]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/gpu/drm/i915/intel_pm.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge remote-tracking branch 'airlied/drm-next' into HEAD
[mirror_ubuntu-artful-kernel.git] / drivers / gpu / drm / i915 / intel_pm.c
CommitLineData
85208be0
ED		 1/*
2 * Copyright © 2012 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Eugeni Dodonov <eugeni.dodonov@intel.com>
25 *
26 */
27
2b4e57bd 28#include <linux/cpufreq.h>
85208be0
ED		 29#include "i915_drv.h"
30#include "intel_drv.h"
eb48eb00
DV		 31#include "../../../platform/x86/intel_ips.h"
32#include <linux/module.h>
f9dcb0df 33#include <linux/vgaarb.h>
f4db9321 34#include <drm/i915_powerwell.h>
8a187455 35#include <linux/pm_runtime.h>
85208be0 36
dc39fff7
BW		 37/**
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.
42 *
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.
46 *
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.
53 */
54#define INTEL_RC6_ENABLE (1<<0)
55#define INTEL_RC6p_ENABLE (1<<1)
56#define INTEL_RC6pp_ENABLE (1<<2)
57
f6750b3c
ED		 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.
85208be0 61 *
f6750b3c
ED		 62 * The benefits of FBC are mostly visible with solid backgrounds and
63 * variation-less patterns.
85208be0 64 *
f6750b3c
ED		 65 * FBC-related functionality can be enabled by the means of the
66 * i915.i915_enable_fbc parameter
85208be0
ED		 67 */
68
1fa61106 69static void i8xx_disable_fbc(struct drm_device *dev)
85208be0
ED		 70{
71 struct drm_i915_private *dev_priv = dev->dev_private;
72 u32 fbc_ctl;
73
74 /* Disable compression */
75 fbc_ctl = I915_READ(FBC_CONTROL);
76 if ((fbc_ctl & FBC_CTL_EN) == 0)
77 return;
78
79 fbc_ctl &= ~FBC_CTL_EN;
80 I915_WRITE(FBC_CONTROL, fbc_ctl);
81
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");
85 return;
86 }
87
88 DRM_DEBUG_KMS("disabled FBC\n");
89}
90
993495ae 91static void i8xx_enable_fbc(struct drm_crtc *crtc)
85208be0
ED		 92{
93 struct drm_device *dev = crtc->dev;
94 struct drm_i915_private *dev_priv = dev->dev_private;
f4510a27 95 struct drm_framebuffer *fb = crtc->primary->fb;
85208be0
ED		 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);
99 int cfb_pitch;
7f2cf220 100 int i;
159f9875 101 u32 fbc_ctl;
85208be0 102
5c3fe8b0 103 cfb_pitch = dev_priv->fbc.size / FBC_LL_SIZE;
85208be0
ED		 104 if (fb->pitches[0] < cfb_pitch)
105 cfb_pitch = fb->pitches[0];
106
42a430f5
VS		 107 /* FBC_CTL wants 32B or 64B units */
108 if (IS_GEN2(dev))
109 cfb_pitch = (cfb_pitch / 32) - 1;
110 else
111 cfb_pitch = (cfb_pitch / 64) - 1;
85208be0
ED		 112
113 /* Clear old tags */
114 for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++)
115 I915_WRITE(FBC_TAG + (i * 4), 0);
116
159f9875
VS		 117 if (IS_GEN4(dev)) {
118 u32 fbc_ctl2;
119
120 /* Set it up... */
121 fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE;
7f2cf220 122 fbc_ctl2 |= FBC_CTL_PLANE(intel_crtc->plane);
159f9875
VS		 123 I915_WRITE(FBC_CONTROL2, fbc_ctl2);
124 I915_WRITE(FBC_FENCE_OFF, crtc->y);
125 }
85208be0
ED		 126
127 /* enable it... */
993495ae
VS		 128 fbc_ctl = I915_READ(FBC_CONTROL);
129 fbc_ctl &= 0x3fff << FBC_CTL_INTERVAL_SHIFT;
130 fbc_ctl |= FBC_CTL_EN | FBC_CTL_PERIODIC;
85208be0
ED		 131 if (IS_I945GM(dev))
132 fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */
133 fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT;
85208be0
ED		 134 fbc_ctl |= obj->fence_reg;
135 I915_WRITE(FBC_CONTROL, fbc_ctl);
136
5cd5410e 137 DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c\n",
84f44ce7 138 cfb_pitch, crtc->y, plane_name(intel_crtc->plane));
85208be0
ED		 139}
140
1fa61106 141static bool i8xx_fbc_enabled(struct drm_device *dev)
85208be0
ED		 142{
143 struct drm_i915_private *dev_priv = dev->dev_private;
144
145 return I915_READ(FBC_CONTROL) & FBC_CTL_EN;
146}
147
993495ae 148static void g4x_enable_fbc(struct drm_crtc *crtc)
85208be0
ED		 149{
150 struct drm_device *dev = crtc->dev;
151 struct drm_i915_private *dev_priv = dev->dev_private;
f4510a27 152 struct drm_framebuffer *fb = crtc->primary->fb;
85208be0
ED		 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);
85208be0
ED		 156 u32 dpfc_ctl;
157
3fa2e0ee
VS		 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;
161 else
162 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
85208be0 163 dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg;
85208be0 164
85208be0
ED		 165 I915_WRITE(DPFC_FENCE_YOFF, crtc->y);
166
167 /* enable it... */
fe74c1a5 168 I915_WRITE(DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
85208be0 169
84f44ce7 170 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
85208be0
ED		 171}
172
1fa61106 173static void g4x_disable_fbc(struct drm_device *dev)
85208be0
ED		 174{
175 struct drm_i915_private *dev_priv = dev->dev_private;
176 u32 dpfc_ctl;
177
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);
183
184 DRM_DEBUG_KMS("disabled FBC\n");
185 }
186}
187
1fa61106 188static bool g4x_fbc_enabled(struct drm_device *dev)
85208be0
ED		 189{
190 struct drm_i915_private *dev_priv = dev->dev_private;
191
192 return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN;
193}
194
195static void sandybridge_blit_fbc_update(struct drm_device *dev)
196{
197 struct drm_i915_private *dev_priv = dev->dev_private;
198 u32 blt_ecoskpd;
199
200 /* Make sure blitter notifies FBC of writes */
940aece4
D		 201
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);
c8d9a590 205
85208be0
ED		 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);
c8d9a590 216
940aece4 217 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_MEDIA);
85208be0
ED		 218}
219
993495ae 220static void ironlake_enable_fbc(struct drm_crtc *crtc)
85208be0
ED		 221{
222 struct drm_device *dev = crtc->dev;
223 struct drm_i915_private *dev_priv = dev->dev_private;
f4510a27 224 struct drm_framebuffer *fb = crtc->primary->fb;
85208be0
ED		 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);
85208be0
ED		 228 u32 dpfc_ctl;
229
46f3dab9 230 dpfc_ctl = DPFC_CTL_PLANE(intel_crtc->plane);
3fa2e0ee
VS		 231 if (drm_format_plane_cpp(fb->pixel_format, 0) == 2)
232 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
233 else
234 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
d629336b
VS		 235 dpfc_ctl |= DPFC_CTL_FENCE_EN;
236 if (IS_GEN5(dev))
237 dpfc_ctl |= obj->fence_reg;
85208be0 238
85208be0 239 I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
f343c5f6 240 I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID);
85208be0
ED		 241 /* enable it... */
242 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
243
244 if (IS_GEN6(dev)) {
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);
249 }
250
84f44ce7 251 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
85208be0
ED		 252}
253
1fa61106 254static void ironlake_disable_fbc(struct drm_device *dev)
85208be0
ED		 255{
256 struct drm_i915_private *dev_priv = dev->dev_private;
257 u32 dpfc_ctl;
258
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);
264
265 DRM_DEBUG_KMS("disabled FBC\n");
266 }
267}
268
1fa61106 269static bool ironlake_fbc_enabled(struct drm_device *dev)
85208be0
ED		 270{
271 struct drm_i915_private *dev_priv = dev->dev_private;
272
273 return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
274}
275
993495ae 276static void gen7_enable_fbc(struct drm_crtc *crtc)
abe959c7
RV		 277{
278 struct drm_device *dev = crtc->dev;
279 struct drm_i915_private *dev_priv = dev->dev_private;
f4510a27 280 struct drm_framebuffer *fb = crtc->primary->fb;
abe959c7
RV		 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);
3fa2e0ee 284 u32 dpfc_ctl;
abe959c7 285
3fa2e0ee
VS		 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;
289 else
290 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
291 dpfc_ctl |= IVB_DPFC_CTL_FENCE_EN;
292
293 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
abe959c7 294
891348b2 295 if (IS_IVYBRIDGE(dev)) {
7dd23ba0 296 /* WaFbcAsynchFlipDisableFbcQueue:ivb */
2adb6db8
VS		 297 I915_WRITE(ILK_DISPLAY_CHICKEN1,
298 I915_READ(ILK_DISPLAY_CHICKEN1) |
299 ILK_FBCQ_DIS);
28554164 300 } else {
2adb6db8 301 /* WaFbcAsynchFlipDisableFbcQueue:hsw,bdw */
8f670bb1
VS		 302 I915_WRITE(CHICKEN_PIPESL_1(intel_crtc->pipe),
303 I915_READ(CHICKEN_PIPESL_1(intel_crtc->pipe)) |
304 HSW_FBCQ_DIS);
891348b2 305 }
b74ea102 306
abe959c7
RV		 307 I915_WRITE(SNB_DPFC_CTL_SA,
308 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
309 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
310
311 sandybridge_blit_fbc_update(dev);
312
b19870ee 313 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
abe959c7
RV		 314}
315
85208be0
ED		 316bool intel_fbc_enabled(struct drm_device *dev)
317{
318 struct drm_i915_private *dev_priv = dev->dev_private;
319
320 if (!dev_priv->display.fbc_enabled)
321 return false;
322
323 return dev_priv->display.fbc_enabled(dev);
324}
325
326static void intel_fbc_work_fn(struct work_struct *__work)
327{
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;
333
334 mutex_lock(&dev->struct_mutex);
5c3fe8b0 335 if (work == dev_priv->fbc.fbc_work) {
85208be0
ED		 336 /* Double check that we haven't switched fb without cancelling
337 * the prior work.
338 */
f4510a27 339 if (work->crtc->primary->fb == work->fb) {
993495ae 340 dev_priv->display.enable_fbc(work->crtc);
85208be0 341
5c3fe8b0 342 dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane;
f4510a27 343 dev_priv->fbc.fb_id = work->crtc->primary->fb->base.id;
5c3fe8b0 344 dev_priv->fbc.y = work->crtc->y;
85208be0
ED		 345 }
346
5c3fe8b0 347 dev_priv->fbc.fbc_work = NULL;
85208be0
ED		 348 }
349 mutex_unlock(&dev->struct_mutex);
350
351 kfree(work);
352}
353
354static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
355{
5c3fe8b0 356 if (dev_priv->fbc.fbc_work == NULL)
85208be0
ED		 357 return;
358
359 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
360
361 /* Synchronisation is provided by struct_mutex and checking of
5c3fe8b0 362 * dev_priv->fbc.fbc_work, so we can perform the cancellation
85208be0
ED		 363 * entirely asynchronously.
364 */
5c3fe8b0 365 if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work))
85208be0 366 /* tasklet was killed before being run, clean up */
5c3fe8b0 367 kfree(dev_priv->fbc.fbc_work);
85208be0
ED		 368
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
372 * necessary to run.
373 */
5c3fe8b0 374 dev_priv->fbc.fbc_work = NULL;
85208be0
ED		 375}
376
993495ae 377static void intel_enable_fbc(struct drm_crtc *crtc)
85208be0
ED		 378{
379 struct intel_fbc_work *work;
380 struct drm_device *dev = crtc->dev;
381 struct drm_i915_private *dev_priv = dev->dev_private;
382
383 if (!dev_priv->display.enable_fbc)
384 return;
385
386 intel_cancel_fbc_work(dev_priv);
387
b14c5679 388 work = kzalloc(sizeof(*work), GFP_KERNEL);
85208be0 389 if (work == NULL) {
6cdcb5e7 390 DRM_ERROR("Failed to allocate FBC work structure\n");
993495ae 391 dev_priv->display.enable_fbc(crtc);
85208be0
ED		 392 return;
393 }
394
395 work->crtc = crtc;
f4510a27 396 work->fb = crtc->primary->fb;
85208be0
ED		 397 INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
398
5c3fe8b0 399 dev_priv->fbc.fbc_work = work;
85208be0 400
85208be0
ED		 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.
406 *
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.
7457d617
DL		 411 *
412 * WaFbcWaitForVBlankBeforeEnable:ilk,snb
85208be0
ED		 413 */
414 schedule_delayed_work(&work->work, msecs_to_jiffies(50));
415}
416
417void intel_disable_fbc(struct drm_device *dev)
418{
419 struct drm_i915_private *dev_priv = dev->dev_private;
420
421 intel_cancel_fbc_work(dev_priv);
422
423 if (!dev_priv->display.disable_fbc)
424 return;
425
426 dev_priv->display.disable_fbc(dev);
5c3fe8b0 427 dev_priv->fbc.plane = -1;
85208be0
ED		 428}
429
29ebf90f
CW		 430static bool set_no_fbc_reason(struct drm_i915_private *dev_priv,
431 enum no_fbc_reason reason)
432{
433 if (dev_priv->fbc.no_fbc_reason == reason)
434 return false;
435
436 dev_priv->fbc.no_fbc_reason = reason;
437 return true;
438}
439
85208be0
ED		 440/**
441 * intel_update_fbc - enable/disable FBC as needed
442 * @dev: the drm_device
443 *
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
449 * - no dual wide
f85da868 450 * - framebuffer <= max_hdisplay in width, max_vdisplay in height
85208be0
ED		 451 *
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
455 * stolen memory.
456 *
457 * We need to enable/disable FBC on a global basis.
458 */
459void intel_update_fbc(struct drm_device *dev)
460{
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;
ef644fda 467 const struct drm_display_mode *adjusted_mode;
37327abd 468 unsigned int max_width, max_height;
85208be0 469
3a77c4c4 470 if (!HAS_FBC(dev)) {
29ebf90f 471 set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED);
85208be0 472 return;
29ebf90f 473 }
85208be0 474
d330a953 475 if (!i915.powersave) {
29ebf90f
CW		 476 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
477 DRM_DEBUG_KMS("fbc disabled per module param\n");
85208be0 478 return;
29ebf90f 479 }
85208be0
ED		 480
481 /*
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.)
489 */
490 list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) {
3490ea5d 491 if (intel_crtc_active(tmp_crtc) &&
4c445e0e 492 to_intel_crtc(tmp_crtc)->primary_enabled) {
85208be0 493 if (crtc) {
29ebf90f
CW		 494 if (set_no_fbc_reason(dev_priv, FBC_MULTIPLE_PIPES))
495 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
85208be0
ED		 496 goto out_disable;
497 }
498 crtc = tmp_crtc;
499 }
500 }
501
f4510a27 502 if (!crtc || crtc->primary->fb == NULL) {
29ebf90f
CW		 503 if (set_no_fbc_reason(dev_priv, FBC_NO_OUTPUT))
504 DRM_DEBUG_KMS("no output, disabling\n");
85208be0
ED		 505 goto out_disable;
506 }
507
508 intel_crtc = to_intel_crtc(crtc);
f4510a27 509 fb = crtc->primary->fb;
85208be0
ED		 510 intel_fb = to_intel_framebuffer(fb);
511 obj = intel_fb->obj;
ef644fda 512 adjusted_mode = &intel_crtc->config.adjusted_mode;
85208be0 513
d330a953 514 if (i915.enable_fbc < 0 &&
8a5729a3 515 INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev)) {
29ebf90f
CW		 516 if (set_no_fbc_reason(dev_priv, FBC_CHIP_DEFAULT))
517 DRM_DEBUG_KMS("disabled per chip default\n");
8a5729a3 518 goto out_disable;
85208be0 519 }
d330a953 520 if (!i915.enable_fbc) {
29ebf90f
CW		 521 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
522 DRM_DEBUG_KMS("fbc disabled per module param\n");
85208be0
ED		 523 goto out_disable;
524 }
ef644fda
VS		 525 if ((adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) ||
526 (adjusted_mode->flags & DRM_MODE_FLAG_DBLSCAN)) {
29ebf90f
CW		 527 if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE))
528 DRM_DEBUG_KMS("mode incompatible with compression, "
529 "disabling\n");
85208be0
ED		 530 goto out_disable;
531 }
f85da868
PZ		 532
533 if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) {
37327abd
VS		 534 max_width = 4096;
535 max_height = 2048;
f85da868 536 } else {
37327abd
VS		 537 max_width = 2048;
538 max_height = 1536;
f85da868 539 }
37327abd
VS		 540 if (intel_crtc->config.pipe_src_w > max_width ||
541 intel_crtc->config.pipe_src_h > max_height) {
29ebf90f
CW		 542 if (set_no_fbc_reason(dev_priv, FBC_MODE_TOO_LARGE))
543 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
85208be0
ED		 544 goto out_disable;
545 }
8f94d24b 546 if ((INTEL_INFO(dev)->gen < 4 || HAS_DDI(dev)) &&
c5a44aa0 547 intel_crtc->plane != PLANE_A) {
29ebf90f 548 if (set_no_fbc_reason(dev_priv, FBC_BAD_PLANE))
c5a44aa0 549 DRM_DEBUG_KMS("plane not A, disabling compression\n");
85208be0
ED		 550 goto out_disable;
551 }
552
553 /* The use of a CPU fence is mandatory in order to detect writes
554 * by the CPU to the scanout and trigger updates to the FBC.
555 */
556 if (obj->tiling_mode != I915_TILING_X ||
557 obj->fence_reg == I915_FENCE_REG_NONE) {
29ebf90f
CW		 558 if (set_no_fbc_reason(dev_priv, FBC_NOT_TILED))
559 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
85208be0
ED		 560 goto out_disable;
561 }
562
563 /* If the kernel debugger is active, always disable compression */
564 if (in_dbg_master())
565 goto out_disable;
566
11be49eb 567 if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) {
29ebf90f
CW		 568 if (set_no_fbc_reason(dev_priv, FBC_STOLEN_TOO_SMALL))
569 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
11be49eb
CW		 570 goto out_disable;
571 }
572
85208be0
ED		 573 /* If the scanout has not changed, don't modify the FBC settings.
574 * Note that we make the fundamental assumption that the fb->obj
575 * cannot be unpinned (and have its GTT offset and fence revoked)
576 * without first being decoupled from the scanout and FBC disabled.
577 */
5c3fe8b0
BW		 578 if (dev_priv->fbc.plane == intel_crtc->plane &&
579 dev_priv->fbc.fb_id == fb->base.id &&
580 dev_priv->fbc.y == crtc->y)
85208be0
ED		 581 return;
582
583 if (intel_fbc_enabled(dev)) {
584 /* We update FBC along two paths, after changing fb/crtc
585 * configuration (modeswitching) and after page-flipping
586 * finishes. For the latter, we know that not only did
587 * we disable the FBC at the start of the page-flip
588 * sequence, but also more than one vblank has passed.
589 *
590 * For the former case of modeswitching, it is possible
591 * to switch between two FBC valid configurations
592 * instantaneously so we do need to disable the FBC
593 * before we can modify its control registers. We also
594 * have to wait for the next vblank for that to take
595 * effect. However, since we delay enabling FBC we can
596 * assume that a vblank has passed since disabling and
597 * that we can safely alter the registers in the deferred
598 * callback.
599 *
600 * In the scenario that we go from a valid to invalid
601 * and then back to valid FBC configuration we have
602 * no strict enforcement that a vblank occurred since
603 * disabling the FBC. However, along all current pipe
604 * disabling paths we do need to wait for a vblank at
605 * some point. And we wait before enabling FBC anyway.
606 */
607 DRM_DEBUG_KMS("disabling active FBC for update\n");
608 intel_disable_fbc(dev);
609 }
610
993495ae 611 intel_enable_fbc(crtc);
29ebf90f 612 dev_priv->fbc.no_fbc_reason = FBC_OK;
85208be0
ED		 613 return;
614
615out_disable:
616 /* Multiple disables should be harmless */
617 if (intel_fbc_enabled(dev)) {
618 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
619 intel_disable_fbc(dev);
620 }
11be49eb 621 i915_gem_stolen_cleanup_compression(dev);
85208be0
ED		 622}
623
c921aba8
DV		 624static void i915_pineview_get_mem_freq(struct drm_device *dev)
625{
50227e1c 626 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV		 627 u32 tmp;
628
629 tmp = I915_READ(CLKCFG);
630
631 switch (tmp & CLKCFG_FSB_MASK) {
632 case CLKCFG_FSB_533:
633 dev_priv->fsb_freq = 533; /* 133*4 */
634 break;
635 case CLKCFG_FSB_800:
636 dev_priv->fsb_freq = 800; /* 200*4 */
637 break;
638 case CLKCFG_FSB_667:
639 dev_priv->fsb_freq = 667; /* 167*4 */
640 break;
641 case CLKCFG_FSB_400:
642 dev_priv->fsb_freq = 400; /* 100*4 */
643 break;
644 }
645
646 switch (tmp & CLKCFG_MEM_MASK) {
647 case CLKCFG_MEM_533:
648 dev_priv->mem_freq = 533;
649 break;
650 case CLKCFG_MEM_667:
651 dev_priv->mem_freq = 667;
652 break;
653 case CLKCFG_MEM_800:
654 dev_priv->mem_freq = 800;
655 break;
656 }
657
658 /* detect pineview DDR3 setting */
659 tmp = I915_READ(CSHRDDR3CTL);
660 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
661}
662
663static void i915_ironlake_get_mem_freq(struct drm_device *dev)
664{
50227e1c 665 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV		 666 u16 ddrpll, csipll;
667
668 ddrpll = I915_READ16(DDRMPLL1);
669 csipll = I915_READ16(CSIPLL0);
670
671 switch (ddrpll & 0xff) {
672 case 0xc:
673 dev_priv->mem_freq = 800;
674 break;
675 case 0x10:
676 dev_priv->mem_freq = 1066;
677 break;
678 case 0x14:
679 dev_priv->mem_freq = 1333;
680 break;
681 case 0x18:
682 dev_priv->mem_freq = 1600;
683 break;
684 default:
685 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
686 ddrpll & 0xff);
687 dev_priv->mem_freq = 0;
688 break;
689 }
690
20e4d407 691 dev_priv->ips.r_t = dev_priv->mem_freq;
c921aba8
DV		 692
693 switch (csipll & 0x3ff) {
694 case 0x00c:
695 dev_priv->fsb_freq = 3200;
696 break;
697 case 0x00e:
698 dev_priv->fsb_freq = 3733;
699 break;
700 case 0x010:
701 dev_priv->fsb_freq = 4266;
702 break;
703 case 0x012:
704 dev_priv->fsb_freq = 4800;
705 break;
706 case 0x014:
707 dev_priv->fsb_freq = 5333;
708 break;
709 case 0x016:
710 dev_priv->fsb_freq = 5866;
711 break;
712 case 0x018:
713 dev_priv->fsb_freq = 6400;
714 break;
715 default:
716 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
717 csipll & 0x3ff);
718 dev_priv->fsb_freq = 0;
719 break;
720 }
721
722 if (dev_priv->fsb_freq == 3200) {
20e4d407 723 dev_priv->ips.c_m = 0;
c921aba8 724 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
20e4d407 725 dev_priv->ips.c_m = 1;
c921aba8 726 } else {
20e4d407 727 dev_priv->ips.c_m = 2;
c921aba8
DV		 728 }
729}
730
b445e3b0
ED		 731static const struct cxsr_latency cxsr_latency_table[] = {
732 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
733 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
734 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
735 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
736 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
737
738 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
739 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
740 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
741 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
742 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
743
744 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
745 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
746 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
747 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
748 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
749
750 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
751 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
752 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
753 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
754 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
755
756 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
757 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
758 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
759 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
760 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
761
762 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
763 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
764 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
765 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
766 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
767};
768
63c62275 769static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
b445e3b0
ED		 770 int is_ddr3,
771 int fsb,
772 int mem)
773{
774 const struct cxsr_latency *latency;
775 int i;
776
777 if (fsb == 0 || mem == 0)
778 return NULL;
779
780 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
781 latency = &cxsr_latency_table[i];
782 if (is_desktop == latency->is_desktop &&
783 is_ddr3 == latency->is_ddr3 &&
784 fsb == latency->fsb_freq && mem == latency->mem_freq)
785 return latency;
786 }
787
788 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
789
790 return NULL;
791}
792
1fa61106 793static void pineview_disable_cxsr(struct drm_device *dev)
b445e3b0
ED		 794{
795 struct drm_i915_private *dev_priv = dev->dev_private;
796
797 /* deactivate cxsr */
798 I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
799}
800
801/*
802 * Latency for FIFO fetches is dependent on several factors:
803 * - memory configuration (speed, channels)
804 * - chipset
805 * - current MCH state
806 * It can be fairly high in some situations, so here we assume a fairly
807 * pessimal value. It's a tradeoff between extra memory fetches (if we
808 * set this value too high, the FIFO will fetch frequently to stay full)
809 * and power consumption (set it too low to save power and we might see
810 * FIFO underruns and display "flicker").
811 *
812 * A value of 5us seems to be a good balance; safe for very low end
813 * platforms but not overly aggressive on lower latency configs.
814 */
815static const int latency_ns = 5000;
816
1fa61106 817static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 818{
819 struct drm_i915_private *dev_priv = dev->dev_private;
820 uint32_t dsparb = I915_READ(DSPARB);
821 int size;
822
823 size = dsparb & 0x7f;
824 if (plane)
825 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
826
827 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
828 plane ? "B" : "A", size);
829
830 return size;
831}
832
feb56b93 833static int i830_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 834{
835 struct drm_i915_private *dev_priv = dev->dev_private;
836 uint32_t dsparb = I915_READ(DSPARB);
837 int size;
838
839 size = dsparb & 0x1ff;
840 if (plane)
841 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
842 size >>= 1; /* Convert to cachelines */
843
844 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
845 plane ? "B" : "A", size);
846
847 return size;
848}
849
1fa61106 850static int i845_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 851{
852 struct drm_i915_private *dev_priv = dev->dev_private;
853 uint32_t dsparb = I915_READ(DSPARB);
854 int size;
855
856 size = dsparb & 0x7f;
857 size >>= 2; /* Convert to cachelines */
858
859 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
860 plane ? "B" : "A",
861 size);
862
863 return size;
864}
865
b445e3b0
ED		 866/* Pineview has different values for various configs */
867static const struct intel_watermark_params pineview_display_wm = {
868 PINEVIEW_DISPLAY_FIFO,
869 PINEVIEW_MAX_WM,
870 PINEVIEW_DFT_WM,
871 PINEVIEW_GUARD_WM,
872 PINEVIEW_FIFO_LINE_SIZE
873};
874static const struct intel_watermark_params pineview_display_hplloff_wm = {
875 PINEVIEW_DISPLAY_FIFO,
876 PINEVIEW_MAX_WM,
877 PINEVIEW_DFT_HPLLOFF_WM,
878 PINEVIEW_GUARD_WM,
879 PINEVIEW_FIFO_LINE_SIZE
880};
881static const struct intel_watermark_params pineview_cursor_wm = {
882 PINEVIEW_CURSOR_FIFO,
883 PINEVIEW_CURSOR_MAX_WM,
884 PINEVIEW_CURSOR_DFT_WM,
885 PINEVIEW_CURSOR_GUARD_WM,
886 PINEVIEW_FIFO_LINE_SIZE,
887};
888static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
889 PINEVIEW_CURSOR_FIFO,
890 PINEVIEW_CURSOR_MAX_WM,
891 PINEVIEW_CURSOR_DFT_WM,
892 PINEVIEW_CURSOR_GUARD_WM,
893 PINEVIEW_FIFO_LINE_SIZE
894};
895static const struct intel_watermark_params g4x_wm_info = {
896 G4X_FIFO_SIZE,
897 G4X_MAX_WM,
898 G4X_MAX_WM,
899 2,
900 G4X_FIFO_LINE_SIZE,
901};
902static const struct intel_watermark_params g4x_cursor_wm_info = {
903 I965_CURSOR_FIFO,
904 I965_CURSOR_MAX_WM,
905 I965_CURSOR_DFT_WM,
906 2,
907 G4X_FIFO_LINE_SIZE,
908};
909static const struct intel_watermark_params valleyview_wm_info = {
910 VALLEYVIEW_FIFO_SIZE,
911 VALLEYVIEW_MAX_WM,
912 VALLEYVIEW_MAX_WM,
913 2,
914 G4X_FIFO_LINE_SIZE,
915};
916static const struct intel_watermark_params valleyview_cursor_wm_info = {
917 I965_CURSOR_FIFO,
918 VALLEYVIEW_CURSOR_MAX_WM,
919 I965_CURSOR_DFT_WM,
920 2,
921 G4X_FIFO_LINE_SIZE,
922};
923static const struct intel_watermark_params i965_cursor_wm_info = {
924 I965_CURSOR_FIFO,
925 I965_CURSOR_MAX_WM,
926 I965_CURSOR_DFT_WM,
927 2,
928 I915_FIFO_LINE_SIZE,
929};
930static const struct intel_watermark_params i945_wm_info = {
931 I945_FIFO_SIZE,
932 I915_MAX_WM,
933 1,
934 2,
935 I915_FIFO_LINE_SIZE
936};
937static const struct intel_watermark_params i915_wm_info = {
938 I915_FIFO_SIZE,
939 I915_MAX_WM,
940 1,
941 2,
942 I915_FIFO_LINE_SIZE
943};
feb56b93 944static const struct intel_watermark_params i830_wm_info = {
b445e3b0
ED		 945 I855GM_FIFO_SIZE,
946 I915_MAX_WM,
947 1,
948 2,
949 I830_FIFO_LINE_SIZE
950};
feb56b93 951static const struct intel_watermark_params i845_wm_info = {
b445e3b0
ED		 952 I830_FIFO_SIZE,
953 I915_MAX_WM,
954 1,
955 2,
956 I830_FIFO_LINE_SIZE
957};
958
b445e3b0
ED		 959/**
960 * intel_calculate_wm - calculate watermark level
961 * @clock_in_khz: pixel clock
962 * @wm: chip FIFO params
963 * @pixel_size: display pixel size
964 * @latency_ns: memory latency for the platform
965 *
966 * Calculate the watermark level (the level at which the display plane will
967 * start fetching from memory again). Each chip has a different display
968 * FIFO size and allocation, so the caller needs to figure that out and pass
969 * in the correct intel_watermark_params structure.
