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>
33 #include <linux/vgaarb.h>
34 #include <drm/i915_powerwell.h>
35 #include <linux/pm_runtime.h>
38 * RC6 is a special power stage which allows the GPU to enter an very
39 * low-voltage mode when idle, using down to 0V while at this stage. This
40 * stage is entered automatically when the GPU is idle when RC6 support is
41 * enabled, and as soon as new workload arises GPU wakes up automatically as well.
43 * There are different RC6 modes available in Intel GPU, which differentiate
44 * among each other with the latency required to enter and leave RC6 and
45 * voltage consumed by the GPU in different states.
47 * The combination of the following flags define which states GPU is allowed
48 * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
49 * RC6pp is deepest RC6. Their support by hardware varies according to the
50 * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
51 * which brings the most power savings; deeper states save more power, but
52 * require higher latency to switch to and wake up.
54 #define INTEL_RC6_ENABLE (1<<0)
55 #define INTEL_RC6p_ENABLE (1<<1)
56 #define INTEL_RC6pp_ENABLE (1<<2)
58 /* FBC, or Frame Buffer Compression, is a technique employed to compress the
59 * framebuffer contents in-memory, aiming at reducing the required bandwidth
60 * during in-memory transfers and, therefore, reduce the power packet.
62 * The benefits of FBC are mostly visible with solid backgrounds and
63 * variation-less patterns.
65 * FBC-related functionality can be enabled by the means of the
66 * i915.i915_enable_fbc parameter
69 static void i8xx_disable_fbc(struct drm_device
*dev
)
71 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
74 /* Disable compression */
75 fbc_ctl
= I915_READ(FBC_CONTROL
);
76 if ((fbc_ctl
& FBC_CTL_EN
) == 0)
79 fbc_ctl
&= ~FBC_CTL_EN
;
80 I915_WRITE(FBC_CONTROL
, fbc_ctl
);
82 /* Wait for compressing bit to clear */
83 if (wait_for((I915_READ(FBC_STATUS
) & FBC_STAT_COMPRESSING
) == 0, 10)) {
84 DRM_DEBUG_KMS("FBC idle timed out\n");
88 DRM_DEBUG_KMS("disabled FBC\n");
91 static void i8xx_enable_fbc(struct drm_crtc
*crtc
)
93 struct drm_device
*dev
= crtc
->dev
;
94 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
95 struct drm_framebuffer
*fb
= crtc
->primary
->fb
;
96 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
97 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
98 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
103 cfb_pitch
= dev_priv
->fbc
.size
/ FBC_LL_SIZE
;
104 if (fb
->pitches
[0] < cfb_pitch
)
105 cfb_pitch
= fb
->pitches
[0];
107 /* FBC_CTL wants 32B or 64B units */
109 cfb_pitch
= (cfb_pitch
/ 32) - 1;
111 cfb_pitch
= (cfb_pitch
/ 64) - 1;
114 for (i
= 0; i
< (FBC_LL_SIZE
/ 32) + 1; i
++)
115 I915_WRITE(FBC_TAG
+ (i
* 4), 0);
121 fbc_ctl2
= FBC_CTL_FENCE_DBL
| FBC_CTL_IDLE_IMM
| FBC_CTL_CPU_FENCE
;
122 fbc_ctl2
|= FBC_CTL_PLANE(intel_crtc
->plane
);
123 I915_WRITE(FBC_CONTROL2
, fbc_ctl2
);
124 I915_WRITE(FBC_FENCE_OFF
, crtc
->y
);
128 fbc_ctl
= I915_READ(FBC_CONTROL
);
129 fbc_ctl
&= 0x3fff << FBC_CTL_INTERVAL_SHIFT
;
130 fbc_ctl
|= FBC_CTL_EN
| FBC_CTL_PERIODIC
;
132 fbc_ctl
|= FBC_CTL_C3_IDLE
; /* 945 needs special SR handling */
133 fbc_ctl
|= (cfb_pitch
& 0xff) << FBC_CTL_STRIDE_SHIFT
;
134 fbc_ctl
|= obj
->fence_reg
;
135 I915_WRITE(FBC_CONTROL
, fbc_ctl
);
137 DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c\n",
138 cfb_pitch
, crtc
->y
, plane_name(intel_crtc
->plane
));
141 static bool i8xx_fbc_enabled(struct drm_device
*dev
)
143 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
145 return I915_READ(FBC_CONTROL
) & FBC_CTL_EN
;
148 static void g4x_enable_fbc(struct drm_crtc
*crtc
)
150 struct drm_device
*dev
= crtc
->dev
;
151 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
152 struct drm_framebuffer
*fb
= crtc
->primary
->fb
;
153 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
154 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
155 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
158 dpfc_ctl
= DPFC_CTL_PLANE(intel_crtc
->plane
) | DPFC_SR_EN
;
159 if (drm_format_plane_cpp(fb
->pixel_format
, 0) == 2)
160 dpfc_ctl
|= DPFC_CTL_LIMIT_2X
;
162 dpfc_ctl
|= DPFC_CTL_LIMIT_1X
;
163 dpfc_ctl
|= DPFC_CTL_FENCE_EN
| obj
->fence_reg
;
165 I915_WRITE(DPFC_FENCE_YOFF
, crtc
->y
);
168 I915_WRITE(DPFC_CONTROL
, dpfc_ctl
| DPFC_CTL_EN
);
170 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc
->plane
));
173 static void g4x_disable_fbc(struct drm_device
*dev
)
175 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
178 /* Disable compression */
179 dpfc_ctl
= I915_READ(DPFC_CONTROL
);
180 if (dpfc_ctl
& DPFC_CTL_EN
) {
181 dpfc_ctl
&= ~DPFC_CTL_EN
;
182 I915_WRITE(DPFC_CONTROL
, dpfc_ctl
);
184 DRM_DEBUG_KMS("disabled FBC\n");
188 static bool g4x_fbc_enabled(struct drm_device
*dev
)
190 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
192 return I915_READ(DPFC_CONTROL
) & DPFC_CTL_EN
;
195 static void sandybridge_blit_fbc_update(struct drm_device
*dev
)
197 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
200 /* Make sure blitter notifies FBC of writes */
202 /* Blitter is part of Media powerwell on VLV. No impact of
203 * his param in other platforms for now */
204 gen6_gt_force_wake_get(dev_priv
, FORCEWAKE_MEDIA
);
206 blt_ecoskpd
= I915_READ(GEN6_BLITTER_ECOSKPD
);
207 blt_ecoskpd
|= GEN6_BLITTER_FBC_NOTIFY
<<
208 GEN6_BLITTER_LOCK_SHIFT
;
209 I915_WRITE(GEN6_BLITTER_ECOSKPD
, blt_ecoskpd
);
210 blt_ecoskpd
|= GEN6_BLITTER_FBC_NOTIFY
;
211 I915_WRITE(GEN6_BLITTER_ECOSKPD
, blt_ecoskpd
);
212 blt_ecoskpd
&= ~(GEN6_BLITTER_FBC_NOTIFY
<<
213 GEN6_BLITTER_LOCK_SHIFT
);
214 I915_WRITE(GEN6_BLITTER_ECOSKPD
, blt_ecoskpd
);
215 POSTING_READ(GEN6_BLITTER_ECOSKPD
);
217 gen6_gt_force_wake_put(dev_priv
, FORCEWAKE_MEDIA
);
220 static void ironlake_enable_fbc(struct drm_crtc
*crtc
)
222 struct drm_device
*dev
= crtc
->dev
;
223 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
224 struct drm_framebuffer
*fb
= crtc
->primary
->fb
;
225 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
226 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
227 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
230 dpfc_ctl
= DPFC_CTL_PLANE(intel_crtc
->plane
);
231 if (drm_format_plane_cpp(fb
->pixel_format
, 0) == 2)
232 dpfc_ctl
|= DPFC_CTL_LIMIT_2X
;
234 dpfc_ctl
|= DPFC_CTL_LIMIT_1X
;
235 dpfc_ctl
|= DPFC_CTL_FENCE_EN
;
237 dpfc_ctl
|= obj
->fence_reg
;
239 I915_WRITE(ILK_DPFC_FENCE_YOFF
, crtc
->y
);
240 I915_WRITE(ILK_FBC_RT_BASE
, i915_gem_obj_ggtt_offset(obj
) | ILK_FBC_RT_VALID
);
242 I915_WRITE(ILK_DPFC_CONTROL
, dpfc_ctl
| DPFC_CTL_EN
);
245 I915_WRITE(SNB_DPFC_CTL_SA
,
246 SNB_CPU_FENCE_ENABLE
| obj
->fence_reg
);
247 I915_WRITE(DPFC_CPU_FENCE_OFFSET
, crtc
->y
);
248 sandybridge_blit_fbc_update(dev
);
251 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc
->plane
));
254 static void ironlake_disable_fbc(struct drm_device
*dev
)
256 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
259 /* Disable compression */
260 dpfc_ctl
= I915_READ(ILK_DPFC_CONTROL
);
261 if (dpfc_ctl
& DPFC_CTL_EN
) {
262 dpfc_ctl
&= ~DPFC_CTL_EN
;
263 I915_WRITE(ILK_DPFC_CONTROL
, dpfc_ctl
);
265 DRM_DEBUG_KMS("disabled FBC\n");
269 static bool ironlake_fbc_enabled(struct drm_device
*dev
)
271 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
273 return I915_READ(ILK_DPFC_CONTROL
) & DPFC_CTL_EN
;
276 static void gen7_enable_fbc(struct drm_crtc
*crtc
)
278 struct drm_device
*dev
= crtc
->dev
;
279 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
280 struct drm_framebuffer
*fb
= crtc
->primary
->fb
;
281 struct intel_framebuffer
*intel_fb
= to_intel_framebuffer(fb
);
282 struct drm_i915_gem_object
*obj
= intel_fb
->obj
;
283 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
286 dpfc_ctl
= IVB_DPFC_CTL_PLANE(intel_crtc
->plane
);
287 if (drm_format_plane_cpp(fb
->pixel_format
, 0) == 2)
288 dpfc_ctl
|= DPFC_CTL_LIMIT_2X
;
290 dpfc_ctl
|= DPFC_CTL_LIMIT_1X
;
291 dpfc_ctl
|= IVB_DPFC_CTL_FENCE_EN
;
293 I915_WRITE(ILK_DPFC_CONTROL
, dpfc_ctl
| DPFC_CTL_EN
);
295 if (IS_IVYBRIDGE(dev
)) {
296 /* WaFbcAsynchFlipDisableFbcQueue:ivb */
297 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
298 I915_READ(ILK_DISPLAY_CHICKEN1
) |
301 /* WaFbcAsynchFlipDisableFbcQueue:hsw,bdw */
302 I915_WRITE(CHICKEN_PIPESL_1(intel_crtc
->pipe
),
303 I915_READ(CHICKEN_PIPESL_1(intel_crtc
->pipe
)) |
307 I915_WRITE(SNB_DPFC_CTL_SA
,
308 SNB_CPU_FENCE_ENABLE
| obj
->fence_reg
);
309 I915_WRITE(DPFC_CPU_FENCE_OFFSET
, crtc
->y
);
311 sandybridge_blit_fbc_update(dev
);
313 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc
->plane
));
316 bool intel_fbc_enabled(struct drm_device
*dev
)
318 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
320 if (!dev_priv
->display
.fbc_enabled
)
323 return dev_priv
->display
.fbc_enabled(dev
);
326 static void intel_fbc_work_fn(struct work_struct
*__work
)
328 struct intel_fbc_work
*work
=
329 container_of(to_delayed_work(__work
),
330 struct intel_fbc_work
, work
);
331 struct drm_device
*dev
= work
->crtc
->dev
;
332 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
334 mutex_lock(&dev
->struct_mutex
);
335 if (work
== dev_priv
->fbc
.fbc_work
) {
336 /* Double check that we haven't switched fb without cancelling
339 if (work
->crtc
->primary
->fb
== work
->fb
) {
340 dev_priv
->display
.enable_fbc(work
->crtc
);
342 dev_priv
->fbc
.plane
= to_intel_crtc(work
->crtc
)->plane
;
343 dev_priv
->fbc
.fb_id
= work
->crtc
->primary
->fb
->base
.id
;
344 dev_priv
->fbc
.y
= work
->crtc
->y
;
347 dev_priv
->fbc
.fbc_work
= NULL
;
349 mutex_unlock(&dev
->struct_mutex
);
354 static void intel_cancel_fbc_work(struct drm_i915_private
*dev_priv
)
356 if (dev_priv
->fbc
.fbc_work
== NULL
)
359 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
361 /* Synchronisation is provided by struct_mutex and checking of
362 * dev_priv->fbc.fbc_work, so we can perform the cancellation
363 * entirely asynchronously.
365 if (cancel_delayed_work(&dev_priv
->fbc
.fbc_work
->work
))
366 /* tasklet was killed before being run, clean up */
367 kfree(dev_priv
->fbc
.fbc_work
);
369 /* Mark the work as no longer wanted so that if it does
370 * wake-up (because the work was already running and waiting
371 * for our mutex), it will discover that is no longer
374 dev_priv
->fbc
.fbc_work
= NULL
;
377 static void intel_enable_fbc(struct drm_crtc
*crtc
)
379 struct intel_fbc_work
*work
;
380 struct drm_device
*dev
= crtc
->dev
;
381 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
383 if (!dev_priv
->display
.enable_fbc
)
386 intel_cancel_fbc_work(dev_priv
);
388 work
= kzalloc(sizeof(*work
), GFP_KERNEL
);
390 DRM_ERROR("Failed to allocate FBC work structure\n");
391 dev_priv
->display
.enable_fbc(crtc
);
396 work
->fb
= crtc
->primary
->fb
;
397 INIT_DELAYED_WORK(&work
->work
, intel_fbc_work_fn
);
399 dev_priv
->fbc
.fbc_work
= work
;
401 /* Delay the actual enabling to let pageflipping cease and the
402 * display to settle before starting the compression. Note that
403 * this delay also serves a second purpose: it allows for a
404 * vblank to pass after disabling the FBC before we attempt
405 * to modify the control registers.
407 * A more complicated solution would involve tracking vblanks
408 * following the termination of the page-flipping sequence
409 * and indeed performing the enable as a co-routine and not
410 * waiting synchronously upon the vblank.
412 * WaFbcWaitForVBlankBeforeEnable:ilk,snb
414 schedule_delayed_work(&work
->work
, msecs_to_jiffies(50));
417 void intel_disable_fbc(struct drm_device
*dev
)
419 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
421 intel_cancel_fbc_work(dev_priv
);
423 if (!dev_priv
->display
.disable_fbc
)
426 dev_priv
->display
.disable_fbc(dev
);
427 dev_priv
->fbc
.plane
= -1;
430 static bool set_no_fbc_reason(struct drm_i915_private
*dev_priv
,
431 enum no_fbc_reason reason
)
433 if (dev_priv
->fbc
.no_fbc_reason
== reason
)
436 dev_priv
->fbc
.no_fbc_reason
= reason
;
441 * intel_update_fbc - enable/disable FBC as needed
442 * @dev: the drm_device
444 * Set up the framebuffer compression hardware at mode set time. We
445 * enable it if possible:
446 * - plane A only (on pre-965)
447 * - no pixel mulitply/line duplication
448 * - no alpha buffer discard
450 * - framebuffer <= max_hdisplay in width, max_vdisplay in height
452 * We can't assume that any compression will take place (worst case),
453 * so the compressed buffer has to be the same size as the uncompressed
454 * one. It also must reside (along with the line length buffer) in
457 * We need to enable/disable FBC on a global basis.
459 void intel_update_fbc(struct drm_device
*dev
)
461 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
462 struct drm_crtc
*crtc
= NULL
, *tmp_crtc
;
463 struct intel_crtc
*intel_crtc
;
464 struct drm_framebuffer
*fb
;
465 struct intel_framebuffer
*intel_fb
;
466 struct drm_i915_gem_object
*obj
;
467 const struct drm_display_mode
*adjusted_mode
;
468 unsigned int max_width
, max_height
;
471 set_no_fbc_reason(dev_priv
, FBC_UNSUPPORTED
);
475 if (!i915
.powersave
) {
476 if (set_no_fbc_reason(dev_priv
, FBC_MODULE_PARAM
))
477 DRM_DEBUG_KMS("fbc disabled per module param\n");
482 * If FBC is already on, we just have to verify that we can
483 * keep it that way...
484 * Need to disable if:
485 * - more than one pipe is active
486 * - changing FBC params (stride, fence, mode)
487 * - new fb is too large to fit in compressed buffer
488 * - going to an unsupported config (interlace, pixel multiply, etc.)
490 for_each_crtc(dev
, tmp_crtc
) {
491 if (intel_crtc_active(tmp_crtc
) &&
492 to_intel_crtc(tmp_crtc
)->primary_enabled
) {
494 if (set_no_fbc_reason(dev_priv
, FBC_MULTIPLE_PIPES
))
495 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
502 if (!crtc
|| crtc
->primary
->fb
== NULL
) {
503 if (set_no_fbc_reason(dev_priv
, FBC_NO_OUTPUT
))
504 DRM_DEBUG_KMS("no output, disabling\n");
508 intel_crtc
= to_intel_crtc(crtc
);
509 fb
= crtc
->primary
->fb
;
510 intel_fb
= to_intel_framebuffer(fb
);
512 adjusted_mode
= &intel_crtc
->config
.adjusted_mode
;
514 if (i915
.enable_fbc
< 0) {
515 if (set_no_fbc_reason(dev_priv
, FBC_CHIP_DEFAULT
))
516 DRM_DEBUG_KMS("disabled per chip default\n");
519 if (!i915
.enable_fbc
) {
520 if (set_no_fbc_reason(dev_priv
, FBC_MODULE_PARAM
))
521 DRM_DEBUG_KMS("fbc disabled per module param\n");
524 if ((adjusted_mode
->flags
& DRM_MODE_FLAG_INTERLACE
) ||
525 (adjusted_mode
->flags
& DRM_MODE_FLAG_DBLSCAN
)) {
526 if (set_no_fbc_reason(dev_priv
, FBC_UNSUPPORTED_MODE
))
527 DRM_DEBUG_KMS("mode incompatible with compression, "
532 if (IS_G4X(dev
) || INTEL_INFO(dev
)->gen
>= 5) {
539 if (intel_crtc
->config
.pipe_src_w
> max_width
||
540 intel_crtc
->config
.pipe_src_h
> max_height
) {
541 if (set_no_fbc_reason(dev_priv
, FBC_MODE_TOO_LARGE
))
542 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
545 if ((INTEL_INFO(dev
)->gen
< 4 || HAS_DDI(dev
)) &&
546 intel_crtc
->plane
!= PLANE_A
) {
547 if (set_no_fbc_reason(dev_priv
, FBC_BAD_PLANE
))
548 DRM_DEBUG_KMS("plane not A, disabling compression\n");
552 /* The use of a CPU fence is mandatory in order to detect writes
553 * by the CPU to the scanout and trigger updates to the FBC.
555 if (obj
->tiling_mode
!= I915_TILING_X
||
556 obj
->fence_reg
== I915_FENCE_REG_NONE
) {
557 if (set_no_fbc_reason(dev_priv
, FBC_NOT_TILED
))
558 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
562 /* If the kernel debugger is active, always disable compression */
566 if (i915_gem_stolen_setup_compression(dev
, intel_fb
->obj
->base
.size
)) {
567 if (set_no_fbc_reason(dev_priv
, FBC_STOLEN_TOO_SMALL
))
568 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
572 /* If the scanout has not changed, don't modify the FBC settings.
573 * Note that we make the fundamental assumption that the fb->obj
574 * cannot be unpinned (and have its GTT offset and fence revoked)
575 * without first being decoupled from the scanout and FBC disabled.
577 if (dev_priv
->fbc
.plane
== intel_crtc
->plane
&&
578 dev_priv
->fbc
.fb_id
== fb
->base
.id
&&
579 dev_priv
->fbc
.y
== crtc
->y
)
582 if (intel_fbc_enabled(dev
)) {
583 /* We update FBC along two paths, after changing fb/crtc
584 * configuration (modeswitching) and after page-flipping
585 * finishes. For the latter, we know that not only did
586 * we disable the FBC at the start of the page-flip
587 * sequence, but also more than one vblank has passed.
589 * For the former case of modeswitching, it is possible
590 * to switch between two FBC valid configurations
591 * instantaneously so we do need to disable the FBC
592 * before we can modify its control registers. We also
593 * have to wait for the next vblank for that to take
594 * effect. However, since we delay enabling FBC we can
595 * assume that a vblank has passed since disabling and
596 * that we can safely alter the registers in the deferred
599 * In the scenario that we go from a valid to invalid
600 * and then back to valid FBC configuration we have
601 * no strict enforcement that a vblank occurred since
602 * disabling the FBC. However, along all current pipe
603 * disabling paths we do need to wait for a vblank at
604 * some point. And we wait before enabling FBC anyway.
606 DRM_DEBUG_KMS("disabling active FBC for update\n");
607 intel_disable_fbc(dev
);
610 intel_enable_fbc(crtc
);
611 dev_priv
->fbc
.no_fbc_reason
= FBC_OK
;
615 /* Multiple disables should be harmless */
616 if (intel_fbc_enabled(dev
)) {
617 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
618 intel_disable_fbc(dev
);
620 i915_gem_stolen_cleanup_compression(dev
);
623 static void i915_pineview_get_mem_freq(struct drm_device
*dev
)
625 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
628 tmp
= I915_READ(CLKCFG
);
630 switch (tmp
& CLKCFG_FSB_MASK
) {
632 dev_priv
->fsb_freq
= 533; /* 133*4 */
635 dev_priv
->fsb_freq
= 800; /* 200*4 */
638 dev_priv
->fsb_freq
= 667; /* 167*4 */
641 dev_priv
->fsb_freq
= 400; /* 100*4 */
645 switch (tmp
& CLKCFG_MEM_MASK
) {
647 dev_priv
->mem_freq
= 533;
650 dev_priv
->mem_freq
= 667;
653 dev_priv
->mem_freq
= 800;
657 /* detect pineview DDR3 setting */
658 tmp
= I915_READ(CSHRDDR3CTL
);
659 dev_priv
->is_ddr3
= (tmp
& CSHRDDR3CTL_DDR3
) ? 1 : 0;
662 static void i915_ironlake_get_mem_freq(struct drm_device
*dev
)
664 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
667 ddrpll
= I915_READ16(DDRMPLL1
);
668 csipll
= I915_READ16(CSIPLL0
);
670 switch (ddrpll
& 0xff) {
672 dev_priv
->mem_freq
= 800;
675 dev_priv
->mem_freq
= 1066;
678 dev_priv
->mem_freq
= 1333;
681 dev_priv
->mem_freq
= 1600;
684 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
686 dev_priv
->mem_freq
= 0;
690 dev_priv
->ips
.r_t
= dev_priv
->mem_freq
;
692 switch (csipll
& 0x3ff) {
694 dev_priv
->fsb_freq
= 3200;
697 dev_priv
->fsb_freq
= 3733;
700 dev_priv
->fsb_freq
= 4266;
703 dev_priv
->fsb_freq
= 4800;
706 dev_priv
->fsb_freq
= 5333;
709 dev_priv
->fsb_freq
= 5866;
712 dev_priv
->fsb_freq
= 6400;
715 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
717 dev_priv
->fsb_freq
= 0;
721 if (dev_priv
->fsb_freq
== 3200) {
722 dev_priv
->ips
.c_m
= 0;
723 } else if (dev_priv
->fsb_freq
> 3200 && dev_priv
->fsb_freq
<= 4800) {
724 dev_priv
->ips
.c_m
= 1;
726 dev_priv
->ips
.c_m
= 2;
730 static const struct cxsr_latency cxsr_latency_table
[] = {
731 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
732 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
733 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
734 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
735 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
737 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
738 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
739 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
740 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
741 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
743 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
744 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
745 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
746 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
747 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
749 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
750 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
751 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
752 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
753 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
755 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
756 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
757 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
758 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
759 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
761 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
762 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
763 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
764 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
765 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
768 static const struct cxsr_latency
*intel_get_cxsr_latency(int is_desktop
,
773 const struct cxsr_latency
*latency
;
776 if (fsb
== 0 || mem
== 0)
779 for (i
= 0; i
< ARRAY_SIZE(cxsr_latency_table
); i
++) {
780 latency
= &cxsr_latency_table
[i
];
781 if (is_desktop
== latency
->is_desktop
&&
782 is_ddr3
== latency
->is_ddr3
&&
783 fsb
== latency
->fsb_freq
&& mem
== latency
->mem_freq
)
787 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
792 static void pineview_disable_cxsr(struct drm_device
*dev
)
794 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
796 /* deactivate cxsr */
797 I915_WRITE(DSPFW3
, I915_READ(DSPFW3
) & ~PINEVIEW_SELF_REFRESH_EN
);
801 * Latency for FIFO fetches is dependent on several factors:
802 * - memory configuration (speed, channels)
804 * - current MCH state
805 * It can be fairly high in some situations, so here we assume a fairly
806 * pessimal value. It's a tradeoff between extra memory fetches (if we
807 * set this value too high, the FIFO will fetch frequently to stay full)
808 * and power consumption (set it too low to save power and we might see
809 * FIFO underruns and display "flicker").
811 * A value of 5us seems to be a good balance; safe for very low end
812 * platforms but not overly aggressive on lower latency configs.
814 static const int latency_ns
= 5000;
816 static int i9xx_get_fifo_size(struct drm_device
*dev
, int plane
)
818 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
819 uint32_t dsparb
= I915_READ(DSPARB
);
822 size
= dsparb
& 0x7f;
824 size
= ((dsparb
>> DSPARB_CSTART_SHIFT
) & 0x7f) - size
;
826 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
827 plane
? "B" : "A", size
);
832 static int i830_get_fifo_size(struct drm_device
*dev
, int plane
)
834 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
835 uint32_t dsparb
= I915_READ(DSPARB
);
838 size
= dsparb
& 0x1ff;
840 size
= ((dsparb
>> DSPARB_BEND_SHIFT
) & 0x1ff) - size
;
841 size
>>= 1; /* Convert to cachelines */
843 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
844 plane
? "B" : "A", size
);
849 static int i845_get_fifo_size(struct drm_device
*dev
, int plane
)
851 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
852 uint32_t dsparb
= I915_READ(DSPARB
);
855 size
= dsparb
& 0x7f;
856 size
>>= 2; /* Convert to cachelines */
858 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb
,
865 /* Pineview has different values for various configs */
866 static const struct intel_watermark_params pineview_display_wm
= {
867 PINEVIEW_DISPLAY_FIFO
,
871 PINEVIEW_FIFO_LINE_SIZE
873 static const struct intel_watermark_params pineview_display_hplloff_wm
= {
874 PINEVIEW_DISPLAY_FIFO
,
876 PINEVIEW_DFT_HPLLOFF_WM
,
878 PINEVIEW_FIFO_LINE_SIZE
880 static const struct intel_watermark_params pineview_cursor_wm
= {
881 PINEVIEW_CURSOR_FIFO
,
882 PINEVIEW_CURSOR_MAX_WM
,
883 PINEVIEW_CURSOR_DFT_WM
,
884 PINEVIEW_CURSOR_GUARD_WM
,
885 PINEVIEW_FIFO_LINE_SIZE
,
887 static const struct intel_watermark_params pineview_cursor_hplloff_wm
= {
888 PINEVIEW_CURSOR_FIFO
,
889 PINEVIEW_CURSOR_MAX_WM
,
890 PINEVIEW_CURSOR_DFT_WM
,
891 PINEVIEW_CURSOR_GUARD_WM
,
892 PINEVIEW_FIFO_LINE_SIZE
894 static const struct intel_watermark_params g4x_wm_info
= {
901 static const struct intel_watermark_params g4x_cursor_wm_info
= {
908 static const struct intel_watermark_params valleyview_wm_info
= {
909 VALLEYVIEW_FIFO_SIZE
,
915 static const struct intel_watermark_params valleyview_cursor_wm_info
= {
917 VALLEYVIEW_CURSOR_MAX_WM
,
922 static const struct intel_watermark_params i965_cursor_wm_info
= {
929 static const struct intel_watermark_params i945_wm_info
= {
936 static const struct intel_watermark_params i915_wm_info
= {
943 static const struct intel_watermark_params i830_wm_info
= {
950 static const struct intel_watermark_params i845_wm_info
= {
959 * intel_calculate_wm - calculate watermark level
960 * @clock_in_khz: pixel clock
961 * @wm: chip FIFO params
962 * @pixel_size: display pixel size
963 * @latency_ns: memory latency for the platform
965 * Calculate the watermark level (the level at which the display plane will
966 * start fetching from memory again). Each chip has a different display
967 * FIFO size and allocation, so the caller needs to figure that out and pass
968 * in the correct intel_watermark_params structure.
