]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/gpu/drm/i915/intel_pm.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drm/i915: Move VLV cmnlane workaround to intel_power_domains_init_hw()
[mirror_ubuntu-bionic-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 231 if (drm_format_plane_cpp(fb->pixel_format, 0) == 2)
5e59f717
BW		 232 dev_priv->fbc.threshold++;
233
234 switch (dev_priv->fbc.threshold) {
235 case 4:
236 case 3:
237 dpfc_ctl |= DPFC_CTL_LIMIT_4X;
238 break;
239 case 2:
3fa2e0ee 240 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
5e59f717
BW		 241 break;
242 case 1:
3fa2e0ee 243 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
5e59f717
BW		 244 break;
245 }
d629336b
VS		 246 dpfc_ctl |= DPFC_CTL_FENCE_EN;
247 if (IS_GEN5(dev))
248 dpfc_ctl |= obj->fence_reg;
85208be0 249
85208be0 250 I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
f343c5f6 251 I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID);
85208be0
ED		 252 /* enable it... */
253 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
254
255 if (IS_GEN6(dev)) {
256 I915_WRITE(SNB_DPFC_CTL_SA,
257 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
258 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
259 sandybridge_blit_fbc_update(dev);
260 }
261
84f44ce7 262 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
85208be0
ED		 263}
264
1fa61106 265static void ironlake_disable_fbc(struct drm_device *dev)
85208be0
ED		 266{
267 struct drm_i915_private *dev_priv = dev->dev_private;
268 u32 dpfc_ctl;
269
270 /* Disable compression */
271 dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
272 if (dpfc_ctl & DPFC_CTL_EN) {
273 dpfc_ctl &= ~DPFC_CTL_EN;
274 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl);
275
276 DRM_DEBUG_KMS("disabled FBC\n");
277 }
278}
279
1fa61106 280static bool ironlake_fbc_enabled(struct drm_device *dev)
85208be0
ED		 281{
282 struct drm_i915_private *dev_priv = dev->dev_private;
283
284 return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
285}
286
993495ae 287static void gen7_enable_fbc(struct drm_crtc *crtc)
abe959c7
RV		 288{
289 struct drm_device *dev = crtc->dev;
290 struct drm_i915_private *dev_priv = dev->dev_private;
f4510a27 291 struct drm_framebuffer *fb = crtc->primary->fb;
abe959c7
RV		 292 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
293 struct drm_i915_gem_object *obj = intel_fb->obj;
294 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3fa2e0ee 295 u32 dpfc_ctl;
abe959c7 296
3fa2e0ee
VS		 297 dpfc_ctl = IVB_DPFC_CTL_PLANE(intel_crtc->plane);
298 if (drm_format_plane_cpp(fb->pixel_format, 0) == 2)
5e59f717
BW		 299 dev_priv->fbc.threshold++;
300
301 switch (dev_priv->fbc.threshold) {
302 case 4:
303 case 3:
304 dpfc_ctl |= DPFC_CTL_LIMIT_4X;
305 break;
306 case 2:
3fa2e0ee 307 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
5e59f717
BW		 308 break;
309 case 1:
3fa2e0ee 310 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
5e59f717
BW		 311 break;
312 }
313
3fa2e0ee
VS		 314 dpfc_ctl |= IVB_DPFC_CTL_FENCE_EN;
315
316 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
abe959c7 317
891348b2 318 if (IS_IVYBRIDGE(dev)) {
7dd23ba0 319 /* WaFbcAsynchFlipDisableFbcQueue:ivb */
2adb6db8
VS		 320 I915_WRITE(ILK_DISPLAY_CHICKEN1,
321 I915_READ(ILK_DISPLAY_CHICKEN1) |
322 ILK_FBCQ_DIS);
28554164 323 } else {
2adb6db8 324 /* WaFbcAsynchFlipDisableFbcQueue:hsw,bdw */
8f670bb1
VS		 325 I915_WRITE(CHICKEN_PIPESL_1(intel_crtc->pipe),
326 I915_READ(CHICKEN_PIPESL_1(intel_crtc->pipe)) |
327 HSW_FBCQ_DIS);
891348b2 328 }
b74ea102 329
abe959c7
RV		 330 I915_WRITE(SNB_DPFC_CTL_SA,
331 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
332 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
333
334 sandybridge_blit_fbc_update(dev);
335
b19870ee 336 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
abe959c7
RV		 337}
338
85208be0
ED		 339bool intel_fbc_enabled(struct drm_device *dev)
340{
341 struct drm_i915_private *dev_priv = dev->dev_private;
342
343 if (!dev_priv->display.fbc_enabled)
344 return false;
345
346 return dev_priv->display.fbc_enabled(dev);
347}
348
349static void intel_fbc_work_fn(struct work_struct *__work)
350{
351 struct intel_fbc_work *work =
352 container_of(to_delayed_work(__work),
353 struct intel_fbc_work, work);
354 struct drm_device *dev = work->crtc->dev;
355 struct drm_i915_private *dev_priv = dev->dev_private;
356
357 mutex_lock(&dev->struct_mutex);
5c3fe8b0 358 if (work == dev_priv->fbc.fbc_work) {
85208be0
ED		 359 /* Double check that we haven't switched fb without cancelling
360 * the prior work.
361 */
f4510a27 362 if (work->crtc->primary->fb == work->fb) {
993495ae 363 dev_priv->display.enable_fbc(work->crtc);
85208be0 364
5c3fe8b0 365 dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane;
f4510a27 366 dev_priv->fbc.fb_id = work->crtc->primary->fb->base.id;
5c3fe8b0 367 dev_priv->fbc.y = work->crtc->y;
85208be0
ED		 368 }
369
5c3fe8b0 370 dev_priv->fbc.fbc_work = NULL;
85208be0
ED		 371 }
372 mutex_unlock(&dev->struct_mutex);
373
374 kfree(work);
375}
376
377static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
378{
5c3fe8b0 379 if (dev_priv->fbc.fbc_work == NULL)
85208be0
ED		 380 return;
381
382 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
383
384 /* Synchronisation is provided by struct_mutex and checking of
5c3fe8b0 385 * dev_priv->fbc.fbc_work, so we can perform the cancellation
85208be0
ED		 386 * entirely asynchronously.
387 */
5c3fe8b0 388 if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work))
85208be0 389 /* tasklet was killed before being run, clean up */
5c3fe8b0 390 kfree(dev_priv->fbc.fbc_work);
85208be0
ED		 391
392 /* Mark the work as no longer wanted so that if it does
393 * wake-up (because the work was already running and waiting
394 * for our mutex), it will discover that is no longer
395 * necessary to run.
396 */
5c3fe8b0 397 dev_priv->fbc.fbc_work = NULL;
85208be0
ED		 398}
399
993495ae 400static void intel_enable_fbc(struct drm_crtc *crtc)
85208be0
ED		 401{
402 struct intel_fbc_work *work;
403 struct drm_device *dev = crtc->dev;
404 struct drm_i915_private *dev_priv = dev->dev_private;
405
406 if (!dev_priv->display.enable_fbc)
407 return;
408
409 intel_cancel_fbc_work(dev_priv);
410
b14c5679 411 work = kzalloc(sizeof(*work), GFP_KERNEL);
85208be0 412 if (work == NULL) {
6cdcb5e7 413 DRM_ERROR("Failed to allocate FBC work structure\n");
993495ae 414 dev_priv->display.enable_fbc(crtc);
85208be0
ED		 415 return;
416 }
417
418 work->crtc = crtc;
f4510a27 419 work->fb = crtc->primary->fb;
85208be0
ED		 420 INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
421
5c3fe8b0 422 dev_priv->fbc.fbc_work = work;
85208be0 423
85208be0
ED		 424 /* Delay the actual enabling to let pageflipping cease and the
425 * display to settle before starting the compression. Note that
426 * this delay also serves a second purpose: it allows for a
427 * vblank to pass after disabling the FBC before we attempt
428 * to modify the control registers.
429 *
430 * A more complicated solution would involve tracking vblanks
431 * following the termination of the page-flipping sequence
432 * and indeed performing the enable as a co-routine and not
433 * waiting synchronously upon the vblank.
7457d617
DL		 434 *
435 * WaFbcWaitForVBlankBeforeEnable:ilk,snb
85208be0
ED		 436 */
437 schedule_delayed_work(&work->work, msecs_to_jiffies(50));
438}
439
440void intel_disable_fbc(struct drm_device *dev)
441{
442 struct drm_i915_private *dev_priv = dev->dev_private;
443
444 intel_cancel_fbc_work(dev_priv);
445
446 if (!dev_priv->display.disable_fbc)
447 return;
448
449 dev_priv->display.disable_fbc(dev);
5c3fe8b0 450 dev_priv->fbc.plane = -1;
85208be0
ED		 451}
452
29ebf90f
CW		 453static bool set_no_fbc_reason(struct drm_i915_private *dev_priv,
454 enum no_fbc_reason reason)
455{
456 if (dev_priv->fbc.no_fbc_reason == reason)
457 return false;
458
459 dev_priv->fbc.no_fbc_reason = reason;
460 return true;
461}
462
85208be0
ED		 463/**
464 * intel_update_fbc - enable/disable FBC as needed
465 * @dev: the drm_device
466 *
467 * Set up the framebuffer compression hardware at mode set time. We
468 * enable it if possible:
469 * - plane A only (on pre-965)
470 * - no pixel mulitply/line duplication
471 * - no alpha buffer discard
472 * - no dual wide
f85da868 473 * - framebuffer <= max_hdisplay in width, max_vdisplay in height
85208be0
ED		 474 *
475 * We can't assume that any compression will take place (worst case),
476 * so the compressed buffer has to be the same size as the uncompressed
477 * one. It also must reside (along with the line length buffer) in
478 * stolen memory.
479 *
480 * We need to enable/disable FBC on a global basis.
481 */
482void intel_update_fbc(struct drm_device *dev)
483{
484 struct drm_i915_private *dev_priv = dev->dev_private;
485 struct drm_crtc *crtc = NULL, *tmp_crtc;
486 struct intel_crtc *intel_crtc;
487 struct drm_framebuffer *fb;
488 struct intel_framebuffer *intel_fb;
489 struct drm_i915_gem_object *obj;
ef644fda 490 const struct drm_display_mode *adjusted_mode;
37327abd 491 unsigned int max_width, max_height;
85208be0 492
3a77c4c4 493 if (!HAS_FBC(dev)) {
29ebf90f 494 set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED);
85208be0 495 return;
29ebf90f 496 }
85208be0 497
d330a953 498 if (!i915.powersave) {
29ebf90f
CW		 499 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
500 DRM_DEBUG_KMS("fbc disabled per module param\n");
85208be0 501 return;
29ebf90f 502 }
85208be0
ED		 503
504 /*
505 * If FBC is already on, we just have to verify that we can
506 * keep it that way...
507 * Need to disable if:
508 * - more than one pipe is active
509 * - changing FBC params (stride, fence, mode)
510 * - new fb is too large to fit in compressed buffer
511 * - going to an unsupported config (interlace, pixel multiply, etc.)
512 */
70e1e0ec 513 for_each_crtc(dev, tmp_crtc) {
3490ea5d 514 if (intel_crtc_active(tmp_crtc) &&
4c445e0e 515 to_intel_crtc(tmp_crtc)->primary_enabled) {
85208be0 516 if (crtc) {
29ebf90f
CW		 517 if (set_no_fbc_reason(dev_priv, FBC_MULTIPLE_PIPES))
518 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
85208be0
ED		 519 goto out_disable;
520 }
521 crtc = tmp_crtc;
522 }
523 }
524
f4510a27 525 if (!crtc || crtc->primary->fb == NULL) {
29ebf90f
CW		 526 if (set_no_fbc_reason(dev_priv, FBC_NO_OUTPUT))
527 DRM_DEBUG_KMS("no output, disabling\n");
85208be0
ED		 528 goto out_disable;
529 }
530
531 intel_crtc = to_intel_crtc(crtc);
f4510a27 532 fb = crtc->primary->fb;
85208be0
ED		 533 intel_fb = to_intel_framebuffer(fb);
534 obj = intel_fb->obj;
ef644fda 535 adjusted_mode = &intel_crtc->config.adjusted_mode;
85208be0 536
0368920e 537 if (i915.enable_fbc < 0) {
29ebf90f
CW		 538 if (set_no_fbc_reason(dev_priv, FBC_CHIP_DEFAULT))
539 DRM_DEBUG_KMS("disabled per chip default\n");
8a5729a3 540 goto out_disable;
85208be0 541 }
d330a953 542 if (!i915.enable_fbc) {
29ebf90f
CW		 543 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
544 DRM_DEBUG_KMS("fbc disabled per module param\n");
85208be0
ED		 545 goto out_disable;
546 }
ef644fda
VS		 547 if ((adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) ||
548 (adjusted_mode->flags & DRM_MODE_FLAG_DBLSCAN)) {
29ebf90f
CW		 549 if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE))
550 DRM_DEBUG_KMS("mode incompatible with compression, "
551 "disabling\n");
85208be0
ED		 552 goto out_disable;
553 }
f85da868 554
032843a5
DS		 555 if (INTEL_INFO(dev)->gen >= 8 || IS_HASWELL(dev)) {
556 max_width = 4096;
557 max_height = 4096;
558 } else if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) {
37327abd
VS		 559 max_width = 4096;
560 max_height = 2048;
f85da868 561 } else {
37327abd
VS		 562 max_width = 2048;
563 max_height = 1536;
f85da868 564 }
37327abd
VS		 565 if (intel_crtc->config.pipe_src_w > max_width ||
566 intel_crtc->config.pipe_src_h > max_height) {
29ebf90f
CW		 567 if (set_no_fbc_reason(dev_priv, FBC_MODE_TOO_LARGE))
568 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
85208be0
ED		 569 goto out_disable;
570 }
8f94d24b 571 if ((INTEL_INFO(dev)->gen < 4 || HAS_DDI(dev)) &&
c5a44aa0 572 intel_crtc->plane != PLANE_A) {
29ebf90f 573 if (set_no_fbc_reason(dev_priv, FBC_BAD_PLANE))
c5a44aa0 574 DRM_DEBUG_KMS("plane not A, disabling compression\n");
85208be0
ED		 575 goto out_disable;
576 }
577
578 /* The use of a CPU fence is mandatory in order to detect writes
579 * by the CPU to the scanout and trigger updates to the FBC.
580 */
581 if (obj->tiling_mode != I915_TILING_X ||
582 obj->fence_reg == I915_FENCE_REG_NONE) {
29ebf90f
CW		 583 if (set_no_fbc_reason(dev_priv, FBC_NOT_TILED))
584 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
85208be0
ED		 585 goto out_disable;
586 }
587
588 /* If the kernel debugger is active, always disable compression */
589 if (in_dbg_master())
590 goto out_disable;
591
5e59f717
BW		 592 if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size,
593 drm_format_plane_cpp(fb->pixel_format, 0))) {
29ebf90f
CW		 594 if (set_no_fbc_reason(dev_priv, FBC_STOLEN_TOO_SMALL))
595 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
11be49eb
CW		 596 goto out_disable;
597 }
598
85208be0
ED		 599 /* If the scanout has not changed, don't modify the FBC settings.
600 * Note that we make the fundamental assumption that the fb->obj
601 * cannot be unpinned (and have its GTT offset and fence revoked)
602 * without first being decoupled from the scanout and FBC disabled.
603 */
5c3fe8b0
BW		 604 if (dev_priv->fbc.plane == intel_crtc->plane &&
605 dev_priv->fbc.fb_id == fb->base.id &&
606 dev_priv->fbc.y == crtc->y)
85208be0
ED		 607 return;
608
609 if (intel_fbc_enabled(dev)) {
610 /* We update FBC along two paths, after changing fb/crtc
611 * configuration (modeswitching) and after page-flipping
612 * finishes. For the latter, we know that not only did
613 * we disable the FBC at the start of the page-flip
614 * sequence, but also more than one vblank has passed.
615 *
616 * For the former case of modeswitching, it is possible
617 * to switch between two FBC valid configurations
618 * instantaneously so we do need to disable the FBC
619 * before we can modify its control registers. We also
620 * have to wait for the next vblank for that to take
621 * effect. However, since we delay enabling FBC we can
622 * assume that a vblank has passed since disabling and
623 * that we can safely alter the registers in the deferred
624 * callback.
625 *
626 * In the scenario that we go from a valid to invalid
627 * and then back to valid FBC configuration we have
628 * no strict enforcement that a vblank occurred since
629 * disabling the FBC. However, along all current pipe
630 * disabling paths we do need to wait for a vblank at
631 * some point. And we wait before enabling FBC anyway.
632 */
633 DRM_DEBUG_KMS("disabling active FBC for update\n");
634 intel_disable_fbc(dev);
635 }
636
993495ae 637 intel_enable_fbc(crtc);
29ebf90f 638 dev_priv->fbc.no_fbc_reason = FBC_OK;
85208be0
ED		 639 return;
640
641out_disable:
642 /* Multiple disables should be harmless */
643 if (intel_fbc_enabled(dev)) {
644 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
645 intel_disable_fbc(dev);
646 }
11be49eb 647 i915_gem_stolen_cleanup_compression(dev);
85208be0
ED		 648}
649
c921aba8
DV		 650static void i915_pineview_get_mem_freq(struct drm_device *dev)
651{
50227e1c 652 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV		 653 u32 tmp;
654
655 tmp = I915_READ(CLKCFG);
656
657 switch (tmp & CLKCFG_FSB_MASK) {
658 case CLKCFG_FSB_533:
659 dev_priv->fsb_freq = 533; /* 133*4 */
660 break;
661 case CLKCFG_FSB_800:
662 dev_priv->fsb_freq = 800; /* 200*4 */
663 break;
664 case CLKCFG_FSB_667:
665 dev_priv->fsb_freq = 667; /* 167*4 */
666 break;
667 case CLKCFG_FSB_400:
668 dev_priv->fsb_freq = 400; /* 100*4 */
669 break;
670 }
671
672 switch (tmp & CLKCFG_MEM_MASK) {
673 case CLKCFG_MEM_533:
674 dev_priv->mem_freq = 533;
675 break;
676 case CLKCFG_MEM_667:
677 dev_priv->mem_freq = 667;
678 break;
679 case CLKCFG_MEM_800:
680 dev_priv->mem_freq = 800;
681 break;
682 }
683
684 /* detect pineview DDR3 setting */
685 tmp = I915_READ(CSHRDDR3CTL);
686 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
687}
688
689static void i915_ironlake_get_mem_freq(struct drm_device *dev)
690{
50227e1c 691 struct drm_i915_private *dev_priv = dev->dev_private;
c921aba8
DV		 692 u16 ddrpll, csipll;
693
694 ddrpll = I915_READ16(DDRMPLL1);
695 csipll = I915_READ16(CSIPLL0);
696
697 switch (ddrpll & 0xff) {
698 case 0xc:
699 dev_priv->mem_freq = 800;
700 break;
701 case 0x10:
702 dev_priv->mem_freq = 1066;
703 break;
704 case 0x14:
705 dev_priv->mem_freq = 1333;
706 break;
707 case 0x18:
708 dev_priv->mem_freq = 1600;
709 break;
710 default:
711 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
712 ddrpll & 0xff);
713 dev_priv->mem_freq = 0;
714 break;
715 }
716
20e4d407 717 dev_priv->ips.r_t = dev_priv->mem_freq;
c921aba8
DV		 718
719 switch (csipll & 0x3ff) {
720 case 0x00c:
721 dev_priv->fsb_freq = 3200;
722 break;
723 case 0x00e:
724 dev_priv->fsb_freq = 3733;
725 break;
726 case 0x010:
727 dev_priv->fsb_freq = 4266;
728 break;
729 case 0x012:
730 dev_priv->fsb_freq = 4800;
731 break;
732 case 0x014:
733 dev_priv->fsb_freq = 5333;
734 break;
735 case 0x016:
736 dev_priv->fsb_freq = 5866;
737 break;
738 case 0x018:
739 dev_priv->fsb_freq = 6400;
740 break;
741 default:
742 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
743 csipll & 0x3ff);
744 dev_priv->fsb_freq = 0;
745 break;
746 }
747
748 if (dev_priv->fsb_freq == 3200) {
20e4d407 749 dev_priv->ips.c_m = 0;
c921aba8 750 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
20e4d407 751 dev_priv->ips.c_m = 1;
c921aba8 752 } else {
20e4d407 753 dev_priv->ips.c_m = 2;
c921aba8
DV		 754 }
755}
756
b445e3b0
ED		 757static const struct cxsr_latency cxsr_latency_table[] = {
758 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
759 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
760 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
761 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
762 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
763
764 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
765 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
766 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
767 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
768 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
769
770 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
771 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
772 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
773 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
774 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
775
776 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
777 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
778 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
779 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
780 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
781
782 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
783 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
784 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
785 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
786 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
787
788 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
789 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
790 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
791 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
792 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
793};
794
63c62275 795static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
b445e3b0
ED		 796 int is_ddr3,
797 int fsb,
798 int mem)
799{
800 const struct cxsr_latency *latency;
801 int i;
802
803 if (fsb == 0 || mem == 0)
804 return NULL;
805
806 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
807 latency = &cxsr_latency_table[i];
808 if (is_desktop == latency->is_desktop &&
809 is_ddr3 == latency->is_ddr3 &&
810 fsb == latency->fsb_freq && mem == latency->mem_freq)
811 return latency;
812 }
813
814 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
815
816 return NULL;
817}
818
1fa61106 819static void pineview_disable_cxsr(struct drm_device *dev)
b445e3b0
ED		 820{
821 struct drm_i915_private *dev_priv = dev->dev_private;
822
823 /* deactivate cxsr */
824 I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
825}
826
827/*
828 * Latency for FIFO fetches is dependent on several factors:
829 * - memory configuration (speed, channels)
830 * - chipset
831 * - current MCH state
832 * It can be fairly high in some situations, so here we assume a fairly
833 * pessimal value. It's a tradeoff between extra memory fetches (if we
834 * set this value too high, the FIFO will fetch frequently to stay full)
835 * and power consumption (set it too low to save power and we might see
836 * FIFO underruns and display "flicker").
837 *
838 * A value of 5us seems to be a good balance; safe for very low end
839 * platforms but not overly aggressive on lower latency configs.
