688 dev_priv
->rps
.efficient_freq
=
4690 ((ddcc_status
>> 8) & 0xff),
4691 dev_priv
->rps
.min_freq
,
4692 dev_priv
->rps
.max_freq
);
4695 if (IS_SKYLAKE(dev
) || IS_KABYLAKE(dev
)) {
4696 /* Store the frequency values in 16.66 MHZ units, which is
4697 the natural hardware unit for SKL */
4698 dev_priv
->rps
.rp0_freq
*= GEN9_FREQ_SCALER
;
4699 dev_priv
->rps
.rp1_freq
*= GEN9_FREQ_SCALER
;
4700 dev_priv
->rps
.min_freq
*= GEN9_FREQ_SCALER
;
4701 dev_priv
->rps
.max_freq
*= GEN9_FREQ_SCALER
;
4702 dev_priv
->rps
.efficient_freq
*= GEN9_FREQ_SCALER
;
4705 dev_priv
->rps
.idle_freq
= dev_priv
->rps
.min_freq
;
4707 /* Preserve min/max settings in case of re-init */
4708 if (dev_priv
->rps
.max_freq_softlimit
== 0)
4709 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
4711 if (dev_priv
->rps
.min_freq_softlimit
== 0) {
4712 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
4713 dev_priv
->rps
.min_freq_softlimit
=
4714 max_t(int, dev_priv
->rps
.efficient_freq
,
4715 intel_freq_opcode(dev_priv
, 450));
4717 dev_priv
->rps
.min_freq_softlimit
=
4718 dev_priv
->rps
.min_freq
;
4722 /* See the Gen9_GT_PM_Programming_Guide doc for the below */
4723 static void gen9_enable_rps(struct drm_device
*dev
)
4725 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4727 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4729 gen6_init_rps_frequencies(dev
);
4731 /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
4732 if (IS_BXT_REVID(dev
, 0, BXT_REVID_A1
)) {
4733 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4737 /* Program defaults and thresholds for RPS*/
4738 I915_WRITE(GEN6_RC_VIDEO_FREQ
,
4739 GEN9_FREQUENCY(dev_priv
->rps
.rp1_freq
));
4741 /* 1 second timeout*/
4742 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
,
4743 GT_INTERVAL_FROM_US(dev_priv
, 1000000));
4745 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 0xa);
4747 /* Leaning on the below call to gen6_set_rps to program/setup the
4748 * Up/Down EI & threshold registers, as well as the RP_CONTROL,
4749 * RP_INTERRUPT_LIMITS & RPNSWREQ registers */
4750 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
4751 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.min_freq_softlimit
);
4753 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4756 static void gen9_enable_rc6(struct drm_device
*dev
)
4758 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4759 struct intel_engine_cs
*ring
;
4760 uint32_t rc6_mask
= 0;
4763 /* 1a: Software RC state - RC0 */
4764 I915_WRITE(GEN6_RC_STATE
, 0);
4766 /* 1b: Get forcewake during program sequence. Although the driver
4767 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
4768 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4770 /* 2a: Disable RC states. */
4771 I915_WRITE(GEN6_RC_CONTROL
, 0);
4773 /* 2b: Program RC6 thresholds.*/
4775 /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */
4776 if (IS_SKYLAKE(dev
))
4777 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 108 << 16);
4779 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 54 << 16);
4780 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
4781 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
4782 for_each_ring(ring
, dev_priv
, unused
)
4783 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
4785 if (HAS_GUC_UCODE(dev
))
4786 I915_WRITE(GUC_MAX_IDLE_COUNT
, 0xA);
4788 I915_WRITE(GEN6_RC_SLEEP
, 0);
4790 /* 2c: Program Coarse Power Gating Policies. */
4791 I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS
, 25);
4792 I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS
, 25);
4794 /* 3a: Enable RC6 */
4795 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
4796 rc6_mask
= GEN6_RC_CTL_RC6_ENABLE
;
4797 DRM_INFO("RC6 %s\n", onoff(rc6_mask
& GEN6_RC_CTL_RC6_ENABLE
));
4798 /* WaRsUseTimeoutMode */
4799 if (IS_SKL_REVID(dev
, 0, SKL_REVID_D0
) ||
4800 IS_BXT_REVID(dev
, 0, BXT_REVID_A1
)) {
4801 I915_WRITE(GEN6_RC6_THRESHOLD
, 625); /* 800us */
4802 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4803 GEN7_RC_CTL_TO_MODE
|
4806 I915_WRITE(GEN6_RC6_THRESHOLD
, 37500); /* 37.5/125ms per EI */
4807 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4808 GEN6_RC_CTL_EI_MODE(1) |
4813 * 3b: Enable Coarse Power Gating only when RC6 is enabled.
4814 * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6.
4816 if (NEEDS_WaRsDisableCoarsePowerGating(dev
))
4817 I915_WRITE(GEN9_PG_ENABLE
, 0);
4819 I915_WRITE(GEN9_PG_ENABLE
, (rc6_mask
& GEN6_RC_CTL_RC6_ENABLE
) ?
4820 (GEN9_RENDER_PG_ENABLE
| GEN9_MEDIA_PG_ENABLE
) : 0);
4822 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4826 static void gen8_enable_rps(struct drm_device
*dev
)
4828 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4829 struct intel_engine_cs
*ring
;
4830 uint32_t rc6_mask
= 0;
4833 /* 1a: Software RC state - RC0 */
4834 I915_WRITE(GEN6_RC_STATE
, 0);
4836 /* 1c & 1d: Get forcewake during program sequence. Although the driver
4837 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
4838 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4840 /* 2a: Disable RC states. */
4841 I915_WRITE(GEN6_RC_CONTROL
, 0);
4843 /* Initialize rps frequencies */
4844 gen6_init_rps_frequencies(dev
);
4846 /* 2b: Program RC6 thresholds.*/
4847 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16);
4848 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
4849 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
4850 for_each_ring(ring
, dev_priv
, unused
)
4851 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
4852 I915_WRITE(GEN6_RC_SLEEP
, 0);
4853 if (IS_BROADWELL(dev
))
4854 I915_WRITE(GEN6_RC6_THRESHOLD
, 625); /* 800us/1.28 for TO */
4856 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000); /* 50/125ms per EI */
4859 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
4860 rc6_mask
= GEN6_RC_CTL_RC6_ENABLE
;
4861 intel_print_rc6_info(dev
, rc6_mask
);
4862 if (IS_BROADWELL(dev
))
4863 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4864 GEN7_RC_CTL_TO_MODE
|
4867 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
4868 GEN6_RC_CTL_EI_MODE(1) |
4871 /* 4 Program defaults and thresholds for RPS*/
4872 I915_WRITE(GEN6_RPNSWREQ
,
4873 HSW_FREQUENCY(dev_priv
->rps
.rp1_freq
));
4874 I915_WRITE(GEN6_RC_VIDEO_FREQ
,
4875 HSW_FREQUENCY(dev_priv
->rps
.rp1_freq
));
4876 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
4877 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 100000000 / 128); /* 1 second timeout */
4879 /* Docs recommend 900MHz, and 300 MHz respectively */
4880 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
,
4881 dev_priv
->rps
.max_freq_softlimit
<< 24 |
4882 dev_priv
->rps
.min_freq_softlimit
<< 16);
4884 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 7600000 / 128); /* 76ms busyness per EI, 90% */
4885 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 31300000 / 128); /* 313ms busyness per EI, 70%*/
4886 I915_WRITE(GEN6_RP_UP_EI
, 66000); /* 84.48ms, XXX: random? */
4887 I915_WRITE(GEN6_RP_DOWN_EI
, 350000); /* 448ms, XXX: random? */
4889 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
4892 I915_WRITE(GEN6_RP_CONTROL
,
4893 GEN6_RP_MEDIA_TURBO
|
4894 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
4895 GEN6_RP_MEDIA_IS_GFX
|
4897 GEN6_RP_UP_BUSY_AVG
|
4898 GEN6_RP_DOWN_IDLE_AVG
);
4900 /* 6: Ring frequency + overclocking (our driver does this later */
4902 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
4903 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.idle_freq
);
4905 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
4908 static void gen6_enable_rps(struct drm_device
*dev
)
4910 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4911 struct intel_engine_cs
*ring
;
4912 u32 rc6vids
, pcu_mbox
= 0, rc6_mask
= 0;
4917 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
4919 /* Here begins a magic sequence of register writes to enable
4920 * auto-downclocking.
4922 * Perhaps there might be some value in exposing these to
4925 I915_WRITE(GEN6_RC_STATE
, 0);
4927 /* Clear the DBG now so we don't confuse earlier errors */
4928 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
4929 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg
);
4930 I915_WRITE(GTFIFODBG
, gtfifodbg
);
4933 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
4935 /* Initialize rps frequencies */
4936 gen6_init_rps_frequencies(dev
);
4938 /* disable the counters and set deterministic thresholds */
4939 I915_WRITE(GEN6_RC_CONTROL
, 0);
4941 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT
, 1000 << 16);
4942 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16 | 30);
4943 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT
, 30);
4944 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
4945 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
4947 for_each_ring(ring
, dev_priv
, i
)
4948 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
4950 I915_WRITE(GEN6_RC_SLEEP
, 0);
4951 I915_WRITE(GEN6_RC1e_THRESHOLD
, 1000);
4952 if (IS_IVYBRIDGE(dev
))
4953 I915_WRITE(GEN6_RC6_THRESHOLD
, 125000);
4955 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000);
4956 I915_WRITE(GEN6_RC6p_THRESHOLD
, 150000);
4957 I915_WRITE(GEN6_RC6pp_THRESHOLD
, 64000); /* unused */
4959 /* Check if we are enabling RC6 */
4960 rc6_mode
= intel_enable_rc6(dev_priv
->dev
);
4961 if (rc6_mode
& INTEL_RC6_ENABLE
)
4962 rc6_mask
|= GEN6_RC_CTL_RC6_ENABLE
;
4964 /* We don't use those on Haswell */
4965 if (!IS_HASWELL(dev
)) {
4966 if (rc6_mode
& INTEL_RC6p_ENABLE
)
4967 rc6_mask
|= GEN6_RC_CTL_RC6p_ENABLE
;
4969 if (rc6_mode
& INTEL_RC6pp_ENABLE
)
4970 rc6_mask
|= GEN6_RC_CTL_RC6pp_ENABLE
;
4973 intel_print_rc6_info(dev
, rc6_mask
);
4975 I915_WRITE(GEN6_RC_CONTROL
,
4977 GEN6_RC_CTL_EI_MODE(1) |
4978 GEN6_RC_CTL_HW_ENABLE
);
4980 /* Power down if completely idle for over 50ms */
4981 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 50000);
4982 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
4984 ret
= sandybridge_pcode_write(dev_priv
, GEN6_PCODE_WRITE_MIN_FREQ_TABLE
, 0);
4986 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
4988 ret
= sandybridge_pcode_read(dev_priv
, GEN6_READ_OC_PARAMS
, &pcu_mbox
);
4989 if (!ret
&& (pcu_mbox
& (1<<31))) { /* OC supported */
4990 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
4991 (dev_priv
->rps
.max_freq_softlimit
& 0xff) * 50,
4992 (pcu_mbox
& 0xff) * 50);
4993 dev_priv
->rps
.max_freq
= pcu_mbox
& 0xff;
4996 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
4997 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.idle_freq
);
5000 ret
= sandybridge_pcode_read(dev_priv
, GEN6_PCODE_READ_RC6VIDS
, &rc6vids
);
5001 if (IS_GEN6(dev
) && ret
) {
5002 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
5003 } else if (IS_GEN6(dev
) && (GEN6_DECODE_RC6_VID(rc6vids
& 0xff) < 450)) {
5004 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
5005 GEN6_DECODE_RC6_VID(rc6vids
& 0xff), 450);
5006 rc6vids
&= 0xffff00;
5007 rc6vids
|= GEN6_ENCODE_RC6_VID(450);
5008 ret
= sandybridge_pcode_write(dev_priv
, GEN6_PCODE_WRITE_RC6VIDS
, rc6vids
);
5010 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
5013 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
5016 static void __gen6_update_ring_freq(struct drm_device
*dev
)
5018 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5020 unsigned int gpu_freq
;
5021 unsigned int max_ia_freq
, min_ring_freq
;
5022 unsigned int max_gpu_freq
, min_gpu_freq
;
5023 int scaling_factor
= 180;
5024 struct cpufreq_policy
*policy
;
5026 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5028 policy
= cpufreq_cpu_get(0);
5030 max_ia_freq
= policy
->cpuinfo
.max_freq
;
5031 cpufreq_cpu_put(policy
);
5034 * Default to measured freq if none found, PCU will ensure we
5037 max_ia_freq
= tsc_khz
;
5040 /* Convert from kHz to MHz */
5041 max_ia_freq
/= 1000;
5043 min_ring_freq
= I915_READ(DCLK
) & 0xf;
5044 /* convert DDR frequency from units of 266.6MHz to bandwidth */
5045 min_ring_freq
= mult_frac(min_ring_freq
, 8, 3);
5047 if (IS_SKYLAKE(dev
) || IS_KABYLAKE(dev
)) {
5048 /* Convert GT frequency to 50 HZ units */
5049 min_gpu_freq
= dev_priv
->rps
.min_freq
/ GEN9_FREQ_SCALER
;
5050 max_gpu_freq
= dev_priv
->rps
.max_freq
/ GEN9_FREQ_SCALER
;
5052 min_gpu_freq
= dev_priv
->rps
.min_freq
;
5053 max_gpu_freq
= dev_priv
->rps
.max_freq
;
5057 * For each potential GPU frequency, load a ring frequency we'd like
5058 * to use for memory access. We do this by specifying the IA frequency
5059 * the PCU should use as a reference to determine the ring frequency.