970 *
971 * As the pixel clock runs, the FIFO will be drained at a rate that depends
972 * on the pixel size. When it reaches the watermark level, it'll start
973 * fetching FIFO line sized based chunks from memory until the FIFO fills
974 * past the watermark point. If the FIFO drains completely, a FIFO underrun
975 * will occur, and a display engine hang could result.
976 */
977static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
978 const struct intel_watermark_params *wm,
979 int fifo_size,
980 int pixel_size,
981 unsigned long latency_ns)
982{
983 long entries_required, wm_size;
984
985 /*
986 * Note: we need to make sure we don't overflow for various clock &
987 * latency values.
988 * clocks go from a few thousand to several hundred thousand.
989 * latency is usually a few thousand
990 */
991 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
992 1000;
993 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
994
995 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
996
997 wm_size = fifo_size - (entries_required + wm->guard_size);
998
999 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
1000
1001 /* Don't promote wm_size to unsigned... */
1002 if (wm_size > (long)wm->max_wm)
1003 wm_size = wm->max_wm;
1004 if (wm_size <= 0)
1005 wm_size = wm->default_wm;
1006 return wm_size;
1007}
1008
1009static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1010{
1011 struct drm_crtc *crtc, *enabled = NULL;
1012
1013 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
3490ea5d 1014 if (intel_crtc_active(crtc)) {
b445e3b0
ED		 1015 if (enabled)
1016 return NULL;
1017 enabled = crtc;
1018 }
1019 }
1020
1021 return enabled;
1022}
1023
46ba614c 1024static void pineview_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1025{
46ba614c 1026 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1027 struct drm_i915_private *dev_priv = dev->dev_private;
1028 struct drm_crtc *crtc;
1029 const struct cxsr_latency *latency;
1030 u32 reg;
1031 unsigned long wm;
1032
1033 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1034 dev_priv->fsb_freq, dev_priv->mem_freq);
1035 if (!latency) {
1036 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1037 pineview_disable_cxsr(dev);
1038 return;
1039 }
1040
1041 crtc = single_enabled_crtc(dev);
1042 if (crtc) {
241bfc38 1043 const struct drm_display_mode *adjusted_mode;
f4510a27 1044 int pixel_size = crtc->primary->fb->bits_per_pixel / 8;
241bfc38
DL		 1045 int clock;
1046
1047 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1048 clock = adjusted_mode->crtc_clock;
b445e3b0
ED		 1049
1050 /* Display SR */
1051 wm = intel_calculate_wm(clock, &pineview_display_wm,
1052 pineview_display_wm.fifo_size,
1053 pixel_size, latency->display_sr);
1054 reg = I915_READ(DSPFW1);
1055 reg &= ~DSPFW_SR_MASK;
1056 reg |= wm << DSPFW_SR_SHIFT;
1057 I915_WRITE(DSPFW1, reg);
1058 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1059
1060 /* cursor SR */
1061 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1062 pineview_display_wm.fifo_size,
1063 pixel_size, latency->cursor_sr);
1064 reg = I915_READ(DSPFW3);
1065 reg &= ~DSPFW_CURSOR_SR_MASK;
1066 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1067 I915_WRITE(DSPFW3, reg);
1068
1069 /* Display HPLL off SR */
1070 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1071 pineview_display_hplloff_wm.fifo_size,
1072 pixel_size, latency->display_hpll_disable);
1073 reg = I915_READ(DSPFW3);
1074 reg &= ~DSPFW_HPLL_SR_MASK;
1075 reg |= wm & DSPFW_HPLL_SR_MASK;
1076 I915_WRITE(DSPFW3, reg);
1077
1078 /* cursor HPLL off SR */
1079 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1080 pineview_display_hplloff_wm.fifo_size,
1081 pixel_size, latency->cursor_hpll_disable);
1082 reg = I915_READ(DSPFW3);
1083 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1084 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1085 I915_WRITE(DSPFW3, reg);
1086 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1087
1088 /* activate cxsr */
1089 I915_WRITE(DSPFW3,
1090 I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1091 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1092 } else {
1093 pineview_disable_cxsr(dev);
1094 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1095 }
1096}
1097
1098static bool g4x_compute_wm0(struct drm_device *dev,
1099 int plane,
1100 const struct intel_watermark_params *display,
1101 int display_latency_ns,
1102 const struct intel_watermark_params *cursor,
1103 int cursor_latency_ns,
1104 int *plane_wm,
1105 int *cursor_wm)
1106{
1107 struct drm_crtc *crtc;
4fe8590a 1108 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1109 int htotal, hdisplay, clock, pixel_size;
1110 int line_time_us, line_count;
1111 int entries, tlb_miss;
1112
1113 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 1114 if (!intel_crtc_active(crtc)) {
b445e3b0
ED		 1115 *cursor_wm = cursor->guard_size;
1116 *plane_wm = display->guard_size;
1117 return false;
1118 }
1119
4fe8590a 1120 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1121 clock = adjusted_mode->crtc_clock;
fec8cba3 1122 htotal = adjusted_mode->crtc_htotal;
37327abd 1123 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
f4510a27 1124 pixel_size = crtc->primary->fb->bits_per_pixel / 8;
b445e3b0
ED		 1125
1126 /* Use the small buffer method to calculate plane watermark */
1127 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1128 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1129 if (tlb_miss > 0)
1130 entries += tlb_miss;
1131 entries = DIV_ROUND_UP(entries, display->cacheline_size);
1132 *plane_wm = entries + display->guard_size;
1133 if (*plane_wm > (int)display->max_wm)
1134 *plane_wm = display->max_wm;
1135
1136 /* Use the large buffer method to calculate cursor watermark */
922044c9 1137 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0 1138 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
7bb836dd 1139 entries = line_count * to_intel_crtc(crtc)->cursor_width * pixel_size;
b445e3b0
ED		 1140 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1141 if (tlb_miss > 0)
1142 entries += tlb_miss;
1143 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1144 *cursor_wm = entries + cursor->guard_size;
1145 if (*cursor_wm > (int)cursor->max_wm)
1146 *cursor_wm = (int)cursor->max_wm;
1147
1148 return true;
1149}
1150
1151/*
1152 * Check the wm result.
1153 *
1154 * If any calculated watermark values is larger than the maximum value that
1155 * can be programmed into the associated watermark register, that watermark
1156 * must be disabled.
1157 */
1158static bool g4x_check_srwm(struct drm_device *dev,
1159 int display_wm, int cursor_wm,
1160 const struct intel_watermark_params *display,
1161 const struct intel_watermark_params *cursor)
1162{
1163 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1164 display_wm, cursor_wm);
1165
1166 if (display_wm > display->max_wm) {
1167 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1168 display_wm, display->max_wm);
1169 return false;
1170 }
1171
1172 if (cursor_wm > cursor->max_wm) {
1173 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1174 cursor_wm, cursor->max_wm);
1175 return false;
1176 }
1177
1178 if (!(display_wm || cursor_wm)) {
1179 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1180 return false;
1181 }
1182
1183 return true;
1184}
1185
1186static bool g4x_compute_srwm(struct drm_device *dev,
1187 int plane,
1188 int latency_ns,
1189 const struct intel_watermark_params *display,
1190 const struct intel_watermark_params *cursor,
1191 int *display_wm, int *cursor_wm)
1192{
1193 struct drm_crtc *crtc;
4fe8590a 1194 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1195 int hdisplay, htotal, pixel_size, clock;
1196 unsigned long line_time_us;
1197 int line_count, line_size;
1198 int small, large;
1199 int entries;
1200
1201 if (!latency_ns) {
1202 *display_wm = *cursor_wm = 0;
1203 return false;
1204 }
1205
1206 crtc = intel_get_crtc_for_plane(dev, plane);
4fe8590a 1207 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1208 clock = adjusted_mode->crtc_clock;
fec8cba3 1209 htotal = adjusted_mode->crtc_htotal;
37327abd 1210 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
f4510a27 1211 pixel_size = crtc->primary->fb->bits_per_pixel / 8;
b445e3b0 1212
922044c9 1213 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1214 line_count = (latency_ns / line_time_us + 1000) / 1000;
1215 line_size = hdisplay * pixel_size;
1216
1217 /* Use the minimum of the small and large buffer method for primary */
1218 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1219 large = line_count * line_size;
1220
1221 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1222 *display_wm = entries + display->guard_size;
1223
1224 /* calculate the self-refresh watermark for display cursor */
7bb836dd 1225 entries = line_count * pixel_size * to_intel_crtc(crtc)->cursor_width;
b445e3b0
ED		 1226 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1227 *cursor_wm = entries + cursor->guard_size;
1228
1229 return g4x_check_srwm(dev,
1230 *display_wm, *cursor_wm,
1231 display, cursor);
1232}
1233
1234static bool vlv_compute_drain_latency(struct drm_device *dev,
1235 int plane,
1236 int *plane_prec_mult,
1237 int *plane_dl,
1238 int *cursor_prec_mult,
1239 int *cursor_dl)
1240{
1241 struct drm_crtc *crtc;
1242 int clock, pixel_size;
1243 int entries;
1244
1245 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 1246 if (!intel_crtc_active(crtc))
b445e3b0
ED		 1247 return false;
1248
241bfc38 1249 clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
f4510a27 1250 pixel_size = crtc->primary->fb->bits_per_pixel / 8; /* BPP */
b445e3b0
ED		 1251
1252 entries = (clock / 1000) * pixel_size;
1253 *plane_prec_mult = (entries > 256) ?
1254 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1255 *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1256 pixel_size);
1257
1258 entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */
1259 *cursor_prec_mult = (entries > 256) ?
1260 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1261 *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1262
1263 return true;
1264}
1265
1266/*
1267 * Update drain latency registers of memory arbiter
1268 *
1269 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1270 * to be programmed. Each plane has a drain latency multiplier and a drain
1271 * latency value.
1272 */
1273
1274static void vlv_update_drain_latency(struct drm_device *dev)
1275{
1276 struct drm_i915_private *dev_priv = dev->dev_private;
1277 int planea_prec, planea_dl, planeb_prec, planeb_dl;
1278 int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1279 int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1280 either 16 or 32 */
1281
1282 /* For plane A, Cursor A */
1283 if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1284 &cursor_prec_mult, &cursora_dl)) {
1285 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1286 DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1287 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1288 DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1289
1290 I915_WRITE(VLV_DDL1, cursora_prec |
1291 (cursora_dl << DDL_CURSORA_SHIFT) |
1292 planea_prec | planea_dl);
1293 }
1294
1295 /* For plane B, Cursor B */
1296 if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1297 &cursor_prec_mult, &cursorb_dl)) {
1298 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1299 DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1300 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1301 DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1302
1303 I915_WRITE(VLV_DDL2, cursorb_prec |
1304 (cursorb_dl << DDL_CURSORB_SHIFT) |
1305 planeb_prec | planeb_dl);
1306 }
1307}
1308
1309#define single_plane_enabled(mask) is_power_of_2(mask)
1310
46ba614c 1311static void valleyview_update_wm(struct drm_crtc *crtc)
b445e3b0 1312{
46ba614c 1313 struct drm_device *dev = crtc->dev;
b445e3b0
ED		 1314 static const int sr_latency_ns = 12000;
1315 struct drm_i915_private *dev_priv = dev->dev_private;
1316 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1317 int plane_sr, cursor_sr;
af6c4575 1318 int ignore_plane_sr, ignore_cursor_sr;
b445e3b0
ED		 1319 unsigned int enabled = 0;
1320
1321 vlv_update_drain_latency(dev);
1322
51cea1f4 1323 if (g4x_compute_wm0(dev, PIPE_A,
b445e3b0
ED		 1324 &valleyview_wm_info, latency_ns,
1325 &valleyview_cursor_wm_info, latency_ns,
1326 &planea_wm, &cursora_wm))
51cea1f4 1327 enabled |= 1 << PIPE_A;
b445e3b0 1328
51cea1f4 1329 if (g4x_compute_wm0(dev, PIPE_B,
b445e3b0
ED		 1330 &valleyview_wm_info, latency_ns,
1331 &valleyview_cursor_wm_info, latency_ns,
1332 &planeb_wm, &cursorb_wm))
51cea1f4 1333 enabled |= 1 << PIPE_B;
b445e3b0 1334
b445e3b0
ED		 1335 if (single_plane_enabled(enabled) &&
1336 g4x_compute_srwm(dev, ffs(enabled) - 1,
1337 sr_latency_ns,
1338 &valleyview_wm_info,
1339 &valleyview_cursor_wm_info,
af6c4575
CW		 1340 &plane_sr, &ignore_cursor_sr) &&
1341 g4x_compute_srwm(dev, ffs(enabled) - 1,
1342 2*sr_latency_ns,
1343 &valleyview_wm_info,
1344 &valleyview_cursor_wm_info,
52bd02d8 1345 &ignore_plane_sr, &cursor_sr)) {
b445e3b0 1346 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
52bd02d8 1347 } else {
b445e3b0
ED		 1348 I915_WRITE(FW_BLC_SELF_VLV,
1349 I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
52bd02d8
CW		 1350 plane_sr = cursor_sr = 0;
1351 }
b445e3b0
ED		 1352
1353 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1354 planea_wm, cursora_wm,
1355 planeb_wm, cursorb_wm,
1356 plane_sr, cursor_sr);
1357
1358 I915_WRITE(DSPFW1,
1359 (plane_sr << DSPFW_SR_SHIFT) |
1360 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1361 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1362 planea_wm);
1363 I915_WRITE(DSPFW2,
8c919b28 1364 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
b445e3b0
ED		 1365 (cursora_wm << DSPFW_CURSORA_SHIFT));
1366 I915_WRITE(DSPFW3,
8c919b28
CW		 1367 (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1368 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
b445e3b0
ED		 1369}
1370
46ba614c 1371static void g4x_update_wm(struct drm_crtc *crtc)
b445e3b0 1372{
46ba614c 1373 struct drm_device *dev = crtc->dev;
b445e3b0
ED		 1374 static const int sr_latency_ns = 12000;
1375 struct drm_i915_private *dev_priv = dev->dev_private;
1376 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1377 int plane_sr, cursor_sr;
1378 unsigned int enabled = 0;
1379
51cea1f4 1380 if (g4x_compute_wm0(dev, PIPE_A,
b445e3b0
ED		 1381 &g4x_wm_info, latency_ns,
1382 &g4x_cursor_wm_info, latency_ns,
1383 &planea_wm, &cursora_wm))
51cea1f4 1384 enabled |= 1 << PIPE_A;
b445e3b0 1385
51cea1f4 1386 if (g4x_compute_wm0(dev, PIPE_B,
b445e3b0
ED		 1387 &g4x_wm_info, latency_ns,
1388 &g4x_cursor_wm_info, latency_ns,
1389 &planeb_wm, &cursorb_wm))
51cea1f4 1390 enabled |= 1 << PIPE_B;
b445e3b0 1391
b445e3b0
ED		 1392 if (single_plane_enabled(enabled) &&
1393 g4x_compute_srwm(dev, ffs(enabled) - 1,
1394 sr_latency_ns,
1395 &g4x_wm_info,
1396 &g4x_cursor_wm_info,
52bd02d8 1397 &plane_sr, &cursor_sr)) {
b445e3b0 1398 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
52bd02d8 1399 } else {
b445e3b0
ED		 1400 I915_WRITE(FW_BLC_SELF,
1401 I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
52bd02d8
CW		 1402 plane_sr = cursor_sr = 0;
1403 }
b445e3b0
ED		 1404
1405 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1406 planea_wm, cursora_wm,
1407 planeb_wm, cursorb_wm,
1408 plane_sr, cursor_sr);
1409
1410 I915_WRITE(DSPFW1,
1411 (plane_sr << DSPFW_SR_SHIFT) |
1412 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1413 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1414 planea_wm);
1415 I915_WRITE(DSPFW2,
8c919b28 1416 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
b445e3b0
ED		 1417 (cursora_wm << DSPFW_CURSORA_SHIFT));
1418 /* HPLL off in SR has some issues on G4x... disable it */
1419 I915_WRITE(DSPFW3,
8c919b28 1420 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
b445e3b0
ED		 1421 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1422}
1423
46ba614c 1424static void i965_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1425{
46ba614c 1426 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1427 struct drm_i915_private *dev_priv = dev->dev_private;
1428 struct drm_crtc *crtc;
1429 int srwm = 1;
1430 int cursor_sr = 16;
1431
1432 /* Calc sr entries for one plane configs */
1433 crtc = single_enabled_crtc(dev);
1434 if (crtc) {
1435 /* self-refresh has much higher latency */
1436 static const int sr_latency_ns = 12000;
4fe8590a
VS		 1437 const struct drm_display_mode *adjusted_mode =
1438 &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1439 int clock = adjusted_mode->crtc_clock;
fec8cba3 1440 int htotal = adjusted_mode->crtc_htotal;
37327abd 1441 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
f4510a27 1442 int pixel_size = crtc->primary->fb->bits_per_pixel / 8;
b445e3b0
ED		 1443 unsigned long line_time_us;
1444 int entries;
1445
922044c9 1446 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1447
1448 /* Use ns/us then divide to preserve precision */
1449 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1450 pixel_size * hdisplay;
1451 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1452 srwm = I965_FIFO_SIZE - entries;
1453 if (srwm < 0)
1454 srwm = 1;
1455 srwm &= 0x1ff;
1456 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1457 entries, srwm);
1458
1459 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
7bb836dd 1460 pixel_size * to_intel_crtc(crtc)->cursor_width;
b445e3b0
ED		 1461 entries = DIV_ROUND_UP(entries,
1462 i965_cursor_wm_info.cacheline_size);
1463 cursor_sr = i965_cursor_wm_info.fifo_size -
1464 (entries + i965_cursor_wm_info.guard_size);
1465
1466 if (cursor_sr > i965_cursor_wm_info.max_wm)
1467 cursor_sr = i965_cursor_wm_info.max_wm;
1468
1469 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1470 "cursor %d\n", srwm, cursor_sr);
1471
1472 if (IS_CRESTLINE(dev))
1473 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1474 } else {
1475 /* Turn off self refresh if both pipes are enabled */
1476 if (IS_CRESTLINE(dev))
1477 I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1478 & ~FW_BLC_SELF_EN);
1479 }
1480
1481 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1482 srwm);
1483
1484 /* 965 has limitations... */
1485 I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1486 (8 << 16) | (8 << 8) | (8 << 0));
1487 I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1488 /* update cursor SR watermark */
1489 I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1490}
1491
46ba614c 1492static void i9xx_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1493{
46ba614c 1494 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1495 struct drm_i915_private *dev_priv = dev->dev_private;
1496 const struct intel_watermark_params *wm_info;
1497 uint32_t fwater_lo;
1498 uint32_t fwater_hi;
1499 int cwm, srwm = 1;
1500 int fifo_size;
1501 int planea_wm, planeb_wm;
1502 struct drm_crtc *crtc, *enabled = NULL;
1503
1504 if (IS_I945GM(dev))
1505 wm_info = &i945_wm_info;
1506 else if (!IS_GEN2(dev))
1507 wm_info = &i915_wm_info;
1508 else
feb56b93 1509 wm_info = &i830_wm_info;
b445e3b0
ED		 1510
1511 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1512 crtc = intel_get_crtc_for_plane(dev, 0);
3490ea5d 1513 if (intel_crtc_active(crtc)) {
241bfc38 1514 const struct drm_display_mode *adjusted_mode;
f4510a27 1515 int cpp = crtc->primary->fb->bits_per_pixel / 8;
b9e0bda3
CW		 1516 if (IS_GEN2(dev))
1517 cpp = 4;
1518
241bfc38
DL		 1519 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1520 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1521 wm_info, fifo_size, cpp,
b445e3b0
ED		 1522 latency_ns);
1523 enabled = crtc;
1524 } else
1525 planea_wm = fifo_size - wm_info->guard_size;
1526
1527 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1528 crtc = intel_get_crtc_for_plane(dev, 1);
3490ea5d 1529 if (intel_crtc_active(crtc)) {
241bfc38 1530 const struct drm_display_mode *adjusted_mode;
f4510a27 1531 int cpp = crtc->primary->fb->bits_per_pixel / 8;
b9e0bda3
CW		 1532 if (IS_GEN2(dev))
1533 cpp = 4;
1534
241bfc38
DL		 1535 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1536 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1537 wm_info, fifo_size, cpp,
b445e3b0
ED		 1538 latency_ns);
1539 if (enabled == NULL)
1540 enabled = crtc;
1541 else
1542 enabled = NULL;
1543 } else
1544 planeb_wm = fifo_size - wm_info->guard_size;
1545
1546 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1547
1548 /*
1549 * Overlay gets an aggressive default since video jitter is bad.
1550 */
1551 cwm = 2;
1552
1553 /* Play safe and disable self-refresh before adjusting watermarks. */
1554 if (IS_I945G(dev) || IS_I945GM(dev))
1555 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1556 else if (IS_I915GM(dev))
3f2dc5ac 1557 I915_WRITE(INSTPM, _MASKED_BIT_DISABLE(INSTPM_SELF_EN));
b445e3b0
ED		 1558
1559 /* Calc sr entries for one plane configs */
1560 if (HAS_FW_BLC(dev) && enabled) {
1561 /* self-refresh has much higher latency */
1562 static const int sr_latency_ns = 6000;
4fe8590a
VS		 1563 const struct drm_display_mode *adjusted_mode =
1564 &to_intel_crtc(enabled)->config.adjusted_mode;
241bfc38 1565 int clock = adjusted_mode->crtc_clock;
fec8cba3 1566 int htotal = adjusted_mode->crtc_htotal;
f727b490 1567 int hdisplay = to_intel_crtc(enabled)->config.pipe_src_w;
f4510a27 1568 int pixel_size = enabled->primary->fb->bits_per_pixel / 8;
b445e3b0
ED		 1569 unsigned long line_time_us;
1570 int entries;
1571
922044c9 1572 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1573
1574 /* Use ns/us then divide to preserve precision */
1575 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1576 pixel_size * hdisplay;
1577 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1578 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1579 srwm = wm_info->fifo_size - entries;
1580 if (srwm < 0)
1581 srwm = 1;
1582
1583 if (IS_I945G(dev) || IS_I945GM(dev))
1584 I915_WRITE(FW_BLC_SELF,
1585 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1586 else if (IS_I915GM(dev))
1587 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1588 }
1589
1590 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1591 planea_wm, planeb_wm, cwm, srwm);
1592
1593 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1594 fwater_hi = (cwm & 0x1f);
1595
1596 /* Set request length to 8 cachelines per fetch */
1597 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1598 fwater_hi = fwater_hi | (1 << 8);
1599
1600 I915_WRITE(FW_BLC, fwater_lo);
1601 I915_WRITE(FW_BLC2, fwater_hi);
1602
1603 if (HAS_FW_BLC(dev)) {
1604 if (enabled) {
1605 if (IS_I945G(dev) || IS_I945GM(dev))
1606 I915_WRITE(FW_BLC_SELF,
1607 FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1608 else if (IS_I915GM(dev))
3f2dc5ac 1609 I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_SELF_EN));
b445e3b0
ED		 1610 DRM_DEBUG_KMS("memory self refresh enabled\n");
1611 } else
1612 DRM_DEBUG_KMS("memory self refresh disabled\n");
1613 }
1614}
1615
feb56b93 1616static void i845_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1617{
46ba614c 1618 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1619 struct drm_i915_private *dev_priv = dev->dev_private;
1620 struct drm_crtc *crtc;
241bfc38 1621 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1622 uint32_t fwater_lo;
1623 int planea_wm;
1624
1625 crtc = single_enabled_crtc(dev);
1626 if (crtc == NULL)
1627 return;
1628
241bfc38
DL		 1629 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1630 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
feb56b93 1631 &i845_wm_info,
b445e3b0 1632 dev_priv->display.get_fifo_size(dev, 0),
b9e0bda3 1633 4, latency_ns);
b445e3b0
ED		 1634 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1635 fwater_lo |= (3<<8) | planea_wm;
1636
1637 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1638
1639 I915_WRITE(FW_BLC, fwater_lo);
1640}
1641
3658729a
VS		 1642static uint32_t ilk_pipe_pixel_rate(struct drm_device *dev,
1643 struct drm_crtc *crtc)
801bcfff
PZ		 1644{
1645 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
fd4daa9c 1646 uint32_t pixel_rate;
801bcfff 1647
241bfc38 1648 pixel_rate = intel_crtc->config.adjusted_mode.crtc_clock;
801bcfff
PZ		 1649
1650 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1651 * adjust the pixel_rate here. */
1652
fd4daa9c 1653 if (intel_crtc->config.pch_pfit.enabled) {
801bcfff 1654 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
fd4daa9c 1655 uint32_t pfit_size = intel_crtc->config.pch_pfit.size;
801bcfff 1656
37327abd
VS		 1657 pipe_w = intel_crtc->config.pipe_src_w;
1658 pipe_h = intel_crtc->config.pipe_src_h;
801bcfff
PZ		 1659 pfit_w = (pfit_size >> 16) & 0xFFFF;
1660 pfit_h = pfit_size & 0xFFFF;
1661 if (pipe_w < pfit_w)
1662 pipe_w = pfit_w;
1663 if (pipe_h < pfit_h)
1664 pipe_h = pfit_h;
1665
1666 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
1667 pfit_w * pfit_h);
1668 }
1669
1670 return pixel_rate;
1671}
1672
37126462 1673/* latency must be in 0.1us units. */
23297044 1674static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
801bcfff
PZ		 1675 uint32_t latency)
1676{
1677 uint64_t ret;
1678
3312ba65
VS		 1679 if (WARN(latency == 0, "Latency value missing\n"))
1680 return UINT_MAX;
1681
801bcfff
PZ		 1682 ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
1683 ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
1684
1685 return ret;
1686}
1687
37126462 1688/* latency must be in 0.1us units. */
23297044 1689static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
801bcfff
PZ		 1690 uint32_t horiz_pixels, uint8_t bytes_per_pixel,
1691 uint32_t latency)
1692{
1693 uint32_t ret;
1694
3312ba65
VS		 1695 if (WARN(latency == 0, "Latency value missing\n"))
1696 return UINT_MAX;
1697
801bcfff
PZ		 1698 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
1699 ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
1700 ret = DIV_ROUND_UP(ret, 64) + 2;
1701 return ret;
1702}
1703
23297044 1704static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
cca32e9a
PZ		 1705 uint8_t bytes_per_pixel)
1706{
1707 return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
1708}
1709
820c1980 1710struct ilk_pipe_wm_parameters {
801bcfff 1711 bool active;
801bcfff
PZ		 1712 uint32_t pipe_htotal;
1713 uint32_t pixel_rate;
c35426d2
VS		 1714 struct intel_plane_wm_parameters pri;
1715 struct intel_plane_wm_parameters spr;
1716 struct intel_plane_wm_parameters cur;
801bcfff
PZ		 1717};
1718
820c1980 1719struct ilk_wm_maximums {
cca32e9a
PZ		 1720 uint16_t pri;
1721 uint16_t spr;
1722 uint16_t cur;
1723 uint16_t fbc;
1724};
1725
240264f4
VS		 1726/* used in computing the new watermarks state */
1727struct intel_wm_config {
1728 unsigned int num_pipes_active;
1729 bool sprites_enabled;
1730 bool sprites_scaled;
240264f4
VS		 1731};
1732
37126462
VS		 1733/*
1734 * For both WM_PIPE and WM_LP.