840 */
841static const int latency_ns = 5000;
842
1fa61106 843static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 844{
845 struct drm_i915_private *dev_priv = dev->dev_private;
846 uint32_t dsparb = I915_READ(DSPARB);
847 int size;
848
849 size = dsparb & 0x7f;
850 if (plane)
851 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
852
853 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
854 plane ? "B" : "A", size);
855
856 return size;
857}
858
feb56b93 859static int i830_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 860{
861 struct drm_i915_private *dev_priv = dev->dev_private;
862 uint32_t dsparb = I915_READ(DSPARB);
863 int size;
864
865 size = dsparb & 0x1ff;
866 if (plane)
867 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
868 size >>= 1; /* Convert to cachelines */
869
870 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
871 plane ? "B" : "A", size);
872
873 return size;
874}
875
1fa61106 876static int i845_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED		 877{
878 struct drm_i915_private *dev_priv = dev->dev_private;
879 uint32_t dsparb = I915_READ(DSPARB);
880 int size;
881
882 size = dsparb & 0x7f;
883 size >>= 2; /* Convert to cachelines */
884
885 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
886 plane ? "B" : "A",
887 size);
888
889 return size;
890}
891
b445e3b0
ED		 892/* Pineview has different values for various configs */
893static const struct intel_watermark_params pineview_display_wm = {
e0f0273e
VS		 894 .fifo_size = PINEVIEW_DISPLAY_FIFO,
895 .max_wm = PINEVIEW_MAX_WM,
896 .default_wm = PINEVIEW_DFT_WM,
897 .guard_size = PINEVIEW_GUARD_WM,
898 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 899};
900static const struct intel_watermark_params pineview_display_hplloff_wm = {
e0f0273e
VS		 901 .fifo_size = PINEVIEW_DISPLAY_FIFO,
902 .max_wm = PINEVIEW_MAX_WM,
903 .default_wm = PINEVIEW_DFT_HPLLOFF_WM,
904 .guard_size = PINEVIEW_GUARD_WM,
905 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 906};
907static const struct intel_watermark_params pineview_cursor_wm = {
e0f0273e
VS		 908 .fifo_size = PINEVIEW_CURSOR_FIFO,
909 .max_wm = PINEVIEW_CURSOR_MAX_WM,
910 .default_wm = PINEVIEW_CURSOR_DFT_WM,
911 .guard_size = PINEVIEW_CURSOR_GUARD_WM,
912 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 913};
914static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
e0f0273e
VS		 915 .fifo_size = PINEVIEW_CURSOR_FIFO,
916 .max_wm = PINEVIEW_CURSOR_MAX_WM,
917 .default_wm = PINEVIEW_CURSOR_DFT_WM,
918 .guard_size = PINEVIEW_CURSOR_GUARD_WM,
919 .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
b445e3b0
ED		 920};
921static const struct intel_watermark_params g4x_wm_info = {
e0f0273e
VS		 922 .fifo_size = G4X_FIFO_SIZE,
923 .max_wm = G4X_MAX_WM,
924 .default_wm = G4X_MAX_WM,
925 .guard_size = 2,
926 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED		 927};
928static const struct intel_watermark_params g4x_cursor_wm_info = {
e0f0273e
VS		 929 .fifo_size = I965_CURSOR_FIFO,
930 .max_wm = I965_CURSOR_MAX_WM,
931 .default_wm = I965_CURSOR_DFT_WM,
932 .guard_size = 2,
933 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED		 934};
935static const struct intel_watermark_params valleyview_wm_info = {
e0f0273e
VS		 936 .fifo_size = VALLEYVIEW_FIFO_SIZE,
937 .max_wm = VALLEYVIEW_MAX_WM,
938 .default_wm = VALLEYVIEW_MAX_WM,
939 .guard_size = 2,
940 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED		 941};
942static const struct intel_watermark_params valleyview_cursor_wm_info = {
e0f0273e
VS		 943 .fifo_size = I965_CURSOR_FIFO,
944 .max_wm = VALLEYVIEW_CURSOR_MAX_WM,
945 .default_wm = I965_CURSOR_DFT_WM,
946 .guard_size = 2,
947 .cacheline_size = G4X_FIFO_LINE_SIZE,
b445e3b0
ED		 948};
949static const struct intel_watermark_params i965_cursor_wm_info = {
e0f0273e
VS		 950 .fifo_size = I965_CURSOR_FIFO,
951 .max_wm = I965_CURSOR_MAX_WM,
952 .default_wm = I965_CURSOR_DFT_WM,
953 .guard_size = 2,
954 .cacheline_size = I915_FIFO_LINE_SIZE,
b445e3b0
ED		 955};
956static const struct intel_watermark_params i945_wm_info = {
e0f0273e
VS		 957 .fifo_size = I945_FIFO_SIZE,
958 .max_wm = I915_MAX_WM,
959 .default_wm = 1,
960 .guard_size = 2,
961 .cacheline_size = I915_FIFO_LINE_SIZE,
b445e3b0
ED		 962};
963static const struct intel_watermark_params i915_wm_info = {
e0f0273e
VS		 964 .fifo_size = I915_FIFO_SIZE,
965 .max_wm = I915_MAX_WM,
966 .default_wm = 1,
967 .guard_size = 2,
968 .cacheline_size = I915_FIFO_LINE_SIZE,
b445e3b0 969};
feb56b93 970static const struct intel_watermark_params i830_wm_info = {
e0f0273e
VS		 971 .fifo_size = I855GM_FIFO_SIZE,
972 .max_wm = I915_MAX_WM,
973 .default_wm = 1,
974 .guard_size = 2,
975 .cacheline_size = I830_FIFO_LINE_SIZE,
b445e3b0 976};
feb56b93 977static const struct intel_watermark_params i845_wm_info = {
e0f0273e
VS		 978 .fifo_size = I830_FIFO_SIZE,
979 .max_wm = I915_MAX_WM,
980 .default_wm = 1,
981 .guard_size = 2,
982 .cacheline_size = I830_FIFO_LINE_SIZE,
b445e3b0
ED		 983};
984
b445e3b0
ED		 985/**
986 * intel_calculate_wm - calculate watermark level
987 * @clock_in_khz: pixel clock
988 * @wm: chip FIFO params
989 * @pixel_size: display pixel size
990 * @latency_ns: memory latency for the platform
991 *
992 * Calculate the watermark level (the level at which the display plane will
993 * start fetching from memory again). Each chip has a different display
994 * FIFO size and allocation, so the caller needs to figure that out and pass
995 * in the correct intel_watermark_params structure.
996 *
997 * As the pixel clock runs, the FIFO will be drained at a rate that depends
998 * on the pixel size. When it reaches the watermark level, it'll start
999 * fetching FIFO line sized based chunks from memory until the FIFO fills
1000 * past the watermark point. If the FIFO drains completely, a FIFO underrun
1001 * will occur, and a display engine hang could result.
1002 */
1003static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
1004 const struct intel_watermark_params *wm,
1005 int fifo_size,
1006 int pixel_size,
1007 unsigned long latency_ns)
1008{
1009 long entries_required, wm_size;
1010
1011 /*
1012 * Note: we need to make sure we don't overflow for various clock &
1013 * latency values.
1014 * clocks go from a few thousand to several hundred thousand.
1015 * latency is usually a few thousand
1016 */
1017 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
1018 1000;
1019 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
1020
1021 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
1022
1023 wm_size = fifo_size - (entries_required + wm->guard_size);
1024
1025 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
1026
1027 /* Don't promote wm_size to unsigned... */
1028 if (wm_size > (long)wm->max_wm)
1029 wm_size = wm->max_wm;
1030 if (wm_size <= 0)
1031 wm_size = wm->default_wm;
1032 return wm_size;
1033}
1034
1035static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1036{
1037 struct drm_crtc *crtc, *enabled = NULL;
1038
70e1e0ec 1039 for_each_crtc(dev, crtc) {
3490ea5d 1040 if (intel_crtc_active(crtc)) {
b445e3b0
ED		 1041 if (enabled)
1042 return NULL;
1043 enabled = crtc;
1044 }
1045 }
1046
1047 return enabled;
1048}
1049
46ba614c 1050static void pineview_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1051{
46ba614c 1052 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1053 struct drm_i915_private *dev_priv = dev->dev_private;
1054 struct drm_crtc *crtc;
1055 const struct cxsr_latency *latency;
1056 u32 reg;
1057 unsigned long wm;
1058
1059 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1060 dev_priv->fsb_freq, dev_priv->mem_freq);
1061 if (!latency) {
1062 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1063 pineview_disable_cxsr(dev);
1064 return;
1065 }
1066
1067 crtc = single_enabled_crtc(dev);
1068 if (crtc) {
241bfc38 1069 const struct drm_display_mode *adjusted_mode;
f4510a27 1070 int pixel_size = crtc->primary->fb->bits_per_pixel / 8;
241bfc38
DL		 1071 int clock;
1072
1073 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1074 clock = adjusted_mode->crtc_clock;
b445e3b0
ED		 1075
1076 /* Display SR */
1077 wm = intel_calculate_wm(clock, &pineview_display_wm,
1078 pineview_display_wm.fifo_size,
1079 pixel_size, latency->display_sr);
1080 reg = I915_READ(DSPFW1);
1081 reg &= ~DSPFW_SR_MASK;
1082 reg |= wm << DSPFW_SR_SHIFT;
1083 I915_WRITE(DSPFW1, reg);
1084 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1085
1086 /* cursor SR */
1087 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1088 pineview_display_wm.fifo_size,
1089 pixel_size, latency->cursor_sr);
1090 reg = I915_READ(DSPFW3);
1091 reg &= ~DSPFW_CURSOR_SR_MASK;
1092 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1093 I915_WRITE(DSPFW3, reg);
1094
1095 /* Display HPLL off SR */
1096 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1097 pineview_display_hplloff_wm.fifo_size,
1098 pixel_size, latency->display_hpll_disable);
1099 reg = I915_READ(DSPFW3);
1100 reg &= ~DSPFW_HPLL_SR_MASK;
1101 reg |= wm & DSPFW_HPLL_SR_MASK;
1102 I915_WRITE(DSPFW3, reg);
1103
1104 /* cursor HPLL off SR */
1105 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1106 pineview_display_hplloff_wm.fifo_size,
1107 pixel_size, latency->cursor_hpll_disable);
1108 reg = I915_READ(DSPFW3);
1109 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1110 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1111 I915_WRITE(DSPFW3, reg);
1112 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1113
1114 /* activate cxsr */
1115 I915_WRITE(DSPFW3,
1116 I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1117 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1118 } else {
1119 pineview_disable_cxsr(dev);
1120 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1121 }
1122}
1123
1124static bool g4x_compute_wm0(struct drm_device *dev,
1125 int plane,
1126 const struct intel_watermark_params *display,
1127 int display_latency_ns,
1128 const struct intel_watermark_params *cursor,
1129 int cursor_latency_ns,
1130 int *plane_wm,
1131 int *cursor_wm)
1132{
1133 struct drm_crtc *crtc;
4fe8590a 1134 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1135 int htotal, hdisplay, clock, pixel_size;
1136 int line_time_us, line_count;
1137 int entries, tlb_miss;
1138
1139 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 1140 if (!intel_crtc_active(crtc)) {
b445e3b0
ED		 1141 *cursor_wm = cursor->guard_size;
1142 *plane_wm = display->guard_size;
1143 return false;
1144 }
1145
4fe8590a 1146 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1147 clock = adjusted_mode->crtc_clock;
fec8cba3 1148 htotal = adjusted_mode->crtc_htotal;
37327abd 1149 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
f4510a27 1150 pixel_size = crtc->primary->fb->bits_per_pixel / 8;
b445e3b0
ED		 1151
1152 /* Use the small buffer method to calculate plane watermark */
1153 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1154 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1155 if (tlb_miss > 0)
1156 entries += tlb_miss;
1157 entries = DIV_ROUND_UP(entries, display->cacheline_size);
1158 *plane_wm = entries + display->guard_size;
1159 if (*plane_wm > (int)display->max_wm)
1160 *plane_wm = display->max_wm;
1161
1162 /* Use the large buffer method to calculate cursor watermark */
922044c9 1163 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0 1164 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
7bb836dd 1165 entries = line_count * to_intel_crtc(crtc)->cursor_width * pixel_size;
b445e3b0
ED		 1166 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1167 if (tlb_miss > 0)
1168 entries += tlb_miss;
1169 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1170 *cursor_wm = entries + cursor->guard_size;
1171 if (*cursor_wm > (int)cursor->max_wm)
1172 *cursor_wm = (int)cursor->max_wm;
1173
1174 return true;
1175}
1176
1177/*
1178 * Check the wm result.
1179 *
1180 * If any calculated watermark values is larger than the maximum value that
1181 * can be programmed into the associated watermark register, that watermark
1182 * must be disabled.
1183 */
1184static bool g4x_check_srwm(struct drm_device *dev,
1185 int display_wm, int cursor_wm,
1186 const struct intel_watermark_params *display,
1187 const struct intel_watermark_params *cursor)
1188{
1189 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1190 display_wm, cursor_wm);
1191
1192 if (display_wm > display->max_wm) {
1193 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1194 display_wm, display->max_wm);
1195 return false;
1196 }
1197
1198 if (cursor_wm > cursor->max_wm) {
1199 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1200 cursor_wm, cursor->max_wm);
1201 return false;
1202 }
1203
1204 if (!(display_wm || cursor_wm)) {
1205 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1206 return false;
1207 }
1208
1209 return true;
1210}
1211
1212static bool g4x_compute_srwm(struct drm_device *dev,
1213 int plane,
1214 int latency_ns,
1215 const struct intel_watermark_params *display,
1216 const struct intel_watermark_params *cursor,
1217 int *display_wm, int *cursor_wm)
1218{
1219 struct drm_crtc *crtc;
4fe8590a 1220 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1221 int hdisplay, htotal, pixel_size, clock;
1222 unsigned long line_time_us;
1223 int line_count, line_size;
1224 int small, large;
1225 int entries;
1226
1227 if (!latency_ns) {
1228 *display_wm = *cursor_wm = 0;
1229 return false;
1230 }
1231
1232 crtc = intel_get_crtc_for_plane(dev, plane);
4fe8590a 1233 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1234 clock = adjusted_mode->crtc_clock;
fec8cba3 1235 htotal = adjusted_mode->crtc_htotal;
37327abd 1236 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
f4510a27 1237 pixel_size = crtc->primary->fb->bits_per_pixel / 8;
b445e3b0 1238
922044c9 1239 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1240 line_count = (latency_ns / line_time_us + 1000) / 1000;
1241 line_size = hdisplay * pixel_size;
1242
1243 /* Use the minimum of the small and large buffer method for primary */
1244 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1245 large = line_count * line_size;
1246
1247 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1248 *display_wm = entries + display->guard_size;
1249
1250 /* calculate the self-refresh watermark for display cursor */
7bb836dd 1251 entries = line_count * pixel_size * to_intel_crtc(crtc)->cursor_width;
b445e3b0
ED		 1252 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1253 *cursor_wm = entries + cursor->guard_size;
1254
1255 return g4x_check_srwm(dev,
1256 *display_wm, *cursor_wm,
1257 display, cursor);
1258}
1259
1260static bool vlv_compute_drain_latency(struct drm_device *dev,
1261 int plane,
1262 int *plane_prec_mult,
1263 int *plane_dl,
1264 int *cursor_prec_mult,
1265 int *cursor_dl)
1266{
1267 struct drm_crtc *crtc;
1268 int clock, pixel_size;
1269 int entries;
1270
1271 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 1272 if (!intel_crtc_active(crtc))
b445e3b0
ED		 1273 return false;
1274
241bfc38 1275 clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
f4510a27 1276 pixel_size = crtc->primary->fb->bits_per_pixel / 8; /* BPP */
b445e3b0
ED		 1277
1278 entries = (clock / 1000) * pixel_size;
1279 *plane_prec_mult = (entries > 256) ?
1280 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1281 *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1282 pixel_size);
1283
1284 entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */
1285 *cursor_prec_mult = (entries > 256) ?
1286 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1287 *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1288
1289 return true;
1290}
1291
1292/*
1293 * Update drain latency registers of memory arbiter
1294 *
1295 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1296 * to be programmed. Each plane has a drain latency multiplier and a drain
1297 * latency value.
1298 */
1299
1300static void vlv_update_drain_latency(struct drm_device *dev)
1301{
1302 struct drm_i915_private *dev_priv = dev->dev_private;
1303 int planea_prec, planea_dl, planeb_prec, planeb_dl;
1304 int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1305 int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1306 either 16 or 32 */
1307
1308 /* For plane A, Cursor A */
1309 if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1310 &cursor_prec_mult, &cursora_dl)) {
1311 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1312 DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1313 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1314 DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1315
1316 I915_WRITE(VLV_DDL1, cursora_prec |
1317 (cursora_dl << DDL_CURSORA_SHIFT) |
1318 planea_prec | planea_dl);
1319 }
1320
1321 /* For plane B, Cursor B */
1322 if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1323 &cursor_prec_mult, &cursorb_dl)) {
1324 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1325 DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1326 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1327 DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1328
1329 I915_WRITE(VLV_DDL2, cursorb_prec |
1330 (cursorb_dl << DDL_CURSORB_SHIFT) |
1331 planeb_prec | planeb_dl);
1332 }
1333}
1334
1335#define single_plane_enabled(mask) is_power_of_2(mask)
1336
46ba614c 1337static void valleyview_update_wm(struct drm_crtc *crtc)
b445e3b0 1338{
46ba614c 1339 struct drm_device *dev = crtc->dev;
b445e3b0
ED		 1340 static const int sr_latency_ns = 12000;
1341 struct drm_i915_private *dev_priv = dev->dev_private;
1342 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1343 int plane_sr, cursor_sr;
af6c4575 1344 int ignore_plane_sr, ignore_cursor_sr;
b445e3b0
ED		 1345 unsigned int enabled = 0;
1346
1347 vlv_update_drain_latency(dev);
1348
51cea1f4 1349 if (g4x_compute_wm0(dev, PIPE_A,
b445e3b0
ED		 1350 &valleyview_wm_info, latency_ns,
1351 &valleyview_cursor_wm_info, latency_ns,
1352 &planea_wm, &cursora_wm))
51cea1f4 1353 enabled |= 1 << PIPE_A;
b445e3b0 1354
51cea1f4 1355 if (g4x_compute_wm0(dev, PIPE_B,
b445e3b0
ED		 1356 &valleyview_wm_info, latency_ns,
1357 &valleyview_cursor_wm_info, latency_ns,
1358 &planeb_wm, &cursorb_wm))
51cea1f4 1359 enabled |= 1 << PIPE_B;
b445e3b0 1360
b445e3b0
ED		 1361 if (single_plane_enabled(enabled) &&
1362 g4x_compute_srwm(dev, ffs(enabled) - 1,
1363 sr_latency_ns,
1364 &valleyview_wm_info,
1365 &valleyview_cursor_wm_info,
af6c4575
CW		 1366 &plane_sr, &ignore_cursor_sr) &&
1367 g4x_compute_srwm(dev, ffs(enabled) - 1,
1368 2*sr_latency_ns,
1369 &valleyview_wm_info,
1370 &valleyview_cursor_wm_info,
52bd02d8 1371 &ignore_plane_sr, &cursor_sr)) {
b445e3b0 1372 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
52bd02d8 1373 } else {
b445e3b0
ED		 1374 I915_WRITE(FW_BLC_SELF_VLV,
1375 I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
52bd02d8
CW		 1376 plane_sr = cursor_sr = 0;
1377 }
b445e3b0
ED		 1378
1379 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1380 planea_wm, cursora_wm,
1381 planeb_wm, cursorb_wm,
1382 plane_sr, cursor_sr);
1383
1384 I915_WRITE(DSPFW1,
1385 (plane_sr << DSPFW_SR_SHIFT) |
1386 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1387 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1388 planea_wm);
1389 I915_WRITE(DSPFW2,
8c919b28 1390 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
b445e3b0
ED		 1391 (cursora_wm << DSPFW_CURSORA_SHIFT));
1392 I915_WRITE(DSPFW3,
8c919b28
CW		 1393 (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1394 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
b445e3b0
ED		 1395}
1396
46ba614c 1397static void g4x_update_wm(struct drm_crtc *crtc)
b445e3b0 1398{
46ba614c 1399 struct drm_device *dev = crtc->dev;
b445e3b0
ED		 1400 static const int sr_latency_ns = 12000;
1401 struct drm_i915_private *dev_priv = dev->dev_private;
1402 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1403 int plane_sr, cursor_sr;
1404 unsigned int enabled = 0;
1405
51cea1f4 1406 if (g4x_compute_wm0(dev, PIPE_A,
b445e3b0
ED		 1407 &g4x_wm_info, latency_ns,
1408 &g4x_cursor_wm_info, latency_ns,
1409 &planea_wm, &cursora_wm))
51cea1f4 1410 enabled |= 1 << PIPE_A;
b445e3b0 1411
51cea1f4 1412 if (g4x_compute_wm0(dev, PIPE_B,
b445e3b0
ED		 1413 &g4x_wm_info, latency_ns,
1414 &g4x_cursor_wm_info, latency_ns,
1415 &planeb_wm, &cursorb_wm))
51cea1f4 1416 enabled |= 1 << PIPE_B;
b445e3b0 1417
b445e3b0
ED		 1418 if (single_plane_enabled(enabled) &&
1419 g4x_compute_srwm(dev, ffs(enabled) - 1,
1420 sr_latency_ns,
1421 &g4x_wm_info,
1422 &g4x_cursor_wm_info,
52bd02d8 1423 &plane_sr, &cursor_sr)) {
b445e3b0 1424 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
52bd02d8 1425 } else {
b445e3b0
ED		 1426 I915_WRITE(FW_BLC_SELF,
1427 I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
52bd02d8
CW		 1428 plane_sr = cursor_sr = 0;
1429 }
b445e3b0
ED		 1430
1431 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1432 planea_wm, cursora_wm,
1433 planeb_wm, cursorb_wm,
1434 plane_sr, cursor_sr);
1435
1436 I915_WRITE(DSPFW1,
1437 (plane_sr << DSPFW_SR_SHIFT) |
1438 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1439 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1440 planea_wm);
1441 I915_WRITE(DSPFW2,
8c919b28 1442 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
b445e3b0
ED		 1443 (cursora_wm << DSPFW_CURSORA_SHIFT));
1444 /* HPLL off in SR has some issues on G4x... disable it */
1445 I915_WRITE(DSPFW3,
8c919b28 1446 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
b445e3b0
ED		 1447 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1448}
1449
46ba614c 1450static void i965_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1451{
46ba614c 1452 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1453 struct drm_i915_private *dev_priv = dev->dev_private;
1454 struct drm_crtc *crtc;
1455 int srwm = 1;
1456 int cursor_sr = 16;
1457
1458 /* Calc sr entries for one plane configs */
1459 crtc = single_enabled_crtc(dev);
1460 if (crtc) {
1461 /* self-refresh has much higher latency */
1462 static const int sr_latency_ns = 12000;
4fe8590a
VS		 1463 const struct drm_display_mode *adjusted_mode =
1464 &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1465 int clock = adjusted_mode->crtc_clock;
fec8cba3 1466 int htotal = adjusted_mode->crtc_htotal;
37327abd 1467 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
f4510a27 1468 int pixel_size = crtc->primary->fb->bits_per_pixel / 8;
b445e3b0
ED		 1469 unsigned long line_time_us;
1470 int entries;
1471
922044c9 1472 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1473
1474 /* Use ns/us then divide to preserve precision */
1475 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1476 pixel_size * hdisplay;
1477 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1478 srwm = I965_FIFO_SIZE - entries;
1479 if (srwm < 0)
1480 srwm = 1;
1481 srwm &= 0x1ff;
1482 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1483 entries, srwm);
1484
1485 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
7bb836dd 1486 pixel_size * to_intel_crtc(crtc)->cursor_width;
b445e3b0
ED		 1487 entries = DIV_ROUND_UP(entries,
1488 i965_cursor_wm_info.cacheline_size);
1489 cursor_sr = i965_cursor_wm_info.fifo_size -
1490 (entries + i965_cursor_wm_info.guard_size);
1491
1492 if (cursor_sr > i965_cursor_wm_info.max_wm)
1493 cursor_sr = i965_cursor_wm_info.max_wm;
1494
1495 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1496 "cursor %d\n", srwm, cursor_sr);
1497
1498 if (IS_CRESTLINE(dev))
1499 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1500 } else {
1501 /* Turn off self refresh if both pipes are enabled */
1502 if (IS_CRESTLINE(dev))
1503 I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1504 & ~FW_BLC_SELF_EN);
1505 }
1506
1507 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1508 srwm);
1509
1510 /* 965 has limitations... */
1511 I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1512 (8 << 16) | (8 << 8) | (8 << 0));
1513 I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1514 /* update cursor SR watermark */
1515 I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1516}
1517
46ba614c 1518static void i9xx_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1519{
46ba614c 1520 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1521 struct drm_i915_private *dev_priv = dev->dev_private;
1522 const struct intel_watermark_params *wm_info;
1523 uint32_t fwater_lo;
1524 uint32_t fwater_hi;
1525 int cwm, srwm = 1;
1526 int fifo_size;
1527 int planea_wm, planeb_wm;
1528 struct drm_crtc *crtc, *enabled = NULL;
1529
1530 if (IS_I945GM(dev))
1531 wm_info = &i945_wm_info;
1532 else if (!IS_GEN2(dev))
1533 wm_info = &i915_wm_info;
1534 else
feb56b93 1535 wm_info = &i830_wm_info;
b445e3b0
ED		 1536
1537 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1538 crtc = intel_get_crtc_for_plane(dev, 0);
3490ea5d 1539 if (intel_crtc_active(crtc)) {
241bfc38 1540 const struct drm_display_mode *adjusted_mode;
f4510a27 1541 int cpp = crtc->primary->fb->bits_per_pixel / 8;
b9e0bda3
CW		 1542 if (IS_GEN2(dev))
1543 cpp = 4;
1544
241bfc38
DL		 1545 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1546 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1547 wm_info, fifo_size, cpp,
b445e3b0
ED		 1548 latency_ns);
1549 enabled = crtc;
1550 } else
1551 planea_wm = fifo_size - wm_info->guard_size;
1552
1553 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1554 crtc = intel_get_crtc_for_plane(dev, 1);
3490ea5d 1555 if (intel_crtc_active(crtc)) {
241bfc38 1556 const struct drm_display_mode *adjusted_mode;
f4510a27 1557 int cpp = crtc->primary->fb->bits_per_pixel / 8;
b9e0bda3
CW		 1558 if (IS_GEN2(dev))
1559 cpp = 4;
1560
241bfc38
DL		 1561 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1562 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1563 wm_info, fifo_size, cpp,
b445e3b0
ED		 1564 latency_ns);
1565 if (enabled == NULL)
1566 enabled = crtc;
1567 else
1568 enabled = NULL;
1569 } else
1570 planeb_wm = fifo_size - wm_info->guard_size;
1571
1572 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1573
2ab1bc9d
DV		 1574 if (IS_I915GM(dev) && enabled) {
1575 struct intel_framebuffer *fb;
1576
1577 fb = to_intel_framebuffer(enabled->primary->fb);
1578
1579 /* self-refresh seems busted with untiled */
1580 if (fb->obj->tiling_mode == I915_TILING_NONE)
1581 enabled = NULL;
1582 }
1583
b445e3b0
ED		 1584 /*
1585 * Overlay gets an aggressive default since video jitter is bad.