5061 for (gpu_freq
= max_gpu_freq
; gpu_freq
>= min_gpu_freq
; gpu_freq
--) {
5062 int diff
= max_gpu_freq
- gpu_freq
;
5063 unsigned int ia_freq
= 0, ring_freq
= 0;
5065 if (IS_SKYLAKE(dev
) || IS_KABYLAKE(dev
)) {
5067 * ring_freq = 2 * GT. ring_freq is in 100MHz units
5068 * No floor required for ring frequency on SKL.
5070 ring_freq
= gpu_freq
;
5071 } else if (INTEL_INFO(dev
)->gen
>= 8) {
5072 /* max(2 * GT, DDR). NB: GT is 50MHz units */
5073 ring_freq
= max(min_ring_freq
, gpu_freq
);
5074 } else if (IS_HASWELL(dev
)) {
5075 ring_freq
= mult_frac(gpu_freq
, 5, 4);
5076 ring_freq
= max(min_ring_freq
, ring_freq
);
5077 /* leave ia_freq as the default, chosen by cpufreq */
5079 /* On older processors, there is no separate ring
5080 * clock domain, so in order to boost the bandwidth
5081 * of the ring, we need to upclock the CPU (ia_freq).
5083 * For GPU frequencies less than 750MHz,
5084 * just use the lowest ring freq.
5086 if (gpu_freq
< min_freq
)
5089 ia_freq
= max_ia_freq
- ((diff
* scaling_factor
) / 2);
5090 ia_freq
= DIV_ROUND_CLOSEST(ia_freq
, 100);
5093 sandybridge_pcode_write(dev_priv
,
5094 GEN6_PCODE_WRITE_MIN_FREQ_TABLE
,
5095 ia_freq
<< GEN6_PCODE_FREQ_IA_RATIO_SHIFT
|
5096 ring_freq
<< GEN6_PCODE_FREQ_RING_RATIO_SHIFT
|
5101 void gen6_update_ring_freq(struct drm_device
*dev
)
5103 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5105 if (!HAS_CORE_RING_FREQ(dev
))
5108 mutex_lock(&dev_priv
->rps
.hw_lock
);
5109 __gen6_update_ring_freq(dev
);
5110 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5113 static int cherryview_rps_max_freq(struct drm_i915_private
*dev_priv
)
5115 struct drm_device
*dev
= dev_priv
->dev
;
5118 val
= vlv_punit_read(dev_priv
, FB_GFX_FMAX_AT_VMAX_FUSE
);
5120 switch (INTEL_INFO(dev
)->eu_total
) {
5122 /* (2 * 4) config */
5123 rp0
= (val
>> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT
);
5126 /* (2 * 6) config */
5127 rp0
= (val
>> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT
);
5130 /* (2 * 8) config */
5132 /* Setting (2 * 8) Min RP0 for any other combination */
5133 rp0
= (val
>> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT
);
5137 rp0
= (rp0
& FB_GFX_FREQ_FUSE_MASK
);
5142 static int cherryview_rps_rpe_freq(struct drm_i915_private
*dev_priv
)
5146 val
= vlv_punit_read(dev_priv
, PUNIT_GPU_DUTYCYCLE_REG
);
5147 rpe
= (val
>> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT
) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK
;
5152 static int cherryview_rps_guar_freq(struct drm_i915_private
*dev_priv
)
5156 val
= vlv_punit_read(dev_priv
, FB_GFX_FMAX_AT_VMAX_FUSE
);
5157 rp1
= (val
& FB_GFX_FREQ_FUSE_MASK
);
5162 static int valleyview_rps_guar_freq(struct drm_i915_private
*dev_priv
)
5166 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FREQ_FUSE
);
5168 rp1
= (val
& FB_GFX_FGUARANTEED_FREQ_FUSE_MASK
) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT
;
5173 static int valleyview_rps_max_freq(struct drm_i915_private
*dev_priv
)
5177 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FREQ_FUSE
);
5179 rp0
= (val
& FB_GFX_MAX_FREQ_FUSE_MASK
) >> FB_GFX_MAX_FREQ_FUSE_SHIFT
;
5181 rp0
= min_t(u32
, rp0
, 0xea);
5186 static int valleyview_rps_rpe_freq(struct drm_i915_private
*dev_priv
)
5190 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FMAX_FUSE_LO
);
5191 rpe
= (val
& FB_FMAX_VMIN_FREQ_LO_MASK
) >> FB_FMAX_VMIN_FREQ_LO_SHIFT
;
5192 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FMAX_FUSE_HI
);
5193 rpe
|= (val
& FB_FMAX_VMIN_FREQ_HI_MASK
) << 5;
5198 static int valleyview_rps_min_freq(struct drm_i915_private
*dev_priv
)
5202 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_LFM
) & 0xff;
5204 * According to the BYT Punit GPU turbo HAS 1.1.6.3 the minimum value
5205 * for the minimum frequency in GPLL mode is 0xc1. Contrary to this on
5206 * a BYT-M B0 the above register contains 0xbf. Moreover when setting
5207 * a frequency Punit will not allow values below 0xc0. Clamp it 0xc0
5208 * to make sure it matches what Punit accepts.
5210 return max_t(u32
, val
, 0xc0);
5213 /* Check that the pctx buffer wasn't move under us. */
5214 static void valleyview_check_pctx(struct drm_i915_private
*dev_priv
)
5216 unsigned long pctx_addr
= I915_READ(VLV_PCBR
) & ~4095;
5218 WARN_ON(pctx_addr
!= dev_priv
->mm
.stolen_base
+
5219 dev_priv
->vlv_pctx
->stolen
->start
);
5223 /* Check that the pcbr address is not empty. */
5224 static void cherryview_check_pctx(struct drm_i915_private
*dev_priv
)
5226 unsigned long pctx_addr
= I915_READ(VLV_PCBR
) & ~4095;
5228 WARN_ON((pctx_addr
>> VLV_PCBR_ADDR_SHIFT
) == 0);
5231 static void cherryview_setup_pctx(struct drm_device
*dev
)
5233 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5234 unsigned long pctx_paddr
, paddr
;
5235 struct i915_gtt
*gtt
= &dev_priv
->gtt
;
5237 int pctx_size
= 32*1024;
5239 pcbr
= I915_READ(VLV_PCBR
);
5240 if ((pcbr
>> VLV_PCBR_ADDR_SHIFT
) == 0) {
5241 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
5242 paddr
= (dev_priv
->mm
.stolen_base
+
5243 (gtt
->stolen_size
- pctx_size
));
5245 pctx_paddr
= (paddr
& (~4095));
5246 I915_WRITE(VLV_PCBR
, pctx_paddr
);
5249 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR
));
5252 static void valleyview_setup_pctx(struct drm_device
*dev
)
5254 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5255 struct drm_i915_gem_object
*pctx
;
5256 unsigned long pctx_paddr
;
5258 int pctx_size
= 24*1024;
5260 mutex_lock(&dev
->struct_mutex
);
5262 pcbr
= I915_READ(VLV_PCBR
);
5264 /* BIOS set it up already, grab the pre-alloc'd space */
5267 pcbr_offset
= (pcbr
& (~4095)) - dev_priv
->mm
.stolen_base
;
5268 pctx
= i915_gem_object_create_stolen_for_preallocated(dev_priv
->dev
,
5270 I915_GTT_OFFSET_NONE
,
5275 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
5278 * From the Gunit register HAS:
5279 * The Gfx driver is expected to program this register and ensure
5280 * proper allocation within Gfx stolen memory. For example, this
5281 * register should be programmed such than the PCBR range does not
5282 * overlap with other ranges, such as the frame buffer, protected
5283 * memory, or any other relevant ranges.