1735 * mem_value must be in 0.1us units.
1736 */
820c1980 1737static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters *params,
cca32e9a
PZ		 1738 uint32_t mem_value,
1739 bool is_lp)
801bcfff 1740{
cca32e9a
PZ		 1741 uint32_t method1, method2;
1742
c35426d2 1743 if (!params->active || !params->pri.enabled)
801bcfff
PZ		 1744 return 0;
1745
23297044 1746 method1 = ilk_wm_method1(params->pixel_rate,
c35426d2 1747 params->pri.bytes_per_pixel,
cca32e9a
PZ		 1748 mem_value);
1749
1750 if (!is_lp)
1751 return method1;
1752
23297044 1753 method2 = ilk_wm_method2(params->pixel_rate,
cca32e9a 1754 params->pipe_htotal,
c35426d2
VS		 1755 params->pri.horiz_pixels,
1756 params->pri.bytes_per_pixel,
cca32e9a
PZ		 1757 mem_value);
1758
1759 return min(method1, method2);
801bcfff
PZ		 1760}
1761
37126462
VS		 1762/*
1763 * For both WM_PIPE and WM_LP.
1764 * mem_value must be in 0.1us units.
1765 */
820c1980 1766static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters *params,
801bcfff
PZ		 1767 uint32_t mem_value)
1768{
1769 uint32_t method1, method2;
1770
c35426d2 1771 if (!params->active || !params->spr.enabled)
801bcfff
PZ		 1772 return 0;
1773
23297044 1774 method1 = ilk_wm_method1(params->pixel_rate,
c35426d2 1775 params->spr.bytes_per_pixel,
801bcfff 1776 mem_value);
23297044 1777 method2 = ilk_wm_method2(params->pixel_rate,
801bcfff 1778 params->pipe_htotal,
c35426d2
VS		 1779 params->spr.horiz_pixels,
1780 params->spr.bytes_per_pixel,
801bcfff
PZ		 1781 mem_value);
1782 return min(method1, method2);
1783}
1784
37126462
VS		 1785/*
1786 * For both WM_PIPE and WM_LP.
1787 * mem_value must be in 0.1us units.
1788 */
820c1980 1789static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters *params,
801bcfff
PZ		 1790 uint32_t mem_value)
1791{
c35426d2 1792 if (!params->active || !params->cur.enabled)
801bcfff
PZ		 1793 return 0;
1794
23297044 1795 return ilk_wm_method2(params->pixel_rate,
801bcfff 1796 params->pipe_htotal,
c35426d2
VS		 1797 params->cur.horiz_pixels,
1798 params->cur.bytes_per_pixel,
801bcfff
PZ		 1799 mem_value);
1800}
1801
cca32e9a 1802/* Only for WM_LP. */
820c1980 1803static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters *params,
1fda9882 1804 uint32_t pri_val)
cca32e9a 1805{
c35426d2 1806 if (!params->active || !params->pri.enabled)
cca32e9a
PZ		 1807 return 0;
1808
23297044 1809 return ilk_wm_fbc(pri_val,
c35426d2
VS		 1810 params->pri.horiz_pixels,
1811 params->pri.bytes_per_pixel);
cca32e9a
PZ		 1812}
1813
158ae64f
VS		 1814static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
1815{
416f4727
VS		 1816 if (INTEL_INFO(dev)->gen >= 8)
1817 return 3072;
1818 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS		 1819 return 768;
1820 else
1821 return 512;
1822}
1823
1824/* Calculate the maximum primary/sprite plane watermark */
1825static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
1826 int level,
240264f4 1827 const struct intel_wm_config *config,
158ae64f
VS		 1828 enum intel_ddb_partitioning ddb_partitioning,
1829 bool is_sprite)
1830{
1831 unsigned int fifo_size = ilk_display_fifo_size(dev);
1832 unsigned int max;
1833
1834 /* if sprites aren't enabled, sprites get nothing */
240264f4 1835 if (is_sprite && !config->sprites_enabled)
158ae64f
VS		 1836 return 0;
1837
1838 /* HSW allows LP1+ watermarks even with multiple pipes */
240264f4 1839 if (level == 0 || config->num_pipes_active > 1) {
158ae64f
VS		 1840 fifo_size /= INTEL_INFO(dev)->num_pipes;
1841
1842 /*
1843 * For some reason the non self refresh
1844 * FIFO size is only half of the self
1845 * refresh FIFO size on ILK/SNB.
1846 */
1847 if (INTEL_INFO(dev)->gen <= 6)
1848 fifo_size /= 2;
1849 }
1850
240264f4 1851 if (config->sprites_enabled) {
158ae64f
VS		 1852 /* level 0 is always calculated with 1:1 split */
1853 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
1854 if (is_sprite)
1855 fifo_size *= 5;
1856 fifo_size /= 6;
1857 } else {
1858 fifo_size /= 2;
1859 }
1860 }
1861
1862 /* clamp to max that the registers can hold */
416f4727
VS		 1863 if (INTEL_INFO(dev)->gen >= 8)
1864 max = level == 0 ? 255 : 2047;
1865 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS		 1866 /* IVB/HSW primary/sprite plane watermarks */
1867 max = level == 0 ? 127 : 1023;
1868 else if (!is_sprite)
1869 /* ILK/SNB primary plane watermarks */
1870 max = level == 0 ? 127 : 511;
1871 else
1872 /* ILK/SNB sprite plane watermarks */
1873 max = level == 0 ? 63 : 255;
1874
1875 return min(fifo_size, max);
1876}
1877
1878/* Calculate the maximum cursor plane watermark */
1879static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
240264f4
VS		 1880 int level,
1881 const struct intel_wm_config *config)
158ae64f
VS		 1882{
1883 /* HSW LP1+ watermarks w/ multiple pipes */
240264f4 1884 if (level > 0 && config->num_pipes_active > 1)
158ae64f
VS		 1885 return 64;
1886
1887 /* otherwise just report max that registers can hold */
1888 if (INTEL_INFO(dev)->gen >= 7)
1889 return level == 0 ? 63 : 255;
1890 else
1891 return level == 0 ? 31 : 63;
1892}
1893
1894/* Calculate the maximum FBC watermark */
d34ff9c6 1895static unsigned int ilk_fbc_wm_max(const struct drm_device *dev)
158ae64f
VS		 1896{
1897 /* max that registers can hold */
416f4727
VS		 1898 if (INTEL_INFO(dev)->gen >= 8)
1899 return 31;
1900 else
1901 return 15;
158ae64f
VS		 1902}
1903
d34ff9c6 1904static void ilk_compute_wm_maximums(const struct drm_device *dev,
34982fe1
VS		 1905 int level,
1906 const struct intel_wm_config *config,
1907 enum intel_ddb_partitioning ddb_partitioning,
820c1980 1908 struct ilk_wm_maximums *max)
158ae64f 1909{
240264f4
VS		 1910 max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
1911 max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
1912 max->cur = ilk_cursor_wm_max(dev, level, config);
416f4727 1913 max->fbc = ilk_fbc_wm_max(dev);
158ae64f
VS		 1914}
1915
d9395655 1916static bool ilk_validate_wm_level(int level,
820c1980 1917 const struct ilk_wm_maximums *max,
d9395655 1918 struct intel_wm_level *result)
a9786a11
VS		 1919{
1920 bool ret;
1921
1922 /* already determined to be invalid? */
1923 if (!result->enable)
1924 return false;
1925
1926 result->enable = result->pri_val <= max->pri &&
1927 result->spr_val <= max->spr &&
1928 result->cur_val <= max->cur;
1929
1930 ret = result->enable;
1931
1932 /*
1933 * HACK until we can pre-compute everything,
1934 * and thus fail gracefully if LP0 watermarks
1935 * are exceeded...
1936 */
1937 if (level == 0 && !result->enable) {
1938 if (result->pri_val > max->pri)
1939 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1940 level, result->pri_val, max->pri);
1941 if (result->spr_val > max->spr)
1942 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1943 level, result->spr_val, max->spr);
1944 if (result->cur_val > max->cur)
1945 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
1946 level, result->cur_val, max->cur);
1947
1948 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
1949 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
1950 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
1951 result->enable = true;
1952 }
1953
a9786a11
VS		 1954 return ret;
1955}
1956
d34ff9c6 1957static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
6f5ddd17 1958 int level,
820c1980 1959 const struct ilk_pipe_wm_parameters *p,
1fd527cc 1960 struct intel_wm_level *result)
6f5ddd17
VS		 1961{
1962 uint16_t pri_latency = dev_priv->wm.pri_latency[level];
1963 uint16_t spr_latency = dev_priv->wm.spr_latency[level];
1964 uint16_t cur_latency = dev_priv->wm.cur_latency[level];
1965
1966 /* WM1+ latency values stored in 0.5us units */
1967 if (level > 0) {
1968 pri_latency *= 5;
1969 spr_latency *= 5;
1970 cur_latency *= 5;
1971 }
1972
1973 result->pri_val = ilk_compute_pri_wm(p, pri_latency, level);
1974 result->spr_val = ilk_compute_spr_wm(p, spr_latency);
1975 result->cur_val = ilk_compute_cur_wm(p, cur_latency);
1976 result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val);
1977 result->enable = true;
1978}
1979
801bcfff
PZ		 1980static uint32_t
1981hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
1f8eeabf
ED		 1982{
1983 struct drm_i915_private *dev_priv = dev->dev_private;
1011d8c4 1984 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
1011d8c4 1985 struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode;
85a02deb 1986 u32 linetime, ips_linetime;
1f8eeabf 1987
801bcfff
PZ		 1988 if (!intel_crtc_active(crtc))
1989 return 0;
1011d8c4 1990
1f8eeabf
ED		 1991 /* The WM are computed with base on how long it takes to fill a single
1992 * row at the given clock rate, multiplied by 8.
1993 * */
fec8cba3
JB		 1994 linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8,
1995 mode->crtc_clock);
1996 ips_linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8,
85a02deb 1997 intel_ddi_get_cdclk_freq(dev_priv));
1f8eeabf 1998
801bcfff
PZ		 1999 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
2000 PIPE_WM_LINETIME_TIME(linetime);
1f8eeabf
ED		 2001}
2002
12b134df
VS		 2003static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[5])
2004{
2005 struct drm_i915_private *dev_priv = dev->dev_private;
2006
a42a5719 2007 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
12b134df
VS		 2008 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2009
2010 wm[0] = (sskpd >> 56) & 0xFF;
2011 if (wm[0] == 0)
2012 wm[0] = sskpd & 0xF;
e5d5019e
VS		 2013 wm[1] = (sskpd >> 4) & 0xFF;
2014 wm[2] = (sskpd >> 12) & 0xFF;
2015 wm[3] = (sskpd >> 20) & 0x1FF;
2016 wm[4] = (sskpd >> 32) & 0x1FF;
63cf9a13
VS		 2017 } else if (INTEL_INFO(dev)->gen >= 6) {
2018 uint32_t sskpd = I915_READ(MCH_SSKPD);
2019
2020 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2021 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2022 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2023 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
3a88d0ac
VS		 2024 } else if (INTEL_INFO(dev)->gen >= 5) {
2025 uint32_t mltr = I915_READ(MLTR_ILK);
2026
2027 /* ILK primary LP0 latency is 700 ns */
2028 wm[0] = 7;
2029 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2030 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
12b134df
VS		 2031 }
2032}
2033
53615a5e
VS		 2034static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2035{
2036 /* ILK sprite LP0 latency is 1300 ns */
2037 if (INTEL_INFO(dev)->gen == 5)
2038 wm[0] = 13;
2039}
2040
2041static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2042{
2043 /* ILK cursor LP0 latency is 1300 ns */
2044 if (INTEL_INFO(dev)->gen == 5)
2045 wm[0] = 13;
2046
2047 /* WaDoubleCursorLP3Latency:ivb */
2048 if (IS_IVYBRIDGE(dev))
2049 wm[3] *= 2;
2050}
2051
ad0d6dc4 2052static int ilk_wm_max_level(const struct drm_device *dev)
26ec971e 2053{
26ec971e 2054 /* how many WM levels are we expecting */
a42a5719 2055 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ad0d6dc4 2056 return 4;
26ec971e 2057 else if (INTEL_INFO(dev)->gen >= 6)
ad0d6dc4 2058 return 3;
26ec971e 2059 else
ad0d6dc4
VS		 2060 return 2;
2061}
2062
2063static void intel_print_wm_latency(struct drm_device *dev,
2064 const char *name,
2065 const uint16_t wm[5])
2066{
2067 int level, max_level = ilk_wm_max_level(dev);
26ec971e
VS		 2068
2069 for (level = 0; level <= max_level; level++) {
2070 unsigned int latency = wm[level];
2071
2072 if (latency == 0) {
2073 DRM_ERROR("%s WM%d latency not provided\n",
2074 name, level);
2075 continue;
2076 }
2077
2078 /* WM1+ latency values in 0.5us units */
2079 if (level > 0)
2080 latency *= 5;
2081
2082 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2083 name, level, wm[level],
2084 latency / 10, latency % 10);
2085 }
2086}
2087
fa50ad61 2088static void ilk_setup_wm_latency(struct drm_device *dev)
53615a5e
VS		 2089{
2090 struct drm_i915_private *dev_priv = dev->dev_private;
2091
2092 intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2093
2094 memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2095 sizeof(dev_priv->wm.pri_latency));
2096 memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2097 sizeof(dev_priv->wm.pri_latency));
2098
2099 intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2100 intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
26ec971e
VS		 2101
2102 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2103 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2104 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
53615a5e
VS		 2105}
2106
820c1980
ID		 2107static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
2108 struct ilk_pipe_wm_parameters *p,
a485bfb8 2109 struct intel_wm_config *config)
1011d8c4 2110{
7c4a395f
VS		 2111 struct drm_device *dev = crtc->dev;
2112 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2113 enum pipe pipe = intel_crtc->pipe;
7c4a395f 2114 struct drm_plane *plane;
1011d8c4 2115
7c4a395f
VS		 2116 p->active = intel_crtc_active(crtc);
2117 if (p->active) {
576b259e 2118 p->pipe_htotal = intel_crtc->config.adjusted_mode.crtc_htotal;
3658729a 2119 p->pixel_rate = ilk_pipe_pixel_rate(dev, crtc);
f4510a27 2120 p->pri.bytes_per_pixel = crtc->primary->fb->bits_per_pixel / 8;
c35426d2 2121 p->cur.bytes_per_pixel = 4;
37327abd 2122 p->pri.horiz_pixels = intel_crtc->config.pipe_src_w;
7bb836dd 2123 p->cur.horiz_pixels = intel_crtc->cursor_width;
c35426d2
VS		 2124 /* TODO: for now, assume primary and cursor planes are always enabled. */
2125 p->pri.enabled = true;
2126 p->cur.enabled = true;
801bcfff
PZ		 2127 }
2128
7c4a395f 2129 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
a485bfb8 2130 config->num_pipes_active += intel_crtc_active(crtc);
7c4a395f 2131
af2b653b 2132 drm_for_each_legacy_plane(plane, &dev->mode_config.plane_list) {
801bcfff 2133 struct intel_plane *intel_plane = to_intel_plane(plane);
801bcfff 2134
7c4a395f
VS		 2135 if (intel_plane->pipe == pipe)
2136 p->spr = intel_plane->wm;
cca32e9a 2137
a485bfb8
VS		 2138 config->sprites_enabled |= intel_plane->wm.enabled;
2139 config->sprites_scaled |= intel_plane->wm.scaled;
cca32e9a 2140 }
801bcfff
PZ		 2141}
2142
0b2ae6d7
VS		 2143/* Compute new watermarks for the pipe */
2144static bool intel_compute_pipe_wm(struct drm_crtc *crtc,
820c1980 2145 const struct ilk_pipe_wm_parameters *params,
0b2ae6d7
VS		 2146 struct intel_pipe_wm *pipe_wm)
2147{
2148 struct drm_device *dev = crtc->dev;
d34ff9c6 2149 const struct drm_i915_private *dev_priv = dev->dev_private;
0b2ae6d7
VS		 2150 int level, max_level = ilk_wm_max_level(dev);
2151 /* LP0 watermark maximums depend on this pipe alone */
2152 struct intel_wm_config config = {
2153 .num_pipes_active = 1,
2154 .sprites_enabled = params->spr.enabled,
2155 .sprites_scaled = params->spr.scaled,
2156 };
820c1980 2157 struct ilk_wm_maximums max;
0b2ae6d7 2158
0b2ae6d7 2159 /* LP0 watermarks always use 1/2 DDB partitioning */
34982fe1 2160 ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
0b2ae6d7 2161
7b39a0b7
VS		 2162 /* ILK/SNB: LP2+ watermarks only w/o sprites */
2163 if (INTEL_INFO(dev)->gen <= 6 && params->spr.enabled)
2164 max_level = 1;
2165
2166 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
2167 if (params->spr.scaled)
2168 max_level = 0;
2169
0b2ae6d7
VS		 2170 for (level = 0; level <= max_level; level++)
2171 ilk_compute_wm_level(dev_priv, level, params,
2172 &pipe_wm->wm[level]);
2173
a42a5719 2174 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 2175 pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
0b2ae6d7
VS		 2176
2177 /* At least LP0 must be valid */
d9395655 2178 return ilk_validate_wm_level(0, &max, &pipe_wm->wm[0]);
0b2ae6d7
VS		 2179}
2180
2181/*
2182 * Merge the watermarks from all active pipes for a specific level.
2183 */
2184static void ilk_merge_wm_level(struct drm_device *dev,
2185 int level,
2186 struct intel_wm_level *ret_wm)
2187{
2188 const struct intel_crtc *intel_crtc;
2189
2190 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2191 const struct intel_wm_level *wm =
2192 &intel_crtc->wm.active.wm[level];
2193
2194 if (!wm->enable)
2195 return;
2196
2197 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2198 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2199 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2200 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2201 }
2202
2203 ret_wm->enable = true;
2204}
2205
2206/*
2207 * Merge all low power watermarks for all active pipes.
2208 */
2209static void ilk_wm_merge(struct drm_device *dev,
0ba22e26 2210 const struct intel_wm_config *config,
820c1980 2211 const struct ilk_wm_maximums *max,
0b2ae6d7
VS		 2212 struct intel_pipe_wm *merged)
2213{
2214 int level, max_level = ilk_wm_max_level(dev);
2215
0ba22e26
VS		 2216 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2217 if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) &&
2218 config->num_pipes_active > 1)
2219 return;
2220
6c8b6c28
VS		 2221 /* ILK: FBC WM must be disabled always */
2222 merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6;
0b2ae6d7
VS		 2223
2224 /* merge each WM1+ level */
2225 for (level = 1; level <= max_level; level++) {
2226 struct intel_wm_level *wm = &merged->wm[level];
2227
2228 ilk_merge_wm_level(dev, level, wm);
2229
d9395655 2230 if (!ilk_validate_wm_level(level, max, wm))
0b2ae6d7
VS		 2231 break;
2232
2233 /*
2234 * The spec says it is preferred to disable
2235 * FBC WMs instead of disabling a WM level.
2236 */
2237 if (wm->fbc_val > max->fbc) {
2238 merged->fbc_wm_enabled = false;
2239 wm->fbc_val = 0;
2240 }
2241 }
6c8b6c28
VS		 2242
2243 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2244 /*
2245 * FIXME this is racy. FBC might get enabled later.
2246 * What we should check here is whether FBC can be
2247 * enabled sometime later.
2248 */
2249 if (IS_GEN5(dev) && !merged->fbc_wm_enabled && intel_fbc_enabled(dev)) {
2250 for (level = 2; level <= max_level; level++) {
2251 struct intel_wm_level *wm = &merged->wm[level];
2252
2253 wm->enable = false;
2254 }
2255 }
0b2ae6d7
VS		 2256}
2257
b380ca3c
VS		 2258static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2259{
2260 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2261 return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2262}
2263
a68d68ee
VS		 2264/* The value we need to program into the WM_LPx latency field */
2265static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level)
2266{
2267 struct drm_i915_private *dev_priv = dev->dev_private;
2268
a42a5719 2269 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
a68d68ee
VS		 2270 return 2 * level;
2271 else
2272 return dev_priv->wm.pri_latency[level];
2273}
2274
820c1980 2275static void ilk_compute_wm_results(struct drm_device *dev,
0362c781 2276 const struct intel_pipe_wm *merged,
609cedef 2277 enum intel_ddb_partitioning partitioning,
820c1980 2278 struct ilk_wm_values *results)
801bcfff 2279{
0b2ae6d7
VS		 2280 struct intel_crtc *intel_crtc;
2281 int level, wm_lp;
cca32e9a 2282
0362c781 2283 results->enable_fbc_wm = merged->fbc_wm_enabled;
609cedef 2284 results->partitioning = partitioning;
cca32e9a 2285
0b2ae6d7 2286 /* LP1+ register values */
cca32e9a 2287 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
1fd527cc 2288 const struct intel_wm_level *r;
801bcfff 2289
b380ca3c 2290 level = ilk_wm_lp_to_level(wm_lp, merged);
0b2ae6d7 2291
0362c781 2292 r = &merged->wm[level];
0b2ae6d7 2293 if (!r->enable)
cca32e9a
PZ		 2294 break;
2295
416f4727 2296 results->wm_lp[wm_lp - 1] = WM3_LP_EN |
a68d68ee 2297 (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) |
416f4727
VS		 2298 (r->pri_val << WM1_LP_SR_SHIFT) |
2299 r->cur_val;
2300
2301 if (INTEL_INFO(dev)->gen >= 8)
2302 results->wm_lp[wm_lp - 1] |=
2303 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
2304 else
2305 results->wm_lp[wm_lp - 1] |=
2306 r->fbc_val << WM1_LP_FBC_SHIFT;
2307
6cef2b8a
VS		 2308 if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) {
2309 WARN_ON(wm_lp != 1);
2310 results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
2311 } else
2312 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
cca32e9a 2313 }
801bcfff 2314
0b2ae6d7
VS		 2315 /* LP0 register values */
2316 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2317 enum pipe pipe = intel_crtc->pipe;
2318 const struct intel_wm_level *r =
2319 &intel_crtc->wm.active.wm[0];
2320
2321 if (WARN_ON(!r->enable))
2322 continue;
2323
2324 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
1011d8c4 2325
0b2ae6d7
VS		 2326 results->wm_pipe[pipe] =
2327 (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2328 (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2329 r->cur_val;
801bcfff
PZ		 2330 }
2331}
2332
861f3389
PZ		 2333/* Find the result with the highest level enabled. Check for enable_fbc_wm in
2334 * case both are at the same level. Prefer r1 in case they're the same. */
820c1980 2335static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev,
198a1e9b
VS		 2336 struct intel_pipe_wm *r1,
2337 struct intel_pipe_wm *r2)
861f3389 2338{
198a1e9b
VS		 2339 int level, max_level = ilk_wm_max_level(dev);
2340 int level1 = 0, level2 = 0;
861f3389 2341
198a1e9b
VS		 2342 for (level = 1; level <= max_level; level++) {
2343 if (r1->wm[level].enable)
2344 level1 = level;
2345 if (r2->wm[level].enable)
2346 level2 = level;
861f3389
PZ		 2347 }
2348
198a1e9b
VS		 2349 if (level1 == level2) {
2350 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
861f3389
PZ		 2351 return r2;
2352 else
2353 return r1;
198a1e9b 2354 } else if (level1 > level2) {
861f3389
PZ		 2355 return r1;
2356 } else {
2357 return r2;
2358 }
2359}
2360
49a687c4
VS		 2361/* dirty bits used to track which watermarks need changes */
2362#define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2363#define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2364#define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2365#define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2366#define WM_DIRTY_FBC (1 << 24)
2367#define WM_DIRTY_DDB (1 << 25)
2368
2369static unsigned int ilk_compute_wm_dirty(struct drm_device *dev,
820c1980
ID		 2370 const struct ilk_wm_values *old,
2371 const struct ilk_wm_values *new)
49a687c4
VS		 2372{
2373 unsigned int dirty = 0;
2374 enum pipe pipe;
2375 int wm_lp;
2376
2377 for_each_pipe(pipe) {
2378 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
2379 dirty |= WM_DIRTY_LINETIME(pipe);
2380 /* Must disable LP1+ watermarks too */
2381 dirty |= WM_DIRTY_LP_ALL;
2382 }
2383
2384 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
2385 dirty |= WM_DIRTY_PIPE(pipe);
2386 /* Must disable LP1+ watermarks too */
2387 dirty |= WM_DIRTY_LP_ALL;
2388 }
2389 }
2390
2391 if (old->enable_fbc_wm != new->enable_fbc_wm) {
2392 dirty |= WM_DIRTY_FBC;
2393 /* Must disable LP1+ watermarks too */
2394 dirty |= WM_DIRTY_LP_ALL;
2395 }
2396
2397 if (old->partitioning != new->partitioning) {
2398 dirty |= WM_DIRTY_DDB;
2399 /* Must disable LP1+ watermarks too */
2400 dirty |= WM_DIRTY_LP_ALL;
2401 }
2402
2403 /* LP1+ watermarks already deemed dirty, no need to continue */
2404 if (dirty & WM_DIRTY_LP_ALL)
2405 return dirty;
2406
2407 /* Find the lowest numbered LP1+ watermark in need of an update... */
2408 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2409 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
2410 old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
2411 break;
2412 }
2413
2414 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2415 for (; wm_lp <= 3; wm_lp++)
2416 dirty |= WM_DIRTY_LP(wm_lp);
2417
2418 return dirty;
2419}
2420
8553c18e
VS		 2421static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
2422 unsigned int dirty)
801bcfff 2423{
820c1980 2424 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e 2425 bool changed = false;
801bcfff 2426
facd619b
VS		 2427 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
2428 previous->wm_lp[2] &= ~WM1_LP_SR_EN;
2429 I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
8553c18e 2430 changed = true;
facd619b
VS		 2431 }
2432 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
2433 previous->wm_lp[1] &= ~WM1_LP_SR_EN;
2434 I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
8553c18e 2435 changed = true;
facd619b
VS		 2436 }
2437 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
2438 previous->wm_lp[0] &= ~WM1_LP_SR_EN;
2439 I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
8553c18e 2440 changed = true;
facd619b 2441 }
801bcfff 2442
facd619b
VS		 2443 /*
2444 * Don't touch WM1S_LP_EN here.
2445 * Doing so could cause underruns.
2446 */
6cef2b8a 2447
8553c18e
VS		 2448 return changed;
2449}
2450
2451/*
2452 * The spec says we shouldn't write when we don't need, because every write
2453 * causes WMs to be re-evaluated, expending some power.