1586 */
1587 cwm = 2;
1588
1589 /* Play safe and disable self-refresh before adjusting watermarks. */
1590 if (IS_I945G(dev) || IS_I945GM(dev))
1591 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1592 else if (IS_I915GM(dev))
3f2dc5ac 1593 I915_WRITE(INSTPM, _MASKED_BIT_DISABLE(INSTPM_SELF_EN));
b445e3b0
ED		 1594
1595 /* Calc sr entries for one plane configs */
1596 if (HAS_FW_BLC(dev) && enabled) {
1597 /* self-refresh has much higher latency */
1598 static const int sr_latency_ns = 6000;
4fe8590a
VS		 1599 const struct drm_display_mode *adjusted_mode =
1600 &to_intel_crtc(enabled)->config.adjusted_mode;
241bfc38 1601 int clock = adjusted_mode->crtc_clock;
fec8cba3 1602 int htotal = adjusted_mode->crtc_htotal;
f727b490 1603 int hdisplay = to_intel_crtc(enabled)->config.pipe_src_w;
f4510a27 1604 int pixel_size = enabled->primary->fb->bits_per_pixel / 8;
b445e3b0
ED		 1605 unsigned long line_time_us;
1606 int entries;
1607
922044c9 1608 line_time_us = max(htotal * 1000 / clock, 1);
b445e3b0
ED		 1609
1610 /* Use ns/us then divide to preserve precision */
1611 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1612 pixel_size * hdisplay;
1613 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1614 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1615 srwm = wm_info->fifo_size - entries;
1616 if (srwm < 0)
1617 srwm = 1;
1618
1619 if (IS_I945G(dev) || IS_I945GM(dev))
1620 I915_WRITE(FW_BLC_SELF,
1621 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1622 else if (IS_I915GM(dev))
1623 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1624 }
1625
1626 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1627 planea_wm, planeb_wm, cwm, srwm);
1628
1629 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1630 fwater_hi = (cwm & 0x1f);
1631
1632 /* Set request length to 8 cachelines per fetch */
1633 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1634 fwater_hi = fwater_hi | (1 << 8);
1635
1636 I915_WRITE(FW_BLC, fwater_lo);
1637 I915_WRITE(FW_BLC2, fwater_hi);
1638
1639 if (HAS_FW_BLC(dev)) {
1640 if (enabled) {
1641 if (IS_I945G(dev) || IS_I945GM(dev))
1642 I915_WRITE(FW_BLC_SELF,
1643 FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1644 else if (IS_I915GM(dev))
3f2dc5ac 1645 I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_SELF_EN));
b445e3b0
ED		 1646 DRM_DEBUG_KMS("memory self refresh enabled\n");
1647 } else
1648 DRM_DEBUG_KMS("memory self refresh disabled\n");
1649 }
1650}
1651
feb56b93 1652static void i845_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1653{
46ba614c 1654 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED		 1655 struct drm_i915_private *dev_priv = dev->dev_private;
1656 struct drm_crtc *crtc;
241bfc38 1657 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED		 1658 uint32_t fwater_lo;
1659 int planea_wm;
1660
1661 crtc = single_enabled_crtc(dev);
1662 if (crtc == NULL)
1663 return;
1664
241bfc38
DL		 1665 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1666 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
feb56b93 1667 &i845_wm_info,
b445e3b0 1668 dev_priv->display.get_fifo_size(dev, 0),
b9e0bda3 1669 4, latency_ns);
b445e3b0
ED		 1670 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1671 fwater_lo |= (3<<8) | planea_wm;
1672
1673 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1674
1675 I915_WRITE(FW_BLC, fwater_lo);
1676}
1677
3658729a
VS		 1678static uint32_t ilk_pipe_pixel_rate(struct drm_device *dev,
1679 struct drm_crtc *crtc)
801bcfff
PZ		 1680{
1681 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
fd4daa9c 1682 uint32_t pixel_rate;
801bcfff 1683
241bfc38 1684 pixel_rate = intel_crtc->config.adjusted_mode.crtc_clock;
801bcfff
PZ		 1685
1686 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1687 * adjust the pixel_rate here. */
1688
fd4daa9c 1689 if (intel_crtc->config.pch_pfit.enabled) {
801bcfff 1690 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
fd4daa9c 1691 uint32_t pfit_size = intel_crtc->config.pch_pfit.size;
801bcfff 1692
37327abd
VS		 1693 pipe_w = intel_crtc->config.pipe_src_w;
1694 pipe_h = intel_crtc->config.pipe_src_h;
801bcfff
PZ		 1695 pfit_w = (pfit_size >> 16) & 0xFFFF;
1696 pfit_h = pfit_size & 0xFFFF;
1697 if (pipe_w < pfit_w)
1698 pipe_w = pfit_w;
1699 if (pipe_h < pfit_h)
1700 pipe_h = pfit_h;
1701
1702 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
1703 pfit_w * pfit_h);
1704 }
1705
1706 return pixel_rate;
1707}
1708
37126462 1709/* latency must be in 0.1us units. */
23297044 1710static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
801bcfff
PZ		 1711 uint32_t latency)
1712{
1713 uint64_t ret;
1714
3312ba65
VS		 1715 if (WARN(latency == 0, "Latency value missing\n"))
1716 return UINT_MAX;
1717
801bcfff
PZ		 1718 ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
1719 ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
1720
1721 return ret;
1722}
1723
37126462 1724/* latency must be in 0.1us units. */
23297044 1725static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
801bcfff
PZ		 1726 uint32_t horiz_pixels, uint8_t bytes_per_pixel,
1727 uint32_t latency)
1728{
1729 uint32_t ret;
1730
3312ba65
VS		 1731 if (WARN(latency == 0, "Latency value missing\n"))
1732 return UINT_MAX;
1733
801bcfff
PZ		 1734 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
1735 ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
1736 ret = DIV_ROUND_UP(ret, 64) + 2;
1737 return ret;
1738}
1739
23297044 1740static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
cca32e9a
PZ		 1741 uint8_t bytes_per_pixel)
1742{
1743 return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
1744}
1745
820c1980 1746struct ilk_pipe_wm_parameters {
801bcfff 1747 bool active;
801bcfff
PZ		 1748 uint32_t pipe_htotal;
1749 uint32_t pixel_rate;
c35426d2
VS		 1750 struct intel_plane_wm_parameters pri;
1751 struct intel_plane_wm_parameters spr;
1752 struct intel_plane_wm_parameters cur;
801bcfff
PZ		 1753};
1754
820c1980 1755struct ilk_wm_maximums {
cca32e9a
PZ		 1756 uint16_t pri;
1757 uint16_t spr;
1758 uint16_t cur;
1759 uint16_t fbc;
1760};
1761
240264f4
VS		 1762/* used in computing the new watermarks state */
1763struct intel_wm_config {
1764 unsigned int num_pipes_active;
1765 bool sprites_enabled;
1766 bool sprites_scaled;
240264f4
VS		 1767};
1768
37126462
VS		 1769/*
1770 * For both WM_PIPE and WM_LP.
1771 * mem_value must be in 0.1us units.
1772 */
820c1980 1773static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters *params,
cca32e9a
PZ		 1774 uint32_t mem_value,
1775 bool is_lp)
801bcfff 1776{
cca32e9a
PZ		 1777 uint32_t method1, method2;
1778
c35426d2 1779 if (!params->active || !params->pri.enabled)
801bcfff
PZ		 1780 return 0;
1781
23297044 1782 method1 = ilk_wm_method1(params->pixel_rate,
c35426d2 1783 params->pri.bytes_per_pixel,
cca32e9a
PZ		 1784 mem_value);
1785
1786 if (!is_lp)
1787 return method1;
1788
23297044 1789 method2 = ilk_wm_method2(params->pixel_rate,
cca32e9a 1790 params->pipe_htotal,
c35426d2
VS		 1791 params->pri.horiz_pixels,
1792 params->pri.bytes_per_pixel,
cca32e9a
PZ		 1793 mem_value);
1794
1795 return min(method1, method2);
801bcfff
PZ		 1796}
1797
37126462
VS		 1798/*
1799 * For both WM_PIPE and WM_LP.
1800 * mem_value must be in 0.1us units.
1801 */
820c1980 1802static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters *params,
801bcfff
PZ		 1803 uint32_t mem_value)
1804{
1805 uint32_t method1, method2;
1806
c35426d2 1807 if (!params->active || !params->spr.enabled)
801bcfff
PZ		 1808 return 0;
1809
23297044 1810 method1 = ilk_wm_method1(params->pixel_rate,
c35426d2 1811 params->spr.bytes_per_pixel,
801bcfff 1812 mem_value);
23297044 1813 method2 = ilk_wm_method2(params->pixel_rate,
801bcfff 1814 params->pipe_htotal,
c35426d2
VS		 1815 params->spr.horiz_pixels,
1816 params->spr.bytes_per_pixel,
801bcfff
PZ		 1817 mem_value);
1818 return min(method1, method2);
1819}
1820
37126462
VS		 1821/*
1822 * For both WM_PIPE and WM_LP.
1823 * mem_value must be in 0.1us units.
1824 */
820c1980 1825static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters *params,
801bcfff
PZ		 1826 uint32_t mem_value)
1827{
c35426d2 1828 if (!params->active || !params->cur.enabled)
801bcfff
PZ		 1829 return 0;
1830
23297044 1831 return ilk_wm_method2(params->pixel_rate,
801bcfff 1832 params->pipe_htotal,
c35426d2
VS		 1833 params->cur.horiz_pixels,
1834 params->cur.bytes_per_pixel,
801bcfff
PZ		 1835 mem_value);
1836}
1837
cca32e9a 1838/* Only for WM_LP. */
820c1980 1839static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters *params,
1fda9882 1840 uint32_t pri_val)
cca32e9a 1841{
c35426d2 1842 if (!params->active || !params->pri.enabled)
cca32e9a
PZ		 1843 return 0;
1844
23297044 1845 return ilk_wm_fbc(pri_val,
c35426d2
VS		 1846 params->pri.horiz_pixels,
1847 params->pri.bytes_per_pixel);
cca32e9a
PZ		 1848}
1849
158ae64f
VS		 1850static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
1851{
416f4727
VS		 1852 if (INTEL_INFO(dev)->gen >= 8)
1853 return 3072;
1854 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS		 1855 return 768;
1856 else
1857 return 512;
1858}
1859
4e975081
VS		 1860static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev,
1861 int level, bool is_sprite)
1862{
1863 if (INTEL_INFO(dev)->gen >= 8)
1864 /* BDW primary/sprite plane watermarks */
1865 return level == 0 ? 255 : 2047;
1866 else if (INTEL_INFO(dev)->gen >= 7)
1867 /* IVB/HSW primary/sprite plane watermarks */
1868 return level == 0 ? 127 : 1023;
1869 else if (!is_sprite)
1870 /* ILK/SNB primary plane watermarks */
1871 return level == 0 ? 127 : 511;
1872 else
1873 /* ILK/SNB sprite plane watermarks */
1874 return level == 0 ? 63 : 255;
1875}
1876
1877static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev,
1878 int level)
1879{
1880 if (INTEL_INFO(dev)->gen >= 7)
1881 return level == 0 ? 63 : 255;
1882 else
1883 return level == 0 ? 31 : 63;
1884}
1885
1886static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev)
1887{
1888 if (INTEL_INFO(dev)->gen >= 8)
1889 return 31;
1890 else
1891 return 15;
1892}
1893
158ae64f
VS		 1894/* Calculate the maximum primary/sprite plane watermark */
1895static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
1896 int level,
240264f4 1897 const struct intel_wm_config *config,
158ae64f
VS		 1898 enum intel_ddb_partitioning ddb_partitioning,
1899 bool is_sprite)
1900{
1901 unsigned int fifo_size = ilk_display_fifo_size(dev);
158ae64f
VS		 1902
1903 /* if sprites aren't enabled, sprites get nothing */
240264f4 1904 if (is_sprite && !config->sprites_enabled)
158ae64f
VS		 1905 return 0;
1906
1907 /* HSW allows LP1+ watermarks even with multiple pipes */
240264f4 1908 if (level == 0 || config->num_pipes_active > 1) {
158ae64f
VS		 1909 fifo_size /= INTEL_INFO(dev)->num_pipes;
1910
1911 /*
1912 * For some reason the non self refresh
1913 * FIFO size is only half of the self
1914 * refresh FIFO size on ILK/SNB.
1915 */
1916 if (INTEL_INFO(dev)->gen <= 6)
1917 fifo_size /= 2;
1918 }
1919
240264f4 1920 if (config->sprites_enabled) {
158ae64f
VS		 1921 /* level 0 is always calculated with 1:1 split */
1922 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
1923 if (is_sprite)
1924 fifo_size *= 5;
1925 fifo_size /= 6;
1926 } else {
1927 fifo_size /= 2;
1928 }
1929 }
1930
1931 /* clamp to max that the registers can hold */
4e975081 1932 return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite));
158ae64f
VS		 1933}
1934
1935/* Calculate the maximum cursor plane watermark */
1936static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
240264f4
VS		 1937 int level,
1938 const struct intel_wm_config *config)
158ae64f
VS		 1939{
1940 /* HSW LP1+ watermarks w/ multiple pipes */
240264f4 1941 if (level > 0 && config->num_pipes_active > 1)
158ae64f
VS		 1942 return 64;
1943
1944 /* otherwise just report max that registers can hold */
4e975081 1945 return ilk_cursor_wm_reg_max(dev, level);
158ae64f
VS		 1946}
1947
d34ff9c6 1948static void ilk_compute_wm_maximums(const struct drm_device *dev,
34982fe1
VS		 1949 int level,
1950 const struct intel_wm_config *config,
1951 enum intel_ddb_partitioning ddb_partitioning,
820c1980 1952 struct ilk_wm_maximums *max)
158ae64f 1953{
240264f4
VS		 1954 max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
1955 max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
1956 max->cur = ilk_cursor_wm_max(dev, level, config);
4e975081 1957 max->fbc = ilk_fbc_wm_reg_max(dev);
158ae64f
VS		 1958}
1959
a3cb4048
VS		 1960static void ilk_compute_wm_reg_maximums(struct drm_device *dev,
1961 int level,
1962 struct ilk_wm_maximums *max)
1963{
1964 max->pri = ilk_plane_wm_reg_max(dev, level, false);
1965 max->spr = ilk_plane_wm_reg_max(dev, level, true);
1966 max->cur = ilk_cursor_wm_reg_max(dev, level);
1967 max->fbc = ilk_fbc_wm_reg_max(dev);
1968}
1969
d9395655 1970static bool ilk_validate_wm_level(int level,
820c1980 1971 const struct ilk_wm_maximums *max,
d9395655 1972 struct intel_wm_level *result)
a9786a11
VS		 1973{
1974 bool ret;
1975
1976 /* already determined to be invalid? */
1977 if (!result->enable)
1978 return false;
1979
1980 result->enable = result->pri_val <= max->pri &&
1981 result->spr_val <= max->spr &&
1982 result->cur_val <= max->cur;
1983
1984 ret = result->enable;
1985
1986 /*
1987 * HACK until we can pre-compute everything,
1988 * and thus fail gracefully if LP0 watermarks
1989 * are exceeded...
1990 */
1991 if (level == 0 && !result->enable) {
1992 if (result->pri_val > max->pri)
1993 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1994 level, result->pri_val, max->pri);
1995 if (result->spr_val > max->spr)
1996 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1997 level, result->spr_val, max->spr);
1998 if (result->cur_val > max->cur)
1999 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
2000 level, result->cur_val, max->cur);
2001
2002 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
2003 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
2004 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
2005 result->enable = true;
2006 }
2007
a9786a11
VS		 2008 return ret;
2009}
2010
d34ff9c6 2011static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
6f5ddd17 2012 int level,
820c1980 2013 const struct ilk_pipe_wm_parameters *p,
1fd527cc 2014 struct intel_wm_level *result)
6f5ddd17
VS		 2015{
2016 uint16_t pri_latency = dev_priv->wm.pri_latency[level];
2017 uint16_t spr_latency = dev_priv->wm.spr_latency[level];
2018 uint16_t cur_latency = dev_priv->wm.cur_latency[level];
2019
2020 /* WM1+ latency values stored in 0.5us units */
2021 if (level > 0) {
2022 pri_latency *= 5;
2023 spr_latency *= 5;
2024 cur_latency *= 5;
2025 }
2026
2027 result->pri_val = ilk_compute_pri_wm(p, pri_latency, level);
2028 result->spr_val = ilk_compute_spr_wm(p, spr_latency);
2029 result->cur_val = ilk_compute_cur_wm(p, cur_latency);
2030 result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val);
2031 result->enable = true;
2032}
2033
801bcfff
PZ		 2034static uint32_t
2035hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
1f8eeabf
ED		 2036{
2037 struct drm_i915_private *dev_priv = dev->dev_private;
1011d8c4 2038 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
1011d8c4 2039 struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode;
85a02deb 2040 u32 linetime, ips_linetime;
1f8eeabf 2041
801bcfff
PZ		 2042 if (!intel_crtc_active(crtc))
2043 return 0;
1011d8c4 2044
1f8eeabf
ED		 2045 /* The WM are computed with base on how long it takes to fill a single
2046 * row at the given clock rate, multiplied by 8.
2047 * */
fec8cba3
JB		 2048 linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8,
2049 mode->crtc_clock);
2050 ips_linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8,
85a02deb 2051 intel_ddi_get_cdclk_freq(dev_priv));
1f8eeabf 2052
801bcfff
PZ		 2053 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
2054 PIPE_WM_LINETIME_TIME(linetime);
1f8eeabf
ED		 2055}
2056
12b134df
VS		 2057static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[5])
2058{
2059 struct drm_i915_private *dev_priv = dev->dev_private;
2060
a42a5719 2061 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
12b134df
VS		 2062 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2063
2064 wm[0] = (sskpd >> 56) & 0xFF;
2065 if (wm[0] == 0)
2066 wm[0] = sskpd & 0xF;
e5d5019e
VS		 2067 wm[1] = (sskpd >> 4) & 0xFF;
2068 wm[2] = (sskpd >> 12) & 0xFF;
2069 wm[3] = (sskpd >> 20) & 0x1FF;
2070 wm[4] = (sskpd >> 32) & 0x1FF;
63cf9a13
VS		 2071 } else if (INTEL_INFO(dev)->gen >= 6) {
2072 uint32_t sskpd = I915_READ(MCH_SSKPD);
2073
2074 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2075 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2076 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2077 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
3a88d0ac
VS		 2078 } else if (INTEL_INFO(dev)->gen >= 5) {
2079 uint32_t mltr = I915_READ(MLTR_ILK);
2080
2081 /* ILK primary LP0 latency is 700 ns */
2082 wm[0] = 7;
2083 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2084 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
12b134df
VS		 2085 }
2086}
2087
53615a5e
VS		 2088static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2089{
2090 /* ILK sprite LP0 latency is 1300 ns */
2091 if (INTEL_INFO(dev)->gen == 5)
2092 wm[0] = 13;
2093}
2094
2095static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2096{
2097 /* ILK cursor LP0 latency is 1300 ns */
2098 if (INTEL_INFO(dev)->gen == 5)
2099 wm[0] = 13;
2100
2101 /* WaDoubleCursorLP3Latency:ivb */
2102 if (IS_IVYBRIDGE(dev))
2103 wm[3] *= 2;
2104}
2105
546c81fd 2106int ilk_wm_max_level(const struct drm_device *dev)
26ec971e 2107{
26ec971e 2108 /* how many WM levels are we expecting */
a42a5719 2109 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ad0d6dc4 2110 return 4;
26ec971e 2111 else if (INTEL_INFO(dev)->gen >= 6)
ad0d6dc4 2112 return 3;
26ec971e 2113 else
ad0d6dc4
VS		 2114 return 2;
2115}
2116
2117static void intel_print_wm_latency(struct drm_device *dev,
2118 const char *name,
2119 const uint16_t wm[5])
2120{
2121 int level, max_level = ilk_wm_max_level(dev);
26ec971e
VS		 2122
2123 for (level = 0; level <= max_level; level++) {
2124 unsigned int latency = wm[level];
2125
2126 if (latency == 0) {
2127 DRM_ERROR("%s WM%d latency not provided\n",
2128 name, level);
2129 continue;
2130 }
2131
2132 /* WM1+ latency values in 0.5us units */
2133 if (level > 0)
2134 latency *= 5;
2135
2136 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2137 name, level, wm[level],
2138 latency / 10, latency % 10);
2139 }
2140}
2141
e95a2f75
VS		 2142static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
2143 uint16_t wm[5], uint16_t min)
2144{
2145 int level, max_level = ilk_wm_max_level(dev_priv->dev);
2146
2147 if (wm[0] >= min)
2148 return false;
2149
2150 wm[0] = max(wm[0], min);
2151 for (level = 1; level <= max_level; level++)
2152 wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
2153
2154 return true;
2155}
2156
2157static void snb_wm_latency_quirk(struct drm_device *dev)
2158{
2159 struct drm_i915_private *dev_priv = dev->dev_private;
2160 bool changed;
2161
2162 /*
2163 * The BIOS provided WM memory latency values are often
2164 * inadequate for high resolution displays. Adjust them.
2165 */
2166 changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
2167 ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
2168 ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
2169
2170 if (!changed)
2171 return;
2172
2173 DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
2174 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2175 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2176 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
2177}
2178
fa50ad61 2179static void ilk_setup_wm_latency(struct drm_device *dev)
53615a5e
VS		 2180{
2181 struct drm_i915_private *dev_priv = dev->dev_private;
2182
2183 intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2184
2185 memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2186 sizeof(dev_priv->wm.pri_latency));
2187 memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2188 sizeof(dev_priv->wm.pri_latency));
2189
2190 intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2191 intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
26ec971e
VS		 2192
2193 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2194 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2195 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
e95a2f75
VS		 2196
2197 if (IS_GEN6(dev))
2198 snb_wm_latency_quirk(dev);
53615a5e
VS		 2199}
2200
820c1980 2201static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
2a44b76b 2202 struct ilk_pipe_wm_parameters *p)
1011d8c4 2203{
7c4a395f
VS		 2204 struct drm_device *dev = crtc->dev;
2205 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2206 enum pipe pipe = intel_crtc->pipe;
7c4a395f 2207 struct drm_plane *plane;
1011d8c4 2208
2a44b76b
VS		 2209 if (!intel_crtc_active(crtc))
2210 return;
801bcfff 2211
2a44b76b
VS		 2212 p->active = true;
2213 p->pipe_htotal = intel_crtc->config.adjusted_mode.crtc_htotal;
2214 p->pixel_rate = ilk_pipe_pixel_rate(dev, crtc);
2215 p->pri.bytes_per_pixel = crtc->primary->fb->bits_per_pixel / 8;
2216 p->cur.bytes_per_pixel = 4;
2217 p->pri.horiz_pixels = intel_crtc->config.pipe_src_w;
2218 p->cur.horiz_pixels = intel_crtc->cursor_width;
2219 /* TODO: for now, assume primary and cursor planes are always enabled. */
2220 p->pri.enabled = true;
2221 p->cur.enabled = true;
7c4a395f 2222
af2b653b 2223 drm_for_each_legacy_plane(plane, &dev->mode_config.plane_list) {
801bcfff 2224 struct intel_plane *intel_plane = to_intel_plane(plane);
801bcfff 2225
2a44b76b 2226 if (intel_plane->pipe == pipe) {
7c4a395f 2227 p->spr = intel_plane->wm;
2a44b76b
VS		 2228 break;
2229 }
2230 }
2231}
2232
2233static void ilk_compute_wm_config(struct drm_device *dev,
2234 struct intel_wm_config *config)
2235{
2236 struct intel_crtc *intel_crtc;
2237
2238 /* Compute the currently _active_ config */
d3fcc808 2239 for_each_intel_crtc(dev, intel_crtc) {
2a44b76b 2240 const struct intel_pipe_wm *wm = &intel_crtc->wm.active;
cca32e9a 2241
2a44b76b
VS		 2242 if (!wm->pipe_enabled)
2243 continue;
cca32e9a 2244
2a44b76b
VS		 2245 config->sprites_enabled |= wm->sprites_enabled;
2246 config->sprites_scaled |= wm->sprites_scaled;
2247 config->num_pipes_active++;
cca32e9a 2248 }
801bcfff
PZ		 2249}
2250
0b2ae6d7
VS		 2251/* Compute new watermarks for the pipe */
2252static bool intel_compute_pipe_wm(struct drm_crtc *crtc,
820c1980 2253 const struct ilk_pipe_wm_parameters *params,
0b2ae6d7
VS		 2254 struct intel_pipe_wm *pipe_wm)
2255{
2256 struct drm_device *dev = crtc->dev;
d34ff9c6 2257 const struct drm_i915_private *dev_priv = dev->dev_private;
0b2ae6d7
VS		 2258 int level, max_level = ilk_wm_max_level(dev);
2259 /* LP0 watermark maximums depend on this pipe alone */
2260 struct intel_wm_config config = {
2261 .num_pipes_active = 1,
2262 .sprites_enabled = params->spr.enabled,
2263 .sprites_scaled = params->spr.scaled,
2264 };
820c1980 2265 struct ilk_wm_maximums max;
0b2ae6d7 2266
2a44b76b
VS		 2267 pipe_wm->pipe_enabled = params->active;
2268 pipe_wm->sprites_enabled = params->spr.enabled;
2269 pipe_wm->sprites_scaled = params->spr.scaled;
2270
7b39a0b7
VS		 2271 /* ILK/SNB: LP2+ watermarks only w/o sprites */
2272 if (INTEL_INFO(dev)->gen <= 6 && params->spr.enabled)
2273 max_level = 1;
2274
2275 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
2276 if (params->spr.scaled)
2277 max_level = 0;
2278
a3cb4048 2279 ilk_compute_wm_level(dev_priv, 0, params, &pipe_wm->wm[0]);
0b2ae6d7 2280
a42a5719 2281 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 2282 pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
0b2ae6d7 2283
a3cb4048
VS		 2284 /* LP0 watermarks always use 1/2 DDB partitioning */
2285 ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
2286
0b2ae6d7 2287 /* At least LP0 must be valid */
a3cb4048
VS		 2288 if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0]))
2289 return false;
2290
2291 ilk_compute_wm_reg_maximums(dev, 1, &max);
2292
2293 for (level = 1; level <= max_level; level++) {
2294 struct intel_wm_level wm = {};
2295
2296 ilk_compute_wm_level(dev_priv, level, params, &wm);
2297
2298 /*
2299 * Disable any watermark level that exceeds the
2300 * register maximums since such watermarks are
2301 * always invalid.