5285 pctx
= i915_gem_object_create_stolen(dev
, pctx_size
);
5287 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
5291 pctx_paddr
= dev_priv
->mm
.stolen_base
+ pctx
->stolen
->start
;
5292 I915_WRITE(VLV_PCBR
, pctx_paddr
);
5295 DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR
));
5296 dev_priv
->vlv_pctx
= pctx
;
5297 mutex_unlock(&dev
->struct_mutex
);
5300 static void valleyview_cleanup_pctx(struct drm_device
*dev
)
5302 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5304 if (WARN_ON(!dev_priv
->vlv_pctx
))
5307 drm_gem_object_unreference_unlocked(&dev_priv
->vlv_pctx
->base
);
5308 dev_priv
->vlv_pctx
= NULL
;
5311 static void valleyview_init_gt_powersave(struct drm_device
*dev
)
5313 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5316 valleyview_setup_pctx(dev
);
5318 mutex_lock(&dev_priv
->rps
.hw_lock
);
5320 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5321 switch ((val
>> 6) & 3) {
5324 dev_priv
->mem_freq
= 800;
5327 dev_priv
->mem_freq
= 1066;
5330 dev_priv
->mem_freq
= 1333;
5333 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv
->mem_freq
);
5335 dev_priv
->rps
.max_freq
= valleyview_rps_max_freq(dev_priv
);
5336 dev_priv
->rps
.rp0_freq
= dev_priv
->rps
.max_freq
;
5337 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
5338 intel_gpu_freq(dev_priv
, dev_priv
->rps
.max_freq
),
5339 dev_priv
->rps
.max_freq
);
5341 dev_priv
->rps
.efficient_freq
= valleyview_rps_rpe_freq(dev_priv
);
5342 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
5343 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5344 dev_priv
->rps
.efficient_freq
);
5346 dev_priv
->rps
.rp1_freq
= valleyview_rps_guar_freq(dev_priv
);
5347 DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n",
5348 intel_gpu_freq(dev_priv
, dev_priv
->rps
.rp1_freq
),
5349 dev_priv
->rps
.rp1_freq
);
5351 dev_priv
->rps
.min_freq
= valleyview_rps_min_freq(dev_priv
);
5352 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
5353 intel_gpu_freq(dev_priv
, dev_priv
->rps
.min_freq
),
5354 dev_priv
->rps
.min_freq
);
5356 dev_priv
->rps
.idle_freq
= dev_priv
->rps
.min_freq
;
5358 /* Preserve min/max settings in case of re-init */
5359 if (dev_priv
->rps
.max_freq_softlimit
== 0)
5360 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
5362 if (dev_priv
->rps
.min_freq_softlimit
== 0)
5363 dev_priv
->rps
.min_freq_softlimit
= dev_priv
->rps
.min_freq
;
5365 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5368 static void cherryview_init_gt_powersave(struct drm_device
*dev
)
5370 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5373 cherryview_setup_pctx(dev
);
5375 mutex_lock(&dev_priv
->rps
.hw_lock
);
5377 mutex_lock(&dev_priv
->sb_lock
);
5378 val
= vlv_cck_read(dev_priv
, CCK_FUSE_REG
);
5379 mutex_unlock(&dev_priv
->sb_lock
);
5381 switch ((val
>> 2) & 0x7) {
5383 dev_priv
->mem_freq
= 2000;
5386 dev_priv
->mem_freq
= 1600;
5389 DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv
->mem_freq
);
5391 dev_priv
->rps
.max_freq
= cherryview_rps_max_freq(dev_priv
);
5392 dev_priv
->rps
.rp0_freq
= dev_priv
->rps
.max_freq
;
5393 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
5394 intel_gpu_freq(dev_priv
, dev_priv
->rps
.max_freq
),
5395 dev_priv
->rps
.max_freq
);
5397 dev_priv
->rps
.efficient_freq
= cherryview_rps_rpe_freq(dev_priv
);
5398 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
5399 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5400 dev_priv
->rps
.efficient_freq
);
5402 dev_priv
->rps
.rp1_freq
= cherryview_rps_guar_freq(dev_priv
);
5403 DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n",
5404 intel_gpu_freq(dev_priv
, dev_priv
->rps
.rp1_freq
),
5405 dev_priv
->rps
.rp1_freq
);
5407 /* PUnit validated range is only [RPe, RP0] */
5408 dev_priv
->rps
.min_freq
= dev_priv
->rps
.efficient_freq
;
5409 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
5410 intel_gpu_freq(dev_priv
, dev_priv
->rps
.min_freq
),
5411 dev_priv
->rps
.min_freq
);
5413 WARN_ONCE((dev_priv
->rps
.max_freq
|
5414 dev_priv
->rps
.efficient_freq
|
5415 dev_priv
->rps
.rp1_freq
|
5416 dev_priv
->rps
.min_freq
) & 1,
5417 "Odd GPU freq values\n");
5419 dev_priv
->rps
.idle_freq
= dev_priv
->rps
.min_freq
;
5421 /* Preserve min/max settings in case of re-init */
5422 if (dev_priv
->rps
.max_freq_softlimit
== 0)
5423 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
5425 if (dev_priv
->rps
.min_freq_softlimit
== 0)
5426 dev_priv
->rps
.min_freq_softlimit
= dev_priv
->rps
.min_freq
;
5428 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5431 static void valleyview_cleanup_gt_powersave(struct drm_device
*dev
)
5433 valleyview_cleanup_pctx(dev
);
5436 static void cherryview_enable_rps(struct drm_device
*dev
)
5438 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5439 struct intel_engine_cs
*ring
;
5440 u32 gtfifodbg
, val
, rc6_mode
= 0, pcbr
;
5443 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5445 gtfifodbg
= I915_READ(GTFIFODBG
);
5447 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5449 I915_WRITE(GTFIFODBG
, gtfifodbg
);
5452 cherryview_check_pctx(dev_priv
);
5454 /* 1a & 1b: Get forcewake during program sequence. Although the driver
5455 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
5456 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
5458 /* Disable RC states. */
5459 I915_WRITE(GEN6_RC_CONTROL
, 0);
5461 /* 2a: Program RC6 thresholds.*/
5462 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16);
5463 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
5464 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
5466 for_each_ring(ring
, dev_priv
, i
)
5467 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
5468 I915_WRITE(GEN6_RC_SLEEP
, 0);
5470 /* TO threshold set to 500 us ( 0x186 * 1.28 us) */
5471 I915_WRITE(GEN6_RC6_THRESHOLD
, 0x186);
5473 /* allows RC6 residency counter to work */
5474 I915_WRITE(VLV_COUNTER_CONTROL
,
5475 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH
|
5476 VLV_MEDIA_RC6_COUNT_EN
|
5477 VLV_RENDER_RC6_COUNT_EN
));
5479 /* For now we assume BIOS is allocating and populating the PCBR */
5480 pcbr
= I915_READ(VLV_PCBR
);
5483 if ((intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
) &&
5484 (pcbr
>> VLV_PCBR_ADDR_SHIFT
))
5485 rc6_mode
= GEN7_RC_CTL_TO_MODE
;
5487 I915_WRITE(GEN6_RC_CONTROL
, rc6_mode
);
5489 /* 4 Program defaults and thresholds for RPS*/
5490 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 1000000);
5491 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 59400);
5492 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 245000);
5493 I915_WRITE(GEN6_RP_UP_EI
, 66000);
5494 I915_WRITE(GEN6_RP_DOWN_EI
, 350000);
5496 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
5499 I915_WRITE(GEN6_RP_CONTROL
,
5500 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
5501 GEN6_RP_MEDIA_IS_GFX
|
5503 GEN6_RP_UP_BUSY_AVG
|
5504 GEN6_RP_DOWN_IDLE_AVG
);
5506 /* Setting Fixed Bias */
5507 val
= VLV_OVERRIDE_EN
|
5509 CHV_BIAS_CPU_50_SOC_50
;
5510 vlv_punit_write(dev_priv
, VLV_TURBO_SOC_OVERRIDE
, val
);
5512 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5514 /* RPS code assumes GPLL is used */
5515 WARN_ONCE((val
& GPLLENABLE
) == 0, "GPLL not enabled\n");
5517 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val
& GPLLENABLE
));
5518 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val
);
5520 dev_priv
->rps
.cur_freq
= (val
>> 8) & 0xff;
5521 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
5522 intel_gpu_freq(dev_priv
, dev_priv
->rps
.cur_freq
),
5523 dev_priv
->rps
.cur_freq
);
5525 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
5526 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5527 dev_priv
->rps
.efficient_freq
);
5529 valleyview_set_rps(dev_priv
->dev
, dev_priv
->rps
.efficient_freq
);
5531 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
5534 static void valleyview_enable_rps(struct drm_device
*dev
)
5536 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5537 struct intel_engine_cs
*ring
;
5538 u32 gtfifodbg
, val
, rc6_mode
= 0;
5541 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
5543 valleyview_check_pctx(dev_priv
);
5545 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
5546 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
5548 I915_WRITE(GTFIFODBG
, gtfifodbg
);
5551 /* If VLV, Forcewake all wells, else re-direct to regular path */
5552 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_ALL
);
5554 /* Disable RC states. */
5555 I915_WRITE(GEN6_RC_CONTROL
, 0);
5557 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 1000000);
5558 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 59400);
5559 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 245000);
5560 I915_WRITE(GEN6_RP_UP_EI
, 66000);
5561 I915_WRITE(GEN6_RP_DOWN_EI
, 350000);
5563 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
5565 I915_WRITE(GEN6_RP_CONTROL
,
5566 GEN6_RP_MEDIA_TURBO
|
5567 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
5568 GEN6_RP_MEDIA_IS_GFX
|
5570 GEN6_RP_UP_BUSY_AVG
|
5571 GEN6_RP_DOWN_IDLE_CONT
);
5573 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 0x00280000);
5574 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
5575 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
5577 for_each_ring(ring
, dev_priv
, i
)
5578 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
5580 I915_WRITE(GEN6_RC6_THRESHOLD
, 0x557);
5582 /* allows RC6 residency counter to work */
5583 I915_WRITE(VLV_COUNTER_CONTROL
,
5584 _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN
|
5585 VLV_RENDER_RC0_COUNT_EN
|
5586 VLV_MEDIA_RC6_COUNT_EN
|
5587 VLV_RENDER_RC6_COUNT_EN
));
5589 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
5590 rc6_mode
= GEN7_RC_CTL_TO_MODE
| VLV_RC_CTL_CTX_RST_PARALLEL
;
5592 intel_print_rc6_info(dev
, rc6_mode
);
5594 I915_WRITE(GEN6_RC_CONTROL
, rc6_mode
);
5596 /* Setting Fixed Bias */
5597 val
= VLV_OVERRIDE_EN
|
5599 VLV_BIAS_CPU_125_SOC_875
;
5600 vlv_punit_write(dev_priv
, VLV_TURBO_SOC_OVERRIDE
, val
);
5602 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5604 /* RPS code assumes GPLL is used */
5605 WARN_ONCE((val
& GPLLENABLE
) == 0, "GPLL not enabled\n");
5607 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val
& GPLLENABLE
));
5608 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val
);
5610 dev_priv
->rps
.cur_freq
= (val
>> 8) & 0xff;
5611 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
5612 intel_gpu_freq(dev_priv
, dev_priv
->rps
.cur_freq
),
5613 dev_priv
->rps
.cur_freq
);
5615 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
5616 intel_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
5617 dev_priv
->rps
.efficient_freq
);
5619 valleyview_set_rps(dev_priv
->dev
, dev_priv
->rps
.efficient_freq
);
5621 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_ALL
);
5624 static unsigned long intel_pxfreq(u32 vidfreq
)
5627 int div
= (vidfreq
& 0x3f0000) >> 16;
5628 int post
= (vidfreq
& 0x3000) >> 12;
5629 int pre
= (vidfreq
& 0x7);
5634 freq
= ((div
* 133333) / ((1<<post
) * pre
));
5639 static const struct cparams
{
5645 { 1, 1333, 301, 28664 },
5646 { 1, 1066, 294, 24460 },
5647 { 1, 800, 294, 25192 },
5648 { 0, 1333, 276, 27605 },
5649 { 0, 1066, 276, 27605 },
5650 { 0, 800, 231, 23784 },
5653 static unsigned long __i915_chipset_val(struct drm_i915_private
*dev_priv
)
5655 u64 total_count
, diff
, ret
;
5656 u32 count1
, count2
, count3
, m
= 0, c
= 0;
5657 unsigned long now
= jiffies_to_msecs(jiffies
), diff1
;
5660 assert_spin_locked(&mchdev_lock
);
5662 diff1
= now
- dev_priv
->ips
.last_time1
;
5664 /* Prevent division-by-zero if we are asking too fast.