2454 */
820c1980
ID		 2455static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
2456 struct ilk_wm_values *results)
8553c18e
VS		 2457{
2458 struct drm_device *dev = dev_priv->dev;
820c1980 2459 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e
VS		 2460 unsigned int dirty;
2461 uint32_t val;
2462
2463 dirty = ilk_compute_wm_dirty(dev, previous, results);
2464 if (!dirty)
2465 return;
2466
2467 _ilk_disable_lp_wm(dev_priv, dirty);
2468
49a687c4 2469 if (dirty & WM_DIRTY_PIPE(PIPE_A))
801bcfff 2470 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
49a687c4 2471 if (dirty & WM_DIRTY_PIPE(PIPE_B))
801bcfff 2472 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
49a687c4 2473 if (dirty & WM_DIRTY_PIPE(PIPE_C))
801bcfff
PZ		 2474 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2475
49a687c4 2476 if (dirty & WM_DIRTY_LINETIME(PIPE_A))
801bcfff 2477 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
49a687c4 2478 if (dirty & WM_DIRTY_LINETIME(PIPE_B))
801bcfff 2479 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
49a687c4 2480 if (dirty & WM_DIRTY_LINETIME(PIPE_C))
801bcfff
PZ		 2481 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2482
49a687c4 2483 if (dirty & WM_DIRTY_DDB) {
a42a5719 2484 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
ac9545fd
VS		 2485 val = I915_READ(WM_MISC);
2486 if (results->partitioning == INTEL_DDB_PART_1_2)
2487 val &= ~WM_MISC_DATA_PARTITION_5_6;
2488 else
2489 val |= WM_MISC_DATA_PARTITION_5_6;
2490 I915_WRITE(WM_MISC, val);
2491 } else {
2492 val = I915_READ(DISP_ARB_CTL2);
2493 if (results->partitioning == INTEL_DDB_PART_1_2)
2494 val &= ~DISP_DATA_PARTITION_5_6;
2495 else
2496 val |= DISP_DATA_PARTITION_5_6;
2497 I915_WRITE(DISP_ARB_CTL2, val);
2498 }
1011d8c4
PZ		 2499 }
2500
49a687c4 2501 if (dirty & WM_DIRTY_FBC) {
cca32e9a
PZ		 2502 val = I915_READ(DISP_ARB_CTL);
2503 if (results->enable_fbc_wm)
2504 val &= ~DISP_FBC_WM_DIS;
2505 else
2506 val |= DISP_FBC_WM_DIS;
2507 I915_WRITE(DISP_ARB_CTL, val);
2508 }
2509
954911eb
ID		 2510 if (dirty & WM_DIRTY_LP(1) &&
2511 previous->wm_lp_spr[0] != results->wm_lp_spr[0])
2512 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2513
2514 if (INTEL_INFO(dev)->gen >= 7) {
6cef2b8a
VS		 2515 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
2516 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2517 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
2518 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2519 }
801bcfff 2520
facd619b 2521 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
801bcfff 2522 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
facd619b 2523 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
801bcfff 2524 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
facd619b 2525 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
801bcfff 2526 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
609cedef
VS		 2527
2528 dev_priv->wm.hw = *results;
801bcfff
PZ		 2529}
2530
8553c18e
VS		 2531static bool ilk_disable_lp_wm(struct drm_device *dev)
2532{
2533 struct drm_i915_private *dev_priv = dev->dev_private;
2534
2535 return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
2536}
2537
820c1980 2538static void ilk_update_wm(struct drm_crtc *crtc)
801bcfff 2539{
7c4a395f 2540 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
46ba614c 2541 struct drm_device *dev = crtc->dev;
801bcfff 2542 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980
ID		 2543 struct ilk_wm_maximums max;
2544 struct ilk_pipe_wm_parameters params = {};
2545 struct ilk_wm_values results = {};
77c122bc 2546 enum intel_ddb_partitioning partitioning;
7c4a395f 2547 struct intel_pipe_wm pipe_wm = {};
198a1e9b 2548 struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
a485bfb8 2549 struct intel_wm_config config = {};
7c4a395f 2550
820c1980 2551 ilk_compute_wm_parameters(crtc, &params, &config);
7c4a395f
VS		 2552
2553 intel_compute_pipe_wm(crtc, &params, &pipe_wm);
2554
2555 if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
2556 return;
861f3389 2557
7c4a395f 2558 intel_crtc->wm.active = pipe_wm;
861f3389 2559
34982fe1 2560 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
0ba22e26 2561 ilk_wm_merge(dev, &config, &max, &lp_wm_1_2);
a485bfb8
VS		 2562
2563 /* 5/6 split only in single pipe config on IVB+ */
ec98c8d1
VS		 2564 if (INTEL_INFO(dev)->gen >= 7 &&
2565 config.num_pipes_active == 1 && config.sprites_enabled) {
34982fe1 2566 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
0ba22e26 2567 ilk_wm_merge(dev, &config, &max, &lp_wm_5_6);
0362c781 2568
820c1980 2569 best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
861f3389 2570 } else {
198a1e9b 2571 best_lp_wm = &lp_wm_1_2;
861f3389
PZ		 2572 }
2573
198a1e9b 2574 partitioning = (best_lp_wm == &lp_wm_1_2) ?
77c122bc 2575 INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
801bcfff 2576
820c1980 2577 ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results);
609cedef 2578
820c1980 2579 ilk_write_wm_values(dev_priv, &results);
1011d8c4
PZ		 2580}
2581
820c1980 2582static void ilk_update_sprite_wm(struct drm_plane *plane,
adf3d35e 2583 struct drm_crtc *crtc,
526682e9 2584 uint32_t sprite_width, int pixel_size,
bdd57d03 2585 bool enabled, bool scaled)
526682e9 2586{
8553c18e 2587 struct drm_device *dev = plane->dev;
adf3d35e 2588 struct intel_plane *intel_plane = to_intel_plane(plane);
526682e9 2589
adf3d35e
VS		 2590 intel_plane->wm.enabled = enabled;
2591 intel_plane->wm.scaled = scaled;
2592 intel_plane->wm.horiz_pixels = sprite_width;
2593 intel_plane->wm.bytes_per_pixel = pixel_size;
526682e9 2594
8553c18e
VS		 2595 /*
2596 * IVB workaround: must disable low power watermarks for at least
2597 * one frame before enabling scaling. LP watermarks can be re-enabled
2598 * when scaling is disabled.
2599 *
2600 * WaCxSRDisabledForSpriteScaling:ivb
2601 */
2602 if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev))
2603 intel_wait_for_vblank(dev, intel_plane->pipe);
2604
820c1980 2605 ilk_update_wm(crtc);
526682e9
PZ		 2606}
2607
243e6a44
VS		 2608static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
2609{
2610 struct drm_device *dev = crtc->dev;
2611 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 2612 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS		 2613 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2614 struct intel_pipe_wm *active = &intel_crtc->wm.active;
2615 enum pipe pipe = intel_crtc->pipe;
2616 static const unsigned int wm0_pipe_reg[] = {
2617 [PIPE_A] = WM0_PIPEA_ILK,
2618 [PIPE_B] = WM0_PIPEB_ILK,
2619 [PIPE_C] = WM0_PIPEC_IVB,
2620 };
2621
2622 hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
a42a5719 2623 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 2624 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
243e6a44
VS		 2625
2626 if (intel_crtc_active(crtc)) {
2627 u32 tmp = hw->wm_pipe[pipe];
2628
2629 /*
2630 * For active pipes LP0 watermark is marked as
2631 * enabled, and LP1+ watermaks as disabled since
2632 * we can't really reverse compute them in case
2633 * multiple pipes are active.
2634 */
2635 active->wm[0].enable = true;
2636 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
2637 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
2638 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
2639 active->linetime = hw->wm_linetime[pipe];
2640 } else {
2641 int level, max_level = ilk_wm_max_level(dev);
2642
2643 /*
2644 * For inactive pipes, all watermark levels
2645 * should be marked as enabled but zeroed,
2646 * which is what we'd compute them to.
2647 */
2648 for (level = 0; level <= max_level; level++)
2649 active->wm[level].enable = true;
2650 }
2651}
2652
2653void ilk_wm_get_hw_state(struct drm_device *dev)
2654{
2655 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 2656 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS		 2657 struct drm_crtc *crtc;
2658
2659 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
2660 ilk_pipe_wm_get_hw_state(crtc);
2661
2662 hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
2663 hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
2664 hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
2665
2666 hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
2667 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
2668 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
2669
a42a5719 2670 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ac9545fd
VS		 2671 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
2672 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
2673 else if (IS_IVYBRIDGE(dev))
2674 hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
2675 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
243e6a44
VS		 2676
2677 hw->enable_fbc_wm =
2678 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
2679}
2680
b445e3b0
ED		 2681/**
2682 * intel_update_watermarks - update FIFO watermark values based on current modes
2683 *
2684 * Calculate watermark values for the various WM regs based on current mode
2685 * and plane configuration.
2686 *
2687 * There are several cases to deal with here:
2688 * - normal (i.e. non-self-refresh)
2689 * - self-refresh (SR) mode
2690 * - lines are large relative to FIFO size (buffer can hold up to 2)
2691 * - lines are small relative to FIFO size (buffer can hold more than 2
2692 * lines), so need to account for TLB latency
2693 *
2694 * The normal calculation is:
2695 * watermark = dotclock * bytes per pixel * latency
2696 * where latency is platform & configuration dependent (we assume pessimal
2697 * values here).
2698 *
2699 * The SR calculation is:
2700 * watermark = (trunc(latency/line time)+1) * surface width *
2701 * bytes per pixel
2702 * where
2703 * line time = htotal / dotclock
2704 * surface width = hdisplay for normal plane and 64 for cursor
2705 * and latency is assumed to be high, as above.
2706 *
2707 * The final value programmed to the register should always be rounded up,
2708 * and include an extra 2 entries to account for clock crossings.
2709 *
2710 * We don't use the sprite, so we can ignore that. And on Crestline we have
2711 * to set the non-SR watermarks to 8.
2712 */
46ba614c 2713void intel_update_watermarks(struct drm_crtc *crtc)
b445e3b0 2714{
46ba614c 2715 struct drm_i915_private *dev_priv = crtc->dev->dev_private;
b445e3b0
ED		 2716
2717 if (dev_priv->display.update_wm)
46ba614c 2718 dev_priv->display.update_wm(crtc);
b445e3b0
ED		 2719}
2720
adf3d35e
VS		 2721void intel_update_sprite_watermarks(struct drm_plane *plane,
2722 struct drm_crtc *crtc,
4c4ff43a 2723 uint32_t sprite_width, int pixel_size,
39db4a4d 2724 bool enabled, bool scaled)
b445e3b0 2725{
adf3d35e 2726 struct drm_i915_private *dev_priv = plane->dev->dev_private;
b445e3b0
ED		 2727
2728 if (dev_priv->display.update_sprite_wm)
adf3d35e 2729 dev_priv->display.update_sprite_wm(plane, crtc, sprite_width,
39db4a4d 2730 pixel_size, enabled, scaled);
b445e3b0
ED		 2731}
2732
2b4e57bd
ED		 2733static struct drm_i915_gem_object *
2734intel_alloc_context_page(struct drm_device *dev)
2735{
2736 struct drm_i915_gem_object *ctx;
2737 int ret;
2738
2739 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2740
2741 ctx = i915_gem_alloc_object(dev, 4096);
2742 if (!ctx) {
2743 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2744 return NULL;
2745 }
2746
c69766f2 2747 ret = i915_gem_obj_ggtt_pin(ctx, 4096, 0);
2b4e57bd
ED		 2748 if (ret) {
2749 DRM_ERROR("failed to pin power context: %d\n", ret);
2750 goto err_unref;
2751 }
2752
2753 ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
2754 if (ret) {
2755 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
2756 goto err_unpin;
2757 }
2758
2759 return ctx;
2760
2761err_unpin:
d7f46fc4 2762 i915_gem_object_ggtt_unpin(ctx);
2b4e57bd
ED		 2763err_unref:
2764 drm_gem_object_unreference(&ctx->base);
2b4e57bd
ED		 2765 return NULL;
2766}
2767
9270388e
DV		 2768/**
2769 * Lock protecting IPS related data structures
9270388e
DV		 2770 */
2771DEFINE_SPINLOCK(mchdev_lock);
2772
2773/* Global for IPS driver to get at the current i915 device. Protected by
2774 * mchdev_lock. */
2775static struct drm_i915_private *i915_mch_dev;
2776
2b4e57bd
ED		 2777bool ironlake_set_drps(struct drm_device *dev, u8 val)
2778{
2779 struct drm_i915_private *dev_priv = dev->dev_private;
2780 u16 rgvswctl;
2781
9270388e
DV		 2782 assert_spin_locked(&mchdev_lock);
2783
2b4e57bd
ED		 2784 rgvswctl = I915_READ16(MEMSWCTL);
2785 if (rgvswctl & MEMCTL_CMD_STS) {
2786 DRM_DEBUG("gpu busy, RCS change rejected\n");
2787 return false; /* still busy with another command */
2788 }
2789
2790 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
2791 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
2792 I915_WRITE16(MEMSWCTL, rgvswctl);
2793 POSTING_READ16(MEMSWCTL);
2794
2795 rgvswctl |= MEMCTL_CMD_STS;
2796 I915_WRITE16(MEMSWCTL, rgvswctl);
2797
2798 return true;
2799}
2800
8090c6b9 2801static void ironlake_enable_drps(struct drm_device *dev)
2b4e57bd
ED		 2802{
2803 struct drm_i915_private *dev_priv = dev->dev_private;
2804 u32 rgvmodectl = I915_READ(MEMMODECTL);
2805 u8 fmax, fmin, fstart, vstart;
2806
9270388e
DV		 2807 spin_lock_irq(&mchdev_lock);
2808
2b4e57bd
ED		 2809 /* Enable temp reporting */
2810 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
2811 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
2812
2813 /* 100ms RC evaluation intervals */
2814 I915_WRITE(RCUPEI, 100000);
2815 I915_WRITE(RCDNEI, 100000);
2816
2817 /* Set max/min thresholds to 90ms and 80ms respectively */
2818 I915_WRITE(RCBMAXAVG, 90000);
2819 I915_WRITE(RCBMINAVG, 80000);
2820
2821 I915_WRITE(MEMIHYST, 1);
2822
2823 /* Set up min, max, and cur for interrupt handling */
2824 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
2825 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
2826 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
2827 MEMMODE_FSTART_SHIFT;
2828
2829 vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
2830 PXVFREQ_PX_SHIFT;
2831
20e4d407
DV		 2832 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
2833 dev_priv->ips.fstart = fstart;
2b4e57bd 2834
20e4d407
DV		 2835 dev_priv->ips.max_delay = fstart;
2836 dev_priv->ips.min_delay = fmin;
2837 dev_priv->ips.cur_delay = fstart;
2b4e57bd
ED		 2838
2839 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2840 fmax, fmin, fstart);
2841
2842 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
2843
2844 /*
2845 * Interrupts will be enabled in ironlake_irq_postinstall
2846 */
2847
2848 I915_WRITE(VIDSTART, vstart);
2849 POSTING_READ(VIDSTART);
2850
2851 rgvmodectl |= MEMMODE_SWMODE_EN;
2852 I915_WRITE(MEMMODECTL, rgvmodectl);
2853
9270388e 2854 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2b4e57bd 2855 DRM_ERROR("stuck trying to change perf mode\n");
9270388e 2856 mdelay(1);
2b4e57bd
ED		 2857
2858 ironlake_set_drps(dev, fstart);
2859
20e4d407 2860 dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
2b4e57bd 2861 I915_READ(0x112e0);
20e4d407
DV		 2862 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
2863 dev_priv->ips.last_count2 = I915_READ(0x112f4);
2864 getrawmonotonic(&dev_priv->ips.last_time2);
9270388e
DV		 2865
2866 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED		 2867}
2868
8090c6b9 2869static void ironlake_disable_drps(struct drm_device *dev)
2b4e57bd
ED		 2870{
2871 struct drm_i915_private *dev_priv = dev->dev_private;
9270388e
DV		 2872 u16 rgvswctl;
2873
2874 spin_lock_irq(&mchdev_lock);
2875
2876 rgvswctl = I915_READ16(MEMSWCTL);
2b4e57bd
ED		 2877
2878 /* Ack interrupts, disable EFC interrupt */
2879 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
2880 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
2881 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
2882 I915_WRITE(DEIIR, DE_PCU_EVENT);
2883 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
2884
2885 /* Go back to the starting frequency */
20e4d407 2886 ironlake_set_drps(dev, dev_priv->ips.fstart);
9270388e 2887 mdelay(1);
2b4e57bd
ED		 2888 rgvswctl |= MEMCTL_CMD_STS;
2889 I915_WRITE(MEMSWCTL, rgvswctl);
9270388e 2890 mdelay(1);
2b4e57bd 2891
9270388e 2892 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED		 2893}
2894
acbe9475
DV		 2895/* There's a funny hw issue where the hw returns all 0 when reading from
2896 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
2897 * ourselves, instead of doing a rmw cycle (which might result in us clearing
2898 * all limits and the gpu stuck at whatever frequency it is at atm).
2899 */
6917c7b9 2900static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 val)
2b4e57bd 2901{
7b9e0ae6 2902 u32 limits;
2b4e57bd 2903
20b46e59
DV		 2904 /* Only set the down limit when we've reached the lowest level to avoid
2905 * getting more interrupts, otherwise leave this clear. This prevents a
2906 * race in the hw when coming out of rc6: There's a tiny window where
2907 * the hw runs at the minimal clock before selecting the desired
2908 * frequency, if the down threshold expires in that window we will not
2909 * receive a down interrupt. */
b39fb297
BW		 2910 limits = dev_priv->rps.max_freq_softlimit << 24;
2911 if (val <= dev_priv->rps.min_freq_softlimit)
2912 limits |= dev_priv->rps.min_freq_softlimit << 16;
20b46e59
DV		 2913
2914 return limits;
2915}
2916
dd75fdc8
CW		 2917static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
2918{
2919 int new_power;
2920
2921 new_power = dev_priv->rps.power;
2922 switch (dev_priv->rps.power) {
2923 case LOW_POWER:
b39fb297 2924 if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq)
dd75fdc8
CW		 2925 new_power = BETWEEN;
2926 break;
2927
2928 case BETWEEN:
b39fb297 2929 if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq)
dd75fdc8 2930 new_power = LOW_POWER;
b39fb297 2931 else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq)
dd75fdc8
CW		 2932 new_power = HIGH_POWER;
2933 break;
2934
2935 case HIGH_POWER:
b39fb297 2936 if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq)
dd75fdc8
CW		 2937 new_power = BETWEEN;
2938 break;
2939 }
2940 /* Max/min bins are special */
b39fb297 2941 if (val == dev_priv->rps.min_freq_softlimit)
dd75fdc8 2942 new_power = LOW_POWER;
b39fb297 2943 if (val == dev_priv->rps.max_freq_softlimit)
dd75fdc8
CW		 2944 new_power = HIGH_POWER;
2945 if (new_power == dev_priv->rps.power)
2946 return;
2947
2948 /* Note the units here are not exactly 1us, but 1280ns. */
2949 switch (new_power) {
2950 case LOW_POWER:
2951 /* Upclock if more than 95% busy over 16ms */
2952 I915_WRITE(GEN6_RP_UP_EI, 12500);
2953 I915_WRITE(GEN6_RP_UP_THRESHOLD, 11800);
2954
2955 /* Downclock if less than 85% busy over 32ms */
2956 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
2957 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 21250);
2958
2959 I915_WRITE(GEN6_RP_CONTROL,
2960 GEN6_RP_MEDIA_TURBO |
2961 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2962 GEN6_RP_MEDIA_IS_GFX |
2963 GEN6_RP_ENABLE |
2964 GEN6_RP_UP_BUSY_AVG |
2965 GEN6_RP_DOWN_IDLE_AVG);
2966 break;
2967
2968 case BETWEEN:
2969 /* Upclock if more than 90% busy over 13ms */
2970 I915_WRITE(GEN6_RP_UP_EI, 10250);
2971 I915_WRITE(GEN6_RP_UP_THRESHOLD, 9225);
2972
2973 /* Downclock if less than 75% busy over 32ms */
2974 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
2975 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 18750);
2976
2977 I915_WRITE(GEN6_RP_CONTROL,
2978 GEN6_RP_MEDIA_TURBO |
2979 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2980 GEN6_RP_MEDIA_IS_GFX |
2981 GEN6_RP_ENABLE |
2982 GEN6_RP_UP_BUSY_AVG |
2983 GEN6_RP_DOWN_IDLE_AVG);
2984 break;
2985
2986 case HIGH_POWER:
2987 /* Upclock if more than 85% busy over 10ms */
2988 I915_WRITE(GEN6_RP_UP_EI, 8000);
2989 I915_WRITE(GEN6_RP_UP_THRESHOLD, 6800);
2990
2991 /* Downclock if less than 60% busy over 32ms */
2992 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
2993 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 15000);
2994
2995 I915_WRITE(GEN6_RP_CONTROL,
2996 GEN6_RP_MEDIA_TURBO |
2997 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2998 GEN6_RP_MEDIA_IS_GFX |
2999 GEN6_RP_ENABLE |
3000 GEN6_RP_UP_BUSY_AVG |
3001 GEN6_RP_DOWN_IDLE_AVG);
3002 break;
3003 }
3004
3005 dev_priv->rps.power = new_power;
3006 dev_priv->rps.last_adj = 0;
3007}
3008
2876ce73
CW		 3009static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val)
3010{
3011 u32 mask = 0;
3012
3013 if (val > dev_priv->rps.min_freq_softlimit)
3014 mask |= GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT;
3015 if (val < dev_priv->rps.max_freq_softlimit)
3016 mask |= GEN6_PM_RP_UP_THRESHOLD;
3017
3018 /* IVB and SNB hard hangs on looping batchbuffer
3019 * if GEN6_PM_UP_EI_EXPIRED is masked.
3020 */
3021 if (INTEL_INFO(dev_priv->dev)->gen <= 7 && !IS_HASWELL(dev_priv->dev))
3022 mask |= GEN6_PM_RP_UP_EI_EXPIRED;
3023
3024 return ~mask;
3025}
3026
b8a5ff8d
JM		 3027/* gen6_set_rps is called to update the frequency request, but should also be
3028 * called when the range (min_delay and max_delay) is modified so that we can
3029 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
20b46e59
DV		 3030void gen6_set_rps(struct drm_device *dev, u8 val)
3031{
3032 struct drm_i915_private *dev_priv = dev->dev_private;
7b9e0ae6 3033
4fc688ce 3034 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
b39fb297
BW		 3035 WARN_ON(val > dev_priv->rps.max_freq_softlimit);
3036 WARN_ON(val < dev_priv->rps.min_freq_softlimit);
004777cb 3037
eb64cad1
CW		 3038 /* min/max delay may still have been modified so be sure to
3039 * write the limits value.
3040 */
3041 if (val != dev_priv->rps.cur_freq) {
3042 gen6_set_rps_thresholds(dev_priv, val);
b8a5ff8d 3043
50e6a2a7 3044 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
eb64cad1
CW		 3045 I915_WRITE(GEN6_RPNSWREQ,
3046 HSW_FREQUENCY(val));
3047 else
3048 I915_WRITE(GEN6_RPNSWREQ,
3049 GEN6_FREQUENCY(val) |
3050 GEN6_OFFSET(0) |
3051 GEN6_AGGRESSIVE_TURBO);
b8a5ff8d 3052 }
7b9e0ae6 3053
7b9e0ae6
CW		 3054 /* Make sure we continue to get interrupts
3055 * until we hit the minimum or maximum frequencies.
3056 */
eb64cad1 3057 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, gen6_rps_limits(dev_priv, val));
2876ce73 3058 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
7b9e0ae6 3059
d5570a72
BW		 3060 POSTING_READ(GEN6_RPNSWREQ);
3061
b39fb297 3062 dev_priv->rps.cur_freq = val;
be2cde9a 3063 trace_intel_gpu_freq_change(val * 50);
2b4e57bd
ED		 3064}
3065
76c3552f
D		 3066/* vlv_set_rps_idle: Set the frequency to Rpn if Gfx clocks are down
3067 *
3068 * * If Gfx is Idle, then
3069 * 1. Mask Turbo interrupts
3070 * 2. Bring up Gfx clock
3071 * 3. Change the freq to Rpn and wait till P-Unit updates freq
3072 * 4. Clear the Force GFX CLK ON bit so that Gfx can down
3073 * 5. Unmask Turbo interrupts
3074*/
3075static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
3076{
3077 /*
3078 * When we are idle. Drop to min voltage state.
3079 */
3080
b39fb297 3081 if (dev_priv->rps.cur_freq <= dev_priv->rps.min_freq_softlimit)
76c3552f
D		 3082 return;
3083
3084 /* Mask turbo interrupt so that they will not come in between */
3085 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
3086
3087 /* Bring up the Gfx clock */
3088 I915_WRITE(VLV_GTLC_SURVIVABILITY_REG,
3089 I915_READ(VLV_GTLC_SURVIVABILITY_REG) |
3090 VLV_GFX_CLK_FORCE_ON_BIT);
3091
3092 if (wait_for(((VLV_GFX_CLK_STATUS_BIT &
3093 I915_READ(VLV_GTLC_SURVIVABILITY_REG)) != 0), 5)) {
3094 DRM_ERROR("GFX_CLK_ON request timed out\n");
3095 return;
3096 }
3097
b39fb297 3098 dev_priv->rps.cur_freq = dev_priv->rps.min_freq_softlimit;
76c3552f
D		 3099
3100 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ,
b39fb297 3101 dev_priv->rps.min_freq_softlimit);
76c3552f
D		 3102
3103 if (wait_for(((vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS))
3104 & GENFREQSTATUS) == 0, 5))
3105 DRM_ERROR("timed out waiting for Punit\n");
3106
3107 /* Release the Gfx clock */
3108 I915_WRITE(VLV_GTLC_SURVIVABILITY_REG,
3109 I915_READ(VLV_GTLC_SURVIVABILITY_REG) &
3110 ~VLV_GFX_CLK_FORCE_ON_BIT);
2876ce73
CW		 3111
3112 I915_WRITE(GEN6_PMINTRMSK,
3113 gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq));
76c3552f
D		 3114}
3115
b29c19b6
CW		 3116void gen6_rps_idle(struct drm_i915_private *dev_priv)
3117{
691bb717
DL		 3118 struct drm_device *dev = dev_priv->dev;
3119
b29c19b6 3120 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 3121 if (dev_priv->rps.enabled) {
691bb717 3122 if (IS_VALLEYVIEW(dev))
76c3552f 3123 vlv_set_rps_idle(dev_priv);
c0951f0c 3124 else
b39fb297 3125 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
c0951f0c
CW		 3126 dev_priv->rps.last_adj = 0;
3127 }
b29c19b6
CW		 3128 mutex_unlock(&dev_priv->rps.hw_lock);
3129}
3130
3131void gen6_rps_boost(struct drm_i915_private *dev_priv)
3132{
691bb717
DL		 3133 struct drm_device *dev = dev_priv->dev;
3134
b29c19b6 3135 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 3136 if (dev_priv->rps.enabled) {
691bb717 3137 if (IS_VALLEYVIEW(dev))
b39fb297 3138 valleyview_set_rps(dev_priv->dev, dev_priv->rps.max_freq_softlimit);
c0951f0c 3139 else
b39fb297 3140 gen6_set_rps(dev_priv->dev, dev_priv->rps.max_freq_softlimit);
c0951f0c
CW		 3141 dev_priv->rps.last_adj = 0;
3142 }
b29c19b6
CW		 3143 mutex_unlock(&dev_priv->rps.hw_lock);
3144}
3145
0a073b84
JB		 3146void valleyview_set_rps(struct drm_device *dev, u8 val)
3147{
3148 struct drm_i915_private *dev_priv = dev->dev_private;
7a67092a 3149
0a073b84 3150 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
b39fb297
BW		 3151 WARN_ON(val > dev_priv->rps.max_freq_softlimit);
3152 WARN_ON(val < dev_priv->rps.min_freq_softlimit);
0a073b84 3153
73008b98 3154 DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
b39fb297
BW		 3155 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
3156 dev_priv->rps.cur_freq,
2ec3815f 3157 vlv_gpu_freq(dev_priv, val), val);
0a073b84 3158
2876ce73
CW		 3159 if (val != dev_priv->rps.cur_freq)
3160 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
0a073b84 3161
09c87db8 3162 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
0a073b84 3163
b39fb297 3164 dev_priv->rps.cur_freq = val;
2ec3815f 3165 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv, val));
0a073b84
JB		 3166}
3167
44fc7d5c 3168static void gen6_disable_rps_interrupts(struct drm_device *dev)
2b4e57bd
ED		 3169{
3170 struct drm_i915_private *dev_priv = dev->dev_private;
3171
2b4e57bd 3172 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
a6706b45
D		 3173 I915_WRITE(GEN6_PMIER, I915_READ(GEN6_PMIER) &
3174 ~dev_priv->pm_rps_events);
2b4e57bd
ED		 3175 /* Complete PM interrupt masking here doesn't race with the rps work
3176 * item again unmasking PM interrupts because that is using a different
3177 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
3178 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
3179
59cdb63d 3180 spin_lock_irq(&dev_priv->irq_lock);
c6a828d3 3181 dev_priv->rps.pm_iir = 0;
59cdb63d 3182 spin_unlock_irq(&dev_priv->irq_lock);
2b4e57bd 3183
a6706b45 3184 I915_WRITE(GEN6_PMIIR, dev_priv->pm_rps_events);
2b4e57bd
ED		 3185}
3186
44fc7d5c 3187static void gen6_disable_rps(struct drm_device *dev)
d20d4f0c
JB		 3188{
3189 struct drm_i915_private *dev_priv = dev->dev_private;
3190
3191 I915_WRITE(GEN6_RC_CONTROL, 0);
44fc7d5c 3192 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
d20d4f0c 3193
44fc7d5c
DV		 3194 gen6_disable_rps_interrupts(dev);
3195}
3196
3197static void valleyview_disable_rps(struct drm_device *dev)
3198{
3199 struct drm_i915_private *dev_priv = dev->dev_private;
3200
3201 I915_WRITE(GEN6_RC_CONTROL, 0);
d20d4f0c 3202
44fc7d5c 3203 gen6_disable_rps_interrupts(dev);
d20d4f0c
JB		 3204}
3205
dc39fff7
BW		 3206static void intel_print_rc6_info(struct drm_device *dev, u32 mode)
3207{
dc39fff7 3208 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
1c79b42f
BW		 3209 (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
3210 (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
3211 (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
dc39fff7
BW		 3212}
3213
2b4e57bd
ED		 3214int intel_enable_rc6(const struct drm_device *dev)
3215{
eb4926e4
DL		 3216 /* No RC6 before Ironlake */
3217 if (INTEL_INFO(dev)->gen < 5)
3218 return 0;
3219
456470eb 3220 /* Respect the kernel parameter if it is set */
d330a953
JN		 3221 if (i915.enable_rc6 >= 0)
3222 return i915.enable_rc6;
2b4e57bd 3223
6567d748
CW		 3224 /* Disable RC6 on Ironlake */
3225 if (INTEL_INFO(dev)->gen == 5)
3226 return 0;
2b4e57bd 3227
8bade1ad 3228 if (IS_IVYBRIDGE(dev))
cca84a1f 3229 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
8bade1ad
BW		 3230
3231 return INTEL_RC6_ENABLE;
2b4e57bd
ED		 3232}
3233
44fc7d5c
DV		 3234static void gen6_enable_rps_interrupts(struct drm_device *dev)
3235{
3236 struct drm_i915_private *dev_priv = dev->dev_private;
3237
3238 spin_lock_irq(&dev_priv->irq_lock);
a0b3335a 3239 WARN_ON(dev_priv->rps.pm_iir);
a6706b45
D		 3240 snb_enable_pm_irq(dev_priv, dev_priv->pm_rps_events);
3241 I915_WRITE(GEN6_PMIIR, dev_priv->pm_rps_events);
44fc7d5c 3242 spin_unlock_irq(&dev_priv->irq_lock);
44fc7d5c
DV		 3243}
3244
3280e8b0
BW		 3245static void parse_rp_state_cap(struct drm_i915_private *dev_priv, u32 rp_state_cap)
3246{
3247 /* All of these values are in units of 50MHz */
3248 dev_priv->rps.cur_freq = 0;
3249 /* static values from HW: RP0 < RPe < RP1 < RPn (min_freq) */
3250 dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff;
3251 dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff;
3252 dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff;
3253 /* XXX: only BYT has a special efficient freq */
3254 dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq;
3255 /* hw_max = RP0 until we check for overclocking */
3256 dev_priv->rps.max_freq = dev_priv->rps.rp0_freq;
3257
3258 /* Preserve min/max settings in case of re-init */
3259 if (dev_priv->rps.max_freq_softlimit == 0)
3260 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
3261
3262 if (dev_priv->rps.min_freq_softlimit == 0)
3263 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
3264}
3265
6edee7f3
BW		 3266static void gen8_enable_rps(struct drm_device *dev)
3267{
3268 struct drm_i915_private *dev_priv = dev->dev_private;
3269 struct intel_ring_buffer *ring;
3270 uint32_t rc6_mask = 0, rp_state_cap;
3271 int unused;
3272
3273 /* 1a: Software RC state - RC0 */
3274 I915_WRITE(GEN6_RC_STATE, 0);
3275
3276 /* 1c & 1d: Get forcewake during program sequence. Although the driver
3277 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
c8d9a590 3278 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW		 3279
3280 /* 2a: Disable RC states. */
3281 I915_WRITE(GEN6_RC_CONTROL, 0);
3282
3283 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3280e8b0 3284 parse_rp_state_cap(dev_priv, rp_state_cap);
6edee7f3
BW		 3285
3286 /* 2b: Program RC6 thresholds.*/
3287 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
3288 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
3289 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
3290 for_each_ring(ring, dev_priv, unused)
3291 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3292 I915_WRITE(GEN6_RC_SLEEP, 0);
3293 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
3294
3295 /* 3: Enable RC6 */
3296 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
3297 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
abbf9d2c 3298 intel_print_rc6_info(dev, rc6_mask);
6edee7f3 3299 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
abbf9d2c
BW		 3300 GEN6_RC_CTL_EI_MODE(1) |
3301 rc6_mask);
6edee7f3
BW		 3302
3303 /* 4 Program defaults and thresholds for RPS*/
f9bdc585
BW		 3304 I915_WRITE(GEN6_RPNSWREQ,
3305 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
3306 I915_WRITE(GEN6_RC_VIDEO_FREQ,
3307 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
6edee7f3
BW		 3308 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
3309 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
3310
3311 /* Docs recommend 900MHz, and 300 MHz respectively */
3312 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
b39fb297
BW		 3313 dev_priv->rps.max_freq_softlimit << 24 |
3314 dev_priv->rps.min_freq_softlimit << 16);
6edee7f3
BW		 3315
3316 I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
3317 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
3318 I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
3319 I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
3320
3321 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3322
3323 /* 5: Enable RPS */
3324 I915_WRITE(GEN6_RP_CONTROL,
3325 GEN6_RP_MEDIA_TURBO |
3326 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3327 GEN6_RP_MEDIA_IS_GFX |
3328 GEN6_RP_ENABLE |
3329 GEN6_RP_UP_BUSY_AVG |
3330 GEN6_RP_DOWN_IDLE_AVG);
3331
3332 /* 6: Ring frequency + overclocking (our driver does this later */
3333
3334 gen6_set_rps(dev, (I915_READ(GEN6_GT_PERF_STATUS) & 0xff00) >> 8);
3335
3336 gen6_enable_rps_interrupts(dev);
3337
c8d9a590 3338 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW		 3339}
3340
79f5b2c7 3341static void gen6_enable_rps(struct drm_device *dev)
2b4e57bd 3342{
79f5b2c7 3343 struct drm_i915_private *dev_priv = dev->dev_private;
b4519513 3344 struct intel_ring_buffer *ring;
2a5913a8 3345 u32 rp_state_cap;
7b9e0ae6 3346 u32 gt_perf_status;
d060c169 3347 u32 rc6vids, pcu_mbox = 0, rc6_mask = 0;
2b4e57bd 3348 u32 gtfifodbg;
2b4e57bd 3349 int rc6_mode;
42c0526c 3350 int i, ret;
2b4e57bd 3351
4fc688ce 3352 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 3353
2b4e57bd
ED		 3354 /* Here begins a magic sequence of register writes to enable
3355 * auto-downclocking.