2302 */
2303 if (!ilk_validate_wm_level(level, &max, &wm))
2304 break;
2305
2306 pipe_wm->wm[level] = wm;
2307 }
2308
2309 return true;
0b2ae6d7
VS		 2310}
2311
2312/*
2313 * Merge the watermarks from all active pipes for a specific level.
2314 */
2315static void ilk_merge_wm_level(struct drm_device *dev,
2316 int level,
2317 struct intel_wm_level *ret_wm)
2318{
2319 const struct intel_crtc *intel_crtc;
2320
d52fea5b
VS		 2321 ret_wm->enable = true;
2322
d3fcc808 2323 for_each_intel_crtc(dev, intel_crtc) {
fe392efd
VS		 2324 const struct intel_pipe_wm *active = &intel_crtc->wm.active;
2325 const struct intel_wm_level *wm = &active->wm[level];
2326
2327 if (!active->pipe_enabled)
2328 continue;
0b2ae6d7 2329
d52fea5b
VS		 2330 /*
2331 * The watermark values may have been used in the past,
2332 * so we must maintain them in the registers for some
2333 * time even if the level is now disabled.
2334 */
0b2ae6d7 2335 if (!wm->enable)
d52fea5b 2336 ret_wm->enable = false;
0b2ae6d7
VS		 2337
2338 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2339 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2340 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2341 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2342 }
0b2ae6d7
VS		 2343}
2344
2345/*
2346 * Merge all low power watermarks for all active pipes.
2347 */
2348static void ilk_wm_merge(struct drm_device *dev,
0ba22e26 2349 const struct intel_wm_config *config,
820c1980 2350 const struct ilk_wm_maximums *max,
0b2ae6d7
VS		 2351 struct intel_pipe_wm *merged)
2352{
2353 int level, max_level = ilk_wm_max_level(dev);
d52fea5b 2354 int last_enabled_level = max_level;
0b2ae6d7 2355
0ba22e26
VS		 2356 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2357 if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) &&
2358 config->num_pipes_active > 1)
2359 return;
2360
6c8b6c28
VS		 2361 /* ILK: FBC WM must be disabled always */
2362 merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6;
0b2ae6d7
VS		 2363
2364 /* merge each WM1+ level */
2365 for (level = 1; level <= max_level; level++) {
2366 struct intel_wm_level *wm = &merged->wm[level];
2367
2368 ilk_merge_wm_level(dev, level, wm);
2369
d52fea5b
VS		 2370 if (level > last_enabled_level)
2371 wm->enable = false;
2372 else if (!ilk_validate_wm_level(level, max, wm))
2373 /* make sure all following levels get disabled */
2374 last_enabled_level = level - 1;
0b2ae6d7
VS		 2375
2376 /*
2377 * The spec says it is preferred to disable
2378 * FBC WMs instead of disabling a WM level.
2379 */
2380 if (wm->fbc_val > max->fbc) {
d52fea5b
VS		 2381 if (wm->enable)
2382 merged->fbc_wm_enabled = false;
0b2ae6d7
VS		 2383 wm->fbc_val = 0;
2384 }
2385 }
6c8b6c28
VS		 2386
2387 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2388 /*
2389 * FIXME this is racy. FBC might get enabled later.
2390 * What we should check here is whether FBC can be
2391 * enabled sometime later.
2392 */
2393 if (IS_GEN5(dev) && !merged->fbc_wm_enabled && intel_fbc_enabled(dev)) {
2394 for (level = 2; level <= max_level; level++) {
2395 struct intel_wm_level *wm = &merged->wm[level];
2396
2397 wm->enable = false;
2398 }
2399 }
0b2ae6d7
VS		 2400}
2401
b380ca3c
VS		 2402static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2403{
2404 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2405 return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2406}
2407
a68d68ee
VS		 2408/* The value we need to program into the WM_LPx latency field */
2409static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level)
2410{
2411 struct drm_i915_private *dev_priv = dev->dev_private;
2412
a42a5719 2413 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
a68d68ee
VS		 2414 return 2 * level;
2415 else
2416 return dev_priv->wm.pri_latency[level];
2417}
2418
820c1980 2419static void ilk_compute_wm_results(struct drm_device *dev,
0362c781 2420 const struct intel_pipe_wm *merged,
609cedef 2421 enum intel_ddb_partitioning partitioning,
820c1980 2422 struct ilk_wm_values *results)
801bcfff 2423{
0b2ae6d7
VS		 2424 struct intel_crtc *intel_crtc;
2425 int level, wm_lp;
cca32e9a 2426
0362c781 2427 results->enable_fbc_wm = merged->fbc_wm_enabled;
609cedef 2428 results->partitioning = partitioning;
cca32e9a 2429
0b2ae6d7 2430 /* LP1+ register values */
cca32e9a 2431 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
1fd527cc 2432 const struct intel_wm_level *r;
801bcfff 2433
b380ca3c 2434 level = ilk_wm_lp_to_level(wm_lp, merged);
0b2ae6d7 2435
0362c781 2436 r = &merged->wm[level];
cca32e9a 2437
d52fea5b
VS		 2438 /*
2439 * Maintain the watermark values even if the level is
2440 * disabled. Doing otherwise could cause underruns.
2441 */
2442 results->wm_lp[wm_lp - 1] =
a68d68ee 2443 (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) |
416f4727
VS		 2444 (r->pri_val << WM1_LP_SR_SHIFT) |
2445 r->cur_val;
2446
d52fea5b
VS		 2447 if (r->enable)
2448 results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN;
2449
416f4727
VS		 2450 if (INTEL_INFO(dev)->gen >= 8)
2451 results->wm_lp[wm_lp - 1] |=
2452 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
2453 else
2454 results->wm_lp[wm_lp - 1] |=
2455 r->fbc_val << WM1_LP_FBC_SHIFT;
2456
d52fea5b
VS		 2457 /*
2458 * Always set WM1S_LP_EN when spr_val != 0, even if the
2459 * level is disabled. Doing otherwise could cause underruns.
2460 */
6cef2b8a
VS		 2461 if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) {
2462 WARN_ON(wm_lp != 1);
2463 results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
2464 } else
2465 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
cca32e9a 2466 }
801bcfff 2467
0b2ae6d7 2468 /* LP0 register values */
d3fcc808 2469 for_each_intel_crtc(dev, intel_crtc) {
0b2ae6d7
VS		 2470 enum pipe pipe = intel_crtc->pipe;
2471 const struct intel_wm_level *r =
2472 &intel_crtc->wm.active.wm[0];
2473
2474 if (WARN_ON(!r->enable))
2475 continue;
2476
2477 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
1011d8c4 2478
0b2ae6d7
VS		 2479 results->wm_pipe[pipe] =
2480 (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2481 (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2482 r->cur_val;
801bcfff
PZ		 2483 }
2484}
2485
861f3389
PZ		 2486/* Find the result with the highest level enabled. Check for enable_fbc_wm in
2487 * case both are at the same level. Prefer r1 in case they're the same. */
820c1980 2488static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev,
198a1e9b
VS		 2489 struct intel_pipe_wm *r1,
2490 struct intel_pipe_wm *r2)
861f3389 2491{
198a1e9b
VS		 2492 int level, max_level = ilk_wm_max_level(dev);
2493 int level1 = 0, level2 = 0;
861f3389 2494
198a1e9b
VS		 2495 for (level = 1; level <= max_level; level++) {
2496 if (r1->wm[level].enable)
2497 level1 = level;
2498 if (r2->wm[level].enable)
2499 level2 = level;
861f3389
PZ		 2500 }
2501
198a1e9b
VS		 2502 if (level1 == level2) {
2503 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
861f3389
PZ		 2504 return r2;
2505 else
2506 return r1;
198a1e9b 2507 } else if (level1 > level2) {
861f3389
PZ		 2508 return r1;
2509 } else {
2510 return r2;
2511 }
2512}
2513
49a687c4
VS		 2514/* dirty bits used to track which watermarks need changes */
2515#define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2516#define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2517#define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2518#define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2519#define WM_DIRTY_FBC (1 << 24)
2520#define WM_DIRTY_DDB (1 << 25)
2521
2522static unsigned int ilk_compute_wm_dirty(struct drm_device *dev,
820c1980
ID		 2523 const struct ilk_wm_values *old,
2524 const struct ilk_wm_values *new)
49a687c4
VS		 2525{
2526 unsigned int dirty = 0;
2527 enum pipe pipe;
2528 int wm_lp;
2529
2530 for_each_pipe(pipe) {
2531 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
2532 dirty |= WM_DIRTY_LINETIME(pipe);
2533 /* Must disable LP1+ watermarks too */
2534 dirty |= WM_DIRTY_LP_ALL;
2535 }
2536
2537 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
2538 dirty |= WM_DIRTY_PIPE(pipe);
2539 /* Must disable LP1+ watermarks too */
2540 dirty |= WM_DIRTY_LP_ALL;
2541 }
2542 }
2543
2544 if (old->enable_fbc_wm != new->enable_fbc_wm) {
2545 dirty |= WM_DIRTY_FBC;
2546 /* Must disable LP1+ watermarks too */
2547 dirty |= WM_DIRTY_LP_ALL;
2548 }
2549
2550 if (old->partitioning != new->partitioning) {
2551 dirty |= WM_DIRTY_DDB;
2552 /* Must disable LP1+ watermarks too */
2553 dirty |= WM_DIRTY_LP_ALL;
2554 }
2555
2556 /* LP1+ watermarks already deemed dirty, no need to continue */
2557 if (dirty & WM_DIRTY_LP_ALL)
2558 return dirty;
2559
2560 /* Find the lowest numbered LP1+ watermark in need of an update... */
2561 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2562 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
2563 old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
2564 break;
2565 }
2566
2567 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2568 for (; wm_lp <= 3; wm_lp++)
2569 dirty |= WM_DIRTY_LP(wm_lp);
2570
2571 return dirty;
2572}
2573
8553c18e
VS		 2574static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
2575 unsigned int dirty)
801bcfff 2576{
820c1980 2577 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e 2578 bool changed = false;
801bcfff 2579
facd619b
VS		 2580 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
2581 previous->wm_lp[2] &= ~WM1_LP_SR_EN;
2582 I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
8553c18e 2583 changed = true;
facd619b
VS		 2584 }
2585 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
2586 previous->wm_lp[1] &= ~WM1_LP_SR_EN;
2587 I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
8553c18e 2588 changed = true;
facd619b
VS		 2589 }
2590 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
2591 previous->wm_lp[0] &= ~WM1_LP_SR_EN;
2592 I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
8553c18e 2593 changed = true;
facd619b 2594 }
801bcfff 2595
facd619b
VS		 2596 /*
2597 * Don't touch WM1S_LP_EN here.
2598 * Doing so could cause underruns.
2599 */
6cef2b8a 2600
8553c18e
VS		 2601 return changed;
2602}
2603
2604/*
2605 * The spec says we shouldn't write when we don't need, because every write
2606 * causes WMs to be re-evaluated, expending some power.
2607 */
820c1980
ID		 2608static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
2609 struct ilk_wm_values *results)
8553c18e
VS		 2610{
2611 struct drm_device *dev = dev_priv->dev;
820c1980 2612 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e
VS		 2613 unsigned int dirty;
2614 uint32_t val;
2615
2616 dirty = ilk_compute_wm_dirty(dev, previous, results);
2617 if (!dirty)
2618 return;
2619
2620 _ilk_disable_lp_wm(dev_priv, dirty);
2621
49a687c4 2622 if (dirty & WM_DIRTY_PIPE(PIPE_A))
801bcfff 2623 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
49a687c4 2624 if (dirty & WM_DIRTY_PIPE(PIPE_B))
801bcfff 2625 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
49a687c4 2626 if (dirty & WM_DIRTY_PIPE(PIPE_C))
801bcfff
PZ		 2627 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2628
49a687c4 2629 if (dirty & WM_DIRTY_LINETIME(PIPE_A))
801bcfff 2630 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
49a687c4 2631 if (dirty & WM_DIRTY_LINETIME(PIPE_B))
801bcfff 2632 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
49a687c4 2633 if (dirty & WM_DIRTY_LINETIME(PIPE_C))
801bcfff
PZ		 2634 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2635
49a687c4 2636 if (dirty & WM_DIRTY_DDB) {
a42a5719 2637 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
ac9545fd
VS		 2638 val = I915_READ(WM_MISC);
2639 if (results->partitioning == INTEL_DDB_PART_1_2)
2640 val &= ~WM_MISC_DATA_PARTITION_5_6;
2641 else
2642 val |= WM_MISC_DATA_PARTITION_5_6;
2643 I915_WRITE(WM_MISC, val);
2644 } else {
2645 val = I915_READ(DISP_ARB_CTL2);
2646 if (results->partitioning == INTEL_DDB_PART_1_2)
2647 val &= ~DISP_DATA_PARTITION_5_6;
2648 else
2649 val |= DISP_DATA_PARTITION_5_6;
2650 I915_WRITE(DISP_ARB_CTL2, val);
2651 }
1011d8c4
PZ		 2652 }
2653
49a687c4 2654 if (dirty & WM_DIRTY_FBC) {
cca32e9a
PZ		 2655 val = I915_READ(DISP_ARB_CTL);
2656 if (results->enable_fbc_wm)
2657 val &= ~DISP_FBC_WM_DIS;
2658 else
2659 val |= DISP_FBC_WM_DIS;
2660 I915_WRITE(DISP_ARB_CTL, val);
2661 }
2662
954911eb
ID		 2663 if (dirty & WM_DIRTY_LP(1) &&
2664 previous->wm_lp_spr[0] != results->wm_lp_spr[0])
2665 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2666
2667 if (INTEL_INFO(dev)->gen >= 7) {
6cef2b8a
VS		 2668 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
2669 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2670 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
2671 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2672 }
801bcfff 2673
facd619b 2674 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
801bcfff 2675 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
facd619b 2676 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
801bcfff 2677 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
facd619b 2678 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
801bcfff 2679 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
609cedef
VS		 2680
2681 dev_priv->wm.hw = *results;
801bcfff
PZ		 2682}
2683
8553c18e
VS		 2684static bool ilk_disable_lp_wm(struct drm_device *dev)
2685{
2686 struct drm_i915_private *dev_priv = dev->dev_private;
2687
2688 return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
2689}
2690
820c1980 2691static void ilk_update_wm(struct drm_crtc *crtc)
801bcfff 2692{
7c4a395f 2693 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
46ba614c 2694 struct drm_device *dev = crtc->dev;
801bcfff 2695 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980
ID		 2696 struct ilk_wm_maximums max;
2697 struct ilk_pipe_wm_parameters params = {};
2698 struct ilk_wm_values results = {};
77c122bc 2699 enum intel_ddb_partitioning partitioning;
7c4a395f 2700 struct intel_pipe_wm pipe_wm = {};
198a1e9b 2701 struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
a485bfb8 2702 struct intel_wm_config config = {};
7c4a395f 2703
2a44b76b 2704 ilk_compute_wm_parameters(crtc, &params);
7c4a395f
VS		 2705
2706 intel_compute_pipe_wm(crtc, &params, &pipe_wm);
2707
2708 if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
2709 return;
861f3389 2710
7c4a395f 2711 intel_crtc->wm.active = pipe_wm;
861f3389 2712
2a44b76b
VS		 2713 ilk_compute_wm_config(dev, &config);
2714
34982fe1 2715 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
0ba22e26 2716 ilk_wm_merge(dev, &config, &max, &lp_wm_1_2);
a485bfb8
VS		 2717
2718 /* 5/6 split only in single pipe config on IVB+ */
ec98c8d1
VS		 2719 if (INTEL_INFO(dev)->gen >= 7 &&
2720 config.num_pipes_active == 1 && config.sprites_enabled) {
34982fe1 2721 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
0ba22e26 2722 ilk_wm_merge(dev, &config, &max, &lp_wm_5_6);
0362c781 2723
820c1980 2724 best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
861f3389 2725 } else {
198a1e9b 2726 best_lp_wm = &lp_wm_1_2;
861f3389
PZ		 2727 }
2728
198a1e9b 2729 partitioning = (best_lp_wm == &lp_wm_1_2) ?
77c122bc 2730 INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
801bcfff 2731
820c1980 2732 ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results);
609cedef 2733
820c1980 2734 ilk_write_wm_values(dev_priv, &results);
1011d8c4
PZ		 2735}
2736
820c1980 2737static void ilk_update_sprite_wm(struct drm_plane *plane,
adf3d35e 2738 struct drm_crtc *crtc,
526682e9 2739 uint32_t sprite_width, int pixel_size,
bdd57d03 2740 bool enabled, bool scaled)
526682e9 2741{
8553c18e 2742 struct drm_device *dev = plane->dev;
adf3d35e 2743 struct intel_plane *intel_plane = to_intel_plane(plane);
526682e9 2744
adf3d35e
VS		 2745 intel_plane->wm.enabled = enabled;
2746 intel_plane->wm.scaled = scaled;
2747 intel_plane->wm.horiz_pixels = sprite_width;
2748 intel_plane->wm.bytes_per_pixel = pixel_size;
526682e9 2749
8553c18e
VS		 2750 /*
2751 * IVB workaround: must disable low power watermarks for at least
2752 * one frame before enabling scaling. LP watermarks can be re-enabled
2753 * when scaling is disabled.
2754 *
2755 * WaCxSRDisabledForSpriteScaling:ivb
2756 */
2757 if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev))
2758 intel_wait_for_vblank(dev, intel_plane->pipe);
2759
820c1980 2760 ilk_update_wm(crtc);
526682e9
PZ		 2761}
2762
243e6a44
VS		 2763static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
2764{
2765 struct drm_device *dev = crtc->dev;
2766 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 2767 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS		 2768 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2769 struct intel_pipe_wm *active = &intel_crtc->wm.active;
2770 enum pipe pipe = intel_crtc->pipe;
2771 static const unsigned int wm0_pipe_reg[] = {
2772 [PIPE_A] = WM0_PIPEA_ILK,
2773 [PIPE_B] = WM0_PIPEB_ILK,
2774 [PIPE_C] = WM0_PIPEC_IVB,
2775 };
2776
2777 hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
a42a5719 2778 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 2779 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
243e6a44 2780
2a44b76b
VS		 2781 active->pipe_enabled = intel_crtc_active(crtc);
2782
2783 if (active->pipe_enabled) {
243e6a44
VS		 2784 u32 tmp = hw->wm_pipe[pipe];
2785
2786 /*
2787 * For active pipes LP0 watermark is marked as
2788 * enabled, and LP1+ watermaks as disabled since
2789 * we can't really reverse compute them in case
2790 * multiple pipes are active.
2791 */
2792 active->wm[0].enable = true;
2793 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
2794 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
2795 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
2796 active->linetime = hw->wm_linetime[pipe];
2797 } else {
2798 int level, max_level = ilk_wm_max_level(dev);
2799
2800 /*
2801 * For inactive pipes, all watermark levels
2802 * should be marked as enabled but zeroed,
2803 * which is what we'd compute them to.
2804 */
2805 for (level = 0; level <= max_level; level++)
2806 active->wm[level].enable = true;
2807 }
2808}
2809
2810void ilk_wm_get_hw_state(struct drm_device *dev)
2811{
2812 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 2813 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS		 2814 struct drm_crtc *crtc;
2815
70e1e0ec 2816 for_each_crtc(dev, crtc)
243e6a44
VS		 2817 ilk_pipe_wm_get_hw_state(crtc);
2818
2819 hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
2820 hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
2821 hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
2822
2823 hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
cfa7698b
VS		 2824 if (INTEL_INFO(dev)->gen >= 7) {
2825 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
2826 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
2827 }
243e6a44 2828
a42a5719 2829 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ac9545fd
VS		 2830 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
2831 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
2832 else if (IS_IVYBRIDGE(dev))
2833 hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
2834 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
243e6a44
VS		 2835
2836 hw->enable_fbc_wm =
2837 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
2838}
2839
b445e3b0
ED		 2840/**
2841 * intel_update_watermarks - update FIFO watermark values based on current modes
2842 *
2843 * Calculate watermark values for the various WM regs based on current mode
2844 * and plane configuration.
2845 *
2846 * There are several cases to deal with here:
2847 * - normal (i.e. non-self-refresh)
2848 * - self-refresh (SR) mode
2849 * - lines are large relative to FIFO size (buffer can hold up to 2)
2850 * - lines are small relative to FIFO size (buffer can hold more than 2
2851 * lines), so need to account for TLB latency
2852 *
2853 * The normal calculation is:
2854 * watermark = dotclock * bytes per pixel * latency
2855 * where latency is platform & configuration dependent (we assume pessimal
2856 * values here).
2857 *
2858 * The SR calculation is:
2859 * watermark = (trunc(latency/line time)+1) * surface width *
2860 * bytes per pixel
2861 * where
2862 * line time = htotal / dotclock
2863 * surface width = hdisplay for normal plane and 64 for cursor
2864 * and latency is assumed to be high, as above.
2865 *
2866 * The final value programmed to the register should always be rounded up,
2867 * and include an extra 2 entries to account for clock crossings.
2868 *
2869 * We don't use the sprite, so we can ignore that. And on Crestline we have
2870 * to set the non-SR watermarks to 8.