970 * As the pixel clock runs, the FIFO will be drained at a rate that depends
971 * on the pixel size. When it reaches the watermark level, it'll start
972 * fetching FIFO line sized based chunks from memory until the FIFO fills
973 * past the watermark point. If the FIFO drains completely, a FIFO underrun
974 * will occur, and a display engine hang could result.
976 static unsigned long intel_calculate_wm(unsigned long clock_in_khz
,
977 const struct intel_watermark_params
*wm
,
980 unsigned long latency_ns
)
982 long entries_required
, wm_size
;
985 * Note: we need to make sure we don't overflow for various clock &
987 * clocks go from a few thousand to several hundred thousand.
988 * latency is usually a few thousand
990 entries_required
= ((clock_in_khz
/ 1000) * pixel_size
* latency_ns
) /
992 entries_required
= DIV_ROUND_UP(entries_required
, wm
->cacheline_size
);
994 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required
);
996 wm_size
= fifo_size
- (entries_required
+ wm
->guard_size
);
998 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size
);
1000 /* Don't promote wm_size to unsigned... */
1001 if (wm_size
> (long)wm
->max_wm
)
1002 wm_size
= wm
->max_wm
;
1004 wm_size
= wm
->default_wm
;
1008 static struct drm_crtc
*single_enabled_crtc(struct drm_device
*dev
)
1010 struct drm_crtc
*crtc
, *enabled
= NULL
;
1012 for_each_crtc(dev
, crtc
) {
1013 if (intel_crtc_active(crtc
)) {
1023 static void pineview_update_wm(struct drm_crtc
*unused_crtc
)
1025 struct drm_device
*dev
= unused_crtc
->dev
;
1026 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1027 struct drm_crtc
*crtc
;
1028 const struct cxsr_latency
*latency
;
1032 latency
= intel_get_cxsr_latency(IS_PINEVIEW_G(dev
), dev_priv
->is_ddr3
,
1033 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
1035 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1036 pineview_disable_cxsr(dev
);
1040 crtc
= single_enabled_crtc(dev
);
1042 const struct drm_display_mode
*adjusted_mode
;
1043 int pixel_size
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
1046 adjusted_mode
= &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1047 clock
= adjusted_mode
->crtc_clock
;
1050 wm
= intel_calculate_wm(clock
, &pineview_display_wm
,
1051 pineview_display_wm
.fifo_size
,
1052 pixel_size
, latency
->display_sr
);
1053 reg
= I915_READ(DSPFW1
);
1054 reg
&= ~DSPFW_SR_MASK
;
1055 reg
|= wm
<< DSPFW_SR_SHIFT
;
1056 I915_WRITE(DSPFW1
, reg
);
1057 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg
);
1060 wm
= intel_calculate_wm(clock
, &pineview_cursor_wm
,
1061 pineview_display_wm
.fifo_size
,
1062 pixel_size
, latency
->cursor_sr
);
1063 reg
= I915_READ(DSPFW3
);
1064 reg
&= ~DSPFW_CURSOR_SR_MASK
;
1065 reg
|= (wm
& 0x3f) << DSPFW_CURSOR_SR_SHIFT
;
1066 I915_WRITE(DSPFW3
, reg
);
1068 /* Display HPLL off SR */
1069 wm
= intel_calculate_wm(clock
, &pineview_display_hplloff_wm
,
1070 pineview_display_hplloff_wm
.fifo_size
,
1071 pixel_size
, latency
->display_hpll_disable
);
1072 reg
= I915_READ(DSPFW3
);
1073 reg
&= ~DSPFW_HPLL_SR_MASK
;
1074 reg
|= wm
& DSPFW_HPLL_SR_MASK
;
1075 I915_WRITE(DSPFW3
, reg
);
1077 /* cursor HPLL off SR */
1078 wm
= intel_calculate_wm(clock
, &pineview_cursor_hplloff_wm
,
1079 pineview_display_hplloff_wm
.fifo_size
,
1080 pixel_size
, latency
->cursor_hpll_disable
);
1081 reg
= I915_READ(DSPFW3
);
1082 reg
&= ~DSPFW_HPLL_CURSOR_MASK
;
1083 reg
|= (wm
& 0x3f) << DSPFW_HPLL_CURSOR_SHIFT
;
1084 I915_WRITE(DSPFW3
, reg
);
1085 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg
);
1089 I915_READ(DSPFW3
) | PINEVIEW_SELF_REFRESH_EN
);
1090 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1092 pineview_disable_cxsr(dev
);
1093 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1097 static bool g4x_compute_wm0(struct drm_device
*dev
,
1099 const struct intel_watermark_params
*display
,
1100 int display_latency_ns
,
1101 const struct intel_watermark_params
*cursor
,
1102 int cursor_latency_ns
,
1106 struct drm_crtc
*crtc
;
1107 const struct drm_display_mode
*adjusted_mode
;
1108 int htotal
, hdisplay
, clock
, pixel_size
;
1109 int line_time_us
, line_count
;
1110 int entries
, tlb_miss
;
1112 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1113 if (!intel_crtc_active(crtc
)) {
1114 *cursor_wm
= cursor
->guard_size
;
1115 *plane_wm
= display
->guard_size
;
1119 adjusted_mode
= &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1120 clock
= adjusted_mode
->crtc_clock
;
1121 htotal
= adjusted_mode
->crtc_htotal
;
1122 hdisplay
= to_intel_crtc(crtc
)->config
.pipe_src_w
;
1123 pixel_size
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
1125 /* Use the small buffer method to calculate plane watermark */
1126 entries
= ((clock
* pixel_size
/ 1000) * display_latency_ns
) / 1000;
1127 tlb_miss
= display
->fifo_size
*display
->cacheline_size
- hdisplay
* 8;
1129 entries
+= tlb_miss
;
1130 entries
= DIV_ROUND_UP(entries
, display
->cacheline_size
);
1131 *plane_wm
= entries
+ display
->guard_size
;
1132 if (*plane_wm
> (int)display
->max_wm
)
1133 *plane_wm
= display
->max_wm
;
1135 /* Use the large buffer method to calculate cursor watermark */
1136 line_time_us
= max(htotal
* 1000 / clock
, 1);
1137 line_count
= (cursor_latency_ns
/ line_time_us
+ 1000) / 1000;
1138 entries
= line_count
* to_intel_crtc(crtc
)->cursor_width
* pixel_size
;
1139 tlb_miss
= cursor
->fifo_size
*cursor
->cacheline_size
- hdisplay
* 8;
1141 entries
+= tlb_miss
;
1142 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
1143 *cursor_wm
= entries
+ cursor
->guard_size
;
1144 if (*cursor_wm
> (int)cursor
->max_wm
)
1145 *cursor_wm
= (int)cursor
->max_wm
;
1151 * Check the wm result.
1153 * If any calculated watermark values is larger than the maximum value that
1154 * can be programmed into the associated watermark register, that watermark
1157 static bool g4x_check_srwm(struct drm_device
*dev
,
1158 int display_wm
, int cursor_wm
,
1159 const struct intel_watermark_params
*display
,
1160 const struct intel_watermark_params
*cursor
)
1162 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1163 display_wm
, cursor_wm
);
1165 if (display_wm
> display
->max_wm
) {
1166 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1167 display_wm
, display
->max_wm
);
1171 if (cursor_wm
> cursor
->max_wm
) {
1172 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1173 cursor_wm
, cursor
->max_wm
);
1177 if (!(display_wm
|| cursor_wm
)) {
1178 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1185 static bool g4x_compute_srwm(struct drm_device
*dev
,
1188 const struct intel_watermark_params
*display
,
1189 const struct intel_watermark_params
*cursor
,
1190 int *display_wm
, int *cursor_wm
)
1192 struct drm_crtc
*crtc
;
1193 const struct drm_display_mode
*adjusted_mode
;
1194 int hdisplay
, htotal
, pixel_size
, clock
;
1195 unsigned long line_time_us
;
1196 int line_count
, line_size
;
1201 *display_wm
= *cursor_wm
= 0;
1205 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1206 adjusted_mode
= &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1207 clock
= adjusted_mode
->crtc_clock
;
1208 htotal
= adjusted_mode
->crtc_htotal
;
1209 hdisplay
= to_intel_crtc(crtc
)->config
.pipe_src_w
;
1210 pixel_size
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
1212 line_time_us
= max(htotal
* 1000 / clock
, 1);
1213 line_count
= (latency_ns
/ line_time_us
+ 1000) / 1000;
1214 line_size
= hdisplay
* pixel_size
;
1216 /* Use the minimum of the small and large buffer method for primary */
1217 small
= ((clock
* pixel_size
/ 1000) * latency_ns
) / 1000;
1218 large
= line_count
* line_size
;
1220 entries
= DIV_ROUND_UP(min(small
, large
), display
->cacheline_size
);
1221 *display_wm
= entries
+ display
->guard_size
;
1223 /* calculate the self-refresh watermark for display cursor */
1224 entries
= line_count
* pixel_size
* to_intel_crtc(crtc
)->cursor_width
;
1225 entries
= DIV_ROUND_UP(entries
, cursor
->cacheline_size
);
1226 *cursor_wm
= entries
+ cursor
->guard_size
;
1228 return g4x_check_srwm(dev
,
1229 *display_wm
, *cursor_wm
,
1233 static bool vlv_compute_drain_latency(struct drm_device
*dev
,
1235 int *plane_prec_mult
,
1237 int *cursor_prec_mult
,
1240 struct drm_crtc
*crtc
;
1241 int clock
, pixel_size
;
1244 crtc
= intel_get_crtc_for_plane(dev
, plane
);
1245 if (!intel_crtc_active(crtc
))
1248 clock
= to_intel_crtc(crtc
)->config
.adjusted_mode
.crtc_clock
;
1249 pixel_size
= crtc
->primary
->fb
->bits_per_pixel
/ 8; /* BPP */
1251 entries
= (clock
/ 1000) * pixel_size
;
1252 *plane_prec_mult
= (entries
> 256) ?
1253 DRAIN_LATENCY_PRECISION_32
: DRAIN_LATENCY_PRECISION_16
;
1254 *plane_dl
= (64 * (*plane_prec_mult
) * 4) / ((clock
/ 1000) *
1257 entries
= (clock
/ 1000) * 4; /* BPP is always 4 for cursor */
1258 *cursor_prec_mult
= (entries
> 256) ?
1259 DRAIN_LATENCY_PRECISION_32
: DRAIN_LATENCY_PRECISION_16
;
1260 *cursor_dl
= (64 * (*cursor_prec_mult
) * 4) / ((clock
/ 1000) * 4);
1266 * Update drain latency registers of memory arbiter
1268 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1269 * to be programmed. Each plane has a drain latency multiplier and a drain
1273 static void vlv_update_drain_latency(struct drm_device
*dev
)
1275 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1276 int planea_prec
, planea_dl
, planeb_prec
, planeb_dl
;
1277 int cursora_prec
, cursora_dl
, cursorb_prec
, cursorb_dl
;
1278 int plane_prec_mult
, cursor_prec_mult
; /* Precision multiplier is
1281 /* For plane A, Cursor A */
1282 if (vlv_compute_drain_latency(dev
, 0, &plane_prec_mult
, &planea_dl
,
1283 &cursor_prec_mult
, &cursora_dl
)) {
1284 cursora_prec
= (cursor_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1285 DDL_CURSORA_PRECISION_32
: DDL_CURSORA_PRECISION_16
;
1286 planea_prec
= (plane_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1287 DDL_PLANEA_PRECISION_32
: DDL_PLANEA_PRECISION_16
;
1289 I915_WRITE(VLV_DDL1
, cursora_prec
|
1290 (cursora_dl
<< DDL_CURSORA_SHIFT
) |
1291 planea_prec
| planea_dl
);
1294 /* For plane B, Cursor B */
1295 if (vlv_compute_drain_latency(dev
, 1, &plane_prec_mult
, &planeb_dl
,
1296 &cursor_prec_mult
, &cursorb_dl
)) {
1297 cursorb_prec
= (cursor_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1298 DDL_CURSORB_PRECISION_32
: DDL_CURSORB_PRECISION_16
;
1299 planeb_prec
= (plane_prec_mult
== DRAIN_LATENCY_PRECISION_32
) ?
1300 DDL_PLANEB_PRECISION_32
: DDL_PLANEB_PRECISION_16
;
1302 I915_WRITE(VLV_DDL2
, cursorb_prec
|
1303 (cursorb_dl
<< DDL_CURSORB_SHIFT
) |
1304 planeb_prec
| planeb_dl
);
1308 #define single_plane_enabled(mask) is_power_of_2(mask)
1310 static void valleyview_update_wm(struct drm_crtc
*crtc
)
1312 struct drm_device
*dev
= crtc
->dev
;
1313 static const int sr_latency_ns
= 12000;
1314 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1315 int planea_wm
, planeb_wm
, cursora_wm
, cursorb_wm
;
1316 int plane_sr
, cursor_sr
;
1317 int ignore_plane_sr
, ignore_cursor_sr
;
1318 unsigned int enabled
= 0;
1320 vlv_update_drain_latency(dev
);
1322 if (g4x_compute_wm0(dev
, PIPE_A
,
1323 &valleyview_wm_info
, latency_ns
,
1324 &valleyview_cursor_wm_info
, latency_ns
,
1325 &planea_wm
, &cursora_wm
))
1326 enabled
|= 1 << PIPE_A
;
1328 if (g4x_compute_wm0(dev
, PIPE_B
,
1329 &valleyview_wm_info
, latency_ns
,
1330 &valleyview_cursor_wm_info
, latency_ns
,
1331 &planeb_wm
, &cursorb_wm
))
1332 enabled
|= 1 << PIPE_B
;
1334 if (single_plane_enabled(enabled
) &&
1335 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1337 &valleyview_wm_info
,
1338 &valleyview_cursor_wm_info
,
1339 &plane_sr
, &ignore_cursor_sr
) &&
1340 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1342 &valleyview_wm_info
,
1343 &valleyview_cursor_wm_info
,
1344 &ignore_plane_sr
, &cursor_sr
)) {
1345 I915_WRITE(FW_BLC_SELF_VLV
, FW_CSPWRDWNEN
);
1347 I915_WRITE(FW_BLC_SELF_VLV
,
1348 I915_READ(FW_BLC_SELF_VLV
) & ~FW_CSPWRDWNEN
);
1349 plane_sr
= cursor_sr
= 0;
1352 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1353 planea_wm
, cursora_wm
,
1354 planeb_wm
, cursorb_wm
,
1355 plane_sr
, cursor_sr
);
1358 (plane_sr
<< DSPFW_SR_SHIFT
) |
1359 (cursorb_wm
<< DSPFW_CURSORB_SHIFT
) |
1360 (planeb_wm
<< DSPFW_PLANEB_SHIFT
) |
1363 (I915_READ(DSPFW2
) & ~DSPFW_CURSORA_MASK
) |
1364 (cursora_wm
<< DSPFW_CURSORA_SHIFT
));
1366 (I915_READ(DSPFW3
) & ~DSPFW_CURSOR_SR_MASK
) |
1367 (cursor_sr
<< DSPFW_CURSOR_SR_SHIFT
));
1370 static void g4x_update_wm(struct drm_crtc
*crtc
)
1372 struct drm_device
*dev
= crtc
->dev
;
1373 static const int sr_latency_ns
= 12000;
1374 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1375 int planea_wm
, planeb_wm
, cursora_wm
, cursorb_wm
;
1376 int plane_sr
, cursor_sr
;
1377 unsigned int enabled
= 0;
1379 if (g4x_compute_wm0(dev
, PIPE_A
,
1380 &g4x_wm_info
, latency_ns
,
1381 &g4x_cursor_wm_info
, latency_ns
,
1382 &planea_wm
, &cursora_wm
))
1383 enabled
|= 1 << PIPE_A
;
1385 if (g4x_compute_wm0(dev
, PIPE_B
,
1386 &g4x_wm_info
, latency_ns
,
1387 &g4x_cursor_wm_info
, latency_ns
,
1388 &planeb_wm
, &cursorb_wm
))
1389 enabled
|= 1 << PIPE_B
;
1391 if (single_plane_enabled(enabled
) &&
1392 g4x_compute_srwm(dev
, ffs(enabled
) - 1,
1395 &g4x_cursor_wm_info
,
1396 &plane_sr
, &cursor_sr
)) {
1397 I915_WRITE(FW_BLC_SELF
, FW_BLC_SELF_EN
);
1399 I915_WRITE(FW_BLC_SELF
,
1400 I915_READ(FW_BLC_SELF
) & ~FW_BLC_SELF_EN
);
1401 plane_sr
= cursor_sr
= 0;
1404 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1405 planea_wm
, cursora_wm
,
1406 planeb_wm
, cursorb_wm
,
1407 plane_sr
, cursor_sr
);
1410 (plane_sr
<< DSPFW_SR_SHIFT
) |
1411 (cursorb_wm
<< DSPFW_CURSORB_SHIFT
) |
1412 (planeb_wm
<< DSPFW_PLANEB_SHIFT
) |
1415 (I915_READ(DSPFW2
) & ~DSPFW_CURSORA_MASK
) |
1416 (cursora_wm
<< DSPFW_CURSORA_SHIFT
));
1417 /* HPLL off in SR has some issues on G4x... disable it */
1419 (I915_READ(DSPFW3
) & ~(DSPFW_HPLL_SR_EN
| DSPFW_CURSOR_SR_MASK
)) |
1420 (cursor_sr
<< DSPFW_CURSOR_SR_SHIFT
));
1423 static void i965_update_wm(struct drm_crtc
*unused_crtc
)
1425 struct drm_device
*dev
= unused_crtc
->dev
;
1426 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1427 struct drm_crtc
*crtc
;
1431 /* Calc sr entries for one plane configs */
1432 crtc
= single_enabled_crtc(dev
);
1434 /* self-refresh has much higher latency */
1435 static const int sr_latency_ns
= 12000;
1436 const struct drm_display_mode
*adjusted_mode
=
1437 &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1438 int clock
= adjusted_mode
->crtc_clock
;
1439 int htotal
= adjusted_mode
->crtc_htotal
;
1440 int hdisplay
= to_intel_crtc(crtc
)->config
.pipe_src_w
;
1441 int pixel_size
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
1442 unsigned long line_time_us
;
1445 line_time_us
= max(htotal
* 1000 / clock
, 1);
1447 /* Use ns/us then divide to preserve precision */
1448 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1449 pixel_size
* hdisplay
;
1450 entries
= DIV_ROUND_UP(entries
, I915_FIFO_LINE_SIZE
);
1451 srwm
= I965_FIFO_SIZE
- entries
;
1455 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1458 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1459 pixel_size
* to_intel_crtc(crtc
)->cursor_width
;
1460 entries
= DIV_ROUND_UP(entries
,
1461 i965_cursor_wm_info
.cacheline_size
);
1462 cursor_sr
= i965_cursor_wm_info
.fifo_size
-
1463 (entries
+ i965_cursor_wm_info
.guard_size
);
1465 if (cursor_sr
> i965_cursor_wm_info
.max_wm
)
1466 cursor_sr
= i965_cursor_wm_info
.max_wm
;
1468 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1469 "cursor %d\n", srwm
, cursor_sr
);
1471 if (IS_CRESTLINE(dev
))
1472 I915_WRITE(FW_BLC_SELF
, FW_BLC_SELF_EN
);
1474 /* Turn off self refresh if both pipes are enabled */
1475 if (IS_CRESTLINE(dev
))
1476 I915_WRITE(FW_BLC_SELF
, I915_READ(FW_BLC_SELF
)
1480 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1483 /* 965 has limitations... */
1484 I915_WRITE(DSPFW1
, (srwm
<< DSPFW_SR_SHIFT
) |
1485 (8 << 16) | (8 << 8) | (8 << 0));
1486 I915_WRITE(DSPFW2
, (8 << 8) | (8 << 0));
1487 /* update cursor SR watermark */
1488 I915_WRITE(DSPFW3
, (cursor_sr
<< DSPFW_CURSOR_SR_SHIFT
));
1491 static void i9xx_update_wm(struct drm_crtc
*unused_crtc
)
1493 struct drm_device
*dev
= unused_crtc
->dev
;
1494 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1495 const struct intel_watermark_params
*wm_info
;
1500 int planea_wm
, planeb_wm
;
1501 struct drm_crtc
*crtc
, *enabled
= NULL
;
1504 wm_info
= &i945_wm_info
;
1505 else if (!IS_GEN2(dev
))
1506 wm_info
= &i915_wm_info
;
1508 wm_info
= &i830_wm_info
;
1510 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 0);
1511 crtc
= intel_get_crtc_for_plane(dev
, 0);
1512 if (intel_crtc_active(crtc
)) {
1513 const struct drm_display_mode
*adjusted_mode
;
1514 int cpp
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
1518 adjusted_mode
= &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1519 planea_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1520 wm_info
, fifo_size
, cpp
,
1524 planea_wm
= fifo_size
- wm_info
->guard_size
;
1526 fifo_size
= dev_priv
->display
.get_fifo_size(dev
, 1);
1527 crtc
= intel_get_crtc_for_plane(dev
, 1);
1528 if (intel_crtc_active(crtc
)) {
1529 const struct drm_display_mode
*adjusted_mode
;
1530 int cpp
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
1534 adjusted_mode
= &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1535 planeb_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1536 wm_info
, fifo_size
, cpp
,
1538 if (enabled
== NULL
)
1543 planeb_wm
= fifo_size
- wm_info
->guard_size
;
1545 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm
, planeb_wm
);
1547 if (IS_I915GM(dev
) && enabled
) {
1548 struct intel_framebuffer
*fb
;
1550 fb
= to_intel_framebuffer(enabled
->primary
->fb
);
1552 /* self-refresh seems busted with untiled */
1553 if (fb
->obj
->tiling_mode
== I915_TILING_NONE
)
1558 * Overlay gets an aggressive default since video jitter is bad.
1562 /* Play safe and disable self-refresh before adjusting watermarks. */
1563 if (IS_I945G(dev
) || IS_I945GM(dev
))
1564 I915_WRITE(FW_BLC_SELF
, FW_BLC_SELF_EN_MASK
| 0);
1565 else if (IS_I915GM(dev
))
1566 I915_WRITE(INSTPM
, _MASKED_BIT_DISABLE(INSTPM_SELF_EN
));
1568 /* Calc sr entries for one plane configs */
1569 if (HAS_FW_BLC(dev
) && enabled
) {
1570 /* self-refresh has much higher latency */
1571 static const int sr_latency_ns
= 6000;
1572 const struct drm_display_mode
*adjusted_mode
=
1573 &to_intel_crtc(enabled
)->config
.adjusted_mode
;
1574 int clock
= adjusted_mode
->crtc_clock
;
1575 int htotal
= adjusted_mode
->crtc_htotal
;
1576 int hdisplay
= to_intel_crtc(enabled
)->config
.pipe_src_w
;
1577 int pixel_size
= enabled
->primary
->fb
->bits_per_pixel
/ 8;
1578 unsigned long line_time_us
;
1581 line_time_us
= max(htotal
* 1000 / clock
, 1);
1583 /* Use ns/us then divide to preserve precision */
1584 entries
= (((sr_latency_ns
/ line_time_us
) + 1000) / 1000) *
1585 pixel_size
* hdisplay
;
1586 entries
= DIV_ROUND_UP(entries
, wm_info
->cacheline_size
);
1587 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries
);
1588 srwm
= wm_info
->fifo_size
- entries
;
1592 if (IS_I945G(dev
) || IS_I945GM(dev
))
1593 I915_WRITE(FW_BLC_SELF
,
1594 FW_BLC_SELF_FIFO_MASK
| (srwm
& 0xff));
1595 else if (IS_I915GM(dev
))
1596 I915_WRITE(FW_BLC_SELF
, srwm
& 0x3f);
1599 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1600 planea_wm
, planeb_wm
, cwm
, srwm
);
1602 fwater_lo
= ((planeb_wm
& 0x3f) << 16) | (planea_wm
& 0x3f);
1603 fwater_hi
= (cwm
& 0x1f);
1605 /* Set request length to 8 cachelines per fetch */
1606 fwater_lo
= fwater_lo
| (1 << 24) | (1 << 8);
1607 fwater_hi
= fwater_hi
| (1 << 8);
1609 I915_WRITE(FW_BLC
, fwater_lo
);
1610 I915_WRITE(FW_BLC2
, fwater_hi
);
1612 if (HAS_FW_BLC(dev
)) {
1614 if (IS_I945G(dev
) || IS_I945GM(dev
))
1615 I915_WRITE(FW_BLC_SELF
,
1616 FW_BLC_SELF_EN_MASK
| FW_BLC_SELF_EN
);
1617 else if (IS_I915GM(dev
))
1618 I915_WRITE(INSTPM
, _MASKED_BIT_ENABLE(INSTPM_SELF_EN
));
1619 DRM_DEBUG_KMS("memory self refresh enabled\n");
1621 DRM_DEBUG_KMS("memory self refresh disabled\n");
1625 static void i845_update_wm(struct drm_crtc
*unused_crtc
)
1627 struct drm_device
*dev
= unused_crtc
->dev
;
1628 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1629 struct drm_crtc
*crtc
;
1630 const struct drm_display_mode
*adjusted_mode
;
1634 crtc
= single_enabled_crtc(dev
);
1638 adjusted_mode
= &to_intel_crtc(crtc
)->config
.adjusted_mode
;
1639 planea_wm
= intel_calculate_wm(adjusted_mode
->crtc_clock
,
1641 dev_priv
->display
.get_fifo_size(dev
, 0),
1643 fwater_lo
= I915_READ(FW_BLC
) & ~0xfff;
1644 fwater_lo
|= (3<<8) | planea_wm
;
1646 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm
);
1648 I915_WRITE(FW_BLC
, fwater_lo
);
1651 static uint32_t ilk_pipe_pixel_rate(struct drm_device
*dev
,
1652 struct drm_crtc
*crtc
)
1654 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
1655 uint32_t pixel_rate
;
1657 pixel_rate
= intel_crtc
->config
.adjusted_mode
.crtc_clock
;
1659 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1660 * adjust the pixel_rate here. */
1662 if (intel_crtc
->config
.pch_pfit
.enabled
) {
1663 uint64_t pipe_w
, pipe_h
, pfit_w
, pfit_h
;
1664 uint32_t pfit_size
= intel_crtc
->config
.pch_pfit
.size
;
1666 pipe_w
= intel_crtc
->config
.pipe_src_w
;
1667 pipe_h
= intel_crtc
->config
.pipe_src_h
;
1668 pfit_w
= (pfit_size
>> 16) & 0xFFFF;
1669 pfit_h
= pfit_size
& 0xFFFF;
1670 if (pipe_w
< pfit_w
)
1672 if (pipe_h
< pfit_h
)
1675 pixel_rate
= div_u64((uint64_t) pixel_rate
* pipe_w
* pipe_h
,
1682 /* latency must be in 0.1us units. */
1683 static uint32_t ilk_wm_method1(uint32_t pixel_rate
, uint8_t bytes_per_pixel
,
1688 if (WARN(latency
== 0, "Latency value missing\n"))
1691 ret
= (uint64_t) pixel_rate
* bytes_per_pixel
* latency
;
1692 ret
= DIV_ROUND_UP_ULL(ret
, 64 * 10000) + 2;
1697 /* latency must be in 0.1us units. */
1698 static uint32_t ilk_wm_method2(uint32_t pixel_rate
, uint32_t pipe_htotal
,
1699 uint32_t horiz_pixels
, uint8_t bytes_per_pixel
,
1704 if (WARN(latency
== 0, "Latency value missing\n"))
1707 ret
= (latency
* pixel_rate
) / (pipe_htotal
* 10000);
1708 ret
= (ret
+ 1) * horiz_pixels
* bytes_per_pixel
;
1709 ret
= DIV_ROUND_UP(ret
, 64) + 2;
1713 static uint32_t ilk_wm_fbc(uint32_t pri_val
, uint32_t horiz_pixels
,
1714 uint8_t bytes_per_pixel
)
1716 return DIV_ROUND_UP(pri_val
* 64, horiz_pixels
* bytes_per_pixel
) + 2;
1719 struct ilk_pipe_wm_parameters
{
1721 uint32_t pipe_htotal
;
1722 uint32_t pixel_rate
;
1723 struct intel_plane_wm_parameters pri
;
1724 struct intel_plane_wm_parameters spr
;
1725 struct intel_plane_wm_parameters cur
;
1728 struct ilk_wm_maximums
{
1735 /* used in computing the new watermarks state */
1736 struct intel_wm_config
{
1737 unsigned int num_pipes_active
;
1738 bool sprites_enabled
;
1739 bool sprites_scaled
;
1743 * For both WM_PIPE and WM_LP.
1744 * mem_value must be in 0.1us units.