3356 *
3357 * Perhaps there might be some value in exposing these to
3358 * userspace...
3359 */
3360 I915_WRITE(GEN6_RC_STATE, 0);
2b4e57bd
ED		 3361
3362 /* Clear the DBG now so we don't confuse earlier errors */
3363 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
3364 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
3365 I915_WRITE(GTFIFODBG, gtfifodbg);
3366 }
3367
c8d9a590 3368 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
2b4e57bd 3369
7b9e0ae6
CW		 3370 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3371 gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
3372
3280e8b0 3373 parse_rp_state_cap(dev_priv, rp_state_cap);
dd0a1aa1 3374
2b4e57bd
ED		 3375 /* disable the counters and set deterministic thresholds */
3376 I915_WRITE(GEN6_RC_CONTROL, 0);
3377
3378 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
3379 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
3380 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
3381 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3382 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3383
b4519513
CW		 3384 for_each_ring(ring, dev_priv, i)
3385 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2b4e57bd
ED		 3386
3387 I915_WRITE(GEN6_RC_SLEEP, 0);
3388 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
29c78f60 3389 if (IS_IVYBRIDGE(dev))
351aa566
SM		 3390 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
3391 else
3392 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
0920a487 3393 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2b4e57bd
ED		 3394 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
3395
5a7dc92a 3396 /* Check if we are enabling RC6 */
2b4e57bd
ED		 3397 rc6_mode = intel_enable_rc6(dev_priv->dev);
3398 if (rc6_mode & INTEL_RC6_ENABLE)
3399 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
3400
5a7dc92a
ED		 3401 /* We don't use those on Haswell */
3402 if (!IS_HASWELL(dev)) {
3403 if (rc6_mode & INTEL_RC6p_ENABLE)
3404 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2b4e57bd 3405
5a7dc92a
ED		 3406 if (rc6_mode & INTEL_RC6pp_ENABLE)
3407 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
3408 }
2b4e57bd 3409
dc39fff7 3410 intel_print_rc6_info(dev, rc6_mask);
2b4e57bd
ED		 3411
3412 I915_WRITE(GEN6_RC_CONTROL,
3413 rc6_mask |
3414 GEN6_RC_CTL_EI_MODE(1) |
3415 GEN6_RC_CTL_HW_ENABLE);
3416
dd75fdc8
CW		 3417 /* Power down if completely idle for over 50ms */
3418 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
2b4e57bd 3419 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2b4e57bd 3420
42c0526c 3421 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
d060c169 3422 if (ret)
42c0526c 3423 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
d060c169
BW		 3424
3425 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
3426 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
3427 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
b39fb297 3428 (dev_priv->rps.max_freq_softlimit & 0xff) * 50,
d060c169 3429 (pcu_mbox & 0xff) * 50);
b39fb297 3430 dev_priv->rps.max_freq = pcu_mbox & 0xff;
2b4e57bd
ED		 3431 }
3432
dd75fdc8 3433 dev_priv->rps.power = HIGH_POWER; /* force a reset */
b39fb297 3434 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
2b4e57bd 3435
44fc7d5c 3436 gen6_enable_rps_interrupts(dev);
2b4e57bd 3437
31643d54
BW		 3438 rc6vids = 0;
3439 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
3440 if (IS_GEN6(dev) && ret) {
3441 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
3442 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
3443 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
3444 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
3445 rc6vids &= 0xffff00;
3446 rc6vids |= GEN6_ENCODE_RC6_VID(450);
3447 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
3448 if (ret)
3449 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
3450 }
3451
c8d9a590 3452 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
2b4e57bd
ED		 3453}
3454
c67a470b 3455void gen6_update_ring_freq(struct drm_device *dev)
2b4e57bd 3456{
79f5b2c7 3457 struct drm_i915_private *dev_priv = dev->dev_private;
2b4e57bd 3458 int min_freq = 15;
3ebecd07
CW		 3459 unsigned int gpu_freq;
3460 unsigned int max_ia_freq, min_ring_freq;
2b4e57bd 3461 int scaling_factor = 180;
eda79642 3462 struct cpufreq_policy *policy;
2b4e57bd 3463
4fc688ce 3464 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 3465
eda79642
BW		 3466 policy = cpufreq_cpu_get(0);
3467 if (policy) {
3468 max_ia_freq = policy->cpuinfo.max_freq;
3469 cpufreq_cpu_put(policy);
3470 } else {
3471 /*
3472 * Default to measured freq if none found, PCU will ensure we
3473 * don't go over
3474 */
2b4e57bd 3475 max_ia_freq = tsc_khz;
eda79642 3476 }
2b4e57bd
ED		 3477
3478 /* Convert from kHz to MHz */
3479 max_ia_freq /= 1000;
3480
153b4b95 3481 min_ring_freq = I915_READ(DCLK) & 0xf;
f6aca45c
BW		 3482 /* convert DDR frequency from units of 266.6MHz to bandwidth */
3483 min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3ebecd07 3484
2b4e57bd
ED		 3485 /*
3486 * For each potential GPU frequency, load a ring frequency we'd like
3487 * to use for memory access. We do this by specifying the IA frequency
3488 * the PCU should use as a reference to determine the ring frequency.
3489 */
4b28a1f3
PZ		 3490 for (gpu_freq = dev_priv->rps.max_freq_softlimit;
3491 gpu_freq >= dev_priv->rps.min_freq_softlimit && gpu_freq != 0;
2b4e57bd 3492 gpu_freq--) {
b39fb297 3493 int diff = dev_priv->rps.max_freq_softlimit - gpu_freq;
3ebecd07
CW		 3494 unsigned int ia_freq = 0, ring_freq = 0;
3495
46c764d4
BW		 3496 if (INTEL_INFO(dev)->gen >= 8) {
3497 /* max(2 * GT, DDR). NB: GT is 50MHz units */
3498 ring_freq = max(min_ring_freq, gpu_freq);
3499 } else if (IS_HASWELL(dev)) {
f6aca45c 3500 ring_freq = mult_frac(gpu_freq, 5, 4);
3ebecd07
CW		 3501 ring_freq = max(min_ring_freq, ring_freq);
3502 /* leave ia_freq as the default, chosen by cpufreq */
3503 } else {
3504 /* On older processors, there is no separate ring
3505 * clock domain, so in order to boost the bandwidth
3506 * of the ring, we need to upclock the CPU (ia_freq).
3507 *
3508 * For GPU frequencies less than 750MHz,
3509 * just use the lowest ring freq.
3510 */
3511 if (gpu_freq < min_freq)
3512 ia_freq = 800;
3513 else
3514 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
3515 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
3516 }
2b4e57bd 3517
42c0526c
BW		 3518 sandybridge_pcode_write(dev_priv,
3519 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3ebecd07
CW		 3520 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
3521 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
3522 gpu_freq);
2b4e57bd 3523 }
2b4e57bd
ED		 3524}
3525
0a073b84
JB		 3526int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
3527{
3528 u32 val, rp0;
3529
64936258 3530 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
0a073b84
JB		 3531
3532 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
3533 /* Clamp to max */
3534 rp0 = min_t(u32, rp0, 0xea);
3535
3536 return rp0;
3537}
3538
3539static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
3540{
3541 u32 val, rpe;
3542
64936258 3543 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
0a073b84 3544 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
64936258 3545 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
0a073b84
JB		 3546 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
3547
3548 return rpe;
3549}
3550
3551int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
3552{
64936258 3553 return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
0a073b84
JB		 3554}
3555
ae48434c
ID		 3556/* Check that the pctx buffer wasn't move under us. */
3557static void valleyview_check_pctx(struct drm_i915_private *dev_priv)
3558{
3559 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
3560
3561 WARN_ON(pctx_addr != dev_priv->mm.stolen_base +
3562 dev_priv->vlv_pctx->stolen->start);
3563}
3564
c9cddffc
JB		 3565static void valleyview_setup_pctx(struct drm_device *dev)
3566{
3567 struct drm_i915_private *dev_priv = dev->dev_private;
3568 struct drm_i915_gem_object *pctx;
3569 unsigned long pctx_paddr;
3570 u32 pcbr;
3571 int pctx_size = 24*1024;
3572
17b0c1f7
ID		 3573 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3574
c9cddffc
JB		 3575 pcbr = I915_READ(VLV_PCBR);
3576 if (pcbr) {
3577 /* BIOS set it up already, grab the pre-alloc'd space */
3578 int pcbr_offset;
3579
3580 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
3581 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
3582 pcbr_offset,
190d6cd5 3583 I915_GTT_OFFSET_NONE,
c9cddffc
JB		 3584 pctx_size);
3585 goto out;
3586 }
3587
3588 /*
3589 * From the Gunit register HAS:
3590 * The Gfx driver is expected to program this register and ensure
3591 * proper allocation within Gfx stolen memory. For example, this
3592 * register should be programmed such than the PCBR range does not
3593 * overlap with other ranges, such as the frame buffer, protected
3594 * memory, or any other relevant ranges.
3595 */
3596 pctx = i915_gem_object_create_stolen(dev, pctx_size);
3597 if (!pctx) {
3598 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
3599 return;
3600 }
3601
3602 pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
3603 I915_WRITE(VLV_PCBR, pctx_paddr);
3604
3605out:
3606 dev_priv->vlv_pctx = pctx;
3607}
3608
ae48434c
ID		 3609static void valleyview_cleanup_pctx(struct drm_device *dev)
3610{
3611 struct drm_i915_private *dev_priv = dev->dev_private;
3612
3613 if (WARN_ON(!dev_priv->vlv_pctx))
3614 return;
3615
3616 drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
3617 dev_priv->vlv_pctx = NULL;
3618}
3619
0a073b84
JB		 3620static void valleyview_enable_rps(struct drm_device *dev)
3621{
3622 struct drm_i915_private *dev_priv = dev->dev_private;
3623 struct intel_ring_buffer *ring;
2a5913a8 3624 u32 gtfifodbg, val, rc6_mode = 0;
0a073b84
JB		 3625 int i;
3626
3627 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3628
ae48434c
ID		 3629 valleyview_check_pctx(dev_priv);
3630
0a073b84 3631 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
f7d85c1e
JB		 3632 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
3633 gtfifodbg);
0a073b84
JB		 3634 I915_WRITE(GTFIFODBG, gtfifodbg);
3635 }
3636
c8d9a590
D		 3637 /* If VLV, Forcewake all wells, else re-direct to regular path */
3638 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
0a073b84
JB		 3639
3640 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
3641 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
3642 I915_WRITE(GEN6_RP_UP_EI, 66000);
3643 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
3644
3645 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3646
3647 I915_WRITE(GEN6_RP_CONTROL,
3648 GEN6_RP_MEDIA_TURBO |
3649 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3650 GEN6_RP_MEDIA_IS_GFX |
3651 GEN6_RP_ENABLE |
3652 GEN6_RP_UP_BUSY_AVG |
3653 GEN6_RP_DOWN_IDLE_CONT);
3654
3655 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
3656 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3657 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3658
3659 for_each_ring(ring, dev_priv, i)
3660 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3661
2f0aa304 3662 I915_WRITE(GEN6_RC6_THRESHOLD, 0x557);
0a073b84
JB		 3663
3664 /* allows RC6 residency counter to work */
49798eb2
JB		 3665 I915_WRITE(VLV_COUNTER_CONTROL,
3666 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
3667 VLV_MEDIA_RC6_COUNT_EN |
3668 VLV_RENDER_RC6_COUNT_EN));
a2b23fe0 3669 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
6b88f295 3670 rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
dc39fff7
BW		 3671
3672 intel_print_rc6_info(dev, rc6_mode);
3673
a2b23fe0 3674 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
0a073b84 3675
64936258 3676 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
0a073b84
JB		 3677
3678 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
3679 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
3680
b39fb297 3681 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
73008b98 3682 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
b39fb297
BW		 3683 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
3684 dev_priv->rps.cur_freq);
0a073b84 3685
2a5913a8
BW		 3686 dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv);
3687 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
73008b98 3688 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
2a5913a8
BW		 3689 vlv_gpu_freq(dev_priv, dev_priv->rps.max_freq),
3690 dev_priv->rps.max_freq);
0a073b84 3691
b39fb297 3692 dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv);
73008b98 3693 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
b39fb297
BW		 3694 vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
3695 dev_priv->rps.efficient_freq);
0a073b84 3696
2a5913a8 3697 dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv);
73008b98 3698 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
2a5913a8
BW		 3699 vlv_gpu_freq(dev_priv, dev_priv->rps.min_freq),
3700 dev_priv->rps.min_freq);
dd0a1aa1
JM		 3701
3702 /* Preserve min/max settings in case of re-init */
b39fb297 3703 if (dev_priv->rps.max_freq_softlimit == 0)
2a5913a8 3704 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
dd0a1aa1 3705
b39fb297 3706 if (dev_priv->rps.min_freq_softlimit == 0)
2a5913a8 3707 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
0a073b84 3708
73008b98 3709 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
b39fb297
BW		 3710 vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
3711 dev_priv->rps.efficient_freq);
0a073b84 3712
b39fb297 3713 valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
0a073b84 3714
44fc7d5c 3715 gen6_enable_rps_interrupts(dev);
0a073b84 3716
c8d9a590 3717 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
0a073b84
JB		 3718}
3719
930ebb46 3720void ironlake_teardown_rc6(struct drm_device *dev)
2b4e57bd
ED		 3721{
3722 struct drm_i915_private *dev_priv = dev->dev_private;
3723
3e373948 3724 if (dev_priv->ips.renderctx) {
d7f46fc4 3725 i915_gem_object_ggtt_unpin(dev_priv->ips.renderctx);
3e373948
DV		 3726 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
3727 dev_priv->ips.renderctx = NULL;
2b4e57bd
ED		 3728 }
3729
3e373948 3730 if (dev_priv->ips.pwrctx) {
d7f46fc4 3731 i915_gem_object_ggtt_unpin(dev_priv->ips.pwrctx);
3e373948
DV		 3732 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
3733 dev_priv->ips.pwrctx = NULL;
2b4e57bd
ED		 3734 }
3735}
3736
930ebb46 3737static void ironlake_disable_rc6(struct drm_device *dev)
2b4e57bd
ED		 3738{
3739 struct drm_i915_private *dev_priv = dev->dev_private;
3740
3741 if (I915_READ(PWRCTXA)) {
3742 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
3743 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
3744 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
3745 50);
3746
3747 I915_WRITE(PWRCTXA, 0);
3748 POSTING_READ(PWRCTXA);
3749
3750 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
3751 POSTING_READ(RSTDBYCTL);
3752 }
2b4e57bd
ED		 3753}
3754
3755static int ironlake_setup_rc6(struct drm_device *dev)
3756{
3757 struct drm_i915_private *dev_priv = dev->dev_private;
3758
3e373948
DV		 3759 if (dev_priv->ips.renderctx == NULL)
3760 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
3761 if (!dev_priv->ips.renderctx)
2b4e57bd
ED		 3762 return -ENOMEM;
3763
3e373948
DV		 3764 if (dev_priv->ips.pwrctx == NULL)
3765 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
3766 if (!dev_priv->ips.pwrctx) {
2b4e57bd
ED		 3767 ironlake_teardown_rc6(dev);
3768 return -ENOMEM;
3769 }
3770
3771 return 0;
3772}
3773
930ebb46 3774static void ironlake_enable_rc6(struct drm_device *dev)
2b4e57bd
ED		 3775{
3776 struct drm_i915_private *dev_priv = dev->dev_private;
6d90c952 3777 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
3e960501 3778 bool was_interruptible;
2b4e57bd
ED		 3779 int ret;
3780
3781 /* rc6 disabled by default due to repeated reports of hanging during
3782 * boot and resume.
3783 */
3784 if (!intel_enable_rc6(dev))
3785 return;
3786
79f5b2c7
DV		 3787 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3788
2b4e57bd 3789 ret = ironlake_setup_rc6(dev);
79f5b2c7 3790 if (ret)
2b4e57bd 3791 return;
2b4e57bd 3792
3e960501
CW		 3793 was_interruptible = dev_priv->mm.interruptible;
3794 dev_priv->mm.interruptible = false;
3795
2b4e57bd
ED		 3796 /*
3797 * GPU can automatically power down the render unit if given a page
3798 * to save state.
3799 */
6d90c952 3800 ret = intel_ring_begin(ring, 6);
2b4e57bd
ED		 3801 if (ret) {
3802 ironlake_teardown_rc6(dev);
3e960501 3803 dev_priv->mm.interruptible = was_interruptible;
2b4e57bd
ED		 3804 return;
3805 }
3806
6d90c952
DV		 3807 intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
3808 intel_ring_emit(ring, MI_SET_CONTEXT);
f343c5f6 3809 intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) |
6d90c952
DV		 3810 MI_MM_SPACE_GTT |
3811 MI_SAVE_EXT_STATE_EN |
3812 MI_RESTORE_EXT_STATE_EN |
3813 MI_RESTORE_INHIBIT);
3814 intel_ring_emit(ring, MI_SUSPEND_FLUSH);
3815 intel_ring_emit(ring, MI_NOOP);
3816 intel_ring_emit(ring, MI_FLUSH);
3817 intel_ring_advance(ring);
2b4e57bd
ED		 3818
3819 /*
3820 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
3821 * does an implicit flush, combined with MI_FLUSH above, it should be
3822 * safe to assume that renderctx is valid
3823 */
3e960501
CW		 3824 ret = intel_ring_idle(ring);
3825 dev_priv->mm.interruptible = was_interruptible;
2b4e57bd 3826 if (ret) {
def27a58 3827 DRM_ERROR("failed to enable ironlake power savings\n");
2b4e57bd 3828 ironlake_teardown_rc6(dev);
2b4e57bd
ED		 3829 return;
3830 }
3831
f343c5f6 3832 I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN);
2b4e57bd 3833 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
dc39fff7
BW		 3834
3835 intel_print_rc6_info(dev, INTEL_RC6_ENABLE);
2b4e57bd
ED		 3836}
3837
dde18883
ED		 3838static unsigned long intel_pxfreq(u32 vidfreq)
3839{
3840 unsigned long freq;
3841 int div = (vidfreq & 0x3f0000) >> 16;
3842 int post = (vidfreq & 0x3000) >> 12;
3843 int pre = (vidfreq & 0x7);
3844
3845 if (!pre)
3846 return 0;
3847
3848 freq = ((div * 133333) / ((1<<post) * pre));
3849
3850 return freq;
3851}
3852
eb48eb00
DV		 3853static const struct cparams {
3854 u16 i;
3855 u16 t;
3856 u16 m;
3857 u16 c;
3858} cparams[] = {
3859 { 1, 1333, 301, 28664 },
3860 { 1, 1066, 294, 24460 },
3861 { 1, 800, 294, 25192 },
3862 { 0, 1333, 276, 27605 },
3863 { 0, 1066, 276, 27605 },
3864 { 0, 800, 231, 23784 },
3865};
3866
f531dcb2 3867static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 3868{
3869 u64 total_count, diff, ret;
3870 u32 count1, count2, count3, m = 0, c = 0;
3871 unsigned long now = jiffies_to_msecs(jiffies), diff1;
3872 int i;
3873
02d71956
DV		 3874 assert_spin_locked(&mchdev_lock);
3875
20e4d407 3876 diff1 = now - dev_priv->ips.last_time1;
eb48eb00
DV		 3877
3878 /* Prevent division-by-zero if we are asking too fast.
3879 * Also, we don't get interesting results if we are polling
3880 * faster than once in 10ms, so just return the saved value
3881 * in such cases.