2871 */
46ba614c 2872void intel_update_watermarks(struct drm_crtc *crtc)
b445e3b0 2873{
46ba614c 2874 struct drm_i915_private *dev_priv = crtc->dev->dev_private;
b445e3b0
ED		 2875
2876 if (dev_priv->display.update_wm)
46ba614c 2877 dev_priv->display.update_wm(crtc);
b445e3b0
ED		 2878}
2879
adf3d35e
VS		 2880void intel_update_sprite_watermarks(struct drm_plane *plane,
2881 struct drm_crtc *crtc,
4c4ff43a 2882 uint32_t sprite_width, int pixel_size,
39db4a4d 2883 bool enabled, bool scaled)
b445e3b0 2884{
adf3d35e 2885 struct drm_i915_private *dev_priv = plane->dev->dev_private;
b445e3b0
ED		 2886
2887 if (dev_priv->display.update_sprite_wm)
adf3d35e 2888 dev_priv->display.update_sprite_wm(plane, crtc, sprite_width,
39db4a4d 2889 pixel_size, enabled, scaled);
b445e3b0
ED		 2890}
2891
2b4e57bd
ED		 2892static struct drm_i915_gem_object *
2893intel_alloc_context_page(struct drm_device *dev)
2894{
2895 struct drm_i915_gem_object *ctx;
2896 int ret;
2897
2898 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2899
2900 ctx = i915_gem_alloc_object(dev, 4096);
2901 if (!ctx) {
2902 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2903 return NULL;
2904 }
2905
c69766f2 2906 ret = i915_gem_obj_ggtt_pin(ctx, 4096, 0);
2b4e57bd
ED		 2907 if (ret) {
2908 DRM_ERROR("failed to pin power context: %d\n", ret);
2909 goto err_unref;
2910 }
2911
2912 ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
2913 if (ret) {
2914 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
2915 goto err_unpin;
2916 }
2917
2918 return ctx;
2919
2920err_unpin:
d7f46fc4 2921 i915_gem_object_ggtt_unpin(ctx);
2b4e57bd
ED		 2922err_unref:
2923 drm_gem_object_unreference(&ctx->base);
2b4e57bd
ED		 2924 return NULL;
2925}
2926
9270388e
DV		 2927/**
2928 * Lock protecting IPS related data structures
9270388e
DV		 2929 */
2930DEFINE_SPINLOCK(mchdev_lock);
2931
2932/* Global for IPS driver to get at the current i915 device. Protected by
2933 * mchdev_lock. */
2934static struct drm_i915_private *i915_mch_dev;
2935
2b4e57bd
ED		 2936bool ironlake_set_drps(struct drm_device *dev, u8 val)
2937{
2938 struct drm_i915_private *dev_priv = dev->dev_private;
2939 u16 rgvswctl;
2940
9270388e
DV		 2941 assert_spin_locked(&mchdev_lock);
2942
2b4e57bd
ED		 2943 rgvswctl = I915_READ16(MEMSWCTL);
2944 if (rgvswctl & MEMCTL_CMD_STS) {
2945 DRM_DEBUG("gpu busy, RCS change rejected\n");
2946 return false; /* still busy with another command */
2947 }
2948
2949 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
2950 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
2951 I915_WRITE16(MEMSWCTL, rgvswctl);
2952 POSTING_READ16(MEMSWCTL);
2953
2954 rgvswctl |= MEMCTL_CMD_STS;
2955 I915_WRITE16(MEMSWCTL, rgvswctl);
2956
2957 return true;
2958}
2959
8090c6b9 2960static void ironlake_enable_drps(struct drm_device *dev)
2b4e57bd
ED		 2961{
2962 struct drm_i915_private *dev_priv = dev->dev_private;
2963 u32 rgvmodectl = I915_READ(MEMMODECTL);
2964 u8 fmax, fmin, fstart, vstart;
2965
9270388e
DV		 2966 spin_lock_irq(&mchdev_lock);
2967
2b4e57bd
ED		 2968 /* Enable temp reporting */
2969 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
2970 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
2971
2972 /* 100ms RC evaluation intervals */
2973 I915_WRITE(RCUPEI, 100000);
2974 I915_WRITE(RCDNEI, 100000);
2975
2976 /* Set max/min thresholds to 90ms and 80ms respectively */
2977 I915_WRITE(RCBMAXAVG, 90000);
2978 I915_WRITE(RCBMINAVG, 80000);
2979
2980 I915_WRITE(MEMIHYST, 1);
2981
2982 /* Set up min, max, and cur for interrupt handling */
2983 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
2984 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
2985 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
2986 MEMMODE_FSTART_SHIFT;
2987
2988 vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
2989 PXVFREQ_PX_SHIFT;
2990
20e4d407
DV		 2991 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
2992 dev_priv->ips.fstart = fstart;
2b4e57bd 2993
20e4d407
DV		 2994 dev_priv->ips.max_delay = fstart;
2995 dev_priv->ips.min_delay = fmin;
2996 dev_priv->ips.cur_delay = fstart;
2b4e57bd
ED		 2997
2998 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2999 fmax, fmin, fstart);
3000
3001 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
3002
3003 /*
3004 * Interrupts will be enabled in ironlake_irq_postinstall
3005 */
3006
3007 I915_WRITE(VIDSTART, vstart);
3008 POSTING_READ(VIDSTART);
3009
3010 rgvmodectl |= MEMMODE_SWMODE_EN;
3011 I915_WRITE(MEMMODECTL, rgvmodectl);
3012
9270388e 3013 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2b4e57bd 3014 DRM_ERROR("stuck trying to change perf mode\n");
9270388e 3015 mdelay(1);
2b4e57bd
ED		 3016
3017 ironlake_set_drps(dev, fstart);
3018
20e4d407 3019 dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
2b4e57bd 3020 I915_READ(0x112e0);
20e4d407
DV		 3021 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
3022 dev_priv->ips.last_count2 = I915_READ(0x112f4);
3023 getrawmonotonic(&dev_priv->ips.last_time2);
9270388e
DV		 3024
3025 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED		 3026}
3027
8090c6b9 3028static void ironlake_disable_drps(struct drm_device *dev)
2b4e57bd
ED		 3029{
3030 struct drm_i915_private *dev_priv = dev->dev_private;
9270388e
DV		 3031 u16 rgvswctl;
3032
3033 spin_lock_irq(&mchdev_lock);
3034
3035 rgvswctl = I915_READ16(MEMSWCTL);
2b4e57bd
ED		 3036
3037 /* Ack interrupts, disable EFC interrupt */
3038 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
3039 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
3040 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
3041 I915_WRITE(DEIIR, DE_PCU_EVENT);
3042 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
3043
3044 /* Go back to the starting frequency */
20e4d407 3045 ironlake_set_drps(dev, dev_priv->ips.fstart);
9270388e 3046 mdelay(1);
2b4e57bd
ED		 3047 rgvswctl |= MEMCTL_CMD_STS;
3048 I915_WRITE(MEMSWCTL, rgvswctl);
9270388e 3049 mdelay(1);
2b4e57bd 3050
9270388e 3051 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED		 3052}
3053
acbe9475
DV		 3054/* There's a funny hw issue where the hw returns all 0 when reading from
3055 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
3056 * ourselves, instead of doing a rmw cycle (which might result in us clearing
3057 * all limits and the gpu stuck at whatever frequency it is at atm).
3058 */
6917c7b9 3059static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 val)
2b4e57bd 3060{
7b9e0ae6 3061 u32 limits;
2b4e57bd 3062
20b46e59
DV		 3063 /* Only set the down limit when we've reached the lowest level to avoid
3064 * getting more interrupts, otherwise leave this clear. This prevents a
3065 * race in the hw when coming out of rc6: There's a tiny window where
3066 * the hw runs at the minimal clock before selecting the desired
3067 * frequency, if the down threshold expires in that window we will not
3068 * receive a down interrupt. */
b39fb297
BW		 3069 limits = dev_priv->rps.max_freq_softlimit << 24;
3070 if (val <= dev_priv->rps.min_freq_softlimit)
3071 limits |= dev_priv->rps.min_freq_softlimit << 16;
20b46e59
DV		 3072
3073 return limits;
3074}
3075
dd75fdc8
CW		 3076static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
3077{
3078 int new_power;
3079
3080 new_power = dev_priv->rps.power;
3081 switch (dev_priv->rps.power) {
3082 case LOW_POWER:
b39fb297 3083 if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq)
dd75fdc8
CW		 3084 new_power = BETWEEN;
3085 break;
3086
3087 case BETWEEN:
b39fb297 3088 if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq)
dd75fdc8 3089 new_power = LOW_POWER;
b39fb297 3090 else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq)
dd75fdc8
CW		 3091 new_power = HIGH_POWER;
3092 break;
3093
3094 case HIGH_POWER:
b39fb297 3095 if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq)
dd75fdc8
CW		 3096 new_power = BETWEEN;
3097 break;
3098 }
3099 /* Max/min bins are special */
b39fb297 3100 if (val == dev_priv->rps.min_freq_softlimit)
dd75fdc8 3101 new_power = LOW_POWER;
b39fb297 3102 if (val == dev_priv->rps.max_freq_softlimit)
dd75fdc8
CW		 3103 new_power = HIGH_POWER;
3104 if (new_power == dev_priv->rps.power)
3105 return;
3106
3107 /* Note the units here are not exactly 1us, but 1280ns. */
3108 switch (new_power) {
3109 case LOW_POWER:
3110 /* Upclock if more than 95% busy over 16ms */
3111 I915_WRITE(GEN6_RP_UP_EI, 12500);
3112 I915_WRITE(GEN6_RP_UP_THRESHOLD, 11800);
3113
3114 /* Downclock if less than 85% busy over 32ms */
3115 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3116 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 21250);
3117
3118 I915_WRITE(GEN6_RP_CONTROL,
3119 GEN6_RP_MEDIA_TURBO |
3120 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3121 GEN6_RP_MEDIA_IS_GFX |
3122 GEN6_RP_ENABLE |
3123 GEN6_RP_UP_BUSY_AVG |
3124 GEN6_RP_DOWN_IDLE_AVG);
3125 break;
3126
3127 case BETWEEN:
3128 /* Upclock if more than 90% busy over 13ms */
3129 I915_WRITE(GEN6_RP_UP_EI, 10250);
3130 I915_WRITE(GEN6_RP_UP_THRESHOLD, 9225);
3131
3132 /* Downclock if less than 75% busy over 32ms */
3133 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3134 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 18750);
3135
3136 I915_WRITE(GEN6_RP_CONTROL,
3137 GEN6_RP_MEDIA_TURBO |
3138 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3139 GEN6_RP_MEDIA_IS_GFX |
3140 GEN6_RP_ENABLE |
3141 GEN6_RP_UP_BUSY_AVG |
3142 GEN6_RP_DOWN_IDLE_AVG);
3143 break;
3144
3145 case HIGH_POWER:
3146 /* Upclock if more than 85% busy over 10ms */
3147 I915_WRITE(GEN6_RP_UP_EI, 8000);
3148 I915_WRITE(GEN6_RP_UP_THRESHOLD, 6800);
3149
3150 /* Downclock if less than 60% busy over 32ms */
3151 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3152 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 15000);
3153
3154 I915_WRITE(GEN6_RP_CONTROL,
3155 GEN6_RP_MEDIA_TURBO |
3156 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3157 GEN6_RP_MEDIA_IS_GFX |
3158 GEN6_RP_ENABLE |
3159 GEN6_RP_UP_BUSY_AVG |
3160 GEN6_RP_DOWN_IDLE_AVG);
3161 break;
3162 }
3163
3164 dev_priv->rps.power = new_power;
3165 dev_priv->rps.last_adj = 0;
3166}
3167
2876ce73
CW		 3168static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val)
3169{
3170 u32 mask = 0;
3171
3172 if (val > dev_priv->rps.min_freq_softlimit)
3173 mask |= GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT;
3174 if (val < dev_priv->rps.max_freq_softlimit)
3175 mask |= GEN6_PM_RP_UP_THRESHOLD;
3176
3177 /* IVB and SNB hard hangs on looping batchbuffer
3178 * if GEN6_PM_UP_EI_EXPIRED is masked.
3179 */
3180 if (INTEL_INFO(dev_priv->dev)->gen <= 7 && !IS_HASWELL(dev_priv->dev))
3181 mask |= GEN6_PM_RP_UP_EI_EXPIRED;
3182
baccd458
D		 3183 if (IS_GEN8(dev_priv->dev))
3184 mask |= GEN8_PMINTR_REDIRECT_TO_NON_DISP;
3185
2876ce73
CW		 3186 return ~mask;
3187}
3188
b8a5ff8d
JM		 3189/* gen6_set_rps is called to update the frequency request, but should also be
3190 * called when the range (min_delay and max_delay) is modified so that we can
3191 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
20b46e59
DV		 3192void gen6_set_rps(struct drm_device *dev, u8 val)
3193{
3194 struct drm_i915_private *dev_priv = dev->dev_private;
7b9e0ae6 3195
4fc688ce 3196 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
b39fb297
BW		 3197 WARN_ON(val > dev_priv->rps.max_freq_softlimit);
3198 WARN_ON(val < dev_priv->rps.min_freq_softlimit);
004777cb 3199
eb64cad1
CW		 3200 /* min/max delay may still have been modified so be sure to
3201 * write the limits value.
3202 */
3203 if (val != dev_priv->rps.cur_freq) {
3204 gen6_set_rps_thresholds(dev_priv, val);
b8a5ff8d 3205
50e6a2a7 3206 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
eb64cad1
CW		 3207 I915_WRITE(GEN6_RPNSWREQ,
3208 HSW_FREQUENCY(val));
3209 else
3210 I915_WRITE(GEN6_RPNSWREQ,
3211 GEN6_FREQUENCY(val) |
3212 GEN6_OFFSET(0) |
3213 GEN6_AGGRESSIVE_TURBO);
b8a5ff8d 3214 }
7b9e0ae6 3215
7b9e0ae6
CW		 3216 /* Make sure we continue to get interrupts
3217 * until we hit the minimum or maximum frequencies.
3218 */
eb64cad1 3219 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, gen6_rps_limits(dev_priv, val));
2876ce73 3220 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
7b9e0ae6 3221
d5570a72
BW		 3222 POSTING_READ(GEN6_RPNSWREQ);
3223
b39fb297 3224 dev_priv->rps.cur_freq = val;
be2cde9a 3225 trace_intel_gpu_freq_change(val * 50);
2b4e57bd
ED		 3226}
3227
76c3552f
D		 3228/* vlv_set_rps_idle: Set the frequency to Rpn if Gfx clocks are down
3229 *
3230 * * If Gfx is Idle, then
3231 * 1. Mask Turbo interrupts
3232 * 2. Bring up Gfx clock
3233 * 3. Change the freq to Rpn and wait till P-Unit updates freq
3234 * 4. Clear the Force GFX CLK ON bit so that Gfx can down
3235 * 5. Unmask Turbo interrupts
3236*/
3237static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
3238{
5549d25f
D		 3239 struct drm_device *dev = dev_priv->dev;
3240
3241 /* Latest VLV doesn't need to force the gfx clock */
3242 if (dev->pdev->revision >= 0xd) {
3243 valleyview_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
3244 return;
3245 }
3246
76c3552f
D		 3247 /*
3248 * When we are idle. Drop to min voltage state.
3249 */
3250
b39fb297 3251 if (dev_priv->rps.cur_freq <= dev_priv->rps.min_freq_softlimit)
76c3552f
D		 3252 return;
3253
3254 /* Mask turbo interrupt so that they will not come in between */
3255 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
3256
650ad970 3257 vlv_force_gfx_clock(dev_priv, true);
76c3552f 3258
b39fb297 3259 dev_priv->rps.cur_freq = dev_priv->rps.min_freq_softlimit;
76c3552f
D		 3260
3261 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ,
b39fb297 3262 dev_priv->rps.min_freq_softlimit);
76c3552f
D		 3263
3264 if (wait_for(((vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS))
3265 & GENFREQSTATUS) == 0, 5))
3266 DRM_ERROR("timed out waiting for Punit\n");
3267
650ad970 3268 vlv_force_gfx_clock(dev_priv, false);
76c3552f 3269
2876ce73
CW		 3270 I915_WRITE(GEN6_PMINTRMSK,
3271 gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq));
76c3552f
D		 3272}
3273
b29c19b6
CW		 3274void gen6_rps_idle(struct drm_i915_private *dev_priv)
3275{
691bb717
DL		 3276 struct drm_device *dev = dev_priv->dev;
3277
b29c19b6 3278 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 3279 if (dev_priv->rps.enabled) {
691bb717 3280 if (IS_VALLEYVIEW(dev))
76c3552f 3281 vlv_set_rps_idle(dev_priv);
c0951f0c 3282 else
b39fb297 3283 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
c0951f0c
CW		 3284 dev_priv->rps.last_adj = 0;
3285 }
b29c19b6
CW		 3286 mutex_unlock(&dev_priv->rps.hw_lock);
3287}
3288
3289void gen6_rps_boost(struct drm_i915_private *dev_priv)
3290{
691bb717
DL		 3291 struct drm_device *dev = dev_priv->dev;
3292
b29c19b6 3293 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 3294 if (dev_priv->rps.enabled) {
691bb717 3295 if (IS_VALLEYVIEW(dev))
b39fb297 3296 valleyview_set_rps(dev_priv->dev, dev_priv->rps.max_freq_softlimit);
c0951f0c 3297 else
b39fb297 3298 gen6_set_rps(dev_priv->dev, dev_priv->rps.max_freq_softlimit);
c0951f0c
CW		 3299 dev_priv->rps.last_adj = 0;
3300 }
b29c19b6
CW		 3301 mutex_unlock(&dev_priv->rps.hw_lock);
3302}
3303
0a073b84
JB		 3304void valleyview_set_rps(struct drm_device *dev, u8 val)
3305{
3306 struct drm_i915_private *dev_priv = dev->dev_private;
7a67092a 3307
0a073b84 3308 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
b39fb297
BW		 3309 WARN_ON(val > dev_priv->rps.max_freq_softlimit);
3310 WARN_ON(val < dev_priv->rps.min_freq_softlimit);
0a073b84 3311
73008b98 3312 DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
b39fb297
BW		 3313 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
3314 dev_priv->rps.cur_freq,
2ec3815f 3315 vlv_gpu_freq(dev_priv, val), val);
0a073b84 3316
2876ce73
CW		 3317 if (val != dev_priv->rps.cur_freq)
3318 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
0a073b84 3319
09c87db8 3320 I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
0a073b84 3321
b39fb297 3322 dev_priv->rps.cur_freq = val;
2ec3815f 3323 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv, val));
0a073b84
JB		 3324}
3325
0961021a
BW		 3326static void gen8_disable_rps_interrupts(struct drm_device *dev)
3327{
3328 struct drm_i915_private *dev_priv = dev->dev_private;
3329
992f191f 3330 I915_WRITE(GEN6_PMINTRMSK, ~GEN8_PMINTR_REDIRECT_TO_NON_DISP);
0961021a
BW		 3331 I915_WRITE(GEN8_GT_IER(2), I915_READ(GEN8_GT_IER(2)) &
3332 ~dev_priv->pm_rps_events);
3333 /* Complete PM interrupt masking here doesn't race with the rps work
3334 * item again unmasking PM interrupts because that is using a different
3335 * register (GEN8_GT_IMR(2)) to mask PM interrupts. The only risk is in
3336 * leaving stale bits in GEN8_GT_IIR(2) and GEN8_GT_IMR(2) which
3337 * gen8_enable_rps will clean up. */
3338
3339 spin_lock_irq(&dev_priv->irq_lock);
3340 dev_priv->rps.pm_iir = 0;
3341 spin_unlock_irq(&dev_priv->irq_lock);
3342
3343 I915_WRITE(GEN8_GT_IIR(2), dev_priv->pm_rps_events);
3344}
3345
44fc7d5c 3346static void gen6_disable_rps_interrupts(struct drm_device *dev)
2b4e57bd
ED		 3347{
3348 struct drm_i915_private *dev_priv = dev->dev_private;
3349
2b4e57bd 3350 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
a6706b45
D		 3351 I915_WRITE(GEN6_PMIER, I915_READ(GEN6_PMIER) &
3352 ~dev_priv->pm_rps_events);
2b4e57bd
ED		 3353 /* Complete PM interrupt masking here doesn't race with the rps work
3354 * item again unmasking PM interrupts because that is using a different
3355 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
3356 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
3357
59cdb63d 3358 spin_lock_irq(&dev_priv->irq_lock);
c6a828d3 3359 dev_priv->rps.pm_iir = 0;
59cdb63d 3360 spin_unlock_irq(&dev_priv->irq_lock);
2b4e57bd 3361
a6706b45 3362 I915_WRITE(GEN6_PMIIR, dev_priv->pm_rps_events);
2b4e57bd
ED		 3363}
3364
44fc7d5c 3365static void gen6_disable_rps(struct drm_device *dev)
d20d4f0c
JB		 3366{
3367 struct drm_i915_private *dev_priv = dev->dev_private;
3368
3369 I915_WRITE(GEN6_RC_CONTROL, 0);
44fc7d5c 3370 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
d20d4f0c 3371
0961021a
BW		 3372 if (IS_BROADWELL(dev))
3373 gen8_disable_rps_interrupts(dev);
3374 else
3375 gen6_disable_rps_interrupts(dev);
44fc7d5c
DV		 3376}
3377
38807746
D		 3378static void cherryview_disable_rps(struct drm_device *dev)
3379{
3380 struct drm_i915_private *dev_priv = dev->dev_private;
3381
3382 I915_WRITE(GEN6_RC_CONTROL, 0);
3383}
3384
44fc7d5c
DV		 3385static void valleyview_disable_rps(struct drm_device *dev)
3386{
3387 struct drm_i915_private *dev_priv = dev->dev_private;
3388
3389 I915_WRITE(GEN6_RC_CONTROL, 0);
d20d4f0c 3390
44fc7d5c 3391 gen6_disable_rps_interrupts(dev);
d20d4f0c
JB		 3392}
3393
dc39fff7
BW		 3394static void intel_print_rc6_info(struct drm_device *dev, u32 mode)
3395{
91ca689a
ID		 3396 if (IS_VALLEYVIEW(dev)) {
3397 if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1)))
3398 mode = GEN6_RC_CTL_RC6_ENABLE;
3399 else
3400 mode = 0;
3401 }
dc39fff7 3402 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
1c79b42f
BW		 3403 (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
3404 (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
3405 (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
dc39fff7
BW		 3406}
3407
e6069ca8 3408static int sanitize_rc6_option(const struct drm_device *dev, int enable_rc6)
2b4e57bd 3409{
eb4926e4
DL		 3410 /* No RC6 before Ironlake */
3411 if (INTEL_INFO(dev)->gen < 5)
3412 return 0;
3413
e6069ca8
ID		 3414 /* RC6 is only on Ironlake mobile not on desktop */
3415 if (INTEL_INFO(dev)->gen == 5 && !IS_IRONLAKE_M(dev))
3416 return 0;
3417
456470eb 3418 /* Respect the kernel parameter if it is set */
e6069ca8
ID		 3419 if (enable_rc6 >= 0) {
3420 int mask;
3421
3422 if (INTEL_INFO(dev)->gen == 6 || IS_IVYBRIDGE(dev))
3423 mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE |
3424 INTEL_RC6pp_ENABLE;
3425 else
3426 mask = INTEL_RC6_ENABLE;
3427
3428 if ((enable_rc6 & mask) != enable_rc6)
3429 DRM_INFO("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
8fd9c1a9 3430 enable_rc6 & mask, enable_rc6, mask);
e6069ca8
ID		 3431
3432 return enable_rc6 & mask;
3433 }
2b4e57bd 3434
6567d748
CW		 3435 /* Disable RC6 on Ironlake */
3436 if (INTEL_INFO(dev)->gen == 5)
3437 return 0;
2b4e57bd 3438
8bade1ad 3439 if (IS_IVYBRIDGE(dev))
cca84a1f 3440 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
8bade1ad
BW		 3441
3442 return INTEL_RC6_ENABLE;
2b4e57bd
ED		 3443}
3444
e6069ca8
ID		 3445int intel_enable_rc6(const struct drm_device *dev)
3446{
3447 return i915.enable_rc6;
3448}
3449
0961021a
BW		 3450static void gen8_enable_rps_interrupts(struct drm_device *dev)
3451{
3452 struct drm_i915_private *dev_priv = dev->dev_private;
3453
3454 spin_lock_irq(&dev_priv->irq_lock);
3455 WARN_ON(dev_priv->rps.pm_iir);
3456 bdw_enable_pm_irq(dev_priv, dev_priv->pm_rps_events);
3457 I915_WRITE(GEN8_GT_IIR(2), dev_priv->pm_rps_events);
3458 spin_unlock_irq(&dev_priv->irq_lock);
3459}
3460
44fc7d5c
DV		 3461static void gen6_enable_rps_interrupts(struct drm_device *dev)
3462{
3463 struct drm_i915_private *dev_priv = dev->dev_private;
3464
3465 spin_lock_irq(&dev_priv->irq_lock);
a0b3335a 3466 WARN_ON(dev_priv->rps.pm_iir);
a6706b45
D		 3467 snb_enable_pm_irq(dev_priv, dev_priv->pm_rps_events);
3468 I915_WRITE(GEN6_PMIIR, dev_priv->pm_rps_events);
44fc7d5c 3469 spin_unlock_irq(&dev_priv->irq_lock);
44fc7d5c
DV		 3470}
3471
3280e8b0
BW		 3472static void parse_rp_state_cap(struct drm_i915_private *dev_priv, u32 rp_state_cap)
3473{
3474 /* All of these values are in units of 50MHz */
3475 dev_priv->rps.cur_freq = 0;
3476 /* static values from HW: RP0 < RPe < RP1 < RPn (min_freq) */
3477 dev_priv->rps.rp1_freq = (rp_state_cap >> 8) & 0xff;
3478 dev_priv->rps.rp0_freq = (rp_state_cap >> 0) & 0xff;
3479 dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff;
3480 /* XXX: only BYT has a special efficient freq */
3481 dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq;
3482 /* hw_max = RP0 until we check for overclocking */
3483 dev_priv->rps.max_freq = dev_priv->rps.rp0_freq;
3484
3485 /* Preserve min/max settings in case of re-init */
3486 if (dev_priv->rps.max_freq_softlimit == 0)
3487 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
3488
3489 if (dev_priv->rps.min_freq_softlimit == 0)
3490 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
3491}
3492
6edee7f3
BW		 3493static void gen8_enable_rps(struct drm_device *dev)
3494{
3495 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 3496 struct intel_engine_cs *ring;
6edee7f3
BW		 3497 uint32_t rc6_mask = 0, rp_state_cap;
3498 int unused;
3499
3500 /* 1a: Software RC state - RC0 */
3501 I915_WRITE(GEN6_RC_STATE, 0);
3502
3503 /* 1c & 1d: Get forcewake during program sequence. Although the driver
3504 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
c8d9a590 3505 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW		 3506
3507 /* 2a: Disable RC states. */
3508 I915_WRITE(GEN6_RC_CONTROL, 0);
3509
3510 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3280e8b0 3511 parse_rp_state_cap(dev_priv, rp_state_cap);
6edee7f3
BW		 3512
3513 /* 2b: Program RC6 thresholds.*/
3514 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
3515 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
3516 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
3517 for_each_ring(ring, dev_priv, unused)
3518 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3519 I915_WRITE(GEN6_RC_SLEEP, 0);
0d68b25e
TR		 3520 if (IS_BROADWELL(dev))
3521 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */
3522 else
3523 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
6edee7f3
BW		 3524
3525 /* 3: Enable RC6 */
3526 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
3527 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
abbf9d2c 3528 intel_print_rc6_info(dev, rc6_mask);
0d68b25e
TR		 3529 if (IS_BROADWELL(dev))
3530 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
3531 GEN7_RC_CTL_TO_MODE |
3532 rc6_mask);
3533 else
3534 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
3535 GEN6_RC_CTL_EI_MODE(1) |
3536 rc6_mask);
6edee7f3
BW		 3537
3538 /* 4 Program defaults and thresholds for RPS*/
f9bdc585
BW		 3539 I915_WRITE(GEN6_RPNSWREQ,
3540 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
3541 I915_WRITE(GEN6_RC_VIDEO_FREQ,
3542 HSW_FREQUENCY(dev_priv->rps.rp1_freq));
6edee7f3
BW		 3543 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
3544 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
3545
3546 /* Docs recommend 900MHz, and 300 MHz respectively */
3547 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
b39fb297
BW		 3548 dev_priv->rps.max_freq_softlimit << 24 |
3549 dev_priv->rps.min_freq_softlimit << 16);
6edee7f3
BW		 3550
3551 I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
3552 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
3553 I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
3554 I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
3555
3556 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3557
3558 /* 5: Enable RPS */
3559 I915_WRITE(GEN6_RP_CONTROL,
3560 GEN6_RP_MEDIA_TURBO |
3561 GEN6_RP_MEDIA_HW_NORMAL_MODE |
7405f42c 3562 GEN6_RP_MEDIA_IS_GFX |
6edee7f3
BW		 3563 GEN6_RP_ENABLE |
3564 GEN6_RP_UP_BUSY_AVG |
3565 GEN6_RP_DOWN_IDLE_AVG);
3566
3567 /* 6: Ring frequency + overclocking (our driver does this later */
3568
3569 gen6_set_rps(dev, (I915_READ(GEN6_GT_PERF_STATUS) & 0xff00) >> 8);
3570
0961021a 3571 gen8_enable_rps_interrupts(dev);
6edee7f3 3572
c8d9a590 3573 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW		 3574}
3575
79f5b2c7 3576static void gen6_enable_rps(struct drm_device *dev)
2b4e57bd 3577{
79f5b2c7 3578 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 3579 struct intel_engine_cs *ring;
2a5913a8 3580 u32 rp_state_cap;
7b9e0ae6 3581 u32 gt_perf_status;
d060c169 3582 u32 rc6vids, pcu_mbox = 0, rc6_mask = 0;
2b4e57bd 3583 u32 gtfifodbg;
2b4e57bd 3584 int rc6_mode;
42c0526c 3585 int i, ret;
2b4e57bd 3586
4fc688ce 3587 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 3588
2b4e57bd
ED		 3589 /* Here begins a magic sequence of register writes to enable
3590 * auto-downclocking.
