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>
34 #define FORCEWAKE_ACK_TIMEOUT_MS 2
36 /* FBC, or Frame Buffer Compression, is a technique employed to compress the
37 * framebuffer contents in-memory, aiming at reducing the required bandwidth
38 * during in-memory transfers and, therefore, reduce the power packet.
40 * The benefits of FBC are mostly visible with solid backgrounds and
41 * variation-less patterns.
43 * FBC-related functionality can be enabled by the means of the
44 * i915.i915_enable_fbc parameter
47 static void i8xx_disable_fbc(struct drm_device
*dev
)
49 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
52 /* Disable compression */
53 fbc_ctl
= I915_READ(FBC_CONTROL
);
54 if ((fbc_ctl
& FBC_CTL_EN
) == 0)
57 fbc_ctl
&= ~FBC_CTL_EN
;
58 I915_WRITE(FBC_CONTROL
, fbc_ctl
);
60 /* Wait for compressing bit to clear */
61 if (wait_for((I915_READ(FBC_STATUS
) & FBC_STAT_COMPRESSING
) == 0, 10)) {
62 DRM_DEBUG_KMS("FBC idle timed out\n");
66 DRM_DEBUG_KMS("disabled FBC\n");
69 static void i8xx_enable_fbc(struct drm_crtc
*crtc
, unsigned long interval
)
71 struct drm_device
*dev
= crtc
->dev
;
72 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
73 struct drm_framebuffer
*fb
= crtc
->fb
;
74 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
75 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
76 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
79 u32 fbc_ctl
, fbc_ctl2
;
81 cfb_pitch
= dev_priv
->cfb_size
/ FBC_LL_SIZE
;
82 if (fb
->pitches
[0] < cfb_pitch
)
83 cfb_pitch
= fb
->pitches
[0];
85 /* FBC_CTL wants 64B units */
86 cfb_pitch
= (cfb_pitch
/ 64) - 1;
87 plane
= intel_crtc
->plane
== 0 ? FBC_CTL_PLANEA
: FBC_CTL_PLANEB
;
90 for (i
= 0; i
< (FBC_LL_SIZE
/ 32) + 1; i
++)
91 I915_WRITE(FBC_TAG
+ (i
* 4), 0);
94 fbc_ctl2
= FBC_CTL_FENCE_DBL
| FBC_CTL_IDLE_IMM
| FBC_CTL_CPU_FENCE
;
96 I915_WRITE(FBC_CONTROL2
, fbc_ctl2
);
97 I915_WRITE(FBC_FENCE_OFF
, crtc
->y
);
100 fbc_ctl
= FBC_CTL_EN
| FBC_CTL_PERIODIC
;
102 fbc_ctl
|= FBC_CTL_C3_IDLE
; /* 945 needs special SR handling */
103 fbc_ctl
|= (cfb_pitch
& 0xff) << FBC_CTL_STRIDE_SHIFT
;
104 fbc_ctl
|= (interval
& 0x2fff) << FBC_CTL_INTERVAL_SHIFT
;
105 fbc_ctl
|= obj
->fence_reg
;
106 I915_WRITE(FBC_CONTROL
, fbc_ctl
);
108 DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %d, ",
109 cfb_pitch
, crtc
->y
, intel_crtc
->plane
);
112 static bool i8xx_fbc_enabled(struct drm_device
*dev
)
114 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
116 return I915_READ(FBC_CONTROL
) & FBC_CTL_EN
;
119 static void g4x_enable_fbc(struct drm_crtc
*crtc
, unsigned long interval
)
121 struct drm_device
*dev
= crtc
->dev
;
122 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
123 struct drm_framebuffer
*fb
= crtc
->fb
;
124 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
125 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
126 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
127 int plane
= intel_crtc
->plane
== 0 ? DPFC_CTL_PLANEA
: DPFC_CTL_PLANEB
;
128 unsigned long stall_watermark
= 200;
131 dpfc_ctl
= plane
| DPFC_SR_EN
| DPFC_CTL_LIMIT_1X
;
132 dpfc_ctl
|= DPFC_CTL_FENCE_EN
| obj
->fence_reg
;
133 I915_WRITE(DPFC_CHICKEN
, DPFC_HT_MODIFY
);
135 I915_WRITE(DPFC_RECOMP_CTL
, DPFC_RECOMP_STALL_EN
|
136 (stall_watermark
<< DPFC_RECOMP_STALL_WM_SHIFT
) |
137 (interval
<< DPFC_RECOMP_TIMER_COUNT_SHIFT
));
138 I915_WRITE(DPFC_FENCE_YOFF
, crtc
->y
);
141 I915_WRITE(DPFC_CONTROL
, I915_READ(DPFC_CONTROL
) | DPFC_CTL_EN
);
143 DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc
->plane
);
146 static void g4x_disable_fbc(struct drm_device
*dev
)
148 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
151 /* Disable compression */
152 dpfc_ctl
= I915_READ(DPFC_CONTROL
);
153 if (dpfc_ctl
& DPFC_CTL_EN
) {
154 dpfc_ctl
&= ~DPFC_CTL_EN
;
155 I915_WRITE(DPFC_CONTROL
, dpfc_ctl
);
157 DRM_DEBUG_KMS("disabled FBC\n");
161 static bool g4x_fbc_enabled(struct drm_device
*dev
)
163 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
165 return I915_READ(DPFC_CONTROL
) & DPFC_CTL_EN
;
168 static void sandybridge_blit_fbc_update(struct drm_device
*dev
)
170 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
173 /* Make sure blitter notifies FBC of writes */
174 gen6_gt_force_wake_get(dev_priv
);
175 blt_ecoskpd
= I915_READ(GEN6_BLITTER_ECOSKPD
);
176 blt_ecoskpd
|= GEN6_BLITTER_FBC_NOTIFY
<<
177 GEN6_BLITTER_LOCK_SHIFT
;
178 I915_WRITE(GEN6_BLITTER_ECOSKPD
, blt_ecoskpd
);
179 blt_ecoskpd
|= GEN6_BLITTER_FBC_NOTIFY
;
180 I915_WRITE(GEN6_BLITTER_ECOSKPD
, blt_ecoskpd
);
181 blt_ecoskpd
&= ~(GEN6_BLITTER_FBC_NOTIFY
<<
182 GEN6_BLITTER_LOCK_SHIFT
);
183 I915_WRITE(GEN6_BLITTER_ECOSKPD
, blt_ecoskpd
);
184 POSTING_READ(GEN6_BLITTER_ECOSKPD
);
185 gen6_gt_force_wake_put(dev_priv
);
188 static void ironlake_enable_fbc(struct drm_crtc
*crtc
, unsigned long interval
)
190 struct drm_device
*dev
= crtc
->dev
;
191 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
192 struct drm_framebuffer
*fb
= crtc
->fb
;
193 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
194 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
195 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
196 int plane
= intel_crtc
->plane
== 0 ? DPFC_CTL_PLANEA
: DPFC_CTL_PLANEB
;
197 unsigned long stall_watermark
= 200;
200 dpfc_ctl
= I915_READ(ILK_DPFC_CONTROL
);
201 dpfc_ctl
&= DPFC_RESERVED
;
202 dpfc_ctl
|= (plane
| DPFC_CTL_LIMIT_1X
);
203 /* Set persistent mode for front-buffer rendering, ala X. */
204 dpfc_ctl
|= DPFC_CTL_PERSISTENT_MODE
;
205 dpfc_ctl
|= (DPFC_CTL_FENCE_EN
| obj
->fence_reg
);
206 I915_WRITE(ILK_DPFC_CHICKEN
, DPFC_HT_MODIFY
);
208 I915_WRITE(ILK_DPFC_RECOMP_CTL
, DPFC_RECOMP_STALL_EN
|
209 (stall_watermark
<< DPFC_RECOMP_STALL_WM_SHIFT
) |
210 (interval
<< DPFC_RECOMP_TIMER_COUNT_SHIFT
));
211 I915_WRITE(ILK_DPFC_FENCE_YOFF
, crtc
->y
);
212 I915_WRITE(ILK_FBC_RT_BASE
, obj
->gtt_offset
| ILK_FBC_RT_VALID
);
214 I915_WRITE(ILK_DPFC_CONTROL
, dpfc_ctl
| DPFC_CTL_EN
);
217 I915_WRITE(SNB_DPFC_CTL_SA
,
218 SNB_CPU_FENCE_ENABLE
| obj
->fence_reg
);
219 I915_WRITE(DPFC_CPU_FENCE_OFFSET
, crtc
->y
);
220 sandybridge_blit_fbc_update(dev
);
223 DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc
->plane
);
226 static void ironlake_disable_fbc(struct drm_device
*dev
)
228 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
231 /* Disable compression */
232 dpfc_ctl
= I915_READ(ILK_DPFC_CONTROL
);
233 if (dpfc_ctl
& DPFC_CTL_EN
) {
234 dpfc_ctl
&= ~DPFC_CTL_EN
;
235 I915_WRITE(ILK_DPFC_CONTROL
, dpfc_ctl
);
237 DRM_DEBUG_KMS("disabled FBC\n");
241 static bool ironlake_fbc_enabled(struct drm_device
*dev
)
243 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
245 return I915_READ(ILK_DPFC_CONTROL
) & DPFC_CTL_EN
;
248 bool intel_fbc_enabled(struct drm_device
*dev
)
250 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
252 if (!dev_priv
->display
.fbc_enabled
)
255 return dev_priv
->display
.fbc_enabled(dev
);
258 static void intel_fbc_work_fn(struct work_struct
*__work
)
260 struct intel_fbc_work
*work
=
261 container_of(to_delayed_work(__work
),
262 struct intel_fbc_work
, work
);
263 struct drm_device
*dev
= work
->crtc
->dev
;
264 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
266 mutex_lock(&dev
->struct_mutex
);
267 if (work
== dev_priv
->fbc_work
) {
268 /* Double check that we haven't switched fb without cancelling
271 if (work
->crtc
->fb
== work
->fb
) {
272 dev_priv
->display
.enable_fbc(work
->crtc
,
275 dev_priv
->cfb_plane
= to_intel_crtc(work
->crtc
)->plane
;
276 dev_priv
->cfb_fb
= work
->crtc
->fb
->base
.id
;
277 dev_priv
->cfb_y
= work
->crtc
->y
;
280 dev_priv
->fbc_work
= NULL
;
282 mutex_unlock(&dev
->struct_mutex
);
287 static void intel_cancel_fbc_work(struct drm_i915_private
*dev_priv
)
289 if (dev_priv
->fbc_work
== NULL
)
292 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
294 /* Synchronisation is provided by struct_mutex and checking of
295 * dev_priv->fbc_work, so we can perform the cancellation
296 * entirely asynchronously.
298 if (cancel_delayed_work(&dev_priv
->fbc_work
->work
))
299 /* tasklet was killed before being run, clean up */
300 kfree(dev_priv
->fbc_work
);
302 /* Mark the work as no longer wanted so that if it does
303 * wake-up (because the work was already running and waiting
304 * for our mutex), it will discover that is no longer
307 dev_priv
->fbc_work
= NULL
;
310 void intel_enable_fbc(struct drm_crtc
*crtc
, unsigned long interval
)
312 struct intel_fbc_work
*work
;
313 struct drm_device
*dev
= crtc
->dev
;
314 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
316 if (!dev_priv
->display
.enable_fbc
)
319 intel_cancel_fbc_work(dev_priv
);
321 work
= kzalloc(sizeof *work
, GFP_KERNEL
);
323 dev_priv
->display
.enable_fbc(crtc
, interval
);
329 work
->interval
= interval
;
330 INIT_DELAYED_WORK(&work
->work
, intel_fbc_work_fn
);
332 dev_priv
->fbc_work
= work
;
334 DRM_DEBUG_KMS("scheduling delayed FBC enable\n");
336 /* Delay the actual enabling to let pageflipping cease and the
337 * display to settle before starting the compression. Note that
338 * this delay also serves a second purpose: it allows for a
339 * vblank to pass after disabling the FBC before we attempt
340 * to modify the control registers.
342 * A more complicated solution would involve tracking vblanks
343 * following the termination of the page-flipping sequence
344 * and indeed performing the enable as a co-routine and not
345 * waiting synchronously upon the vblank.
347 schedule_delayed_work(&work
->work
, msecs_to_jiffies(50));
350 void intel_disable_fbc(struct drm_device
*dev
)
352 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
354 intel_cancel_fbc_work(dev_priv
);
356 if (!dev_priv
->display
.disable_fbc
)
359 dev_priv
->display
.disable_fbc(dev
);
360 dev_priv
->cfb_plane
= -1;
364 * intel_update_fbc - enable/disable FBC as needed
365 * @dev: the drm_device
367 * Set up the framebuffer compression hardware at mode set time. We
368 * enable it if possible:
369 * - plane A only (on pre-965)
370 * - no pixel mulitply/line duplication
371 * - no alpha buffer discard
373 * - framebuffer <= 2048 in width, 1536 in height
375 * We can't assume that any compression will take place (worst case),
376 * so the compressed buffer has to be the same size as the uncompressed
377 * one. It also must reside (along with the line length buffer) in
380 * We need to enable/disable FBC on a global basis.
382 void intel_update_fbc(struct drm_device
*dev
)
384 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
385 struct drm_crtc
*crtc
= NULL
, *tmp_crtc
;
386 struct intel_crtc
*intel_crtc
;
387 struct drm_framebuffer
*fb
;
388 struct intel_framebuffer
*intel_fb
;
389 struct drm_i915_gem_object
*obj
;
395 if (!I915_HAS_FBC(dev
))
399 * If FBC is already on, we just have to verify that we can
400 * keep it that way...
401 * Need to disable if:
402 * - more than one pipe is active
403 * - changing FBC params (stride, fence, mode)
404 * - new fb is too large to fit in compressed buffer
405 * - going to an unsupported config (interlace, pixel multiply, etc.)
407 list_for_each_entry(tmp_crtc
, &dev
->mode_config
.crtc_list
, head
) {
408 if (tmp_crtc
->enabled
&&
409 !to_intel_crtc(tmp_crtc
)->primary_disabled
&&
412 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
413 dev_priv
->no_fbc_reason
= FBC_MULTIPLE_PIPES
;
420 if (!crtc
|| crtc
->fb
== NULL
) {
421 DRM_DEBUG_KMS("no output, disabling\n");
422 dev_priv
->no_fbc_reason
= FBC_NO_OUTPUT
;
426 intel_crtc
= to_intel_crtc(crtc
);
428 intel_fb
= to_intel_framebuffer(fb
);
431 enable_fbc
= i915_enable_fbc
;
432 if (enable_fbc
< 0) {
433 DRM_DEBUG_KMS("fbc set to per-chip default\n");
435 if (INTEL_INFO(dev
)->gen
<= 6)
439 DRM_DEBUG_KMS("fbc disabled per module param\n");
440 dev_priv
->no_fbc_reason
= FBC_MODULE_PARAM
;
443 if (intel_fb
->obj
->base
.size
> dev_priv
->cfb_size
) {
444 DRM_DEBUG_KMS("framebuffer too large, disabling "
446 dev_priv
->no_fbc_reason
= FBC_STOLEN_TOO_SMALL
;
449 if ((crtc
->mode
.flags
& DRM_MODE_FLAG_INTERLACE
) ||
450 (crtc
->mode
.flags
& DRM_MODE_FLAG_DBLSCAN
)) {
451 DRM_DEBUG_KMS("mode incompatible with compression, "
453 dev_priv
->no_fbc_reason
= FBC_UNSUPPORTED_MODE
;
456 if ((crtc
->mode
.hdisplay
> 2048) ||
457 (crtc
->mode
.vdisplay
> 1536)) {
458 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
459 dev_priv
->no_fbc_reason
= FBC_MODE_TOO_LARGE
;
462 if ((IS_I915GM(dev
) || IS_I945GM(dev
)) && intel_crtc
->plane
!= 0) {
463 DRM_DEBUG_KMS("plane not 0, disabling compression\n");
464 dev_priv
->no_fbc_reason
= FBC_BAD_PLANE
;
468 /* The use of a CPU fence is mandatory in order to detect writes
469 * by the CPU to the scanout and trigger updates to the FBC.
471 if (obj
->tiling_mode
!= I915_TILING_X
||
472 obj
->fence_reg
== I915_FENCE_REG_NONE
) {
473 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
474 dev_priv
->no_fbc_reason
= FBC_NOT_TILED
;
478 /* If the kernel debugger is active, always disable compression */
482 /* If the scanout has not changed, don't modify the FBC settings.
483 * Note that we make the fundamental assumption that the fb->obj
484 * cannot be unpinned (and have its GTT offset and fence revoked)
485 * without first being decoupled from the scanout and FBC disabled.
487 if (dev_priv
->cfb_plane
== intel_crtc
->plane
&&
488 dev_priv
->cfb_fb
== fb
->base
.id
&&
489 dev_priv
->cfb_y
== crtc
->y
)
492 if (intel_fbc_enabled(dev
)) {
493 /* We update FBC along two paths, after changing fb/crtc
494 * configuration (modeswitching) and after page-flipping
495 * finishes. For the latter, we know that not only did
496 * we disable the FBC at the start of the page-flip
497 * sequence, but also more than one vblank has passed.