3882 */
3883 if (diff1 <= 10)
20e4d407 3884 return dev_priv->ips.chipset_power;
eb48eb00
DV		 3885
3886 count1 = I915_READ(DMIEC);
3887 count2 = I915_READ(DDREC);
3888 count3 = I915_READ(CSIEC);
3889
3890 total_count = count1 + count2 + count3;
3891
3892 /* FIXME: handle per-counter overflow */
20e4d407
DV		 3893 if (total_count < dev_priv->ips.last_count1) {
3894 diff = ~0UL - dev_priv->ips.last_count1;
eb48eb00
DV		 3895 diff += total_count;
3896 } else {
20e4d407 3897 diff = total_count - dev_priv->ips.last_count1;
eb48eb00
DV		 3898 }
3899
3900 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
20e4d407
DV		 3901 if (cparams[i].i == dev_priv->ips.c_m &&
3902 cparams[i].t == dev_priv->ips.r_t) {
eb48eb00
DV		 3903 m = cparams[i].m;
3904 c = cparams[i].c;
3905 break;
3906 }
3907 }
3908
3909 diff = div_u64(diff, diff1);
3910 ret = ((m * diff) + c);
3911 ret = div_u64(ret, 10);
3912
20e4d407
DV		 3913 dev_priv->ips.last_count1 = total_count;
3914 dev_priv->ips.last_time1 = now;
eb48eb00 3915
20e4d407 3916 dev_priv->ips.chipset_power = ret;
eb48eb00
DV		 3917
3918 return ret;
3919}
3920
f531dcb2
CW		 3921unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
3922{
3d13ef2e 3923 struct drm_device *dev = dev_priv->dev;
f531dcb2
CW		 3924 unsigned long val;
3925
3d13ef2e 3926 if (INTEL_INFO(dev)->gen != 5)
f531dcb2
CW		 3927 return 0;
3928
3929 spin_lock_irq(&mchdev_lock);
3930
3931 val = __i915_chipset_val(dev_priv);
3932
3933 spin_unlock_irq(&mchdev_lock);
3934
3935 return val;
3936}
3937
eb48eb00
DV		 3938unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
3939{
3940 unsigned long m, x, b;
3941 u32 tsfs;
3942
3943 tsfs = I915_READ(TSFS);
3944
3945 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
3946 x = I915_READ8(TR1);
3947
3948 b = tsfs & TSFS_INTR_MASK;
3949
3950 return ((m * x) / 127) - b;
3951}
3952
3953static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
3954{
3d13ef2e 3955 struct drm_device *dev = dev_priv->dev;
eb48eb00
DV		 3956 static const struct v_table {
3957 u16 vd; /* in .1 mil */
3958 u16 vm; /* in .1 mil */
3959 } v_table[] = {
3960 { 0, 0, },
3961 { 375, 0, },
3962 { 500, 0, },
3963 { 625, 0, },
3964 { 750, 0, },
3965 { 875, 0, },
3966 { 1000, 0, },
3967 { 1125, 0, },
3968 { 4125, 3000, },
3969 { 4125, 3000, },
3970 { 4125, 3000, },
3971 { 4125, 3000, },
3972 { 4125, 3000, },
3973 { 4125, 3000, },
3974 { 4125, 3000, },
3975 { 4125, 3000, },
3976 { 4125, 3000, },
3977 { 4125, 3000, },
3978 { 4125, 3000, },
3979 { 4125, 3000, },
3980 { 4125, 3000, },
3981 { 4125, 3000, },
3982 { 4125, 3000, },
3983 { 4125, 3000, },
3984 { 4125, 3000, },
3985 { 4125, 3000, },
3986 { 4125, 3000, },
3987 { 4125, 3000, },
3988 { 4125, 3000, },
3989 { 4125, 3000, },
3990 { 4125, 3000, },
3991 { 4125, 3000, },
3992 { 4250, 3125, },
3993 { 4375, 3250, },
3994 { 4500, 3375, },
3995 { 4625, 3500, },
3996 { 4750, 3625, },
3997 { 4875, 3750, },
3998 { 5000, 3875, },
3999 { 5125, 4000, },
4000 { 5250, 4125, },
4001 { 5375, 4250, },
4002 { 5500, 4375, },
4003 { 5625, 4500, },
4004 { 5750, 4625, },
4005 { 5875, 4750, },
4006 { 6000, 4875, },
4007 { 6125, 5000, },
4008 { 6250, 5125, },
4009 { 6375, 5250, },
4010 { 6500, 5375, },
4011 { 6625, 5500, },
4012 { 6750, 5625, },
4013 { 6875, 5750, },
4014 { 7000, 5875, },
4015 { 7125, 6000, },
4016 { 7250, 6125, },
4017 { 7375, 6250, },
4018 { 7500, 6375, },
4019 { 7625, 6500, },
4020 { 7750, 6625, },
4021 { 7875, 6750, },
4022 { 8000, 6875, },
4023 { 8125, 7000, },
4024 { 8250, 7125, },
4025 { 8375, 7250, },
4026 { 8500, 7375, },
4027 { 8625, 7500, },
4028 { 8750, 7625, },
4029 { 8875, 7750, },
4030 { 9000, 7875, },
4031 { 9125, 8000, },
4032 { 9250, 8125, },
4033 { 9375, 8250, },
4034 { 9500, 8375, },
4035 { 9625, 8500, },
4036 { 9750, 8625, },
4037 { 9875, 8750, },
4038 { 10000, 8875, },
4039 { 10125, 9000, },
4040 { 10250, 9125, },
4041 { 10375, 9250, },
4042 { 10500, 9375, },
4043 { 10625, 9500, },
4044 { 10750, 9625, },
4045 { 10875, 9750, },
4046 { 11000, 9875, },
4047 { 11125, 10000, },
4048 { 11250, 10125, },
4049 { 11375, 10250, },
4050 { 11500, 10375, },
4051 { 11625, 10500, },
4052 { 11750, 10625, },
4053 { 11875, 10750, },
4054 { 12000, 10875, },
4055 { 12125, 11000, },
4056 { 12250, 11125, },
4057 { 12375, 11250, },
4058 { 12500, 11375, },
4059 { 12625, 11500, },
4060 { 12750, 11625, },
4061 { 12875, 11750, },
4062 { 13000, 11875, },
4063 { 13125, 12000, },
4064 { 13250, 12125, },
4065 { 13375, 12250, },
4066 { 13500, 12375, },
4067 { 13625, 12500, },
4068 { 13750, 12625, },
4069 { 13875, 12750, },
4070 { 14000, 12875, },
4071 { 14125, 13000, },
4072 { 14250, 13125, },
4073 { 14375, 13250, },
4074 { 14500, 13375, },
4075 { 14625, 13500, },
4076 { 14750, 13625, },
4077 { 14875, 13750, },
4078 { 15000, 13875, },
4079 { 15125, 14000, },
4080 { 15250, 14125, },
4081 { 15375, 14250, },
4082 { 15500, 14375, },
4083 { 15625, 14500, },
4084 { 15750, 14625, },
4085 { 15875, 14750, },
4086 { 16000, 14875, },
4087 { 16125, 15000, },
4088 };
3d13ef2e 4089 if (INTEL_INFO(dev)->is_mobile)
eb48eb00
DV		 4090 return v_table[pxvid].vm;
4091 else
4092 return v_table[pxvid].vd;
4093}
4094
02d71956 4095static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 4096{
4097 struct timespec now, diff1;
4098 u64 diff;
4099 unsigned long diffms;
4100 u32 count;
4101
02d71956 4102 assert_spin_locked(&mchdev_lock);
eb48eb00
DV		 4103
4104 getrawmonotonic(&now);
20e4d407 4105 diff1 = timespec_sub(now, dev_priv->ips.last_time2);
eb48eb00
DV		 4106
4107 /* Don't divide by 0 */
4108 diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
4109 if (!diffms)
4110 return;
4111
4112 count = I915_READ(GFXEC);
4113
20e4d407
DV		 4114 if (count < dev_priv->ips.last_count2) {
4115 diff = ~0UL - dev_priv->ips.last_count2;
eb48eb00
DV		 4116 diff += count;
4117 } else {
20e4d407 4118 diff = count - dev_priv->ips.last_count2;
eb48eb00
DV		 4119 }
4120
20e4d407
DV		 4121 dev_priv->ips.last_count2 = count;
4122 dev_priv->ips.last_time2 = now;
eb48eb00
DV		 4123
4124 /* More magic constants... */
4125 diff = diff * 1181;
4126 diff = div_u64(diff, diffms * 10);
20e4d407 4127 dev_priv->ips.gfx_power = diff;
eb48eb00
DV		 4128}
4129
02d71956
DV		 4130void i915_update_gfx_val(struct drm_i915_private *dev_priv)
4131{
3d13ef2e
DL		 4132 struct drm_device *dev = dev_priv->dev;
4133
4134 if (INTEL_INFO(dev)->gen != 5)
02d71956
DV		 4135 return;
4136
9270388e 4137 spin_lock_irq(&mchdev_lock);
02d71956
DV		 4138
4139 __i915_update_gfx_val(dev_priv);
4140
9270388e 4141 spin_unlock_irq(&mchdev_lock);
02d71956
DV		 4142}
4143
f531dcb2 4144static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 4145{
4146 unsigned long t, corr, state1, corr2, state2;
4147 u32 pxvid, ext_v;
4148
02d71956
DV		 4149 assert_spin_locked(&mchdev_lock);
4150
b39fb297 4151 pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_freq * 4));
eb48eb00
DV		 4152 pxvid = (pxvid >> 24) & 0x7f;
4153 ext_v = pvid_to_extvid(dev_priv, pxvid);
4154
4155 state1 = ext_v;
4156
4157 t = i915_mch_val(dev_priv);
4158
4159 /* Revel in the empirically derived constants */
4160
4161 /* Correction factor in 1/100000 units */
4162 if (t > 80)
4163 corr = ((t * 2349) + 135940);
4164 else if (t >= 50)
4165 corr = ((t * 964) + 29317);
4166 else /* < 50 */
4167 corr = ((t * 301) + 1004);
4168
4169 corr = corr * ((150142 * state1) / 10000 - 78642);
4170 corr /= 100000;
20e4d407 4171 corr2 = (corr * dev_priv->ips.corr);
eb48eb00
DV		 4172
4173 state2 = (corr2 * state1) / 10000;
4174 state2 /= 100; /* convert to mW */
4175
02d71956 4176 __i915_update_gfx_val(dev_priv);
eb48eb00 4177
20e4d407 4178 return dev_priv->ips.gfx_power + state2;
eb48eb00
DV		 4179}
4180
f531dcb2
CW		 4181unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
4182{
3d13ef2e 4183 struct drm_device *dev = dev_priv->dev;
f531dcb2
CW		 4184 unsigned long val;
4185
3d13ef2e 4186 if (INTEL_INFO(dev)->gen != 5)
f531dcb2
CW		 4187 return 0;
4188
4189 spin_lock_irq(&mchdev_lock);
4190
4191 val = __i915_gfx_val(dev_priv);
4192
4193 spin_unlock_irq(&mchdev_lock);
4194
4195 return val;
4196}
4197
eb48eb00
DV		 4198/**
4199 * i915_read_mch_val - return value for IPS use
4200 *
4201 * Calculate and return a value for the IPS driver to use when deciding whether
4202 * we have thermal and power headroom to increase CPU or GPU power budget.
4203 */
4204unsigned long i915_read_mch_val(void)
4205{
4206 struct drm_i915_private *dev_priv;
4207 unsigned long chipset_val, graphics_val, ret = 0;
4208
9270388e 4209 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4210 if (!i915_mch_dev)
4211 goto out_unlock;
4212 dev_priv = i915_mch_dev;
4213
f531dcb2
CW		 4214 chipset_val = __i915_chipset_val(dev_priv);
4215 graphics_val = __i915_gfx_val(dev_priv);
eb48eb00
DV		 4216
4217 ret = chipset_val + graphics_val;
4218
4219out_unlock:
9270388e 4220 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4221
4222 return ret;
4223}
4224EXPORT_SYMBOL_GPL(i915_read_mch_val);
4225
4226/**
4227 * i915_gpu_raise - raise GPU frequency limit
4228 *
4229 * Raise the limit; IPS indicates we have thermal headroom.
4230 */
4231bool i915_gpu_raise(void)
4232{
4233 struct drm_i915_private *dev_priv;
4234 bool ret = true;
4235
9270388e 4236 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4237 if (!i915_mch_dev) {
4238 ret = false;
4239 goto out_unlock;
4240 }
4241 dev_priv = i915_mch_dev;
4242
20e4d407
DV		 4243 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
4244 dev_priv->ips.max_delay--;
eb48eb00
DV		 4245
4246out_unlock:
9270388e 4247 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4248
4249 return ret;
4250}
4251EXPORT_SYMBOL_GPL(i915_gpu_raise);
4252
4253/**
4254 * i915_gpu_lower - lower GPU frequency limit
4255 *
4256 * IPS indicates we're close to a thermal limit, so throttle back the GPU
4257 * frequency maximum.
4258 */
4259bool i915_gpu_lower(void)
4260{
4261 struct drm_i915_private *dev_priv;
4262 bool ret = true;
4263
9270388e 4264 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4265 if (!i915_mch_dev) {
4266 ret = false;
4267 goto out_unlock;
4268 }
4269 dev_priv = i915_mch_dev;
4270
20e4d407
DV		 4271 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
4272 dev_priv->ips.max_delay++;
eb48eb00
DV		 4273
4274out_unlock:
9270388e 4275 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4276
4277 return ret;
4278}
4279EXPORT_SYMBOL_GPL(i915_gpu_lower);
4280
4281/**
4282 * i915_gpu_busy - indicate GPU business to IPS
4283 *
4284 * Tell the IPS driver whether or not the GPU is busy.
4285 */
4286bool i915_gpu_busy(void)
4287{
4288 struct drm_i915_private *dev_priv;
f047e395 4289 struct intel_ring_buffer *ring;
eb48eb00 4290 bool ret = false;
f047e395 4291 int i;
eb48eb00 4292
9270388e 4293 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4294 if (!i915_mch_dev)
4295 goto out_unlock;
4296 dev_priv = i915_mch_dev;
4297
f047e395
CW		 4298 for_each_ring(ring, dev_priv, i)
4299 ret |= !list_empty(&ring->request_list);
eb48eb00
DV		 4300
4301out_unlock:
9270388e 4302 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4303
4304 return ret;
4305}
4306EXPORT_SYMBOL_GPL(i915_gpu_busy);
4307
4308/**
4309 * i915_gpu_turbo_disable - disable graphics turbo
4310 *
4311 * Disable graphics turbo by resetting the max frequency and setting the
4312 * current frequency to the default.
4313 */
4314bool i915_gpu_turbo_disable(void)
4315{
4316 struct drm_i915_private *dev_priv;
4317 bool ret = true;
4318
9270388e 4319 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4320 if (!i915_mch_dev) {
4321 ret = false;
4322 goto out_unlock;
4323 }
4324 dev_priv = i915_mch_dev;
4325
20e4d407 4326 dev_priv->ips.max_delay = dev_priv->ips.fstart;
eb48eb00 4327
20e4d407 4328 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
eb48eb00
DV		 4329 ret = false;
4330
4331out_unlock:
9270388e 4332 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4333
4334 return ret;
4335}
4336EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
4337
4338/**
4339 * Tells the intel_ips driver that the i915 driver is now loaded, if
4340 * IPS got loaded first.
4341 *
4342 * This awkward dance is so that neither module has to depend on the
4343 * other in order for IPS to do the appropriate communication of
4344 * GPU turbo limits to i915.
4345 */
4346static void
4347ips_ping_for_i915_load(void)
4348{
4349 void (*link)(void);
4350
4351 link = symbol_get(ips_link_to_i915_driver);
4352 if (link) {
4353 link();
4354 symbol_put(ips_link_to_i915_driver);
4355 }
4356}
4357
4358void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
4359{
02d71956
DV		 4360 /* We only register the i915 ips part with intel-ips once everything is
4361 * set up, to avoid intel-ips sneaking in and reading bogus values. */
9270388e 4362 spin_lock_irq(&mchdev_lock);
eb48eb00 4363 i915_mch_dev = dev_priv;
9270388e 4364 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4365
4366 ips_ping_for_i915_load();
4367}
4368
4369void intel_gpu_ips_teardown(void)
4370{
9270388e 4371 spin_lock_irq(&mchdev_lock);
eb48eb00 4372 i915_mch_dev = NULL;
9270388e 4373 spin_unlock_irq(&mchdev_lock);
eb48eb00 4374}
76c3552f 4375
8090c6b9 4376static void intel_init_emon(struct drm_device *dev)
dde18883
ED		 4377{
4378 struct drm_i915_private *dev_priv = dev->dev_private;
4379 u32 lcfuse;
4380 u8 pxw[16];
4381 int i;
4382
4383 /* Disable to program */
4384 I915_WRITE(ECR, 0);
4385 POSTING_READ(ECR);
4386
4387 /* Program energy weights for various events */
4388 I915_WRITE(SDEW, 0x15040d00);
4389 I915_WRITE(CSIEW0, 0x007f0000);
4390 I915_WRITE(CSIEW1, 0x1e220004);
4391 I915_WRITE(CSIEW2, 0x04000004);
4392
4393 for (i = 0; i < 5; i++)
4394 I915_WRITE(PEW + (i * 4), 0);
4395 for (i = 0; i < 3; i++)
4396 I915_WRITE(DEW + (i * 4), 0);
4397
4398 /* Program P-state weights to account for frequency power adjustment */
4399 for (i = 0; i < 16; i++) {
4400 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
4401 unsigned long freq = intel_pxfreq(pxvidfreq);
4402 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
4403 PXVFREQ_PX_SHIFT;
4404 unsigned long val;
4405
4406 val = vid * vid;
4407 val *= (freq / 1000);
4408 val *= 255;
4409 val /= (127*127*900);
4410 if (val > 0xff)
4411 DRM_ERROR("bad pxval: %ld\n", val);
4412 pxw[i] = val;
4413 }
4414 /* Render standby states get 0 weight */
4415 pxw[14] = 0;
4416 pxw[15] = 0;
4417
4418 for (i = 0; i < 4; i++) {
4419 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
4420 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
4421 I915_WRITE(PXW + (i * 4), val);
4422 }
4423
4424 /* Adjust magic regs to magic values (more experimental results) */
4425 I915_WRITE(OGW0, 0);
4426 I915_WRITE(OGW1, 0);
4427 I915_WRITE(EG0, 0x00007f00);
4428 I915_WRITE(EG1, 0x0000000e);
4429 I915_WRITE(EG2, 0x000e0000);
4430 I915_WRITE(EG3, 0x68000300);
4431 I915_WRITE(EG4, 0x42000000);
4432 I915_WRITE(EG5, 0x00140031);
4433 I915_WRITE(EG6, 0);
4434 I915_WRITE(EG7, 0);
4435
4436 for (i = 0; i < 8; i++)
4437 I915_WRITE(PXWL + (i * 4), 0);
4438
4439 /* Enable PMON + select events */
4440 I915_WRITE(ECR, 0x80000019);
4441
4442 lcfuse = I915_READ(LCFUSE02);
4443
20e4d407 4444 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
dde18883
ED		 4445}
4446
ae48434c
ID		 4447void intel_init_gt_powersave(struct drm_device *dev)
4448{
4449 if (IS_VALLEYVIEW(dev))
4450 valleyview_setup_pctx(dev);
4451}
4452
4453void intel_cleanup_gt_powersave(struct drm_device *dev)
4454{
4455 if (IS_VALLEYVIEW(dev))
4456 valleyview_cleanup_pctx(dev);
4457}
4458
8090c6b9
DV		 4459void intel_disable_gt_powersave(struct drm_device *dev)
4460{
1a01ab3b
JB		 4461 struct drm_i915_private *dev_priv = dev->dev_private;
4462
fd0c0642
DV		 4463 /* Interrupts should be disabled already to avoid re-arming. */
4464 WARN_ON(dev->irq_enabled);
4465
930ebb46 4466 if (IS_IRONLAKE_M(dev)) {
8090c6b9 4467 ironlake_disable_drps(dev);
930ebb46 4468 ironlake_disable_rc6(dev);
0a073b84 4469 } else if (INTEL_INFO(dev)->gen >= 6) {
1a01ab3b 4470 cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
250848ca 4471 cancel_work_sync(&dev_priv->rps.work);
4fc688ce 4472 mutex_lock(&dev_priv->rps.hw_lock);
d20d4f0c
JB		 4473 if (IS_VALLEYVIEW(dev))
4474 valleyview_disable_rps(dev);
4475 else
4476 gen6_disable_rps(dev);
c0951f0c 4477 dev_priv->rps.enabled = false;
4fc688ce 4478 mutex_unlock(&dev_priv->rps.hw_lock);
930ebb46 4479 }
8090c6b9
DV		 4480}
4481
1a01ab3b
JB		 4482static void intel_gen6_powersave_work(struct work_struct *work)
4483{
4484 struct drm_i915_private *dev_priv =
4485 container_of(work, struct drm_i915_private,
4486 rps.delayed_resume_work.work);
4487 struct drm_device *dev = dev_priv->dev;
4488
4fc688ce 4489 mutex_lock(&dev_priv->rps.hw_lock);
0a073b84
JB		 4490
4491 if (IS_VALLEYVIEW(dev)) {
4492 valleyview_enable_rps(dev);
6edee7f3
BW		 4493 } else if (IS_BROADWELL(dev)) {
4494 gen8_enable_rps(dev);
4495 gen6_update_ring_freq(dev);
0a073b84
JB		 4496 } else {
4497 gen6_enable_rps(dev);
4498 gen6_update_ring_freq(dev);
4499 }
c0951f0c 4500 dev_priv->rps.enabled = true;
4fc688ce 4501 mutex_unlock(&dev_priv->rps.hw_lock);
1a01ab3b
JB		 4502}
4503
8090c6b9
DV		 4504void intel_enable_gt_powersave(struct drm_device *dev)
4505{
1a01ab3b
JB		 4506 struct drm_i915_private *dev_priv = dev->dev_private;
4507
8090c6b9
DV		 4508 if (IS_IRONLAKE_M(dev)) {
4509 ironlake_enable_drps(dev);
4510 ironlake_enable_rc6(dev);
4511 intel_init_emon(dev);
0a073b84 4512 } else if (IS_GEN6(dev) || IS_GEN7(dev)) {
1a01ab3b
JB		 4513 /*
4514 * PCU communication is slow and this doesn't need to be
4515 * done at any specific time, so do this out of our fast path
4516 * to make resume and init faster.
4517 */
4518 schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
4519 round_jiffies_up_relative(HZ));
8090c6b9
DV		 4520 }
4521}
4522
3107bd48
DV		 4523static void ibx_init_clock_gating(struct drm_device *dev)
4524{
4525 struct drm_i915_private *dev_priv = dev->dev_private;
4526
4527 /*
4528 * On Ibex Peak and Cougar Point, we need to disable clock
4529 * gating for the panel power sequencer or it will fail to
4530 * start up when no ports are active.
4531 */
4532 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
4533}
4534
0e088b8f
VS		 4535static void g4x_disable_trickle_feed(struct drm_device *dev)
4536{
4537 struct drm_i915_private *dev_priv = dev->dev_private;
4538 int pipe;
4539
4540 for_each_pipe(pipe) {
4541 I915_WRITE(DSPCNTR(pipe),
4542 I915_READ(DSPCNTR(pipe)) |
4543 DISPPLANE_TRICKLE_FEED_DISABLE);
1dba99f4 4544 intel_flush_primary_plane(dev_priv, pipe);
0e088b8f
VS		 4545 }
4546}
4547
017636cc
VS		 4548static void ilk_init_lp_watermarks(struct drm_device *dev)
4549{
4550 struct drm_i915_private *dev_priv = dev->dev_private;
4551
4552 I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
4553 I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
4554 I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
4555
4556 /*
4557 * Don't touch WM1S_LP_EN here.
4558 * Doing so could cause underruns.
4559 */
4560}
4561
1fa61106 4562static void ironlake_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 4563{
4564 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 4565 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 4566
f1e8fa56
DL		 4567 /*
4568 * Required for FBC
4569 * WaFbcDisableDpfcClockGating:ilk
4570 */
4d47e4f5
DL		 4571 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
4572 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
4573 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED		 4574
4575 I915_WRITE(PCH_3DCGDIS0,
4576 MARIUNIT_CLOCK_GATE_DISABLE |
4577 SVSMUNIT_CLOCK_GATE_DISABLE);
4578 I915_WRITE(PCH_3DCGDIS1,
4579 VFMUNIT_CLOCK_GATE_DISABLE);
4580
6f1d69b0
ED		 4581 /*
4582 * According to the spec the following bits should be set in
4583 * order to enable memory self-refresh
4584 * The bit 22/21 of 0x42004
4585 * The bit 5 of 0x42020
4586 * The bit 15 of 0x45000
4587 */
4588 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4589 (I915_READ(ILK_DISPLAY_CHICKEN2) |
4590 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4d47e4f5 4591 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED		 4592 I915_WRITE(DISP_ARB_CTL,
4593 (I915_READ(DISP_ARB_CTL) |
4594 DISP_FBC_WM_DIS));
017636cc
VS		 4595
4596 ilk_init_lp_watermarks(dev);
6f1d69b0
ED		 4597
4598 /*
4599 * Based on the document from hardware guys the following bits
4600 * should be set unconditionally in order to enable FBC.
4601 * The bit 22 of 0x42000
4602 * The bit 22 of 0x42004
4603 * The bit 7,8,9 of 0x42020.
4604 */
4605 if (IS_IRONLAKE_M(dev)) {
4bb35334 4606 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6f1d69b0
ED		 4607 I915_WRITE(ILK_DISPLAY_CHICKEN1,
4608 I915_READ(ILK_DISPLAY_CHICKEN1) |
4609 ILK_FBCQ_DIS);
4610 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4611 I915_READ(ILK_DISPLAY_CHICKEN2) |
4612 ILK_DPARB_GATE);
6f1d69b0
ED		 4613 }
4614
4d47e4f5
DL		 4615 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
4616
6f1d69b0
ED		 4617 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4618 I915_READ(ILK_DISPLAY_CHICKEN2) |
4619 ILK_ELPIN_409_SELECT);
4620 I915_WRITE(_3D_CHICKEN2,
4621 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
4622 _3D_CHICKEN2_WM_READ_PIPELINED);
4358a374 4623
ecdb4eb7 4624 /* WaDisableRenderCachePipelinedFlush:ilk */
4358a374
DV		 4625 I915_WRITE(CACHE_MODE_0,
4626 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3107bd48 4627
0e088b8f 4628 g4x_disable_trickle_feed(dev);
bdad2b2f 4629
3107bd48
DV		 4630 ibx_init_clock_gating(dev);
4631}
4632
4633static void cpt_init_clock_gating(struct drm_device *dev)
4634{
4635 struct drm_i915_private *dev_priv = dev->dev_private;
4636 int pipe;
3f704fa2 4637 uint32_t val;
3107bd48
DV		 4638
4639 /*
4640 * On Ibex Peak and Cougar Point, we need to disable clock
4641 * gating for the panel power sequencer or it will fail to
4642 * start up when no ports are active.
4643 */
cd664078
JB		 4644 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
4645 PCH_DPLUNIT_CLOCK_GATE_DISABLE |
4646 PCH_CPUNIT_CLOCK_GATE_DISABLE);
3107bd48
DV		 4647 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
4648 DPLS_EDP_PPS_FIX_DIS);
335c07b7
TI		 4649 /* The below fixes the weird display corruption, a few pixels shifted
4650 * downward, on (only) LVDS of some HP laptops with IVY.
4651 */
3f704fa2 4652 for_each_pipe(pipe) {
dc4bd2d1
PZ		 4653 val = I915_READ(TRANS_CHICKEN2(pipe));
4654 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
4655 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
41aa3448 4656 if (dev_priv->vbt.fdi_rx_polarity_inverted)
3f704fa2 4657 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
dc4bd2d1
PZ		 4658 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
4659 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
4660 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3f704fa2
PZ		 4661 I915_WRITE(TRANS_CHICKEN2(pipe), val);
4662 }
3107bd48
DV		 4663 /* WADP0ClockGatingDisable */
4664 for_each_pipe(pipe) {
4665 I915_WRITE(TRANS_CHICKEN1(pipe),
4666 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
4667 }
6f1d69b0
ED		 4668}
4669
1d7aaa0c
DV		 4670static void gen6_check_mch_setup(struct drm_device *dev)
4671{
4672 struct drm_i915_private *dev_priv = dev->dev_private;
4673 uint32_t tmp;
4674
4675 tmp = I915_READ(MCH_SSKPD);
4676 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
4677 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
4678 DRM_INFO("This can cause pipe underruns and display issues.\n");
4679 DRM_INFO("Please upgrade your BIOS to fix this.\n");
4680 }
4681}
4682
1fa61106 4683static void gen6_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 4684{
4685 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 4686 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 4687
231e54f6 4688 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6f1d69b0
ED		 4689
4690 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4691 I915_READ(ILK_DISPLAY_CHICKEN2) |
4692 ILK_ELPIN_409_SELECT);
4693
ecdb4eb7 4694 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4283908e
DV		 4695 I915_WRITE(_3D_CHICKEN,
4696 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
4697
ecdb4eb7 4698 /* WaSetupGtModeTdRowDispatch:snb */
6547fbdb
DV		 4699 if (IS_SNB_GT1(dev))
4700 I915_WRITE(GEN6_GT_MODE,
4701 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
4702
8d85d272
VS		 4703 /*
4704 * BSpec recoomends 8x4 when MSAA is used,
4705 * however in practice 16x4 seems fastest.
c5c98a58
VS		 4706 *
4707 * Note that PS/WM thread counts depend on the WIZ hashing
4708 * disable bit, which we don't touch here, but it's good
4709 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
8d85d272
VS		 4710 */
4711 I915_WRITE(GEN6_GT_MODE,
4712 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
4713
017636cc 4714 ilk_init_lp_watermarks(dev);
6f1d69b0 4715
6f1d69b0 4716 I915_WRITE(CACHE_MODE_0,
50743298 4717 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
6f1d69b0
ED		 4718
4719 I915_WRITE(GEN6_UCGCTL1,
4720 I915_READ(GEN6_UCGCTL1) |
4721 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
4722 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
4723
4724 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4725 * gating disable must be set. Failure to set it results in
4726 * flickering pixels due to Z write ordering failures after
4727 * some amount of runtime in the Mesa "fire" demo, and Unigine
4728 * Sanctuary and Tropics, and apparently anything else with
4729 * alpha test or pixel discard.
4730 *
4731 * According to the spec, bit 11 (RCCUNIT) must also be set,
4732 * but we didn't debug actual testcases to find it out.
0f846f81 4733 *
ef59318c
VS		 4734 * WaDisableRCCUnitClockGating:snb
4735 * WaDisableRCPBUnitClockGating:snb
6f1d69b0
ED		 4736 */
4737 I915_WRITE(GEN6_UCGCTL2,
4738 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
4739 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4740
5eb146dd 4741 /* WaStripsFansDisableFastClipPerformanceFix:snb */
743b57d8
VS		 4742 I915_WRITE(_3D_CHICKEN3,
4743 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL));
6f1d69b0 4744
e927ecde
VS		 4745 /*
4746 * Bspec says:
4747 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
4748 * 3DSTATE_SF number of SF output attributes is more than 16."
4749 */
4750 I915_WRITE(_3D_CHICKEN3,
4751 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH));
4752
6f1d69b0
ED		 4753 /*
4754 * According to the spec the following bits should be
4755 * set in order to enable memory self-refresh and fbc:
4756 * The bit21 and bit22 of 0x42000
4757 * The bit21 and bit22 of 0x42004
4758 * The bit5 and bit7 of 0x42020
4759 * The bit14 of 0x70180
4760 * The bit14 of 0x71180
4bb35334
DL		 4761 *
4762 * WaFbcAsynchFlipDisableFbcQueue:snb
6f1d69b0
ED		 4763 */
4764 I915_WRITE(ILK_DISPLAY_CHICKEN1,
4765 I915_READ(ILK_DISPLAY_CHICKEN1) |
4766 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
4767 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4768 I915_READ(ILK_DISPLAY_CHICKEN2) |
4769 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
231e54f6
DL		 4770 I915_WRITE(ILK_DSPCLK_GATE_D,
4771 I915_READ(ILK_DSPCLK_GATE_D) |
4772 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
4773 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
6f1d69b0 4774
0e088b8f 4775 g4x_disable_trickle_feed(dev);
f8f2ac9a 4776
3107bd48 4777 cpt_init_clock_gating(dev);
1d7aaa0c
DV		 4778
4779 gen6_check_mch_setup(dev);
6f1d69b0
ED		 4780}
4781
4782static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
4783{
4784 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
4785
3aad9059 4786 /*
46680e0a 4787 * WaVSThreadDispatchOverride:ivb,vlv
3aad9059
VS		 4788 *
4789 * This actually overrides the dispatch
4790 * mode for all thread types.