3591 *
3592 * Perhaps there might be some value in exposing these to
3593 * userspace...
3594 */
3595 I915_WRITE(GEN6_RC_STATE, 0);
2b4e57bd
ED		 3596
3597 /* Clear the DBG now so we don't confuse earlier errors */
3598 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
3599 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
3600 I915_WRITE(GTFIFODBG, gtfifodbg);
3601 }
3602
c8d9a590 3603 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
2b4e57bd 3604
7b9e0ae6
CW		 3605 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3606 gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
3607
3280e8b0 3608 parse_rp_state_cap(dev_priv, rp_state_cap);
dd0a1aa1 3609
2b4e57bd
ED		 3610 /* disable the counters and set deterministic thresholds */
3611 I915_WRITE(GEN6_RC_CONTROL, 0);
3612
3613 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
3614 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
3615 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
3616 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3617 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3618
b4519513
CW		 3619 for_each_ring(ring, dev_priv, i)
3620 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2b4e57bd
ED		 3621
3622 I915_WRITE(GEN6_RC_SLEEP, 0);
3623 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
29c78f60 3624 if (IS_IVYBRIDGE(dev))
351aa566
SM		 3625 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
3626 else
3627 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
0920a487 3628 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2b4e57bd
ED		 3629 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
3630
5a7dc92a 3631 /* Check if we are enabling RC6 */
2b4e57bd
ED		 3632 rc6_mode = intel_enable_rc6(dev_priv->dev);
3633 if (rc6_mode & INTEL_RC6_ENABLE)
3634 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
3635
5a7dc92a
ED		 3636 /* We don't use those on Haswell */
3637 if (!IS_HASWELL(dev)) {
3638 if (rc6_mode & INTEL_RC6p_ENABLE)
3639 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2b4e57bd 3640
5a7dc92a
ED		 3641 if (rc6_mode & INTEL_RC6pp_ENABLE)
3642 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
3643 }
2b4e57bd 3644
dc39fff7 3645 intel_print_rc6_info(dev, rc6_mask);
2b4e57bd
ED		 3646
3647 I915_WRITE(GEN6_RC_CONTROL,
3648 rc6_mask |
3649 GEN6_RC_CTL_EI_MODE(1) |
3650 GEN6_RC_CTL_HW_ENABLE);
3651
dd75fdc8
CW		 3652 /* Power down if completely idle for over 50ms */
3653 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
2b4e57bd 3654 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2b4e57bd 3655
42c0526c 3656 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
d060c169 3657 if (ret)
42c0526c 3658 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
d060c169
BW		 3659
3660 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
3661 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
3662 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
b39fb297 3663 (dev_priv->rps.max_freq_softlimit & 0xff) * 50,
d060c169 3664 (pcu_mbox & 0xff) * 50);
b39fb297 3665 dev_priv->rps.max_freq = pcu_mbox & 0xff;
2b4e57bd
ED		 3666 }
3667
dd75fdc8 3668 dev_priv->rps.power = HIGH_POWER; /* force a reset */
b39fb297 3669 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
2b4e57bd 3670
44fc7d5c 3671 gen6_enable_rps_interrupts(dev);
2b4e57bd 3672
31643d54
BW		 3673 rc6vids = 0;
3674 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
3675 if (IS_GEN6(dev) && ret) {
3676 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
3677 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
3678 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
3679 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
3680 rc6vids &= 0xffff00;
3681 rc6vids |= GEN6_ENCODE_RC6_VID(450);
3682 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
3683 if (ret)
3684 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
3685 }
3686
c8d9a590 3687 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
2b4e57bd
ED		 3688}
3689
c2bc2fc5 3690static void __gen6_update_ring_freq(struct drm_device *dev)
2b4e57bd 3691{
79f5b2c7 3692 struct drm_i915_private *dev_priv = dev->dev_private;
2b4e57bd 3693 int min_freq = 15;
3ebecd07
CW		 3694 unsigned int gpu_freq;
3695 unsigned int max_ia_freq, min_ring_freq;
2b4e57bd 3696 int scaling_factor = 180;
eda79642 3697 struct cpufreq_policy *policy;
2b4e57bd 3698
4fc688ce 3699 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 3700
eda79642
BW		 3701 policy = cpufreq_cpu_get(0);
3702 if (policy) {
3703 max_ia_freq = policy->cpuinfo.max_freq;
3704 cpufreq_cpu_put(policy);
3705 } else {
3706 /*
3707 * Default to measured freq if none found, PCU will ensure we
3708 * don't go over
3709 */
2b4e57bd 3710 max_ia_freq = tsc_khz;
eda79642 3711 }
2b4e57bd
ED		 3712
3713 /* Convert from kHz to MHz */
3714 max_ia_freq /= 1000;
3715
153b4b95 3716 min_ring_freq = I915_READ(DCLK) & 0xf;
f6aca45c
BW		 3717 /* convert DDR frequency from units of 266.6MHz to bandwidth */
3718 min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3ebecd07 3719
2b4e57bd
ED		 3720 /*
3721 * For each potential GPU frequency, load a ring frequency we'd like
3722 * to use for memory access. We do this by specifying the IA frequency
3723 * the PCU should use as a reference to determine the ring frequency.
3724 */
b39fb297 3725 for (gpu_freq = dev_priv->rps.max_freq_softlimit; gpu_freq >= dev_priv->rps.min_freq_softlimit;
2b4e57bd 3726 gpu_freq--) {
b39fb297 3727 int diff = dev_priv->rps.max_freq_softlimit - gpu_freq;
3ebecd07
CW		 3728 unsigned int ia_freq = 0, ring_freq = 0;
3729
46c764d4
BW		 3730 if (INTEL_INFO(dev)->gen >= 8) {
3731 /* max(2 * GT, DDR). NB: GT is 50MHz units */
3732 ring_freq = max(min_ring_freq, gpu_freq);
3733 } else if (IS_HASWELL(dev)) {
f6aca45c 3734 ring_freq = mult_frac(gpu_freq, 5, 4);
3ebecd07
CW		 3735 ring_freq = max(min_ring_freq, ring_freq);
3736 /* leave ia_freq as the default, chosen by cpufreq */
3737 } else {
3738 /* On older processors, there is no separate ring
3739 * clock domain, so in order to boost the bandwidth
3740 * of the ring, we need to upclock the CPU (ia_freq).
3741 *
3742 * For GPU frequencies less than 750MHz,
3743 * just use the lowest ring freq.
3744 */
3745 if (gpu_freq < min_freq)
3746 ia_freq = 800;
3747 else
3748 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
3749 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
3750 }
2b4e57bd 3751
42c0526c
BW		 3752 sandybridge_pcode_write(dev_priv,
3753 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3ebecd07
CW		 3754 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
3755 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
3756 gpu_freq);
2b4e57bd 3757 }
2b4e57bd
ED		 3758}
3759
c2bc2fc5
ID		 3760void gen6_update_ring_freq(struct drm_device *dev)
3761{
3762 struct drm_i915_private *dev_priv = dev->dev_private;
3763
3764 if (INTEL_INFO(dev)->gen < 6 || IS_VALLEYVIEW(dev))
3765 return;
3766
3767 mutex_lock(&dev_priv->rps.hw_lock);
3768 __gen6_update_ring_freq(dev);
3769 mutex_unlock(&dev_priv->rps.hw_lock);
3770}
3771
2b6b3a09
D		 3772int cherryview_rps_max_freq(struct drm_i915_private *dev_priv)
3773{
3774 u32 val, rp0;
3775
3776 val = vlv_punit_read(dev_priv, PUNIT_GPU_STATUS_REG);
3777 rp0 = (val >> PUNIT_GPU_STATUS_MAX_FREQ_SHIFT) & PUNIT_GPU_STATUS_MAX_FREQ_MASK;
3778
3779 return rp0;
3780}
3781
3782static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv)
3783{
3784 u32 val, rpe;
3785
3786 val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG);
3787 rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK;
3788
3789 return rpe;
3790}
3791
3792int cherryview_rps_min_freq(struct drm_i915_private *dev_priv)
3793{
3794 u32 val, rpn;
3795
3796 val = vlv_punit_read(dev_priv, PUNIT_GPU_STATUS_REG);
3797 rpn = (val >> PUNIT_GPU_STATIS_GFX_MIN_FREQ_SHIFT) & PUNIT_GPU_STATUS_GFX_MIN_FREQ_MASK;
3798 return rpn;
3799}
3800
0a073b84
JB		 3801int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
3802{
3803 u32 val, rp0;
3804
64936258 3805 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
0a073b84
JB		 3806
3807 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
3808 /* Clamp to max */
3809 rp0 = min_t(u32, rp0, 0xea);
3810
3811 return rp0;
3812}
3813
3814static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
3815{
3816 u32 val, rpe;
3817
64936258 3818 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
0a073b84 3819 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
64936258 3820 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
0a073b84
JB		 3821 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
3822
3823 return rpe;
3824}
3825
3826int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
3827{
64936258 3828 return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
0a073b84
JB		 3829}
3830
ae48434c
ID		 3831/* Check that the pctx buffer wasn't move under us. */
3832static void valleyview_check_pctx(struct drm_i915_private *dev_priv)
3833{
3834 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
3835
3836 WARN_ON(pctx_addr != dev_priv->mm.stolen_base +
3837 dev_priv->vlv_pctx->stolen->start);
3838}
3839
38807746
D		 3840
3841/* Check that the pcbr address is not empty. */
3842static void cherryview_check_pctx(struct drm_i915_private *dev_priv)
3843{
3844 unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
3845
3846 WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0);
3847}
3848
3849static void cherryview_setup_pctx(struct drm_device *dev)
3850{
3851 struct drm_i915_private *dev_priv = dev->dev_private;
3852 unsigned long pctx_paddr, paddr;
3853 struct i915_gtt *gtt = &dev_priv->gtt;
3854 u32 pcbr;
3855 int pctx_size = 32*1024;
3856
3857 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3858
3859 pcbr = I915_READ(VLV_PCBR);
3860 if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) {
3861 paddr = (dev_priv->mm.stolen_base +
3862 (gtt->stolen_size - pctx_size));
3863
3864 pctx_paddr = (paddr & (~4095));
3865 I915_WRITE(VLV_PCBR, pctx_paddr);
3866 }
3867}
3868
c9cddffc
JB		 3869static void valleyview_setup_pctx(struct drm_device *dev)
3870{
3871 struct drm_i915_private *dev_priv = dev->dev_private;
3872 struct drm_i915_gem_object *pctx;
3873 unsigned long pctx_paddr;
3874 u32 pcbr;
3875 int pctx_size = 24*1024;
3876
17b0c1f7
ID		 3877 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3878
c9cddffc
JB		 3879 pcbr = I915_READ(VLV_PCBR);
3880 if (pcbr) {
3881 /* BIOS set it up already, grab the pre-alloc'd space */
3882 int pcbr_offset;
3883
3884 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
3885 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
3886 pcbr_offset,
190d6cd5 3887 I915_GTT_OFFSET_NONE,
c9cddffc
JB		 3888 pctx_size);
3889 goto out;
3890 }
3891
3892 /*
3893 * From the Gunit register HAS:
3894 * The Gfx driver is expected to program this register and ensure
3895 * proper allocation within Gfx stolen memory. For example, this
3896 * register should be programmed such than the PCBR range does not
3897 * overlap with other ranges, such as the frame buffer, protected
3898 * memory, or any other relevant ranges.
3899 */
3900 pctx = i915_gem_object_create_stolen(dev, pctx_size);
3901 if (!pctx) {
3902 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
3903 return;
3904 }
3905
3906 pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
3907 I915_WRITE(VLV_PCBR, pctx_paddr);
3908
3909out:
3910 dev_priv->vlv_pctx = pctx;
3911}
3912
ae48434c
ID		 3913static void valleyview_cleanup_pctx(struct drm_device *dev)
3914{
3915 struct drm_i915_private *dev_priv = dev->dev_private;
3916
3917 if (WARN_ON(!dev_priv->vlv_pctx))
3918 return;
3919
3920 drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
3921 dev_priv->vlv_pctx = NULL;
3922}
3923
4e80519e
ID		 3924static void valleyview_init_gt_powersave(struct drm_device *dev)
3925{
3926 struct drm_i915_private *dev_priv = dev->dev_private;
3927
3928 valleyview_setup_pctx(dev);
3929
3930 mutex_lock(&dev_priv->rps.hw_lock);
3931
3932 dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv);
3933 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
3934 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
3935 vlv_gpu_freq(dev_priv, dev_priv->rps.max_freq),
3936 dev_priv->rps.max_freq);
3937
3938 dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv);
3939 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
3940 vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
3941 dev_priv->rps.efficient_freq);
3942
3943 dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv);
3944 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
3945 vlv_gpu_freq(dev_priv, dev_priv->rps.min_freq),
3946 dev_priv->rps.min_freq);
3947
3948 /* Preserve min/max settings in case of re-init */
3949 if (dev_priv->rps.max_freq_softlimit == 0)
3950 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
3951
3952 if (dev_priv->rps.min_freq_softlimit == 0)
3953 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
3954
3955 mutex_unlock(&dev_priv->rps.hw_lock);
3956}
3957
38807746
D		 3958static void cherryview_init_gt_powersave(struct drm_device *dev)
3959{
2b6b3a09
D		 3960 struct drm_i915_private *dev_priv = dev->dev_private;
3961
38807746 3962 cherryview_setup_pctx(dev);
2b6b3a09
D		 3963
3964 mutex_lock(&dev_priv->rps.hw_lock);
3965
3966 dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv);
3967 dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
3968 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
3969 vlv_gpu_freq(dev_priv, dev_priv->rps.max_freq),
3970 dev_priv->rps.max_freq);
3971
3972 dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv);
3973 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
3974 vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
3975 dev_priv->rps.efficient_freq);
3976
3977 dev_priv->rps.min_freq = cherryview_rps_min_freq(dev_priv);
3978 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
3979 vlv_gpu_freq(dev_priv, dev_priv->rps.min_freq),
3980 dev_priv->rps.min_freq);
3981
3982 /* Preserve min/max settings in case of re-init */
3983 if (dev_priv->rps.max_freq_softlimit == 0)
3984 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
3985
3986 if (dev_priv->rps.min_freq_softlimit == 0)
3987 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
3988
3989 mutex_unlock(&dev_priv->rps.hw_lock);
38807746
D		 3990}
3991
4e80519e
ID		 3992static void valleyview_cleanup_gt_powersave(struct drm_device *dev)
3993{
3994 valleyview_cleanup_pctx(dev);
3995}
3996
38807746
D		 3997static void cherryview_enable_rps(struct drm_device *dev)
3998{
3999 struct drm_i915_private *dev_priv = dev->dev_private;
4000 struct intel_engine_cs *ring;
2b6b3a09 4001 u32 gtfifodbg, val, rc6_mode = 0, pcbr;
38807746
D		 4002 int i;
4003
4004 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4005
4006 gtfifodbg = I915_READ(GTFIFODBG);
4007 if (gtfifodbg) {
4008 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
4009 gtfifodbg);
4010 I915_WRITE(GTFIFODBG, gtfifodbg);
4011 }
4012
4013 cherryview_check_pctx(dev_priv);
4014
4015 /* 1a & 1b: Get forcewake during program sequence. Although the driver
4016 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
4017 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
4018
4019 /* 2a: Program RC6 thresholds.*/
4020 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
4021 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
4022 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
4023
4024 for_each_ring(ring, dev_priv, i)
4025 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
4026 I915_WRITE(GEN6_RC_SLEEP, 0);
4027
4028 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
4029
4030 /* allows RC6 residency counter to work */
4031 I915_WRITE(VLV_COUNTER_CONTROL,
4032 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
4033 VLV_MEDIA_RC6_COUNT_EN |
4034 VLV_RENDER_RC6_COUNT_EN));
4035
4036 /* For now we assume BIOS is allocating and populating the PCBR */
4037 pcbr = I915_READ(VLV_PCBR);
4038
4039 DRM_DEBUG_DRIVER("PCBR offset : 0x%x\n", pcbr);
4040
4041 /* 3: Enable RC6 */
4042 if ((intel_enable_rc6(dev) & INTEL_RC6_ENABLE) &&
4043 (pcbr >> VLV_PCBR_ADDR_SHIFT))
4044 rc6_mode = GEN6_RC_CTL_EI_MODE(1);
4045
4046 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
4047
2b6b3a09
D		 4048 /* 4 Program defaults and thresholds for RPS*/
4049 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
4050 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
4051 I915_WRITE(GEN6_RP_UP_EI, 66000);
4052 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
4053
4054 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
4055
7405f42c
TR		 4056 /* WaDisablePwrmtrEvent:chv (pre-production hw) */
4057 I915_WRITE(0xA80C, I915_READ(0xA80C) & 0x00ffffff);
4058 I915_WRITE(0xA810, I915_READ(0xA810) & 0xffffff00);
4059
2b6b3a09
D		 4060 /* 5: Enable RPS */
4061 I915_WRITE(GEN6_RP_CONTROL,
4062 GEN6_RP_MEDIA_HW_NORMAL_MODE |
7405f42c 4063 GEN6_RP_MEDIA_IS_GFX | /* WaSetMaskForGfxBusyness:chv (pre-production hw ?) */
2b6b3a09
D		 4064 GEN6_RP_ENABLE |
4065 GEN6_RP_UP_BUSY_AVG |
4066 GEN6_RP_DOWN_IDLE_AVG);
4067
4068 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
4069
4070 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
4071 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
4072
4073 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
4074 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
4075 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
4076 dev_priv->rps.cur_freq);
4077
4078 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
4079 vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
4080 dev_priv->rps.efficient_freq);
4081
4082 valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
4083
38807746
D		 4084 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
4085}
4086
0a073b84
JB		 4087static void valleyview_enable_rps(struct drm_device *dev)
4088{
4089 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 4090 struct intel_engine_cs *ring;
2a5913a8 4091 u32 gtfifodbg, val, rc6_mode = 0;
0a073b84
JB		 4092 int i;
4093
4094 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4095
ae48434c
ID		 4096 valleyview_check_pctx(dev_priv);
4097
0a073b84 4098 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
f7d85c1e
JB		 4099 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
4100 gtfifodbg);
0a073b84
JB		 4101 I915_WRITE(GTFIFODBG, gtfifodbg);
4102 }
4103
c8d9a590
D		 4104 /* If VLV, Forcewake all wells, else re-direct to regular path */
4105 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
0a073b84
JB		 4106
4107 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
4108 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
4109 I915_WRITE(GEN6_RP_UP_EI, 66000);
4110 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
4111
4112 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
4113
4114 I915_WRITE(GEN6_RP_CONTROL,
4115 GEN6_RP_MEDIA_TURBO |
4116 GEN6_RP_MEDIA_HW_NORMAL_MODE |
4117 GEN6_RP_MEDIA_IS_GFX |
4118 GEN6_RP_ENABLE |
4119 GEN6_RP_UP_BUSY_AVG |
4120 GEN6_RP_DOWN_IDLE_CONT);
4121
4122 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
4123 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
4124 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
4125
4126 for_each_ring(ring, dev_priv, i)
4127 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
4128
2f0aa304 4129 I915_WRITE(GEN6_RC6_THRESHOLD, 0x557);
0a073b84
JB		 4130
4131 /* allows RC6 residency counter to work */
49798eb2
JB		 4132 I915_WRITE(VLV_COUNTER_CONTROL,
4133 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
4134 VLV_MEDIA_RC6_COUNT_EN |
4135 VLV_RENDER_RC6_COUNT_EN));
a2b23fe0 4136 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
6b88f295 4137 rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
dc39fff7
BW		 4138
4139 intel_print_rc6_info(dev, rc6_mode);
4140
a2b23fe0 4141 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
0a073b84 4142
64936258 4143 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
0a073b84
JB		 4144
4145 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
4146 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
4147
b39fb297 4148 dev_priv->rps.cur_freq = (val >> 8) & 0xff;
73008b98 4149 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
b39fb297
BW		 4150 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
4151 dev_priv->rps.cur_freq);
0a073b84 4152
73008b98 4153 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
b39fb297
BW		 4154 vlv_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
4155 dev_priv->rps.efficient_freq);
0a073b84 4156
b39fb297 4157 valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
0a073b84 4158
44fc7d5c 4159 gen6_enable_rps_interrupts(dev);
0a073b84 4160
c8d9a590 4161 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
0a073b84
JB		 4162}
4163
930ebb46 4164void ironlake_teardown_rc6(struct drm_device *dev)
2b4e57bd
ED		 4165{
4166 struct drm_i915_private *dev_priv = dev->dev_private;
4167
3e373948 4168 if (dev_priv->ips.renderctx) {
d7f46fc4 4169 i915_gem_object_ggtt_unpin(dev_priv->ips.renderctx);
3e373948
DV		 4170 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
4171 dev_priv->ips.renderctx = NULL;
2b4e57bd
ED		 4172 }
4173
3e373948 4174 if (dev_priv->ips.pwrctx) {
d7f46fc4 4175 i915_gem_object_ggtt_unpin(dev_priv->ips.pwrctx);
3e373948
DV		 4176 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
4177 dev_priv->ips.pwrctx = NULL;
2b4e57bd
ED		 4178 }
4179}
4180
930ebb46 4181static void ironlake_disable_rc6(struct drm_device *dev)
2b4e57bd
ED		 4182{
4183 struct drm_i915_private *dev_priv = dev->dev_private;
4184
4185 if (I915_READ(PWRCTXA)) {
4186 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
4187 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
4188 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
4189 50);
4190
4191 I915_WRITE(PWRCTXA, 0);
4192 POSTING_READ(PWRCTXA);
4193
4194 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
4195 POSTING_READ(RSTDBYCTL);
4196 }
2b4e57bd
ED		 4197}
4198
4199static int ironlake_setup_rc6(struct drm_device *dev)
4200{
4201 struct drm_i915_private *dev_priv = dev->dev_private;
4202
3e373948
DV		 4203 if (dev_priv->ips.renderctx == NULL)
4204 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
4205 if (!dev_priv->ips.renderctx)
2b4e57bd
ED		 4206 return -ENOMEM;
4207
3e373948
DV		 4208 if (dev_priv->ips.pwrctx == NULL)
4209 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
4210 if (!dev_priv->ips.pwrctx) {
2b4e57bd
ED		 4211 ironlake_teardown_rc6(dev);
4212 return -ENOMEM;
4213 }
4214
4215 return 0;
4216}
4217
930ebb46 4218static void ironlake_enable_rc6(struct drm_device *dev)
2b4e57bd
ED		 4219{
4220 struct drm_i915_private *dev_priv = dev->dev_private;
a4872ba6 4221 struct intel_engine_cs *ring = &dev_priv->ring[RCS];
3e960501 4222 bool was_interruptible;
2b4e57bd
ED		 4223 int ret;
4224
4225 /* rc6 disabled by default due to repeated reports of hanging during
4226 * boot and resume.