1746 static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters
*params
,
1750 uint32_t method1
, method2
;
1752 if (!params
->active
|| !params
->pri
.enabled
)
1755 method1
= ilk_wm_method1(params
->pixel_rate
,
1756 params
->pri
.bytes_per_pixel
,
1762 method2
= ilk_wm_method2(params
->pixel_rate
,
1763 params
->pipe_htotal
,
1764 params
->pri
.horiz_pixels
,
1765 params
->pri
.bytes_per_pixel
,
1768 return min(method1
, method2
);
1772 * For both WM_PIPE and WM_LP.
1773 * mem_value must be in 0.1us units.
1775 static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters
*params
,
1778 uint32_t method1
, method2
;
1780 if (!params
->active
|| !params
->spr
.enabled
)
1783 method1
= ilk_wm_method1(params
->pixel_rate
,
1784 params
->spr
.bytes_per_pixel
,
1786 method2
= ilk_wm_method2(params
->pixel_rate
,
1787 params
->pipe_htotal
,
1788 params
->spr
.horiz_pixels
,
1789 params
->spr
.bytes_per_pixel
,
1791 return min(method1
, method2
);
1795 * For both WM_PIPE and WM_LP.
1796 * mem_value must be in 0.1us units.
1798 static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters
*params
,
1801 if (!params
->active
|| !params
->cur
.enabled
)
1804 return ilk_wm_method2(params
->pixel_rate
,
1805 params
->pipe_htotal
,
1806 params
->cur
.horiz_pixels
,
1807 params
->cur
.bytes_per_pixel
,
1811 /* Only for WM_LP. */
1812 static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters
*params
,
1815 if (!params
->active
|| !params
->pri
.enabled
)
1818 return ilk_wm_fbc(pri_val
,
1819 params
->pri
.horiz_pixels
,
1820 params
->pri
.bytes_per_pixel
);
1823 static unsigned int ilk_display_fifo_size(const struct drm_device
*dev
)
1825 if (INTEL_INFO(dev
)->gen
>= 8)
1827 else if (INTEL_INFO(dev
)->gen
>= 7)
1833 static unsigned int ilk_plane_wm_reg_max(const struct drm_device
*dev
,
1834 int level
, bool is_sprite
)
1836 if (INTEL_INFO(dev
)->gen
>= 8)
1837 /* BDW primary/sprite plane watermarks */
1838 return level
== 0 ? 255 : 2047;
1839 else if (INTEL_INFO(dev
)->gen
>= 7)
1840 /* IVB/HSW primary/sprite plane watermarks */
1841 return level
== 0 ? 127 : 1023;
1842 else if (!is_sprite
)
1843 /* ILK/SNB primary plane watermarks */
1844 return level
== 0 ? 127 : 511;
1846 /* ILK/SNB sprite plane watermarks */
1847 return level
== 0 ? 63 : 255;
1850 static unsigned int ilk_cursor_wm_reg_max(const struct drm_device
*dev
,
1853 if (INTEL_INFO(dev
)->gen
>= 7)
1854 return level
== 0 ? 63 : 255;
1856 return level
== 0 ? 31 : 63;
1859 static unsigned int ilk_fbc_wm_reg_max(const struct drm_device
*dev
)
1861 if (INTEL_INFO(dev
)->gen
>= 8)
1867 /* Calculate the maximum primary/sprite plane watermark */
1868 static unsigned int ilk_plane_wm_max(const struct drm_device
*dev
,
1870 const struct intel_wm_config
*config
,
1871 enum intel_ddb_partitioning ddb_partitioning
,
1874 unsigned int fifo_size
= ilk_display_fifo_size(dev
);
1876 /* if sprites aren't enabled, sprites get nothing */
1877 if (is_sprite
&& !config
->sprites_enabled
)
1880 /* HSW allows LP1+ watermarks even with multiple pipes */
1881 if (level
== 0 || config
->num_pipes_active
> 1) {
1882 fifo_size
/= INTEL_INFO(dev
)->num_pipes
;
1885 * For some reason the non self refresh
1886 * FIFO size is only half of the self
1887 * refresh FIFO size on ILK/SNB.
1889 if (INTEL_INFO(dev
)->gen
<= 6)
1893 if (config
->sprites_enabled
) {
1894 /* level 0 is always calculated with 1:1 split */
1895 if (level
> 0 && ddb_partitioning
== INTEL_DDB_PART_5_6
) {
1904 /* clamp to max that the registers can hold */
1905 return min(fifo_size
, ilk_plane_wm_reg_max(dev
, level
, is_sprite
));
1908 /* Calculate the maximum cursor plane watermark */
1909 static unsigned int ilk_cursor_wm_max(const struct drm_device
*dev
,
1911 const struct intel_wm_config
*config
)
1913 /* HSW LP1+ watermarks w/ multiple pipes */
1914 if (level
> 0 && config
->num_pipes_active
> 1)
1917 /* otherwise just report max that registers can hold */
1918 return ilk_cursor_wm_reg_max(dev
, level
);
1921 static void ilk_compute_wm_maximums(const struct drm_device
*dev
,
1923 const struct intel_wm_config
*config
,
1924 enum intel_ddb_partitioning ddb_partitioning
,
1925 struct ilk_wm_maximums
*max
)
1927 max
->pri
= ilk_plane_wm_max(dev
, level
, config
, ddb_partitioning
, false);
1928 max
->spr
= ilk_plane_wm_max(dev
, level
, config
, ddb_partitioning
, true);
1929 max
->cur
= ilk_cursor_wm_max(dev
, level
, config
);
1930 max
->fbc
= ilk_fbc_wm_reg_max(dev
);
1933 static void ilk_compute_wm_reg_maximums(struct drm_device
*dev
,
1935 struct ilk_wm_maximums
*max
)
1937 max
->pri
= ilk_plane_wm_reg_max(dev
, level
, false);
1938 max
->spr
= ilk_plane_wm_reg_max(dev
, level
, true);
1939 max
->cur
= ilk_cursor_wm_reg_max(dev
, level
);
1940 max
->fbc
= ilk_fbc_wm_reg_max(dev
);
1943 static bool ilk_validate_wm_level(int level
,
1944 const struct ilk_wm_maximums
*max
,
1945 struct intel_wm_level
*result
)
1949 /* already determined to be invalid? */
1950 if (!result
->enable
)
1953 result
->enable
= result
->pri_val
<= max
->pri
&&
1954 result
->spr_val
<= max
->spr
&&
1955 result
->cur_val
<= max
->cur
;
1957 ret
= result
->enable
;
1960 * HACK until we can pre-compute everything,
1961 * and thus fail gracefully if LP0 watermarks
1964 if (level
== 0 && !result
->enable
) {
1965 if (result
->pri_val
> max
->pri
)
1966 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1967 level
, result
->pri_val
, max
->pri
);
1968 if (result
->spr_val
> max
->spr
)
1969 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1970 level
, result
->spr_val
, max
->spr
);
1971 if (result
->cur_val
> max
->cur
)
1972 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
1973 level
, result
->cur_val
, max
->cur
);
1975 result
->pri_val
= min_t(uint32_t, result
->pri_val
, max
->pri
);
1976 result
->spr_val
= min_t(uint32_t, result
->spr_val
, max
->spr
);
1977 result
->cur_val
= min_t(uint32_t, result
->cur_val
, max
->cur
);
1978 result
->enable
= true;
1984 static void ilk_compute_wm_level(const struct drm_i915_private
*dev_priv
,
1986 const struct ilk_pipe_wm_parameters
*p
,
1987 struct intel_wm_level
*result
)
1989 uint16_t pri_latency
= dev_priv
->wm
.pri_latency
[level
];
1990 uint16_t spr_latency
= dev_priv
->wm
.spr_latency
[level
];
1991 uint16_t cur_latency
= dev_priv
->wm
.cur_latency
[level
];
1993 /* WM1+ latency values stored in 0.5us units */
2000 result
->pri_val
= ilk_compute_pri_wm(p
, pri_latency
, level
);
2001 result
->spr_val
= ilk_compute_spr_wm(p
, spr_latency
);
2002 result
->cur_val
= ilk_compute_cur_wm(p
, cur_latency
);
2003 result
->fbc_val
= ilk_compute_fbc_wm(p
, result
->pri_val
);
2004 result
->enable
= true;
2008 hsw_compute_linetime_wm(struct drm_device
*dev
, struct drm_crtc
*crtc
)
2010 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2011 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2012 struct drm_display_mode
*mode
= &intel_crtc
->config
.adjusted_mode
;
2013 u32 linetime
, ips_linetime
;
2015 if (!intel_crtc_active(crtc
))
2018 /* The WM are computed with base on how long it takes to fill a single
2019 * row at the given clock rate, multiplied by 8.
2021 linetime
= DIV_ROUND_CLOSEST(mode
->crtc_htotal
* 1000 * 8,
2023 ips_linetime
= DIV_ROUND_CLOSEST(mode
->crtc_htotal
* 1000 * 8,
2024 intel_ddi_get_cdclk_freq(dev_priv
));
2026 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime
) |
2027 PIPE_WM_LINETIME_TIME(linetime
);
2030 static void intel_read_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2032 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2034 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
)) {
2035 uint64_t sskpd
= I915_READ64(MCH_SSKPD
);
2037 wm
[0] = (sskpd
>> 56) & 0xFF;
2039 wm
[0] = sskpd
& 0xF;
2040 wm
[1] = (sskpd
>> 4) & 0xFF;
2041 wm
[2] = (sskpd
>> 12) & 0xFF;
2042 wm
[3] = (sskpd
>> 20) & 0x1FF;
2043 wm
[4] = (sskpd
>> 32) & 0x1FF;
2044 } else if (INTEL_INFO(dev
)->gen
>= 6) {
2045 uint32_t sskpd
= I915_READ(MCH_SSKPD
);
2047 wm
[0] = (sskpd
>> SSKPD_WM0_SHIFT
) & SSKPD_WM_MASK
;
2048 wm
[1] = (sskpd
>> SSKPD_WM1_SHIFT
) & SSKPD_WM_MASK
;
2049 wm
[2] = (sskpd
>> SSKPD_WM2_SHIFT
) & SSKPD_WM_MASK
;
2050 wm
[3] = (sskpd
>> SSKPD_WM3_SHIFT
) & SSKPD_WM_MASK
;
2051 } else if (INTEL_INFO(dev
)->gen
>= 5) {
2052 uint32_t mltr
= I915_READ(MLTR_ILK
);
2054 /* ILK primary LP0 latency is 700 ns */
2056 wm
[1] = (mltr
>> MLTR_WM1_SHIFT
) & ILK_SRLT_MASK
;
2057 wm
[2] = (mltr
>> MLTR_WM2_SHIFT
) & ILK_SRLT_MASK
;
2061 static void intel_fixup_spr_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2063 /* ILK sprite LP0 latency is 1300 ns */
2064 if (INTEL_INFO(dev
)->gen
== 5)
2068 static void intel_fixup_cur_wm_latency(struct drm_device
*dev
, uint16_t wm
[5])
2070 /* ILK cursor LP0 latency is 1300 ns */
2071 if (INTEL_INFO(dev
)->gen
== 5)
2074 /* WaDoubleCursorLP3Latency:ivb */
2075 if (IS_IVYBRIDGE(dev
))
2079 int ilk_wm_max_level(const struct drm_device
*dev
)
2081 /* how many WM levels are we expecting */
2082 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2084 else if (INTEL_INFO(dev
)->gen
>= 6)
2090 static void intel_print_wm_latency(struct drm_device
*dev
,
2092 const uint16_t wm
[5])
2094 int level
, max_level
= ilk_wm_max_level(dev
);
2096 for (level
= 0; level
<= max_level
; level
++) {
2097 unsigned int latency
= wm
[level
];
2100 DRM_ERROR("%s WM%d latency not provided\n",
2105 /* WM1+ latency values in 0.5us units */
2109 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2110 name
, level
, wm
[level
],
2111 latency
/ 10, latency
% 10);
2115 static bool ilk_increase_wm_latency(struct drm_i915_private
*dev_priv
,
2116 uint16_t wm
[5], uint16_t min
)
2118 int level
, max_level
= ilk_wm_max_level(dev_priv
->dev
);
2123 wm
[0] = max(wm
[0], min
);
2124 for (level
= 1; level
<= max_level
; level
++)
2125 wm
[level
] = max_t(uint16_t, wm
[level
], DIV_ROUND_UP(min
, 5));
2130 static void snb_wm_latency_quirk(struct drm_device
*dev
)
2132 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2136 * The BIOS provided WM memory latency values are often
2137 * inadequate for high resolution displays. Adjust them.
2139 changed
= ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.pri_latency
, 12) |
2140 ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.spr_latency
, 12) |
2141 ilk_increase_wm_latency(dev_priv
, dev_priv
->wm
.cur_latency
, 12);
2146 DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
2147 intel_print_wm_latency(dev
, "Primary", dev_priv
->wm
.pri_latency
);
2148 intel_print_wm_latency(dev
, "Sprite", dev_priv
->wm
.spr_latency
);
2149 intel_print_wm_latency(dev
, "Cursor", dev_priv
->wm
.cur_latency
);
2152 static void ilk_setup_wm_latency(struct drm_device
*dev
)
2154 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2156 intel_read_wm_latency(dev
, dev_priv
->wm
.pri_latency
);
2158 memcpy(dev_priv
->wm
.spr_latency
, dev_priv
->wm
.pri_latency
,
2159 sizeof(dev_priv
->wm
.pri_latency
));
2160 memcpy(dev_priv
->wm
.cur_latency
, dev_priv
->wm
.pri_latency
,
2161 sizeof(dev_priv
->wm
.pri_latency
));
2163 intel_fixup_spr_wm_latency(dev
, dev_priv
->wm
.spr_latency
);
2164 intel_fixup_cur_wm_latency(dev
, dev_priv
->wm
.cur_latency
);
2166 intel_print_wm_latency(dev
, "Primary", dev_priv
->wm
.pri_latency
);
2167 intel_print_wm_latency(dev
, "Sprite", dev_priv
->wm
.spr_latency
);
2168 intel_print_wm_latency(dev
, "Cursor", dev_priv
->wm
.cur_latency
);
2171 snb_wm_latency_quirk(dev
);
2174 static void ilk_compute_wm_parameters(struct drm_crtc
*crtc
,
2175 struct ilk_pipe_wm_parameters
*p
)
2177 struct drm_device
*dev
= crtc
->dev
;
2178 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2179 enum pipe pipe
= intel_crtc
->pipe
;
2180 struct drm_plane
*plane
;
2182 if (!intel_crtc_active(crtc
))
2186 p
->pipe_htotal
= intel_crtc
->config
.adjusted_mode
.crtc_htotal
;
2187 p
->pixel_rate
= ilk_pipe_pixel_rate(dev
, crtc
);
2188 p
->pri
.bytes_per_pixel
= crtc
->primary
->fb
->bits_per_pixel
/ 8;
2189 p
->cur
.bytes_per_pixel
= 4;
2190 p
->pri
.horiz_pixels
= intel_crtc
->config
.pipe_src_w
;
2191 p
->cur
.horiz_pixels
= intel_crtc
->cursor_width
;
2192 /* TODO: for now, assume primary and cursor planes are always enabled. */
2193 p
->pri
.enabled
= true;
2194 p
->cur
.enabled
= true;
2196 drm_for_each_legacy_plane(plane
, &dev
->mode_config
.plane_list
) {
2197 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
2199 if (intel_plane
->pipe
== pipe
) {
2200 p
->spr
= intel_plane
->wm
;
2206 static void ilk_compute_wm_config(struct drm_device
*dev
,
2207 struct intel_wm_config
*config
)
2209 struct intel_crtc
*intel_crtc
;
2211 /* Compute the currently _active_ config */
2212 for_each_intel_crtc(dev
, intel_crtc
) {
2213 const struct intel_pipe_wm
*wm
= &intel_crtc
->wm
.active
;
2215 if (!wm
->pipe_enabled
)
2218 config
->sprites_enabled
|= wm
->sprites_enabled
;
2219 config
->sprites_scaled
|= wm
->sprites_scaled
;
2220 config
->num_pipes_active
++;
2224 /* Compute new watermarks for the pipe */
2225 static bool intel_compute_pipe_wm(struct drm_crtc
*crtc
,
2226 const struct ilk_pipe_wm_parameters
*params
,
2227 struct intel_pipe_wm
*pipe_wm
)
2229 struct drm_device
*dev
= crtc
->dev
;
2230 const struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2231 int level
, max_level
= ilk_wm_max_level(dev
);
2232 /* LP0 watermark maximums depend on this pipe alone */
2233 struct intel_wm_config config
= {
2234 .num_pipes_active
= 1,
2235 .sprites_enabled
= params
->spr
.enabled
,
2236 .sprites_scaled
= params
->spr
.scaled
,
2238 struct ilk_wm_maximums max
;
2240 pipe_wm
->pipe_enabled
= params
->active
;
2241 pipe_wm
->sprites_enabled
= params
->spr
.enabled
;
2242 pipe_wm
->sprites_scaled
= params
->spr
.scaled
;
2244 /* ILK/SNB: LP2+ watermarks only w/o sprites */
2245 if (INTEL_INFO(dev
)->gen
<= 6 && params
->spr
.enabled
)
2248 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
2249 if (params
->spr
.scaled
)
2252 ilk_compute_wm_level(dev_priv
, 0, params
, &pipe_wm
->wm
[0]);
2254 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2255 pipe_wm
->linetime
= hsw_compute_linetime_wm(dev
, crtc
);
2257 /* LP0 watermarks always use 1/2 DDB partitioning */
2258 ilk_compute_wm_maximums(dev
, 0, &config
, INTEL_DDB_PART_1_2
, &max
);
2260 /* At least LP0 must be valid */
2261 if (!ilk_validate_wm_level(0, &max
, &pipe_wm
->wm
[0]))
2264 ilk_compute_wm_reg_maximums(dev
, 1, &max
);
2266 for (level
= 1; level
<= max_level
; level
++) {
2267 struct intel_wm_level wm
= {};
2269 ilk_compute_wm_level(dev_priv
, level
, params
, &wm
);
2272 * Disable any watermark level that exceeds the
2273 * register maximums since such watermarks are
2276 if (!ilk_validate_wm_level(level
, &max
, &wm
))
2279 pipe_wm
->wm
[level
] = wm
;
2286 * Merge the watermarks from all active pipes for a specific level.
2288 static void ilk_merge_wm_level(struct drm_device
*dev
,
2290 struct intel_wm_level
*ret_wm
)
2292 const struct intel_crtc
*intel_crtc
;
2294 ret_wm
->enable
= true;
2296 for_each_intel_crtc(dev
, intel_crtc
) {
2297 const struct intel_pipe_wm
*active
= &intel_crtc
->wm
.active
;
2298 const struct intel_wm_level
*wm
= &active
->wm
[level
];
2300 if (!active
->pipe_enabled
)
2304 * The watermark values may have been used in the past,
2305 * so we must maintain them in the registers for some
2306 * time even if the level is now disabled.
2309 ret_wm
->enable
= false;
2311 ret_wm
->pri_val
= max(ret_wm
->pri_val
, wm
->pri_val
);
2312 ret_wm
->spr_val
= max(ret_wm
->spr_val
, wm
->spr_val
);
2313 ret_wm
->cur_val
= max(ret_wm
->cur_val
, wm
->cur_val
);
2314 ret_wm
->fbc_val
= max(ret_wm
->fbc_val
, wm
->fbc_val
);
2319 * Merge all low power watermarks for all active pipes.
2321 static void ilk_wm_merge(struct drm_device
*dev
,
2322 const struct intel_wm_config
*config
,
2323 const struct ilk_wm_maximums
*max
,
2324 struct intel_pipe_wm
*merged
)
2326 int level
, max_level
= ilk_wm_max_level(dev
);
2327 int last_enabled_level
= max_level
;
2329 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2330 if ((INTEL_INFO(dev
)->gen
<= 6 || IS_IVYBRIDGE(dev
)) &&
2331 config
->num_pipes_active
> 1)
2334 /* ILK: FBC WM must be disabled always */
2335 merged
->fbc_wm_enabled
= INTEL_INFO(dev
)->gen
>= 6;
2337 /* merge each WM1+ level */
2338 for (level
= 1; level
<= max_level
; level
++) {
2339 struct intel_wm_level
*wm
= &merged
->wm
[level
];
2341 ilk_merge_wm_level(dev
, level
, wm
);
2343 if (level
> last_enabled_level
)
2345 else if (!ilk_validate_wm_level(level
, max
, wm
))
2346 /* make sure all following levels get disabled */
2347 last_enabled_level
= level
- 1;
2350 * The spec says it is preferred to disable
2351 * FBC WMs instead of disabling a WM level.
2353 if (wm
->fbc_val
> max
->fbc
) {
2355 merged
->fbc_wm_enabled
= false;
2360 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2362 * FIXME this is racy. FBC might get enabled later.
2363 * What we should check here is whether FBC can be
2364 * enabled sometime later.
2366 if (IS_GEN5(dev
) && !merged
->fbc_wm_enabled
&& intel_fbc_enabled(dev
)) {
2367 for (level
= 2; level
<= max_level
; level
++) {
2368 struct intel_wm_level
*wm
= &merged
->wm
[level
];
2375 static int ilk_wm_lp_to_level(int wm_lp
, const struct intel_pipe_wm
*pipe_wm
)
2377 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2378 return wm_lp
+ (wm_lp
>= 2 && pipe_wm
->wm
[4].enable
);
2381 /* The value we need to program into the WM_LPx latency field */
2382 static unsigned int ilk_wm_lp_latency(struct drm_device
*dev
, int level
)
2384 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2386 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2389 return dev_priv
->wm
.pri_latency
[level
];
2392 static void ilk_compute_wm_results(struct drm_device
*dev
,
2393 const struct intel_pipe_wm
*merged
,
2394 enum intel_ddb_partitioning partitioning
,
2395 struct ilk_wm_values
*results
)
2397 struct intel_crtc
*intel_crtc
;
2400 results
->enable_fbc_wm
= merged
->fbc_wm_enabled
;
2401 results
->partitioning
= partitioning
;
2403 /* LP1+ register values */
2404 for (wm_lp
= 1; wm_lp
<= 3; wm_lp
++) {
2405 const struct intel_wm_level
*r
;
2407 level
= ilk_wm_lp_to_level(wm_lp
, merged
);
2409 r
= &merged
->wm
[level
];
2412 * Maintain the watermark values even if the level is
2413 * disabled. Doing otherwise could cause underruns.
2415 results
->wm_lp
[wm_lp
- 1] =
2416 (ilk_wm_lp_latency(dev
, level
) << WM1_LP_LATENCY_SHIFT
) |
2417 (r
->pri_val
<< WM1_LP_SR_SHIFT
) |
2421 results
->wm_lp
[wm_lp
- 1] |= WM1_LP_SR_EN
;
2423 if (INTEL_INFO(dev
)->gen
>= 8)
2424 results
->wm_lp
[wm_lp
- 1] |=
2425 r
->fbc_val
<< WM1_LP_FBC_SHIFT_BDW
;
2427 results
->wm_lp
[wm_lp
- 1] |=
2428 r
->fbc_val
<< WM1_LP_FBC_SHIFT
;
2431 * Always set WM1S_LP_EN when spr_val != 0, even if the
2432 * level is disabled. Doing otherwise could cause underruns.
2434 if (INTEL_INFO(dev
)->gen
<= 6 && r
->spr_val
) {
2435 WARN_ON(wm_lp
!= 1);
2436 results
->wm_lp_spr
[wm_lp
- 1] = WM1S_LP_EN
| r
->spr_val
;
2438 results
->wm_lp_spr
[wm_lp
- 1] = r
->spr_val
;
2441 /* LP0 register values */
2442 for_each_intel_crtc(dev
, intel_crtc
) {
2443 enum pipe pipe
= intel_crtc
->pipe
;
2444 const struct intel_wm_level
*r
=
2445 &intel_crtc
->wm
.active
.wm
[0];
2447 if (WARN_ON(!r
->enable
))
2450 results
->wm_linetime
[pipe
] = intel_crtc
->wm
.active
.linetime
;
2452 results
->wm_pipe
[pipe
] =
2453 (r
->pri_val
<< WM0_PIPE_PLANE_SHIFT
) |
2454 (r
->spr_val
<< WM0_PIPE_SPRITE_SHIFT
) |
2459 /* Find the result with the highest level enabled. Check for enable_fbc_wm in
2460 * case both are at the same level. Prefer r1 in case they're the same. */
2461 static struct intel_pipe_wm
*ilk_find_best_result(struct drm_device
*dev
,
2462 struct intel_pipe_wm
*r1
,
2463 struct intel_pipe_wm
*r2
)
2465 int level
, max_level
= ilk_wm_max_level(dev
);
2466 int level1
= 0, level2
= 0;
2468 for (level
= 1; level
<= max_level
; level
++) {
2469 if (r1
->wm
[level
].enable
)
2471 if (r2
->wm
[level
].enable
)
2475 if (level1
== level2
) {
2476 if (r2
->fbc_wm_enabled
&& !r1
->fbc_wm_enabled
)
2480 } else if (level1
> level2
) {
2487 /* dirty bits used to track which watermarks need changes */
2488 #define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2489 #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2490 #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2491 #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2492 #define WM_DIRTY_FBC (1 << 24)
2493 #define WM_DIRTY_DDB (1 << 25)
2495 static unsigned int ilk_compute_wm_dirty(struct drm_device
*dev
,
2496 const struct ilk_wm_values
*old
,
2497 const struct ilk_wm_values
*new)
2499 unsigned int dirty
= 0;
2503 for_each_pipe(pipe
) {
2504 if (old
->wm_linetime
[pipe
] != new->wm_linetime
[pipe
]) {
2505 dirty
|= WM_DIRTY_LINETIME(pipe
);
2506 /* Must disable LP1+ watermarks too */
2507 dirty
|= WM_DIRTY_LP_ALL
;
2510 if (old
->wm_pipe
[pipe
] != new->wm_pipe
[pipe
]) {
2511 dirty
|= WM_DIRTY_PIPE(pipe
);
2512 /* Must disable LP1+ watermarks too */
2513 dirty
|= WM_DIRTY_LP_ALL
;
2517 if (old
->enable_fbc_wm
!= new->enable_fbc_wm
) {
2518 dirty
|= WM_DIRTY_FBC
;
2519 /* Must disable LP1+ watermarks too */
2520 dirty
|= WM_DIRTY_LP_ALL
;
2523 if (old
->partitioning
!= new->partitioning
) {
2524 dirty
|= WM_DIRTY_DDB
;
2525 /* Must disable LP1+ watermarks too */
2526 dirty
|= WM_DIRTY_LP_ALL
;
2529 /* LP1+ watermarks already deemed dirty, no need to continue */
2530 if (dirty
& WM_DIRTY_LP_ALL
)
2533 /* Find the lowest numbered LP1+ watermark in need of an update... */
2534 for (wm_lp
= 1; wm_lp
<= 3; wm_lp
++) {
2535 if (old
->wm_lp
[wm_lp
- 1] != new->wm_lp
[wm_lp
- 1] ||
2536 old
->wm_lp_spr
[wm_lp
- 1] != new->wm_lp_spr
[wm_lp
- 1])
2540 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2541 for (; wm_lp
<= 3; wm_lp
++)
2542 dirty
|= WM_DIRTY_LP(wm_lp
);
2547 static bool _ilk_disable_lp_wm(struct drm_i915_private
*dev_priv
,
2550 struct ilk_wm_values
*previous
= &dev_priv
->wm
.hw
;
2551 bool changed
= false;
2553 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp
[2] & WM1_LP_SR_EN
) {
2554 previous
->wm_lp
[2] &= ~WM1_LP_SR_EN
;
2555 I915_WRITE(WM3_LP_ILK
, previous
->wm_lp
[2]);
2558 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp
[1] & WM1_LP_SR_EN
) {
2559 previous
->wm_lp
[1] &= ~WM1_LP_SR_EN
;
2560 I915_WRITE(WM2_LP_ILK
, previous
->wm_lp
[1]);
2563 if (dirty
& WM_DIRTY_LP(1) && previous
->wm_lp
[0] & WM1_LP_SR_EN
) {
2564 previous
->wm_lp
[0] &= ~WM1_LP_SR_EN
;
2565 I915_WRITE(WM1_LP_ILK
, previous
->wm_lp
[0]);
2570 * Don't touch WM1S_LP_EN here.
2571 * Doing so could cause underruns.
2578 * The spec says we shouldn't write when we don't need, because every write
2579 * causes WMs to be re-evaluated, expending some power.