5665 * Also, we don't get interesting results if we are polling
5666 * faster than once in 10ms, so just return the saved value
5670 return dev_priv
->ips
.chipset_power
;
5672 count1
= I915_READ(DMIEC
);
5673 count2
= I915_READ(DDREC
);
5674 count3
= I915_READ(CSIEC
);
5676 total_count
= count1
+ count2
+ count3
;
5678 /* FIXME: handle per-counter overflow */
5679 if (total_count
< dev_priv
->ips
.last_count1
) {
5680 diff
= ~0UL - dev_priv
->ips
.last_count1
;
5681 diff
+= total_count
;
5683 diff
= total_count
- dev_priv
->ips
.last_count1
;
5686 for (i
= 0; i
< ARRAY_SIZE(cparams
); i
++) {
5687 if (cparams
[i
].i
== dev_priv
->ips
.c_m
&&
5688 cparams
[i
].t
== dev_priv
->ips
.r_t
) {
5695 diff
= div_u64(diff
, diff1
);
5696 ret
= ((m
* diff
) + c
);
5697 ret
= div_u64(ret
, 10);
5699 dev_priv
->ips
.last_count1
= total_count
;
5700 dev_priv
->ips
.last_time1
= now
;
5702 dev_priv
->ips
.chipset_power
= ret
;
5707 unsigned long i915_chipset_val(struct drm_i915_private
*dev_priv
)
5709 struct drm_device
*dev
= dev_priv
->dev
;
5712 if (INTEL_INFO(dev
)->gen
!= 5)
5715 spin_lock_irq(&mchdev_lock
);
5717 val
= __i915_chipset_val(dev_priv
);
5719 spin_unlock_irq(&mchdev_lock
);
5724 unsigned long i915_mch_val(struct drm_i915_private
*dev_priv
)
5726 unsigned long m
, x
, b
;
5729 tsfs
= I915_READ(TSFS
);
5731 m
= ((tsfs
& TSFS_SLOPE_MASK
) >> TSFS_SLOPE_SHIFT
);
5732 x
= I915_READ8(TR1
);
5734 b
= tsfs
& TSFS_INTR_MASK
;
5736 return ((m
* x
) / 127) - b
;
5739 static int _pxvid_to_vd(u8 pxvid
)
5744 if (pxvid
>= 8 && pxvid
< 31)
5747 return (pxvid
+ 2) * 125;
5750 static u32
pvid_to_extvid(struct drm_i915_private
*dev_priv
, u8 pxvid
)
5752 struct drm_device
*dev
= dev_priv
->dev
;
5753 const int vd
= _pxvid_to_vd(pxvid
);
5754 const int vm
= vd
- 1125;
5756 if (INTEL_INFO(dev
)->is_mobile
)
5757 return vm
> 0 ? vm
: 0;
5762 static void __i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
5764 u64 now
, diff
, diffms
;
5767 assert_spin_locked(&mchdev_lock
);
5769 now
= ktime_get_raw_ns();
5770 diffms
= now
- dev_priv
->ips
.last_time2
;
5771 do_div(diffms
, NSEC_PER_MSEC
);
5773 /* Don't divide by 0 */
5777 count
= I915_READ(GFXEC
);
5779 if (count
< dev_priv
->ips
.last_count2
) {
5780 diff
= ~0UL - dev_priv
->ips
.last_count2
;
5783 diff
= count
- dev_priv
->ips
.last_count2
;
5786 dev_priv
->ips
.last_count2
= count
;
5787 dev_priv
->ips
.last_time2
= now
;
5789 /* More magic constants... */
5791 diff
= div_u64(diff
, diffms
* 10);
5792 dev_priv
->ips
.gfx_power
= diff
;
5795 void i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
5797 struct drm_device
*dev
= dev_priv
->dev
;
5799 if (INTEL_INFO(dev
)->gen
!= 5)
5802 spin_lock_irq(&mchdev_lock
);
5804 __i915_update_gfx_val(dev_priv
);
5806 spin_unlock_irq(&mchdev_lock
);
5809 static unsigned long __i915_gfx_val(struct drm_i915_private
*dev_priv
)
5811 unsigned long t
, corr
, state1
, corr2
, state2
;
5814 assert_spin_locked(&mchdev_lock
);
5816 pxvid
= I915_READ(PXVFREQ(dev_priv
->rps
.cur_freq
));
5817 pxvid
= (pxvid
>> 24) & 0x7f;
5818 ext_v
= pvid_to_extvid(dev_priv
, pxvid
);
5822 t
= i915_mch_val(dev_priv
);
5824 /* Revel in the empirically derived constants */
5826 /* Correction factor in 1/100000 units */
5828 corr
= ((t
* 2349) + 135940);
5830 corr
= ((t
* 964) + 29317);
5832 corr
= ((t
* 301) + 1004);
5834 corr
= corr
* ((150142 * state1
) / 10000 - 78642);
5836 corr2
= (corr
* dev_priv
->ips
.corr
);
5838 state2
= (corr2
* state1
) / 10000;
5839 state2
/= 100; /* convert to mW */
5841 __i915_update_gfx_val(dev_priv
);
5843 return dev_priv
->ips
.gfx_power
+ state2
;
5846 unsigned long i915_gfx_val(struct drm_i915_private
*dev_priv
)
5848 struct drm_device
*dev
= dev_priv
->dev
;
5851 if (INTEL_INFO(dev
)->gen
!= 5)
5854 spin_lock_irq(&mchdev_lock
);
5856 val
= __i915_gfx_val(dev_priv
);
5858 spin_unlock_irq(&mchdev_lock
);
5864 * i915_read_mch_val - return value for IPS use
5866 * Calculate and return a value for the IPS driver to use when deciding whether
5867 * we have thermal and power headroom to increase CPU or GPU power budget.
5869 unsigned long i915_read_mch_val(void)
5871 struct drm_i915_private
*dev_priv
;
5872 unsigned long chipset_val
, graphics_val
, ret
= 0;
5874 spin_lock_irq(&mchdev_lock
);
5877 dev_priv
= i915_mch_dev
;
5879 chipset_val
= __i915_chipset_val(dev_priv
);
5880 graphics_val
= __i915_gfx_val(dev_priv
);
5882 ret
= chipset_val
+ graphics_val
;
5885 spin_unlock_irq(&mchdev_lock
);
5889 EXPORT_SYMBOL_GPL(i915_read_mch_val
);
5892 * i915_gpu_raise - raise GPU frequency limit
5894 * Raise the limit; IPS indicates we have thermal headroom.
5896 bool i915_gpu_raise(void)
5898 struct drm_i915_private
*dev_priv
;
5901 spin_lock_irq(&mchdev_lock
);
5902 if (!i915_mch_dev
) {
5906 dev_priv
= i915_mch_dev
;
5908 if (dev_priv
->ips
.max_delay
> dev_priv
->ips
.fmax
)
5909 dev_priv
->ips
.max_delay
--;
5912 spin_unlock_irq(&mchdev_lock
);
5916 EXPORT_SYMBOL_GPL(i915_gpu_raise
);
5919 * i915_gpu_lower - lower GPU frequency limit
5921 * IPS indicates we're close to a thermal limit, so throttle back the GPU
5922 * frequency maximum.
5924 bool i915_gpu_lower(void)
5926 struct drm_i915_private
*dev_priv
;
5929 spin_lock_irq(&mchdev_lock
);
5930 if (!i915_mch_dev
) {
5934 dev_priv
= i915_mch_dev
;
5936 if (dev_priv
->ips
.max_delay
< dev_priv
->ips
.min_delay
)
5937 dev_priv
->ips
.max_delay
++;
5940 spin_unlock_irq(&mchdev_lock
);
5944 EXPORT_SYMBOL_GPL(i915_gpu_lower
);
5947 * i915_gpu_busy - indicate GPU business to IPS
5949 * Tell the IPS driver whether or not the GPU is busy.
5951 bool i915_gpu_busy(void)
5953 struct drm_i915_private
*dev_priv
;
5954 struct intel_engine_cs
*ring
;
5958 spin_lock_irq(&mchdev_lock
);
5961 dev_priv
= i915_mch_dev
;
5963 for_each_ring(ring
, dev_priv
, i
)
5964 ret
|= !list_empty(&ring
->request_list
);
5967 spin_unlock_irq(&mchdev_lock
);
5971 EXPORT_SYMBOL_GPL(i915_gpu_busy
);
5974 * i915_gpu_turbo_disable - disable graphics turbo
5976 * Disable graphics turbo by resetting the max frequency and setting the
5977 * current frequency to the default.
5979 bool i915_gpu_turbo_disable(void)
5981 struct drm_i915_private
*dev_priv
;
5984 spin_lock_irq(&mchdev_lock
);
5985 if (!i915_mch_dev
) {
5989 dev_priv
= i915_mch_dev
;
5991 dev_priv
->ips
.max_delay
= dev_priv
->ips
.fstart
;
5993 if (!ironlake_set_drps(dev_priv
->dev
, dev_priv
->ips
.fstart
))
5997 spin_unlock_irq(&mchdev_lock
);
6001 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable
);
6004 * Tells the intel_ips driver that the i915 driver is now loaded, if
6005 * IPS got loaded first.
6007 * This awkward dance is so that neither module has to depend on the
6008 * other in order for IPS to do the appropriate communication of
6009 * GPU turbo limits to i915.