4227 */
4228 if (!intel_enable_rc6(dev))
4229 return;
4230
79f5b2c7
DV		 4231 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
4232
2b4e57bd 4233 ret = ironlake_setup_rc6(dev);
79f5b2c7 4234 if (ret)
2b4e57bd 4235 return;
2b4e57bd 4236
3e960501
CW		 4237 was_interruptible = dev_priv->mm.interruptible;
4238 dev_priv->mm.interruptible = false;
4239
2b4e57bd
ED		 4240 /*
4241 * GPU can automatically power down the render unit if given a page
4242 * to save state.
4243 */
6d90c952 4244 ret = intel_ring_begin(ring, 6);
2b4e57bd
ED		 4245 if (ret) {
4246 ironlake_teardown_rc6(dev);
3e960501 4247 dev_priv->mm.interruptible = was_interruptible;
2b4e57bd
ED		 4248 return;
4249 }
4250
6d90c952
DV		 4251 intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
4252 intel_ring_emit(ring, MI_SET_CONTEXT);
f343c5f6 4253 intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) |
6d90c952
DV		 4254 MI_MM_SPACE_GTT |
4255 MI_SAVE_EXT_STATE_EN |
4256 MI_RESTORE_EXT_STATE_EN |
4257 MI_RESTORE_INHIBIT);
4258 intel_ring_emit(ring, MI_SUSPEND_FLUSH);
4259 intel_ring_emit(ring, MI_NOOP);
4260 intel_ring_emit(ring, MI_FLUSH);
4261 intel_ring_advance(ring);
2b4e57bd
ED		 4262
4263 /*
4264 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
4265 * does an implicit flush, combined with MI_FLUSH above, it should be
4266 * safe to assume that renderctx is valid
4267 */
3e960501
CW		 4268 ret = intel_ring_idle(ring);
4269 dev_priv->mm.interruptible = was_interruptible;
2b4e57bd 4270 if (ret) {
def27a58 4271 DRM_ERROR("failed to enable ironlake power savings\n");
2b4e57bd 4272 ironlake_teardown_rc6(dev);
2b4e57bd
ED		 4273 return;
4274 }
4275
f343c5f6 4276 I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN);
2b4e57bd 4277 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
dc39fff7 4278
91ca689a 4279 intel_print_rc6_info(dev, GEN6_RC_CTL_RC6_ENABLE);
2b4e57bd
ED		 4280}
4281
dde18883
ED		 4282static unsigned long intel_pxfreq(u32 vidfreq)
4283{
4284 unsigned long freq;
4285 int div = (vidfreq & 0x3f0000) >> 16;
4286 int post = (vidfreq & 0x3000) >> 12;
4287 int pre = (vidfreq & 0x7);
4288
4289 if (!pre)
4290 return 0;
4291
4292 freq = ((div * 133333) / ((1<<post) * pre));
4293
4294 return freq;
4295}
4296
eb48eb00
DV		 4297static const struct cparams {
4298 u16 i;
4299 u16 t;
4300 u16 m;
4301 u16 c;
4302} cparams[] = {
4303 { 1, 1333, 301, 28664 },
4304 { 1, 1066, 294, 24460 },
4305 { 1, 800, 294, 25192 },
4306 { 0, 1333, 276, 27605 },
4307 { 0, 1066, 276, 27605 },
4308 { 0, 800, 231, 23784 },
4309};
4310
f531dcb2 4311static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 4312{
4313 u64 total_count, diff, ret;
4314 u32 count1, count2, count3, m = 0, c = 0;
4315 unsigned long now = jiffies_to_msecs(jiffies), diff1;
4316 int i;
4317
02d71956
DV		 4318 assert_spin_locked(&mchdev_lock);
4319
20e4d407 4320 diff1 = now - dev_priv->ips.last_time1;
eb48eb00
DV		 4321
4322 /* Prevent division-by-zero if we are asking too fast.
4323 * Also, we don't get interesting results if we are polling
4324 * faster than once in 10ms, so just return the saved value
4325 * in such cases.
4326 */
4327 if (diff1 <= 10)
20e4d407 4328 return dev_priv->ips.chipset_power;
eb48eb00
DV		 4329
4330 count1 = I915_READ(DMIEC);
4331 count2 = I915_READ(DDREC);
4332 count3 = I915_READ(CSIEC);
4333
4334 total_count = count1 + count2 + count3;
4335
4336 /* FIXME: handle per-counter overflow */
20e4d407
DV		 4337 if (total_count < dev_priv->ips.last_count1) {
4338 diff = ~0UL - dev_priv->ips.last_count1;
eb48eb00
DV		 4339 diff += total_count;
4340 } else {
20e4d407 4341 diff = total_count - dev_priv->ips.last_count1;
eb48eb00
DV		 4342 }
4343
4344 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
20e4d407
DV		 4345 if (cparams[i].i == dev_priv->ips.c_m &&
4346 cparams[i].t == dev_priv->ips.r_t) {
eb48eb00
DV		 4347 m = cparams[i].m;
4348 c = cparams[i].c;
4349 break;
4350 }
4351 }
4352
4353 diff = div_u64(diff, diff1);
4354 ret = ((m * diff) + c);
4355 ret = div_u64(ret, 10);
4356
20e4d407
DV		 4357 dev_priv->ips.last_count1 = total_count;
4358 dev_priv->ips.last_time1 = now;
eb48eb00 4359
20e4d407 4360 dev_priv->ips.chipset_power = ret;
eb48eb00
DV		 4361
4362 return ret;
4363}
4364
f531dcb2
CW		 4365unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
4366{
3d13ef2e 4367 struct drm_device *dev = dev_priv->dev;
f531dcb2
CW		 4368 unsigned long val;
4369
3d13ef2e 4370 if (INTEL_INFO(dev)->gen != 5)
f531dcb2
CW		 4371 return 0;
4372
4373 spin_lock_irq(&mchdev_lock);
4374
4375 val = __i915_chipset_val(dev_priv);
4376
4377 spin_unlock_irq(&mchdev_lock);
4378
4379 return val;
4380}
4381
eb48eb00
DV		 4382unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
4383{
4384 unsigned long m, x, b;
4385 u32 tsfs;
4386
4387 tsfs = I915_READ(TSFS);
4388
4389 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
4390 x = I915_READ8(TR1);
4391
4392 b = tsfs & TSFS_INTR_MASK;
4393
4394 return ((m * x) / 127) - b;
4395}
4396
4397static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
4398{
3d13ef2e 4399 struct drm_device *dev = dev_priv->dev;
eb48eb00
DV		 4400 static const struct v_table {
4401 u16 vd; /* in .1 mil */
4402 u16 vm; /* in .1 mil */
4403 } v_table[] = {
4404 { 0, 0, },
4405 { 375, 0, },
4406 { 500, 0, },
4407 { 625, 0, },
4408 { 750, 0, },
4409 { 875, 0, },
4410 { 1000, 0, },
4411 { 1125, 0, },
4412 { 4125, 3000, },
4413 { 4125, 3000, },
4414 { 4125, 3000, },
4415 { 4125, 3000, },
4416 { 4125, 3000, },
4417 { 4125, 3000, },
4418 { 4125, 3000, },
4419 { 4125, 3000, },
4420 { 4125, 3000, },
4421 { 4125, 3000, },
4422 { 4125, 3000, },
4423 { 4125, 3000, },
4424 { 4125, 3000, },
4425 { 4125, 3000, },
4426 { 4125, 3000, },
4427 { 4125, 3000, },
4428 { 4125, 3000, },
4429 { 4125, 3000, },
4430 { 4125, 3000, },
4431 { 4125, 3000, },
4432 { 4125, 3000, },
4433 { 4125, 3000, },
4434 { 4125, 3000, },
4435 { 4125, 3000, },
4436 { 4250, 3125, },
4437 { 4375, 3250, },
4438 { 4500, 3375, },
4439 { 4625, 3500, },
4440 { 4750, 3625, },
4441 { 4875, 3750, },
4442 { 5000, 3875, },
4443 { 5125, 4000, },
4444 { 5250, 4125, },
4445 { 5375, 4250, },
4446 { 5500, 4375, },
4447 { 5625, 4500, },
4448 { 5750, 4625, },
4449 { 5875, 4750, },
4450 { 6000, 4875, },
4451 { 6125, 5000, },
4452 { 6250, 5125, },
4453 { 6375, 5250, },
4454 { 6500, 5375, },
4455 { 6625, 5500, },
4456 { 6750, 5625, },
4457 { 6875, 5750, },
4458 { 7000, 5875, },
4459 { 7125, 6000, },
4460 { 7250, 6125, },
4461 { 7375, 6250, },
4462 { 7500, 6375, },
4463 { 7625, 6500, },
4464 { 7750, 6625, },
4465 { 7875, 6750, },
4466 { 8000, 6875, },
4467 { 8125, 7000, },
4468 { 8250, 7125, },
4469 { 8375, 7250, },
4470 { 8500, 7375, },
4471 { 8625, 7500, },
4472 { 8750, 7625, },
4473 { 8875, 7750, },
4474 { 9000, 7875, },
4475 { 9125, 8000, },
4476 { 9250, 8125, },
4477 { 9375, 8250, },
4478 { 9500, 8375, },
4479 { 9625, 8500, },
4480 { 9750, 8625, },
4481 { 9875, 8750, },
4482 { 10000, 8875, },
4483 { 10125, 9000, },
4484 { 10250, 9125, },
4485 { 10375, 9250, },
4486 { 10500, 9375, },
4487 { 10625, 9500, },
4488 { 10750, 9625, },
4489 { 10875, 9750, },
4490 { 11000, 9875, },
4491 { 11125, 10000, },
4492 { 11250, 10125, },
4493 { 11375, 10250, },
4494 { 11500, 10375, },
4495 { 11625, 10500, },
4496 { 11750, 10625, },
4497 { 11875, 10750, },
4498 { 12000, 10875, },
4499 { 12125, 11000, },
4500 { 12250, 11125, },
4501 { 12375, 11250, },
4502 { 12500, 11375, },
4503 { 12625, 11500, },
4504 { 12750, 11625, },
4505 { 12875, 11750, },
4506 { 13000, 11875, },
4507 { 13125, 12000, },
4508 { 13250, 12125, },
4509 { 13375, 12250, },
4510 { 13500, 12375, },
4511 { 13625, 12500, },
4512 { 13750, 12625, },
4513 { 13875, 12750, },
4514 { 14000, 12875, },
4515 { 14125, 13000, },
4516 { 14250, 13125, },
4517 { 14375, 13250, },
4518 { 14500, 13375, },
4519 { 14625, 13500, },
4520 { 14750, 13625, },
4521 { 14875, 13750, },
4522 { 15000, 13875, },
4523 { 15125, 14000, },
4524 { 15250, 14125, },
4525 { 15375, 14250, },
4526 { 15500, 14375, },
4527 { 15625, 14500, },
4528 { 15750, 14625, },
4529 { 15875, 14750, },
4530 { 16000, 14875, },
4531 { 16125, 15000, },
4532 };
3d13ef2e 4533 if (INTEL_INFO(dev)->is_mobile)
eb48eb00
DV		 4534 return v_table[pxvid].vm;
4535 else
4536 return v_table[pxvid].vd;
4537}
4538
02d71956 4539static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 4540{
4541 struct timespec now, diff1;
4542 u64 diff;
4543 unsigned long diffms;
4544 u32 count;
4545
02d71956 4546 assert_spin_locked(&mchdev_lock);
eb48eb00
DV		 4547
4548 getrawmonotonic(&now);
20e4d407 4549 diff1 = timespec_sub(now, dev_priv->ips.last_time2);
eb48eb00
DV		 4550
4551 /* Don't divide by 0 */
4552 diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
4553 if (!diffms)
4554 return;
4555
4556 count = I915_READ(GFXEC);
4557
20e4d407
DV		 4558 if (count < dev_priv->ips.last_count2) {
4559 diff = ~0UL - dev_priv->ips.last_count2;
eb48eb00
DV		 4560 diff += count;
4561 } else {
20e4d407 4562 diff = count - dev_priv->ips.last_count2;
eb48eb00
DV		 4563 }
4564
20e4d407
DV		 4565 dev_priv->ips.last_count2 = count;
4566 dev_priv->ips.last_time2 = now;
eb48eb00
DV		 4567
4568 /* More magic constants... */
4569 diff = diff * 1181;
4570 diff = div_u64(diff, diffms * 10);
20e4d407 4571 dev_priv->ips.gfx_power = diff;
eb48eb00
DV		 4572}
4573
02d71956
DV		 4574void i915_update_gfx_val(struct drm_i915_private *dev_priv)
4575{
3d13ef2e
DL		 4576 struct drm_device *dev = dev_priv->dev;
4577
4578 if (INTEL_INFO(dev)->gen != 5)
02d71956
DV		 4579 return;
4580
9270388e 4581 spin_lock_irq(&mchdev_lock);
02d71956
DV		 4582
4583 __i915_update_gfx_val(dev_priv);
4584
9270388e 4585 spin_unlock_irq(&mchdev_lock);
02d71956
DV		 4586}
4587
f531dcb2 4588static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV		 4589{
4590 unsigned long t, corr, state1, corr2, state2;
4591 u32 pxvid, ext_v;
4592
02d71956
DV		 4593 assert_spin_locked(&mchdev_lock);
4594
b39fb297 4595 pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_freq * 4));
eb48eb00
DV		 4596 pxvid = (pxvid >> 24) & 0x7f;
4597 ext_v = pvid_to_extvid(dev_priv, pxvid);
4598
4599 state1 = ext_v;
4600
4601 t = i915_mch_val(dev_priv);
4602
4603 /* Revel in the empirically derived constants */
4604
4605 /* Correction factor in 1/100000 units */
4606 if (t > 80)
4607 corr = ((t * 2349) + 135940);
4608 else if (t >= 50)
4609 corr = ((t * 964) + 29317);
4610 else /* < 50 */
4611 corr = ((t * 301) + 1004);
4612
4613 corr = corr * ((150142 * state1) / 10000 - 78642);
4614 corr /= 100000;
20e4d407 4615 corr2 = (corr * dev_priv->ips.corr);
eb48eb00
DV		 4616
4617 state2 = (corr2 * state1) / 10000;
4618 state2 /= 100; /* convert to mW */
4619
02d71956 4620 __i915_update_gfx_val(dev_priv);
eb48eb00 4621
20e4d407 4622 return dev_priv->ips.gfx_power + state2;
eb48eb00
DV		 4623}
4624
f531dcb2
CW		 4625unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
4626{
3d13ef2e 4627 struct drm_device *dev = dev_priv->dev;
f531dcb2
CW		 4628 unsigned long val;
4629
3d13ef2e 4630 if (INTEL_INFO(dev)->gen != 5)
f531dcb2
CW		 4631 return 0;
4632
4633 spin_lock_irq(&mchdev_lock);
4634
4635 val = __i915_gfx_val(dev_priv);
4636
4637 spin_unlock_irq(&mchdev_lock);
4638
4639 return val;
4640}
4641
eb48eb00
DV		 4642/**
4643 * i915_read_mch_val - return value for IPS use
4644 *
4645 * Calculate and return a value for the IPS driver to use when deciding whether
4646 * we have thermal and power headroom to increase CPU or GPU power budget.
4647 */
4648unsigned long i915_read_mch_val(void)
4649{
4650 struct drm_i915_private *dev_priv;
4651 unsigned long chipset_val, graphics_val, ret = 0;
4652
9270388e 4653 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4654 if (!i915_mch_dev)
4655 goto out_unlock;
4656 dev_priv = i915_mch_dev;
4657
f531dcb2
CW		 4658 chipset_val = __i915_chipset_val(dev_priv);
4659 graphics_val = __i915_gfx_val(dev_priv);
eb48eb00
DV		 4660
4661 ret = chipset_val + graphics_val;
4662
4663out_unlock:
9270388e 4664 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4665
4666 return ret;
4667}
4668EXPORT_SYMBOL_GPL(i915_read_mch_val);
4669
4670/**
4671 * i915_gpu_raise - raise GPU frequency limit
4672 *
4673 * Raise the limit; IPS indicates we have thermal headroom.
4674 */
4675bool i915_gpu_raise(void)
4676{
4677 struct drm_i915_private *dev_priv;
4678 bool ret = true;
4679
9270388e 4680 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4681 if (!i915_mch_dev) {
4682 ret = false;
4683 goto out_unlock;
4684 }
4685 dev_priv = i915_mch_dev;
4686
20e4d407
DV		 4687 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
4688 dev_priv->ips.max_delay--;
eb48eb00
DV		 4689
4690out_unlock:
9270388e 4691 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4692
4693 return ret;
4694}
4695EXPORT_SYMBOL_GPL(i915_gpu_raise);
4696
4697/**
4698 * i915_gpu_lower - lower GPU frequency limit
4699 *
4700 * IPS indicates we're close to a thermal limit, so throttle back the GPU
4701 * frequency maximum.
4702 */
4703bool i915_gpu_lower(void)
4704{
4705 struct drm_i915_private *dev_priv;
4706 bool ret = true;
4707
9270388e 4708 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4709 if (!i915_mch_dev) {
4710 ret = false;
4711 goto out_unlock;
4712 }
4713 dev_priv = i915_mch_dev;
4714
20e4d407
DV		 4715 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
4716 dev_priv->ips.max_delay++;
eb48eb00
DV		 4717
4718out_unlock:
9270388e 4719 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4720
4721 return ret;
4722}
4723EXPORT_SYMBOL_GPL(i915_gpu_lower);
4724
4725/**
4726 * i915_gpu_busy - indicate GPU business to IPS
4727 *
4728 * Tell the IPS driver whether or not the GPU is busy.
4729 */
4730bool i915_gpu_busy(void)
4731{
4732 struct drm_i915_private *dev_priv;
a4872ba6 4733 struct intel_engine_cs *ring;
eb48eb00 4734 bool ret = false;
f047e395 4735 int i;
eb48eb00 4736
9270388e 4737 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4738 if (!i915_mch_dev)
4739 goto out_unlock;
4740 dev_priv = i915_mch_dev;
4741
f047e395
CW		 4742 for_each_ring(ring, dev_priv, i)
4743 ret |= !list_empty(&ring->request_list);
eb48eb00
DV		 4744
4745out_unlock:
9270388e 4746 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4747
4748 return ret;
4749}
4750EXPORT_SYMBOL_GPL(i915_gpu_busy);
4751
4752/**
4753 * i915_gpu_turbo_disable - disable graphics turbo
4754 *
4755 * Disable graphics turbo by resetting the max frequency and setting the
4756 * current frequency to the default.
4757 */
4758bool i915_gpu_turbo_disable(void)
4759{
4760 struct drm_i915_private *dev_priv;
4761 bool ret = true;
4762
9270388e 4763 spin_lock_irq(&mchdev_lock);
eb48eb00
DV		 4764 if (!i915_mch_dev) {
4765 ret = false;
4766 goto out_unlock;
4767 }
4768 dev_priv = i915_mch_dev;
4769
20e4d407 4770 dev_priv->ips.max_delay = dev_priv->ips.fstart;
eb48eb00 4771
20e4d407 4772 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
eb48eb00
DV		 4773 ret = false;
4774
4775out_unlock:
9270388e 4776 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4777
4778 return ret;
4779}
4780EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
4781
4782/**
4783 * Tells the intel_ips driver that the i915 driver is now loaded, if
4784 * IPS got loaded first.
4785 *
4786 * This awkward dance is so that neither module has to depend on the
4787 * other in order for IPS to do the appropriate communication of
4788 * GPU turbo limits to i915.