499 * For the former case of modeswitching, it is possible
500 * to switch between two FBC valid configurations
501 * instantaneously so we do need to disable the FBC
502 * before we can modify its control registers. We also
503 * have to wait for the next vblank for that to take
504 * effect. However, since we delay enabling FBC we can
505 * assume that a vblank has passed since disabling and
506 * that we can safely alter the registers in the deferred
509 * In the scenario that we go from a valid to invalid
510 * and then back to valid FBC configuration we have
511 * no strict enforcement that a vblank occurred since
512 * disabling the FBC. However, along all current pipe
513 * disabling paths we do need to wait for a vblank at
514 * some point. And we wait before enabling FBC anyway.
516 DRM_DEBUG_KMS("disabling active FBC for update\n");
517 intel_disable_fbc(dev
);
520 intel_enable_fbc(crtc
, 500);
524 /* Multiple disables should be harmless */
525 if (intel_fbc_enabled(dev
)) {
526 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
527 intel_disable_fbc(dev
);
531 static void i915_pineview_get_mem_freq(struct drm_device
*dev
)
533 drm_i915_private_t
*dev_priv
= dev
->dev_private
;
536 tmp
= I915_READ(CLKCFG
);
538 switch (tmp
& CLKCFG_FSB_MASK
) {
540 dev_priv
->fsb_freq
= 533; /* 133*4 */
543 dev_priv
->fsb_freq
= 800; /* 200*4 */
546 dev_priv
->fsb_freq
= 667; /* 167*4 */
549 dev_priv
->fsb_freq
= 400; /* 100*4 */
553 switch (tmp
& CLKCFG_MEM_MASK
) {
555 dev_priv
->mem_freq
= 533;
558 dev_priv
->mem_freq
= 667;
561 dev_priv
->mem_freq
= 800;
565 /* detect pineview DDR3 setting */
566 tmp
= I915_READ(CSHRDDR3CTL
);
567 dev_priv
->is_ddr3
= (tmp
& CSHRDDR3CTL_DDR3
) ? 1 : 0;
570 static void i915_ironlake_get_mem_freq(struct drm_device
*dev
)
572 drm_i915_private_t
*dev_priv
= dev
->dev_private
;
575 ddrpll
= I915_READ16(DDRMPLL1
);
576 csipll
= I915_READ16(CSIPLL0
);
578 switch (ddrpll
& 0xff) {
580 dev_priv
->mem_freq
= 800;
583 dev_priv
->mem_freq
= 1066;
586 dev_priv
->mem_freq
= 1333;
589 dev_priv
->mem_freq
= 1600;
592 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
594 dev_priv
->mem_freq
= 0;
598 dev_priv
->ips
.r_t
= dev_priv
->mem_freq
;
600 switch (csipll
& 0x3ff) {
602 dev_priv
->fsb_freq
= 3200;
605 dev_priv
->fsb_freq
= 3733;
608 dev_priv
->fsb_freq
= 4266;
611 dev_priv
->fsb_freq
= 4800;
614 dev_priv
->fsb_freq
= 5333;
617 dev_priv
->fsb_freq
= 5866;
620 dev_priv
->fsb_freq
= 6400;
623 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
625 dev_priv
->fsb_freq
= 0;
629 if (dev_priv
->fsb_freq
== 3200) {
630 dev_priv
->ips
.c_m
= 0;
631 } else if (dev_priv
->fsb_freq
> 3200 && dev_priv
->fsb_freq
<= 4800) {
632 dev_priv
->ips
.c_m
= 1;
634 dev_priv
->ips
.c_m
= 2;
638 static const struct cxsr_latency cxsr_latency_table
[] = {
639 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
640 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
641 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
642 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
643 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
645 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
646 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
647 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
648 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
649 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
651 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
652 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
653 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
654 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
655 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
657 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
658 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
659 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
660 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
661 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
663 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
664 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
665 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
666 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
667 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
669 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
670 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
671 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
672 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
673 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
676 static const struct cxsr_latency
*intel_get_cxsr_latency(int is_desktop
,
681 const struct cxsr_latency
*latency
;
684 if (fsb
== 0 || mem
== 0)
687 for (i
= 0; i
< ARRAY_SIZE(cxsr_latency_table
); i
++) {
688 latency
= &cxsr_latency_table
[i
];
689 if (is_desktop
== latency
->is_desktop
&&
690 is_ddr3
== latency
->is_ddr3
&&
691 fsb
== latency
->fsb_freq
&& mem
== latency
->mem_freq
)
695 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
700 static void pineview_disable_cxsr(struct drm_device
*dev
)
702 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
704 /* deactivate cxsr */
705 I915_WRITE(DSPFW3
, I915_READ(DSPFW3
) & ~PINEVIEW_SELF_REFRESH_EN
);
709 * Latency for FIFO fetches is dependent on several factors:
710 * - memory configuration (speed, channels)
712 * - current MCH state
713 * It can be fairly high in some situations, so here we assume a fairly
714 * pessimal value. It's a tradeoff between extra memory fetches (if we
715 * set this value too high, the FIFO will fetch frequently to stay full)
716 * and power consumption (set it too low to save power and we might see
717 * FIFO underruns and display "flicker").
719 * A value of 5us seems to be a good balance; safe for very low end
720 * platforms but not overly aggressive on lower latency configs.
722 static const int latency_ns
= 5000;
724 static int i9xx_get_fifo_size(struct drm_device
*dev
, int plane
)
726 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
727 uint32_t dsparb
= I915_READ(DSPARB
);
730 size
= dsparb
& 0x7f;
732 size
= ((dsparb
>> DSPARB_CSTART_SHIFT
) & 0x7f) - size
;
734 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
735 plane
? "B" : "A", size
);
740 static int i85x_get_fifo_size(struct drm_device
*dev
, int plane
)
742 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
743 uint32_t dsparb
= I915_READ(DSPARB
);
746 size
= dsparb
& 0x1ff;
748 size
= ((dsparb
>> DSPARB_BEND_SHIFT
) & 0x1ff) - size
;
749 size
>>= 1; /* Convert to cachelines */
751 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
752 plane
? "B" : "A", size
);
757 static int i845_get_fifo_size(struct drm_device
*dev
, int plane
)
759 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
760 uint32_t dsparb
= I915_READ(DSPARB
);
763 size
= dsparb
& 0x7f;
764 size
>>= 2; /* Convert to cachelines */
766 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
773 static int i830_get_fifo_size(struct drm_device
*dev
, int plane
)
775 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
776 uint32_t dsparb
= I915_READ(DSPARB
);
779 size
= dsparb
& 0x7f;
780 size
>>= 1; /* Convert to cachelines */
782 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
783 plane
? "B" : "A", size
);
788 /* Pineview has different values for various configs */
789 static const struct intel_watermark_params pineview_display_wm
= {
790 PINEVIEW_DISPLAY_FIFO
,
794 PINEVIEW_FIFO_LINE_SIZE
796 static const struct intel_watermark_params pineview_display_hplloff_wm
= {
797 PINEVIEW_DISPLAY_FIFO
,
799 PINEVIEW_DFT_HPLLOFF_WM
,
801 PINEVIEW_FIFO_LINE_SIZE
803 static const struct intel_watermark_params pineview_cursor_wm
= {
804 PINEVIEW_CURSOR_FIFO
,
805 PINEVIEW_CURSOR_MAX_WM
,
806 PINEVIEW_CURSOR_DFT_WM
,
807 PINEVIEW_CURSOR_GUARD_WM
,
808 PINEVIEW_FIFO_LINE_SIZE
,
810 static const struct intel_watermark_params pineview_cursor_hplloff_wm
= {
811 PINEVIEW_CURSOR_FIFO
,
812 PINEVIEW_CURSOR_MAX_WM
,
813 PINEVIEW_CURSOR_DFT_WM
,
814 PINEVIEW_CURSOR_GUARD_WM
,
815 PINEVIEW_FIFO_LINE_SIZE
817 static const struct intel_watermark_params g4x_wm_info
= {
824 static const struct intel_watermark_params g4x_cursor_wm_info
= {
831 static const struct intel_watermark_params valleyview_wm_info
= {
832 VALLEYVIEW_FIFO_SIZE
,
838 static const struct intel_watermark_params valleyview_cursor_wm_info
= {
840 VALLEYVIEW_CURSOR_MAX_WM
,
845 static const struct intel_watermark_params i965_cursor_wm_info
= {
852 static const struct intel_watermark_params i945_wm_info
= {
859 static const struct intel_watermark_params i915_wm_info
= {
866 static const struct intel_watermark_params i855_wm_info
= {
873 static const struct intel_watermark_params i830_wm_info
= {
881 static const struct intel_watermark_params ironlake_display_wm_info
= {
888 static const struct intel_watermark_params ironlake_cursor_wm_info
= {
895 static const struct intel_watermark_params ironlake_display_srwm_info
= {
897 ILK_DISPLAY_MAX_SRWM
,
898 ILK_DISPLAY_DFT_SRWM
,
902 static const struct intel_watermark_params ironlake_cursor_srwm_info
= {
910 static const struct intel_watermark_params sandybridge_display_wm_info
= {
917 static const struct intel_watermark_params sandybridge_cursor_wm_info
= {
924 static const struct intel_watermark_params sandybridge_display_srwm_info
= {
926 SNB_DISPLAY_MAX_SRWM
,
927 SNB_DISPLAY_DFT_SRWM
,
931 static const struct intel_watermark_params sandybridge_cursor_srwm_info
= {
941 * intel_calculate_wm - calculate watermark level
942 * @clock_in_khz: pixel clock
943 * @wm: chip FIFO params
944 * @pixel_size: display pixel size
945 * @latency_ns: memory latency for the platform
947 * Calculate the watermark level (the level at which the display plane will
948 * start fetching from memory again). Each chip has a different display
949 * FIFO size and allocation, so the caller needs to figure that out and pass
950 * in the correct intel_watermark_params structure.
952 * As the pixel clock runs, the FIFO will be drained at a rate that depends
953 * on the pixel size. When it reaches the watermark level, it'll start
954 * fetching FIFO line sized based chunks from memory until the FIFO fills
955 * past the watermark point. If the FIFO drains completely, a FIFO underrun
956 * will occur, and a display engine hang could result.
958 static unsigned long intel_calculate_wm(unsigned long clock_in_khz
,
959 const struct intel_watermark_params
*wm
,
962 unsigned long latency_ns
)
964 long entries_required
, wm_size
;
967 * Note: we need to make sure we don't overflow for various clock &
969 * clocks go from a few thousand to several hundred thousand.
970 * latency is usually a few thousand
972 entries_required
= ((clock_in_khz
/ 1000) * pixel_size
* latency_ns
) /
974 entries_required
= DIV_ROUND_UP(entries_required
, wm
->cacheline_size
);
976 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required
);
978 wm_size
= fifo_size
- (entries_required
+ wm
->guard_size
);
980 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size
);
982 /* Don't promote wm_size to unsigned... */
983 if (wm_size
> (long)wm
->max_wm
)
984 wm_size
= wm
->max_wm
;
986 wm_size
= wm
->default_wm
;
990 static struct drm_crtc
*single_enabled_crtc(struct drm_device
*dev
)
992 struct drm_crtc
*crtc
, *enabled
= NULL
;
994 list_for_each_entry(crtc
, &dev
->mode_config
.crtc_list
, head
) {
995 if (crtc
->enabled
&& crtc
->fb
) {
1005 static void pineview_update_wm(struct drm_device
*dev
)
1007 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1008 struct drm_crtc
*crtc
;
1009 const struct cxsr_latency
*latency
;
1013 latency
= intel_get_cxsr_latency(IS_PINEVIEW_G(dev
), dev_priv
->is_ddr3
,
1014 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
1016 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1017 pineview_disable_cxsr(dev
);
1021 crtc
= single_enabled_crtc(dev
);
1023 int clock
= crtc
->mode
.clock
;
1024 int pixel_size
= crtc
->fb
->bits_per_pixel
/ 8;
1027 wm
= intel_calculate_wm(clock
, &pineview_display_wm
,
1028 pineview_display_wm
.fifo_size
,
1029 pixel_size
, latency
->display_sr
);
1030 reg
= I915_READ(DSPFW1
);
1031 reg
&= ~DSPFW_SR_MASK
;
1032 reg
|= wm
<< DSPFW_SR_SHIFT
;
1033 I915_WRITE(DSPFW1
, reg
);
1034 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg
);
1037 wm
= intel_calculate_wm(clock
, &pineview_cursor_wm
,
1038 pineview_display_wm
.fifo_size
,
1039 pixel_size
, latency
->cursor_sr
);
1040 reg
= I915_READ(DSPFW3
);
1041 reg
&= ~DSPFW_CURSOR_SR_MASK
;
1042 reg
|= (wm
& 0x3f) << DSPFW_CURSOR_SR_SHIFT
;
1043 I915_WRITE(DSPFW3
, reg
);
1045 /* Display HPLL off SR */
1046 wm
= intel_calculate_wm(clock
, &pineview_display_hplloff_wm
,
1047 pineview_display_hplloff_wm
.fifo_size
,
1048 pixel_size
, latency
->display_hpll_disable
);
1049 reg
= I915_READ(DSPFW3
);
1050 reg
&= ~DSPFW_HPLL_SR_MASK
;
1051 reg
|= wm
& DSPFW_HPLL_SR_MASK
;
1052 I915_WRITE(DSPFW3
, reg
);
1054 /* cursor HPLL off SR */
1055 wm
= intel_calculate_wm(clock
, &pineview_cursor_hplloff_wm
,
1056 pineview_display_hplloff_wm
.fifo_size
,
1057 pixel_size
, latency
->cursor_hpll_disable
);
1058 reg
= I915_READ(DSPFW3
);
1059 reg
&= ~DSPFW_HPLL_CURSOR_MASK
;
1060 reg
|= (wm
& 0x3f) << DSPFW_HPLL_CURSOR_SHIFT
;
1061 I915_WRITE(DSPFW3
, reg
);
1062 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg
);
1066 I915_READ(DSPFW3
) | PINEVIEW_SELF_REFRESH_EN
);
1067 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1069 pineview_disable_cxsr(dev
);
1070 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1074 static bool g4x_compute_wm0(struct drm_device
*dev
,
1076 const struct intel_watermark_params
*display
,
1077 int display_latency_ns
,
1078 const struct intel_watermark_params
*cursor
,
1079 int cursor_latency_ns
,
1083 struct drm_crtc
*crtc
;
1084 int htotal
, hdisplay
, clock
, pixel_size
;
1085 int line_time_us
, line_count
;
1086 int entries
, tlb_miss
;
1088 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1089 if (crtc
->fb
== NULL
|| !crtc
->enabled
) {
1090 *cursor_wm
= cursor
->guard_size
;
1091 *plane_wm
= display
->guard_size
;
1095 htotal
= crtc
->mode
.htotal
;
1096 hdisplay
= crtc
->mode
.hdisplay
;
1097 clock
= crtc
->mode
.clock
;
1098 pixel_size
= crtc
->fb
->bits_per_pixel
/ 8;
1100 /* Use the small buffer method to calculate plane watermark */
1101 entries
= ((clock
* pixel_size
/ 1000) * display_latency_ns
) / 1000;
1102 tlb_miss
= display
->fifo_size
*display
->cacheline_size
- hdisplay
* 8;
1104 entries
+= tlb_miss
;
1105 entries
= DIV_ROUND_UP(entries
, display
->cacheline_size
);
1106 *plane_wm
= entries
+ display
->guard_size
;
1107 if (*plane_wm
> (int)display
->max_wm
)
1108 *plane_wm
= display
->max_wm
;
1110 /* Use the large buffer method to calculate cursor watermark */
1111 line_time_us
= ((htotal
* 1000) / clock
);
1112 line_count
= (cursor_latency_ns
/ line_time_us
+ 1000) / 1000;
1113 entries
= line_count
* 64 * pixel_size
;
1114 tlb_miss
= cursor
->fifo_size
*cursor
->cacheline_size
- hdisplay
* 8;
1116 entries
+= tlb_miss
;
1117 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
1118 *cursor_wm
= entries
+ cursor
->guard_size
;
1119 if (*cursor_wm
> (int)cursor
->max_wm
)
1120 *cursor_wm
= (int)cursor
->max_wm
;
1126 * Check the wm result.