2581 static void ilk_write_wm_values(struct drm_i915_private
*dev_priv
,
2582 struct ilk_wm_values
*results
)
2584 struct drm_device
*dev
= dev_priv
->dev
;
2585 struct ilk_wm_values
*previous
= &dev_priv
->wm
.hw
;
2589 dirty
= ilk_compute_wm_dirty(dev
, previous
, results
);
2593 _ilk_disable_lp_wm(dev_priv
, dirty
);
2595 if (dirty
& WM_DIRTY_PIPE(PIPE_A
))
2596 I915_WRITE(WM0_PIPEA_ILK
, results
->wm_pipe
[0]);
2597 if (dirty
& WM_DIRTY_PIPE(PIPE_B
))
2598 I915_WRITE(WM0_PIPEB_ILK
, results
->wm_pipe
[1]);
2599 if (dirty
& WM_DIRTY_PIPE(PIPE_C
))
2600 I915_WRITE(WM0_PIPEC_IVB
, results
->wm_pipe
[2]);
2602 if (dirty
& WM_DIRTY_LINETIME(PIPE_A
))
2603 I915_WRITE(PIPE_WM_LINETIME(PIPE_A
), results
->wm_linetime
[0]);
2604 if (dirty
& WM_DIRTY_LINETIME(PIPE_B
))
2605 I915_WRITE(PIPE_WM_LINETIME(PIPE_B
), results
->wm_linetime
[1]);
2606 if (dirty
& WM_DIRTY_LINETIME(PIPE_C
))
2607 I915_WRITE(PIPE_WM_LINETIME(PIPE_C
), results
->wm_linetime
[2]);
2609 if (dirty
& WM_DIRTY_DDB
) {
2610 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
)) {
2611 val
= I915_READ(WM_MISC
);
2612 if (results
->partitioning
== INTEL_DDB_PART_1_2
)
2613 val
&= ~WM_MISC_DATA_PARTITION_5_6
;
2615 val
|= WM_MISC_DATA_PARTITION_5_6
;
2616 I915_WRITE(WM_MISC
, val
);
2618 val
= I915_READ(DISP_ARB_CTL2
);
2619 if (results
->partitioning
== INTEL_DDB_PART_1_2
)
2620 val
&= ~DISP_DATA_PARTITION_5_6
;
2622 val
|= DISP_DATA_PARTITION_5_6
;
2623 I915_WRITE(DISP_ARB_CTL2
, val
);
2627 if (dirty
& WM_DIRTY_FBC
) {
2628 val
= I915_READ(DISP_ARB_CTL
);
2629 if (results
->enable_fbc_wm
)
2630 val
&= ~DISP_FBC_WM_DIS
;
2632 val
|= DISP_FBC_WM_DIS
;
2633 I915_WRITE(DISP_ARB_CTL
, val
);
2636 if (dirty
& WM_DIRTY_LP(1) &&
2637 previous
->wm_lp_spr
[0] != results
->wm_lp_spr
[0])
2638 I915_WRITE(WM1S_LP_ILK
, results
->wm_lp_spr
[0]);
2640 if (INTEL_INFO(dev
)->gen
>= 7) {
2641 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp_spr
[1] != results
->wm_lp_spr
[1])
2642 I915_WRITE(WM2S_LP_IVB
, results
->wm_lp_spr
[1]);
2643 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp_spr
[2] != results
->wm_lp_spr
[2])
2644 I915_WRITE(WM3S_LP_IVB
, results
->wm_lp_spr
[2]);
2647 if (dirty
& WM_DIRTY_LP(1) && previous
->wm_lp
[0] != results
->wm_lp
[0])
2648 I915_WRITE(WM1_LP_ILK
, results
->wm_lp
[0]);
2649 if (dirty
& WM_DIRTY_LP(2) && previous
->wm_lp
[1] != results
->wm_lp
[1])
2650 I915_WRITE(WM2_LP_ILK
, results
->wm_lp
[1]);
2651 if (dirty
& WM_DIRTY_LP(3) && previous
->wm_lp
[2] != results
->wm_lp
[2])
2652 I915_WRITE(WM3_LP_ILK
, results
->wm_lp
[2]);
2654 dev_priv
->wm
.hw
= *results
;
2657 static bool ilk_disable_lp_wm(struct drm_device
*dev
)
2659 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2661 return _ilk_disable_lp_wm(dev_priv
, WM_DIRTY_LP_ALL
);
2664 static void ilk_update_wm(struct drm_crtc
*crtc
)
2666 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2667 struct drm_device
*dev
= crtc
->dev
;
2668 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2669 struct ilk_wm_maximums max
;
2670 struct ilk_pipe_wm_parameters params
= {};
2671 struct ilk_wm_values results
= {};
2672 enum intel_ddb_partitioning partitioning
;
2673 struct intel_pipe_wm pipe_wm
= {};
2674 struct intel_pipe_wm lp_wm_1_2
= {}, lp_wm_5_6
= {}, *best_lp_wm
;
2675 struct intel_wm_config config
= {};
2677 ilk_compute_wm_parameters(crtc
, ¶ms
);
2679 intel_compute_pipe_wm(crtc
, ¶ms
, &pipe_wm
);
2681 if (!memcmp(&intel_crtc
->wm
.active
, &pipe_wm
, sizeof(pipe_wm
)))
2684 intel_crtc
->wm
.active
= pipe_wm
;
2686 ilk_compute_wm_config(dev
, &config
);
2688 ilk_compute_wm_maximums(dev
, 1, &config
, INTEL_DDB_PART_1_2
, &max
);
2689 ilk_wm_merge(dev
, &config
, &max
, &lp_wm_1_2
);
2691 /* 5/6 split only in single pipe config on IVB+ */
2692 if (INTEL_INFO(dev
)->gen
>= 7 &&
2693 config
.num_pipes_active
== 1 && config
.sprites_enabled
) {
2694 ilk_compute_wm_maximums(dev
, 1, &config
, INTEL_DDB_PART_5_6
, &max
);
2695 ilk_wm_merge(dev
, &config
, &max
, &lp_wm_5_6
);
2697 best_lp_wm
= ilk_find_best_result(dev
, &lp_wm_1_2
, &lp_wm_5_6
);
2699 best_lp_wm
= &lp_wm_1_2
;
2702 partitioning
= (best_lp_wm
== &lp_wm_1_2
) ?
2703 INTEL_DDB_PART_1_2
: INTEL_DDB_PART_5_6
;
2705 ilk_compute_wm_results(dev
, best_lp_wm
, partitioning
, &results
);
2707 ilk_write_wm_values(dev_priv
, &results
);
2710 static void ilk_update_sprite_wm(struct drm_plane
*plane
,
2711 struct drm_crtc
*crtc
,
2712 uint32_t sprite_width
, int pixel_size
,
2713 bool enabled
, bool scaled
)
2715 struct drm_device
*dev
= plane
->dev
;
2716 struct intel_plane
*intel_plane
= to_intel_plane(plane
);
2718 intel_plane
->wm
.enabled
= enabled
;
2719 intel_plane
->wm
.scaled
= scaled
;
2720 intel_plane
->wm
.horiz_pixels
= sprite_width
;
2721 intel_plane
->wm
.bytes_per_pixel
= pixel_size
;
2724 * IVB workaround: must disable low power watermarks for at least
2725 * one frame before enabling scaling. LP watermarks can be re-enabled
2726 * when scaling is disabled.
2728 * WaCxSRDisabledForSpriteScaling:ivb
2730 if (IS_IVYBRIDGE(dev
) && scaled
&& ilk_disable_lp_wm(dev
))
2731 intel_wait_for_vblank(dev
, intel_plane
->pipe
);
2733 ilk_update_wm(crtc
);
2736 static void ilk_pipe_wm_get_hw_state(struct drm_crtc
*crtc
)
2738 struct drm_device
*dev
= crtc
->dev
;
2739 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2740 struct ilk_wm_values
*hw
= &dev_priv
->wm
.hw
;
2741 struct intel_crtc
*intel_crtc
= to_intel_crtc(crtc
);
2742 struct intel_pipe_wm
*active
= &intel_crtc
->wm
.active
;
2743 enum pipe pipe
= intel_crtc
->pipe
;
2744 static const unsigned int wm0_pipe_reg
[] = {
2745 [PIPE_A
] = WM0_PIPEA_ILK
,
2746 [PIPE_B
] = WM0_PIPEB_ILK
,
2747 [PIPE_C
] = WM0_PIPEC_IVB
,
2750 hw
->wm_pipe
[pipe
] = I915_READ(wm0_pipe_reg
[pipe
]);
2751 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2752 hw
->wm_linetime
[pipe
] = I915_READ(PIPE_WM_LINETIME(pipe
));
2754 active
->pipe_enabled
= intel_crtc_active(crtc
);
2756 if (active
->pipe_enabled
) {
2757 u32 tmp
= hw
->wm_pipe
[pipe
];
2760 * For active pipes LP0 watermark is marked as
2761 * enabled, and LP1+ watermaks as disabled since
2762 * we can't really reverse compute them in case
2763 * multiple pipes are active.
2765 active
->wm
[0].enable
= true;
2766 active
->wm
[0].pri_val
= (tmp
& WM0_PIPE_PLANE_MASK
) >> WM0_PIPE_PLANE_SHIFT
;
2767 active
->wm
[0].spr_val
= (tmp
& WM0_PIPE_SPRITE_MASK
) >> WM0_PIPE_SPRITE_SHIFT
;
2768 active
->wm
[0].cur_val
= tmp
& WM0_PIPE_CURSOR_MASK
;
2769 active
->linetime
= hw
->wm_linetime
[pipe
];
2771 int level
, max_level
= ilk_wm_max_level(dev
);
2774 * For inactive pipes, all watermark levels
2775 * should be marked as enabled but zeroed,
2776 * which is what we'd compute them to.
2778 for (level
= 0; level
<= max_level
; level
++)
2779 active
->wm
[level
].enable
= true;
2783 void ilk_wm_get_hw_state(struct drm_device
*dev
)
2785 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2786 struct ilk_wm_values
*hw
= &dev_priv
->wm
.hw
;
2787 struct drm_crtc
*crtc
;
2789 for_each_crtc(dev
, crtc
)
2790 ilk_pipe_wm_get_hw_state(crtc
);
2792 hw
->wm_lp
[0] = I915_READ(WM1_LP_ILK
);
2793 hw
->wm_lp
[1] = I915_READ(WM2_LP_ILK
);
2794 hw
->wm_lp
[2] = I915_READ(WM3_LP_ILK
);
2796 hw
->wm_lp_spr
[0] = I915_READ(WM1S_LP_ILK
);
2797 if (INTEL_INFO(dev
)->gen
>= 7) {
2798 hw
->wm_lp_spr
[1] = I915_READ(WM2S_LP_IVB
);
2799 hw
->wm_lp_spr
[2] = I915_READ(WM3S_LP_IVB
);
2802 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
2803 hw
->partitioning
= (I915_READ(WM_MISC
) & WM_MISC_DATA_PARTITION_5_6
) ?
2804 INTEL_DDB_PART_5_6
: INTEL_DDB_PART_1_2
;
2805 else if (IS_IVYBRIDGE(dev
))
2806 hw
->partitioning
= (I915_READ(DISP_ARB_CTL2
) & DISP_DATA_PARTITION_5_6
) ?
2807 INTEL_DDB_PART_5_6
: INTEL_DDB_PART_1_2
;
2810 !(I915_READ(DISP_ARB_CTL
) & DISP_FBC_WM_DIS
);
2814 * intel_update_watermarks - update FIFO watermark values based on current modes
2816 * Calculate watermark values for the various WM regs based on current mode
2817 * and plane configuration.
2819 * There are several cases to deal with here:
2820 * - normal (i.e. non-self-refresh)
2821 * - self-refresh (SR) mode
2822 * - lines are large relative to FIFO size (buffer can hold up to 2)
2823 * - lines are small relative to FIFO size (buffer can hold more than 2
2824 * lines), so need to account for TLB latency
2826 * The normal calculation is:
2827 * watermark = dotclock * bytes per pixel * latency
2828 * where latency is platform & configuration dependent (we assume pessimal
2831 * The SR calculation is:
2832 * watermark = (trunc(latency/line time)+1) * surface width *
2835 * line time = htotal / dotclock
2836 * surface width = hdisplay for normal plane and 64 for cursor
2837 * and latency is assumed to be high, as above.
2839 * The final value programmed to the register should always be rounded up,
2840 * and include an extra 2 entries to account for clock crossings.
2842 * We don't use the sprite, so we can ignore that. And on Crestline we have
2843 * to set the non-SR watermarks to 8.
2845 void intel_update_watermarks(struct drm_crtc
*crtc
)
2847 struct drm_i915_private
*dev_priv
= crtc
->dev
->dev_private
;
2849 if (dev_priv
->display
.update_wm
)
2850 dev_priv
->display
.update_wm(crtc
);
2853 void intel_update_sprite_watermarks(struct drm_plane
*plane
,
2854 struct drm_crtc
*crtc
,
2855 uint32_t sprite_width
, int pixel_size
,
2856 bool enabled
, bool scaled
)
2858 struct drm_i915_private
*dev_priv
= plane
->dev
->dev_private
;
2860 if (dev_priv
->display
.update_sprite_wm
)
2861 dev_priv
->display
.update_sprite_wm(plane
, crtc
, sprite_width
,
2862 pixel_size
, enabled
, scaled
);
2865 static struct drm_i915_gem_object
*
2866 intel_alloc_context_page(struct drm_device
*dev
)
2868 struct drm_i915_gem_object
*ctx
;
2871 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
2873 ctx
= i915_gem_alloc_object(dev
, 4096);
2875 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2879 ret
= i915_gem_obj_ggtt_pin(ctx
, 4096, 0);
2881 DRM_ERROR("failed to pin power context: %d\n", ret
);
2885 ret
= i915_gem_object_set_to_gtt_domain(ctx
, 1);
2887 DRM_ERROR("failed to set-domain on power context: %d\n", ret
);
2894 i915_gem_object_ggtt_unpin(ctx
);
2896 drm_gem_object_unreference(&ctx
->base
);
2901 * Lock protecting IPS related data structures
2903 DEFINE_SPINLOCK(mchdev_lock
);
2905 /* Global for IPS driver to get at the current i915 device. Protected by
2907 static struct drm_i915_private
*i915_mch_dev
;
2909 bool ironlake_set_drps(struct drm_device
*dev
, u8 val
)
2911 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2914 assert_spin_locked(&mchdev_lock
);
2916 rgvswctl
= I915_READ16(MEMSWCTL
);
2917 if (rgvswctl
& MEMCTL_CMD_STS
) {
2918 DRM_DEBUG("gpu busy, RCS change rejected\n");
2919 return false; /* still busy with another command */
2922 rgvswctl
= (MEMCTL_CMD_CHFREQ
<< MEMCTL_CMD_SHIFT
) |
2923 (val
<< MEMCTL_FREQ_SHIFT
) | MEMCTL_SFCAVM
;
2924 I915_WRITE16(MEMSWCTL
, rgvswctl
);
2925 POSTING_READ16(MEMSWCTL
);
2927 rgvswctl
|= MEMCTL_CMD_STS
;
2928 I915_WRITE16(MEMSWCTL
, rgvswctl
);
2933 static void ironlake_enable_drps(struct drm_device
*dev
)
2935 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
2936 u32 rgvmodectl
= I915_READ(MEMMODECTL
);
2937 u8 fmax
, fmin
, fstart
, vstart
;
2939 spin_lock_irq(&mchdev_lock
);
2941 /* Enable temp reporting */
2942 I915_WRITE16(PMMISC
, I915_READ(PMMISC
) | MCPPCE_EN
);
2943 I915_WRITE16(TSC1
, I915_READ(TSC1
) | TSE
);
2945 /* 100ms RC evaluation intervals */
2946 I915_WRITE(RCUPEI
, 100000);
2947 I915_WRITE(RCDNEI
, 100000);
2949 /* Set max/min thresholds to 90ms and 80ms respectively */
2950 I915_WRITE(RCBMAXAVG
, 90000);
2951 I915_WRITE(RCBMINAVG
, 80000);
2953 I915_WRITE(MEMIHYST
, 1);
2955 /* Set up min, max, and cur for interrupt handling */
2956 fmax
= (rgvmodectl
& MEMMODE_FMAX_MASK
) >> MEMMODE_FMAX_SHIFT
;
2957 fmin
= (rgvmodectl
& MEMMODE_FMIN_MASK
);
2958 fstart
= (rgvmodectl
& MEMMODE_FSTART_MASK
) >>
2959 MEMMODE_FSTART_SHIFT
;
2961 vstart
= (I915_READ(PXVFREQ_BASE
+ (fstart
* 4)) & PXVFREQ_PX_MASK
) >>
2964 dev_priv
->ips
.fmax
= fmax
; /* IPS callback will increase this */
2965 dev_priv
->ips
.fstart
= fstart
;
2967 dev_priv
->ips
.max_delay
= fstart
;
2968 dev_priv
->ips
.min_delay
= fmin
;
2969 dev_priv
->ips
.cur_delay
= fstart
;
2971 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2972 fmax
, fmin
, fstart
);
2974 I915_WRITE(MEMINTREN
, MEMINT_CX_SUPR_EN
| MEMINT_EVAL_CHG_EN
);
2977 * Interrupts will be enabled in ironlake_irq_postinstall
2980 I915_WRITE(VIDSTART
, vstart
);
2981 POSTING_READ(VIDSTART
);
2983 rgvmodectl
|= MEMMODE_SWMODE_EN
;
2984 I915_WRITE(MEMMODECTL
, rgvmodectl
);
2986 if (wait_for_atomic((I915_READ(MEMSWCTL
) & MEMCTL_CMD_STS
) == 0, 10))
2987 DRM_ERROR("stuck trying to change perf mode\n");
2990 ironlake_set_drps(dev
, fstart
);
2992 dev_priv
->ips
.last_count1
= I915_READ(0x112e4) + I915_READ(0x112e8) +
2994 dev_priv
->ips
.last_time1
= jiffies_to_msecs(jiffies
);
2995 dev_priv
->ips
.last_count2
= I915_READ(0x112f4);
2996 getrawmonotonic(&dev_priv
->ips
.last_time2
);
2998 spin_unlock_irq(&mchdev_lock
);
3001 static void ironlake_disable_drps(struct drm_device
*dev
)
3003 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3006 spin_lock_irq(&mchdev_lock
);
3008 rgvswctl
= I915_READ16(MEMSWCTL
);
3010 /* Ack interrupts, disable EFC interrupt */
3011 I915_WRITE(MEMINTREN
, I915_READ(MEMINTREN
) & ~MEMINT_EVAL_CHG_EN
);
3012 I915_WRITE(MEMINTRSTS
, MEMINT_EVAL_CHG
);
3013 I915_WRITE(DEIER
, I915_READ(DEIER
) & ~DE_PCU_EVENT
);
3014 I915_WRITE(DEIIR
, DE_PCU_EVENT
);
3015 I915_WRITE(DEIMR
, I915_READ(DEIMR
) | DE_PCU_EVENT
);
3017 /* Go back to the starting frequency */
3018 ironlake_set_drps(dev
, dev_priv
->ips
.fstart
);
3020 rgvswctl
|= MEMCTL_CMD_STS
;
3021 I915_WRITE(MEMSWCTL
, rgvswctl
);
3024 spin_unlock_irq(&mchdev_lock
);
3027 /* There's a funny hw issue where the hw returns all 0 when reading from
3028 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
3029 * ourselves, instead of doing a rmw cycle (which might result in us clearing
3030 * all limits and the gpu stuck at whatever frequency it is at atm).
3032 static u32
gen6_rps_limits(struct drm_i915_private
*dev_priv
, u8 val
)
3036 /* Only set the down limit when we've reached the lowest level to avoid
3037 * getting more interrupts, otherwise leave this clear. This prevents a
3038 * race in the hw when coming out of rc6: There's a tiny window where
3039 * the hw runs at the minimal clock before selecting the desired
3040 * frequency, if the down threshold expires in that window we will not
3041 * receive a down interrupt. */
3042 limits
= dev_priv
->rps
.max_freq_softlimit
<< 24;
3043 if (val
<= dev_priv
->rps
.min_freq_softlimit
)
3044 limits
|= dev_priv
->rps
.min_freq_softlimit
<< 16;
3049 static void gen6_set_rps_thresholds(struct drm_i915_private
*dev_priv
, u8 val
)
3053 new_power
= dev_priv
->rps
.power
;
3054 switch (dev_priv
->rps
.power
) {
3056 if (val
> dev_priv
->rps
.efficient_freq
+ 1 && val
> dev_priv
->rps
.cur_freq
)
3057 new_power
= BETWEEN
;
3061 if (val
<= dev_priv
->rps
.efficient_freq
&& val
< dev_priv
->rps
.cur_freq
)
3062 new_power
= LOW_POWER
;
3063 else if (val
>= dev_priv
->rps
.rp0_freq
&& val
> dev_priv
->rps
.cur_freq
)
3064 new_power
= HIGH_POWER
;
3068 if (val
< (dev_priv
->rps
.rp1_freq
+ dev_priv
->rps
.rp0_freq
) >> 1 && val
< dev_priv
->rps
.cur_freq
)
3069 new_power
= BETWEEN
;
3072 /* Max/min bins are special */
3073 if (val
== dev_priv
->rps
.min_freq_softlimit
)
3074 new_power
= LOW_POWER
;
3075 if (val
== dev_priv
->rps
.max_freq_softlimit
)
3076 new_power
= HIGH_POWER
;
3077 if (new_power
== dev_priv
->rps
.power
)
3080 /* Note the units here are not exactly 1us, but 1280ns. */
3081 switch (new_power
) {
3083 /* Upclock if more than 95% busy over 16ms */
3084 I915_WRITE(GEN6_RP_UP_EI
, 12500);
3085 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 11800);
3087 /* Downclock if less than 85% busy over 32ms */
3088 I915_WRITE(GEN6_RP_DOWN_EI
, 25000);
3089 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 21250);
3091 I915_WRITE(GEN6_RP_CONTROL
,
3092 GEN6_RP_MEDIA_TURBO
|
3093 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
3094 GEN6_RP_MEDIA_IS_GFX
|
3096 GEN6_RP_UP_BUSY_AVG
|
3097 GEN6_RP_DOWN_IDLE_AVG
);
3101 /* Upclock if more than 90% busy over 13ms */
3102 I915_WRITE(GEN6_RP_UP_EI
, 10250);
3103 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 9225);
3105 /* Downclock if less than 75% busy over 32ms */
3106 I915_WRITE(GEN6_RP_DOWN_EI
, 25000);
3107 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 18750);
3109 I915_WRITE(GEN6_RP_CONTROL
,
3110 GEN6_RP_MEDIA_TURBO
|
3111 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
3112 GEN6_RP_MEDIA_IS_GFX
|
3114 GEN6_RP_UP_BUSY_AVG
|
3115 GEN6_RP_DOWN_IDLE_AVG
);
3119 /* Upclock if more than 85% busy over 10ms */
3120 I915_WRITE(GEN6_RP_UP_EI
, 8000);
3121 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 6800);
3123 /* Downclock if less than 60% busy over 32ms */
3124 I915_WRITE(GEN6_RP_DOWN_EI
, 25000);
3125 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 15000);
3127 I915_WRITE(GEN6_RP_CONTROL
,
3128 GEN6_RP_MEDIA_TURBO
|
3129 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
3130 GEN6_RP_MEDIA_IS_GFX
|
3132 GEN6_RP_UP_BUSY_AVG
|
3133 GEN6_RP_DOWN_IDLE_AVG
);
3137 dev_priv
->rps
.power
= new_power
;
3138 dev_priv
->rps
.last_adj
= 0;
3141 static u32
gen6_rps_pm_mask(struct drm_i915_private
*dev_priv
, u8 val
)
3145 if (val
> dev_priv
->rps
.min_freq_softlimit
)
3146 mask
|= GEN6_PM_RP_DOWN_THRESHOLD
| GEN6_PM_RP_DOWN_TIMEOUT
;
3147 if (val
< dev_priv
->rps
.max_freq_softlimit
)
3148 mask
|= GEN6_PM_RP_UP_THRESHOLD
;
3150 /* IVB and SNB hard hangs on looping batchbuffer
3151 * if GEN6_PM_UP_EI_EXPIRED is masked.
3153 if (INTEL_INFO(dev_priv
->dev
)->gen
<= 7 && !IS_HASWELL(dev_priv
->dev
))
3154 mask
|= GEN6_PM_RP_UP_EI_EXPIRED
;
3156 if (IS_GEN8(dev_priv
->dev
))
3157 mask
|= GEN8_PMINTR_REDIRECT_TO_NON_DISP
;
3162 /* gen6_set_rps is called to update the frequency request, but should also be
3163 * called when the range (min_delay and max_delay) is modified so that we can
3164 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
3165 void gen6_set_rps(struct drm_device
*dev
, u8 val
)
3167 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3169 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
3170 WARN_ON(val
> dev_priv
->rps
.max_freq_softlimit
);
3171 WARN_ON(val
< dev_priv
->rps
.min_freq_softlimit
);
3173 /* min/max delay may still have been modified so be sure to
3174 * write the limits value.
3176 if (val
!= dev_priv
->rps
.cur_freq
) {
3177 gen6_set_rps_thresholds(dev_priv
, val
);
3179 if (IS_HASWELL(dev
) || IS_BROADWELL(dev
))
3180 I915_WRITE(GEN6_RPNSWREQ
,
3181 HSW_FREQUENCY(val
));
3183 I915_WRITE(GEN6_RPNSWREQ
,
3184 GEN6_FREQUENCY(val
) |
3186 GEN6_AGGRESSIVE_TURBO
);
3189 /* Make sure we continue to get interrupts
3190 * until we hit the minimum or maximum frequencies.
3192 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
, gen6_rps_limits(dev_priv
, val
));
3193 I915_WRITE(GEN6_PMINTRMSK
, gen6_rps_pm_mask(dev_priv
, val
));
3195 POSTING_READ(GEN6_RPNSWREQ
);
3197 dev_priv
->rps
.cur_freq
= val
;
3198 trace_intel_gpu_freq_change(val
* 50);
3201 /* vlv_set_rps_idle: Set the frequency to Rpn if Gfx clocks are down
3203 * * If Gfx is Idle, then
3204 * 1. Mask Turbo interrupts
3205 * 2. Bring up Gfx clock
3206 * 3. Change the freq to Rpn and wait till P-Unit updates freq
3207 * 4. Clear the Force GFX CLK ON bit so that Gfx can down
3208 * 5. Unmask Turbo interrupts
3210 static void vlv_set_rps_idle(struct drm_i915_private
*dev_priv
)
3213 * When we are idle. Drop to min voltage state.