6012 ips_ping_for_i915_load(void)
6016 link
= symbol_get(ips_link_to_i915_driver
);
6019 symbol_put(ips_link_to_i915_driver
);
6023 void intel_gpu_ips_init(struct drm_i915_private
*dev_priv
)
6025 /* We only register the i915 ips part with intel-ips once everything is
6026 * set up, to avoid intel-ips sneaking in and reading bogus values. */
6027 spin_lock_irq(&mchdev_lock
);
6028 i915_mch_dev
= dev_priv
;
6029 spin_unlock_irq(&mchdev_lock
);
6031 ips_ping_for_i915_load();
6034 void intel_gpu_ips_teardown(void)
6036 spin_lock_irq(&mchdev_lock
);
6037 i915_mch_dev
= NULL
;
6038 spin_unlock_irq(&mchdev_lock
);
6041 static void intel_init_emon(struct drm_device
*dev
)
6043 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6048 /* Disable to program */
6052 /* Program energy weights for various events */
6053 I915_WRITE(SDEW
, 0x15040d00);
6054 I915_WRITE(CSIEW0
, 0x007f0000);
6055 I915_WRITE(CSIEW1
, 0x1e220004);
6056 I915_WRITE(CSIEW2
, 0x04000004);
6058 for (i
= 0; i
< 5; i
++)
6059 I915_WRITE(PEW(i
), 0);
6060 for (i
= 0; i
< 3; i
++)
6061 I915_WRITE(DEW(i
), 0);
6063 /* Program P-state weights to account for frequency power adjustment */
6064 for (i
= 0; i
< 16; i
++) {
6065 u32 pxvidfreq
= I915_READ(PXVFREQ(i
));
6066 unsigned long freq
= intel_pxfreq(pxvidfreq
);
6067 unsigned long vid
= (pxvidfreq
& PXVFREQ_PX_MASK
) >>
6072 val
*= (freq
/ 1000);
6074 val
/= (127*127*900);
6076 DRM_ERROR("bad pxval: %ld\n", val
);
6079 /* Render standby states get 0 weight */
6083 for (i
= 0; i
< 4; i
++) {
6084 u32 val
= (pxw
[i
*4] << 24) | (pxw
[(i
*4)+1] << 16) |
6085 (pxw
[(i
*4)+2] << 8) | (pxw
[(i
*4)+3]);
6086 I915_WRITE(PXW(i
), val
);
6089 /* Adjust magic regs to magic values (more experimental results) */
6090 I915_WRITE(OGW0
, 0);
6091 I915_WRITE(OGW1
, 0);
6092 I915_WRITE(EG0
, 0x00007f00);
6093 I915_WRITE(EG1
, 0x0000000e);
6094 I915_WRITE(EG2
, 0x000e0000);
6095 I915_WRITE(EG3
, 0x68000300);
6096 I915_WRITE(EG4
, 0x42000000);
6097 I915_WRITE(EG5
, 0x00140031);
6101 for (i
= 0; i
< 8; i
++)
6102 I915_WRITE(PXWL(i
), 0);
6104 /* Enable PMON + select events */
6105 I915_WRITE(ECR
, 0x80000019);
6107 lcfuse
= I915_READ(LCFUSE02
);
6109 dev_priv
->ips
.corr
= (lcfuse
& LCFUSE_HIV_MASK
);
6112 void intel_init_gt_powersave(struct drm_device
*dev
)
6114 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6117 * RPM depends on RC6 to save restore the GT HW context, so make RC6 a
6120 if (!i915
.enable_rc6
) {
6121 DRM_INFO("RC6 disabled, disabling runtime PM support\n");
6122 intel_runtime_pm_get(dev_priv
);
6125 if (IS_CHERRYVIEW(dev
))
6126 cherryview_init_gt_powersave(dev
);
6127 else if (IS_VALLEYVIEW(dev
))
6128 valleyview_init_gt_powersave(dev
);
6131 void intel_cleanup_gt_powersave(struct drm_device
*dev
)
6133 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6135 if (IS_CHERRYVIEW(dev
))
6137 else if (IS_VALLEYVIEW(dev
))
6138 valleyview_cleanup_gt_powersave(dev
);
6140 if (!i915
.enable_rc6
)
6141 intel_runtime_pm_put(dev_priv
);
6144 static void gen6_suspend_rps(struct drm_device
*dev
)
6146 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6148 flush_delayed_work(&dev_priv
->rps
.delayed_resume_work
);
6150 gen6_disable_rps_interrupts(dev
);
6154 * intel_suspend_gt_powersave - suspend PM work and helper threads
6157 * We don't want to disable RC6 or other features here, we just want
6158 * to make sure any work we've queued has finished and won't bother
6159 * us while we're suspended.
6161 void intel_suspend_gt_powersave(struct drm_device
*dev
)
6163 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6165 if (INTEL_INFO(dev
)->gen
< 6)
6168 gen6_suspend_rps(dev
);
6170 /* Force GPU to min freq during suspend */
6171 gen6_rps_idle(dev_priv
);
6174 void intel_disable_gt_powersave(struct drm_device
*dev
)
6176 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6178 if (IS_IRONLAKE_M(dev
)) {
6179 ironlake_disable_drps(dev
);
6180 } else if (INTEL_INFO(dev
)->gen
>= 6) {
6181 intel_suspend_gt_powersave(dev
);
6183 mutex_lock(&dev_priv
->rps
.hw_lock
);
6184 if (INTEL_INFO(dev
)->gen
>= 9)
6185 gen9_disable_rps(dev
);
6186 else if (IS_CHERRYVIEW(dev
))
6187 cherryview_disable_rps(dev
);
6188 else if (IS_VALLEYVIEW(dev
))
6189 valleyview_disable_rps(dev
);
6191 gen6_disable_rps(dev
);
6193 dev_priv
->rps
.enabled
= false;
6194 mutex_unlock(&dev_priv
->rps
.hw_lock
);
6198 static void intel_gen6_powersave_work(struct work_struct
*work
)
6200 struct drm_i915_private
*dev_priv
=
6201 container_of(work
, struct drm_i915_private
,
6202 rps
.delayed_resume_work
.work
);
6203 struct drm_device
*dev
= dev_priv
->dev
;
6205 mutex_lock(&dev_priv
->rps
.hw_lock
);
6207 gen6_reset_rps_interrupts(dev
);
6209 if (IS_CHERRYVIEW(dev
)) {
6210 cherryview_enable_rps(dev
);
6211 } else if (IS_VALLEYVIEW(dev
)) {
6212 valleyview_enable_rps(dev
);
6213 } else if (INTEL_INFO(dev
)->gen
>= 9) {
6214 gen9_enable_rc6(dev
);
6215 gen9_enable_rps(dev
);
6216 if (IS_SKYLAKE(dev
) || IS_KABYLAKE(dev
))
6217 __gen6_update_ring_freq(dev
);
6218 } else if (IS_BROADWELL(dev
)) {
6219 gen8_enable_rps(dev
);
6220 __gen6_update_ring_freq(dev
);
6222 gen6_enable_rps(dev
);
6223 __gen6_update_ring_freq(dev
);
6226 WARN_ON(dev_priv
->rps
.max_freq
< dev_priv
->rps
.min_freq
);
6227 WARN_ON(dev_priv
->rps
.idle_freq
> dev_priv
->rps
.max_freq
);
6229 WARN_ON(dev_priv
->rps
.efficient_freq
< dev_priv
->rps
.min_freq
);
6230 WARN_ON(dev_priv
->rps
.efficient_freq
> dev_priv
->rps
.max_freq
);
6232 dev_priv
->rps
.enabled
= true;
6234 gen6_enable_rps_interrupts(dev
);
6236 mutex_unlock(&dev_priv
->rps
.hw_lock
);
6238 intel_runtime_pm_put(dev_priv
);
6241 void intel_enable_gt_powersave(struct drm_device
*dev
)
6243 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6245 /* Powersaving is controlled by the host when inside a VM */
6246 if (intel_vgpu_active(dev
))
6249 if (IS_IRONLAKE_M(dev
)) {
6250 ironlake_enable_drps(dev
);
6251 mutex_lock(&dev
->struct_mutex
);
6252 intel_init_emon(dev
);
6253 mutex_unlock(&dev
->struct_mutex
);
6254 } else if (INTEL_INFO(dev
)->gen
>= 6) {
6256 * PCU communication is slow and this doesn't need to be
6257 * done at any specific time, so do this out of our fast path
6258 * to make resume and init faster.
6260 * We depend on the HW RC6 power context save/restore
6261 * mechanism when entering D3 through runtime PM suspend. So
6262 * disable RPM until RPS/RC6 is properly setup. We can only
6263 * get here via the driver load/system resume/runtime resume
6264 * paths, so the _noresume version is enough (and in case of
6265 * runtime resume it's necessary).
6267 if (schedule_delayed_work(&dev_priv
->rps
.delayed_resume_work
,
6268 round_jiffies_up_relative(HZ
)))
6269 intel_runtime_pm_get_noresume(dev_priv
);
6273 void intel_reset_gt_powersave(struct drm_device
*dev
)
6275 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6277 if (INTEL_INFO(dev
)->gen
< 6)
6280 gen6_suspend_rps(dev
);
6281 dev_priv
->rps
.enabled
= false;
6284 static void ibx_init_clock_gating(struct drm_device
*dev
)
6286 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6289 * On Ibex Peak and Cougar Point, we need to disable clock
6290 * gating for the panel power sequencer or it will fail to
6291 * start up when no ports are active.
6293 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
);
6296 static void g4x_disable_trickle_feed(struct drm_device
*dev
)
6298 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6301 for_each_pipe(dev_priv
, pipe
) {
6302 I915_WRITE(DSPCNTR(pipe
),
6303 I915_READ(DSPCNTR(pipe
)) |
6304 DISPPLANE_TRICKLE_FEED_DISABLE
);
6306 I915_WRITE(DSPSURF(pipe
), I915_READ(DSPSURF(pipe
)));
6307 POSTING_READ(DSPSURF(pipe
));
6311 static void ilk_init_lp_watermarks(struct drm_device
*dev
)
6313 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6315 I915_WRITE(WM3_LP_ILK
, I915_READ(WM3_LP_ILK
) & ~WM1_LP_SR_EN
);
6316 I915_WRITE(WM2_LP_ILK
, I915_READ(WM2_LP_ILK
) & ~WM1_LP_SR_EN
);
6317 I915_WRITE(WM1_LP_ILK
, I915_READ(WM1_LP_ILK
) & ~WM1_LP_SR_EN
);
6320 * Don't touch WM1S_LP_EN here.
6321 * Doing so could cause underruns.
6325 static void ironlake_init_clock_gating(struct drm_device
*dev
)
6327 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6328 uint32_t dspclk_gate
= ILK_VRHUNIT_CLOCK_GATE_DISABLE
;
6332 * WaFbcDisableDpfcClockGating:ilk
6334 dspclk_gate
|= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE
|
6335 ILK_DPFCUNIT_CLOCK_GATE_DISABLE
|
6336 ILK_DPFDUNIT_CLOCK_GATE_ENABLE
;
6338 I915_WRITE(PCH_3DCGDIS0
,
6339 MARIUNIT_CLOCK_GATE_DISABLE
|
6340 SVSMUNIT_CLOCK_GATE_DISABLE
);
6341 I915_WRITE(PCH_3DCGDIS1
,
6342 VFMUNIT_CLOCK_GATE_DISABLE
);
6345 * According to the spec the following bits should be set in
6346 * order to enable memory self-refresh
6347 * The bit 22/21 of 0x42004
6348 * The bit 5 of 0x42020
6349 * The bit 15 of 0x45000
6351 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6352 (I915_READ(ILK_DISPLAY_CHICKEN2
) |
6353 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
));
6354 dspclk_gate
|= ILK_DPARBUNIT_CLOCK_GATE_ENABLE
;
6355 I915_WRITE(DISP_ARB_CTL
,
6356 (I915_READ(DISP_ARB_CTL
) |
6359 ilk_init_lp_watermarks(dev
);
6362 * Based on the document from hardware guys the following bits
6363 * should be set unconditionally in order to enable FBC.
6364 * The bit 22 of 0x42000
6365 * The bit 22 of 0x42004
6366 * The bit 7,8,9 of 0x42020.
6368 if (IS_IRONLAKE_M(dev
)) {
6369 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6370 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
6371 I915_READ(ILK_DISPLAY_CHICKEN1
) |
6373 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6374 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6378 I915_WRITE(ILK_DSPCLK_GATE_D
, dspclk_gate
);
6380 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6381 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6382 ILK_ELPIN_409_SELECT
);
6383 I915_WRITE(_3D_CHICKEN2
,
6384 _3D_CHICKEN2_WM_READ_PIPELINED
<< 16 |
6385 _3D_CHICKEN2_WM_READ_PIPELINED
);
6387 /* WaDisableRenderCachePipelinedFlush:ilk */
6388 I915_WRITE(CACHE_MODE_0
,
6389 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE
));
6391 /* WaDisable_RenderCache_OperationalFlush:ilk */
6392 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6394 g4x_disable_trickle_feed(dev
);
6396 ibx_init_clock_gating(dev
);
6399 static void cpt_init_clock_gating(struct drm_device
*dev
)
6401 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6406 * On Ibex Peak and Cougar Point, we need to disable clock
6407 * gating for the panel power sequencer or it will fail to
6408 * start up when no ports are active.