1128 * If any calculated watermark values is larger than the maximum value that
1129 * can be programmed into the associated watermark register, that watermark
1132 static bool g4x_check_srwm(struct drm_device
*dev
,
1133 int display_wm
, int cursor_wm
,
1134 const struct intel_watermark_params
*display
,
1135 const struct intel_watermark_params
*cursor
)
1137 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1138 display_wm
, cursor_wm
);
1140 if (display_wm
> display
->max_wm
) {
1141 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1142 display_wm
, display
->max_wm
);
1146 if (cursor_wm
> cursor
->max_wm
) {
1147 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1148 cursor_wm
, cursor
->max_wm
);
1152 if (!(display_wm
|| cursor_wm
)) {
1153 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1160 static bool g4x_compute_srwm(struct drm_device
*dev
,
1163 const struct intel_watermark_params
*display
,
1164 const struct intel_watermark_params
*cursor
,
1165 int *display_wm
, int *cursor_wm
)
1167 struct drm_crtc
*crtc
;
1168 int hdisplay
, htotal
, pixel_size
, clock
;
1169 unsigned long line_time_us
;
1170 int line_count
, line_size
;
1175 *display_wm
= *cursor_wm
= 0;
1179 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1180 hdisplay
= crtc
->mode
.hdisplay
;
1181 htotal
= crtc
->mode
.htotal
;
1182 clock
= crtc
->mode
.clock
;
1183 pixel_size
= crtc
->fb
->bits_per_pixel
/ 8;
1185 line_time_us
= (htotal
* 1000) / clock
;
1186 line_count
= (latency_ns
/ line_time_us
+ 1000) / 1000;
1187 line_size
= hdisplay
* pixel_size
;
1189 /* Use the minimum of the small and large buffer method for primary */
1190 small
= ((clock
* pixel_size
/ 1000) * latency_ns
) / 1000;
1191 large
= line_count
* line_size
;
1193 entries
= DIV_ROUND_UP(min(small
, large
), display
->cacheline_size
);
1194 *display_wm
= entries
+ display
->guard_size
;
1196 /* calculate the self-refresh watermark for display cursor */
1197 entries
= line_count
* pixel_size
* 64;
1198 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
1199 *cursor_wm
= entries
+ cursor
->guard_size
;
1201 return g4x_check_srwm(dev
,
1202 *display_wm
, *cursor_wm
,
1206 static bool vlv_compute_drain_latency(struct drm_device
*dev
,
1208 int *plane_prec_mult
,
1210 int *cursor_prec_mult
,
1213 struct drm_crtc
*crtc
;
1214 int clock
, pixel_size
;
1217 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1218 if (crtc
->fb
== NULL
|| !crtc
->enabled
)
1221 clock
= crtc
->mode
.clock
; /* VESA DOT Clock */
1222 pixel_size
= crtc
->fb
->bits_per_pixel
/ 8; /* BPP */
1224 entries
= (clock
/ 1000) * pixel_size
;
1225 *plane_prec_mult
= (entries
> 256) ?
1226 DRAIN_LATENCY_PRECISION_32
: DRAIN_LATENCY_PRECISION_16
;
1227 *plane_dl
= (64 * (*plane_prec_mult
) * 4) / ((clock
/ 1000) *
1230 entries
= (clock
/ 1000) * 4; /* BPP is always 4 for cursor */
1231 *cursor_prec_mult
= (entries
> 256) ?
1232 DRAIN_LATENCY_PRECISION_32
: DRAIN_LATENCY_PRECISION_16
;
1233 *cursor_dl
= (64 * (*cursor_prec_mult
) * 4) / ((clock
/ 1000) * 4);
1239 * Update drain latency registers of memory arbiter
1241 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1242 * to be programmed. Each plane has a drain latency multiplier and a drain
1246 static void vlv_update_drain_latency(struct drm_device
*dev
)
1248 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1249 int planea_prec
, planea_dl
, planeb_prec
, planeb_dl
;
1250 int cursora_prec
, cursora_dl
, cursorb_prec
, cursorb_dl
;
1251 int plane_prec_mult
, cursor_prec_mult
; /* Precision multiplier is
1254 /* For plane A, Cursor A */
1255 if (vlv_compute_drain_latency(dev
, 0, &plane_prec_mult
, &planea_dl
,
1256 &cursor_prec_mult
, &cursora_dl
)) {
1257 cursora_prec
= (cursor_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1258 DDL_CURSORA_PRECISION_32
: DDL_CURSORA_PRECISION_16
;
1259 planea_prec
= (plane_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1260 DDL_PLANEA_PRECISION_32
: DDL_PLANEA_PRECISION_16
;
1262 I915_WRITE(VLV_DDL1
, cursora_prec
|
1263 (cursora_dl
<< DDL_CURSORA_SHIFT
) |
1264 planea_prec
| planea_dl
);
1267 /* For plane B, Cursor B */
1268 if (vlv_compute_drain_latency(dev
, 1, &plane_prec_mult
, &planeb_dl
,
1269 &cursor_prec_mult
, &cursorb_dl
)) {
1270 cursorb_prec
= (cursor_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1271 DDL_CURSORB_PRECISION_32
: DDL_CURSORB_PRECISION_16
;
1272 planeb_prec
= (plane_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1273 DDL_PLANEB_PRECISION_32
: DDL_PLANEB_PRECISION_16
;
1275 I915_WRITE(VLV_DDL2
, cursorb_prec
|
1276 (cursorb_dl
<< DDL_CURSORB_SHIFT
) |
1277 planeb_prec
| planeb_dl
);
1281 #define single_plane_enabled(mask) is_power_of_2(mask)
1283 static void valleyview_update_wm(struct drm_device
*dev
)
1285 static const int sr_latency_ns
= 12000;
1286 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1287 int planea_wm
, planeb_wm
, cursora_wm
, cursorb_wm
;
1288 int plane_sr
, cursor_sr
;
1289 unsigned int enabled
= 0;
1291 vlv_update_drain_latency(dev
);
1293 if (g4x_compute_wm0(dev
, 0,
1294 &valleyview_wm_info
, latency_ns
,
1295 &valleyview_cursor_wm_info
, latency_ns
,
1296 &planea_wm
, &cursora_wm
))
1299 if (g4x_compute_wm0(dev
, 1,
1300 &valleyview_wm_info
, latency_ns
,
1301 &valleyview_cursor_wm_info
, latency_ns
,
1302 &planeb_wm
, &cursorb_wm
))
1305 plane_sr
= cursor_sr
= 0;
1306 if (single_plane_enabled(enabled
) &&
1307 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1309 &valleyview_wm_info
,
1310 &valleyview_cursor_wm_info
,
1311 &plane_sr
, &cursor_sr
))
1312 I915_WRITE(FW_BLC_SELF_VLV
, FW_CSPWRDWNEN
);
1314 I915_WRITE(FW_BLC_SELF_VLV
,
1315 I915_READ(FW_BLC_SELF_VLV
) & ~FW_CSPWRDWNEN
);
1317 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1318 planea_wm
, cursora_wm
,
1319 planeb_wm
, cursorb_wm
,
1320 plane_sr
, cursor_sr
);
1323 (plane_sr
<< DSPFW_SR_SHIFT
) |
1324 (cursorb_wm
<< DSPFW_CURSORB_SHIFT
) |
1325 (planeb_wm
<< DSPFW_PLANEB_SHIFT
) |
1328 (I915_READ(DSPFW2
) & DSPFW_CURSORA_MASK
) |
1329 (cursora_wm
<< DSPFW_CURSORA_SHIFT
));
1331 (I915_READ(DSPFW3
) | (cursor_sr
<< DSPFW_CURSOR_SR_SHIFT
)));
1334 static void g4x_update_wm(struct drm_device
*dev
)
1336 static const int sr_latency_ns
= 12000;
1337 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1338 int planea_wm
, planeb_wm
, cursora_wm
, cursorb_wm
;
1339 int plane_sr
, cursor_sr
;
1340 unsigned int enabled
= 0;
1342 if (g4x_compute_wm0(dev
, 0,
1343 &g4x_wm_info
, latency_ns
,
1344 &g4x_cursor_wm_info
, latency_ns
,
1345 &planea_wm
, &cursora_wm
))
1348 if (g4x_compute_wm0(dev
, 1,
1349 &g4x_wm_info
, latency_ns
,
1350 &g4x_cursor_wm_info
, latency_ns
,
1351 &planeb_wm
, &cursorb_wm
))
1354 plane_sr
= cursor_sr
= 0;
1355 if (single_plane_enabled(enabled
) &&
1356 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1359 &g4x_cursor_wm_info
,
1360 &plane_sr
, &cursor_sr
))
1361 I915_WRITE(FW_BLC_SELF
, FW_BLC_SELF_EN
);
1363 I915_WRITE(FW_BLC_SELF
,
1364 I915_READ(FW_BLC_SELF
) & ~FW_BLC_SELF_EN
);
1366 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1367 planea_wm
, cursora_wm
,
1368 planeb_wm
, cursorb_wm
,
1369 plane_sr
, cursor_sr
);
1372 (plane_sr
<< DSPFW_SR_SHIFT
) |
1373 (cursorb_wm
<< DSPFW_CURSORB_SHIFT
) |
1374 (planeb_wm
<< DSPFW_PLANEB_SHIFT
) |
1377 (I915_READ(DSPFW2
) & DSPFW_CURSORA_MASK
) |
1378 (cursora_wm
<< DSPFW_CURSORA_SHIFT
));
1379 /* HPLL off in SR has some issues on G4x... disable it */
1381 (I915_READ(DSPFW3
) & ~DSPFW_HPLL_SR_EN
) |
1382 (cursor_sr
<< DSPFW_CURSOR_SR_SHIFT
));
1385 static void i965_update_wm(struct drm_device
*dev
)
1387 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1388 struct drm_crtc
*crtc
;
1392 /* Calc sr entries for one plane configs */
1393 crtc
= single_enabled_crtc(dev
);
1395 /* self-refresh has much higher latency */
1396 static const int sr_latency_ns
= 12000;
1397 int clock
= crtc
->mode
.clock
;
1398 int htotal
= crtc
->mode
.htotal
;
1399 int hdisplay
= crtc
->mode
.hdisplay
;
1400 int pixel_size
= crtc
->fb
->bits_per_pixel
/ 8;
1401 unsigned long line_time_us
;
1404 line_time_us
= ((htotal
* 1000) / clock
);
1406 /* Use ns/us then divide to preserve precision */
1407 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1408 pixel_size
* hdisplay
;
1409 entries
= DIV_ROUND_UP(entries
, I915_FIFO_LINE_SIZE
);
1410 srwm
= I965_FIFO_SIZE
- entries
;
1414 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1417 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1419 entries
= DIV_ROUND_UP(entries
,
1420 i965_cursor_wm_info
.cacheline_size
);
1421 cursor_sr
= i965_cursor_wm_info
.fifo_size
-
1422 (entries
+ i965_cursor_wm_info
.guard_size
);
1424 if (cursor_sr
> i965_cursor_wm_info
.max_wm
)
1425 cursor_sr
= i965_cursor_wm_info
.max_wm
;
1427 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1428 "cursor %d\n", srwm
, cursor_sr
);
1430 if (IS_CRESTLINE(dev
))
1431 I915_WRITE(FW_BLC_SELF
, FW_BLC_SELF_EN
);
1433 /* Turn off self refresh if both pipes are enabled */
1434 if (IS_CRESTLINE(dev
))
1435 I915_WRITE(FW_BLC_SELF
, I915_READ(FW_BLC_SELF
)
1439 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1442 /* 965 has limitations... */
1443 I915_WRITE(DSPFW1
, (srwm
<< DSPFW_SR_SHIFT
) |
1444 (8 << 16) | (8 << 8) | (8 << 0));
1445 I915_WRITE(DSPFW2
, (8 << 8) | (8 << 0));
1446 /* update cursor SR watermark */
1447 I915_WRITE(DSPFW3
, (cursor_sr
<< DSPFW_CURSOR_SR_SHIFT
));
1450 static void i9xx_update_wm(struct drm_device
*dev
)
1452 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1453 const struct intel_watermark_params
*wm_info
;
1458 int planea_wm
, planeb_wm
;
1459 struct drm_crtc
*crtc
, *enabled
= NULL
;
1462 wm_info
= &i945_wm_info
;
1463 else if (!IS_GEN2(dev
))
1464 wm_info
= &i915_wm_info
;
1466 wm_info
= &i855_wm_info
;
1468 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 0);
1469 crtc
= intel_get_crtc_for_plane(dev
, 0);
1470 if (crtc
->enabled
&& crtc
->fb
) {
1471 planea_wm
= intel_calculate_wm(crtc
->mode
.clock
,
1473 crtc
->fb
->bits_per_pixel
/ 8,
1477 planea_wm
= fifo_size
- wm_info
->guard_size
;
1479 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 1);
1480 crtc
= intel_get_crtc_for_plane(dev
, 1);
1481 if (crtc
->enabled
&& crtc
->fb
) {
1482 planeb_wm
= intel_calculate_wm(crtc
->mode
.clock
,
1484 crtc
->fb
->bits_per_pixel
/ 8,
1486 if (enabled
== NULL
)
1491 planeb_wm
= fifo_size
- wm_info
->guard_size
;
1493 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm
, planeb_wm
);
1496 * Overlay gets an aggressive default since video jitter is bad.
1500 /* Play safe and disable self-refresh before adjusting watermarks. */
1501 if (IS_I945G(dev
) || IS_I945GM(dev
))
1502 I915_WRITE(FW_BLC_SELF
, FW_BLC_SELF_EN_MASK
| 0);
1503 else if (IS_I915GM(dev
))
1504 I915_WRITE(INSTPM
, I915_READ(INSTPM
) & ~INSTPM_SELF_EN
);
1506 /* Calc sr entries for one plane configs */
1507 if (HAS_FW_BLC(dev
) && enabled
) {
1508 /* self-refresh has much higher latency */
1509 static const int sr_latency_ns
= 6000;
1510 int clock
= enabled
->mode
.clock
;
1511 int htotal
= enabled
->mode
.htotal
;
1512 int hdisplay
= enabled
->mode
.hdisplay
;
1513 int pixel_size
= enabled
->fb
->bits_per_pixel
/ 8;
1514 unsigned long line_time_us
;
1517 line_time_us
= (htotal
* 1000) / clock
;
1519 /* Use ns/us then divide to preserve precision */
1520 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1521 pixel_size
* hdisplay
;
1522 entries
= DIV_ROUND_UP(entries
, wm_info
->cacheline_size
);
1523 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries
);
1524 srwm
= wm_info
->fifo_size
- entries
;
1528 if (IS_I945G(dev
) || IS_I945GM(dev
))
1529 I915_WRITE(FW_BLC_SELF
,
1530 FW_BLC_SELF_FIFO_MASK
| (srwm
& 0xff));
1531 else if (IS_I915GM(dev
))
1532 I915_WRITE(FW_BLC_SELF
, srwm
& 0x3f);
1535 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1536 planea_wm
, planeb_wm
, cwm
, srwm
);
1538 fwater_lo
= ((planeb_wm
& 0x3f) << 16) | (planea_wm
& 0x3f);
1539 fwater_hi
= (cwm
& 0x1f);
1541 /* Set request length to 8 cachelines per fetch */
1542 fwater_lo
= fwater_lo
| (1 << 24) | (1 << 8);
1543 fwater_hi
= fwater_hi
| (1 << 8);
1545 I915_WRITE(FW_BLC
, fwater_lo
);
1546 I915_WRITE(FW_BLC2
, fwater_hi
);
1548 if (HAS_FW_BLC(dev
)) {
1550 if (IS_I945G(dev
) || IS_I945GM(dev
))
1551 I915_WRITE(FW_BLC_SELF
,
1552 FW_BLC_SELF_EN_MASK
| FW_BLC_SELF_EN
);
1553 else if (IS_I915GM(dev
))
1554 I915_WRITE(INSTPM
, I915_READ(INSTPM
) | INSTPM_SELF_EN
);
1555 DRM_DEBUG_KMS("memory self refresh enabled\n");
1557 DRM_DEBUG_KMS("memory self refresh disabled\n");
1561 static void i830_update_wm(struct drm_device
*dev
)
1563 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1564 struct drm_crtc
*crtc
;
1568 crtc
= single_enabled_crtc(dev
);
1572 planea_wm
= intel_calculate_wm(crtc
->mode
.clock
, &i830_wm_info
,
1573 dev_priv
->display
.get_fifo_size(dev
, 0),
1574 crtc
->fb
->bits_per_pixel
/ 8,
1576 fwater_lo
= I915_READ(FW_BLC
) & ~0xfff;
1577 fwater_lo
|= (3<<8) | planea_wm
;
1579 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm
);
1581 I915_WRITE(FW_BLC
, fwater_lo
);
1584 #define ILK_LP0_PLANE_LATENCY 700
1585 #define ILK_LP0_CURSOR_LATENCY 1300
1588 * Check the wm result.