3216 if (dev_priv
->rps
.cur_freq
<= dev_priv
->rps
.min_freq_softlimit
)
3219 /* Mask turbo interrupt so that they will not come in between */
3220 I915_WRITE(GEN6_PMINTRMSK
, 0xffffffff);
3222 vlv_force_gfx_clock(dev_priv
, true);
3224 dev_priv
->rps
.cur_freq
= dev_priv
->rps
.min_freq_softlimit
;
3226 vlv_punit_write(dev_priv
, PUNIT_REG_GPU_FREQ_REQ
,
3227 dev_priv
->rps
.min_freq_softlimit
);
3229 if (wait_for(((vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
))
3230 & GENFREQSTATUS
) == 0, 5))
3231 DRM_ERROR("timed out waiting for Punit\n");
3233 vlv_force_gfx_clock(dev_priv
, false);
3235 I915_WRITE(GEN6_PMINTRMSK
,
3236 gen6_rps_pm_mask(dev_priv
, dev_priv
->rps
.cur_freq
));
3239 void gen6_rps_idle(struct drm_i915_private
*dev_priv
)
3241 struct drm_device
*dev
= dev_priv
->dev
;
3243 mutex_lock(&dev_priv
->rps
.hw_lock
);
3244 if (dev_priv
->rps
.enabled
) {
3245 if (IS_VALLEYVIEW(dev
))
3246 vlv_set_rps_idle(dev_priv
);
3248 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.min_freq_softlimit
);
3249 dev_priv
->rps
.last_adj
= 0;
3251 mutex_unlock(&dev_priv
->rps
.hw_lock
);
3254 void gen6_rps_boost(struct drm_i915_private
*dev_priv
)
3256 struct drm_device
*dev
= dev_priv
->dev
;
3258 mutex_lock(&dev_priv
->rps
.hw_lock
);
3259 if (dev_priv
->rps
.enabled
) {
3260 if (IS_VALLEYVIEW(dev
))
3261 valleyview_set_rps(dev_priv
->dev
, dev_priv
->rps
.max_freq_softlimit
);
3263 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.max_freq_softlimit
);
3264 dev_priv
->rps
.last_adj
= 0;
3266 mutex_unlock(&dev_priv
->rps
.hw_lock
);
3269 void valleyview_set_rps(struct drm_device
*dev
, u8 val
)
3271 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3273 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
3274 WARN_ON(val
> dev_priv
->rps
.max_freq_softlimit
);
3275 WARN_ON(val
< dev_priv
->rps
.min_freq_softlimit
);
3277 DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
3278 vlv_gpu_freq(dev_priv
, dev_priv
->rps
.cur_freq
),
3279 dev_priv
->rps
.cur_freq
,
3280 vlv_gpu_freq(dev_priv
, val
), val
);
3282 if (val
!= dev_priv
->rps
.cur_freq
)
3283 vlv_punit_write(dev_priv
, PUNIT_REG_GPU_FREQ_REQ
, val
);
3285 I915_WRITE(GEN6_PMINTRMSK
, gen6_rps_pm_mask(dev_priv
, val
));
3287 dev_priv
->rps
.cur_freq
= val
;
3288 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv
, val
));
3291 static void gen8_disable_rps_interrupts(struct drm_device
*dev
)
3293 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3295 I915_WRITE(GEN6_PMINTRMSK
, ~GEN8_PMINTR_REDIRECT_TO_NON_DISP
);
3296 I915_WRITE(GEN8_GT_IER(2), I915_READ(GEN8_GT_IER(2)) &
3297 ~dev_priv
->pm_rps_events
);
3298 /* Complete PM interrupt masking here doesn't race with the rps work
3299 * item again unmasking PM interrupts because that is using a different
3300 * register (GEN8_GT_IMR(2)) to mask PM interrupts. The only risk is in
3301 * leaving stale bits in GEN8_GT_IIR(2) and GEN8_GT_IMR(2) which
3302 * gen8_enable_rps will clean up. */
3304 spin_lock_irq(&dev_priv
->irq_lock
);
3305 dev_priv
->rps
.pm_iir
= 0;
3306 spin_unlock_irq(&dev_priv
->irq_lock
);
3308 I915_WRITE(GEN8_GT_IIR(2), dev_priv
->pm_rps_events
);
3311 static void gen6_disable_rps_interrupts(struct drm_device
*dev
)
3313 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3315 I915_WRITE(GEN6_PMINTRMSK
, 0xffffffff);
3316 I915_WRITE(GEN6_PMIER
, I915_READ(GEN6_PMIER
) &
3317 ~dev_priv
->pm_rps_events
);
3318 /* Complete PM interrupt masking here doesn't race with the rps work
3319 * item again unmasking PM interrupts because that is using a different
3320 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
3321 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
3323 spin_lock_irq(&dev_priv
->irq_lock
);
3324 dev_priv
->rps
.pm_iir
= 0;
3325 spin_unlock_irq(&dev_priv
->irq_lock
);
3327 I915_WRITE(GEN6_PMIIR
, dev_priv
->pm_rps_events
);
3330 static void gen6_disable_rps(struct drm_device
*dev
)
3332 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3334 I915_WRITE(GEN6_RC_CONTROL
, 0);
3335 I915_WRITE(GEN6_RPNSWREQ
, 1 << 31);
3337 if (IS_BROADWELL(dev
))
3338 gen8_disable_rps_interrupts(dev
);
3340 gen6_disable_rps_interrupts(dev
);
3343 static void valleyview_disable_rps(struct drm_device
*dev
)
3345 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3347 I915_WRITE(GEN6_RC_CONTROL
, 0);
3349 gen6_disable_rps_interrupts(dev
);
3352 static void intel_print_rc6_info(struct drm_device
*dev
, u32 mode
)
3354 if (IS_VALLEYVIEW(dev
)) {
3355 if (mode
& (GEN7_RC_CTL_TO_MODE
| GEN6_RC_CTL_EI_MODE(1)))
3356 mode
= GEN6_RC_CTL_RC6_ENABLE
;
3360 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
3361 (mode
& GEN6_RC_CTL_RC6_ENABLE
) ? "on" : "off",
3362 (mode
& GEN6_RC_CTL_RC6p_ENABLE
) ? "on" : "off",
3363 (mode
& GEN6_RC_CTL_RC6pp_ENABLE
) ? "on" : "off");
3366 static int sanitize_rc6_option(const struct drm_device
*dev
, int enable_rc6
)
3368 /* No RC6 before Ironlake */
3369 if (INTEL_INFO(dev
)->gen
< 5)
3372 /* RC6 is only on Ironlake mobile not on desktop */
3373 if (INTEL_INFO(dev
)->gen
== 5 && !IS_IRONLAKE_M(dev
))
3376 /* Respect the kernel parameter if it is set */
3377 if (enable_rc6
>= 0) {
3380 if (INTEL_INFO(dev
)->gen
== 6 || IS_IVYBRIDGE(dev
))
3381 mask
= INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
|
3384 mask
= INTEL_RC6_ENABLE
;
3386 if ((enable_rc6
& mask
) != enable_rc6
)
3387 DRM_INFO("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
3388 enable_rc6
& mask
, enable_rc6
, mask
);
3390 return enable_rc6
& mask
;
3393 /* Disable RC6 on Ironlake */
3394 if (INTEL_INFO(dev
)->gen
== 5)
3397 if (IS_IVYBRIDGE(dev
))
3398 return (INTEL_RC6_ENABLE
| INTEL_RC6p_ENABLE
);
3400 return INTEL_RC6_ENABLE
;
3403 int intel_enable_rc6(const struct drm_device
*dev
)
3405 return i915
.enable_rc6
;
3408 static void gen8_enable_rps_interrupts(struct drm_device
*dev
)
3410 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3412 spin_lock_irq(&dev_priv
->irq_lock
);
3413 WARN_ON(dev_priv
->rps
.pm_iir
);
3414 bdw_enable_pm_irq(dev_priv
, dev_priv
->pm_rps_events
);
3415 I915_WRITE(GEN8_GT_IIR(2), dev_priv
->pm_rps_events
);
3416 spin_unlock_irq(&dev_priv
->irq_lock
);
3419 static void gen6_enable_rps_interrupts(struct drm_device
*dev
)
3421 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3423 spin_lock_irq(&dev_priv
->irq_lock
);
3424 WARN_ON(dev_priv
->rps
.pm_iir
);
3425 snb_enable_pm_irq(dev_priv
, dev_priv
->pm_rps_events
);
3426 I915_WRITE(GEN6_PMIIR
, dev_priv
->pm_rps_events
);
3427 spin_unlock_irq(&dev_priv
->irq_lock
);
3430 static void parse_rp_state_cap(struct drm_i915_private
*dev_priv
, u32 rp_state_cap
)
3432 /* All of these values are in units of 50MHz */
3433 dev_priv
->rps
.cur_freq
= 0;
3434 /* static values from HW: RP0 < RPe < RP1 < RPn (min_freq) */
3435 dev_priv
->rps
.rp1_freq
= (rp_state_cap
>> 8) & 0xff;
3436 dev_priv
->rps
.rp0_freq
= (rp_state_cap
>> 0) & 0xff;
3437 dev_priv
->rps
.min_freq
= (rp_state_cap
>> 16) & 0xff;
3438 /* XXX: only BYT has a special efficient freq */
3439 dev_priv
->rps
.efficient_freq
= dev_priv
->rps
.rp1_freq
;
3440 /* hw_max = RP0 until we check for overclocking */
3441 dev_priv
->rps
.max_freq
= dev_priv
->rps
.rp0_freq
;
3443 /* Preserve min/max settings in case of re-init */
3444 if (dev_priv
->rps
.max_freq_softlimit
== 0)
3445 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
3447 if (dev_priv
->rps
.min_freq_softlimit
== 0)
3448 dev_priv
->rps
.min_freq_softlimit
= dev_priv
->rps
.min_freq
;
3451 static void gen8_enable_rps(struct drm_device
*dev
)
3453 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3454 struct intel_engine_cs
*ring
;
3455 uint32_t rc6_mask
= 0, rp_state_cap
;
3458 /* 1a: Software RC state - RC0 */
3459 I915_WRITE(GEN6_RC_STATE
, 0);
3461 /* 1c & 1d: Get forcewake during program sequence. Although the driver
3462 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
3463 gen6_gt_force_wake_get(dev_priv
, FORCEWAKE_ALL
);
3465 /* 2a: Disable RC states. */
3466 I915_WRITE(GEN6_RC_CONTROL
, 0);
3468 rp_state_cap
= I915_READ(GEN6_RP_STATE_CAP
);
3469 parse_rp_state_cap(dev_priv
, rp_state_cap
);
3471 /* 2b: Program RC6 thresholds.*/
3472 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16);
3473 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000); /* 12500 * 1280ns */
3474 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25); /* 25 * 1280ns */
3475 for_each_ring(ring
, dev_priv
, unused
)
3476 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
3477 I915_WRITE(GEN6_RC_SLEEP
, 0);
3478 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000); /* 50/125ms per EI */
3481 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
3482 rc6_mask
= GEN6_RC_CTL_RC6_ENABLE
;
3483 intel_print_rc6_info(dev
, rc6_mask
);
3484 I915_WRITE(GEN6_RC_CONTROL
, GEN6_RC_CTL_HW_ENABLE
|
3485 GEN6_RC_CTL_EI_MODE(1) |
3488 /* 4 Program defaults and thresholds for RPS*/
3489 I915_WRITE(GEN6_RPNSWREQ
,
3490 HSW_FREQUENCY(dev_priv
->rps
.rp1_freq
));
3491 I915_WRITE(GEN6_RC_VIDEO_FREQ
,
3492 HSW_FREQUENCY(dev_priv
->rps
.rp1_freq
));
3493 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
3494 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 100000000 / 128); /* 1 second timeout */
3496 /* Docs recommend 900MHz, and 300 MHz respectively */
3497 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS
,
3498 dev_priv
->rps
.max_freq_softlimit
<< 24 |
3499 dev_priv
->rps
.min_freq_softlimit
<< 16);
3501 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 7600000 / 128); /* 76ms busyness per EI, 90% */
3502 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 31300000 / 128); /* 313ms busyness per EI, 70%*/
3503 I915_WRITE(GEN6_RP_UP_EI
, 66000); /* 84.48ms, XXX: random? */
3504 I915_WRITE(GEN6_RP_DOWN_EI
, 350000); /* 448ms, XXX: random? */
3506 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
3509 I915_WRITE(GEN6_RP_CONTROL
,
3510 GEN6_RP_MEDIA_TURBO
|
3511 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
3512 GEN6_RP_MEDIA_IS_GFX
|
3514 GEN6_RP_UP_BUSY_AVG
|
3515 GEN6_RP_DOWN_IDLE_AVG
);
3517 /* 6: Ring frequency + overclocking (our driver does this later */
3519 gen6_set_rps(dev
, (I915_READ(GEN6_GT_PERF_STATUS
) & 0xff00) >> 8);
3521 gen8_enable_rps_interrupts(dev
);
3523 gen6_gt_force_wake_put(dev_priv
, FORCEWAKE_ALL
);
3526 static void gen6_enable_rps(struct drm_device
*dev
)
3528 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3529 struct intel_engine_cs
*ring
;
3532 u32 rc6vids
, pcu_mbox
= 0, rc6_mask
= 0;
3537 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
3539 /* Here begins a magic sequence of register writes to enable
3540 * auto-downclocking.
3542 * Perhaps there might be some value in exposing these to
3545 I915_WRITE(GEN6_RC_STATE
, 0);
3547 /* Clear the DBG now so we don't confuse earlier errors */
3548 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
3549 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg
);
3550 I915_WRITE(GTFIFODBG
, gtfifodbg
);
3553 gen6_gt_force_wake_get(dev_priv
, FORCEWAKE_ALL
);
3555 rp_state_cap
= I915_READ(GEN6_RP_STATE_CAP
);
3556 gt_perf_status
= I915_READ(GEN6_GT_PERF_STATUS
);
3558 parse_rp_state_cap(dev_priv
, rp_state_cap
);
3560 /* disable the counters and set deterministic thresholds */
3561 I915_WRITE(GEN6_RC_CONTROL
, 0);
3563 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT
, 1000 << 16);
3564 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 40 << 16 | 30);
3565 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT
, 30);
3566 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
3567 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
3569 for_each_ring(ring
, dev_priv
, i
)
3570 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
3572 I915_WRITE(GEN6_RC_SLEEP
, 0);
3573 I915_WRITE(GEN6_RC1e_THRESHOLD
, 1000);
3574 if (IS_IVYBRIDGE(dev
))
3575 I915_WRITE(GEN6_RC6_THRESHOLD
, 125000);
3577 I915_WRITE(GEN6_RC6_THRESHOLD
, 50000);
3578 I915_WRITE(GEN6_RC6p_THRESHOLD
, 150000);
3579 I915_WRITE(GEN6_RC6pp_THRESHOLD
, 64000); /* unused */
3581 /* Check if we are enabling RC6 */
3582 rc6_mode
= intel_enable_rc6(dev_priv
->dev
);
3583 if (rc6_mode
& INTEL_RC6_ENABLE
)
3584 rc6_mask
|= GEN6_RC_CTL_RC6_ENABLE
;
3586 /* We don't use those on Haswell */
3587 if (!IS_HASWELL(dev
)) {
3588 if (rc6_mode
& INTEL_RC6p_ENABLE
)
3589 rc6_mask
|= GEN6_RC_CTL_RC6p_ENABLE
;
3591 if (rc6_mode
& INTEL_RC6pp_ENABLE
)
3592 rc6_mask
|= GEN6_RC_CTL_RC6pp_ENABLE
;
3595 intel_print_rc6_info(dev
, rc6_mask
);
3597 I915_WRITE(GEN6_RC_CONTROL
,
3599 GEN6_RC_CTL_EI_MODE(1) |
3600 GEN6_RC_CTL_HW_ENABLE
);
3602 /* Power down if completely idle for over 50ms */
3603 I915_WRITE(GEN6_RP_DOWN_TIMEOUT
, 50000);
3604 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
3606 ret
= sandybridge_pcode_write(dev_priv
, GEN6_PCODE_WRITE_MIN_FREQ_TABLE
, 0);
3608 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
3610 ret
= sandybridge_pcode_read(dev_priv
, GEN6_READ_OC_PARAMS
, &pcu_mbox
);
3611 if (!ret
&& (pcu_mbox
& (1<<31))) { /* OC supported */
3612 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
3613 (dev_priv
->rps
.max_freq_softlimit
& 0xff) * 50,
3614 (pcu_mbox
& 0xff) * 50);
3615 dev_priv
->rps
.max_freq
= pcu_mbox
& 0xff;
3618 dev_priv
->rps
.power
= HIGH_POWER
; /* force a reset */
3619 gen6_set_rps(dev_priv
->dev
, dev_priv
->rps
.min_freq_softlimit
);
3621 gen6_enable_rps_interrupts(dev
);
3624 ret
= sandybridge_pcode_read(dev_priv
, GEN6_PCODE_READ_RC6VIDS
, &rc6vids
);
3625 if (IS_GEN6(dev
) && ret
) {
3626 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
3627 } else if (IS_GEN6(dev
) && (GEN6_DECODE_RC6_VID(rc6vids
& 0xff) < 450)) {
3628 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
3629 GEN6_DECODE_RC6_VID(rc6vids
& 0xff), 450);
3630 rc6vids
&= 0xffff00;
3631 rc6vids
|= GEN6_ENCODE_RC6_VID(450);
3632 ret
= sandybridge_pcode_write(dev_priv
, GEN6_PCODE_WRITE_RC6VIDS
, rc6vids
);
3634 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
3637 gen6_gt_force_wake_put(dev_priv
, FORCEWAKE_ALL
);
3640 static void __gen6_update_ring_freq(struct drm_device
*dev
)
3642 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3644 unsigned int gpu_freq
;
3645 unsigned int max_ia_freq
, min_ring_freq
;
3646 int scaling_factor
= 180;
3647 struct cpufreq_policy
*policy
;
3649 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
3651 policy
= cpufreq_cpu_get(0);
3653 max_ia_freq
= policy
->cpuinfo
.max_freq
;
3654 cpufreq_cpu_put(policy
);
3657 * Default to measured freq if none found, PCU will ensure we
3660 max_ia_freq
= tsc_khz
;
3663 /* Convert from kHz to MHz */
3664 max_ia_freq
/= 1000;
3666 min_ring_freq
= I915_READ(DCLK
) & 0xf;
3667 /* convert DDR frequency from units of 266.6MHz to bandwidth */
3668 min_ring_freq
= mult_frac(min_ring_freq
, 8, 3);
3671 * For each potential GPU frequency, load a ring frequency we'd like
3672 * to use for memory access. We do this by specifying the IA frequency
3673 * the PCU should use as a reference to determine the ring frequency.
3675 for (gpu_freq
= dev_priv
->rps
.max_freq_softlimit
; gpu_freq
>= dev_priv
->rps
.min_freq_softlimit
;
3677 int diff
= dev_priv
->rps
.max_freq_softlimit
- gpu_freq
;
3678 unsigned int ia_freq
= 0, ring_freq
= 0;
3680 if (INTEL_INFO(dev
)->gen
>= 8) {
3681 /* max(2 * GT, DDR). NB: GT is 50MHz units */
3682 ring_freq
= max(min_ring_freq
, gpu_freq
);
3683 } else if (IS_HASWELL(dev
)) {
3684 ring_freq
= mult_frac(gpu_freq
, 5, 4);
3685 ring_freq
= max(min_ring_freq
, ring_freq
);
3686 /* leave ia_freq as the default, chosen by cpufreq */
3688 /* On older processors, there is no separate ring
3689 * clock domain, so in order to boost the bandwidth
3690 * of the ring, we need to upclock the CPU (ia_freq).
3692 * For GPU frequencies less than 750MHz,
3693 * just use the lowest ring freq.
3695 if (gpu_freq
< min_freq
)
3698 ia_freq
= max_ia_freq
- ((diff
* scaling_factor
) / 2);
3699 ia_freq
= DIV_ROUND_CLOSEST(ia_freq
, 100);
3702 sandybridge_pcode_write(dev_priv
,
3703 GEN6_PCODE_WRITE_MIN_FREQ_TABLE
,
3704 ia_freq
<< GEN6_PCODE_FREQ_IA_RATIO_SHIFT
|
3705 ring_freq
<< GEN6_PCODE_FREQ_RING_RATIO_SHIFT
|
3710 void gen6_update_ring_freq(struct drm_device
*dev
)
3712 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3714 if (INTEL_INFO(dev
)->gen
< 6 || IS_VALLEYVIEW(dev
))
3717 mutex_lock(&dev_priv
->rps
.hw_lock
);
3718 __gen6_update_ring_freq(dev
);
3719 mutex_unlock(&dev_priv
->rps
.hw_lock
);
3722 int valleyview_rps_max_freq(struct drm_i915_private
*dev_priv
)
3726 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FREQ_FUSE
);
3728 rp0
= (val
& FB_GFX_MAX_FREQ_FUSE_MASK
) >> FB_GFX_MAX_FREQ_FUSE_SHIFT
;
3730 rp0
= min_t(u32
, rp0
, 0xea);
3735 static int valleyview_rps_rpe_freq(struct drm_i915_private
*dev_priv
)
3739 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FMAX_FUSE_LO
);
3740 rpe
= (val
& FB_FMAX_VMIN_FREQ_LO_MASK
) >> FB_FMAX_VMIN_FREQ_LO_SHIFT
;
3741 val
= vlv_nc_read(dev_priv
, IOSF_NC_FB_GFX_FMAX_FUSE_HI
);
3742 rpe
|= (val
& FB_FMAX_VMIN_FREQ_HI_MASK
) << 5;
3747 int valleyview_rps_min_freq(struct drm_i915_private
*dev_priv
)
3749 return vlv_punit_read(dev_priv
, PUNIT_REG_GPU_LFM
) & 0xff;
3752 /* Check that the pctx buffer wasn't move under us. */
3753 static void valleyview_check_pctx(struct drm_i915_private
*dev_priv
)
3755 unsigned long pctx_addr
= I915_READ(VLV_PCBR
) & ~4095;
3757 WARN_ON(pctx_addr
!= dev_priv
->mm
.stolen_base
+
3758 dev_priv
->vlv_pctx
->stolen
->start
);
3761 static void valleyview_setup_pctx(struct drm_device
*dev
)
3763 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3764 struct drm_i915_gem_object
*pctx
;
3765 unsigned long pctx_paddr
;
3767 int pctx_size
= 24*1024;
3769 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
3771 pcbr
= I915_READ(VLV_PCBR
);
3773 /* BIOS set it up already, grab the pre-alloc'd space */
3776 pcbr_offset
= (pcbr
& (~4095)) - dev_priv
->mm
.stolen_base
;
3777 pctx
= i915_gem_object_create_stolen_for_preallocated(dev_priv
->dev
,
3779 I915_GTT_OFFSET_NONE
,
3785 * From the Gunit register HAS:
3786 * The Gfx driver is expected to program this register and ensure
3787 * proper allocation within Gfx stolen memory. For example, this
3788 * register should be programmed such than the PCBR range does not
3789 * overlap with other ranges, such as the frame buffer, protected
3790 * memory, or any other relevant ranges.
3792 pctx
= i915_gem_object_create_stolen(dev
, pctx_size
);
3794 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
3798 pctx_paddr
= dev_priv
->mm
.stolen_base
+ pctx
->stolen
->start
;
3799 I915_WRITE(VLV_PCBR
, pctx_paddr
);
3802 dev_priv
->vlv_pctx
= pctx
;
3805 static void valleyview_cleanup_pctx(struct drm_device
*dev
)
3807 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3809 if (WARN_ON(!dev_priv
->vlv_pctx
))
3812 drm_gem_object_unreference(&dev_priv
->vlv_pctx
->base
);
3813 dev_priv
->vlv_pctx
= NULL
;
3816 static void valleyview_init_gt_powersave(struct drm_device
*dev
)
3818 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3820 valleyview_setup_pctx(dev
);
3822 mutex_lock(&dev_priv
->rps
.hw_lock
);
3824 dev_priv
->rps
.max_freq
= valleyview_rps_max_freq(dev_priv
);
3825 dev_priv
->rps
.rp0_freq
= dev_priv
->rps
.max_freq
;
3826 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
3827 vlv_gpu_freq(dev_priv
, dev_priv
->rps
.max_freq
),
3828 dev_priv
->rps
.max_freq
);
3830 dev_priv
->rps
.efficient_freq
= valleyview_rps_rpe_freq(dev_priv
);
3831 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
3832 vlv_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
3833 dev_priv
->rps
.efficient_freq
);
3835 dev_priv
->rps
.min_freq
= valleyview_rps_min_freq(dev_priv
);
3836 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
3837 vlv_gpu_freq(dev_priv
, dev_priv
->rps
.min_freq
),
3838 dev_priv
->rps
.min_freq
);
3840 /* Preserve min/max settings in case of re-init */
3841 if (dev_priv
->rps
.max_freq_softlimit
== 0)
3842 dev_priv
->rps
.max_freq_softlimit
= dev_priv
->rps
.max_freq
;
3844 if (dev_priv
->rps
.min_freq_softlimit
== 0)
3845 dev_priv
->rps
.min_freq_softlimit
= dev_priv
->rps
.min_freq
;
3847 mutex_unlock(&dev_priv
->rps
.hw_lock
);
3850 static void valleyview_cleanup_gt_powersave(struct drm_device
*dev
)
3852 valleyview_cleanup_pctx(dev
);
3855 static void valleyview_enable_rps(struct drm_device
*dev
)
3857 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3858 struct intel_engine_cs
*ring
;
3859 u32 gtfifodbg
, val
, rc6_mode
= 0;
3862 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
3864 valleyview_check_pctx(dev_priv
);
3866 if ((gtfifodbg
= I915_READ(GTFIFODBG
))) {
3867 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
3869 I915_WRITE(GTFIFODBG
, gtfifodbg
);
3872 /* If VLV, Forcewake all wells, else re-direct to regular path */
3873 gen6_gt_force_wake_get(dev_priv
, FORCEWAKE_ALL
);
3875 I915_WRITE(GEN6_RP_UP_THRESHOLD
, 59400);
3876 I915_WRITE(GEN6_RP_DOWN_THRESHOLD
, 245000);
3877 I915_WRITE(GEN6_RP_UP_EI
, 66000);
3878 I915_WRITE(GEN6_RP_DOWN_EI
, 350000);
3880 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS
, 10);
3882 I915_WRITE(GEN6_RP_CONTROL
,
3883 GEN6_RP_MEDIA_TURBO
|
3884 GEN6_RP_MEDIA_HW_NORMAL_MODE
|
3885 GEN6_RP_MEDIA_IS_GFX
|
3887 GEN6_RP_UP_BUSY_AVG
|
3888 GEN6_RP_DOWN_IDLE_CONT
);
3890 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT
, 0x00280000);
3891 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL
, 125000);
3892 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS
, 25);
3894 for_each_ring(ring
, dev_priv
, i
)
3895 I915_WRITE(RING_MAX_IDLE(ring
->mmio_base
), 10);
3897 I915_WRITE(GEN6_RC6_THRESHOLD
, 0x557);
3899 /* allows RC6 residency counter to work */
3900 I915_WRITE(VLV_COUNTER_CONTROL
,
3901 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH
|
3902 VLV_MEDIA_RC6_COUNT_EN
|
3903 VLV_RENDER_RC6_COUNT_EN
));
3904 if (intel_enable_rc6(dev
) & INTEL_RC6_ENABLE
)
3905 rc6_mode
= GEN7_RC_CTL_TO_MODE
| VLV_RC_CTL_CTX_RST_PARALLEL
;
3907 intel_print_rc6_info(dev
, rc6_mode
);
3909 I915_WRITE(GEN6_RC_CONTROL
, rc6_mode
);
3911 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
3913 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val
& 0x10 ? "yes" : "no");
3914 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val
);
3916 dev_priv
->rps
.cur_freq
= (val
>> 8) & 0xff;
3917 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
3918 vlv_gpu_freq(dev_priv
, dev_priv
->rps
.cur_freq
),
3919 dev_priv
->rps
.cur_freq
);
3921 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
3922 vlv_gpu_freq(dev_priv
, dev_priv
->rps
.efficient_freq
),
3923 dev_priv
->rps
.efficient_freq
);
3925 valleyview_set_rps(dev_priv
->dev
, dev_priv
->rps
.efficient_freq
);
3927 gen6_enable_rps_interrupts(dev
);
3929 gen6_gt_force_wake_put(dev_priv
, FORCEWAKE_ALL
);
3932 void ironlake_teardown_rc6(struct drm_device
*dev
)
3934 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3936 if (dev_priv
->ips
.renderctx
) {
3937 i915_gem_object_ggtt_unpin(dev_priv
->ips
.renderctx
);
3938 drm_gem_object_unreference(&dev_priv
->ips
.renderctx
->base
);
3939 dev_priv
->ips
.renderctx
= NULL
;
3942 if (dev_priv
->ips
.pwrctx
) {
3943 i915_gem_object_ggtt_unpin(dev_priv
->ips
.pwrctx
);
3944 drm_gem_object_unreference(&dev_priv
->ips
.pwrctx
->base
);
3945 dev_priv
->ips
.pwrctx
= NULL
;
3949 static void ironlake_disable_rc6(struct drm_device
*dev
)
3951 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3953 if (I915_READ(PWRCTXA
)) {
3954 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
3955 I915_WRITE(RSTDBYCTL
, I915_READ(RSTDBYCTL
) | RCX_SW_EXIT
);
3956 wait_for(((I915_READ(RSTDBYCTL
) & RSX_STATUS_MASK
) == RSX_STATUS_ON
),
3959 I915_WRITE(PWRCTXA
, 0);
3960 POSTING_READ(PWRCTXA
);
3962 I915_WRITE(RSTDBYCTL
, I915_READ(RSTDBYCTL
) & ~RCX_SW_EXIT
);
3963 POSTING_READ(RSTDBYCTL
);
3967 static int ironlake_setup_rc6(struct drm_device
*dev
)
3969 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3971 if (dev_priv
->ips
.renderctx
== NULL
)
3972 dev_priv
->ips
.renderctx
= intel_alloc_context_page(dev
);
3973 if (!dev_priv
->ips
.renderctx
)
3976 if (dev_priv
->ips
.pwrctx
== NULL
)
3977 dev_priv
->ips
.pwrctx
= intel_alloc_context_page(dev
);
3978 if (!dev_priv
->ips
.pwrctx
) {
3979 ironlake_teardown_rc6(dev
);
3986 static void ironlake_enable_rc6(struct drm_device
*dev
)
3988 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
3989 struct intel_engine_cs
*ring
= &dev_priv
->ring
[RCS
];
3990 bool was_interruptible
;
3993 /* rc6 disabled by default due to repeated reports of hanging during
3996 if (!intel_enable_rc6(dev
))
3999 WARN_ON(!mutex_is_locked(&dev
->struct_mutex
));
4001 ret
= ironlake_setup_rc6(dev
);
4005 was_interruptible
= dev_priv
->mm
.interruptible
;
4006 dev_priv
->mm
.interruptible
= false;
4009 * GPU can automatically power down the render unit if given a page
4012 ret
= intel_ring_begin(ring
, 6);
4014 ironlake_teardown_rc6(dev
);
4015 dev_priv
->mm
.interruptible
= was_interruptible
;
4019 intel_ring_emit(ring
, MI_SUSPEND_FLUSH
| MI_SUSPEND_FLUSH_EN
);
4020 intel_ring_emit(ring
, MI_SET_CONTEXT
);
4021 intel_ring_emit(ring
, i915_gem_obj_ggtt_offset(dev_priv
->ips
.renderctx
) |
4023 MI_SAVE_EXT_STATE_EN
|
4024 MI_RESTORE_EXT_STATE_EN
|
4025 MI_RESTORE_INHIBIT
);
4026 intel_ring_emit(ring
, MI_SUSPEND_FLUSH
);
4027 intel_ring_emit(ring
, MI_NOOP
);
4028 intel_ring_emit(ring
, MI_FLUSH
);
4029 intel_ring_advance(ring
);
4032 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
4033 * does an implicit flush, combined with MI_FLUSH above, it should be
4034 * safe to assume that renderctx is valid
4036 ret
= intel_ring_idle(ring
);
4037 dev_priv
->mm
.interruptible
= was_interruptible
;
4039 DRM_ERROR("failed to enable ironlake power savings\n");
4040 ironlake_teardown_rc6(dev
);
4044 I915_WRITE(PWRCTXA
, i915_gem_obj_ggtt_offset(dev_priv
->ips
.pwrctx
) | PWRCTX_EN
);
4045 I915_WRITE(RSTDBYCTL
, I915_READ(RSTDBYCTL
) & ~RCX_SW_EXIT
);
4047 intel_print_rc6_info(dev
, GEN6_RC_CTL_RC6_ENABLE
);
4050 static unsigned long intel_pxfreq(u32 vidfreq
)
4053 int div
= (vidfreq
& 0x3f0000) >> 16;
4054 int post
= (vidfreq
& 0x3000) >> 12;
4055 int pre
= (vidfreq
& 0x7);
4060 freq
= ((div
* 133333) / ((1<<post
) * pre
));
4065 static const struct cparams
{
4071 { 1, 1333, 301, 28664 },
4072 { 1, 1066, 294, 24460 },
4073 { 1, 800, 294, 25192 },
4074 { 0, 1333, 276, 27605 },
4075 { 0, 1066, 276, 27605 },
4076 { 0, 800, 231, 23784 },
4079 static unsigned long __i915_chipset_val(struct drm_i915_private
*dev_priv
)
4081 u64 total_count
, diff
, ret
;
4082 u32 count1
, count2
, count3
, m
= 0, c
= 0;
4083 unsigned long now
= jiffies_to_msecs(jiffies
), diff1
;
4086 assert_spin_locked(&mchdev_lock
);
4088 diff1
= now
- dev_priv
->ips
.last_time1
;
4090 /* Prevent division-by-zero if we are asking too fast.