6410 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
|
6411 PCH_DPLUNIT_CLOCK_GATE_DISABLE
|
6412 PCH_CPUNIT_CLOCK_GATE_DISABLE
);
6413 I915_WRITE(SOUTH_CHICKEN2
, I915_READ(SOUTH_CHICKEN2
) |
6414 DPLS_EDP_PPS_FIX_DIS
);
6415 /* The below fixes the weird display corruption, a few pixels shifted
6416 * downward, on (only) LVDS of some HP laptops with IVY.
6418 for_each_pipe(dev_priv
, pipe
) {
6419 val
= I915_READ(TRANS_CHICKEN2(pipe
));
6420 val
|= TRANS_CHICKEN2_TIMING_OVERRIDE
;
6421 val
&= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED
;
6422 if (dev_priv
->vbt
.fdi_rx_polarity_inverted
)
6423 val
|= TRANS_CHICKEN2_FDI_POLARITY_REVERSED
;
6424 val
&= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK
;
6425 val
&= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER
;
6426 val
&= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH
;
6427 I915_WRITE(TRANS_CHICKEN2(pipe
), val
);
6429 /* WADP0ClockGatingDisable */
6430 for_each_pipe(dev_priv
, pipe
) {
6431 I915_WRITE(TRANS_CHICKEN1(pipe
),
6432 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE
);
6436 static void gen6_check_mch_setup(struct drm_device
*dev
)
6438 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6441 tmp
= I915_READ(MCH_SSKPD
);
6442 if ((tmp
& MCH_SSKPD_WM0_MASK
) != MCH_SSKPD_WM0_VAL
)
6443 DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n",
6447 static void gen6_init_clock_gating(struct drm_device
*dev
)
6449 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6450 uint32_t dspclk_gate
= ILK_VRHUNIT_CLOCK_GATE_DISABLE
;
6452 I915_WRITE(ILK_DSPCLK_GATE_D
, dspclk_gate
);
6454 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6455 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6456 ILK_ELPIN_409_SELECT
);
6458 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
6459 I915_WRITE(_3D_CHICKEN
,
6460 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB
));
6462 /* WaDisable_RenderCache_OperationalFlush:snb */
6463 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6466 * BSpec recoomends 8x4 when MSAA is used,
6467 * however in practice 16x4 seems fastest.
6469 * Note that PS/WM thread counts depend on the WIZ hashing
6470 * disable bit, which we don't touch here, but it's good
6471 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6473 I915_WRITE(GEN6_GT_MODE
,
6474 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6476 ilk_init_lp_watermarks(dev
);
6478 I915_WRITE(CACHE_MODE_0
,
6479 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB
));
6481 I915_WRITE(GEN6_UCGCTL1
,
6482 I915_READ(GEN6_UCGCTL1
) |
6483 GEN6_BLBUNIT_CLOCK_GATE_DISABLE
|
6484 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
6486 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
6487 * gating disable must be set. Failure to set it results in
6488 * flickering pixels due to Z write ordering failures after
6489 * some amount of runtime in the Mesa "fire" demo, and Unigine
6490 * Sanctuary and Tropics, and apparently anything else with
6491 * alpha test or pixel discard.
6493 * According to the spec, bit 11 (RCCUNIT) must also be set,
6494 * but we didn't debug actual testcases to find it out.
6496 * WaDisableRCCUnitClockGating:snb
6497 * WaDisableRCPBUnitClockGating:snb
6499 I915_WRITE(GEN6_UCGCTL2
,
6500 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE
|
6501 GEN6_RCCUNIT_CLOCK_GATE_DISABLE
);
6503 /* WaStripsFansDisableFastClipPerformanceFix:snb */
6504 I915_WRITE(_3D_CHICKEN3
,
6505 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL
));
6509 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
6510 * 3DSTATE_SF number of SF output attributes is more than 16."
6512 I915_WRITE(_3D_CHICKEN3
,
6513 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH
));
6516 * According to the spec the following bits should be
6517 * set in order to enable memory self-refresh and fbc:
6518 * The bit21 and bit22 of 0x42000
6519 * The bit21 and bit22 of 0x42004
6520 * The bit5 and bit7 of 0x42020
6521 * The bit14 of 0x70180
6522 * The bit14 of 0x71180
6524 * WaFbcAsynchFlipDisableFbcQueue:snb
6526 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
6527 I915_READ(ILK_DISPLAY_CHICKEN1
) |
6528 ILK_FBCQ_DIS
| ILK_PABSTRETCH_DIS
);
6529 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
6530 I915_READ(ILK_DISPLAY_CHICKEN2
) |
6531 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
);
6532 I915_WRITE(ILK_DSPCLK_GATE_D
,
6533 I915_READ(ILK_DSPCLK_GATE_D
) |
6534 ILK_DPARBUNIT_CLOCK_GATE_ENABLE
|
6535 ILK_DPFDUNIT_CLOCK_GATE_ENABLE
);
6537 g4x_disable_trickle_feed(dev
);
6539 cpt_init_clock_gating(dev
);
6541 gen6_check_mch_setup(dev
);
6544 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private
*dev_priv
)
6546 uint32_t reg
= I915_READ(GEN7_FF_THREAD_MODE
);
6549 * WaVSThreadDispatchOverride:ivb,vlv
6551 * This actually overrides the dispatch
6552 * mode for all thread types.
6554 reg
&= ~GEN7_FF_SCHED_MASK
;
6555 reg
|= GEN7_FF_TS_SCHED_HW
;
6556 reg
|= GEN7_FF_VS_SCHED_HW
;
6557 reg
|= GEN7_FF_DS_SCHED_HW
;
6559 I915_WRITE(GEN7_FF_THREAD_MODE
, reg
);
6562 static void lpt_init_clock_gating(struct drm_device
*dev
)
6564 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6567 * TODO: this bit should only be enabled when really needed, then
6568 * disabled when not needed anymore in order to save power.
6570 if (HAS_PCH_LPT_LP(dev
))
6571 I915_WRITE(SOUTH_DSPCLK_GATE_D
,
6572 I915_READ(SOUTH_DSPCLK_GATE_D
) |
6573 PCH_LP_PARTITION_LEVEL_DISABLE
);
6575 /* WADPOClockGatingDisable:hsw */
6576 I915_WRITE(TRANS_CHICKEN1(PIPE_A
),
6577 I915_READ(TRANS_CHICKEN1(PIPE_A
)) |
6578 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE
);
6581 static void lpt_suspend_hw(struct drm_device
*dev
)
6583 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6585 if (HAS_PCH_LPT_LP(dev
)) {
6586 uint32_t val
= I915_READ(SOUTH_DSPCLK_GATE_D
);
6588 val
&= ~PCH_LP_PARTITION_LEVEL_DISABLE
;
6589 I915_WRITE(SOUTH_DSPCLK_GATE_D
, val
);
6593 static void broadwell_init_clock_gating(struct drm_device
*dev
)
6595 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6599 ilk_init_lp_watermarks(dev
);
6601 /* WaSwitchSolVfFArbitrationPriority:bdw */
6602 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) | HSW_ECOCHK_ARB_PRIO_SOL
);
6604 /* WaPsrDPAMaskVBlankInSRD:bdw */
6605 I915_WRITE(CHICKEN_PAR1_1
,
6606 I915_READ(CHICKEN_PAR1_1
) | DPA_MASK_VBLANK_SRD
);
6608 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
6609 for_each_pipe(dev_priv
, pipe
) {
6610 I915_WRITE(CHICKEN_PIPESL_1(pipe
),
6611 I915_READ(CHICKEN_PIPESL_1(pipe
)) |
6612 BDW_DPRS_MASK_VBLANK_SRD
);
6615 /* WaVSRefCountFullforceMissDisable:bdw */
6616 /* WaDSRefCountFullforceMissDisable:bdw */
6617 I915_WRITE(GEN7_FF_THREAD_MODE
,
6618 I915_READ(GEN7_FF_THREAD_MODE
) &
6619 ~(GEN8_FF_DS_REF_CNT_FFME
| GEN7_FF_VS_REF_CNT_FFME
));
6621 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
6622 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE
));
6624 /* WaDisableSDEUnitClockGating:bdw */
6625 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
6626 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
6629 * WaProgramL3SqcReg1Default:bdw
6630 * WaTempDisableDOPClkGating:bdw
6632 misccpctl
= I915_READ(GEN7_MISCCPCTL
);
6633 I915_WRITE(GEN7_MISCCPCTL
, misccpctl
& ~GEN7_DOP_CLOCK_GATE_ENABLE
);
6634 I915_WRITE(GEN8_L3SQCREG1
, BDW_WA_L3SQCREG1_DEFAULT
);
6635 I915_WRITE(GEN7_MISCCPCTL
, misccpctl
);
6638 * WaGttCachingOffByDefault:bdw
6639 * GTT cache may not work with big pages, so if those
6640 * are ever enabled GTT cache may need to be disabled.
6642 I915_WRITE(HSW_GTT_CACHE_EN
, GTT_CACHE_EN_ALL
);
6644 lpt_init_clock_gating(dev
);
6647 static void haswell_init_clock_gating(struct drm_device
*dev
)
6649 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6651 ilk_init_lp_watermarks(dev
);
6653 /* L3 caching of data atomics doesn't work -- disable it. */
6654 I915_WRITE(HSW_SCRATCH1
, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE
);
6655 I915_WRITE(HSW_ROW_CHICKEN3
,
6656 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE
));
6658 /* This is required by WaCatErrorRejectionIssue:hsw */
6659 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
6660 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
6661 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
6663 /* WaVSRefCountFullforceMissDisable:hsw */
6664 I915_WRITE(GEN7_FF_THREAD_MODE
,
6665 I915_READ(GEN7_FF_THREAD_MODE
) & ~GEN7_FF_VS_REF_CNT_FFME
);
6667 /* WaDisable_RenderCache_OperationalFlush:hsw */
6668 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6670 /* enable HiZ Raw Stall Optimization */
6671 I915_WRITE(CACHE_MODE_0_GEN7
,
6672 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE
));
6674 /* WaDisable4x2SubspanOptimization:hsw */
6675 I915_WRITE(CACHE_MODE_1
,
6676 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
6679 * BSpec recommends 8x4 when MSAA is used,
6680 * however in practice 16x4 seems fastest.