4791 */
6f1d69b0
ED		 4792 reg &= ~GEN7_FF_SCHED_MASK;
4793 reg |= GEN7_FF_TS_SCHED_HW;
4794 reg |= GEN7_FF_VS_SCHED_HW;
4795 reg |= GEN7_FF_DS_SCHED_HW;
4796
4797 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
4798}
4799
17a303ec
PZ		 4800static void lpt_init_clock_gating(struct drm_device *dev)
4801{
4802 struct drm_i915_private *dev_priv = dev->dev_private;
4803
4804 /*
4805 * TODO: this bit should only be enabled when really needed, then
4806 * disabled when not needed anymore in order to save power.
4807 */
4808 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
4809 I915_WRITE(SOUTH_DSPCLK_GATE_D,
4810 I915_READ(SOUTH_DSPCLK_GATE_D) |
4811 PCH_LP_PARTITION_LEVEL_DISABLE);
0a790cdb
PZ		 4812
4813 /* WADPOClockGatingDisable:hsw */
4814 I915_WRITE(_TRANSA_CHICKEN1,
4815 I915_READ(_TRANSA_CHICKEN1) |
4816 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
17a303ec
PZ		 4817}
4818
7d708ee4
ID		 4819static void lpt_suspend_hw(struct drm_device *dev)
4820{
4821 struct drm_i915_private *dev_priv = dev->dev_private;
4822
4823 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) {
4824 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
4825
4826 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
4827 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
4828 }
4829}
4830
1020a5c2
BW		 4831static void gen8_init_clock_gating(struct drm_device *dev)
4832{
4833 struct drm_i915_private *dev_priv = dev->dev_private;
07d27e20 4834 enum pipe pipe;
1020a5c2
BW		 4835
4836 I915_WRITE(WM3_LP_ILK, 0);
4837 I915_WRITE(WM2_LP_ILK, 0);
4838 I915_WRITE(WM1_LP_ILK, 0);
50ed5fbd
BW		 4839
4840 /* FIXME(BDW): Check all the w/a, some might only apply to
4841 * pre-production hw. */
4842
c8966e10
KG		 4843 /* WaDisablePartialInstShootdown:bdw */
4844 I915_WRITE(GEN8_ROW_CHICKEN,
4845 _MASKED_BIT_ENABLE(PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE));
4846
1411e6a5
KG		 4847 /* WaDisableThreadStallDopClockGating:bdw */
4848 /* FIXME: Unclear whether we really need this on production bdw. */
4849 I915_WRITE(GEN8_ROW_CHICKEN,
4850 _MASKED_BIT_ENABLE(STALL_DOP_GATING_DISABLE));
4851
4167e32c
DL		 4852 /*
4853 * This GEN8_CENTROID_PIXEL_OPT_DIS W/A is only needed for
4854 * pre-production hardware
4855 */
fd392b60
BW		 4856 I915_WRITE(HALF_SLICE_CHICKEN3,
4857 _MASKED_BIT_ENABLE(GEN8_CENTROID_PIXEL_OPT_DIS));
bf66347c
BW		 4858 I915_WRITE(HALF_SLICE_CHICKEN3,
4859 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS));
4afe8d33
BW		 4860 I915_WRITE(GAMTARBMODE, _MASKED_BIT_ENABLE(ARB_MODE_BWGTLB_DISABLE));
4861
7f88da0c
BW		 4862 I915_WRITE(_3D_CHICKEN3,
4863 _3D_CHICKEN_SDE_LIMIT_FIFO_POLY_DEPTH(2));
4864
a75f3628
BW		 4865 I915_WRITE(COMMON_SLICE_CHICKEN2,
4866 _MASKED_BIT_ENABLE(GEN8_CSC2_SBE_VUE_CACHE_CONSERVATIVE));
4867
4c2e7a5f
BW		 4868 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4869 _MASKED_BIT_ENABLE(GEN7_SINGLE_SUBSCAN_DISPATCH_ENABLE));
4870
ab57fff1 4871 /* WaSwitchSolVfFArbitrationPriority:bdw */
50ed5fbd 4872 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
fe4ab3ce 4873
ab57fff1 4874 /* WaPsrDPAMaskVBlankInSRD:bdw */
fe4ab3ce
BW		 4875 I915_WRITE(CHICKEN_PAR1_1,
4876 I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
4877
ab57fff1 4878 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
07d27e20
DL		 4879 for_each_pipe(pipe) {
4880 I915_WRITE(CHICKEN_PIPESL_1(pipe),
c7c65622 4881 I915_READ(CHICKEN_PIPESL_1(pipe)) |
8f670bb1 4882 BDW_DPRS_MASK_VBLANK_SRD);
fe4ab3ce 4883 }
63801f21
BW		 4884
4885 /* Use Force Non-Coherent whenever executing a 3D context. This is a
4886 * workaround for for a possible hang in the unlikely event a TLB
4887 * invalidation occurs during a PSD flush.
4888 */
4889 I915_WRITE(HDC_CHICKEN0,
4890 I915_READ(HDC_CHICKEN0) |
4891 _MASKED_BIT_ENABLE(HDC_FORCE_NON_COHERENT));
ab57fff1
BW		 4892
4893 /* WaVSRefCountFullforceMissDisable:bdw */
4894 /* WaDSRefCountFullforceMissDisable:bdw */
4895 I915_WRITE(GEN7_FF_THREAD_MODE,
4896 I915_READ(GEN7_FF_THREAD_MODE) &
4897 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
36075a4c
VS		 4898
4899 /*
4900 * BSpec recommends 8x4 when MSAA is used,
4901 * however in practice 16x4 seems fastest.
c5c98a58
VS		 4902 *
4903 * Note that PS/WM thread counts depend on the WIZ hashing
4904 * disable bit, which we don't touch here, but it's good
4905 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
36075a4c
VS		 4906 */
4907 I915_WRITE(GEN7_GT_MODE,
4908 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
295e8bb7
VS		 4909
4910 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
4911 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
4f1ca9e9
VS		 4912
4913 /* WaDisableSDEUnitClockGating:bdw */
4914 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
4915 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
5d708680
DL		 4916
4917 /* Wa4x4STCOptimizationDisable:bdw */
4918 I915_WRITE(CACHE_MODE_1,
4919 _MASKED_BIT_ENABLE(GEN8_4x4_STC_OPTIMIZATION_DISABLE));
1020a5c2
BW		 4920}
4921
cad2a2d7
ED		 4922static void haswell_init_clock_gating(struct drm_device *dev)
4923{
4924 struct drm_i915_private *dev_priv = dev->dev_private;
cad2a2d7 4925
017636cc 4926 ilk_init_lp_watermarks(dev);
cad2a2d7 4927
f3fc4884
FJ		 4928 /* L3 caching of data atomics doesn't work -- disable it. */
4929 I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
4930 I915_WRITE(HSW_ROW_CHICKEN3,
4931 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
4932
ecdb4eb7 4933 /* This is required by WaCatErrorRejectionIssue:hsw */
cad2a2d7
ED		 4934 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4935 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4936 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4937
e36ea7ff
VS		 4938 /* WaVSRefCountFullforceMissDisable:hsw */
4939 I915_WRITE(GEN7_FF_THREAD_MODE,
4940 I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);
cad2a2d7 4941
fe27c606
CW		 4942 /* enable HiZ Raw Stall Optimization */
4943 I915_WRITE(CACHE_MODE_0_GEN7,
4944 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
4945
ecdb4eb7 4946 /* WaDisable4x2SubspanOptimization:hsw */
cad2a2d7
ED		 4947 I915_WRITE(CACHE_MODE_1,
4948 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
1544d9d5 4949
a12c4967
VS		 4950 /*
4951 * BSpec recommends 8x4 when MSAA is used,
4952 * however in practice 16x4 seems fastest.
c5c98a58
VS		 4953 *
4954 * Note that PS/WM thread counts depend on the WIZ hashing
4955 * disable bit, which we don't touch here, but it's good
4956 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a12c4967
VS		 4957 */
4958 I915_WRITE(GEN7_GT_MODE,
4959 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
4960
ecdb4eb7 4961 /* WaSwitchSolVfFArbitrationPriority:hsw */
e3dff585
BW		 4962 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
4963
90a88643
PZ		 4964 /* WaRsPkgCStateDisplayPMReq:hsw */
4965 I915_WRITE(CHICKEN_PAR1_1,
4966 I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
1544d9d5 4967
17a303ec 4968 lpt_init_clock_gating(dev);
cad2a2d7
ED		 4969}
4970
1fa61106 4971static void ivybridge_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 4972{
4973 struct drm_i915_private *dev_priv = dev->dev_private;
20848223 4974 uint32_t snpcr;
6f1d69b0 4975
017636cc 4976 ilk_init_lp_watermarks(dev);
6f1d69b0 4977
231e54f6 4978 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 4979
ecdb4eb7 4980 /* WaDisableEarlyCull:ivb */
87f8020e
JB		 4981 I915_WRITE(_3D_CHICKEN3,
4982 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4983
ecdb4eb7 4984 /* WaDisableBackToBackFlipFix:ivb */
6f1d69b0
ED		 4985 I915_WRITE(IVB_CHICKEN3,
4986 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4987 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4988
ecdb4eb7 4989 /* WaDisablePSDDualDispatchEnable:ivb */
12f3382b
JB		 4990 if (IS_IVB_GT1(dev))
4991 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4992 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 4993
ecdb4eb7 4994 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6f1d69b0
ED		 4995 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4996 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4997
ecdb4eb7 4998 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6f1d69b0
ED		 4999 I915_WRITE(GEN7_L3CNTLREG1,
5000 GEN7_WA_FOR_GEN7_L3_CONTROL);
5001 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
8ab43976
JB		 5002 GEN7_WA_L3_CHICKEN_MODE);
5003 if (IS_IVB_GT1(dev))
5004 I915_WRITE(GEN7_ROW_CHICKEN2,
5005 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2
VS		 5006 else {
5007 /* must write both registers */
5008 I915_WRITE(GEN7_ROW_CHICKEN2,
5009 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
8ab43976
JB		 5010 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
5011 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2 5012 }
6f1d69b0 5013
ecdb4eb7 5014 /* WaForceL3Serialization:ivb */
61939d97
JB		 5015 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5016 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5017
1b80a19a 5018 /*
0f846f81 5019 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 5020 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
0f846f81
JB		 5021 */
5022 I915_WRITE(GEN6_UCGCTL2,
28acf3b2 5023 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 5024
ecdb4eb7 5025 /* This is required by WaCatErrorRejectionIssue:ivb */
6f1d69b0
ED		 5026 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5027 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5028 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5029
0e088b8f 5030 g4x_disable_trickle_feed(dev);
6f1d69b0
ED		 5031
5032 gen7_setup_fixed_func_scheduler(dev_priv);
97e1930f 5033
22721343
CW		 5034 if (0) { /* causes HiZ corruption on ivb:gt1 */
5035 /* enable HiZ Raw Stall Optimization */
5036 I915_WRITE(CACHE_MODE_0_GEN7,
5037 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
5038 }
116f2b6d 5039
ecdb4eb7 5040 /* WaDisable4x2SubspanOptimization:ivb */
97e1930f
DV		 5041 I915_WRITE(CACHE_MODE_1,
5042 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
20848223 5043
a607c1a4
VS		 5044 /*
5045 * BSpec recommends 8x4 when MSAA is used,
5046 * however in practice 16x4 seems fastest.
c5c98a58
VS		 5047 *
5048 * Note that PS/WM thread counts depend on the WIZ hashing
5049 * disable bit, which we don't touch here, but it's good
5050 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a607c1a4
VS		 5051 */
5052 I915_WRITE(GEN7_GT_MODE,
5053 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
5054
20848223
BW		 5055 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
5056 snpcr &= ~GEN6_MBC_SNPCR_MASK;
5057 snpcr |= GEN6_MBC_SNPCR_MED;
5058 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
3107bd48 5059
ab5c608b
BW		 5060 if (!HAS_PCH_NOP(dev))
5061 cpt_init_clock_gating(dev);
1d7aaa0c
DV		 5062
5063 gen6_check_mch_setup(dev);
6f1d69b0
ED		 5064}
5065
1fa61106 5066static void valleyview_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5067{
5068 struct drm_i915_private *dev_priv = dev->dev_private;
85b1d7b3
JB		 5069 u32 val;
5070
5071 mutex_lock(&dev_priv->rps.hw_lock);
5072 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5073 mutex_unlock(&dev_priv->rps.hw_lock);
5074 switch ((val >> 6) & 3) {
5075 case 0:
f64a28a7 5076 case 1:
f6d51948 5077 dev_priv->mem_freq = 800;
85b1d7b3 5078 break;
f64a28a7 5079 case 2:
f6d51948 5080 dev_priv->mem_freq = 1066;
85b1d7b3 5081 break;
f64a28a7 5082 case 3:
2325991e 5083 dev_priv->mem_freq = 1333;
f64a28a7 5084 break;
85b1d7b3
JB		 5085 }
5086 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
6f1d69b0 5087
d60c4473
ID		 5088 dev_priv->vlv_cdclk_freq = valleyview_cur_cdclk(dev_priv);
5089 DRM_DEBUG_DRIVER("Current CD clock rate: %d MHz",
5090 dev_priv->vlv_cdclk_freq);
5091
d7fe0cc0 5092 I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 5093
ecdb4eb7 5094 /* WaDisableEarlyCull:vlv */
87f8020e
JB		 5095 I915_WRITE(_3D_CHICKEN3,
5096 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5097
ecdb4eb7 5098 /* WaDisableBackToBackFlipFix:vlv */
6f1d69b0
ED		 5099 I915_WRITE(IVB_CHICKEN3,
5100 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5101 CHICKEN3_DGMG_DONE_FIX_DISABLE);
5102
fad7d36e 5103 /* WaPsdDispatchEnable:vlv */
ecdb4eb7 5104 /* WaDisablePSDDualDispatchEnable:vlv */
12f3382b 5105 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
d3bc0303
JB		 5106 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
5107 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 5108
ecdb4eb7 5109 /* WaForceL3Serialization:vlv */
61939d97
JB		 5110 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5111 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5112
ecdb4eb7 5113 /* WaDisableDopClockGating:vlv */
8ab43976
JB		 5114 I915_WRITE(GEN7_ROW_CHICKEN2,
5115 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5116
ecdb4eb7 5117 /* This is required by WaCatErrorRejectionIssue:vlv */
6f1d69b0
ED		 5118 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5119 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5120 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5121
46680e0a
VS		 5122 gen7_setup_fixed_func_scheduler(dev_priv);
5123
3c0edaeb 5124 /*
0f846f81 5125 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 5126 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
0f846f81
JB		 5127 */
5128 I915_WRITE(GEN6_UCGCTL2,
3c0edaeb 5129 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 5130
c5c32cda 5131 /* WaDisableL3Bank2xClockGate:vlv */
e3f33d46
JB		 5132 I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
5133
e0d8d59b 5134 I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
6f1d69b0 5135
afd58e79
VS		 5136 /*
5137 * BSpec says this must be set, even though
5138 * WaDisable4x2SubspanOptimization isn't listed for VLV.
5139 */
6b26c86d
DV		 5140 I915_WRITE(CACHE_MODE_1,
5141 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
7983117f 5142
031994ee
VS		 5143 /*
5144 * WaIncreaseL3CreditsForVLVB0:vlv
5145 * This is the hardware default actually.
5146 */
5147 I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);
5148
2d809570 5149 /*
ecdb4eb7 5150 * WaDisableVLVClockGating_VBIIssue:vlv
2d809570
JB		 5151 * Disable clock gating on th GCFG unit to prevent a delay
5152 * in the reporting of vblank events.
5153 */
7a0d1eed 5154 I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
6f1d69b0
ED		 5155}
5156
1fa61106 5157static void g4x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5158{
5159 struct drm_i915_private *dev_priv = dev->dev_private;
5160 uint32_t dspclk_gate;
5161
5162 I915_WRITE(RENCLK_GATE_D1, 0);
5163 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
5164 GS_UNIT_CLOCK_GATE_DISABLE |
5165 CL_UNIT_CLOCK_GATE_DISABLE);
5166 I915_WRITE(RAMCLK_GATE_D, 0);
5167 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
5168 OVRUNIT_CLOCK_GATE_DISABLE |
5169 OVCUNIT_CLOCK_GATE_DISABLE;
5170 if (IS_GM45(dev))
5171 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
5172 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4358a374
DV		 5173
5174 /* WaDisableRenderCachePipelinedFlush */
5175 I915_WRITE(CACHE_MODE_0,
5176 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
de1aa629 5177
0e088b8f 5178 g4x_disable_trickle_feed(dev);
6f1d69b0
ED		 5179}
5180
1fa61106 5181static void crestline_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5182{
5183 struct drm_i915_private *dev_priv = dev->dev_private;
5184
5185 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
5186 I915_WRITE(RENCLK_GATE_D2, 0);
5187 I915_WRITE(DSPCLK_GATE_D, 0);
5188 I915_WRITE(RAMCLK_GATE_D, 0);
5189 I915_WRITE16(DEUC, 0);
20f94967
VS		 5190 I915_WRITE(MI_ARB_STATE,
5191 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED		 5192}
5193
1fa61106 5194static void broadwater_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5195{
5196 struct drm_i915_private *dev_priv = dev->dev_private;
5197
5198 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
5199 I965_RCC_CLOCK_GATE_DISABLE |
5200 I965_RCPB_CLOCK_GATE_DISABLE |
5201 I965_ISC_CLOCK_GATE_DISABLE |
5202 I965_FBC_CLOCK_GATE_DISABLE);
5203 I915_WRITE(RENCLK_GATE_D2, 0);
20f94967
VS		 5204 I915_WRITE(MI_ARB_STATE,
5205 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED		 5206}
5207
1fa61106 5208static void gen3_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5209{
5210 struct drm_i915_private *dev_priv = dev->dev_private;
5211 u32 dstate = I915_READ(D_STATE);
5212
5213 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
5214 DSTATE_DOT_CLOCK_GATING;
5215 I915_WRITE(D_STATE, dstate);
13a86b85
CW		 5216
5217 if (IS_PINEVIEW(dev))
5218 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
974a3b0f
DV		 5219
5220 /* IIR "flip pending" means done if this bit is set */
5221 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
6f1d69b0
ED		 5222}
5223
1fa61106 5224static void i85x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5225{
5226 struct drm_i915_private *dev_priv = dev->dev_private;
5227
5228 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
5229}
5230
1fa61106 5231static void i830_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5232{
5233 struct drm_i915_private *dev_priv = dev->dev_private;
5234
5235 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
5236}
5237
6f1d69b0
ED		 5238void intel_init_clock_gating(struct drm_device *dev)
5239{
5240 struct drm_i915_private *dev_priv = dev->dev_private;
5241
5242 dev_priv->display.init_clock_gating(dev);
6f1d69b0
ED		 5243}
5244
7d708ee4
ID		 5245void intel_suspend_hw(struct drm_device *dev)
5246{
5247 if (HAS_PCH_LPT(dev))
5248 lpt_suspend_hw(dev);
5249}
5250
c1ca727f
ID		 5251#define for_each_power_well(i, power_well, domain_mask, power_domains) \
5252 for (i = 0; \
5253 i < (power_domains)->power_well_count && \
5254 ((power_well) = &(power_domains)->power_wells[i]); \
5255 i++) \
5256 if ((power_well)->domains & (domain_mask))
5257
5258#define for_each_power_well_rev(i, power_well, domain_mask, power_domains) \
5259 for (i = (power_domains)->power_well_count - 1; \
5260 i >= 0 && ((power_well) = &(power_domains)->power_wells[i]);\
5261 i--) \
5262 if ((power_well)->domains & (domain_mask))
5263
15d199ea
PZ		 5264/**
5265 * We should only use the power well if we explicitly asked the hardware to
5266 * enable it, so check if it's enabled and also check if we've requested it to
5267 * be enabled.
5268 */
da7e29bd 5269static bool hsw_power_well_enabled(struct drm_i915_private *dev_priv,
c1ca727f
ID		 5270 struct i915_power_well *power_well)
5271{
c1ca727f
ID		 5272 return I915_READ(HSW_PWR_WELL_DRIVER) ==
5273 (HSW_PWR_WELL_ENABLE_REQUEST | HSW_PWR_WELL_STATE_ENABLED);
5274}
5275
da7e29bd 5276bool intel_display_power_enabled_sw(struct drm_i915_private *dev_priv,
ddf9c536
ID		 5277 enum intel_display_power_domain domain)
5278{
ddf9c536
ID		 5279 struct i915_power_domains *power_domains;
5280
5281 power_domains = &dev_priv->power_domains;
5282
5283 return power_domains->domain_use_count[domain];
5284}
5285
da7e29bd 5286bool intel_display_power_enabled(struct drm_i915_private *dev_priv,
b97186f0 5287 enum intel_display_power_domain domain)
15d199ea 5288{
c1ca727f
ID		 5289 struct i915_power_domains *power_domains;
5290 struct i915_power_well *power_well;
5291 bool is_enabled;
5292 int i;
15d199ea 5293
882244a3
PZ		 5294 if (dev_priv->pm.suspended)
5295 return false;
5296
c1ca727f
ID		 5297 power_domains = &dev_priv->power_domains;
5298
5299 is_enabled = true;
5300
5301 mutex_lock(&power_domains->lock);
5302 for_each_power_well_rev(i, power_well, BIT(domain), power_domains) {
6f3ef5dd
ID		 5303 if (power_well->always_on)
5304 continue;
5305
c6cb582e 5306 if (!power_well->ops->is_enabled(dev_priv, power_well)) {
c1ca727f
ID		 5307 is_enabled = false;
5308 break;
5309 }
5310 }
5311 mutex_unlock(&power_domains->lock);
5312
5313 return is_enabled;
15d199ea
PZ		 5314}
5315
93c73e8c
ID		 5316/*
5317 * Starting with Haswell, we have a "Power Down Well" that can be turned off
5318 * when not needed anymore. We have 4 registers that can request the power well
5319 * to be enabled, and it will only be disabled if none of the registers is
5320 * requesting it to be enabled.
5321 */
d5e8fdc8
PZ		 5322static void hsw_power_well_post_enable(struct drm_i915_private *dev_priv)
5323{
5324 struct drm_device *dev = dev_priv->dev;
5325 unsigned long irqflags;
5326
f9dcb0df
PZ		 5327 /*
5328 * After we re-enable the power well, if we touch VGA register 0x3d5
5329 * we'll get unclaimed register interrupts. This stops after we write
5330 * anything to the VGA MSR register. The vgacon module uses this
5331 * register all the time, so if we unbind our driver and, as a
5332 * consequence, bind vgacon, we'll get stuck in an infinite loop at
5333 * console_unlock(). So make here we touch the VGA MSR register, making
5334 * sure vgacon can keep working normally without triggering interrupts
5335 * and error messages.
5336 */
5337 vga_get_uninterruptible(dev->pdev, VGA_RSRC_LEGACY_IO);
5338 outb(inb(VGA_MSR_READ), VGA_MSR_WRITE);
5339 vga_put(dev->pdev, VGA_RSRC_LEGACY_IO);
5340
d5e8fdc8
PZ		 5341 if (IS_BROADWELL(dev)) {
5342 spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
5343 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_B),
5344 dev_priv->de_irq_mask[PIPE_B]);
5345 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_B),
5346 ~dev_priv->de_irq_mask[PIPE_B] |
5347 GEN8_PIPE_VBLANK);
5348 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_C),
5349 dev_priv->de_irq_mask[PIPE_C]);
5350 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_C),
5351 ~dev_priv->de_irq_mask[PIPE_C] |
5352 GEN8_PIPE_VBLANK);
5353 POSTING_READ(GEN8_DE_PIPE_IER(PIPE_C));
5354 spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
5355 }
5356}
5357
dd7c0b66
ID		 5358static void reset_vblank_counter(struct drm_device *dev, enum pipe pipe)
5359{
5360 assert_spin_locked(&dev->vbl_lock);
5361
5362 dev->vblank[pipe].last = 0;
5363}
5364
d5e8fdc8
PZ		 5365static void hsw_power_well_post_disable(struct drm_i915_private *dev_priv)
5366{
5367 struct drm_device *dev = dev_priv->dev;
07d27e20 5368 enum pipe pipe;
d5e8fdc8
PZ		 5369 unsigned long irqflags;
5370
5371 /*
5372 * After this, the registers on the pipes that are part of the power
5373 * well will become zero, so we have to adjust our counters according to
5374 * that.
5375 *
5376 * FIXME: Should we do this in general in drm_vblank_post_modeset?
5377 */
5378 spin_lock_irqsave(&dev->vbl_lock, irqflags);
07d27e20
DL		 5379 for_each_pipe(pipe)
5380 if (pipe != PIPE_A)
dd7c0b66 5381 reset_vblank_counter(dev, pipe);
d5e8fdc8
PZ		 5382 spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
5383}
5384
da7e29bd 5385static void hsw_set_power_well(struct drm_i915_private *dev_priv,
c1ca727f 5386 struct i915_power_well *power_well, bool enable)
d0d3e513 5387{
fa42e23c
PZ		 5388 bool is_enabled, enable_requested;
5389 uint32_t tmp;
d0d3e513 5390
fa42e23c 5391 tmp = I915_READ(HSW_PWR_WELL_DRIVER);
6aedd1f5
PZ		 5392 is_enabled = tmp & HSW_PWR_WELL_STATE_ENABLED;
5393 enable_requested = tmp & HSW_PWR_WELL_ENABLE_REQUEST;
d0d3e513 5394
fa42e23c
PZ		 5395 if (enable) {
5396 if (!enable_requested)
6aedd1f5
PZ		 5397 I915_WRITE(HSW_PWR_WELL_DRIVER,
5398 HSW_PWR_WELL_ENABLE_REQUEST);
d0d3e513 5399
fa42e23c
PZ		 5400 if (!is_enabled) {
5401 DRM_DEBUG_KMS("Enabling power well\n");
5402 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
6aedd1f5 5403 HSW_PWR_WELL_STATE_ENABLED), 20))
fa42e23c
PZ		 5404 DRM_ERROR("Timeout enabling power well\n");
5405 }
596cc11e 5406
d5e8fdc8 5407 hsw_power_well_post_enable(dev_priv);
fa42e23c
PZ		 5408 } else {
5409 if (enable_requested) {
5410 I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
9dbd8feb 5411 POSTING_READ(HSW_PWR_WELL_DRIVER);
fa42e23c 5412 DRM_DEBUG_KMS("Requesting to disable the power well\n");
9dbd8feb 5413
d5e8fdc8 5414 hsw_power_well_post_disable(dev_priv);
d0d3e513
ED		 5415 }
5416 }
fa42e23c 5417}
d0d3e513 5418
c6cb582e
ID		 5419static void hsw_power_well_sync_hw(struct drm_i915_private *dev_priv,
5420 struct i915_power_well *power_well)
5421{
5422 hsw_set_power_well(dev_priv, power_well, power_well->count > 0);
5423
5424 /*
5425 * We're taking over the BIOS, so clear any requests made by it since
5426 * the driver is in charge now.
5427 */
5428 if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE_REQUEST)
5429 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
5430}
5431
5432static void hsw_power_well_enable(struct drm_i915_private *dev_priv,
5433 struct i915_power_well *power_well)
5434{
c6cb582e
ID		 5435 hsw_set_power_well(dev_priv, power_well, true);
5436}
5437
5438static void hsw_power_well_disable(struct drm_i915_private *dev_priv,
5439 struct i915_power_well *power_well)
5440{
5441 hsw_set_power_well(dev_priv, power_well, false);
c6cb582e
ID		 5442}
5443
a45f4466
ID		 5444static void i9xx_always_on_power_well_noop(struct drm_i915_private *dev_priv,
5445 struct i915_power_well *power_well)
5446{
5447}
5448
5449static bool i9xx_always_on_power_well_enabled(struct drm_i915_private *dev_priv,
5450 struct i915_power_well *power_well)
5451{
5452 return true;
5453}
5454
77961eb9
ID		 5455static void vlv_set_power_well(struct drm_i915_private *dev_priv,
5456 struct i915_power_well *power_well, bool enable)
5457{
5458 enum punit_power_well power_well_id = power_well->data;
5459 u32 mask;
5460 u32 state;
5461 u32 ctrl;
5462
5463 mask = PUNIT_PWRGT_MASK(power_well_id);
5464 state = enable ? PUNIT_PWRGT_PWR_ON(power_well_id) :
5465 PUNIT_PWRGT_PWR_GATE(power_well_id);
5466
5467 mutex_lock(&dev_priv->rps.hw_lock);
5468
5469#define COND \
5470 ((vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_STATUS) & mask) == state)
5471
5472 if (COND)
5473 goto out;
5474
5475 ctrl = vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_CTRL);
5476 ctrl &= ~mask;
5477 ctrl |= state;
5478 vlv_punit_write(dev_priv, PUNIT_REG_PWRGT_CTRL, ctrl);
5479
5480 if (wait_for(COND, 100))
5481 DRM_ERROR("timout setting power well state %08x (%08x)\n",
5482 state,
5483 vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_CTRL));
5484
5485#undef COND
5486
5487out:
5488 mutex_unlock(&dev_priv->rps.hw_lock);
5489}
5490
5491static void vlv_power_well_sync_hw(struct drm_i915_private *dev_priv,
5492 struct i915_power_well *power_well)
5493{
5494 vlv_set_power_well(dev_priv, power_well, power_well->count > 0);
5495}
5496
5497static void vlv_power_well_enable(struct drm_i915_private *dev_priv,
5498 struct i915_power_well *power_well)
5499{
5500 vlv_set_power_well(dev_priv, power_well, true);
5501}
5502
5503static void vlv_power_well_disable(struct drm_i915_private *dev_priv,
5504 struct i915_power_well *power_well)
5505{
5506 vlv_set_power_well(dev_priv, power_well, false);
5507}
5508
5509static bool vlv_power_well_enabled(struct drm_i915_private *dev_priv,
5510 struct i915_power_well *power_well)
5511{
5512 int power_well_id = power_well->data;
5513 bool enabled = false;
5514 u32 mask;
5515 u32 state;
5516 u32 ctrl;
5517
5518 mask = PUNIT_PWRGT_MASK(power_well_id);
5519 ctrl = PUNIT_PWRGT_PWR_ON(power_well_id);
5520
5521 mutex_lock(&dev_priv->rps.hw_lock);
5522
5523 state = vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_STATUS) & mask;
5524 /*
5525 * We only ever set the power-on and power-gate states, anything
5526 * else is unexpected.