4091 * Also, we don't get interesting results if we are polling
4092 * faster than once in 10ms, so just return the saved value
4096 return dev_priv
->ips
.chipset_power
;
4098 count1
= I915_READ(DMIEC
);
4099 count2
= I915_READ(DDREC
);
4100 count3
= I915_READ(CSIEC
);
4102 total_count
= count1
+ count2
+ count3
;
4104 /* FIXME: handle per-counter overflow */
4105 if (total_count
< dev_priv
->ips
.last_count1
) {
4106 diff
= ~0UL - dev_priv
->ips
.last_count1
;
4107 diff
+= total_count
;
4109 diff
= total_count
- dev_priv
->ips
.last_count1
;
4112 for (i
= 0; i
< ARRAY_SIZE(cparams
); i
++) {
4113 if (cparams
[i
].i
== dev_priv
->ips
.c_m
&&
4114 cparams
[i
].t
== dev_priv
->ips
.r_t
) {
4121 diff
= div_u64(diff
, diff1
);
4122 ret
= ((m
* diff
) + c
);
4123 ret
= div_u64(ret
, 10);
4125 dev_priv
->ips
.last_count1
= total_count
;
4126 dev_priv
->ips
.last_time1
= now
;
4128 dev_priv
->ips
.chipset_power
= ret
;
4133 unsigned long i915_chipset_val(struct drm_i915_private
*dev_priv
)
4135 struct drm_device
*dev
= dev_priv
->dev
;
4138 if (INTEL_INFO(dev
)->gen
!= 5)
4141 spin_lock_irq(&mchdev_lock
);
4143 val
= __i915_chipset_val(dev_priv
);
4145 spin_unlock_irq(&mchdev_lock
);
4150 unsigned long i915_mch_val(struct drm_i915_private
*dev_priv
)
4152 unsigned long m
, x
, b
;
4155 tsfs
= I915_READ(TSFS
);
4157 m
= ((tsfs
& TSFS_SLOPE_MASK
) >> TSFS_SLOPE_SHIFT
);
4158 x
= I915_READ8(TR1
);
4160 b
= tsfs
& TSFS_INTR_MASK
;
4162 return ((m
* x
) / 127) - b
;
4165 static u16
pvid_to_extvid(struct drm_i915_private
*dev_priv
, u8 pxvid
)
4167 struct drm_device
*dev
= dev_priv
->dev
;
4168 static const struct v_table
{
4169 u16 vd
; /* in .1 mil */
4170 u16 vm
; /* in .1 mil */
4301 if (INTEL_INFO(dev
)->is_mobile
)
4302 return v_table
[pxvid
].vm
;
4304 return v_table
[pxvid
].vd
;
4307 static void __i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
4309 struct timespec now
, diff1
;
4311 unsigned long diffms
;
4314 assert_spin_locked(&mchdev_lock
);
4316 getrawmonotonic(&now
);
4317 diff1
= timespec_sub(now
, dev_priv
->ips
.last_time2
);
4319 /* Don't divide by 0 */
4320 diffms
= diff1
.tv_sec
* 1000 + diff1
.tv_nsec
/ 1000000;
4324 count
= I915_READ(GFXEC
);
4326 if (count
< dev_priv
->ips
.last_count2
) {
4327 diff
= ~0UL - dev_priv
->ips
.last_count2
;
4330 diff
= count
- dev_priv
->ips
.last_count2
;
4333 dev_priv
->ips
.last_count2
= count
;
4334 dev_priv
->ips
.last_time2
= now
;
4336 /* More magic constants... */
4338 diff
= div_u64(diff
, diffms
* 10);
4339 dev_priv
->ips
.gfx_power
= diff
;
4342 void i915_update_gfx_val(struct drm_i915_private
*dev_priv
)
4344 struct drm_device
*dev
= dev_priv
->dev
;
4346 if (INTEL_INFO(dev
)->gen
!= 5)
4349 spin_lock_irq(&mchdev_lock
);
4351 __i915_update_gfx_val(dev_priv
);
4353 spin_unlock_irq(&mchdev_lock
);
4356 static unsigned long __i915_gfx_val(struct drm_i915_private
*dev_priv
)
4358 unsigned long t
, corr
, state1
, corr2
, state2
;
4361 assert_spin_locked(&mchdev_lock
);
4363 pxvid
= I915_READ(PXVFREQ_BASE
+ (dev_priv
->rps
.cur_freq
* 4));
4364 pxvid
= (pxvid
>> 24) & 0x7f;
4365 ext_v
= pvid_to_extvid(dev_priv
, pxvid
);
4369 t
= i915_mch_val(dev_priv
);
4371 /* Revel in the empirically derived constants */
4373 /* Correction factor in 1/100000 units */
4375 corr
= ((t
* 2349) + 135940);
4377 corr
= ((t
* 964) + 29317);
4379 corr
= ((t
* 301) + 1004);
4381 corr
= corr
* ((150142 * state1
) / 10000 - 78642);
4383 corr2
= (corr
* dev_priv
->ips
.corr
);
4385 state2
= (corr2
* state1
) / 10000;
4386 state2
/= 100; /* convert to mW */
4388 __i915_update_gfx_val(dev_priv
);
4390 return dev_priv
->ips
.gfx_power
+ state2
;
4393 unsigned long i915_gfx_val(struct drm_i915_private
*dev_priv
)
4395 struct drm_device
*dev
= dev_priv
->dev
;
4398 if (INTEL_INFO(dev
)->gen
!= 5)
4401 spin_lock_irq(&mchdev_lock
);
4403 val
= __i915_gfx_val(dev_priv
);
4405 spin_unlock_irq(&mchdev_lock
);
4411 * i915_read_mch_val - return value for IPS use
4413 * Calculate and return a value for the IPS driver to use when deciding whether
4414 * we have thermal and power headroom to increase CPU or GPU power budget.
4416 unsigned long i915_read_mch_val(void)
4418 struct drm_i915_private
*dev_priv
;
4419 unsigned long chipset_val
, graphics_val
, ret
= 0;
4421 spin_lock_irq(&mchdev_lock
);
4424 dev_priv
= i915_mch_dev
;
4426 chipset_val
= __i915_chipset_val(dev_priv
);
4427 graphics_val
= __i915_gfx_val(dev_priv
);
4429 ret
= chipset_val
+ graphics_val
;
4432 spin_unlock_irq(&mchdev_lock
);
4436 EXPORT_SYMBOL_GPL(i915_read_mch_val
);
4439 * i915_gpu_raise - raise GPU frequency limit
4441 * Raise the limit; IPS indicates we have thermal headroom.
4443 bool i915_gpu_raise(void)
4445 struct drm_i915_private
*dev_priv
;
4448 spin_lock_irq(&mchdev_lock
);
4449 if (!i915_mch_dev
) {
4453 dev_priv
= i915_mch_dev
;
4455 if (dev_priv
->ips
.max_delay
> dev_priv
->ips
.fmax
)
4456 dev_priv
->ips
.max_delay
--;
4459 spin_unlock_irq(&mchdev_lock
);
4463 EXPORT_SYMBOL_GPL(i915_gpu_raise
);
4466 * i915_gpu_lower - lower GPU frequency limit
4468 * IPS indicates we're close to a thermal limit, so throttle back the GPU
4469 * frequency maximum.
4471 bool i915_gpu_lower(void)
4473 struct drm_i915_private
*dev_priv
;
4476 spin_lock_irq(&mchdev_lock
);
4477 if (!i915_mch_dev
) {
4481 dev_priv
= i915_mch_dev
;
4483 if (dev_priv
->ips
.max_delay
< dev_priv
->ips
.min_delay
)
4484 dev_priv
->ips
.max_delay
++;
4487 spin_unlock_irq(&mchdev_lock
);
4491 EXPORT_SYMBOL_GPL(i915_gpu_lower
);
4494 * i915_gpu_busy - indicate GPU business to IPS
4496 * Tell the IPS driver whether or not the GPU is busy.
4498 bool i915_gpu_busy(void)
4500 struct drm_i915_private
*dev_priv
;
4501 struct intel_engine_cs
*ring
;
4505 spin_lock_irq(&mchdev_lock
);
4508 dev_priv
= i915_mch_dev
;
4510 for_each_ring(ring
, dev_priv
, i
)
4511 ret
|= !list_empty(&ring
->request_list
);
4514 spin_unlock_irq(&mchdev_lock
);
4518 EXPORT_SYMBOL_GPL(i915_gpu_busy
);
4521 * i915_gpu_turbo_disable - disable graphics turbo
4523 * Disable graphics turbo by resetting the max frequency and setting the
4524 * current frequency to the default.
4526 bool i915_gpu_turbo_disable(void)
4528 struct drm_i915_private
*dev_priv
;
4531 spin_lock_irq(&mchdev_lock
);
4532 if (!i915_mch_dev
) {
4536 dev_priv
= i915_mch_dev
;
4538 dev_priv
->ips
.max_delay
= dev_priv
->ips
.fstart
;
4540 if (!ironlake_set_drps(dev_priv
->dev
, dev_priv
->ips
.fstart
))
4544 spin_unlock_irq(&mchdev_lock
);
4548 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable
);
4551 * Tells the intel_ips driver that the i915 driver is now loaded, if
4552 * IPS got loaded first.
4554 * This awkward dance is so that neither module has to depend on the
4555 * other in order for IPS to do the appropriate communication of
4556 * GPU turbo limits to i915.
4559 ips_ping_for_i915_load(void)
4563 link
= symbol_get(ips_link_to_i915_driver
);
4566 symbol_put(ips_link_to_i915_driver
);
4570 void intel_gpu_ips_init(struct drm_i915_private
*dev_priv
)
4572 /* We only register the i915 ips part with intel-ips once everything is
4573 * set up, to avoid intel-ips sneaking in and reading bogus values. */
4574 spin_lock_irq(&mchdev_lock
);
4575 i915_mch_dev
= dev_priv
;
4576 spin_unlock_irq(&mchdev_lock
);
4578 ips_ping_for_i915_load();
4581 void intel_gpu_ips_teardown(void)
4583 spin_lock_irq(&mchdev_lock
);
4584 i915_mch_dev
= NULL
;
4585 spin_unlock_irq(&mchdev_lock
);
4588 static void intel_init_emon(struct drm_device
*dev
)
4590 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4595 /* Disable to program */
4599 /* Program energy weights for various events */
4600 I915_WRITE(SDEW
, 0x15040d00);
4601 I915_WRITE(CSIEW0
, 0x007f0000);
4602 I915_WRITE(CSIEW1
, 0x1e220004);
4603 I915_WRITE(CSIEW2
, 0x04000004);
4605 for (i
= 0; i
< 5; i
++)
4606 I915_WRITE(PEW
+ (i
* 4), 0);
4607 for (i
= 0; i
< 3; i
++)
4608 I915_WRITE(DEW
+ (i
* 4), 0);
4610 /* Program P-state weights to account for frequency power adjustment */
4611 for (i
= 0; i
< 16; i
++) {
4612 u32 pxvidfreq
= I915_READ(PXVFREQ_BASE
+ (i
* 4));
4613 unsigned long freq
= intel_pxfreq(pxvidfreq
);
4614 unsigned long vid
= (pxvidfreq
& PXVFREQ_PX_MASK
) >>
4619 val
*= (freq
/ 1000);
4621 val
/= (127*127*900);
4623 DRM_ERROR("bad pxval: %ld\n", val
);
4626 /* Render standby states get 0 weight */
4630 for (i
= 0; i
< 4; i
++) {
4631 u32 val
= (pxw
[i
*4] << 24) | (pxw
[(i
*4)+1] << 16) |
4632 (pxw
[(i
*4)+2] << 8) | (pxw
[(i
*4)+3]);
4633 I915_WRITE(PXW
+ (i
* 4), val
);
4636 /* Adjust magic regs to magic values (more experimental results) */
4637 I915_WRITE(OGW0
, 0);
4638 I915_WRITE(OGW1
, 0);
4639 I915_WRITE(EG0
, 0x00007f00);
4640 I915_WRITE(EG1
, 0x0000000e);
4641 I915_WRITE(EG2
, 0x000e0000);
4642 I915_WRITE(EG3
, 0x68000300);
4643 I915_WRITE(EG4
, 0x42000000);
4644 I915_WRITE(EG5
, 0x00140031);
4648 for (i
= 0; i
< 8; i
++)
4649 I915_WRITE(PXWL
+ (i
* 4), 0);
4651 /* Enable PMON + select events */
4652 I915_WRITE(ECR
, 0x80000019);
4654 lcfuse
= I915_READ(LCFUSE02
);
4656 dev_priv
->ips
.corr
= (lcfuse
& LCFUSE_HIV_MASK
);
4659 void intel_init_gt_powersave(struct drm_device
*dev
)
4661 i915
.enable_rc6
= sanitize_rc6_option(dev
, i915
.enable_rc6
);
4663 if (IS_VALLEYVIEW(dev
))
4664 valleyview_init_gt_powersave(dev
);
4667 void intel_cleanup_gt_powersave(struct drm_device
*dev
)
4669 if (IS_VALLEYVIEW(dev
))
4670 valleyview_cleanup_gt_powersave(dev
);
4673 void intel_disable_gt_powersave(struct drm_device
*dev
)
4675 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4677 /* Interrupts should be disabled already to avoid re-arming. */
4678 WARN_ON(dev
->irq_enabled
);
4680 if (IS_IRONLAKE_M(dev
)) {
4681 ironlake_disable_drps(dev
);
4682 ironlake_disable_rc6(dev
);
4683 } else if (IS_GEN6(dev
) || IS_GEN7(dev
) || IS_BROADWELL(dev
)) {
4684 if (cancel_delayed_work_sync(&dev_priv
->rps
.delayed_resume_work
))
4685 intel_runtime_pm_put(dev_priv
);
4687 cancel_work_sync(&dev_priv
->rps
.work
);
4688 mutex_lock(&dev_priv
->rps
.hw_lock
);
4689 if (IS_VALLEYVIEW(dev
))
4690 valleyview_disable_rps(dev
);
4692 gen6_disable_rps(dev
);
4693 dev_priv
->rps
.enabled
= false;
4694 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4698 static void intel_gen6_powersave_work(struct work_struct
*work
)
4700 struct drm_i915_private
*dev_priv
=
4701 container_of(work
, struct drm_i915_private
,
4702 rps
.delayed_resume_work
.work
);
4703 struct drm_device
*dev
= dev_priv
->dev
;
4705 mutex_lock(&dev_priv
->rps
.hw_lock
);
4707 if (IS_VALLEYVIEW(dev
)) {
4708 valleyview_enable_rps(dev
);
4709 } else if (IS_BROADWELL(dev
)) {
4710 gen8_enable_rps(dev
);
4711 __gen6_update_ring_freq(dev
);
4713 gen6_enable_rps(dev
);
4714 __gen6_update_ring_freq(dev
);
4716 dev_priv
->rps
.enabled
= true;
4717 mutex_unlock(&dev_priv
->rps
.hw_lock
);
4719 intel_runtime_pm_put(dev_priv
);
4722 void intel_enable_gt_powersave(struct drm_device
*dev
)
4724 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4726 if (IS_IRONLAKE_M(dev
)) {
4727 mutex_lock(&dev
->struct_mutex
);
4728 ironlake_enable_drps(dev
);
4729 ironlake_enable_rc6(dev
);
4730 intel_init_emon(dev
);
4731 mutex_unlock(&dev
->struct_mutex
);
4732 } else if (IS_GEN6(dev
) || IS_GEN7(dev
) || IS_BROADWELL(dev
)) {
4734 * PCU communication is slow and this doesn't need to be
4735 * done at any specific time, so do this out of our fast path
4736 * to make resume and init faster.
4738 * We depend on the HW RC6 power context save/restore
4739 * mechanism when entering D3 through runtime PM suspend. So
4740 * disable RPM until RPS/RC6 is properly setup. We can only
4741 * get here via the driver load/system resume/runtime resume
4742 * paths, so the _noresume version is enough (and in case of
4743 * runtime resume it's necessary).
4745 if (schedule_delayed_work(&dev_priv
->rps
.delayed_resume_work
,
4746 round_jiffies_up_relative(HZ
)))
4747 intel_runtime_pm_get_noresume(dev_priv
);
4751 void intel_reset_gt_powersave(struct drm_device
*dev
)
4753 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4755 dev_priv
->rps
.enabled
= false;
4756 intel_enable_gt_powersave(dev
);
4759 static void ibx_init_clock_gating(struct drm_device
*dev
)
4761 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4764 * On Ibex Peak and Cougar Point, we need to disable clock
4765 * gating for the panel power sequencer or it will fail to
4766 * start up when no ports are active.
4768 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
);
4771 static void g4x_disable_trickle_feed(struct drm_device
*dev
)
4773 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4776 for_each_pipe(pipe
) {
4777 I915_WRITE(DSPCNTR(pipe
),
4778 I915_READ(DSPCNTR(pipe
)) |
4779 DISPPLANE_TRICKLE_FEED_DISABLE
);
4780 intel_flush_primary_plane(dev_priv
, pipe
);
4784 static void ilk_init_lp_watermarks(struct drm_device
*dev
)
4786 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4788 I915_WRITE(WM3_LP_ILK
, I915_READ(WM3_LP_ILK
) & ~WM1_LP_SR_EN
);
4789 I915_WRITE(WM2_LP_ILK
, I915_READ(WM2_LP_ILK
) & ~WM1_LP_SR_EN
);
4790 I915_WRITE(WM1_LP_ILK
, I915_READ(WM1_LP_ILK
) & ~WM1_LP_SR_EN
);
4793 * Don't touch WM1S_LP_EN here.
4794 * Doing so could cause underruns.
4798 static void ironlake_init_clock_gating(struct drm_device
*dev
)
4800 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4801 uint32_t dspclk_gate
= ILK_VRHUNIT_CLOCK_GATE_DISABLE
;
4805 * WaFbcDisableDpfcClockGating:ilk
4807 dspclk_gate
|= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE
|
4808 ILK_DPFCUNIT_CLOCK_GATE_DISABLE
|
4809 ILK_DPFDUNIT_CLOCK_GATE_ENABLE
;
4811 I915_WRITE(PCH_3DCGDIS0
,
4812 MARIUNIT_CLOCK_GATE_DISABLE
|
4813 SVSMUNIT_CLOCK_GATE_DISABLE
);
4814 I915_WRITE(PCH_3DCGDIS1
,
4815 VFMUNIT_CLOCK_GATE_DISABLE
);
4818 * According to the spec the following bits should be set in
4819 * order to enable memory self-refresh
4820 * The bit 22/21 of 0x42004
4821 * The bit 5 of 0x42020
4822 * The bit 15 of 0x45000
4824 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
4825 (I915_READ(ILK_DISPLAY_CHICKEN2
) |
4826 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
));
4827 dspclk_gate
|= ILK_DPARBUNIT_CLOCK_GATE_ENABLE
;
4828 I915_WRITE(DISP_ARB_CTL
,
4829 (I915_READ(DISP_ARB_CTL
) |
4832 ilk_init_lp_watermarks(dev
);
4835 * Based on the document from hardware guys the following bits
4836 * should be set unconditionally in order to enable FBC.
4837 * The bit 22 of 0x42000
4838 * The bit 22 of 0x42004
4839 * The bit 7,8,9 of 0x42020.
4841 if (IS_IRONLAKE_M(dev
)) {
4842 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
4843 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
4844 I915_READ(ILK_DISPLAY_CHICKEN1
) |
4846 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
4847 I915_READ(ILK_DISPLAY_CHICKEN2
) |
4851 I915_WRITE(ILK_DSPCLK_GATE_D
, dspclk_gate
);
4853 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
4854 I915_READ(ILK_DISPLAY_CHICKEN2
) |
4855 ILK_ELPIN_409_SELECT
);
4856 I915_WRITE(_3D_CHICKEN2
,
4857 _3D_CHICKEN2_WM_READ_PIPELINED
<< 16 |
4858 _3D_CHICKEN2_WM_READ_PIPELINED
);
4860 /* WaDisableRenderCachePipelinedFlush:ilk */
4861 I915_WRITE(CACHE_MODE_0
,
4862 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE
));
4864 /* WaDisable_RenderCache_OperationalFlush:ilk */
4865 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
4867 g4x_disable_trickle_feed(dev
);
4869 ibx_init_clock_gating(dev
);
4872 static void cpt_init_clock_gating(struct drm_device
*dev
)
4874 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4879 * On Ibex Peak and Cougar Point, we need to disable clock
4880 * gating for the panel power sequencer or it will fail to
4881 * start up when no ports are active.
4883 I915_WRITE(SOUTH_DSPCLK_GATE_D
, PCH_DPLSUNIT_CLOCK_GATE_DISABLE
|
4884 PCH_DPLUNIT_CLOCK_GATE_DISABLE
|
4885 PCH_CPUNIT_CLOCK_GATE_DISABLE
);
4886 I915_WRITE(SOUTH_CHICKEN2
, I915_READ(SOUTH_CHICKEN2
) |
4887 DPLS_EDP_PPS_FIX_DIS
);
4888 /* The below fixes the weird display corruption, a few pixels shifted
4889 * downward, on (only) LVDS of some HP laptops with IVY.
4891 for_each_pipe(pipe
) {
4892 val
= I915_READ(TRANS_CHICKEN2(pipe
));
4893 val
|= TRANS_CHICKEN2_TIMING_OVERRIDE
;
4894 val
&= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED
;
4895 if (dev_priv
->vbt
.fdi_rx_polarity_inverted
)
4896 val
|= TRANS_CHICKEN2_FDI_POLARITY_REVERSED
;
4897 val
&= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK
;
4898 val
&= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER
;
4899 val
&= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH
;
4900 I915_WRITE(TRANS_CHICKEN2(pipe
), val
);
4902 /* WADP0ClockGatingDisable */
4903 for_each_pipe(pipe
) {
4904 I915_WRITE(TRANS_CHICKEN1(pipe
),
4905 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE
);
4909 static void gen6_check_mch_setup(struct drm_device
*dev
)
4911 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4914 tmp
= I915_READ(MCH_SSKPD
);
4915 if ((tmp
& MCH_SSKPD_WM0_MASK
) != MCH_SSKPD_WM0_VAL
) {
4916 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp
);
4917 DRM_INFO("This can cause pipe underruns and display issues.\n");
4918 DRM_INFO("Please upgrade your BIOS to fix this.\n");
4922 static void gen6_init_clock_gating(struct drm_device
*dev
)
4924 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
4925 uint32_t dspclk_gate
= ILK_VRHUNIT_CLOCK_GATE_DISABLE
;
4927 I915_WRITE(ILK_DSPCLK_GATE_D
, dspclk_gate
);
4929 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
4930 I915_READ(ILK_DISPLAY_CHICKEN2
) |
4931 ILK_ELPIN_409_SELECT
);
4933 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4934 I915_WRITE(_3D_CHICKEN
,
4935 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB
));
4937 /* WaSetupGtModeTdRowDispatch:snb */
4938 if (IS_SNB_GT1(dev
))
4939 I915_WRITE(GEN6_GT_MODE
,
4940 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE
));
4942 /* WaDisable_RenderCache_OperationalFlush:snb */
4943 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
4946 * BSpec recoomends 8x4 when MSAA is used,
4947 * however in practice 16x4 seems fastest.
4949 * Note that PS/WM thread counts depend on the WIZ hashing
4950 * disable bit, which we don't touch here, but it's good
4951 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
4953 I915_WRITE(GEN6_GT_MODE
,
4954 GEN6_WIZ_HASHING_MASK
| GEN6_WIZ_HASHING_16x4
);
4956 ilk_init_lp_watermarks(dev
);
4958 I915_WRITE(CACHE_MODE_0
,
4959 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB
));
4961 I915_WRITE(GEN6_UCGCTL1
,
4962 I915_READ(GEN6_UCGCTL1
) |
4963 GEN6_BLBUNIT_CLOCK_GATE_DISABLE
|
4964 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
4966 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4967 * gating disable must be set. Failure to set it results in
4968 * flickering pixels due to Z write ordering failures after
4969 * some amount of runtime in the Mesa "fire" demo, and Unigine
4970 * Sanctuary and Tropics, and apparently anything else with
4971 * alpha test or pixel discard.
4973 * According to the spec, bit 11 (RCCUNIT) must also be set,
4974 * but we didn't debug actual testcases to find it out.
4976 * WaDisableRCCUnitClockGating:snb
4977 * WaDisableRCPBUnitClockGating:snb
4979 I915_WRITE(GEN6_UCGCTL2
,
4980 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE
|
4981 GEN6_RCCUNIT_CLOCK_GATE_DISABLE
);
4983 /* WaStripsFansDisableFastClipPerformanceFix:snb */
4984 I915_WRITE(_3D_CHICKEN3
,
4985 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL
));
4989 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
4990 * 3DSTATE_SF number of SF output attributes is more than 16."
4992 I915_WRITE(_3D_CHICKEN3
,
4993 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH
));
4996 * According to the spec the following bits should be
4997 * set in order to enable memory self-refresh and fbc:
4998 * The bit21 and bit22 of 0x42000
4999 * The bit21 and bit22 of 0x42004
5000 * The bit5 and bit7 of 0x42020
5001 * The bit14 of 0x70180
5002 * The bit14 of 0x71180
5004 * WaFbcAsynchFlipDisableFbcQueue:snb
5006 I915_WRITE(ILK_DISPLAY_CHICKEN1
,
5007 I915_READ(ILK_DISPLAY_CHICKEN1
) |
5008 ILK_FBCQ_DIS
| ILK_PABSTRETCH_DIS
);
5009 I915_WRITE(ILK_DISPLAY_CHICKEN2
,
5010 I915_READ(ILK_DISPLAY_CHICKEN2
) |
5011 ILK_DPARB_GATE
| ILK_VSDPFD_FULL
);
5012 I915_WRITE(ILK_DSPCLK_GATE_D
,
5013 I915_READ(ILK_DSPCLK_GATE_D
) |
5014 ILK_DPARBUNIT_CLOCK_GATE_ENABLE
|
5015 ILK_DPFDUNIT_CLOCK_GATE_ENABLE
);
5017 g4x_disable_trickle_feed(dev
);
5019 cpt_init_clock_gating(dev
);
5021 gen6_check_mch_setup(dev
);
5024 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private
*dev_priv
)
5026 uint32_t reg
= I915_READ(GEN7_FF_THREAD_MODE
);
5029 * WaVSThreadDispatchOverride:ivb,vlv
5031 * This actually overrides the dispatch
5032 * mode for all thread types.