6682 * Note that PS/WM thread counts depend on the WIZ hashing
6683 * disable bit, which we don't touch here, but it's good
6684 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6686 I915_WRITE(GEN7_GT_MODE
,
6687 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6689 /* WaSampleCChickenBitEnable:hsw */
6690 I915_WRITE(HALF_SLICE_CHICKEN3
,
6691 _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE
));
6693 /* WaSwitchSolVfFArbitrationPriority:hsw */
6694 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) | HSW_ECOCHK_ARB_PRIO_SOL
);
6696 /* WaRsPkgCStateDisplayPMReq:hsw */
6697 I915_WRITE(CHICKEN_PAR1_1
,
6698 I915_READ(CHICKEN_PAR1_1
) | FORCE_ARB_IDLE_PLANES
);
6700 lpt_init_clock_gating(dev
);
6703 static void ivybridge_init_clock_gating(struct drm_device
*dev
)
6705 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6708 ilk_init_lp_watermarks(dev
);
6710 I915_WRITE(ILK_DSPCLK_GATE_D
, ILK_VRHUNIT_CLOCK_GATE_DISABLE
);
6712 /* WaDisableEarlyCull:ivb */
6713 I915_WRITE(_3D_CHICKEN3
,
6714 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL
));
6716 /* WaDisableBackToBackFlipFix:ivb */
6717 I915_WRITE(IVB_CHICKEN3
,
6718 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
6719 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
6721 /* WaDisablePSDDualDispatchEnable:ivb */
6722 if (IS_IVB_GT1(dev
))
6723 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
6724 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
));
6726 /* WaDisable_RenderCache_OperationalFlush:ivb */
6727 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6729 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6730 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1
,
6731 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
6733 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6734 I915_WRITE(GEN7_L3CNTLREG1
,
6735 GEN7_WA_FOR_GEN7_L3_CONTROL
);
6736 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER
,
6737 GEN7_WA_L3_CHICKEN_MODE
);
6738 if (IS_IVB_GT1(dev
))
6739 I915_WRITE(GEN7_ROW_CHICKEN2
,
6740 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6742 /* must write both registers */
6743 I915_WRITE(GEN7_ROW_CHICKEN2
,
6744 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6745 I915_WRITE(GEN7_ROW_CHICKEN2_GT2
,
6746 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6749 /* WaForceL3Serialization:ivb */
6750 I915_WRITE(GEN7_L3SQCREG4
, I915_READ(GEN7_L3SQCREG4
) &
6751 ~L3SQ_URB_READ_CAM_MATCH_DISABLE
);
6754 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
6755 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
6757 I915_WRITE(GEN6_UCGCTL2
,
6758 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
6760 /* This is required by WaCatErrorRejectionIssue:ivb */
6761 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
6762 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
6763 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
6765 g4x_disable_trickle_feed(dev
);
6767 gen7_setup_fixed_func_scheduler(dev_priv
);
6769 if (0) { /* causes HiZ corruption on ivb:gt1 */
6770 /* enable HiZ Raw Stall Optimization */
6771 I915_WRITE(CACHE_MODE_0_GEN7
,
6772 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE
));
6775 /* WaDisable4x2SubspanOptimization:ivb */
6776 I915_WRITE(CACHE_MODE_1
,
6777 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
6780 * BSpec recommends 8x4 when MSAA is used,
6781 * however in practice 16x4 seems fastest.
6783 * Note that PS/WM thread counts depend on the WIZ hashing
6784 * disable bit, which we don't touch here, but it's good
6785 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6787 I915_WRITE(GEN7_GT_MODE
,
6788 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6790 snpcr
= I915_READ(GEN6_MBCUNIT_SNPCR
);
6791 snpcr
&= ~GEN6_MBC_SNPCR_MASK
;
6792 snpcr
|= GEN6_MBC_SNPCR_MED
;
6793 I915_WRITE(GEN6_MBCUNIT_SNPCR
, snpcr
);
6795 if (!HAS_PCH_NOP(dev
))
6796 cpt_init_clock_gating(dev
);
6798 gen6_check_mch_setup(dev
);
6801 static void vlv_init_display_clock_gating(struct drm_i915_private
*dev_priv
)
6803 I915_WRITE(DSPCLK_GATE_D
, VRHUNIT_CLOCK_GATE_DISABLE
);
6806 * Disable trickle feed and enable pnd deadline calculation
6808 I915_WRITE(MI_ARB_VLV
, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
);
6809 I915_WRITE(CBR1_VLV
, 0);
6812 static void valleyview_init_clock_gating(struct drm_device
*dev
)
6814 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6816 vlv_init_display_clock_gating(dev_priv
);
6818 /* WaDisableEarlyCull:vlv */
6819 I915_WRITE(_3D_CHICKEN3
,
6820 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL
));
6822 /* WaDisableBackToBackFlipFix:vlv */
6823 I915_WRITE(IVB_CHICKEN3
,
6824 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
6825 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
6827 /* WaPsdDispatchEnable:vlv */
6828 /* WaDisablePSDDualDispatchEnable:vlv */
6829 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
6830 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP
|
6831 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
));
6833 /* WaDisable_RenderCache_OperationalFlush:vlv */
6834 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6836 /* WaForceL3Serialization:vlv */
6837 I915_WRITE(GEN7_L3SQCREG4
, I915_READ(GEN7_L3SQCREG4
) &
6838 ~L3SQ_URB_READ_CAM_MATCH_DISABLE
);
6840 /* WaDisableDopClockGating:vlv */
6841 I915_WRITE(GEN7_ROW_CHICKEN2
,
6842 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
6844 /* This is required by WaCatErrorRejectionIssue:vlv */
6845 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
6846 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
6847 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
6849 gen7_setup_fixed_func_scheduler(dev_priv
);
6852 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
6853 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
6855 I915_WRITE(GEN6_UCGCTL2
,
6856 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
6858 /* WaDisableL3Bank2xClockGate:vlv
6859 * Disabling L3 clock gating- MMIO 940c[25] = 1
6860 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
6861 I915_WRITE(GEN7_UCGCTL4
,
6862 I915_READ(GEN7_UCGCTL4
) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE
);
6865 * BSpec says this must be set, even though
6866 * WaDisable4x2SubspanOptimization isn't listed for VLV.
6868 I915_WRITE(CACHE_MODE_1
,
6869 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
6872 * BSpec recommends 8x4 when MSAA is used,
6873 * however in practice 16x4 seems fastest.
6875 * Note that PS/WM thread counts depend on the WIZ hashing
6876 * disable bit, which we don't touch here, but it's good
6877 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
6879 I915_WRITE(GEN7_GT_MODE
,
6880 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
));
6883 * WaIncreaseL3CreditsForVLVB0:vlv
6884 * This is the hardware default actually.
6886 I915_WRITE(GEN7_L3SQCREG1
, VLV_B0_WA_L3SQCREG1_VALUE
);
6889 * WaDisableVLVClockGating_VBIIssue:vlv
6890 * Disable clock gating on th GCFG unit to prevent a delay
6891 * in the reporting of vblank events.
6893 I915_WRITE(VLV_GUNIT_CLOCK_GATE
, GCFG_DIS
);
6896 static void cherryview_init_clock_gating(struct drm_device
*dev
)
6898 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6900 vlv_init_display_clock_gating(dev_priv
);
6902 /* WaVSRefCountFullforceMissDisable:chv */
6903 /* WaDSRefCountFullforceMissDisable:chv */
6904 I915_WRITE(GEN7_FF_THREAD_MODE
,
6905 I915_READ(GEN7_FF_THREAD_MODE
) &
6906 ~(GEN8_FF_DS_REF_CNT_FFME
| GEN7_FF_VS_REF_CNT_FFME
));
6908 /* WaDisableSemaphoreAndSyncFlipWait:chv */
6909 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
6910 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE
));
6912 /* WaDisableCSUnitClockGating:chv */
6913 I915_WRITE(GEN6_UCGCTL1
, I915_READ(GEN6_UCGCTL1
) |
6914 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
6916 /* WaDisableSDEUnitClockGating:chv */
6917 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
6918 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
6921 * GTT cache may not work with big pages, so if those
6922 * are ever enabled GTT cache may need to be disabled.
6924 I915_WRITE(HSW_GTT_CACHE_EN
, GTT_CACHE_EN_ALL
);
6927 static void g4x_init_clock_gating(struct drm_device
*dev
)
6929 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6930 uint32_t dspclk_gate
;
6932 I915_WRITE(RENCLK_GATE_D1
, 0);
6933 I915_WRITE(RENCLK_GATE_D2
, VF_UNIT_CLOCK_GATE_DISABLE
|
6934 GS_UNIT_CLOCK_GATE_DISABLE
|
6935 CL_UNIT_CLOCK_GATE_DISABLE
);
6936 I915_WRITE(RAMCLK_GATE_D
, 0);
6937 dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
|
6938 OVRUNIT_CLOCK_GATE_DISABLE
|
6939 OVCUNIT_CLOCK_GATE_DISABLE
;
6941 dspclk_gate
|= DSSUNIT_CLOCK_GATE_DISABLE
;
6942 I915_WRITE(DSPCLK_GATE_D
, dspclk_gate
);
6944 /* WaDisableRenderCachePipelinedFlush */
6945 I915_WRITE(CACHE_MODE_0
,
6946 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE
));
6948 /* WaDisable_RenderCache_OperationalFlush:g4x */
6949 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6951 g4x_disable_trickle_feed(dev
);
6954 static void crestline_init_clock_gating(struct drm_device
*dev
)
6956 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6958 I915_WRITE(RENCLK_GATE_D1
, I965_RCC_CLOCK_GATE_DISABLE
);
6959 I915_WRITE(RENCLK_GATE_D2
, 0);
6960 I915_WRITE(DSPCLK_GATE_D
, 0);
6961 I915_WRITE(RAMCLK_GATE_D
, 0);
6962 I915_WRITE16(DEUC
, 0);
6963 I915_WRITE(MI_ARB_STATE
,
6964 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
6966 /* WaDisable_RenderCache_OperationalFlush:gen4 */
6967 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6970 static void broadwater_init_clock_gating(struct drm_device
*dev
)
6972 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6974 I915_WRITE(RENCLK_GATE_D1
, I965_RCZ_CLOCK_GATE_DISABLE
|
6975 I965_RCC_CLOCK_GATE_DISABLE
|
6976 I965_RCPB_CLOCK_GATE_DISABLE
|
6977 I965_ISC_CLOCK_GATE_DISABLE
|
6978 I965_FBC_CLOCK_GATE_DISABLE
);
6979 I915_WRITE(RENCLK_GATE_D2
, 0);
6980 I915_WRITE(MI_ARB_STATE
,
6981 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
6983 /* WaDisable_RenderCache_OperationalFlush:gen4 */
6984 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
6987 static void gen3_init_clock_gating(struct drm_device
*dev
)
6989 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6990 u32 dstate
= I915_READ(D_STATE
);
6992 dstate
|= DSTATE_PLL_D3_OFF
| DSTATE_GFX_CLOCK_GATING
|
6993 DSTATE_DOT_CLOCK_GATING