1590 * If any calculated watermark values is larger than the maximum value that
1591 * can be programmed into the associated watermark register, that watermark
1594 static bool ironlake_check_srwm(struct drm_device
*dev
, int level
,
1595 int fbc_wm
, int display_wm
, int cursor_wm
,
1596 const struct intel_watermark_params
*display
,
1597 const struct intel_watermark_params
*cursor
)
1599 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1601 DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d,"
1602 " cursor %d\n", level
, display_wm
, fbc_wm
, cursor_wm
);
1604 if (fbc_wm
> SNB_FBC_MAX_SRWM
) {
1605 DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n",
1606 fbc_wm
, SNB_FBC_MAX_SRWM
, level
);
1608 /* fbc has it's own way to disable FBC WM */
1609 I915_WRITE(DISP_ARB_CTL
,
1610 I915_READ(DISP_ARB_CTL
) | DISP_FBC_WM_DIS
);
1614 if (display_wm
> display
->max_wm
) {
1615 DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n",
1616 display_wm
, SNB_DISPLAY_MAX_SRWM
, level
);
1620 if (cursor_wm
> cursor
->max_wm
) {
1621 DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n",
1622 cursor_wm
, SNB_CURSOR_MAX_SRWM
, level
);
1626 if (!(fbc_wm
|| display_wm
|| cursor_wm
)) {
1627 DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level
, level
);
1635 * Compute watermark values of WM[1-3],
1637 static bool ironlake_compute_srwm(struct drm_device
*dev
, int level
, int plane
,
1639 const struct intel_watermark_params
*display
,
1640 const struct intel_watermark_params
*cursor
,
1641 int *fbc_wm
, int *display_wm
, int *cursor_wm
)
1643 struct drm_crtc
*crtc
;
1644 unsigned long line_time_us
;
1645 int hdisplay
, htotal
, pixel_size
, clock
;
1646 int line_count
, line_size
;
1651 *fbc_wm
= *display_wm
= *cursor_wm
= 0;
1655 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1656 hdisplay
= crtc
->mode
.hdisplay
;
1657 htotal
= crtc
->mode
.htotal
;
1658 clock
= crtc
->mode
.clock
;
1659 pixel_size
= crtc
->fb
->bits_per_pixel
/ 8;
1661 line_time_us
= (htotal
* 1000) / clock
;
1662 line_count
= (latency_ns
/ line_time_us
+ 1000) / 1000;
1663 line_size
= hdisplay
* pixel_size
;
1665 /* Use the minimum of the small and large buffer method for primary */
1666 small
= ((clock
* pixel_size
/ 1000) * latency_ns
) / 1000;
1667 large
= line_count
* line_size
;
1669 entries
= DIV_ROUND_UP(min(small
, large
), display
->cacheline_size
);
1670 *display_wm
= entries
+ display
->guard_size
;
1674 * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2
1676 *fbc_wm
= DIV_ROUND_UP(*display_wm
* 64, line_size
) + 2;
1678 /* calculate the self-refresh watermark for display cursor */
1679 entries
= line_count
* pixel_size
* 64;
1680 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
1681 *cursor_wm
= entries
+ cursor
->guard_size
;
1683 return ironlake_check_srwm(dev
, level
,
1684 *fbc_wm
, *display_wm
, *cursor_wm
,
1688 static void ironlake_update_wm(struct drm_device
*dev
)
1690 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1691 int fbc_wm
, plane_wm
, cursor_wm
;
1692 unsigned int enabled
;
1695 if (g4x_compute_wm0(dev
, 0,
1696 &ironlake_display_wm_info
,
1697 ILK_LP0_PLANE_LATENCY
,
1698 &ironlake_cursor_wm_info
,
1699 ILK_LP0_CURSOR_LATENCY
,
1700 &plane_wm
, &cursor_wm
)) {
1701 I915_WRITE(WM0_PIPEA_ILK
,
1702 (plane_wm
<< WM0_PIPE_PLANE_SHIFT
) | cursor_wm
);
1703 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1704 " plane %d, " "cursor: %d\n",
1705 plane_wm
, cursor_wm
);
1709 if (g4x_compute_wm0(dev
, 1,
1710 &ironlake_display_wm_info
,
1711 ILK_LP0_PLANE_LATENCY
,
1712 &ironlake_cursor_wm_info
,
1713 ILK_LP0_CURSOR_LATENCY
,
1714 &plane_wm
, &cursor_wm
)) {
1715 I915_WRITE(WM0_PIPEB_ILK
,
1716 (plane_wm
<< WM0_PIPE_PLANE_SHIFT
) | cursor_wm
);
1717 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1718 " plane %d, cursor: %d\n",
1719 plane_wm
, cursor_wm
);
1724 * Calculate and update the self-refresh watermark only when one
1725 * display plane is used.
1727 I915_WRITE(WM3_LP_ILK
, 0);
1728 I915_WRITE(WM2_LP_ILK
, 0);
1729 I915_WRITE(WM1_LP_ILK
, 0);
1731 if (!single_plane_enabled(enabled
))
1733 enabled
= ffs(enabled
) - 1;
1736 if (!ironlake_compute_srwm(dev
, 1, enabled
,
1737 ILK_READ_WM1_LATENCY() * 500,
1738 &ironlake_display_srwm_info
,
1739 &ironlake_cursor_srwm_info
,
1740 &fbc_wm
, &plane_wm
, &cursor_wm
))
1743 I915_WRITE(WM1_LP_ILK
,
1745 (ILK_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT
) |
1746 (fbc_wm
<< WM1_LP_FBC_SHIFT
) |
1747 (plane_wm
<< WM1_LP_SR_SHIFT
) |
1751 if (!ironlake_compute_srwm(dev
, 2, enabled
,
1752 ILK_READ_WM2_LATENCY() * 500,
1753 &ironlake_display_srwm_info
,
1754 &ironlake_cursor_srwm_info
,
1755 &fbc_wm
, &plane_wm
, &cursor_wm
))
1758 I915_WRITE(WM2_LP_ILK
,
1760 (ILK_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT
) |
1761 (fbc_wm
<< WM1_LP_FBC_SHIFT
) |
1762 (plane_wm
<< WM1_LP_SR_SHIFT
) |
1766 * WM3 is unsupported on ILK, probably because we don't have latency
1767 * data for that power state
1771 static void sandybridge_update_wm(struct drm_device
*dev
)
1773 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1774 int latency
= SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
1776 int fbc_wm
, plane_wm
, cursor_wm
;
1777 unsigned int enabled
;
1780 if (g4x_compute_wm0(dev
, 0,
1781 &sandybridge_display_wm_info
, latency
,
1782 &sandybridge_cursor_wm_info
, latency
,
1783 &plane_wm
, &cursor_wm
)) {
1784 val
= I915_READ(WM0_PIPEA_ILK
);
1785 val
&= ~(WM0_PIPE_PLANE_MASK
| WM0_PIPE_CURSOR_MASK
);
1786 I915_WRITE(WM0_PIPEA_ILK
, val
|
1787 ((plane_wm
<< WM0_PIPE_PLANE_SHIFT
) | cursor_wm
));
1788 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1789 " plane %d, " "cursor: %d\n",
1790 plane_wm
, cursor_wm
);
1794 if (g4x_compute_wm0(dev
, 1,
1795 &sandybridge_display_wm_info
, latency
,
1796 &sandybridge_cursor_wm_info
, latency
,
1797 &plane_wm
, &cursor_wm
)) {
1798 val
= I915_READ(WM0_PIPEB_ILK
);
1799 val
&= ~(WM0_PIPE_PLANE_MASK
| WM0_PIPE_CURSOR_MASK
);
1800 I915_WRITE(WM0_PIPEB_ILK
, val
|
1801 ((plane_wm
<< WM0_PIPE_PLANE_SHIFT
) | cursor_wm
));
1802 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1803 " plane %d, cursor: %d\n",
1804 plane_wm
, cursor_wm
);
1808 if ((dev_priv
->num_pipe
== 3) &&
1809 g4x_compute_wm0(dev
, 2,
1810 &sandybridge_display_wm_info
, latency
,
1811 &sandybridge_cursor_wm_info
, latency
,
1812 &plane_wm
, &cursor_wm
)) {
1813 val
= I915_READ(WM0_PIPEC_IVB
);
1814 val
&= ~(WM0_PIPE_PLANE_MASK
| WM0_PIPE_CURSOR_MASK
);
1815 I915_WRITE(WM0_PIPEC_IVB
, val
|
1816 ((plane_wm
<< WM0_PIPE_PLANE_SHIFT
) | cursor_wm
));
1817 DRM_DEBUG_KMS("FIFO watermarks For pipe C -"
1818 " plane %d, cursor: %d\n",
1819 plane_wm
, cursor_wm
);
1824 * Calculate and update the self-refresh watermark only when one
1825 * display plane is used.
1827 * SNB support 3 levels of watermark.
1829 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1830 * and disabled in the descending order
1833 I915_WRITE(WM3_LP_ILK
, 0);
1834 I915_WRITE(WM2_LP_ILK
, 0);
1835 I915_WRITE(WM1_LP_ILK
, 0);
1837 if (!single_plane_enabled(enabled
) ||
1838 dev_priv
->sprite_scaling_enabled
)
1840 enabled
= ffs(enabled
) - 1;
1843 if (!ironlake_compute_srwm(dev
, 1, enabled
,
1844 SNB_READ_WM1_LATENCY() * 500,
1845 &sandybridge_display_srwm_info
,
1846 &sandybridge_cursor_srwm_info
,
1847 &fbc_wm
, &plane_wm
, &cursor_wm
))
1850 I915_WRITE(WM1_LP_ILK
,
1852 (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT
) |
1853 (fbc_wm
<< WM1_LP_FBC_SHIFT
) |
1854 (plane_wm
<< WM1_LP_SR_SHIFT
) |
1858 if (!ironlake_compute_srwm(dev
, 2, enabled
,
1859 SNB_READ_WM2_LATENCY() * 500,
1860 &sandybridge_display_srwm_info
,
1861 &sandybridge_cursor_srwm_info
,
1862 &fbc_wm
, &plane_wm
, &cursor_wm
))
1865 I915_WRITE(WM2_LP_ILK
,
1867 (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT
) |
1868 (fbc_wm
<< WM1_LP_FBC_SHIFT
) |
1869 (plane_wm
<< WM1_LP_SR_SHIFT
) |
1873 if (!ironlake_compute_srwm(dev
, 3, enabled
,
1874 SNB_READ_WM3_LATENCY() * 500,
1875 &sandybridge_display_srwm_info
,
1876 &sandybridge_cursor_srwm_info
,
1877 &fbc_wm
, &plane_wm
, &cursor_wm
))
1880 I915_WRITE(WM3_LP_ILK
,
1882 (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT
) |
1883 (fbc_wm
<< WM1_LP_FBC_SHIFT
) |
1884 (plane_wm
<< WM1_LP_SR_SHIFT
) |
1889 haswell_update_linetime_wm(struct drm_device
*dev
, int pipe
,
1890 struct drm_display_mode
*mode
)
1892 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1895 temp
= I915_READ(PIPE_WM_LINETIME(pipe
));
1896 temp
&= ~PIPE_WM_LINETIME_MASK
;
1898 /* The WM are computed with base on how long it takes to fill a single
1899 * row at the given clock rate, multiplied by 8.
1901 temp
|= PIPE_WM_LINETIME_TIME(
1902 ((mode
->crtc_hdisplay
* 1000) / mode
->clock
) * 8);
1904 /* IPS watermarks are only used by pipe A, and are ignored by
1905 * pipes B and C. They are calculated similarly to the common
1906 * linetime values, except that we are using CD clock frequency
1907 * in MHz instead of pixel rate for the division.
1909 * This is a placeholder for the IPS watermark calculation code.
1912 I915_WRITE(PIPE_WM_LINETIME(pipe
), temp
);
1916 sandybridge_compute_sprite_wm(struct drm_device
*dev
, int plane
,
1917 uint32_t sprite_width
, int pixel_size
,
1918 const struct intel_watermark_params
*display
,
1919 int display_latency_ns
, int *sprite_wm
)
1921 struct drm_crtc
*crtc
;
1923 int entries
, tlb_miss
;
1925 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1926 if (crtc
->fb
== NULL
|| !crtc
->enabled
) {
1927 *sprite_wm
= display
->guard_size
;
1931 clock
= crtc
->mode
.clock
;
1933 /* Use the small buffer method to calculate the sprite watermark */
1934 entries
= ((clock
* pixel_size
/ 1000) * display_latency_ns
) / 1000;
1935 tlb_miss
= display
->fifo_size
*display
->cacheline_size
-
1938 entries
+= tlb_miss
;
1939 entries
= DIV_ROUND_UP(entries
, display
->cacheline_size
);
1940 *sprite_wm
= entries
+ display
->guard_size
;
1941 if (*sprite_wm
> (int)display
->max_wm
)
1942 *sprite_wm
= display
->max_wm
;
1948 sandybridge_compute_sprite_srwm(struct drm_device
*dev
, int plane
,
1949 uint32_t sprite_width
, int pixel_size
,
1950 const struct intel_watermark_params
*display
,
1951 int latency_ns
, int *sprite_wm
)
1953 struct drm_crtc
*crtc
;
1954 unsigned long line_time_us
;
1956 int line_count
, line_size
;
1965 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1966 clock
= crtc
->mode
.clock
;
1972 line_time_us
= (sprite_width
* 1000) / clock
;
1973 if (!line_time_us
) {
1978 line_count
= (latency_ns
/ line_time_us
+ 1000) / 1000;
1979 line_size
= sprite_width
* pixel_size
;
1981 /* Use the minimum of the small and large buffer method for primary */
1982 small
= ((clock
* pixel_size
/ 1000) * latency_ns
) / 1000;
1983 large
= line_count
* line_size
;
1985 entries
= DIV_ROUND_UP(min(small
, large
), display
->cacheline_size
);
1986 *sprite_wm
= entries
+ display
->guard_size
;
1988 return *sprite_wm
> 0x3ff ? false : true;
1991 static void sandybridge_update_sprite_wm(struct drm_device
*dev
, int pipe
,
1992 uint32_t sprite_width
, int pixel_size
)
1994 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1995 int latency
= SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
2002 reg
= WM0_PIPEA_ILK
;
2005 reg
= WM0_PIPEB_ILK
;
2008 reg
= WM0_PIPEC_IVB
;
2011 return; /* bad pipe */
2014 ret
= sandybridge_compute_sprite_wm(dev
, pipe
, sprite_width
, pixel_size
,
2015 &sandybridge_display_wm_info
,
2016 latency
, &sprite_wm
);
2018 DRM_DEBUG_KMS("failed to compute sprite wm for pipe %d\n",
2023 val
= I915_READ(reg
);
2024 val
&= ~WM0_PIPE_SPRITE_MASK
;
2025 I915_WRITE(reg
, val
| (sprite_wm
<< WM0_PIPE_SPRITE_SHIFT
));
2026 DRM_DEBUG_KMS("sprite watermarks For pipe %d - %d\n", pipe
, sprite_wm
);
2029 ret
= sandybridge_compute_sprite_srwm(dev
, pipe
, sprite_width
,
2031 &sandybridge_display_srwm_info
,
2032 SNB_READ_WM1_LATENCY() * 500,
2035 DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %d\n",
2039 I915_WRITE(WM1S_LP_ILK
, sprite_wm
);
2041 /* Only IVB has two more LP watermarks for sprite */
2042 if (!IS_IVYBRIDGE(dev
))
2045 ret
= sandybridge_compute_sprite_srwm(dev
, pipe
, sprite_width
,
2047 &sandybridge_display_srwm_info
,
2048 SNB_READ_WM2_LATENCY() * 500,
2051 DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %d\n",
2055 I915_WRITE(WM2S_LP_IVB
, sprite_wm
);
2057 ret
= sandybridge_compute_sprite_srwm(dev
, pipe
, sprite_width
,
2059 &sandybridge_display_srwm_info
,
2060 SNB_READ_WM3_LATENCY() * 500,
2063 DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %d\n",
2067 I915_WRITE(WM3S_LP_IVB
, sprite_wm
);
2071 * intel_update_watermarks - update FIFO watermark values based on current modes
2073 * Calculate watermark values for the various WM regs based on current mode
2074 * and plane configuration.
2076 * There are several cases to deal with here:
2077 * - normal (i.e. non-self-refresh)
2078 * - self-refresh (SR) mode
2079 * - lines are large relative to FIFO size (buffer can hold up to 2)
2080 * - lines are small relative to FIFO size (buffer can hold more than 2
2081 * lines), so need to account for TLB latency
2083 * The normal calculation is:
2084 * watermark = dotclock * bytes per pixel * latency
2085 * where latency is platform & configuration dependent (we assume pessimal
2088 * The SR calculation is:
2089 * watermark = (trunc(latency/line time)+1) * surface width *
2092 * line time = htotal / dotclock
2093 * surface width = hdisplay for normal plane and 64 for cursor
2094 * and latency is assumed to be high, as above.
2096 * The final value programmed to the register should always be rounded up,
2097 * and include an extra 2 entries to account for clock crossings.
2099 * We don't use the sprite, so we can ignore that. And on Crestline we have
2100 * to set the non-SR watermarks to 8.