5034 reg
&= ~GEN7_FF_SCHED_MASK
;
5035 reg
|= GEN7_FF_TS_SCHED_HW
;
5036 reg
|= GEN7_FF_VS_SCHED_HW
;
5037 reg
|= GEN7_FF_DS_SCHED_HW
;
5039 I915_WRITE(GEN7_FF_THREAD_MODE
, reg
);
5042 static void lpt_init_clock_gating(struct drm_device
*dev
)
5044 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5047 * TODO: this bit should only be enabled when really needed, then
5048 * disabled when not needed anymore in order to save power.
5050 if (dev_priv
->pch_id
== INTEL_PCH_LPT_LP_DEVICE_ID_TYPE
)
5051 I915_WRITE(SOUTH_DSPCLK_GATE_D
,
5052 I915_READ(SOUTH_DSPCLK_GATE_D
) |
5053 PCH_LP_PARTITION_LEVEL_DISABLE
);
5055 /* WADPOClockGatingDisable:hsw */
5056 I915_WRITE(_TRANSA_CHICKEN1
,
5057 I915_READ(_TRANSA_CHICKEN1
) |
5058 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE
);
5061 static void lpt_suspend_hw(struct drm_device
*dev
)
5063 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5065 if (dev_priv
->pch_id
== INTEL_PCH_LPT_LP_DEVICE_ID_TYPE
) {
5066 uint32_t val
= I915_READ(SOUTH_DSPCLK_GATE_D
);
5068 val
&= ~PCH_LP_PARTITION_LEVEL_DISABLE
;
5069 I915_WRITE(SOUTH_DSPCLK_GATE_D
, val
);
5073 static void gen8_init_clock_gating(struct drm_device
*dev
)
5075 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5078 I915_WRITE(WM3_LP_ILK
, 0);
5079 I915_WRITE(WM2_LP_ILK
, 0);
5080 I915_WRITE(WM1_LP_ILK
, 0);
5082 /* FIXME(BDW): Check all the w/a, some might only apply to
5083 * pre-production hw. */
5085 /* WaDisablePartialInstShootdown:bdw */
5086 I915_WRITE(GEN8_ROW_CHICKEN
,
5087 _MASKED_BIT_ENABLE(PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE
));
5089 /* WaDisableThreadStallDopClockGating:bdw */
5090 /* FIXME: Unclear whether we really need this on production bdw. */
5091 I915_WRITE(GEN8_ROW_CHICKEN
,
5092 _MASKED_BIT_ENABLE(STALL_DOP_GATING_DISABLE
));
5095 * This GEN8_CENTROID_PIXEL_OPT_DIS W/A is only needed for
5096 * pre-production hardware
5098 I915_WRITE(HALF_SLICE_CHICKEN3
,
5099 _MASKED_BIT_ENABLE(GEN8_CENTROID_PIXEL_OPT_DIS
));
5100 I915_WRITE(HALF_SLICE_CHICKEN3
,
5101 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS
));
5102 I915_WRITE(GAMTARBMODE
, _MASKED_BIT_ENABLE(ARB_MODE_BWGTLB_DISABLE
));
5104 I915_WRITE(_3D_CHICKEN3
,
5105 _3D_CHICKEN_SDE_LIMIT_FIFO_POLY_DEPTH(2));
5107 I915_WRITE(COMMON_SLICE_CHICKEN2
,
5108 _MASKED_BIT_ENABLE(GEN8_CSC2_SBE_VUE_CACHE_CONSERVATIVE
));
5110 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
5111 _MASKED_BIT_ENABLE(GEN7_SINGLE_SUBSCAN_DISPATCH_ENABLE
));
5113 /* WaDisableDopClockGating:bdw May not be needed for production */
5114 I915_WRITE(GEN7_ROW_CHICKEN2
,
5115 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
5117 /* WaSwitchSolVfFArbitrationPriority:bdw */
5118 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) | HSW_ECOCHK_ARB_PRIO_SOL
);
5120 /* WaPsrDPAMaskVBlankInSRD:bdw */
5121 I915_WRITE(CHICKEN_PAR1_1
,
5122 I915_READ(CHICKEN_PAR1_1
) | DPA_MASK_VBLANK_SRD
);
5124 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
5125 for_each_pipe(pipe
) {
5126 I915_WRITE(CHICKEN_PIPESL_1(pipe
),
5127 I915_READ(CHICKEN_PIPESL_1(pipe
)) |
5128 BDW_DPRS_MASK_VBLANK_SRD
);
5131 /* Use Force Non-Coherent whenever executing a 3D context. This is a
5132 * workaround for for a possible hang in the unlikely event a TLB
5133 * invalidation occurs during a PSD flush.
5135 I915_WRITE(HDC_CHICKEN0
,
5136 I915_READ(HDC_CHICKEN0
) |
5137 _MASKED_BIT_ENABLE(HDC_FORCE_NON_COHERENT
));
5139 /* WaVSRefCountFullforceMissDisable:bdw */
5140 /* WaDSRefCountFullforceMissDisable:bdw */
5141 I915_WRITE(GEN7_FF_THREAD_MODE
,
5142 I915_READ(GEN7_FF_THREAD_MODE
) &
5143 ~(GEN8_FF_DS_REF_CNT_FFME
| GEN7_FF_VS_REF_CNT_FFME
));
5146 * BSpec recommends 8x4 when MSAA is used,
5147 * however in practice 16x4 seems fastest.
5149 * Note that PS/WM thread counts depend on the WIZ hashing
5150 * disable bit, which we don't touch here, but it's good
5151 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
5153 I915_WRITE(GEN7_GT_MODE
,
5154 GEN6_WIZ_HASHING_MASK
| GEN6_WIZ_HASHING_16x4
);
5156 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
5157 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE
));
5159 /* WaDisableSDEUnitClockGating:bdw */
5160 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
5161 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
5163 /* Wa4x4STCOptimizationDisable:bdw */
5164 I915_WRITE(CACHE_MODE_1
,
5165 _MASKED_BIT_ENABLE(GEN8_4x4_STC_OPTIMIZATION_DISABLE
));
5168 static void haswell_init_clock_gating(struct drm_device
*dev
)
5170 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5172 ilk_init_lp_watermarks(dev
);
5174 /* L3 caching of data atomics doesn't work -- disable it. */
5175 I915_WRITE(HSW_SCRATCH1
, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE
);
5176 I915_WRITE(HSW_ROW_CHICKEN3
,
5177 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE
));
5179 /* This is required by WaCatErrorRejectionIssue:hsw */
5180 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
5181 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
5182 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
5184 /* WaVSRefCountFullforceMissDisable:hsw */
5185 I915_WRITE(GEN7_FF_THREAD_MODE
,
5186 I915_READ(GEN7_FF_THREAD_MODE
) & ~GEN7_FF_VS_REF_CNT_FFME
);
5188 /* WaDisable_RenderCache_OperationalFlush:hsw */
5189 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
5191 /* enable HiZ Raw Stall Optimization */
5192 I915_WRITE(CACHE_MODE_0_GEN7
,
5193 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE
));
5195 /* WaDisable4x2SubspanOptimization:hsw */
5196 I915_WRITE(CACHE_MODE_1
,
5197 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
5200 * BSpec recommends 8x4 when MSAA is used,
5201 * however in practice 16x4 seems fastest.
5203 * Note that PS/WM thread counts depend on the WIZ hashing
5204 * disable bit, which we don't touch here, but it's good
5205 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
5207 I915_WRITE(GEN7_GT_MODE
,
5208 GEN6_WIZ_HASHING_MASK
| GEN6_WIZ_HASHING_16x4
);
5210 /* WaSwitchSolVfFArbitrationPriority:hsw */
5211 I915_WRITE(GAM_ECOCHK
, I915_READ(GAM_ECOCHK
) | HSW_ECOCHK_ARB_PRIO_SOL
);
5213 /* WaRsPkgCStateDisplayPMReq:hsw */
5214 I915_WRITE(CHICKEN_PAR1_1
,
5215 I915_READ(CHICKEN_PAR1_1
) | FORCE_ARB_IDLE_PLANES
);
5217 lpt_init_clock_gating(dev
);
5220 static void ivybridge_init_clock_gating(struct drm_device
*dev
)
5222 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5225 ilk_init_lp_watermarks(dev
);
5227 I915_WRITE(ILK_DSPCLK_GATE_D
, ILK_VRHUNIT_CLOCK_GATE_DISABLE
);
5229 /* WaDisableEarlyCull:ivb */
5230 I915_WRITE(_3D_CHICKEN3
,
5231 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL
));
5233 /* WaDisableBackToBackFlipFix:ivb */
5234 I915_WRITE(IVB_CHICKEN3
,
5235 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
5236 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
5238 /* WaDisablePSDDualDispatchEnable:ivb */
5239 if (IS_IVB_GT1(dev
))
5240 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
5241 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
));
5243 /* WaDisable_RenderCache_OperationalFlush:ivb */
5244 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
5246 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
5247 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1
,
5248 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
5250 /* WaApplyL3ControlAndL3ChickenMode:ivb */
5251 I915_WRITE(GEN7_L3CNTLREG1
,
5252 GEN7_WA_FOR_GEN7_L3_CONTROL
);
5253 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER
,
5254 GEN7_WA_L3_CHICKEN_MODE
);
5255 if (IS_IVB_GT1(dev
))
5256 I915_WRITE(GEN7_ROW_CHICKEN2
,
5257 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
5259 /* must write both registers */
5260 I915_WRITE(GEN7_ROW_CHICKEN2
,
5261 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
5262 I915_WRITE(GEN7_ROW_CHICKEN2_GT2
,
5263 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
5266 /* WaForceL3Serialization:ivb */
5267 I915_WRITE(GEN7_L3SQCREG4
, I915_READ(GEN7_L3SQCREG4
) &
5268 ~L3SQ_URB_READ_CAM_MATCH_DISABLE
);
5271 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5272 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
5274 I915_WRITE(GEN6_UCGCTL2
,
5275 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
5277 /* This is required by WaCatErrorRejectionIssue:ivb */
5278 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
5279 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
5280 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
5282 g4x_disable_trickle_feed(dev
);
5284 gen7_setup_fixed_func_scheduler(dev_priv
);
5286 if (0) { /* causes HiZ corruption on ivb:gt1 */
5287 /* enable HiZ Raw Stall Optimization */
5288 I915_WRITE(CACHE_MODE_0_GEN7
,
5289 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE
));
5292 /* WaDisable4x2SubspanOptimization:ivb */
5293 I915_WRITE(CACHE_MODE_1
,
5294 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
5297 * BSpec recommends 8x4 when MSAA is used,
5298 * however in practice 16x4 seems fastest.
5300 * Note that PS/WM thread counts depend on the WIZ hashing
5301 * disable bit, which we don't touch here, but it's good
5302 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
5304 I915_WRITE(GEN7_GT_MODE
,
5305 GEN6_WIZ_HASHING_MASK
| GEN6_WIZ_HASHING_16x4
);
5307 snpcr
= I915_READ(GEN6_MBCUNIT_SNPCR
);
5308 snpcr
&= ~GEN6_MBC_SNPCR_MASK
;
5309 snpcr
|= GEN6_MBC_SNPCR_MED
;
5310 I915_WRITE(GEN6_MBCUNIT_SNPCR
, snpcr
);
5312 if (!HAS_PCH_NOP(dev
))
5313 cpt_init_clock_gating(dev
);
5315 gen6_check_mch_setup(dev
);
5318 static void valleyview_init_clock_gating(struct drm_device
*dev
)
5320 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5323 mutex_lock(&dev_priv
->rps
.hw_lock
);
5324 val
= vlv_punit_read(dev_priv
, PUNIT_REG_GPU_FREQ_STS
);
5325 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5326 switch ((val
>> 6) & 3) {
5329 dev_priv
->mem_freq
= 800;
5332 dev_priv
->mem_freq
= 1066;
5335 dev_priv
->mem_freq
= 1333;
5338 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv
->mem_freq
);
5340 dev_priv
->vlv_cdclk_freq
= valleyview_cur_cdclk(dev_priv
);
5341 DRM_DEBUG_DRIVER("Current CD clock rate: %d MHz",
5342 dev_priv
->vlv_cdclk_freq
);
5344 I915_WRITE(DSPCLK_GATE_D
, VRHUNIT_CLOCK_GATE_DISABLE
);
5346 /* WaDisableEarlyCull:vlv */
5347 I915_WRITE(_3D_CHICKEN3
,
5348 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL
));
5350 /* WaDisableBackToBackFlipFix:vlv */
5351 I915_WRITE(IVB_CHICKEN3
,
5352 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE
|
5353 CHICKEN3_DGMG_DONE_FIX_DISABLE
);
5355 /* WaPsdDispatchEnable:vlv */
5356 /* WaDisablePSDDualDispatchEnable:vlv */
5357 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1
,
5358 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP
|
5359 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
));
5361 /* WaDisable_RenderCache_OperationalFlush:vlv */
5362 I915_WRITE(CACHE_MODE_0_GEN7
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
5364 /* WaForceL3Serialization:vlv */
5365 I915_WRITE(GEN7_L3SQCREG4
, I915_READ(GEN7_L3SQCREG4
) &
5366 ~L3SQ_URB_READ_CAM_MATCH_DISABLE
);
5368 /* WaDisableDopClockGating:vlv */
5369 I915_WRITE(GEN7_ROW_CHICKEN2
,
5370 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
5372 /* This is required by WaCatErrorRejectionIssue:vlv */
5373 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
,
5374 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG
) |
5375 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB
);
5377 gen7_setup_fixed_func_scheduler(dev_priv
);
5380 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5381 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
5383 I915_WRITE(GEN6_UCGCTL2
,
5384 GEN6_RCZUNIT_CLOCK_GATE_DISABLE
);
5386 /* WaDisableL3Bank2xClockGate:vlv
5387 * Disabling L3 clock gating- MMIO 940c[25] = 1
5388 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
5389 I915_WRITE(GEN7_UCGCTL4
,
5390 I915_READ(GEN7_UCGCTL4
) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE
);
5392 I915_WRITE(MI_ARB_VLV
, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
);
5395 * BSpec says this must be set, even though
5396 * WaDisable4x2SubspanOptimization isn't listed for VLV.
5398 I915_WRITE(CACHE_MODE_1
,
5399 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
));
5402 * WaIncreaseL3CreditsForVLVB0:vlv
5403 * This is the hardware default actually.
5405 I915_WRITE(GEN7_L3SQCREG1
, VLV_B0_WA_L3SQCREG1_VALUE
);
5408 * WaDisableVLVClockGating_VBIIssue:vlv
5409 * Disable clock gating on th GCFG unit to prevent a delay
5410 * in the reporting of vblank events.
5412 I915_WRITE(VLV_GUNIT_CLOCK_GATE
, GCFG_DIS
);
5415 static void cherryview_init_clock_gating(struct drm_device
*dev
)
5417 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5419 I915_WRITE(DSPCLK_GATE_D
, VRHUNIT_CLOCK_GATE_DISABLE
);
5421 I915_WRITE(MI_ARB_VLV
, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
);
5423 /* WaDisablePartialInstShootdown:chv */
5424 I915_WRITE(GEN8_ROW_CHICKEN
,
5425 _MASKED_BIT_ENABLE(PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE
));
5427 /* WaDisableThreadStallDopClockGating:chv */
5428 I915_WRITE(GEN8_ROW_CHICKEN
,
5429 _MASKED_BIT_ENABLE(STALL_DOP_GATING_DISABLE
));
5431 /* WaVSRefCountFullforceMissDisable:chv */
5432 /* WaDSRefCountFullforceMissDisable:chv */
5433 I915_WRITE(GEN7_FF_THREAD_MODE
,
5434 I915_READ(GEN7_FF_THREAD_MODE
) &
5435 ~(GEN8_FF_DS_REF_CNT_FFME
| GEN7_FF_VS_REF_CNT_FFME
));
5437 /* WaDisableSemaphoreAndSyncFlipWait:chv */
5438 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
5439 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE
));
5441 /* WaDisableCSUnitClockGating:chv */
5442 I915_WRITE(GEN6_UCGCTL1
, I915_READ(GEN6_UCGCTL1
) |
5443 GEN6_CSUNIT_CLOCK_GATE_DISABLE
);
5445 /* WaDisableSDEUnitClockGating:chv */
5446 I915_WRITE(GEN8_UCGCTL6
, I915_READ(GEN8_UCGCTL6
) |
5447 GEN8_SDEUNIT_CLOCK_GATE_DISABLE
);
5449 /* WaDisableSamplerPowerBypass:chv (pre-production hw) */
5450 I915_WRITE(HALF_SLICE_CHICKEN3
,
5451 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS
));
5453 /* WaDisableGunitClockGating:chv (pre-production hw) */
5454 I915_WRITE(VLV_GUNIT_CLOCK_GATE
, I915_READ(VLV_GUNIT_CLOCK_GATE
) |
5457 /* WaDisableFfDopClockGating:chv (pre-production hw) */
5458 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL
,
5459 _MASKED_BIT_ENABLE(GEN8_FF_DOP_CLOCK_GATE_DISABLE
));
5461 /* WaDisableDopClockGating:chv (pre-production hw) */
5462 I915_WRITE(GEN7_ROW_CHICKEN2
,
5463 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE
));
5464 I915_WRITE(GEN6_UCGCTL1
, I915_READ(GEN6_UCGCTL1
) |
5465 GEN6_EU_TCUNIT_CLOCK_GATE_DISABLE
);
5468 static void g4x_init_clock_gating(struct drm_device
*dev
)
5470 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5471 uint32_t dspclk_gate
;
5473 I915_WRITE(RENCLK_GATE_D1
, 0);
5474 I915_WRITE(RENCLK_GATE_D2
, VF_UNIT_CLOCK_GATE_DISABLE
|
5475 GS_UNIT_CLOCK_GATE_DISABLE
|
5476 CL_UNIT_CLOCK_GATE_DISABLE
);
5477 I915_WRITE(RAMCLK_GATE_D
, 0);
5478 dspclk_gate
= VRHUNIT_CLOCK_GATE_DISABLE
|
5479 OVRUNIT_CLOCK_GATE_DISABLE
|
5480 OVCUNIT_CLOCK_GATE_DISABLE
;
5482 dspclk_gate
|= DSSUNIT_CLOCK_GATE_DISABLE
;
5483 I915_WRITE(DSPCLK_GATE_D
, dspclk_gate
);
5485 /* WaDisableRenderCachePipelinedFlush */
5486 I915_WRITE(CACHE_MODE_0
,
5487 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE
));
5489 /* WaDisable_RenderCache_OperationalFlush:g4x */
5490 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
5492 g4x_disable_trickle_feed(dev
);
5495 static void crestline_init_clock_gating(struct drm_device
*dev
)
5497 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5499 I915_WRITE(RENCLK_GATE_D1
, I965_RCC_CLOCK_GATE_DISABLE
);
5500 I915_WRITE(RENCLK_GATE_D2
, 0);
5501 I915_WRITE(DSPCLK_GATE_D
, 0);
5502 I915_WRITE(RAMCLK_GATE_D
, 0);
5503 I915_WRITE16(DEUC
, 0);
5504 I915_WRITE(MI_ARB_STATE
,
5505 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
5507 /* WaDisable_RenderCache_OperationalFlush:gen4 */
5508 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
5511 static void broadwater_init_clock_gating(struct drm_device
*dev
)
5513 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5515 I915_WRITE(RENCLK_GATE_D1
, I965_RCZ_CLOCK_GATE_DISABLE
|
5516 I965_RCC_CLOCK_GATE_DISABLE
|
5517 I965_RCPB_CLOCK_GATE_DISABLE
|
5518 I965_ISC_CLOCK_GATE_DISABLE
|
5519 I965_FBC_CLOCK_GATE_DISABLE
);
5520 I915_WRITE(RENCLK_GATE_D2
, 0);
5521 I915_WRITE(MI_ARB_STATE
,
5522 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE
));
5524 /* WaDisable_RenderCache_OperationalFlush:gen4 */
5525 I915_WRITE(CACHE_MODE_0
, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE
));
5528 static void gen3_init_clock_gating(struct drm_device
*dev
)
5530 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5531 u32 dstate
= I915_READ(D_STATE
);
5533 dstate
|= DSTATE_PLL_D3_OFF
| DSTATE_GFX_CLOCK_GATING
|
5534 DSTATE_DOT_CLOCK_GATING
;
5535 I915_WRITE(D_STATE
, dstate
);
5537 if (IS_PINEVIEW(dev
))
5538 I915_WRITE(ECOSKPD
, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY
));
5540 /* IIR "flip pending" means done if this bit is set */
5541 I915_WRITE(ECOSKPD
, _MASKED_BIT_DISABLE(ECO_FLIP_DONE
));
5543 /* interrupts should cause a wake up from C3 */
5544 I915_WRITE(INSTPM
, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN
));
5546 /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
5547 I915_WRITE(MI_ARB_STATE
, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE
));
5550 static void i85x_init_clock_gating(struct drm_device
*dev
)
5552 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5554 I915_WRITE(RENCLK_GATE_D1
, SV_CLOCK_GATE_DISABLE
);
5556 /* interrupts should cause a wake up from C3 */
5557 I915_WRITE(MI_STATE
, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN
) |
5558 _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE
));
5561 static void i830_init_clock_gating(struct drm_device
*dev
)
5563 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5565 I915_WRITE(DSPCLK_GATE_D
, OVRUNIT_CLOCK_GATE_DISABLE
);
5568 void intel_init_clock_gating(struct drm_device
*dev
)
5570 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
5572 dev_priv
->display
.init_clock_gating(dev
);
5575 void intel_suspend_hw(struct drm_device
*dev
)
5577 if (HAS_PCH_LPT(dev
))
5578 lpt_suspend_hw(dev
);
5581 #define for_each_power_well(i, power_well, domain_mask, power_domains) \
5583 i < (power_domains)->power_well_count && \
5584 ((power_well) = &(power_domains)->power_wells[i]); \
5586 if ((power_well)->domains & (domain_mask))
5588 #define for_each_power_well_rev(i, power_well, domain_mask, power_domains) \
5589 for (i = (power_domains)->power_well_count - 1; \
5590 i >= 0 && ((power_well) = &(power_domains)->power_wells[i]);\
5592 if ((power_well)->domains & (domain_mask))
5595 * We should only use the power well if we explicitly asked the hardware to
5596 * enable it, so check if it's enabled and also check if we've requested it to
5599 static bool hsw_power_well_enabled(struct drm_i915_private
*dev_priv
,
5600 struct i915_power_well
*power_well
)
5602 return I915_READ(HSW_PWR_WELL_DRIVER
) ==
5603 (HSW_PWR_WELL_ENABLE_REQUEST
| HSW_PWR_WELL_STATE_ENABLED
);
5606 bool intel_display_power_enabled_unlocked(struct drm_i915_private
*dev_priv
,
5607 enum intel_display_power_domain domain
)
5609 struct i915_power_domains
*power_domains
;
5610 struct i915_power_well
*power_well
;
5614 if (dev_priv
->pm
.suspended
)
5617 power_domains
= &dev_priv
->power_domains
;
5621 for_each_power_well_rev(i
, power_well
, BIT(domain
), power_domains
) {
5622 if (power_well
->always_on
)
5625 if (!power_well
->hw_enabled
) {
5634 bool intel_display_power_enabled(struct drm_i915_private
*dev_priv
,
5635 enum intel_display_power_domain domain
)
5637 struct i915_power_domains
*power_domains
;
5640 power_domains
= &dev_priv
->power_domains
;
5642 mutex_lock(&power_domains
->lock
);
5643 ret
= intel_display_power_enabled_unlocked(dev_priv
, domain
);
5644 mutex_unlock(&power_domains
->lock
);
5650 * Starting with Haswell, we have a "Power Down Well" that can be turned off
5651 * when not needed anymore. We have 4 registers that can request the power well
5652 * to be enabled, and it will only be disabled if none of the registers is
5653 * requesting it to be enabled.
5655 static void hsw_power_well_post_enable(struct drm_i915_private
*dev_priv
)
5657 struct drm_device
*dev
= dev_priv
->dev
;
5658 unsigned long irqflags
;
5661 * After we re-enable the power well, if we touch VGA register 0x3d5
5662 * we'll get unclaimed register interrupts. This stops after we write
5663 * anything to the VGA MSR register. The vgacon module uses this
5664 * register all the time, so if we unbind our driver and, as a
5665 * consequence, bind vgacon, we'll get stuck in an infinite loop at
5666 * console_unlock(). So make here we touch the VGA MSR register, making
5667 * sure vgacon can keep working normally without triggering interrupts
5668 * and error messages.
5670 vga_get_uninterruptible(dev
->pdev
, VGA_RSRC_LEGACY_IO
);
5671 outb(inb(VGA_MSR_READ
), VGA_MSR_WRITE
);
5672 vga_put(dev
->pdev
, VGA_RSRC_LEGACY_IO
);
5674 if (IS_BROADWELL(dev
)) {
5675 spin_lock_irqsave(&dev_priv
->irq_lock
, irqflags
);
5676 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_B
),
5677 dev_priv
->de_irq_mask
[PIPE_B
]);
5678 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_B
),
5679 ~dev_priv
->de_irq_mask
[PIPE_B
] |
5681 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_C
),
5682 dev_priv
->de_irq_mask
[PIPE_C
]);
5683 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_C
),
5684 ~dev_priv
->de_irq_mask
[PIPE_C
] |
5686 POSTING_READ(GEN8_DE_PIPE_IER(PIPE_C
));
5687 spin_unlock_irqrestore(&dev_priv
->irq_lock
, irqflags
);
5691 static void hsw_set_power_well(struct drm_i915_private
*dev_priv
,
5692 struct i915_power_well
*power_well
, bool enable
)
5694 bool is_enabled
, enable_requested
;
5697 tmp
= I915_READ(HSW_PWR_WELL_DRIVER
);
5698 is_enabled
= tmp
& HSW_PWR_WELL_STATE_ENABLED
;
5699 enable_requested
= tmp
& HSW_PWR_WELL_ENABLE_REQUEST
;
5702 if (!enable_requested
)
5703 I915_WRITE(HSW_PWR_WELL_DRIVER
,
5704 HSW_PWR_WELL_ENABLE_REQUEST
);
5707 DRM_DEBUG_KMS("Enabling power well\n");
5708 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER
) &
5709 HSW_PWR_WELL_STATE_ENABLED
), 20))
5710 DRM_ERROR("Timeout enabling power well\n");
5713 hsw_power_well_post_enable(dev_priv
);
5715 if (enable_requested
) {
5716 I915_WRITE(HSW_PWR_WELL_DRIVER
, 0);
5717 POSTING_READ(HSW_PWR_WELL_DRIVER
);
5718 DRM_DEBUG_KMS("Requesting to disable the power well\n");
5723 static void hsw_power_well_sync_hw(struct drm_i915_private
*dev_priv
,
5724 struct i915_power_well
*power_well
)
5726 hsw_set_power_well(dev_priv
, power_well
, power_well
->count
> 0);
5729 * We're taking over the BIOS, so clear any requests made by it since
5730 * the driver is in charge now.
5732 if (I915_READ(HSW_PWR_WELL_BIOS
) & HSW_PWR_WELL_ENABLE_REQUEST
)
5733 I915_WRITE(HSW_PWR_WELL_BIOS
, 0);
5736 static void hsw_power_well_enable(struct drm_i915_private
*dev_priv
,
5737 struct i915_power_well
*power_well
)
5739 hsw_set_power_well(dev_priv
, power_well
, true);
5742 static void hsw_power_well_disable(struct drm_i915_private
*dev_priv
,
5743 struct i915_power_well
*power_well
)
5745 hsw_set_power_well(dev_priv
, power_well
, false);
5748 static void i9xx_always_on_power_well_noop(struct drm_i915_private
*dev_priv
,
5749 struct i915_power_well
*power_well
)
5753 static bool i9xx_always_on_power_well_enabled(struct drm_i915_private
*dev_priv
,
5754 struct i915_power_well
*power_well
)
5759 void __vlv_set_power_well(struct drm_i915_private
*dev_priv
,
5760 enum punit_power_well power_well_id
, bool enable
)
5762 struct drm_device
*dev
= dev_priv
->dev
;
5768 if (power_well_id
== PUNIT_POWER_WELL_DPIO_CMN_BC
) {
5771 * Enable the CRI clock source so we can get at the
5772 * display and the reference clock for VGA
5773 * hotplug / manual detection.