;
6994 I915_WRITE(D_STATE
, dstate
);
6996 if (IS_PINEVIEW(dev
))
6997 I915_WRITE(ECOSKPD
, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY
));
6999 /* IIR "flip pending" means done if this bit is set */
7000 I915_WRITE(ECOSKPD
, _MASKED_BIT_DISABLE(ECO_FLIP_DONE
));
7002 /* interrupts should cause a wake up from C3 */
7003 I915_WRITE(INSTPM
, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN
));
7005 /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
7006 I915_WRITE(MI_ARB_STATE
, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE
));
7008 I915_WRITE(MI_ARB_STATE
,
7009 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
7012 static void i85x_init_clock_gating(struct drm_device
*dev
)
7014 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7016 I915_WRITE(RENCLK_GATE_D1
, SV_CLOCK_GATE_DISABLE
);
7018 /* interrupts should cause a wake up from C3 */
7019 I915_WRITE(MI_STATE
, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN
) |
7020 _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE
));
7022 I915_WRITE(MEM_MODE
,
7023 _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE
));
7026 static void i830_init_clock_gating(struct drm_device
*dev
)
7028 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7030 I915_WRITE(DSPCLK_GATE_D
, OVRUNIT_CLOCK_GATE_DISABLE
);
7032 I915_WRITE(MEM_MODE
,
7033 _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE
) |
7034 _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE
));
7037 void intel_init_clock_gating(struct drm_device
*dev
)
7039 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7041 if (dev_priv
->display
.init_clock_gating
)
7042 dev_priv
->display
.init_clock_gating(dev
);
7045 void intel_suspend_hw(struct drm_device
*dev
)
7047 if (HAS_PCH_LPT(dev
))
7048 lpt_suspend_hw(dev
);
7051 /* Set up chip specific power management-related functions */
7052 void intel_init_pm(struct drm_device
*dev
)
7054 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7056 intel_fbc_init(dev_priv
);
7059 if (IS_PINEVIEW(dev
))
7060 i915_pineview_get_mem_freq(dev
);
7061 else if (IS_GEN5(dev
))
7062 i915_ironlake_get_mem_freq(dev
);
7064 /* For FIFO watermark updates */
7065 if (INTEL_INFO(dev
)->gen
>= 9) {
7066 skl_setup_wm_latency(dev
);
7068 if (IS_BROXTON(dev
))
7069 dev_priv
->display
.init_clock_gating
=
7070 bxt_init_clock_gating
;
7071 dev_priv
->display
.update_wm
= skl_update_wm
;
7072 } else if (HAS_PCH_SPLIT(dev
)) {
7073 ilk_setup_wm_latency(dev
);
7075 if ((IS_GEN5(dev
) && dev_priv
->wm
.pri_latency
[1] &&
7076 dev_priv
->wm
.spr_latency
[1] && dev_priv
->wm
.cur_latency
[1]) ||
7077 (!IS_GEN5(dev
) && dev_priv
->wm
.pri_latency
[0] &&
7078 dev_priv
->wm
.spr_latency
[0] && dev_priv
->wm
.cur_latency
[0])) {
7079 dev_priv
->display
.update_wm
= ilk_update_wm
;
7080 dev_priv
->display
.compute_pipe_wm
= ilk_compute_pipe_wm
;
7081 dev_priv
->display
.program_watermarks
= ilk_program_watermarks
;
7083 DRM_DEBUG_KMS("Failed to read display plane latency. "
7088 dev_priv
->display
.init_clock_gating
= ironlake_init_clock_gating
;
7089 else if (IS_GEN6(dev
))
7090 dev_priv
->display
.init_clock_gating
= gen6_init_clock_gating
;
7091 else if (IS_IVYBRIDGE(dev
))
7092 dev_priv
->display
.init_clock_gating
= ivybridge_init_clock_gating
;
7093 else if (IS_HASWELL(dev
))
7094 dev_priv
->display
.init_clock_gating
= haswell_init_clock_gating
;
7095 else if (INTEL_INFO(dev
)->gen
== 8)
7096 dev_priv
->display
.init_clock_gating
= broadwell_init_clock_gating
;
7097 } else if (IS_CHERRYVIEW(dev
)) {
7098 vlv_setup_wm_latency(dev
);
7100 dev_priv
->display
.update_wm
= vlv_update_wm
;
7101 dev_priv
->display
.init_clock_gating
=
7102 cherryview_init_clock_gating
;
7103 } else if (IS_VALLEYVIEW(dev
)) {
7104 vlv_setup_wm_latency(dev
);
7106 dev_priv
->display
.update_wm
= vlv_update_wm
;
7107 dev_priv
->display
.init_clock_gating
=
7108 valleyview_init_clock_gating
;
7109 } else if (IS_PINEVIEW(dev
)) {
7110 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev
),
7113 dev_priv
->mem_freq
)) {
7114 DRM_INFO("failed to find known CxSR latency "
7115 "(found ddr%s fsb freq %d, mem freq %d), "
7117 (dev_priv
->is_ddr3
== 1) ? "3" : "2",
7118 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
7119 /* Disable CxSR and never update its watermark again */
7120 intel_set_memory_cxsr(dev_priv
, false);
7121 dev_priv
->display
.update_wm
= NULL
;
7123 dev_priv
->display
.update_wm
= pineview_update_wm
;
7124 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
7125 } else if (IS_G4X(dev
)) {
7126 dev_priv
->display
.update_wm
= g4x_update_wm
;
7127 dev_priv
->display
.init_clock_gating
= g4x_init_clock_gating
;
7128 } else if (IS_GEN4(dev
)) {
7129 dev_priv
->display
.update_wm
= i965_update_wm
;
7130 if (IS_CRESTLINE(dev
))
7131 dev_priv
->display
.init_clock_gating
= crestline_init_clock_gating
;
7132 else if (IS_BROADWATER(dev
))
7133 dev_priv
->display
.init_clock_gating
= broadwater_init_clock_gating
;
7134 } else if (IS_GEN3(dev
)) {
7135 dev_priv
->display
.update_wm
= i9xx_update_wm
;
7136 dev_priv
->display
.get_fifo_size
= i9xx_get_fifo_size
;
7137 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
7138 } else if (IS_GEN2(dev
)) {
7139 if (INTEL_INFO(dev
)->num_pipes
== 1) {
7140 dev_priv
->display
.update_wm
= i845_update_wm
;
7141 dev_priv
->display
.get_fifo_size
= i845_get_fifo_size
;
7143 dev_priv
->display
.update_wm
= i9xx_update_wm
;
7144 dev_priv
->display
.get_fifo_size
= i830_get_fifo_size
;
7147 if (IS_I85X(dev
) || IS_I865G(dev
))
7148 dev_priv
->display
.init_clock_gating
= i85x_init_clock_gating
;
7150 dev_priv
->display
.init_clock_gating
= i830_init_clock_gating
;
7152 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
7156 int sandybridge_pcode_read(struct drm_i915_private
*dev_priv
, u32 mbox
, u32
*val
)
7158 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
7160 if (I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) {
7161 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
7165 I915_WRITE(GEN6_PCODE_DATA
, *val
);
7166 I915_WRITE(GEN6_PCODE_DATA1
, 0);
7167 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
| mbox
);
7169 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
7171 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox
);
7175 *val
= I915_READ(GEN6_PCODE_DATA
);
7176 I915_WRITE(GEN6_PCODE_DATA
, 0);
7181 int sandybridge_pcode_write(struct drm_i915_private
*dev_priv
, u32 mbox
, u32 val
)
7183 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
7185 if (I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) {
7186 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
7190 I915_WRITE(GEN6_PCODE_DATA
, val
);
7191 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
| mbox
);
7193 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
7195 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox
);
7199 I915_WRITE(GEN6_PCODE_DATA
, 0);
7204 static int vlv_gpu_freq_div(unsigned int czclk_freq
)
7206 switch (czclk_freq
) {
7221 static int byt_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
7223 int div
, czclk_freq
= DIV_ROUND_CLOSEST(dev_priv
->czclk_freq
, 1000);
7225 div
= vlv_gpu_freq_div(czclk_freq
);
7229 return DIV_ROUND_CLOSEST(czclk_freq
* (val
+ 6 - 0xbd), div
);
7232 static int byt_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
7234 int mul
, czclk_freq
= DIV_ROUND_CLOSEST(dev_priv
->czclk_freq
, 1000);
7236 mul
= vlv_gpu_freq_div(czclk_freq
);
7240 return DIV_ROUND_CLOSEST(mul
* val
, czclk_freq
) + 0xbd - 6;
7243 static int chv_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
7245 int div
, czclk_freq
= DIV_ROUND_CLOSEST(dev_priv
->czclk_freq
, 1000);
7247 div
= vlv_gpu_freq_div(czclk_freq
);
7252 return DIV_ROUND_CLOSEST(czclk_freq
* val
, 2 * div
) / 2;
7255 static int chv_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
7257 int mul
, czclk_freq
= DIV_ROUND_CLOSEST(dev_priv
->czclk_freq
, 1000);
7259 mul
= vlv_gpu_freq_div(czclk_freq
);
7264 /* CHV needs even values */
7265 return DIV_ROUND_CLOSEST(val
* 2 * mul
, czclk_freq
) * 2;
7268 int intel_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
7270 if (IS_GEN9(dev_priv
->dev
))
7271 return DIV_ROUND_CLOSEST(val
* GT_FREQUENCY_MULTIPLIER
,
7273 else if (IS_CHERRYVIEW(dev_priv
->dev
))
7274 return chv_gpu_freq(dev_priv
, val
);
7275 else if (IS_VALLEYVIEW(dev_priv
->dev
))
7276 return byt_gpu_freq(dev_priv
, val
);
7278 return val
* GT_FREQUENCY_MULTIPLIER
;
7281 int intel_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
7283 if (IS_GEN9(dev_priv
->dev
))
7284 return DIV_ROUND_CLOSEST(val
* GEN9_FREQ_SCALER
,
7285 GT_FREQUENCY_MULTIPLIER
);
7286 else if (IS_CHERRYVIEW(dev_priv
->dev
))
7287 return chv_freq_opcode(dev_priv
, val
);
7288 else if (IS_VALLEYVIEW(dev_priv
->dev
))
7289 return byt_freq_opcode(dev_priv
, val
);
7291 return DIV_ROUND_CLOSEST(val
, GT_FREQUENCY_MULTIPLIER
);
7294 struct request_boost
{
7295 struct work_struct work
;
7296 struct drm_i915_gem_request
*req
;
7299 static void __intel_rps_boost_work(struct work_struct
*work
)
7301 struct request_boost
*boost
= container_of(work
, struct request_boost
, work
);
7302 struct drm_i915_gem_request
*req
= boost
->req
;
7304 if (!i915_gem_request_completed(req
, true))
7305 gen6_rps_boost(to_i915(req
->ring
->dev
), NULL
,
7306 req
->emitted_jiffies
);
7308 i915_gem_request_unreference__unlocked(req
);
7312 void intel_queue_rps_boost_for_request(struct drm_device
*dev
,
7313 struct drm_i915_gem_request
*req
)
7315 struct request_boost
*boost
;
7317 if (req
== NULL
|| INTEL_INFO(dev
)->gen
< 6)
7320 if (i915_gem_request_completed(req
, true))
7323 boost
= kmalloc(sizeof(*boost
), GFP_ATOMIC
);
7327 i915_gem_request_reference(req
);
7330 INIT_WORK(&boost
->work
, __intel_rps_boost_work
);
7331 queue_work(to_i915(dev
)->wq
, &boost
->work
);
7334 void intel_pm_setup(struct drm_device
*dev
)
7336 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
7338 mutex_init(&dev_priv
->rps
.hw_lock
);
7339 spin_lock_init(&dev_priv
->rps
.client_lock
);
7341 INIT_DELAYED_WORK(&dev_priv
->rps
.delayed_resume_work
,
7342 intel_gen6_powersave_work
);
7343 INIT_LIST_HEAD(&dev_priv
->rps
.clients
);
7344 INIT_LIST_HEAD(&dev_priv
->rps
.semaphores
.link
);
7345 INIT_LIST_HEAD(&dev_priv
->rps
.mmioflips
.link
);
7347 dev_priv
->pm
.suspended
= false;
7348 atomic_set(&dev_priv
->pm
.wakeref_count
, 0);
7349 atomic_set(&dev_priv
->pm
.atomic_seq
, 0);