2102 void intel_update_watermarks(struct drm_device
*dev
)
2104 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2106 if (dev_priv
->display
.update_wm
)
2107 dev_priv
->display
.update_wm(dev
);
2110 void intel_update_linetime_watermarks(struct drm_device
*dev
,
2111 int pipe
, struct drm_display_mode
*mode
)
2113 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2115 if (dev_priv
->display
.update_linetime_wm
)
2116 dev_priv
->display
.update_linetime_wm(dev
, pipe
, mode
);
2119 void intel_update_sprite_watermarks(struct drm_device
*dev
, int pipe
,
2120 uint32_t sprite_width
, int pixel_size
)
2122 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2124 if (dev_priv
->display
.update_sprite_wm
)
2125 dev_priv
->display
.update_sprite_wm(dev
, pipe
, sprite_width
,
2129 static struct drm_i915_gem_object
*
2130 intel_alloc_context_page(struct drm_device
*dev
)
2132 struct drm_i915_gem_object
*ctx
;
2135 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
2137 ctx
= i915_gem_alloc_object(dev
, 4096);
2139 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2143 ret
= i915_gem_object_pin(ctx
, 4096, true, false);
2145 DRM_ERROR("failed to pin power context: %d\n", ret
);
2149 ret
= i915_gem_object_set_to_gtt_domain(ctx
, 1);
2151 DRM_ERROR("failed to set-domain on power context: %d\n", ret
);
2158 i915_gem_object_unpin(ctx
);
2160 drm_gem_object_unreference(&ctx
->base
);
2161 mutex_unlock(&dev
->struct_mutex
);
2166 * Lock protecting IPS related data structures
2168 DEFINE_SPINLOCK(mchdev_lock
);
2170 /* Global for IPS driver to get at the current i915 device. Protected by
2172 static struct drm_i915_private
*i915_mch_dev
;
2174 bool ironlake_set_drps(struct drm_device
*dev
, u8 val
)
2176 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2179 assert_spin_locked(&mchdev_lock
);
2181 rgvswctl
= I915_READ16(MEMSWCTL
);
2182 if (rgvswctl
& MEMCTL_CMD_STS
) {
2183 DRM_DEBUG("gpu busy, RCS change rejected\n");
2184 return false; /* still busy with another command */
2187 rgvswctl
= (MEMCTL_CMD_CHFREQ
<< MEMCTL_CMD_SHIFT
) |
2188 (val
<< MEMCTL_FREQ_SHIFT
) | MEMCTL_SFCAVM
;
2189 I915_WRITE16(MEMSWCTL
, rgvswctl
);
2190 POSTING_READ16(MEMSWCTL
);
2192 rgvswctl
|= MEMCTL_CMD_STS
;
2193 I915_WRITE16(MEMSWCTL
, rgvswctl
);
2198 static void ironlake_enable_drps(struct drm_device
*dev
)
2200 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2201 u32 rgvmodectl
= I915_READ(MEMMODECTL
);
2202 u8 fmax
, fmin
, fstart
, vstart
;
2204 spin_lock_irq(&mchdev_lock
);
2206 /* Enable temp reporting */
2207 I915_WRITE16(PMMISC
, I915_READ(PMMISC
) | MCPPCE_EN
);
2208 I915_WRITE16(TSC1
, I915_READ(TSC1
) | TSE
);
2210 /* 100ms RC evaluation intervals */
2211 I915_WRITE(RCUPEI
, 100000);
2212 I915_WRITE(RCDNEI
, 100000);
2214 /* Set max/min thresholds to 90ms and 80ms respectively */
2215 I915_WRITE(RCBMAXAVG
, 90000);
2216 I915_WRITE(RCBMINAVG
, 80000);
2218 I915_WRITE(MEMIHYST
, 1);
2220 /* Set up min, max, and cur for interrupt handling */
2221 fmax
= (rgvmodectl
& MEMMODE_FMAX_MASK
) >> MEMMODE_FMAX_SHIFT
;
2222 fmin
= (rgvmodectl
& MEMMODE_FMIN_MASK
);
2223 fstart
= (rgvmodectl
& MEMMODE_FSTART_MASK
) >>
2224 MEMMODE_FSTART_SHIFT
;
2226 vstart
= (I915_READ(PXVFREQ_BASE
+ (fstart
* 4)) & PXVFREQ_PX_MASK
) >>
2229 dev_priv
->ips
.fmax
= fmax
; /* IPS callback will increase this */
2230 dev_priv
->ips
.fstart
= fstart
;
2232 dev_priv
->ips
.max_delay
= fstart
;
2233 dev_priv
->ips
.min_delay
= fmin
;
2234 dev_priv
->ips
.cur_delay
= fstart
;
2236 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2237 fmax
, fmin
, fstart
);
2239 I915_WRITE(MEMINTREN
, MEMINT_CX_SUPR_EN
| MEMINT_EVAL_CHG_EN
);
2242 * Interrupts will be enabled in ironlake_irq_postinstall
2245 I915_WRITE(VIDSTART
, vstart
);
2246 POSTING_READ(VIDSTART
);
2248 rgvmodectl
|= MEMMODE_SWMODE_EN
;
2249 I915_WRITE(MEMMODECTL
, rgvmodectl
);
2251 if (wait_for_atomic((I915_READ(MEMSWCTL
) & MEMCTL_CMD_STS
) == 0, 10))
2252 DRM_ERROR("stuck trying to change perf mode\n");
2255 ironlake_set_drps(dev
, fstart
);
2257 dev_priv
->ips
.last_count1
= I915_READ(0x112e4) + I915_READ(0x112e8) +
2259 dev_priv
->ips
.last_time1
= jiffies_to_msecs(jiffies
);
2260 dev_priv
->ips
.last_count2
= I915_READ(0x112f4);
2261 getrawmonotonic(&dev_priv
->ips
.last_time2
);
2263 spin_unlock_irq(&mchdev_lock
);
2266 static void ironlake_disable_drps(struct drm_device
*dev
)
2268 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2271 spin_lock_irq(&mchdev_lock
);
2273 rgvswctl
= I915_READ16(MEMSWCTL
);
2275 /* Ack interrupts, disable EFC interrupt */
2276 I915_WRITE(MEMINTREN
, I915_READ(MEMINTREN
) & ~MEMINT_EVAL_CHG_EN
);
2277 I915_WRITE(MEMINTRSTS
, MEMINT_EVAL_CHG
);
2278 I915_WRITE(DEIER
, I915_READ(DEIER
) & ~DE_PCU_EVENT
);
2279 I915_WRITE(DEIIR
, DE_PCU_EVENT
);
2280 I915_WRITE(DEIMR
, I915_READ(DEIMR
) | DE_PCU_EVENT
);
2282 /* Go back to the starting frequency */
2283 ironlake_set_drps(dev
, dev_priv
->ips
.fstart
);
2285 rgvswctl
|= MEMCTL_CMD_STS
;
2286 I915_WRITE(MEMSWCTL
, rgvswctl
);
2289 spin_unlock_irq(&mchdev_lock
);
2292 /* There's a funny hw issue where the hw returns all 0 when reading from
2293 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
2294 * ourselves, instead of doing a rmw cycle (which might result in us clearing
2295 * all limits and the gpu stuck at whatever frequency it is at atm).
2297 static u32
gen6_rps_limits(struct drm_i915_private
*dev_priv
, u8
*val
)
2303 if (*val
>= dev_priv
->rps
.max_delay
)
2304 *val
= dev_priv
->rps
.max_delay
;
2305 limits
|= dev_priv
->rps
.max_delay
<< 24;
2307 /* Only set the down limit when we've reached the lowest level to avoid
2308 * getting more interrupts, otherwise leave this clear. This prevents a
2309 * race in the hw when coming out of rc6: There's a tiny window where
2310 * the hw runs at the minimal clock before selecting the desired
2311 * frequency, if the down threshold expires in that window we will not
2312 * receive a down interrupt. */
2313 if (*val
<= dev_priv
->rps
.min_delay
) {
2314 *val
= dev_priv
->rps
.min_delay
;
2315 limits
|= dev_priv
->rps
.min_delay
<< 16;
2321 void gen6_set_rps(struct drm_device
*dev
, u8 val
)
2323 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2324 u32 limits
= gen6_rps_limits(dev_priv
, &val
);
2326 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
2328 if (val
== dev_priv
->rps
.cur_delay
)
2331 I915_WRITE(GEN6_RPNSWREQ
,
2332 GEN6_FREQUENCY(val
) |
2334 GEN6_AGGRESSIVE_TURBO
);
2336 /* Make sure we continue to get interrupts
2337 * until we hit the minimum or maximum frequencies.
2339 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
, limits
);
2341 POSTING_READ(GEN6_RPNSWREQ
);
2343 dev_priv
->rps
.cur_delay
= val
;
2345 trace_intel_gpu_freq_change(val
* 50);
2348 static void gen6_disable_rps(struct drm_device
*dev
)
2350 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2352 I915_WRITE(GEN6_RC_CONTROL
, 0);
2353 I915_WRITE(GEN6_RPNSWREQ
, 1 << 31);
2354 I915_WRITE(GEN6_PMINTRMSK
, 0xffffffff);
2355 I915_WRITE(GEN6_PMIER
, 0);
2356 /* Complete PM interrupt masking here doesn't race with the rps work
2357 * item again unmasking PM interrupts because that is using a different
2358 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
2359 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
2361 spin_lock_irq(&dev_priv
->rps
.lock
);
2362 dev_priv
->rps
.pm_iir
= 0;
2363 spin_unlock_irq(&dev_priv
->rps
.lock
);
2365 I915_WRITE(GEN6_PMIIR
, I915_READ(GEN6_PMIIR
));
2368 int intel_enable_rc6(const struct drm_device
*dev
)
2370 /* Respect the kernel parameter if it is set */
2371 if (i915_enable_rc6
>= 0)
2372 return i915_enable_rc6
;
2374 if (INTEL_INFO(dev
)->gen
== 5) {
2375 #ifdef CONFIG_INTEL_IOMMU
2376 /* Disable rc6 on ilk if VT-d is on. */
2377 if (intel_iommu_gfx_mapped
)
2380 DRM_DEBUG_DRIVER("Ironlake: only RC6 available\n");
2381 return INTEL_RC6_ENABLE
;
2384 if (IS_HASWELL(dev
)) {
2385 DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
2386 return INTEL_RC6_ENABLE
;
2389 /* snb/ivb have more than one rc6 state. */
2390 if (INTEL_INFO(dev
)->gen
== 6) {
2391 DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
2392 return INTEL_RC6_ENABLE
;
2395 DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n");
2396 return (INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
);
2399 static void gen6_enable_rps(struct drm_device
*dev
)
2401 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2402 struct intel_ring_buffer
*ring
;
2405 u32 pcu_mbox
, rc6_mask
= 0;
2410 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
2412 /* Here begins a magic sequence of register writes to enable
2413 * auto-downclocking.
2415 * Perhaps there might be some value in exposing these to
2418 I915_WRITE(GEN6_RC_STATE
, 0);
2420 /* Clear the DBG now so we don't confuse earlier errors */
2421 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
2422 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg
);
2423 I915_WRITE(GTFIFODBG
, gtfifodbg
);
2426 gen6_gt_force_wake_get(dev_priv
);
2428 rp_state_cap
= I915_READ(GEN6_RP_STATE_CAP
);
2429 gt_perf_status
= I915_READ(GEN6_GT_PERF_STATUS
);
2431 /* In units of 100MHz */
2432 dev_priv
->rps
.max_delay
= rp_state_cap
& 0xff;
2433 dev_priv
->rps
.min_delay
= (rp_state_cap
& 0xff0000) >> 16;
2434 dev_priv
->rps
.cur_delay
= 0;
2436 /* disable the counters and set deterministic thresholds */
2437 I915_WRITE(GEN6_RC_CONTROL
, 0);
2439 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT
, 1000 << 16);
2440 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16 | 30);
2441 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT
, 30);
2442 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
2443 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
2445 for_each_ring(ring
, dev_priv
, i
)
2446 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
2448 I915_WRITE(GEN6_RC_SLEEP
, 0);
2449 I915_WRITE(GEN6_RC1e_THRESHOLD
, 1000);
2450 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000);
2451 I915_WRITE(GEN6_RC6p_THRESHOLD
, 100000);
2452 I915_WRITE(GEN6_RC6pp_THRESHOLD
, 64000); /* unused */
2454 /* Check if we are enabling RC6 */
2455 rc6_mode
= intel_enable_rc6(dev_priv
->dev
);
2456 if (rc6_mode
& INTEL_RC6_ENABLE
)
2457 rc6_mask
|= GEN6_RC_CTL_RC6_ENABLE
;
2459 /* We don't use those on Haswell */
2460 if (!IS_HASWELL(dev
)) {
2461 if (rc6_mode
& INTEL_RC6p_ENABLE
)
2462 rc6_mask
|= GEN6_RC_CTL_RC6p_ENABLE
;
2464 if (rc6_mode
& INTEL_RC6pp_ENABLE
)
2465 rc6_mask
|= GEN6_RC_CTL_RC6pp_ENABLE
;
2468 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
2469 (rc6_mask
& GEN6_RC_CTL_RC6_ENABLE
) ? "on" : "off",
2470 (rc6_mask
& GEN6_RC_CTL_RC6p_ENABLE
) ? "on" : "off",
2471 (rc6_mask
& GEN6_RC_CTL_RC6pp_ENABLE
) ? "on" : "off");
2473 I915_WRITE(GEN6_RC_CONTROL
,
2475 GEN6_RC_CTL_EI_MODE(1) |
2476 GEN6_RC_CTL_HW_ENABLE
);
2478 I915_WRITE(GEN6_RPNSWREQ
,
2479 GEN6_FREQUENCY(10) |
2481 GEN6_AGGRESSIVE_TURBO
);
2482 I915_WRITE(GEN6_RC_VIDEO_FREQ
,
2483 GEN6_FREQUENCY(12));
2485 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 1000000);
2486 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
,
2487 dev_priv
->rps
.max_delay
<< 24 |
2488 dev_priv
->rps
.min_delay
<< 16);
2490 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 59400);
2491 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 245000);
2492 I915_WRITE(GEN6_RP_UP_EI
, 66000);
2493 I915_WRITE(GEN6_RP_DOWN_EI
, 350000);
2495 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
2496 I915_WRITE(GEN6_RP_CONTROL
,
2497 GEN6_RP_MEDIA_TURBO
|
2498 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
2499 GEN6_RP_MEDIA_IS_GFX
|
2501 GEN6_RP_UP_BUSY_AVG
|
2502 (IS_HASWELL(dev
) ? GEN7_RP_DOWN_IDLE_AVG
: GEN6_RP_DOWN_IDLE_CONT
));
2504 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
2506 DRM_ERROR("timeout waiting for pcode mailbox to become idle\n");
2508 I915_WRITE(GEN6_PCODE_DATA
, 0);
2509 I915_WRITE(GEN6_PCODE_MAILBOX
,
2511 GEN6_PCODE_WRITE_MIN_FREQ_TABLE
);
2512 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
2514 DRM_ERROR("timeout waiting for pcode mailbox to finish\n");
2516 /* Check for overclock support */
2517 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
2519 DRM_ERROR("timeout waiting for pcode mailbox to become idle\n");
2520 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_READ_OC_PARAMS
);
2521 pcu_mbox
= I915_READ(GEN6_PCODE_DATA
);
2522 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
2524 DRM_ERROR("timeout waiting for pcode mailbox to finish\n");
2525 if (pcu_mbox
& (1<<31)) { /* OC supported */
2526 dev_priv
->rps
.max_delay
= pcu_mbox
& 0xff;
2527 DRM_DEBUG_DRIVER("overclocking supported, adjusting frequency max to %dMHz\n", pcu_mbox
* 50);
2530 gen6_set_rps(dev_priv
->dev
, (gt_perf_status
& 0xff00) >> 8);
2532 /* requires MSI enabled */
2533 I915_WRITE(GEN6_PMIER
, GEN6_PM_DEFERRED_EVENTS
);
2534 spin_lock_irq(&dev_priv
->rps
.lock
);
2535 WARN_ON(dev_priv
->rps
.pm_iir
!= 0);
2536 I915_WRITE(GEN6_PMIMR
, 0);
2537 spin_unlock_irq(&dev_priv
->rps
.lock
);
2538 /* enable all PM interrupts */
2539 I915_WRITE(GEN6_PMINTRMSK
, 0);
2541 gen6_gt_force_wake_put(dev_priv
);
2544 static void gen6_update_ring_freq(struct drm_device
*dev
)
2546 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2548 int gpu_freq
, ia_freq
, max_ia_freq
;
2549 int scaling_factor
= 180;
2551 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
2553 max_ia_freq
= cpufreq_quick_get_max(0);
2555 * Default to measured freq if none found, PCU will ensure we don't go
2559 max_ia_freq
= tsc_khz
;
2561 /* Convert from kHz to MHz */
2562 max_ia_freq
/= 1000;
2565 * For each potential GPU frequency, load a ring frequency we'd like
2566 * to use for memory access. We do this by specifying the IA frequency
2567 * the PCU should use as a reference to determine the ring frequency.