5775 I915_WRITE(DPLL(PIPE_B
), I915_READ(DPLL(PIPE_B
)) |
5776 DPLL_REFA_CLK_ENABLE_VLV
|
5777 DPLL_INTEGRATED_CRI_CLK_VLV
);
5778 udelay(1); /* >10ns for cmnreset, >0ns for sidereset */
5781 assert_pll_disabled(dev_priv
, pipe
);
5782 /* Assert common reset */
5783 I915_WRITE(DPIO_CTL
, I915_READ(DPIO_CTL
) &
5788 mask
= PUNIT_PWRGT_MASK(power_well_id
);
5789 state
= enable
? PUNIT_PWRGT_PWR_ON(power_well_id
) :
5790 PUNIT_PWRGT_PWR_GATE(power_well_id
);
5792 mutex_lock(&dev_priv
->rps
.hw_lock
);
5795 ((vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_STATUS) & mask) == state)
5800 ctrl
= vlv_punit_read(dev_priv
, PUNIT_REG_PWRGT_CTRL
);
5803 vlv_punit_write(dev_priv
, PUNIT_REG_PWRGT_CTRL
, ctrl
);
5805 if (wait_for(COND
, 100))
5806 DRM_ERROR("timout setting power well state %08x (%08x)\n",
5808 vlv_punit_read(dev_priv
, PUNIT_REG_PWRGT_CTRL
));
5813 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5816 * From VLV2A0_DP_eDP_DPIO_driver_vbios_notes_10.docx -
5817 * 6. De-assert cmn_reset/side_reset. Same as VLV X0.
5818 * a. GUnit 0x2110 bit[0] set to 1 (def 0)
5819 * b. The other bits such as sfr settings / modesel may all
5822 * This should only be done on init and resume from S3 with
5823 * both PLLs disabled, or we risk losing DPIO and PLL
5826 if (power_well_id
== PUNIT_POWER_WELL_DPIO_CMN_BC
&& enable
)
5827 I915_WRITE(DPIO_CTL
, I915_READ(DPIO_CTL
) | DPIO_CMNRST
);
5830 static void vlv_set_power_well(struct drm_i915_private
*dev_priv
,
5831 struct i915_power_well
*power_well
, bool enable
)
5833 enum punit_power_well power_well_id
= power_well
->data
;
5835 __vlv_set_power_well(dev_priv
, power_well_id
, enable
);
5838 static void vlv_power_well_sync_hw(struct drm_i915_private
*dev_priv
,
5839 struct i915_power_well
*power_well
)
5841 vlv_set_power_well(dev_priv
, power_well
, power_well
->count
> 0);
5844 static void vlv_power_well_enable(struct drm_i915_private
*dev_priv
,
5845 struct i915_power_well
*power_well
)
5847 vlv_set_power_well(dev_priv
, power_well
, true);
5850 static void vlv_power_well_disable(struct drm_i915_private
*dev_priv
,
5851 struct i915_power_well
*power_well
)
5853 vlv_set_power_well(dev_priv
, power_well
, false);
5856 static bool vlv_power_well_enabled(struct drm_i915_private
*dev_priv
,
5857 struct i915_power_well
*power_well
)
5859 int power_well_id
= power_well
->data
;
5860 bool enabled
= false;
5865 mask
= PUNIT_PWRGT_MASK(power_well_id
);
5866 ctrl
= PUNIT_PWRGT_PWR_ON(power_well_id
);
5868 mutex_lock(&dev_priv
->rps
.hw_lock
);
5870 state
= vlv_punit_read(dev_priv
, PUNIT_REG_PWRGT_STATUS
) & mask
;
5872 * We only ever set the power-on and power-gate states, anything
5873 * else is unexpected.
5875 WARN_ON(state
!= PUNIT_PWRGT_PWR_ON(power_well_id
) &&
5876 state
!= PUNIT_PWRGT_PWR_GATE(power_well_id
));
5881 * A transient state at this point would mean some unexpected party
5882 * is poking at the power controls too.
5884 ctrl
= vlv_punit_read(dev_priv
, PUNIT_REG_PWRGT_CTRL
) & mask
;
5885 WARN_ON(ctrl
!= state
);
5887 mutex_unlock(&dev_priv
->rps
.hw_lock
);
5892 static void vlv_display_power_well_enable(struct drm_i915_private
*dev_priv
,
5893 struct i915_power_well
*power_well
)
5895 WARN_ON_ONCE(power_well
->data
!= PUNIT_POWER_WELL_DISP2D
);
5897 vlv_set_power_well(dev_priv
, power_well
, true);
5899 spin_lock_irq(&dev_priv
->irq_lock
);
5900 valleyview_enable_display_irqs(dev_priv
);
5901 spin_unlock_irq(&dev_priv
->irq_lock
);
5904 * During driver initialization/resume we can avoid restoring the
5905 * part of the HW/SW state that will be inited anyway explicitly.
5907 if (dev_priv
->power_domains
.initializing
)
5910 intel_hpd_init(dev_priv
->dev
);
5912 i915_redisable_vga_power_on(dev_priv
->dev
);
5915 static void vlv_display_power_well_disable(struct drm_i915_private
*dev_priv
,
5916 struct i915_power_well
*power_well
)
5918 WARN_ON_ONCE(power_well
->data
!= PUNIT_POWER_WELL_DISP2D
);
5920 spin_lock_irq(&dev_priv
->irq_lock
);
5921 valleyview_disable_display_irqs(dev_priv
);
5922 spin_unlock_irq(&dev_priv
->irq_lock
);
5924 vlv_set_power_well(dev_priv
, power_well
, false);
5927 static void check_power_well_state(struct drm_i915_private
*dev_priv
,
5928 struct i915_power_well
*power_well
)
5930 bool enabled
= power_well
->ops
->is_enabled(dev_priv
, power_well
);
5932 if (power_well
->always_on
|| !i915
.disable_power_well
) {
5939 if (enabled
!= (power_well
->count
> 0))
5945 WARN(1, "state mismatch for '%s' (always_on %d hw state %d use-count %d disable_power_well %d\n",
5946 power_well
->name
, power_well
->always_on
, enabled
,
5947 power_well
->count
, i915
.disable_power_well
);
5950 void intel_display_power_get(struct drm_i915_private
*dev_priv
,
5951 enum intel_display_power_domain domain
)
5953 struct i915_power_domains
*power_domains
;
5954 struct i915_power_well
*power_well
;
5957 intel_runtime_pm_get(dev_priv
);
5959 power_domains
= &dev_priv
->power_domains
;
5961 mutex_lock(&power_domains
->lock
);
5963 for_each_power_well(i
, power_well
, BIT(domain
), power_domains
) {
5964 if (!power_well
->count
++) {
5965 DRM_DEBUG_KMS("enabling %s\n", power_well
->name
);
5966 power_well
->ops
->enable(dev_priv
, power_well
);
5967 power_well
->hw_enabled
= true;
5970 check_power_well_state(dev_priv
, power_well
);
5973 power_domains
->domain_use_count
[domain
]++;
5975 mutex_unlock(&power_domains
->lock
);
5978 void intel_display_power_put(struct drm_i915_private
*dev_priv
,
5979 enum intel_display_power_domain domain
)
5981 struct i915_power_domains
*power_domains
;
5982 struct i915_power_well
*power_well
;
5985 power_domains
= &dev_priv
->power_domains
;
5987 mutex_lock(&power_domains
->lock
);
5989 WARN_ON(!power_domains
->domain_use_count
[domain
]);
5990 power_domains
->domain_use_count
[domain
]--;
5992 for_each_power_well_rev(i
, power_well
, BIT(domain
), power_domains
) {
5993 WARN_ON(!power_well
->count
);
5995 if (!--power_well
->count
&& i915
.disable_power_well
) {
5996 DRM_DEBUG_KMS("disabling %s\n", power_well
->name
);
5997 power_well
->hw_enabled
= false;
5998 power_well
->ops
->disable(dev_priv
, power_well
);
6001 check_power_well_state(dev_priv
, power_well
);
6004 mutex_unlock(&power_domains
->lock
);
6006 intel_runtime_pm_put(dev_priv
);
6009 static struct i915_power_domains
*hsw_pwr
;
6011 /* Display audio driver power well request */
6012 int i915_request_power_well(void)
6014 struct drm_i915_private
*dev_priv
;
6019 dev_priv
= container_of(hsw_pwr
, struct drm_i915_private
,
6021 intel_display_power_get(dev_priv
, POWER_DOMAIN_AUDIO
);
6024 EXPORT_SYMBOL_GPL(i915_request_power_well
);
6026 /* Display audio driver power well release */
6027 int i915_release_power_well(void)
6029 struct drm_i915_private
*dev_priv
;
6034 dev_priv
= container_of(hsw_pwr
, struct drm_i915_private
,
6036 intel_display_power_put(dev_priv
, POWER_DOMAIN_AUDIO
);
6039 EXPORT_SYMBOL_GPL(i915_release_power_well
);
6041 #define POWER_DOMAIN_MASK (BIT(POWER_DOMAIN_NUM) - 1)
6043 #define HSW_ALWAYS_ON_POWER_DOMAINS ( \
6044 BIT(POWER_DOMAIN_PIPE_A) | \
6045 BIT(POWER_DOMAIN_TRANSCODER_EDP) | \
6046 BIT(POWER_DOMAIN_PORT_DDI_A_2_LANES) | \
6047 BIT(POWER_DOMAIN_PORT_DDI_A_4_LANES) | \
6048 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
6049 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6050 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
6051 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6052 BIT(POWER_DOMAIN_PORT_DDI_D_2_LANES) | \
6053 BIT(POWER_DOMAIN_PORT_DDI_D_4_LANES) | \
6054 BIT(POWER_DOMAIN_PORT_CRT) | \
6055 BIT(POWER_DOMAIN_INIT))
6056 #define HSW_DISPLAY_POWER_DOMAINS ( \
6057 (POWER_DOMAIN_MASK & ~HSW_ALWAYS_ON_POWER_DOMAINS) | \
6058 BIT(POWER_DOMAIN_INIT))
6060 #define BDW_ALWAYS_ON_POWER_DOMAINS ( \
6061 HSW_ALWAYS_ON_POWER_DOMAINS | \
6062 BIT(POWER_DOMAIN_PIPE_A_PANEL_FITTER))
6063 #define BDW_DISPLAY_POWER_DOMAINS ( \
6064 (POWER_DOMAIN_MASK & ~BDW_ALWAYS_ON_POWER_DOMAINS) | \
6065 BIT(POWER_DOMAIN_INIT))
6067 #define VLV_ALWAYS_ON_POWER_DOMAINS BIT(POWER_DOMAIN_INIT)
6068 #define VLV_DISPLAY_POWER_DOMAINS POWER_DOMAIN_MASK
6070 #define VLV_DPIO_CMN_BC_POWER_DOMAINS ( \
6071 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
6072 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6073 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
6074 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6075 BIT(POWER_DOMAIN_PORT_CRT) | \
6076 BIT(POWER_DOMAIN_INIT))
6078 #define VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS ( \
6079 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
6080 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6081 BIT(POWER_DOMAIN_INIT))
6083 #define VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS ( \
6084 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6085 BIT(POWER_DOMAIN_INIT))
6087 #define VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS ( \
6088 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
6089 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6090 BIT(POWER_DOMAIN_INIT))
6092 #define VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS ( \
6093 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6094 BIT(POWER_DOMAIN_INIT))
6096 static const struct i915_power_well_ops i9xx_always_on_power_well_ops
= {
6097 .sync_hw
= i9xx_always_on_power_well_noop
,
6098 .enable
= i9xx_always_on_power_well_noop
,
6099 .disable
= i9xx_always_on_power_well_noop
,
6100 .is_enabled
= i9xx_always_on_power_well_enabled
,
6103 static struct i915_power_well i9xx_always_on_power_well
[] = {
6105 .name
= "always-on",
6107 .domains
= POWER_DOMAIN_MASK
,
6108 .ops
= &i9xx_always_on_power_well_ops
,
6112 static const struct i915_power_well_ops hsw_power_well_ops
= {
6113 .sync_hw
= hsw_power_well_sync_hw
,
6114 .enable
= hsw_power_well_enable
,
6115 .disable
= hsw_power_well_disable
,
6116 .is_enabled
= hsw_power_well_enabled
,
6119 static struct i915_power_well hsw_power_wells
[] = {
6121 .name
= "always-on",
6123 .domains
= HSW_ALWAYS_ON_POWER_DOMAINS
,
6124 .ops
= &i9xx_always_on_power_well_ops
,
6128 .domains
= HSW_DISPLAY_POWER_DOMAINS
,
6129 .ops
= &hsw_power_well_ops
,
6133 static struct i915_power_well bdw_power_wells
[] = {
6135 .name
= "always-on",
6137 .domains
= BDW_ALWAYS_ON_POWER_DOMAINS
,
6138 .ops
= &i9xx_always_on_power_well_ops
,
6142 .domains
= BDW_DISPLAY_POWER_DOMAINS
,
6143 .ops
= &hsw_power_well_ops
,
6147 static const struct i915_power_well_ops vlv_display_power_well_ops
= {
6148 .sync_hw
= vlv_power_well_sync_hw
,
6149 .enable
= vlv_display_power_well_enable
,
6150 .disable
= vlv_display_power_well_disable
,
6151 .is_enabled
= vlv_power_well_enabled
,
6154 static const struct i915_power_well_ops vlv_dpio_power_well_ops
= {
6155 .sync_hw
= vlv_power_well_sync_hw
,
6156 .enable
= vlv_power_well_enable
,
6157 .disable
= vlv_power_well_disable
,
6158 .is_enabled
= vlv_power_well_enabled
,
6161 static struct i915_power_well vlv_power_wells
[] = {
6163 .name
= "always-on",
6165 .domains
= VLV_ALWAYS_ON_POWER_DOMAINS
,
6166 .ops
= &i9xx_always_on_power_well_ops
,
6170 .domains
= VLV_DISPLAY_POWER_DOMAINS
,
6171 .data
= PUNIT_POWER_WELL_DISP2D
,
6172 .ops
= &vlv_display_power_well_ops
,
6175 .name
= "dpio-tx-b-01",
6176 .domains
= VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS
|
6177 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS
|
6178 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS
|
6179 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS
,
6180 .ops
= &vlv_dpio_power_well_ops
,
6181 .data
= PUNIT_POWER_WELL_DPIO_TX_B_LANES_01
,
6184 .name
= "dpio-tx-b-23",
6185 .domains
= VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS
|
6186 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS
|
6187 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS
|
6188 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS
,
6189 .ops
= &vlv_dpio_power_well_ops
,
6190 .data
= PUNIT_POWER_WELL_DPIO_TX_B_LANES_23
,
6193 .name
= "dpio-tx-c-01",
6194 .domains
= VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS
|
6195 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS
|
6196 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS
|
6197 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS
,
6198 .ops
= &vlv_dpio_power_well_ops
,
6199 .data
= PUNIT_POWER_WELL_DPIO_TX_C_LANES_01
,
6202 .name
= "dpio-tx-c-23",
6203 .domains
= VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS
|
6204 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS
|
6205 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS
|
6206 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS
,
6207 .ops
= &vlv_dpio_power_well_ops
,
6208 .data
= PUNIT_POWER_WELL_DPIO_TX_C_LANES_23
,
6211 .name
= "dpio-common",
6212 .domains
= VLV_DPIO_CMN_BC_POWER_DOMAINS
,
6213 .data
= PUNIT_POWER_WELL_DPIO_CMN_BC
,
6214 .ops
= &vlv_dpio_power_well_ops
,
6218 #define set_power_wells(power_domains, __power_wells) ({ \
6219 (power_domains)->power_wells = (__power_wells); \
6220 (power_domains)->power_well_count = ARRAY_SIZE(__power_wells); \
6223 int intel_power_domains_init(struct drm_i915_private
*dev_priv
)
6225 struct i915_power_domains
*power_domains
= &dev_priv
->power_domains
;
6227 mutex_init(&power_domains
->lock
);
6230 * The enabling order will be from lower to higher indexed wells,
6231 * the disabling order is reversed.
6233 if (IS_HASWELL(dev_priv
->dev
)) {
6234 set_power_wells(power_domains
, hsw_power_wells
);
6235 hsw_pwr
= power_domains
;
6236 } else if (IS_BROADWELL(dev_priv
->dev
)) {
6237 set_power_wells(power_domains
, bdw_power_wells
);
6238 hsw_pwr
= power_domains
;
6239 } else if (IS_VALLEYVIEW(dev_priv
->dev
)) {
6240 set_power_wells(power_domains
, vlv_power_wells
);
6242 set_power_wells(power_domains
, i9xx_always_on_power_well
);
6248 void intel_power_domains_remove(struct drm_i915_private
*dev_priv
)
6253 static void intel_power_domains_resume(struct drm_i915_private
*dev_priv
)
6255 struct i915_power_domains
*power_domains
= &dev_priv
->power_domains
;
6256 struct i915_power_well
*power_well
;
6259 mutex_lock(&power_domains
->lock
);
6260 for_each_power_well(i
, power_well
, POWER_DOMAIN_MASK
, power_domains
) {
6261 power_well
->ops
->sync_hw(dev_priv
, power_well
);
6262 power_well
->hw_enabled
= power_well
->ops
->is_enabled(dev_priv
,
6265 mutex_unlock(&power_domains
->lock
);
6268 void intel_power_domains_init_hw(struct drm_i915_private
*dev_priv
)
6270 struct i915_power_domains
*power_domains
= &dev_priv
->power_domains
;
6272 power_domains
->initializing
= true;
6273 /* For now, we need the power well to be always enabled. */
6274 intel_display_set_init_power(dev_priv
, true);
6275 intel_power_domains_resume(dev_priv
);
6276 power_domains
->initializing
= false;
6279 void intel_aux_display_runtime_get(struct drm_i915_private
*dev_priv
)
6281 intel_runtime_pm_get(dev_priv
);
6284 void intel_aux_display_runtime_put(struct drm_i915_private
*dev_priv
)
6286 intel_runtime_pm_put(dev_priv
);
6289 void intel_runtime_pm_get(struct drm_i915_private
*dev_priv
)
6291 struct drm_device
*dev
= dev_priv
->dev
;
6292 struct device
*device
= &dev
->pdev
->dev
;
6294 if (!HAS_RUNTIME_PM(dev
))
6297 pm_runtime_get_sync(device
);
6298 WARN(dev_priv
->pm
.suspended
, "Device still suspended.\n");
6301 void intel_runtime_pm_get_noresume(struct drm_i915_private
*dev_priv
)
6303 struct drm_device
*dev
= dev_priv
->dev
;
6304 struct device
*device
= &dev
->pdev
->dev
;
6306 if (!HAS_RUNTIME_PM(dev
))
6309 WARN(dev_priv
->pm
.suspended
, "Getting nosync-ref while suspended.\n");
6310 pm_runtime_get_noresume(device
);
6313 void intel_runtime_pm_put(struct drm_i915_private
*dev_priv
)
6315 struct drm_device
*dev
= dev_priv
->dev
;
6316 struct device
*device
= &dev
->pdev
->dev
;
6318 if (!HAS_RUNTIME_PM(dev
))
6321 pm_runtime_mark_last_busy(device
);
6322 pm_runtime_put_autosuspend(device
);
6325 void intel_init_runtime_pm(struct drm_i915_private
*dev_priv
)
6327 struct drm_device
*dev
= dev_priv
->dev
;
6328 struct device
*device
= &dev
->pdev
->dev
;
6330 if (!HAS_RUNTIME_PM(dev
))
6333 pm_runtime_set_active(device
);
6336 * RPM depends on RC6 to save restore the GT HW context, so make RC6 a
6339 if (!intel_enable_rc6(dev
)) {
6340 DRM_INFO("RC6 disabled, disabling runtime PM support\n");
6344 pm_runtime_set_autosuspend_delay(device
, 10000); /* 10s */
6345 pm_runtime_mark_last_busy(device
);
6346 pm_runtime_use_autosuspend(device
);
6348 pm_runtime_put_autosuspend(device
);
6351 void intel_fini_runtime_pm(struct drm_i915_private
*dev_priv
)
6353 struct drm_device
*dev
= dev_priv
->dev
;
6354 struct device
*device
= &dev
->pdev
->dev
;
6356 if (!HAS_RUNTIME_PM(dev
))
6359 if (!intel_enable_rc6(dev
))
6362 /* Make sure we're not suspended first. */
6363 pm_runtime_get_sync(device
);
6364 pm_runtime_disable(device
);
6367 /* Set up chip specific power management-related functions */
6368 void intel_init_pm(struct drm_device
*dev
)
6370 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6373 if (INTEL_INFO(dev
)->gen
>= 7) {
6374 dev_priv
->display
.fbc_enabled
= ironlake_fbc_enabled
;
6375 dev_priv
->display
.enable_fbc
= gen7_enable_fbc
;
6376 dev_priv
->display
.disable_fbc
= ironlake_disable_fbc
;
6377 } else if (INTEL_INFO(dev
)->gen
>= 5) {
6378 dev_priv
->display
.fbc_enabled
= ironlake_fbc_enabled
;
6379 dev_priv
->display
.enable_fbc
= ironlake_enable_fbc
;
6380 dev_priv
->display
.disable_fbc
= ironlake_disable_fbc
;
6381 } else if (IS_GM45(dev
)) {
6382 dev_priv
->display
.fbc_enabled
= g4x_fbc_enabled
;
6383 dev_priv
->display
.enable_fbc
= g4x_enable_fbc
;
6384 dev_priv
->display
.disable_fbc
= g4x_disable_fbc
;
6386 dev_priv
->display
.fbc_enabled
= i8xx_fbc_enabled
;
6387 dev_priv
->display
.enable_fbc
= i8xx_enable_fbc
;
6388 dev_priv
->display
.disable_fbc
= i8xx_disable_fbc
;
6390 /* This value was pulled out of someone's hat */
6391 I915_WRITE(FBC_CONTROL
, 500 << FBC_CTL_INTERVAL_SHIFT
);
6396 if (IS_PINEVIEW(dev
))
6397 i915_pineview_get_mem_freq(dev
);
6398 else if (IS_GEN5(dev
))
6399 i915_ironlake_get_mem_freq(dev
);
6401 /* For FIFO watermark updates */
6402 if (HAS_PCH_SPLIT(dev
)) {
6403 ilk_setup_wm_latency(dev
);
6405 if ((IS_GEN5(dev
) && dev_priv
->wm
.pri_latency
[1] &&
6406 dev_priv
->wm
.spr_latency
[1] && dev_priv
->wm
.cur_latency
[1]) ||
6407 (!IS_GEN5(dev
) && dev_priv
->wm
.pri_latency
[0] &&
6408 dev_priv
->wm
.spr_latency
[0] && dev_priv
->wm
.cur_latency
[0])) {
6409 dev_priv
->display
.update_wm
= ilk_update_wm
;
6410 dev_priv
->display
.update_sprite_wm
= ilk_update_sprite_wm
;
6412 DRM_DEBUG_KMS("Failed to read display plane latency. "
6417 dev_priv
->display
.init_clock_gating
= ironlake_init_clock_gating
;
6418 else if (IS_GEN6(dev
))
6419 dev_priv
->display
.init_clock_gating
= gen6_init_clock_gating
;
6420 else if (IS_IVYBRIDGE(dev
))
6421 dev_priv
->display
.init_clock_gating
= ivybridge_init_clock_gating
;
6422 else if (IS_HASWELL(dev
))
6423 dev_priv
->display
.init_clock_gating
= haswell_init_clock_gating
;
6424 else if (INTEL_INFO(dev
)->gen
== 8)
6425 dev_priv
->display
.init_clock_gating
= gen8_init_clock_gating
;
6426 } else if (IS_CHERRYVIEW(dev
)) {
6427 dev_priv
->display
.update_wm
= valleyview_update_wm
;
6428 dev_priv
->display
.init_clock_gating
=
6429 cherryview_init_clock_gating
;
6430 } else if (IS_VALLEYVIEW(dev
)) {
6431 dev_priv
->display
.update_wm
= valleyview_update_wm
;
6432 dev_priv
->display
.init_clock_gating
=
6433 valleyview_init_clock_gating
;
6434 } else if (IS_PINEVIEW(dev
)) {
6435 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev
),
6438 dev_priv
->mem_freq
)) {
6439 DRM_INFO("failed to find known CxSR latency "
6440 "(found ddr%s fsb freq %d, mem freq %d), "
6442 (dev_priv
->is_ddr3
== 1) ? "3" : "2",
6443 dev_priv
->fsb_freq
, dev_priv
->mem_freq
);
6444 /* Disable CxSR and never update its watermark again */
6445 pineview_disable_cxsr(dev
);
6446 dev_priv
->display
.update_wm
= NULL
;
6448 dev_priv
->display
.update_wm
= pineview_update_wm
;
6449 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
6450 } else if (IS_G4X(dev
)) {
6451 dev_priv
->display
.update_wm
= g4x_update_wm
;
6452 dev_priv
->display
.init_clock_gating
= g4x_init_clock_gating
;
6453 } else if (IS_GEN4(dev
)) {
6454 dev_priv
->display
.update_wm
= i965_update_wm
;
6455 if (IS_CRESTLINE(dev
))
6456 dev_priv
->display
.init_clock_gating
= crestline_init_clock_gating
;
6457 else if (IS_BROADWATER(dev
))
6458 dev_priv
->display
.init_clock_gating
= broadwater_init_clock_gating
;
6459 } else if (IS_GEN3(dev
)) {
6460 dev_priv
->display
.update_wm
= i9xx_update_wm
;
6461 dev_priv
->display
.get_fifo_size
= i9xx_get_fifo_size
;
6462 dev_priv
->display
.init_clock_gating
= gen3_init_clock_gating
;
6463 } else if (IS_GEN2(dev
)) {
6464 if (INTEL_INFO(dev
)->num_pipes
== 1) {
6465 dev_priv
->display
.update_wm
= i845_update_wm
;
6466 dev_priv
->display
.get_fifo_size
= i845_get_fifo_size
;
6468 dev_priv
->display
.update_wm
= i9xx_update_wm
;
6469 dev_priv
->display
.get_fifo_size
= i830_get_fifo_size
;
6472 if (IS_I85X(dev
) || IS_I865G(dev
))
6473 dev_priv
->display
.init_clock_gating
= i85x_init_clock_gating
;
6475 dev_priv
->display
.init_clock_gating
= i830_init_clock_gating
;
6477 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
6481 int sandybridge_pcode_read(struct drm_i915_private
*dev_priv
, u8 mbox
, u32
*val
)
6483 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
6485 if (I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) {
6486 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
6490 I915_WRITE(GEN6_PCODE_DATA
, *val
);
6491 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
| mbox
);
6493 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
6495 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox
);
6499 *val
= I915_READ(GEN6_PCODE_DATA
);
6500 I915_WRITE(GEN6_PCODE_DATA
, 0);
6505 int sandybridge_pcode_write(struct drm_i915_private
*dev_priv
, u8 mbox
, u32 val
)
6507 WARN_ON(!mutex_is_locked(&dev_priv
->rps
.hw_lock
));
6509 if (I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) {
6510 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
6514 I915_WRITE(GEN6_PCODE_DATA
, val
);
6515 I915_WRITE(GEN6_PCODE_MAILBOX
, GEN6_PCODE_READY
| mbox
);
6517 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX
) & GEN6_PCODE_READY
) == 0,
6519 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox
);
6523 I915_WRITE(GEN6_PCODE_DATA
, 0);
6528 int vlv_gpu_freq(struct drm_i915_private
*dev_priv
, int val
)
6533 switch (dev_priv
->mem_freq
) {
6547 return DIV_ROUND_CLOSEST(dev_priv
->mem_freq
* (val
+ 6 - 0xbd), 4 * div
);
6550 int vlv_freq_opcode(struct drm_i915_private
*dev_priv
, int val
)
6555 switch (dev_priv
->mem_freq
) {
6569 return DIV_ROUND_CLOSEST(4 * mul
* val
, dev_priv
->mem_freq
) + 0xbd - 6;
6572 void intel_pm_setup(struct drm_device
*dev
)
6574 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
6576 mutex_init(&dev_priv
->rps
.hw_lock
);
6578 INIT_DELAYED_WORK(&dev_priv
->rps
.delayed_resume_work
,
6579 intel_gen6_powersave_work
);
6581 dev_priv
->pm
.suspended
= false;
6582 dev_priv
->pm
.irqs_disabled
= false;