5527 */
5528 WARN_ON(state != PUNIT_PWRGT_PWR_ON(power_well_id) &&
5529 state != PUNIT_PWRGT_PWR_GATE(power_well_id));
5530 if (state == ctrl)
5531 enabled = true;
5532
5533 /*
5534 * A transient state at this point would mean some unexpected party
5535 * is poking at the power controls too.
5536 */
5537 ctrl = vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_CTRL) & mask;
5538 WARN_ON(ctrl != state);
5539
5540 mutex_unlock(&dev_priv->rps.hw_lock);
5541
5542 return enabled;
5543}
5544
5545static void vlv_display_power_well_enable(struct drm_i915_private *dev_priv,
5546 struct i915_power_well *power_well)
5547{
5548 WARN_ON_ONCE(power_well->data != PUNIT_POWER_WELL_DISP2D);
5549
5550 vlv_set_power_well(dev_priv, power_well, true);
5551
5552 spin_lock_irq(&dev_priv->irq_lock);
5553 valleyview_enable_display_irqs(dev_priv);
5554 spin_unlock_irq(&dev_priv->irq_lock);
5555
5556 /*
5557 * During driver initialization we need to defer enabling hotplug
5558 * processing until fbdev is set up.
5559 */
5560 if (dev_priv->enable_hotplug_processing)
5561 intel_hpd_init(dev_priv->dev);
5562
5563 i915_redisable_vga_power_on(dev_priv->dev);
5564}
5565
5566static void vlv_display_power_well_disable(struct drm_i915_private *dev_priv,
5567 struct i915_power_well *power_well)
5568{
5569 struct drm_device *dev = dev_priv->dev;
5570 enum pipe pipe;
5571
5572 WARN_ON_ONCE(power_well->data != PUNIT_POWER_WELL_DISP2D);
5573
5574 spin_lock_irq(&dev_priv->irq_lock);
5575 for_each_pipe(pipe)
5576 __intel_set_cpu_fifo_underrun_reporting(dev, pipe, false);
5577
5578 valleyview_disable_display_irqs(dev_priv);
5579 spin_unlock_irq(&dev_priv->irq_lock);
5580
5581 spin_lock_irq(&dev->vbl_lock);
5582 for_each_pipe(pipe)
5583 reset_vblank_counter(dev, pipe);
5584 spin_unlock_irq(&dev->vbl_lock);
5585
5586 vlv_set_power_well(dev_priv, power_well, false);
5587}
5588
25eaa003
ID		 5589static void check_power_well_state(struct drm_i915_private *dev_priv,
5590 struct i915_power_well *power_well)
5591{
5592 bool enabled = power_well->ops->is_enabled(dev_priv, power_well);
5593
5594 if (power_well->always_on || !i915.disable_power_well) {
5595 if (!enabled)
5596 goto mismatch;
5597
5598 return;
5599 }
5600
5601 if (enabled != (power_well->count > 0))
5602 goto mismatch;
5603
5604 return;
5605
5606mismatch:
5607 WARN(1, "state mismatch for '%s' (always_on %d hw state %d use-count %d disable_power_well %d\n",
5608 power_well->name, power_well->always_on, enabled,
5609 power_well->count, i915.disable_power_well);
5610}
5611
da7e29bd 5612void intel_display_power_get(struct drm_i915_private *dev_priv,
6765625e
VS		 5613 enum intel_display_power_domain domain)
5614{
83c00f55 5615 struct i915_power_domains *power_domains;
c1ca727f
ID		 5616 struct i915_power_well *power_well;
5617 int i;
6765625e 5618
9e6ea71a
PZ		 5619 intel_runtime_pm_get(dev_priv);
5620
83c00f55
ID		 5621 power_domains = &dev_priv->power_domains;
5622
5623 mutex_lock(&power_domains->lock);
1da51581 5624
25eaa003
ID		 5625 for_each_power_well(i, power_well, BIT(domain), power_domains) {
5626 if (!power_well->count++) {
5627 DRM_DEBUG_KMS("enabling %s\n", power_well->name);
c6cb582e 5628 power_well->ops->enable(dev_priv, power_well);
25eaa003
ID		 5629 }
5630
5631 check_power_well_state(dev_priv, power_well);
5632 }
1da51581 5633
ddf9c536
ID		 5634 power_domains->domain_use_count[domain]++;
5635
83c00f55 5636 mutex_unlock(&power_domains->lock);
6765625e
VS		 5637}
5638
da7e29bd 5639void intel_display_power_put(struct drm_i915_private *dev_priv,
6765625e
VS		 5640 enum intel_display_power_domain domain)
5641{
83c00f55 5642 struct i915_power_domains *power_domains;
c1ca727f
ID		 5643 struct i915_power_well *power_well;
5644 int i;
6765625e 5645
83c00f55
ID		 5646 power_domains = &dev_priv->power_domains;
5647
5648 mutex_lock(&power_domains->lock);
1da51581 5649
1da51581
ID		 5650 WARN_ON(!power_domains->domain_use_count[domain]);
5651 power_domains->domain_use_count[domain]--;
ddf9c536 5652
70bf407c
ID		 5653 for_each_power_well_rev(i, power_well, BIT(domain), power_domains) {
5654 WARN_ON(!power_well->count);
5655
25eaa003
ID		 5656 if (!--power_well->count && i915.disable_power_well) {
5657 DRM_DEBUG_KMS("disabling %s\n", power_well->name);
c6cb582e 5658 power_well->ops->disable(dev_priv, power_well);
25eaa003
ID		 5659 }
5660
5661 check_power_well_state(dev_priv, power_well);
70bf407c 5662 }
1da51581 5663
83c00f55 5664 mutex_unlock(&power_domains->lock);
9e6ea71a
PZ		 5665
5666 intel_runtime_pm_put(dev_priv);
6765625e
VS		 5667}
5668
83c00f55 5669static struct i915_power_domains *hsw_pwr;
a38911a3
WX		 5670
5671/* Display audio driver power well request */
5672void i915_request_power_well(void)
5673{
b4ed4484
ID		 5674 struct drm_i915_private *dev_priv;
5675
a38911a3
WX		 5676 if (WARN_ON(!hsw_pwr))
5677 return;
5678
b4ed4484
ID		 5679 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
5680 power_domains);
da7e29bd 5681 intel_display_power_get(dev_priv, POWER_DOMAIN_AUDIO);
a38911a3
WX		 5682}
5683EXPORT_SYMBOL_GPL(i915_request_power_well);
5684
5685/* Display audio driver power well release */
5686void i915_release_power_well(void)
5687{
b4ed4484
ID		 5688 struct drm_i915_private *dev_priv;
5689
a38911a3
WX		 5690 if (WARN_ON(!hsw_pwr))
5691 return;
5692
b4ed4484
ID		 5693 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
5694 power_domains);
da7e29bd 5695 intel_display_power_put(dev_priv, POWER_DOMAIN_AUDIO);
a38911a3
WX		 5696}
5697EXPORT_SYMBOL_GPL(i915_release_power_well);
5698
efcad917
ID		 5699#define POWER_DOMAIN_MASK (BIT(POWER_DOMAIN_NUM) - 1)
5700
5701#define HSW_ALWAYS_ON_POWER_DOMAINS ( \
5702 BIT(POWER_DOMAIN_PIPE_A) | \
f5938f36 5703 BIT(POWER_DOMAIN_TRANSCODER_EDP) | \
319be8ae
ID		 5704 BIT(POWER_DOMAIN_PORT_DDI_A_2_LANES) | \
5705 BIT(POWER_DOMAIN_PORT_DDI_A_4_LANES) | \
5706 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
5707 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
5708 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
5709 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
5710 BIT(POWER_DOMAIN_PORT_DDI_D_2_LANES) | \
5711 BIT(POWER_DOMAIN_PORT_DDI_D_4_LANES) | \
5712 BIT(POWER_DOMAIN_PORT_CRT) | \
f5938f36 5713 BIT(POWER_DOMAIN_INIT))
efcad917
ID		 5714#define HSW_DISPLAY_POWER_DOMAINS ( \
5715 (POWER_DOMAIN_MASK & ~HSW_ALWAYS_ON_POWER_DOMAINS) | \
5716 BIT(POWER_DOMAIN_INIT))
5717
5718#define BDW_ALWAYS_ON_POWER_DOMAINS ( \
5719 HSW_ALWAYS_ON_POWER_DOMAINS | \
5720 BIT(POWER_DOMAIN_PIPE_A_PANEL_FITTER))
5721#define BDW_DISPLAY_POWER_DOMAINS ( \
5722 (POWER_DOMAIN_MASK & ~BDW_ALWAYS_ON_POWER_DOMAINS) | \
5723 BIT(POWER_DOMAIN_INIT))
5724
77961eb9
ID		 5725#define VLV_ALWAYS_ON_POWER_DOMAINS BIT(POWER_DOMAIN_INIT)
5726#define VLV_DISPLAY_POWER_DOMAINS POWER_DOMAIN_MASK
5727
5728#define VLV_DPIO_CMN_BC_POWER_DOMAINS ( \
5729 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
5730 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
5731 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
5732 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
5733 BIT(POWER_DOMAIN_PORT_CRT) | \
5734 BIT(POWER_DOMAIN_INIT))
5735
5736#define VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS ( \
5737 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
5738 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
5739 BIT(POWER_DOMAIN_INIT))
5740
5741#define VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS ( \
5742 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
5743 BIT(POWER_DOMAIN_INIT))
5744
5745#define VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS ( \
5746 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
5747 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
5748 BIT(POWER_DOMAIN_INIT))
5749
5750#define VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS ( \
5751 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
5752 BIT(POWER_DOMAIN_INIT))
5753
a45f4466
ID		 5754static const struct i915_power_well_ops i9xx_always_on_power_well_ops = {
5755 .sync_hw = i9xx_always_on_power_well_noop,
5756 .enable = i9xx_always_on_power_well_noop,
5757 .disable = i9xx_always_on_power_well_noop,
5758 .is_enabled = i9xx_always_on_power_well_enabled,
5759};
c6cb582e 5760
1c2256df
ID		 5761static struct i915_power_well i9xx_always_on_power_well[] = {
5762 {
5763 .name = "always-on",
5764 .always_on = 1,
5765 .domains = POWER_DOMAIN_MASK,
c6cb582e 5766 .ops = &i9xx_always_on_power_well_ops,
1c2256df
ID		 5767 },
5768};
5769
c6cb582e
ID		 5770static const struct i915_power_well_ops hsw_power_well_ops = {
5771 .sync_hw = hsw_power_well_sync_hw,
5772 .enable = hsw_power_well_enable,
5773 .disable = hsw_power_well_disable,
5774 .is_enabled = hsw_power_well_enabled,
5775};
5776
c1ca727f 5777static struct i915_power_well hsw_power_wells[] = {
6f3ef5dd
ID		 5778 {
5779 .name = "always-on",
5780 .always_on = 1,
5781 .domains = HSW_ALWAYS_ON_POWER_DOMAINS,
c6cb582e 5782 .ops = &i9xx_always_on_power_well_ops,
6f3ef5dd 5783 },
c1ca727f
ID		 5784 {
5785 .name = "display",
efcad917 5786 .domains = HSW_DISPLAY_POWER_DOMAINS,
c6cb582e 5787 .ops = &hsw_power_well_ops,
c1ca727f
ID		 5788 },
5789};
5790
5791static struct i915_power_well bdw_power_wells[] = {
6f3ef5dd
ID		 5792 {
5793 .name = "always-on",
5794 .always_on = 1,
5795 .domains = BDW_ALWAYS_ON_POWER_DOMAINS,
c6cb582e 5796 .ops = &i9xx_always_on_power_well_ops,
6f3ef5dd 5797 },
c1ca727f
ID		 5798 {
5799 .name = "display",
efcad917 5800 .domains = BDW_DISPLAY_POWER_DOMAINS,
c6cb582e 5801 .ops = &hsw_power_well_ops,
c1ca727f
ID		 5802 },
5803};
5804
77961eb9
ID		 5805static const struct i915_power_well_ops vlv_display_power_well_ops = {
5806 .sync_hw = vlv_power_well_sync_hw,
5807 .enable = vlv_display_power_well_enable,
5808 .disable = vlv_display_power_well_disable,
5809 .is_enabled = vlv_power_well_enabled,
5810};
5811
5812static const struct i915_power_well_ops vlv_dpio_power_well_ops = {
5813 .sync_hw = vlv_power_well_sync_hw,
5814 .enable = vlv_power_well_enable,
5815 .disable = vlv_power_well_disable,
5816 .is_enabled = vlv_power_well_enabled,
5817};
5818
5819static struct i915_power_well vlv_power_wells[] = {
5820 {
5821 .name = "always-on",
5822 .always_on = 1,
5823 .domains = VLV_ALWAYS_ON_POWER_DOMAINS,
5824 .ops = &i9xx_always_on_power_well_ops,
5825 },
5826 {
5827 .name = "display",
5828 .domains = VLV_DISPLAY_POWER_DOMAINS,
5829 .data = PUNIT_POWER_WELL_DISP2D,
5830 .ops = &vlv_display_power_well_ops,
5831 },
5832 {
5833 .name = "dpio-common",
5834 .domains = VLV_DPIO_CMN_BC_POWER_DOMAINS,
5835 .data = PUNIT_POWER_WELL_DPIO_CMN_BC,
5836 .ops = &vlv_dpio_power_well_ops,
5837 },
5838 {
5839 .name = "dpio-tx-b-01",
5840 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
5841 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
5842 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
5843 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
5844 .ops = &vlv_dpio_power_well_ops,
5845 .data = PUNIT_POWER_WELL_DPIO_TX_B_LANES_01,
5846 },
5847 {
5848 .name = "dpio-tx-b-23",
5849 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
5850 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
5851 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
5852 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
5853 .ops = &vlv_dpio_power_well_ops,
5854 .data = PUNIT_POWER_WELL_DPIO_TX_B_LANES_23,
5855 },
5856 {
5857 .name = "dpio-tx-c-01",
5858 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
5859 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
5860 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
5861 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
5862 .ops = &vlv_dpio_power_well_ops,
5863 .data = PUNIT_POWER_WELL_DPIO_TX_C_LANES_01,
5864 },
5865 {
5866 .name = "dpio-tx-c-23",
5867 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
5868 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
5869 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
5870 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
5871 .ops = &vlv_dpio_power_well_ops,
5872 .data = PUNIT_POWER_WELL_DPIO_TX_C_LANES_23,
5873 },
5874};
5875
c1ca727f
ID		 5876#define set_power_wells(power_domains, __power_wells) ({ \
5877 (power_domains)->power_wells = (__power_wells); \
5878 (power_domains)->power_well_count = ARRAY_SIZE(__power_wells); \
5879})
5880
da7e29bd 5881int intel_power_domains_init(struct drm_i915_private *dev_priv)
a38911a3 5882{
83c00f55 5883 struct i915_power_domains *power_domains = &dev_priv->power_domains;
c1ca727f 5884
83c00f55 5885 mutex_init(&power_domains->lock);
a38911a3 5886
c1ca727f
ID		 5887 /*
5888 * The enabling order will be from lower to higher indexed wells,
5889 * the disabling order is reversed.
5890 */
da7e29bd 5891 if (IS_HASWELL(dev_priv->dev)) {
c1ca727f
ID		 5892 set_power_wells(power_domains, hsw_power_wells);
5893 hsw_pwr = power_domains;
da7e29bd 5894 } else if (IS_BROADWELL(dev_priv->dev)) {
c1ca727f
ID		 5895 set_power_wells(power_domains, bdw_power_wells);
5896 hsw_pwr = power_domains;
77961eb9
ID		 5897 } else if (IS_VALLEYVIEW(dev_priv->dev)) {
5898 set_power_wells(power_domains, vlv_power_wells);
c1ca727f 5899 } else {
1c2256df 5900 set_power_wells(power_domains, i9xx_always_on_power_well);
c1ca727f 5901 }
a38911a3
WX		 5902
5903 return 0;
5904}
5905
da7e29bd 5906void intel_power_domains_remove(struct drm_i915_private *dev_priv)
a38911a3
WX		 5907{
5908 hsw_pwr = NULL;
5909}
5910
da7e29bd 5911static void intel_power_domains_resume(struct drm_i915_private *dev_priv)
9cdb826c 5912{
83c00f55
ID		 5913 struct i915_power_domains *power_domains = &dev_priv->power_domains;
5914 struct i915_power_well *power_well;
c1ca727f 5915 int i;
9cdb826c 5916
83c00f55 5917 mutex_lock(&power_domains->lock);
a45f4466
ID		 5918 for_each_power_well(i, power_well, POWER_DOMAIN_MASK, power_domains)
5919 power_well->ops->sync_hw(dev_priv, power_well);
83c00f55 5920 mutex_unlock(&power_domains->lock);
a38911a3
WX		 5921}
5922
da7e29bd 5923void intel_power_domains_init_hw(struct drm_i915_private *dev_priv)
d0d3e513 5924{
fa42e23c 5925 /* For now, we need the power well to be always enabled. */
da7e29bd
ID		 5926 intel_display_set_init_power(dev_priv, true);
5927 intel_power_domains_resume(dev_priv);
d0d3e513
ED		 5928}
5929
c67a470b
PZ		 5930void intel_aux_display_runtime_get(struct drm_i915_private *dev_priv)
5931{
d361ae26 5932 intel_runtime_pm_get(dev_priv);
c67a470b
PZ		 5933}
5934
5935void intel_aux_display_runtime_put(struct drm_i915_private *dev_priv)
5936{
d361ae26 5937 intel_runtime_pm_put(dev_priv);
c67a470b
PZ		 5938}
5939
8a187455
PZ		 5940void intel_runtime_pm_get(struct drm_i915_private *dev_priv)
5941{
5942 struct drm_device *dev = dev_priv->dev;
5943 struct device *device = &dev->pdev->dev;
5944
5945 if (!HAS_RUNTIME_PM(dev))
5946 return;
5947
5948 pm_runtime_get_sync(device);
5949 WARN(dev_priv->pm.suspended, "Device still suspended.\n");
5950}
5951
5952void intel_runtime_pm_put(struct drm_i915_private *dev_priv)
5953{
5954 struct drm_device *dev = dev_priv->dev;
5955 struct device *device = &dev->pdev->dev;
5956
5957 if (!HAS_RUNTIME_PM(dev))
5958 return;
5959
5960 pm_runtime_mark_last_busy(device);
5961 pm_runtime_put_autosuspend(device);
5962}
5963
5964void intel_init_runtime_pm(struct drm_i915_private *dev_priv)
5965{
5966 struct drm_device *dev = dev_priv->dev;
5967 struct device *device = &dev->pdev->dev;
5968
8a187455
PZ		 5969 if (!HAS_RUNTIME_PM(dev))
5970 return;
5971
5972 pm_runtime_set_active(device);
5973
5974 pm_runtime_set_autosuspend_delay(device, 10000); /* 10s */
5975 pm_runtime_mark_last_busy(device);
5976 pm_runtime_use_autosuspend(device);
ba0239e0
PZ		 5977
5978 pm_runtime_put_autosuspend(device);
8a187455
PZ		 5979}
5980
5981void intel_fini_runtime_pm(struct drm_i915_private *dev_priv)
5982{
5983 struct drm_device *dev = dev_priv->dev;
5984 struct device *device = &dev->pdev->dev;
5985
5986 if (!HAS_RUNTIME_PM(dev))
5987 return;
5988
5989 /* Make sure we're not suspended first. */
5990 pm_runtime_get_sync(device);
5991 pm_runtime_disable(device);
5992}
5993
1fa61106
ED		 5994/* Set up chip specific power management-related functions */
5995void intel_init_pm(struct drm_device *dev)
5996{
5997 struct drm_i915_private *dev_priv = dev->dev_private;
5998
3a77c4c4 5999 if (HAS_FBC(dev)) {
40045465 6000 if (INTEL_INFO(dev)->gen >= 7) {
1fa61106 6001 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
40045465
VS		 6002 dev_priv->display.enable_fbc = gen7_enable_fbc;
6003 dev_priv->display.disable_fbc = ironlake_disable_fbc;
6004 } else if (INTEL_INFO(dev)->gen >= 5) {
6005 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
6006 dev_priv->display.enable_fbc = ironlake_enable_fbc;
1fa61106
ED		 6007 dev_priv->display.disable_fbc = ironlake_disable_fbc;
6008 } else if (IS_GM45(dev)) {
6009 dev_priv->display.fbc_enabled = g4x_fbc_enabled;
6010 dev_priv->display.enable_fbc = g4x_enable_fbc;
6011 dev_priv->display.disable_fbc = g4x_disable_fbc;
40045465 6012 } else {
1fa61106
ED		 6013 dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
6014 dev_priv->display.enable_fbc = i8xx_enable_fbc;
6015 dev_priv->display.disable_fbc = i8xx_disable_fbc;
993495ae
VS		 6016
6017 /* This value was pulled out of someone's hat */
6018 I915_WRITE(FBC_CONTROL, 500 << FBC_CTL_INTERVAL_SHIFT);
1fa61106 6019 }
1fa61106
ED		 6020 }
6021
c921aba8
DV		 6022 /* For cxsr */
6023 if (IS_PINEVIEW(dev))
6024 i915_pineview_get_mem_freq(dev);
6025 else if (IS_GEN5(dev))
6026 i915_ironlake_get_mem_freq(dev);
6027
1fa61106
ED		 6028 /* For FIFO watermark updates */
6029 if (HAS_PCH_SPLIT(dev)) {
fa50ad61 6030 ilk_setup_wm_latency(dev);
53615a5e 6031
bd602544
VS		 6032 if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] &&
6033 dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
6034 (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] &&
6035 dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
6036 dev_priv->display.update_wm = ilk_update_wm;
6037 dev_priv->display.update_sprite_wm = ilk_update_sprite_wm;
6038 } else {
6039 DRM_DEBUG_KMS("Failed to read display plane latency. "
6040 "Disable CxSR\n");
6041 }
6042
6043 if (IS_GEN5(dev))
1fa61106 6044 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
bd602544 6045 else if (IS_GEN6(dev))
1fa61106 6046 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
bd602544 6047 else if (IS_IVYBRIDGE(dev))
1fa61106 6048 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
bd602544 6049 else if (IS_HASWELL(dev))
cad2a2d7 6050 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
bd602544 6051 else if (INTEL_INFO(dev)->gen == 8)
1020a5c2 6052 dev_priv->display.init_clock_gating = gen8_init_clock_gating;
1fa61106
ED		 6053 } else if (IS_VALLEYVIEW(dev)) {
6054 dev_priv->display.update_wm = valleyview_update_wm;
6055 dev_priv->display.init_clock_gating =
6056 valleyview_init_clock_gating;
1fa61106
ED		 6057 } else if (IS_PINEVIEW(dev)) {
6058 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
6059 dev_priv->is_ddr3,
6060 dev_priv->fsb_freq,
6061 dev_priv->mem_freq)) {
6062 DRM_INFO("failed to find known CxSR latency "
6063 "(found ddr%s fsb freq %d, mem freq %d), "
6064 "disabling CxSR\n",
6065 (dev_priv->is_ddr3 == 1) ? "3" : "2",
6066 dev_priv->fsb_freq, dev_priv->mem_freq);
6067 /* Disable CxSR and never update its watermark again */
6068 pineview_disable_cxsr(dev);
6069 dev_priv->display.update_wm = NULL;
6070 } else
6071 dev_priv->display.update_wm = pineview_update_wm;
6072 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
6073 } else if (IS_G4X(dev)) {
6074 dev_priv->display.update_wm = g4x_update_wm;
6075 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
6076 } else if (IS_GEN4(dev)) {
6077 dev_priv->display.update_wm = i965_update_wm;
6078 if (IS_CRESTLINE(dev))
6079 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
6080 else if (IS_BROADWATER(dev))
6081 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
6082 } else if (IS_GEN3(dev)) {
6083 dev_priv->display.update_wm = i9xx_update_wm;
6084 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
6085 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
feb56b93
DV		 6086 } else if (IS_GEN2(dev)) {
6087 if (INTEL_INFO(dev)->num_pipes == 1) {
6088 dev_priv->display.update_wm = i845_update_wm;
1fa61106 6089 dev_priv->display.get_fifo_size = i845_get_fifo_size;
feb56b93
DV		 6090 } else {
6091 dev_priv->display.update_wm = i9xx_update_wm;
1fa61106 6092 dev_priv->display.get_fifo_size = i830_get_fifo_size;
feb56b93
DV		 6093 }
6094
6095 if (IS_I85X(dev) || IS_I865G(dev))
6096 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
6097 else
6098 dev_priv->display.init_clock_gating = i830_init_clock_gating;
6099 } else {
6100 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
1fa61106
ED		 6101 }
6102}
6103
42c0526c
BW		 6104int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
6105{
4fc688ce 6106 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW		 6107
6108 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
6109 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
6110 return -EAGAIN;
6111 }
6112
6113 I915_WRITE(GEN6_PCODE_DATA, *val);
6114 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
6115
6116 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
6117 500)) {
6118 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
6119 return -ETIMEDOUT;
6120 }
6121
6122 *val = I915_READ(GEN6_PCODE_DATA);
6123 I915_WRITE(GEN6_PCODE_DATA, 0);
6124
6125 return 0;
6126}
6127
6128int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
6129{
4fc688ce 6130 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW		 6131
6132 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
6133 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
6134 return -EAGAIN;
6135 }
6136
6137 I915_WRITE(GEN6_PCODE_DATA, val);
6138 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
6139
6140 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
6141 500)) {
6142 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
6143 return -ETIMEDOUT;
6144 }
6145
6146 I915_WRITE(GEN6_PCODE_DATA, 0);
6147
6148 return 0;
6149}
a0e4e199 6150
2ec3815f 6151int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val)
855ba3be 6152{
07ab118b 6153 int div;
855ba3be 6154
07ab118b 6155 /* 4 x czclk */
2ec3815f 6156 switch (dev_priv->mem_freq) {
855ba3be 6157 case 800:
07ab118b 6158 div = 10;
855ba3be
JB		 6159 break;
6160 case 1066:
07ab118b 6161 div = 12;
855ba3be
JB		 6162 break;
6163 case 1333:
07ab118b 6164 div = 16;
855ba3be
JB		 6165 break;
6166 default:
6167 return -1;
6168 }
6169
2ec3815f 6170 return DIV_ROUND_CLOSEST(dev_priv->mem_freq * (val + 6 - 0xbd), 4 * div);
855ba3be
JB		 6171}
6172
2ec3815f 6173int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val)
855ba3be 6174{
07ab118b 6175 int mul;
855ba3be 6176
07ab118b 6177 /* 4 x czclk */
2ec3815f 6178 switch (dev_priv->mem_freq) {
855ba3be 6179 case 800:
07ab118b 6180 mul = 10;
855ba3be
JB		 6181 break;
6182 case 1066:
07ab118b 6183 mul = 12;
855ba3be
JB		 6184 break;
6185 case 1333:
07ab118b 6186 mul = 16;
855ba3be
JB		 6187 break;
6188 default:
6189 return -1;
6190 }
6191
2ec3815f 6192 return DIV_ROUND_CLOSEST(4 * mul * val, dev_priv->mem_freq) + 0xbd - 6;
855ba3be
JB		 6193}
6194
f742a552 6195void intel_pm_setup(struct drm_device *dev)
907b28c5
CW		 6196{
6197 struct drm_i915_private *dev_priv = dev->dev_private;
6198
f742a552
DV		 6199 mutex_init(&dev_priv->rps.hw_lock);
6200
907b28c5
CW		 6201 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
6202 intel_gen6_powersave_work);
5d584b2e 6203
33688d95 6204 dev_priv->pm.suspended = false;
5d584b2e 6205 dev_priv->pm.irqs_disabled = false;
907b28c5 6206}
Ubuntu Artful kernel mirror