2569 for (gpu_freq
= dev_priv
->rps
.max_delay
; gpu_freq
>= dev_priv
->rps
.min_delay
;
2571 int diff
= dev_priv
->rps
.max_delay
- gpu_freq
;
2574 * For GPU frequencies less than 750MHz, just use the lowest
2577 if (gpu_freq
< min_freq
)
2580 ia_freq
= max_ia_freq
- ((diff
* scaling_factor
) / 2);
2581 ia_freq
= DIV_ROUND_CLOSEST(ia_freq
, 100);
2583 I915_WRITE(GEN6_PCODE_DATA
,
2584 (ia_freq
<< GEN6_PCODE_FREQ_IA_RATIO_SHIFT
) |
2586 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
|
2587 GEN6_PCODE_WRITE_MIN_FREQ_TABLE
);
2588 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) &
2589 GEN6_PCODE_READY
) == 0, 10)) {
2590 DRM_ERROR("pcode write of freq table timed out\n");
2596 void ironlake_teardown_rc6(struct drm_device
*dev
)
2598 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2600 if (dev_priv
->renderctx
) {
2601 i915_gem_object_unpin(dev_priv
->renderctx
);
2602 drm_gem_object_unreference(&dev_priv
->renderctx
->base
);
2603 dev_priv
->renderctx
= NULL
;
2606 if (dev_priv
->pwrctx
) {
2607 i915_gem_object_unpin(dev_priv
->pwrctx
);
2608 drm_gem_object_unreference(&dev_priv
->pwrctx
->base
);
2609 dev_priv
->pwrctx
= NULL
;
2613 static void ironlake_disable_rc6(struct drm_device
*dev
)
2615 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2617 if (I915_READ(PWRCTXA
)) {
2618 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
2619 I915_WRITE(RSTDBYCTL
, I915_READ(RSTDBYCTL
) | RCX_SW_EXIT
);
2620 wait_for(((I915_READ(RSTDBYCTL
) & RSX_STATUS_MASK
) == RSX_STATUS_ON
),
2623 I915_WRITE(PWRCTXA
, 0);
2624 POSTING_READ(PWRCTXA
);
2626 I915_WRITE(RSTDBYCTL
, I915_READ(RSTDBYCTL
) & ~RCX_SW_EXIT
);
2627 POSTING_READ(RSTDBYCTL
);
2631 static int ironlake_setup_rc6(struct drm_device
*dev
)
2633 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2635 if (dev_priv
->renderctx
== NULL
)
2636 dev_priv
->renderctx
= intel_alloc_context_page(dev
);
2637 if (!dev_priv
->renderctx
)
2640 if (dev_priv
->pwrctx
== NULL
)
2641 dev_priv
->pwrctx
= intel_alloc_context_page(dev
);
2642 if (!dev_priv
->pwrctx
) {
2643 ironlake_teardown_rc6(dev
);
2650 static void ironlake_enable_rc6(struct drm_device
*dev
)
2652 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2653 struct intel_ring_buffer
*ring
= &dev_priv
->ring
[RCS
];
2656 /* rc6 disabled by default due to repeated reports of hanging during
2659 if (!intel_enable_rc6(dev
))
2662 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
2664 ret
= ironlake_setup_rc6(dev
);
2669 * GPU can automatically power down the render unit if given a page
2672 ret
= intel_ring_begin(ring
, 6);
2674 ironlake_teardown_rc6(dev
);
2678 intel_ring_emit(ring
, MI_SUSPEND_FLUSH
| MI_SUSPEND_FLUSH_EN
);
2679 intel_ring_emit(ring
, MI_SET_CONTEXT
);
2680 intel_ring_emit(ring
, dev_priv
->renderctx
->gtt_offset
|
2682 MI_SAVE_EXT_STATE_EN
|
2683 MI_RESTORE_EXT_STATE_EN
|
2684 MI_RESTORE_INHIBIT
);
2685 intel_ring_emit(ring
, MI_SUSPEND_FLUSH
);
2686 intel_ring_emit(ring
, MI_NOOP
);
2687 intel_ring_emit(ring
, MI_FLUSH
);
2688 intel_ring_advance(ring
);
2691 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
2692 * does an implicit flush, combined with MI_FLUSH above, it should be
2693 * safe to assume that renderctx is valid
2695 ret
= intel_wait_ring_idle(ring
);
2697 DRM_ERROR("failed to enable ironlake power power savings\n");
2698 ironlake_teardown_rc6(dev
);
2702 I915_WRITE(PWRCTXA
, dev_priv
->pwrctx
->gtt_offset
| PWRCTX_EN
);
2703 I915_WRITE(RSTDBYCTL
, I915_READ(RSTDBYCTL
) & ~RCX_SW_EXIT
);
2706 static unsigned long intel_pxfreq(u32 vidfreq
)
2709 int div
= (vidfreq
& 0x3f0000) >> 16;
2710 int post
= (vidfreq
& 0x3000) >> 12;
2711 int pre
= (vidfreq
& 0x7);
2716 freq
= ((div
* 133333) / ((1<<post
) * pre
));
2721 static const struct cparams
{
2727 { 1, 1333, 301, 28664 },
2728 { 1, 1066, 294, 24460 },
2729 { 1, 800, 294, 25192 },
2730 { 0, 1333, 276, 27605 },
2731 { 0, 1066, 276, 27605 },
2732 { 0, 800, 231, 23784 },
2735 unsigned long i915_chipset_val(struct drm_i915_private
*dev_priv
)
2737 u64 total_count
, diff
, ret
;
2738 u32 count1
, count2
, count3
, m
= 0, c
= 0;
2739 unsigned long now
= jiffies_to_msecs(jiffies
), diff1
;
2742 assert_spin_locked(&mchdev_lock
);
2744 diff1
= now
- dev_priv
->ips
.last_time1
;
2746 /* Prevent division-by-zero if we are asking too fast.
2747 * Also, we don't get interesting results if we are polling
2748 * faster than once in 10ms, so just return the saved value
2752 return dev_priv
->ips
.chipset_power
;
2754 count1
= I915_READ(DMIEC
);
2755 count2
= I915_READ(DDREC
);
2756 count3
= I915_READ(CSIEC
);
2758 total_count
= count1
+ count2
+ count3
;
2760 /* FIXME: handle per-counter overflow */
2761 if (total_count
< dev_priv
->ips
.last_count1
) {
2762 diff
= ~0UL - dev_priv
->ips
.last_count1
;
2763 diff
+= total_count
;
2765 diff
= total_count
- dev_priv
->ips
.last_count1
;
2768 for (i
= 0; i
< ARRAY_SIZE(cparams
); i
++) {
2769 if (cparams
[i
].i
== dev_priv
->ips
.c_m
&&
2770 cparams
[i
].t
== dev_priv
->ips
.r_t
) {
2777 diff
= div_u64(diff
, diff1
);
2778 ret
= ((m
* diff
) + c
);
2779 ret
= div_u64(ret
, 10);
2781 dev_priv
->ips
.last_count1
= total_count
;
2782 dev_priv
->ips
.last_time1
= now
;
2784 dev_priv
->ips
.chipset_power
= ret
;
2789 unsigned long i915_mch_val(struct drm_i915_private
*dev_priv
)
2791 unsigned long m
, x
, b
;
2794 tsfs
= I915_READ(TSFS
);
2796 m
= ((tsfs
& TSFS_SLOPE_MASK
) >> TSFS_SLOPE_SHIFT
);
2797 x
= I915_READ8(TR1
);
2799 b
= tsfs
& TSFS_INTR_MASK
;
2801 return ((m
* x
) / 127) - b
;
2804 static u16
pvid_to_extvid(struct drm_i915_private
*dev_priv
, u8 pxvid
)
2806 static const struct v_table
{
2807 u16 vd
; /* in .1 mil */
2808 u16 vm
; /* in .1 mil */
2939 if (dev_priv
->info
->is_mobile
)
2940 return v_table
[pxvid
].vm
;
2942 return v_table
[pxvid
].vd
;
2945 static void __i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
2947 struct timespec now
, diff1
;
2949 unsigned long diffms
;
2952 assert_spin_locked(&mchdev_lock
);
2954 getrawmonotonic(&now
);
2955 diff1
= timespec_sub(now
, dev_priv
->ips
.last_time2
);
2957 /* Don't divide by 0 */
2958 diffms
= diff1
.tv_sec
* 1000 + diff1
.tv_nsec
/ 1000000;
2962 count
= I915_READ(GFXEC
);
2964 if (count
< dev_priv
->ips
.last_count2
) {
2965 diff
= ~0UL - dev_priv
->ips
.last_count2
;
2968 diff
= count
- dev_priv
->ips
.last_count2
;
2971 dev_priv
->ips
.last_count2
= count
;
2972 dev_priv
->ips
.last_time2
= now
;
2974 /* More magic constants... */
2976 diff
= div_u64(diff
, diffms
* 10);
2977 dev_priv
->ips
.gfx_power
= diff
;
2980 void i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
2982 if (dev_priv
->info
->gen
!= 5)
2985 spin_lock_irq(&mchdev_lock
);
2987 __i915_update_gfx_val(dev_priv
);
2989 spin_unlock_irq(&mchdev_lock
);
2992 unsigned long i915_gfx_val(struct drm_i915_private
*dev_priv
)
2994 unsigned long t
, corr
, state1
, corr2
, state2
;
2997 assert_spin_locked(&mchdev_lock
);
2999 pxvid
= I915_READ(PXVFREQ_BASE
+ (dev_priv
->rps
.cur_delay
* 4));
3000 pxvid
= (pxvid
>> 24) & 0x7f;
3001 ext_v
= pvid_to_extvid(dev_priv
, pxvid
);
3005 t
= i915_mch_val(dev_priv
);
3007 /* Revel in the empirically derived constants */
3009 /* Correction factor in 1/100000 units */
3011 corr
= ((t
* 2349) + 135940);
3013 corr
= ((t
* 964) + 29317);
3015 corr
= ((t
* 301) + 1004);
3017 corr
= corr
* ((150142 * state1
) / 10000 - 78642);
3019 corr2
= (corr
* dev_priv
->ips
.corr
);
3021 state2
= (corr2
* state1
) / 10000;
3022 state2
/= 100; /* convert to mW */
3024 __i915_update_gfx_val(dev_priv
);
3026 return dev_priv
->ips
.gfx_power
+ state2
;
3030 * i915_read_mch_val - return value for IPS use
3032 * Calculate and return a value for the IPS driver to use when deciding whether
3033 * we have thermal and power headroom to increase CPU or GPU power budget.
3035 unsigned long i915_read_mch_val(void)
3037 struct drm_i915_private
*dev_priv
;
3038 unsigned long chipset_val
, graphics_val
, ret
= 0;
3040 spin_lock_irq(&mchdev_lock
);
3043 dev_priv
= i915_mch_dev
;
3045 chipset_val
= i915_chipset_val(dev_priv
);
3046 graphics_val
= i915_gfx_val(dev_priv
);
3048 ret
= chipset_val
+ graphics_val
;
3051 spin_unlock_irq(&mchdev_lock
);
3055 EXPORT_SYMBOL_GPL(i915_read_mch_val
);
3058 * i915_gpu_raise - raise GPU frequency limit
3060 * Raise the limit; IPS indicates we have thermal headroom.
3062 bool i915_gpu_raise(void)
3064 struct drm_i915_private
*dev_priv
;
3067 spin_lock_irq(&mchdev_lock
);
3068 if (!i915_mch_dev
) {
3072 dev_priv
= i915_mch_dev
;
3074 if (dev_priv
->ips
.max_delay
> dev_priv
->ips
.fmax
)
3075 dev_priv
->ips
.max_delay
--;
3078 spin_unlock_irq(&mchdev_lock
);
3082 EXPORT_SYMBOL_GPL(i915_gpu_raise
);
3085 * i915_gpu_lower - lower GPU frequency limit
3087 * IPS indicates we're close to a thermal limit, so throttle back the GPU
3088 * frequency maximum.
3090 bool i915_gpu_lower(void)
3092 struct drm_i915_private
*dev_priv
;
3095 spin_lock_irq(&mchdev_lock
);
3096 if (!i915_mch_dev
) {
3100 dev_priv
= i915_mch_dev
;
3102 if (dev_priv
->ips
.max_delay
< dev_priv
->ips
.min_delay
)
3103 dev_priv
->ips
.max_delay
++;
3106 spin_unlock_irq(&mchdev_lock
);
3110 EXPORT_SYMBOL_GPL(i915_gpu_lower
);
3113 * i915_gpu_busy - indicate GPU business to IPS
3115 * Tell the IPS driver whether or not the GPU is busy.
3117 bool i915_gpu_busy(void)
3119 struct drm_i915_private
*dev_priv
;
3120 struct intel_ring_buffer
*ring
;
3124 spin_lock_irq(&mchdev_lock
);
3127 dev_priv
= i915_mch_dev
;
3129 for_each_ring(ring
, dev_priv
, i
)
3130 ret
|= !list_empty(&ring
->request_list
);
3133 spin_unlock_irq(&mchdev_lock
);
3137 EXPORT_SYMBOL_GPL(i915_gpu_busy
);
3140 * i915_gpu_turbo_disable - disable graphics turbo
3142 * Disable graphics turbo by resetting the max frequency and setting the
3143 * current frequency to the default.
3145 bool i915_gpu_turbo_disable(void)
3147 struct drm_i915_private
*dev_priv
;
3150 spin_lock_irq(&mchdev_lock
);
3151 if (!i915_mch_dev
) {
3155 dev_priv
= i915_mch_dev
;
3157 dev_priv
->ips
.max_delay
= dev_priv
->ips
.fstart
;
3159 if (!ironlake_set_drps(dev_priv
->dev
, dev_priv
->ips
.fstart
))
3163 spin_unlock_irq(&mchdev_lock
);
3167 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable
);
3170 * Tells the intel_ips driver that the i915 driver is now loaded, if
3171 * IPS got loaded first.
3173 * This awkward dance is so that neither module has to depend on the
3174 * other in order for IPS to do the appropriate communication of
3175 * GPU turbo limits to i915.
3178 ips_ping_for_i915_load(void)
3182 link
= symbol_get(ips_link_to_i915_driver
);
3185 symbol_put(ips_link_to_i915_driver
);
3189 void intel_gpu_ips_init(struct drm_i915_private
*dev_priv
)
3191 /* We only register the i915 ips part with intel-ips once everything is
3192 * set up, to avoid intel-ips sneaking in and reading bogus values. */
3193 spin_lock_irq(&mchdev_lock
);
3194 i915_mch_dev
= dev_priv
;
3195 spin_unlock_irq(&mchdev_lock
);
3197 ips_ping_for_i915_load();
3200 void intel_gpu_ips_teardown(void)
3202 spin_lock_irq(&mchdev_lock
);
3203 i915_mch_dev
= NULL
;
3204 spin_unlock_irq(&mchdev_lock
);
3206 static void intel_init_emon(struct drm_device
*dev
)
3208 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3213 /* Disable to program */
3217 /* Program energy weights for various events */
3218 I915_WRITE(SDEW
, 0x15040d00);
3219 I915_WRITE(CSIEW0
, 0x007f0000);
3220 I915_WRITE(CSIEW1
, 0x1e220004);
3221 I915_WRITE(CSIEW2
, 0x04000004);
3223 for (i
= 0; i
< 5; i
++)
3224 I915_WRITE(PEW
+ (i
* 4), 0);
3225 for (i
= 0; i
< 3; i
++)
3226 I915_WRITE(DEW
+ (i
* 4), 0);
3228 /* Program P-state weights to account for frequency power adjustment */
3229 for (i
= 0; i
< 16; i
++) {
3230 u32 pxvidfreq
= I915_READ(PXVFREQ_BASE
+ (i
* 4));
3231 unsigned long freq
= intel_pxfreq(pxvidfreq
);
3232 unsigned long vid
= (pxvidfreq
& PXVFREQ_PX_MASK
) >>
3237 val
*= (freq
/ 1000);
3239 val
/= (127*127*900);
3241 DRM_ERROR("bad pxval: %ld\n", val
);
3244 /* Render standby states get 0 weight */
3248 for (i
= 0; i
< 4; i
++) {
3249 u32 val
= (pxw
[i
*4] << 24) | (pxw
[(i
*4)+1] << 16) |
3250 (pxw
[(i
*4)+2] << 8) | (pxw
[(i
*4)+3]);
3251 I915_WRITE(PXW
+ (i
* 4), val
);
3254 /* Adjust magic regs to magic values (more experimental results) */
3255 I915_WRITE(OGW0
, 0);
3256 I915_WRITE(OGW1
, 0);
3257 I915_WRITE(EG0
, 0x00007f00);
3258 I915_WRITE(EG1
, 0x0000000e);
3259 I915_WRITE(EG2
, 0x000e0000);
3260 I915_WRITE(EG3
, 0x68000300);
3261 I915_WRITE(EG4
, 0x42000000);
3262 I915_WRITE(EG5
, 0x00140031);
3266 for (i
= 0; i
< 8; i
++)
3267 I915_WRITE(PXWL
+ (i
* 4), 0);
3269 /* Enable PMON + select events */
3270 I915_WRITE(ECR
, 0x80000019);
3272 lcfuse
= I915_READ(LCFUSE02
);
3274 dev_priv
->ips
.corr
= (lcfuse
& LCFUSE_HIV_MASK
);
3277 void intel_disable_gt_powersave(struct drm_device
*dev
)
3279 if (IS_IRONLAKE_M(dev
)) {
3280 ironlake_disable_drps(dev
);
3281 ironlake_disable_rc6(dev
);
3282 } else if (INTEL_INFO(dev
)->gen
>= 6 && !IS_VALLEYVIEW(dev
)) {
3283 gen6_disable_rps(dev
);
3287 void intel_enable_gt_powersave(struct drm_device
*dev
)
3289 if (IS_IRONLAKE_M(dev
)) {
3290 ironlake_enable_drps(dev
);
3291 ironlake_enable_rc6(dev
);
3292 intel_init_emon(dev
);
3293 } else if ((IS_GEN6(dev
) || IS_GEN7(dev
)) && !IS_VALLEYVIEW(dev
)) {
3294 gen6_enable_rps(dev
);
3295 gen6_update_ring_freq(dev
);
3299 static void ironlake_init_clock_gating(struct drm_device
*dev
)
3301 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3302 uint32_t dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
;
3304 /* Required for FBC */
3305 dspclk_gate
|= DPFCUNIT_CLOCK_GATE_DISABLE
|
3306 DPFCRUNIT_CLOCK_GATE_DISABLE
|
3307 DPFDUNIT_CLOCK_GATE_DISABLE
;
3308 /* Required for CxSR */
3309 dspclk_gate
|= DPARBUNIT_CLOCK_GATE_DISABLE
;
3311 I915_WRITE(PCH_3DCGDIS0
,
3312 MARIUNIT_CLOCK_GATE_DISABLE
|
3313 SVSMUNIT_CLOCK_GATE_DISABLE
);
3314 I915_WRITE(PCH_3DCGDIS1
,
3315 VFMUNIT_CLOCK_GATE_DISABLE
);
3317 I915_WRITE(PCH_DSPCLK_GATE_D
, dspclk_gate
);
3320 * According to the spec the following bits should be set in
3321 * order to enable memory self-refresh
3322 * The bit 22/21 of 0x42004
3323 * The bit 5 of 0x42020
3324 * The bit 15 of 0x45000
3326 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
3327 (I915_READ(ILK_DISPLAY_CHICKEN2
) |
3328 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
));
3329 I915_WRITE(ILK_DSPCLK_GATE
,
3330 (I915_READ(ILK_DSPCLK_GATE
) |
3331 ILK_DPARB_CLK_GATE
));
3332 I915_WRITE(DISP_ARB_CTL
,
3333 (I915_READ(DISP_ARB_CTL
) |
3335 I915_WRITE(WM3_LP_ILK
, 0);
3336 I915_WRITE(WM2_LP_ILK
, 0);
3337 I915_WRITE(WM1_LP_ILK
, 0);
3340 * Based on the document from hardware guys the following bits
3341 * should be set unconditionally in order to enable FBC.
3342 * The bit 22 of 0x42000
3343 * The bit 22 of 0x42004
3344 * The bit 7,8,9 of 0x42020.
3346 if (IS_IRONLAKE_M(dev
)) {
3347 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
3348 I915_READ(ILK_DISPLAY_CHICKEN1
) |
3350 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
3351 I915_READ(ILK_DISPLAY_CHICKEN2
) |
3353 I915_WRITE(ILK_DSPCLK_GATE
,
3354 I915_READ(ILK_DSPCLK_GATE
) |
3360 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
3361 I915_READ(ILK_DISPLAY_CHICKEN2
) |
3362 ILK_ELPIN_409_SELECT
);
3363 I915_WRITE(_3D_CHICKEN2
,
3364 _3D_CHICKEN2_WM_READ_PIPELINED
<< 16 |
3365 _3D_CHICKEN2_WM_READ_PIPELINED
);
3368 static void gen6_init_clock_gating(struct drm_device
*dev
)
3370 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3372 uint32_t dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
;
3374 I915_WRITE(PCH_DSPCLK_GATE_D
, dspclk_gate
);
3376 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
3377 I915_READ(ILK_DISPLAY_CHICKEN2
) |
3378 ILK_ELPIN_409_SELECT
);
3380 I915_WRITE(WM3_LP_ILK
, 0);
3381 I915_WRITE(WM2_LP_ILK
, 0);
3382 I915_WRITE(WM1_LP_ILK
, 0);
3384 I915_WRITE(CACHE_MODE_0
,
3385 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB
));
3387 I915_WRITE(GEN6_UCGCTL1
,
3388 I915_READ(GEN6_UCGCTL1
) |
3389 GEN6_BLBUNIT_CLOCK_GATE_DISABLE
|
3390 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
3392 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3393 * gating disable must be set. Failure to set it results in
3394 * flickering pixels due to Z write ordering failures after
3395 * some amount of runtime in the Mesa "fire" demo, and Unigine
3396 * Sanctuary and Tropics, and apparently anything else with
3397 * alpha test or pixel discard.
3399 * According to the spec, bit 11 (RCCUNIT) must also be set,
3400 * but we didn't debug actual testcases to find it out.
3402 * Also apply WaDisableVDSUnitClockGating and
3403 * WaDisableRCPBUnitClockGating.
3405 I915_WRITE(GEN6_UCGCTL2
,
3406 GEN7_VDSUNIT_CLOCK_GATE_DISABLE
|
3407 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE
|
3408 GEN6_RCCUNIT_CLOCK_GATE_DISABLE
);
3410 /* Bspec says we need to always set all mask bits. */
3411 I915_WRITE(_3D_CHICKEN
, (0xFFFF << 16) |
3412 _3D_CHICKEN_SF_DISABLE_FASTCLIP_CULL
);
3415 * According to the spec the following bits should be
3416 * set in order to enable memory self-refresh and fbc:
3417 * The bit21 and bit22 of 0x42000
3418 * The bit21 and bit22 of 0x42004
3419 * The bit5 and bit7 of 0x42020
3420 * The bit14 of 0x70180
3421 * The bit14 of 0x71180
3423 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
3424 I915_READ(ILK_DISPLAY_CHICKEN1
) |
3425 ILK_FBCQ_DIS
| ILK_PABSTRETCH_DIS
);
3426 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
3427 I915_READ(ILK_DISPLAY_CHICKEN2
) |
3428 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
);
3429 I915_WRITE(ILK_DSPCLK_GATE
,
3430 I915_READ(ILK_DSPCLK_GATE
) |
3431 ILK_DPARB_CLK_GATE
|
3434 I915_WRITE(GEN6_MBCTL
, I915_READ(GEN6_MBCTL
) |
3435 GEN6_MBCTL_ENABLE_BOOT_FETCH
);
3437 for_each_pipe(pipe
) {
3438 I915_WRITE(DSPCNTR(pipe
),
3439 I915_READ(DSPCNTR(pipe
)) |
3440 DISPPLANE_TRICKLE_FEED_DISABLE
);
3441 intel_flush_display_plane(dev_priv
, pipe
);
3445 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private
*dev_priv
)
3447 uint32_t reg
= I915_READ(GEN7_FF_THREAD_MODE
);
3449 reg
&= ~GEN7_FF_SCHED_MASK
;
3450 reg
|= GEN7_FF_TS_SCHED_HW
;
3451 reg
|= GEN7_FF_VS_SCHED_HW
;
3452 reg
|= GEN7_FF_DS_SCHED_HW
;
3454 I915_WRITE(GEN7_FF_THREAD_MODE
, reg
);
3457 static void haswell_init_clock_gating(struct drm_device
*dev
)
3459 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3461 uint32_t dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
;
3463 I915_WRITE(PCH_DSPCLK_GATE_D
, dspclk_gate
);
3465 I915_WRITE(WM3_LP_ILK
, 0);
3466 I915_WRITE(WM2_LP_ILK
, 0);
3467 I915_WRITE(WM1_LP_ILK
, 0);
3469 /* According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3470 * This implements the WaDisableRCZUnitClockGating workaround.
3472 I915_WRITE(GEN6_UCGCTL2
, GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
3474 I915_WRITE(ILK_DSPCLK_GATE
, IVB_VRHUNIT_CLK_GATE
);
3476 I915_WRITE(IVB_CHICKEN3
,
3477 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
3478 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
3480 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3481 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1
,
3482 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
3484 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3485 I915_WRITE(GEN7_L3CNTLREG1
,
3486 GEN7_WA_FOR_GEN7_L3_CONTROL
);
3487 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER
,
3488 GEN7_WA_L3_CHICKEN_MODE
);
3490 /* This is required by WaCatErrorRejectionIssue */
3491 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
3492 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
3493 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
3495 for_each_pipe(pipe
) {
3496 I915_WRITE(DSPCNTR(pipe
),
3497 I915_READ(DSPCNTR(pipe
)) |
3498 DISPPLANE_TRICKLE_FEED_DISABLE
);
3499 intel_flush_display_plane(dev_priv
, pipe
);
3502 gen7_setup_fixed_func_scheduler(dev_priv
);
3504 /* WaDisable4x2SubspanOptimization */
3505 I915_WRITE(CACHE_MODE_1
,
3506 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
3508 /* XXX: This is a workaround for early silicon revisions and should be
3513 WM_DBG_DISALLOW_MULTIPLE_LP
|
3514 WM_DBG_DISALLOW_SPRITE
|
3515 WM_DBG_DISALLOW_MAXFIFO
);
3519 static void ivybridge_init_clock_gating(struct drm_device
*dev
)
3521 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3523 uint32_t dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
;
3526 I915_WRITE(PCH_DSPCLK_GATE_D
, dspclk_gate
);
3528 I915_WRITE(WM3_LP_ILK
, 0);
3529 I915_WRITE(WM2_LP_ILK
, 0);
3530 I915_WRITE(WM1_LP_ILK
, 0);
3532 I915_WRITE(ILK_DSPCLK_GATE
, IVB_VRHUNIT_CLK_GATE
);
3534 I915_WRITE(IVB_CHICKEN3
,
3535 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
3536 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
3538 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3539 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1
,
3540 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
3542 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3543 I915_WRITE(GEN7_L3CNTLREG1
,
3544 GEN7_WA_FOR_GEN7_L3_CONTROL
);
3545 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER
,
3546 GEN7_WA_L3_CHICKEN_MODE
);
3548 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3549 * gating disable must be set. Failure to set it results in
3550 * flickering pixels due to Z write ordering failures after
3551 * some amount of runtime in the Mesa "fire" demo, and Unigine
3552 * Sanctuary and Tropics, and apparently anything else with
3553 * alpha test or pixel discard.
3555 * According to the spec, bit 11 (RCCUNIT) must also be set,
3556 * but we didn't debug actual testcases to find it out.
3558 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3559 * This implements the WaDisableRCZUnitClockGating workaround.
3561 I915_WRITE(GEN6_UCGCTL2
,
3562 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
|
3563 GEN6_RCCUNIT_CLOCK_GATE_DISABLE
);
3565 /* This is required by WaCatErrorRejectionIssue */
3566 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
3567 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
3568 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
3570 for_each_pipe(pipe
) {
3571 I915_WRITE(DSPCNTR(pipe
),
3572 I915_READ(DSPCNTR(pipe
)) |
3573 DISPPLANE_TRICKLE_FEED_DISABLE
);
3574 intel_flush_display_plane(dev_priv
, pipe
);
3577 I915_WRITE(GEN6_MBCTL
, I915_READ(GEN6_MBCTL
) |
3578 GEN6_MBCTL_ENABLE_BOOT_FETCH
);
3580 gen7_setup_fixed_func_scheduler(dev_priv
);
3582 /* WaDisable4x2SubspanOptimization */
3583 I915_WRITE(CACHE_MODE_1
,
3584 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
3586 snpcr
= I915_READ(GEN6_MBCUNIT_SNPCR
);
3587 snpcr
&= ~GEN6_MBC_SNPCR_MASK
;
3588 snpcr
|= GEN6_MBC_SNPCR_MED
;
3589 I915_WRITE(GEN6_MBCUNIT_SNPCR
, snpcr
);
3592 static void valleyview_init_clock_gating(struct drm_device
*dev
)
3594 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3596 uint32_t dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
;
3598 I915_WRITE(PCH_DSPCLK_GATE_D
, dspclk_gate
);
3600 I915_WRITE(WM3_LP_ILK
, 0);
3601 I915_WRITE(WM2_LP_ILK
, 0);
3602 I915_WRITE(WM1_LP_ILK
, 0);
3604 I915_WRITE(ILK_DSPCLK_GATE
, IVB_VRHUNIT_CLK_GATE
);
3606 I915_WRITE(IVB_CHICKEN3
,
3607 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
3608 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
3610 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3611 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1
,
3612 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
3614 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3615 I915_WRITE(GEN7_L3CNTLREG1
, GEN7_WA_FOR_GEN7_L3_CONTROL
);
3616 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER
, GEN7_WA_L3_CHICKEN_MODE
);
3618 /* This is required by WaCatErrorRejectionIssue */
3619 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
3620 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
3621 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
3623 I915_WRITE(GEN6_MBCTL
, I915_READ(GEN6_MBCTL
) |
3624 GEN6_MBCTL_ENABLE_BOOT_FETCH
);
3627 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3628 * gating disable must be set. Failure to set it results in
3629 * flickering pixels due to Z write ordering failures after
3630 * some amount of runtime in the Mesa "fire" demo, and Unigine
3631 * Sanctuary and Tropics, and apparently anything else with
3632 * alpha test or pixel discard.
3634 * According to the spec, bit 11 (RCCUNIT) must also be set,
3635 * but we didn't debug actual testcases to find it out.
3637 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3638 * This implements the WaDisableRCZUnitClockGating workaround.
3640 * Also apply WaDisableVDSUnitClockGating and
3641 * WaDisableRCPBUnitClockGating.
3643 I915_WRITE(GEN6_UCGCTL2
,
3644 GEN7_VDSUNIT_CLOCK_GATE_DISABLE
|
3645 GEN7_TDLUNIT_CLOCK_GATE_DISABLE
|
3646 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
|
3647 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE
|
3648 GEN6_RCCUNIT_CLOCK_GATE_DISABLE
);
3650 I915_WRITE(GEN7_UCGCTL4
, GEN7_L3BANK2X_CLOCK_GATE_DISABLE
);
3652 for_each_pipe(pipe
) {
3653 I915_WRITE(DSPCNTR(pipe
),
3654 I915_READ(DSPCNTR(pipe
)) |
3655 DISPPLANE_TRICKLE_FEED_DISABLE
);
3656 intel_flush_display_plane(dev_priv
, pipe
);
3659 I915_WRITE(CACHE_MODE_1
,
3660 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
3663 * On ValleyView, the GUnit needs to signal the GT
3664 * when flip and other events complete. So enable
3665 * all the GUnit->GT interrupts here
3667 I915_WRITE(VLV_DPFLIPSTAT
, PIPEB_LINE_COMPARE_INT_EN
|
3668 PIPEB_HLINE_INT_EN
| PIPEB_VBLANK_INT_EN
|
3669 SPRITED_FLIPDONE_INT_EN
| SPRITEC_FLIPDONE_INT_EN
|
3670 PLANEB_FLIPDONE_INT_EN
| PIPEA_LINE_COMPARE_INT_EN
|
3671 PIPEA_HLINE_INT_EN
| PIPEA_VBLANK_INT_EN
|
3672 SPRITEB_FLIPDONE_INT_EN
| SPRITEA_FLIPDONE_INT_EN
|
3673 PLANEA_FLIPDONE_INT_EN
);
3676 static void g4x_init_clock_gating(struct drm_device
*dev
)
3678 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3679 uint32_t dspclk_gate
;
3681 I915_WRITE(RENCLK_GATE_D1
, 0);
3682 I915_WRITE(RENCLK_GATE_D2
, VF_UNIT_CLOCK_GATE_DISABLE
|
3683 GS_UNIT_CLOCK_GATE_DISABLE
|
3684 CL_UNIT_CLOCK_GATE_DISABLE
);
3685 I915_WRITE(RAMCLK_GATE_D
, 0);
3686 dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
|
3687 OVRUNIT_CLOCK_GATE_DISABLE
|
3688 OVCUNIT_CLOCK_GATE_DISABLE
;
3690 dspclk_gate
|= DSSUNIT_CLOCK_GATE_DISABLE
;
3691 I915_WRITE(DSPCLK_GATE_D
, dspclk_gate
);
3694 static void crestline_init_clock_gating(struct drm_device
*dev
)
3696 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3698 I915_WRITE(RENCLK_GATE_D1
, I965_RCC_CLOCK_GATE_DISABLE
);
3699 I915_WRITE(RENCLK_GATE_D2
, 0);
3700 I915_WRITE(DSPCLK_GATE_D
, 0);
3701 I915_WRITE(RAMCLK_GATE_D
, 0);
3702 I915_WRITE16(DEUC
, 0);
3705 static void broadwater_init_clock_gating(struct drm_device
*dev
)
3707 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3709 I915_WRITE(RENCLK_GATE_D1
, I965_RCZ_CLOCK_GATE_DISABLE
|
3710 I965_RCC_CLOCK_GATE_DISABLE
|
3711 I965_RCPB_CLOCK_GATE_DISABLE
|
3712 I965_ISC_CLOCK_GATE_DISABLE
|
3713 I965_FBC_CLOCK_GATE_DISABLE
);
3714 I915_WRITE(RENCLK_GATE_D2
, 0);
3717 static void gen3_init_clock_gating(struct drm_device
*dev
)
3719 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3720 u32 dstate
= I915_READ(D_STATE
);
3722 dstate
|= DSTATE_PLL_D3_OFF
| DSTATE_GFX_CLOCK_GATING
|
3723 DSTATE_DOT_CLOCK_GATING
;
3724 I915_WRITE(D_STATE
, dstate
);
3726 if (IS_PINEVIEW(dev
))
3727 I915_WRITE(ECOSKPD
, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY
));
3730 static void i85x_init_clock_gating(struct drm_device
*dev
)
3732 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3734 I915_WRITE(RENCLK_GATE_D1
, SV_CLOCK_GATE_DISABLE
);
3737 static void i830_init_clock_gating(struct drm_device
*dev
)
3739 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3741 I915_WRITE(DSPCLK_GATE_D
, OVRUNIT_CLOCK_GATE_DISABLE
);
3744 static void ibx_init_clock_gating(struct drm_device
*dev
)
3746 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3749 * On Ibex Peak and Cougar Point, we need to disable clock
3750 * gating for the panel power sequencer or it will fail to
3751 * start up when no ports are active.
3753 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
);
3756 static void cpt_init_clock_gating(struct drm_device
*dev
)
3758 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3762 * On Ibex Peak and Cougar Point, we need to disable clock
3763 * gating for the panel power sequencer or it will fail to
3764 * start up when no ports are active.
3766 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
);
3767 I915_WRITE(SOUTH_CHICKEN2
, I915_READ(SOUTH_CHICKEN2
) |
3768 DPLS_EDP_PPS_FIX_DIS
);
3769 /* Without this, mode sets may fail silently on FDI */
3771 I915_WRITE(TRANS_CHICKEN2(pipe
), TRANS_AUTOTRAIN_GEN_STALL_DIS
);
3774 void intel_init_clock_gating(struct drm_device
*dev
)
3776 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3778 dev_priv
->display
.init_clock_gating(dev
);
3780 if (dev_priv
->display
.init_pch_clock_gating
)
3781 dev_priv
->display
.init_pch_clock_gating(dev
);
3784 /* Starting with Haswell, we have different power wells for
3785 * different parts of the GPU. This attempts to enable them all.
3787 void intel_init_power_wells(struct drm_device
*dev
)
3789 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3790 unsigned long power_wells
[] = {
3797 if (!IS_HASWELL(dev
))
3800 mutex_lock(&dev
->struct_mutex
);
3802 for (i
= 0; i
< ARRAY_SIZE(power_wells
); i
++) {
3803 int well
= I915_READ(power_wells
[i
]);
3805 if ((well
& HSW_PWR_WELL_STATE
) == 0) {
3806 I915_WRITE(power_wells
[i
], well
& HSW_PWR_WELL_ENABLE
);
3807 if (wait_for(I915_READ(power_wells
[i
] & HSW_PWR_WELL_STATE
), 20))
3808 DRM_ERROR("Error enabling power well %lx\n", power_wells
[i
]);
3812 mutex_unlock(&dev
->struct_mutex
);
3815 /* Set up chip specific power management-related functions */
3816 void intel_init_pm(struct drm_device
*dev
)
3818 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3820 if (I915_HAS_FBC(dev
)) {
3821 if (HAS_PCH_SPLIT(dev
)) {
3822 dev_priv
->display
.fbc_enabled
= ironlake_fbc_enabled
;
3823 dev_priv
->display
.enable_fbc
= ironlake_enable_fbc
;
3824 dev_priv
->display
.disable_fbc
= ironlake_disable_fbc
;
3825 } else if (IS_GM45(dev
)) {
3826 dev_priv
->display
.fbc_enabled
= g4x_fbc_enabled
;
3827 dev_priv
->display
.enable_fbc
= g4x_enable_fbc
;
3828 dev_priv
->display
.disable_fbc
= g4x_disable_fbc
;
3829 } else if (IS_CRESTLINE(dev
)) {
3830 dev_priv
->display
.fbc_enabled
= i8xx_fbc_enabled
;
3831 dev_priv
->display
.enable_fbc
= i8xx_enable_fbc
;
3832 dev_priv
->display
.disable_fbc
= i8xx_disable_fbc
;
3834 /* 855GM needs testing */
3838 if (IS_PINEVIEW(dev
))
3839 i915_pineview_get_mem_freq(dev
);
3840 else if (IS_GEN5(dev
))
3841 i915_ironlake_get_mem_freq(dev
);
3843 /* For FIFO watermark updates */
3844 if (HAS_PCH_SPLIT(dev
)) {
3845 if (HAS_PCH_IBX(dev
))
3846 dev_priv
->display
.init_pch_clock_gating
= ibx_init_clock_gating
;
3847 else if (HAS_PCH_CPT(dev
))
3848 dev_priv
->display
.init_pch_clock_gating
= cpt_init_clock_gating
;
3851 if (I915_READ(MLTR_ILK
) & ILK_SRLT_MASK
)
3852 dev_priv
->display
.update_wm
= ironlake_update_wm
;
3854 DRM_DEBUG_KMS("Failed to get proper latency. "
3856 dev_priv
->display
.update_wm
= NULL
;
3858 dev_priv
->display
.init_clock_gating
= ironlake_init_clock_gating
;
3859 } else if (IS_GEN6(dev
)) {
3860 if (SNB_READ_WM0_LATENCY()) {
3861 dev_priv
->display
.update_wm
= sandybridge_update_wm
;
3862 dev_priv
->display
.update_sprite_wm
= sandybridge_update_sprite_wm
;
3864 DRM_DEBUG_KMS("Failed to read display plane latency. "
3866 dev_priv
->display
.update_wm
= NULL
;
3868 dev_priv
->display
.init_clock_gating
= gen6_init_clock_gating
;
3869 } else if (IS_IVYBRIDGE(dev
)) {
3870 /* FIXME: detect B0+ stepping and use auto training */
3871 if (SNB_READ_WM0_LATENCY()) {
3872 dev_priv
->display
.update_wm
= sandybridge_update_wm
;
3873 dev_priv
->display
.update_sprite_wm
= sandybridge_update_sprite_wm
;
3875 DRM_DEBUG_KMS("Failed to read display plane latency. "
3877 dev_priv
->display
.update_wm
= NULL
;
3879 dev_priv
->display
.init_clock_gating
= ivybridge_init_clock_gating
;
3880 } else if (IS_HASWELL(dev
)) {
3881 if (SNB_READ_WM0_LATENCY()) {
3882 dev_priv
->display
.update_wm
= sandybridge_update_wm
;
3883 dev_priv
->display
.update_sprite_wm
= sandybridge_update_sprite_wm
;
3884 dev_priv
->display
.update_linetime_wm
= haswell_update_linetime_wm
;
3886 DRM_DEBUG_KMS("Failed to read display plane latency. "
3888 dev_priv
->display
.update_wm
= NULL
;
3890 dev_priv
->display
.init_clock_gating
= haswell_init_clock_gating
;
3892 dev_priv
->display
.update_wm
= NULL
;
3893 } else if (IS_VALLEYVIEW(dev
)) {
3894 dev_priv
->display
.update_wm
= valleyview_update_wm
;
3895 dev_priv
->display
.init_clock_gating
=
3896 valleyview_init_clock_gating
;
3897 } else if (IS_PINEVIEW(dev
)) {
3898 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev
),
3901 dev_priv
->mem_freq
)) {
3902 DRM_INFO("failed to find known CxSR latency "
3903 "(found ddr%s fsb freq %d, mem freq %d), "
3905 (dev_priv
->is_ddr3
== 1) ? "3" : "2",
3906 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
3907 /* Disable CxSR and never update its watermark again */
3908 pineview_disable_cxsr(dev
);
3909 dev_priv
->display
.update_wm
= NULL
;
3911 dev_priv
->display
.update_wm
= pineview_update_wm
;
3912 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
3913 } else if (IS_G4X(dev
)) {
3914 dev_priv
->display
.update_wm
= g4x_update_wm
;
3915 dev_priv
->display
.init_clock_gating
= g4x_init_clock_gating
;
3916 } else if (IS_GEN4(dev
)) {
3917 dev_priv
->display
.update_wm
= i965_update_wm
;
3918 if (IS_CRESTLINE(dev
))
3919 dev_priv
->display
.init_clock_gating
= crestline_init_clock_gating
;
3920 else if (IS_BROADWATER(dev
))
3921 dev_priv
->display
.init_clock_gating
= broadwater_init_clock_gating
;
3922 } else if (IS_GEN3(dev
)) {
3923 dev_priv
->display
.update_wm
= i9xx_update_wm
;
3924 dev_priv
->display
.get_fifo_size
= i9xx_get_fifo_size
;
3925 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
3926 } else if (IS_I865G(dev
)) {
3927 dev_priv
->display
.update_wm
= i830_update_wm
;
3928 dev_priv
->display
.init_clock_gating
= i85x_init_clock_gating
;
3929 dev_priv
->display
.get_fifo_size
= i830_get_fifo_size
;
3930 } else if (IS_I85X(dev
)) {
3931 dev_priv
->display
.update_wm
= i9xx_update_wm
;
3932 dev_priv
->display
.get_fifo_size
= i85x_get_fifo_size
;
3933 dev_priv
->display
.init_clock_gating
= i85x_init_clock_gating
;
3935 dev_priv
->display
.update_wm
= i830_update_wm
;
3936 dev_priv
->display
.init_clock_gating
= i830_init_clock_gating
;
3938 dev_priv
->display
.get_fifo_size
= i845_get_fifo_size
;
3940 dev_priv
->display
.get_fifo_size
= i830_get_fifo_size
;
3944 static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private
*dev_priv
)
3946 u32 gt_thread_status_mask
;
3948 if (IS_HASWELL(dev_priv
->dev
))
3949 gt_thread_status_mask
= GEN6_GT_THREAD_STATUS_CORE_MASK_HSW
;
3951 gt_thread_status_mask
= GEN6_GT_THREAD_STATUS_CORE_MASK
;
3953 /* w/a for a sporadic read returning 0 by waiting for the GT
3954 * thread to wake up.
3956 if (wait_for_atomic_us((I915_READ_NOTRACE(GEN6_GT_THREAD_STATUS_REG
) & gt_thread_status_mask
) == 0, 500))
3957 DRM_ERROR("GT thread status wait timed out\n");
3960 static void __gen6_gt_force_wake_get(struct drm_i915_private
*dev_priv
)
3964 if (IS_HASWELL(dev_priv
->dev
))
3965 forcewake_ack
= FORCEWAKE_ACK_HSW
;
3967 forcewake_ack
= FORCEWAKE_ACK
;
3969 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack
) & 1) == 0,
3970 FORCEWAKE_ACK_TIMEOUT_MS
))
3971 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
3973 I915_WRITE_NOTRACE(FORCEWAKE
, 1);
3974 POSTING_READ(ECOBUS
); /* something from same cacheline, but !FORCEWAKE */
3976 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack
) & 1),
3977 FORCEWAKE_ACK_TIMEOUT_MS
))
3978 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
3980 __gen6_gt_wait_for_thread_c0(dev_priv
);
3983 static void __gen6_gt_force_wake_mt_get(struct drm_i915_private
*dev_priv
)
3987 if (IS_HASWELL(dev_priv
->dev
))
3988 forcewake_ack
= FORCEWAKE_ACK_HSW
;
3990 forcewake_ack
= FORCEWAKE_MT_ACK
;
3992 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack
) & 1) == 0,
3993 FORCEWAKE_ACK_TIMEOUT_MS
))
3994 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
3996 I915_WRITE_NOTRACE(FORCEWAKE_MT
, _MASKED_BIT_ENABLE(1));
3997 POSTING_READ(ECOBUS
); /* something from same cacheline, but !FORCEWAKE */
3999 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack
) & 1),
4000 FORCEWAKE_ACK_TIMEOUT_MS
))
4001 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4003 __gen6_gt_wait_for_thread_c0(dev_priv
);
4007 * Generally this is called implicitly by the register read function. However,
4008 * if some sequence requires the GT to not power down then this function should
4009 * be called at the beginning of the sequence followed by a call to
4010 * gen6_gt_force_wake_put() at the end of the sequence.
4012 void gen6_gt_force_wake_get(struct drm_i915_private
*dev_priv
)
4014 unsigned long irqflags
;
4016 spin_lock_irqsave(&dev_priv
->gt_lock
, irqflags
);
4017 if (dev_priv
->forcewake_count
++ == 0)
4018 dev_priv
->gt
.force_wake_get(dev_priv
);
4019 spin_unlock_irqrestore(&dev_priv
->gt_lock
, irqflags
);
4022 void gen6_gt_check_fifodbg(struct drm_i915_private
*dev_priv
)
4025 gtfifodbg
= I915_READ_NOTRACE(GTFIFODBG
);
4026 if (WARN(gtfifodbg
& GT_FIFO_CPU_ERROR_MASK
,
4027 "MMIO read or write has been dropped %x\n", gtfifodbg
))
4028 I915_WRITE_NOTRACE(GTFIFODBG
, GT_FIFO_CPU_ERROR_MASK
);
4031 static void __gen6_gt_force_wake_put(struct drm_i915_private
*dev_priv
)
4033 I915_WRITE_NOTRACE(FORCEWAKE
, 0);
4034 /* gen6_gt_check_fifodbg doubles as the POSTING_READ */
4035 gen6_gt_check_fifodbg(dev_priv
);
4038 static void __gen6_gt_force_wake_mt_put(struct drm_i915_private
*dev_priv
)
4040 I915_WRITE_NOTRACE(FORCEWAKE_MT
, _MASKED_BIT_DISABLE(1));
4041 /* gen6_gt_check_fifodbg doubles as the POSTING_READ */
4042 gen6_gt_check_fifodbg(dev_priv
);
4046 * see gen6_gt_force_wake_get()
4048 void gen6_gt_force_wake_put(struct drm_i915_private
*dev_priv
)
4050 unsigned long irqflags
;
4052 spin_lock_irqsave(&dev_priv
->gt_lock
, irqflags
);
4053 if (--dev_priv
->forcewake_count
== 0)
4054 dev_priv
->gt
.force_wake_put(dev_priv
);
4055 spin_unlock_irqrestore(&dev_priv
->gt_lock
, irqflags
);
4058 int __gen6_gt_wait_for_fifo(struct drm_i915_private
*dev_priv
)
4062 if (dev_priv
->gt_fifo_count
< GT_FIFO_NUM_RESERVED_ENTRIES
) {
4064 u32 fifo
= I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES
);
4065 while (fifo
<= GT_FIFO_NUM_RESERVED_ENTRIES
&& loop
--) {
4067 fifo
= I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES
);
4069 if (WARN_ON(loop
< 0 && fifo
<= GT_FIFO_NUM_RESERVED_ENTRIES
))
4071 dev_priv
->gt_fifo_count
= fifo
;
4073 dev_priv
->gt_fifo_count
--;
4078 static void vlv_force_wake_get(struct drm_i915_private
*dev_priv
)
4080 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV
) & 1) == 0,
4081 FORCEWAKE_ACK_TIMEOUT_MS
))
4082 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4084 I915_WRITE_NOTRACE(FORCEWAKE_VLV
, _MASKED_BIT_ENABLE(1));
4086 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV
) & 1),
4087 FORCEWAKE_ACK_TIMEOUT_MS
))
4088 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4090 __gen6_gt_wait_for_thread_c0(dev_priv
);
4093 static void vlv_force_wake_put(struct drm_i915_private
*dev_priv
)
4095 I915_WRITE_NOTRACE(FORCEWAKE_VLV
, _MASKED_BIT_DISABLE(1));
4096 /* The below doubles as a POSTING_READ */
4097 gen6_gt_check_fifodbg(dev_priv
);
4100 void intel_gt_init(struct drm_device
*dev
)
4102 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4104 spin_lock_init(&dev_priv
->gt_lock
);
4106 if (IS_VALLEYVIEW(dev
)) {
4107 dev_priv
->gt
.force_wake_get
= vlv_force_wake_get
;
4108 dev_priv
->gt
.force_wake_put
= vlv_force_wake_put
;
4109 } else if (INTEL_INFO(dev
)->gen
>= 6) {
4110 dev_priv
->gt
.force_wake_get
= __gen6_gt_force_wake_get
;
4111 dev_priv
->gt
.force_wake_put
= __gen6_gt_force_wake_put
;
4113 /* IVB configs may use multi-threaded forcewake */
4114 if (IS_IVYBRIDGE(dev
) || IS_HASWELL(dev
)) {
4117 /* A small trick here - if the bios hasn't configured
4118 * MT forcewake, and if the device is in RC6, then
4119 * force_wake_mt_get will not wake the device and the
4120 * ECOBUS read will return zero. Which will be
4121 * (correctly) interpreted by the test below as MT
4122 * forcewake being disabled.
4124 mutex_lock(&dev
->struct_mutex
);
4125 __gen6_gt_force_wake_mt_get(dev_priv
);
4126 ecobus
= I915_READ_NOTRACE(ECOBUS
);
4127 __gen6_gt_force_wake_mt_put(dev_priv
);
4128 mutex_unlock(&dev
->struct_mutex
);
4130 if (ecobus
& FORCEWAKE_MT_ENABLE
) {
4131 DRM_DEBUG_KMS("Using MT version of forcewake\n");
4132 dev_priv
->gt
.force_wake_get
=
4133 __gen6_gt_force_wake_mt_get
;
4134 dev_priv
->gt
.force_wake_put
=
4135 __gen6_gt_force_wake_mt_put
;