4789 */
4790static void
4791ips_ping_for_i915_load(void)
4792{
4793 void (*link)(void);
4794
4795 link = symbol_get(ips_link_to_i915_driver);
4796 if (link) {
4797 link();
4798 symbol_put(ips_link_to_i915_driver);
4799 }
4800}
4801
4802void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
4803{
02d71956
DV		 4804 /* We only register the i915 ips part with intel-ips once everything is
4805 * set up, to avoid intel-ips sneaking in and reading bogus values. */
9270388e 4806 spin_lock_irq(&mchdev_lock);
eb48eb00 4807 i915_mch_dev = dev_priv;
9270388e 4808 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV		 4809
4810 ips_ping_for_i915_load();
4811}
4812
4813void intel_gpu_ips_teardown(void)
4814{
9270388e 4815 spin_lock_irq(&mchdev_lock);
eb48eb00 4816 i915_mch_dev = NULL;
9270388e 4817 spin_unlock_irq(&mchdev_lock);
eb48eb00 4818}
76c3552f 4819
8090c6b9 4820static void intel_init_emon(struct drm_device *dev)
dde18883
ED		 4821{
4822 struct drm_i915_private *dev_priv = dev->dev_private;
4823 u32 lcfuse;
4824 u8 pxw[16];
4825 int i;
4826
4827 /* Disable to program */
4828 I915_WRITE(ECR, 0);
4829 POSTING_READ(ECR);
4830
4831 /* Program energy weights for various events */
4832 I915_WRITE(SDEW, 0x15040d00);
4833 I915_WRITE(CSIEW0, 0x007f0000);
4834 I915_WRITE(CSIEW1, 0x1e220004);
4835 I915_WRITE(CSIEW2, 0x04000004);
4836
4837 for (i = 0; i < 5; i++)
4838 I915_WRITE(PEW + (i * 4), 0);
4839 for (i = 0; i < 3; i++)
4840 I915_WRITE(DEW + (i * 4), 0);
4841
4842 /* Program P-state weights to account for frequency power adjustment */
4843 for (i = 0; i < 16; i++) {
4844 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
4845 unsigned long freq = intel_pxfreq(pxvidfreq);
4846 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
4847 PXVFREQ_PX_SHIFT;
4848 unsigned long val;
4849
4850 val = vid * vid;
4851 val *= (freq / 1000);
4852 val *= 255;
4853 val /= (127*127*900);
4854 if (val > 0xff)
4855 DRM_ERROR("bad pxval: %ld\n", val);
4856 pxw[i] = val;
4857 }
4858 /* Render standby states get 0 weight */
4859 pxw[14] = 0;
4860 pxw[15] = 0;
4861
4862 for (i = 0; i < 4; i++) {
4863 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
4864 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
4865 I915_WRITE(PXW + (i * 4), val);
4866 }
4867
4868 /* Adjust magic regs to magic values (more experimental results) */
4869 I915_WRITE(OGW0, 0);
4870 I915_WRITE(OGW1, 0);
4871 I915_WRITE(EG0, 0x00007f00);
4872 I915_WRITE(EG1, 0x0000000e);
4873 I915_WRITE(EG2, 0x000e0000);
4874 I915_WRITE(EG3, 0x68000300);
4875 I915_WRITE(EG4, 0x42000000);
4876 I915_WRITE(EG5, 0x00140031);
4877 I915_WRITE(EG6, 0);
4878 I915_WRITE(EG7, 0);
4879
4880 for (i = 0; i < 8; i++)
4881 I915_WRITE(PXWL + (i * 4), 0);
4882
4883 /* Enable PMON + select events */
4884 I915_WRITE(ECR, 0x80000019);
4885
4886 lcfuse = I915_READ(LCFUSE02);
4887
20e4d407 4888 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
dde18883
ED		 4889}
4890
ae48434c
ID		 4891void intel_init_gt_powersave(struct drm_device *dev)
4892{
e6069ca8
ID		 4893 i915.enable_rc6 = sanitize_rc6_option(dev, i915.enable_rc6);
4894
38807746
D		 4895 if (IS_CHERRYVIEW(dev))
4896 cherryview_init_gt_powersave(dev);
4897 else if (IS_VALLEYVIEW(dev))
4e80519e 4898 valleyview_init_gt_powersave(dev);
ae48434c
ID		 4899}
4900
4901void intel_cleanup_gt_powersave(struct drm_device *dev)
4902{
38807746
D		 4903 if (IS_CHERRYVIEW(dev))
4904 return;
4905 else if (IS_VALLEYVIEW(dev))
4e80519e 4906 valleyview_cleanup_gt_powersave(dev);
ae48434c
ID		 4907}
4908
156c7ca0
JB		 4909/**
4910 * intel_suspend_gt_powersave - suspend PM work and helper threads
4911 * @dev: drm device
4912 *
4913 * We don't want to disable RC6 or other features here, we just want
4914 * to make sure any work we've queued has finished and won't bother
4915 * us while we're suspended.
4916 */
4917void intel_suspend_gt_powersave(struct drm_device *dev)
4918{
4919 struct drm_i915_private *dev_priv = dev->dev_private;
4920
4921 /* Interrupts should be disabled already to avoid re-arming. */
e11aa362 4922 WARN_ON(dev->irq_enabled && !dev_priv->pm.irqs_disabled);
156c7ca0
JB		 4923
4924 flush_delayed_work(&dev_priv->rps.delayed_resume_work);
4925
4926 cancel_work_sync(&dev_priv->rps.work);
4927}
4928
8090c6b9
DV		 4929void intel_disable_gt_powersave(struct drm_device *dev)
4930{
1a01ab3b
JB		 4931 struct drm_i915_private *dev_priv = dev->dev_private;
4932
fd0c0642 4933 /* Interrupts should be disabled already to avoid re-arming. */
e11aa362 4934 WARN_ON(dev->irq_enabled && !dev_priv->pm.irqs_disabled);
fd0c0642 4935
930ebb46 4936 if (IS_IRONLAKE_M(dev)) {
8090c6b9 4937 ironlake_disable_drps(dev);
930ebb46 4938 ironlake_disable_rc6(dev);
38807746 4939 } else if (INTEL_INFO(dev)->gen >= 6) {
10d8d366 4940 intel_suspend_gt_powersave(dev);
e494837a 4941
4fc688ce 4942 mutex_lock(&dev_priv->rps.hw_lock);
38807746
D		 4943 if (IS_CHERRYVIEW(dev))
4944 cherryview_disable_rps(dev);
4945 else if (IS_VALLEYVIEW(dev))
d20d4f0c
JB		 4946 valleyview_disable_rps(dev);
4947 else
4948 gen6_disable_rps(dev);
c0951f0c 4949 dev_priv->rps.enabled = false;
4fc688ce 4950 mutex_unlock(&dev_priv->rps.hw_lock);
930ebb46 4951 }
8090c6b9
DV		 4952}
4953
1a01ab3b
JB		 4954static void intel_gen6_powersave_work(struct work_struct *work)
4955{
4956 struct drm_i915_private *dev_priv =
4957 container_of(work, struct drm_i915_private,
4958 rps.delayed_resume_work.work);
4959 struct drm_device *dev = dev_priv->dev;
4960
4fc688ce 4961 mutex_lock(&dev_priv->rps.hw_lock);
0a073b84 4962
38807746
D		 4963 if (IS_CHERRYVIEW(dev)) {
4964 cherryview_enable_rps(dev);
4965 } else if (IS_VALLEYVIEW(dev)) {
0a073b84 4966 valleyview_enable_rps(dev);
6edee7f3
BW		 4967 } else if (IS_BROADWELL(dev)) {
4968 gen8_enable_rps(dev);
c2bc2fc5 4969 __gen6_update_ring_freq(dev);
0a073b84
JB		 4970 } else {
4971 gen6_enable_rps(dev);
c2bc2fc5 4972 __gen6_update_ring_freq(dev);
0a073b84 4973 }
c0951f0c 4974 dev_priv->rps.enabled = true;
4fc688ce 4975 mutex_unlock(&dev_priv->rps.hw_lock);
c6df39b5
ID		 4976
4977 intel_runtime_pm_put(dev_priv);
1a01ab3b
JB		 4978}
4979
8090c6b9
DV		 4980void intel_enable_gt_powersave(struct drm_device *dev)
4981{
1a01ab3b
JB		 4982 struct drm_i915_private *dev_priv = dev->dev_private;
4983
8090c6b9 4984 if (IS_IRONLAKE_M(dev)) {
dc1d0136 4985 mutex_lock(&dev->struct_mutex);
8090c6b9
DV		 4986 ironlake_enable_drps(dev);
4987 ironlake_enable_rc6(dev);
4988 intel_init_emon(dev);
dc1d0136 4989 mutex_unlock(&dev->struct_mutex);
38807746 4990 } else if (INTEL_INFO(dev)->gen >= 6) {
1a01ab3b
JB		 4991 /*
4992 * PCU communication is slow and this doesn't need to be
4993 * done at any specific time, so do this out of our fast path
4994 * to make resume and init faster.
c6df39b5
ID		 4995 *
4996 * We depend on the HW RC6 power context save/restore
4997 * mechanism when entering D3 through runtime PM suspend. So
4998 * disable RPM until RPS/RC6 is properly setup. We can only
4999 * get here via the driver load/system resume/runtime resume
5000 * paths, so the _noresume version is enough (and in case of
5001 * runtime resume it's necessary).
1a01ab3b 5002 */
c6df39b5
ID		 5003 if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
5004 round_jiffies_up_relative(HZ)))
5005 intel_runtime_pm_get_noresume(dev_priv);
8090c6b9
DV		 5006 }
5007}
5008
c6df39b5
ID		 5009void intel_reset_gt_powersave(struct drm_device *dev)
5010{
5011 struct drm_i915_private *dev_priv = dev->dev_private;
5012
5013 dev_priv->rps.enabled = false;
5014 intel_enable_gt_powersave(dev);
5015}
5016
3107bd48
DV		 5017static void ibx_init_clock_gating(struct drm_device *dev)
5018{
5019 struct drm_i915_private *dev_priv = dev->dev_private;
5020
5021 /*
5022 * On Ibex Peak and Cougar Point, we need to disable clock
5023 * gating for the panel power sequencer or it will fail to
5024 * start up when no ports are active.
5025 */
5026 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
5027}
5028
0e088b8f
VS		 5029static void g4x_disable_trickle_feed(struct drm_device *dev)
5030{
5031 struct drm_i915_private *dev_priv = dev->dev_private;
5032 int pipe;
5033
5034 for_each_pipe(pipe) {
5035 I915_WRITE(DSPCNTR(pipe),
5036 I915_READ(DSPCNTR(pipe)) |
5037 DISPPLANE_TRICKLE_FEED_DISABLE);
1dba99f4 5038 intel_flush_primary_plane(dev_priv, pipe);
0e088b8f
VS		 5039 }
5040}
5041
017636cc
VS		 5042static void ilk_init_lp_watermarks(struct drm_device *dev)
5043{
5044 struct drm_i915_private *dev_priv = dev->dev_private;
5045
5046 I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
5047 I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
5048 I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
5049
5050 /*
5051 * Don't touch WM1S_LP_EN here.
5052 * Doing so could cause underruns.
5053 */
5054}
5055
1fa61106 5056static void ironlake_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5057{
5058 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 5059 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 5060
f1e8fa56
DL		 5061 /*
5062 * Required for FBC
5063 * WaFbcDisableDpfcClockGating:ilk
5064 */
4d47e4f5
DL		 5065 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
5066 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
5067 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED		 5068
5069 I915_WRITE(PCH_3DCGDIS0,
5070 MARIUNIT_CLOCK_GATE_DISABLE |
5071 SVSMUNIT_CLOCK_GATE_DISABLE);
5072 I915_WRITE(PCH_3DCGDIS1,
5073 VFMUNIT_CLOCK_GATE_DISABLE);
5074
6f1d69b0
ED		 5075 /*
5076 * According to the spec the following bits should be set in
5077 * order to enable memory self-refresh
5078 * The bit 22/21 of 0x42004
5079 * The bit 5 of 0x42020
5080 * The bit 15 of 0x45000
5081 */
5082 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5083 (I915_READ(ILK_DISPLAY_CHICKEN2) |
5084 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4d47e4f5 5085 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED		 5086 I915_WRITE(DISP_ARB_CTL,
5087 (I915_READ(DISP_ARB_CTL) |
5088 DISP_FBC_WM_DIS));
017636cc
VS		 5089
5090 ilk_init_lp_watermarks(dev);
6f1d69b0
ED		 5091
5092 /*
5093 * Based on the document from hardware guys the following bits
5094 * should be set unconditionally in order to enable FBC.
5095 * The bit 22 of 0x42000
5096 * The bit 22 of 0x42004
5097 * The bit 7,8,9 of 0x42020.
5098 */
5099 if (IS_IRONLAKE_M(dev)) {
4bb35334 5100 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6f1d69b0
ED		 5101 I915_WRITE(ILK_DISPLAY_CHICKEN1,
5102 I915_READ(ILK_DISPLAY_CHICKEN1) |
5103 ILK_FBCQ_DIS);
5104 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5105 I915_READ(ILK_DISPLAY_CHICKEN2) |
5106 ILK_DPARB_GATE);
6f1d69b0
ED		 5107 }
5108
4d47e4f5
DL		 5109 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
5110
6f1d69b0
ED		 5111 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5112 I915_READ(ILK_DISPLAY_CHICKEN2) |
5113 ILK_ELPIN_409_SELECT);
5114 I915_WRITE(_3D_CHICKEN2,
5115 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
5116 _3D_CHICKEN2_WM_READ_PIPELINED);
4358a374 5117
ecdb4eb7 5118 /* WaDisableRenderCachePipelinedFlush:ilk */
4358a374
DV		 5119 I915_WRITE(CACHE_MODE_0,
5120 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3107bd48 5121
4e04632e
AG		 5122 /* WaDisable_RenderCache_OperationalFlush:ilk */
5123 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
5124
0e088b8f 5125 g4x_disable_trickle_feed(dev);
bdad2b2f 5126
3107bd48
DV		 5127 ibx_init_clock_gating(dev);
5128}
5129
5130static void cpt_init_clock_gating(struct drm_device *dev)
5131{
5132 struct drm_i915_private *dev_priv = dev->dev_private;
5133 int pipe;
3f704fa2 5134 uint32_t val;
3107bd48
DV		 5135
5136 /*
5137 * On Ibex Peak and Cougar Point, we need to disable clock
5138 * gating for the panel power sequencer or it will fail to
5139 * start up when no ports are active.
5140 */
cd664078
JB		 5141 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
5142 PCH_DPLUNIT_CLOCK_GATE_DISABLE |
5143 PCH_CPUNIT_CLOCK_GATE_DISABLE);
3107bd48
DV		 5144 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
5145 DPLS_EDP_PPS_FIX_DIS);
335c07b7
TI		 5146 /* The below fixes the weird display corruption, a few pixels shifted
5147 * downward, on (only) LVDS of some HP laptops with IVY.
5148 */
3f704fa2 5149 for_each_pipe(pipe) {
dc4bd2d1
PZ		 5150 val = I915_READ(TRANS_CHICKEN2(pipe));
5151 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
5152 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
41aa3448 5153 if (dev_priv->vbt.fdi_rx_polarity_inverted)
3f704fa2 5154 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
dc4bd2d1
PZ		 5155 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
5156 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
5157 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3f704fa2
PZ		 5158 I915_WRITE(TRANS_CHICKEN2(pipe), val);
5159 }
3107bd48
DV		 5160 /* WADP0ClockGatingDisable */
5161 for_each_pipe(pipe) {
5162 I915_WRITE(TRANS_CHICKEN1(pipe),
5163 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
5164 }
6f1d69b0
ED		 5165}
5166
1d7aaa0c
DV		 5167static void gen6_check_mch_setup(struct drm_device *dev)
5168{
5169 struct drm_i915_private *dev_priv = dev->dev_private;
5170 uint32_t tmp;
5171
5172 tmp = I915_READ(MCH_SSKPD);
5173 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
5174 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
5175 DRM_INFO("This can cause pipe underruns and display issues.\n");
5176 DRM_INFO("Please upgrade your BIOS to fix this.\n");
5177 }
5178}
5179
1fa61106 5180static void gen6_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5181{
5182 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 5183 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 5184
231e54f6 5185 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6f1d69b0
ED		 5186
5187 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5188 I915_READ(ILK_DISPLAY_CHICKEN2) |
5189 ILK_ELPIN_409_SELECT);
5190
ecdb4eb7 5191 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4283908e
DV		 5192 I915_WRITE(_3D_CHICKEN,
5193 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
5194
ecdb4eb7 5195 /* WaSetupGtModeTdRowDispatch:snb */
6547fbdb
DV		 5196 if (IS_SNB_GT1(dev))
5197 I915_WRITE(GEN6_GT_MODE,
5198 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
5199
4e04632e
AG		 5200 /* WaDisable_RenderCache_OperationalFlush:snb */
5201 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
5202
8d85d272
VS		 5203 /*
5204 * BSpec recoomends 8x4 when MSAA is used,
5205 * however in practice 16x4 seems fastest.
c5c98a58
VS		 5206 *
5207 * Note that PS/WM thread counts depend on the WIZ hashing
5208 * disable bit, which we don't touch here, but it's good
5209 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
8d85d272
VS		 5210 */
5211 I915_WRITE(GEN6_GT_MODE,
5212 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
5213
017636cc 5214 ilk_init_lp_watermarks(dev);
6f1d69b0 5215
6f1d69b0 5216 I915_WRITE(CACHE_MODE_0,
50743298 5217 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
6f1d69b0
ED		 5218
5219 I915_WRITE(GEN6_UCGCTL1,
5220 I915_READ(GEN6_UCGCTL1) |
5221 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
5222 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
5223
5224 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
5225 * gating disable must be set. Failure to set it results in
5226 * flickering pixels due to Z write ordering failures after
5227 * some amount of runtime in the Mesa "fire" demo, and Unigine
5228 * Sanctuary and Tropics, and apparently anything else with
5229 * alpha test or pixel discard.
5230 *
5231 * According to the spec, bit 11 (RCCUNIT) must also be set,
5232 * but we didn't debug actual testcases to find it out.
0f846f81 5233 *
ef59318c
VS		 5234 * WaDisableRCCUnitClockGating:snb
5235 * WaDisableRCPBUnitClockGating:snb
6f1d69b0
ED		 5236 */
5237 I915_WRITE(GEN6_UCGCTL2,
5238 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
5239 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
5240
5eb146dd 5241 /* WaStripsFansDisableFastClipPerformanceFix:snb */
743b57d8
VS		 5242 I915_WRITE(_3D_CHICKEN3,
5243 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL));
6f1d69b0 5244
e927ecde
VS		 5245 /*
5246 * Bspec says:
5247 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
5248 * 3DSTATE_SF number of SF output attributes is more than 16."
5249 */
5250 I915_WRITE(_3D_CHICKEN3,
5251 _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH));
5252
6f1d69b0
ED		 5253 /*
5254 * According to the spec the following bits should be
5255 * set in order to enable memory self-refresh and fbc:
5256 * The bit21 and bit22 of 0x42000
5257 * The bit21 and bit22 of 0x42004
5258 * The bit5 and bit7 of 0x42020
5259 * The bit14 of 0x70180
5260 * The bit14 of 0x71180
4bb35334
DL		 5261 *
5262 * WaFbcAsynchFlipDisableFbcQueue:snb
6f1d69b0
ED		 5263 */
5264 I915_WRITE(ILK_DISPLAY_CHICKEN1,
5265 I915_READ(ILK_DISPLAY_CHICKEN1) |
5266 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
5267 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5268 I915_READ(ILK_DISPLAY_CHICKEN2) |
5269 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
231e54f6
DL		 5270 I915_WRITE(ILK_DSPCLK_GATE_D,
5271 I915_READ(ILK_DSPCLK_GATE_D) |
5272 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
5273 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
6f1d69b0 5274
0e088b8f 5275 g4x_disable_trickle_feed(dev);
f8f2ac9a 5276
3107bd48 5277 cpt_init_clock_gating(dev);
1d7aaa0c
DV		 5278
5279 gen6_check_mch_setup(dev);
6f1d69b0
ED		 5280}
5281
5282static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
5283{
5284 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
5285
3aad9059 5286 /*
46680e0a 5287 * WaVSThreadDispatchOverride:ivb,vlv
3aad9059
VS		 5288 *
5289 * This actually overrides the dispatch
5290 * mode for all thread types.
5291 */
6f1d69b0
ED		 5292 reg &= ~GEN7_FF_SCHED_MASK;
5293 reg |= GEN7_FF_TS_SCHED_HW;
5294 reg |= GEN7_FF_VS_SCHED_HW;
5295 reg |= GEN7_FF_DS_SCHED_HW;
5296
5297 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
5298}
5299
17a303ec
PZ		 5300static void lpt_init_clock_gating(struct drm_device *dev)
5301{
5302 struct drm_i915_private *dev_priv = dev->dev_private;
5303
5304 /*
5305 * TODO: this bit should only be enabled when really needed, then
5306 * disabled when not needed anymore in order to save power.
5307 */
5308 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
5309 I915_WRITE(SOUTH_DSPCLK_GATE_D,
5310 I915_READ(SOUTH_DSPCLK_GATE_D) |
5311 PCH_LP_PARTITION_LEVEL_DISABLE);
0a790cdb
PZ		 5312
5313 /* WADPOClockGatingDisable:hsw */
5314 I915_WRITE(_TRANSA_CHICKEN1,
5315 I915_READ(_TRANSA_CHICKEN1) |
5316 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
17a303ec
PZ		 5317}
5318
7d708ee4
ID		 5319static void lpt_suspend_hw(struct drm_device *dev)
5320{
5321 struct drm_i915_private *dev_priv = dev->dev_private;
5322
5323 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) {
5324 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
5325
5326 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
5327 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
5328 }
5329}
5330
1020a5c2
BW		 5331static void gen8_init_clock_gating(struct drm_device *dev)
5332{
5333 struct drm_i915_private *dev_priv = dev->dev_private;
07d27e20 5334 enum pipe pipe;
1020a5c2
BW		 5335
5336 I915_WRITE(WM3_LP_ILK, 0);
5337 I915_WRITE(WM2_LP_ILK, 0);
5338 I915_WRITE(WM1_LP_ILK, 0);
50ed5fbd
BW		 5339
5340 /* FIXME(BDW): Check all the w/a, some might only apply to
5341 * pre-production hw. */
5342
c8966e10
KG		 5343 /* WaDisablePartialInstShootdown:bdw */
5344 I915_WRITE(GEN8_ROW_CHICKEN,
5345 _MASKED_BIT_ENABLE(PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE));
5346
1411e6a5
KG		 5347 /* WaDisableThreadStallDopClockGating:bdw */
5348 /* FIXME: Unclear whether we really need this on production bdw. */
5349 I915_WRITE(GEN8_ROW_CHICKEN,
5350 _MASKED_BIT_ENABLE(STALL_DOP_GATING_DISABLE));
5351
4167e32c
DL		 5352 /*
5353 * This GEN8_CENTROID_PIXEL_OPT_DIS W/A is only needed for
5354 * pre-production hardware
5355 */
fd392b60
BW		 5356 I915_WRITE(HALF_SLICE_CHICKEN3,
5357 _MASKED_BIT_ENABLE(GEN8_CENTROID_PIXEL_OPT_DIS));
bf66347c
BW		 5358 I915_WRITE(HALF_SLICE_CHICKEN3,
5359 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS));
4afe8d33
BW		 5360 I915_WRITE(GAMTARBMODE, _MASKED_BIT_ENABLE(ARB_MODE_BWGTLB_DISABLE));
5361
7f88da0c
BW		 5362 I915_WRITE(_3D_CHICKEN3,
5363 _3D_CHICKEN_SDE_LIMIT_FIFO_POLY_DEPTH(2));
5364
a75f3628
BW		 5365 I915_WRITE(COMMON_SLICE_CHICKEN2,
5366 _MASKED_BIT_ENABLE(GEN8_CSC2_SBE_VUE_CACHE_CONSERVATIVE));
5367
4c2e7a5f
BW		 5368 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5369 _MASKED_BIT_ENABLE(GEN7_SINGLE_SUBSCAN_DISPATCH_ENABLE));
5370
242a4018
BW		 5371 /* WaDisableDopClockGating:bdw May not be needed for production */
5372 I915_WRITE(GEN7_ROW_CHICKEN2,
5373 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5374
ab57fff1 5375 /* WaSwitchSolVfFArbitrationPriority:bdw */
50ed5fbd 5376 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
fe4ab3ce 5377
ab57fff1 5378 /* WaPsrDPAMaskVBlankInSRD:bdw */
fe4ab3ce
BW		 5379 I915_WRITE(CHICKEN_PAR1_1,
5380 I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
5381
ab57fff1 5382 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
07d27e20
DL		 5383 for_each_pipe(pipe) {
5384 I915_WRITE(CHICKEN_PIPESL_1(pipe),
c7c65622 5385 I915_READ(CHICKEN_PIPESL_1(pipe)) |
8f670bb1 5386 BDW_DPRS_MASK_VBLANK_SRD);
fe4ab3ce 5387 }
63801f21
BW		 5388
5389 /* Use Force Non-Coherent whenever executing a 3D context. This is a
5390 * workaround for for a possible hang in the unlikely event a TLB
5391 * invalidation occurs during a PSD flush.
5392 */
5393 I915_WRITE(HDC_CHICKEN0,
5394 I915_READ(HDC_CHICKEN0) |
5395 _MASKED_BIT_ENABLE(HDC_FORCE_NON_COHERENT));
ab57fff1
BW		 5396
5397 /* WaVSRefCountFullforceMissDisable:bdw */
5398 /* WaDSRefCountFullforceMissDisable:bdw */
5399 I915_WRITE(GEN7_FF_THREAD_MODE,
5400 I915_READ(GEN7_FF_THREAD_MODE) &
5401 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
36075a4c
VS		 5402
5403 /*
5404 * BSpec recommends 8x4 when MSAA is used,
5405 * however in practice 16x4 seems fastest.
c5c98a58
VS		 5406 *
5407 * Note that PS/WM thread counts depend on the WIZ hashing
5408 * disable bit, which we don't touch here, but it's good
5409 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
36075a4c
VS		 5410 */
5411 I915_WRITE(GEN7_GT_MODE,
5412 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
295e8bb7
VS		 5413
5414 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
5415 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
4f1ca9e9
VS		 5416
5417 /* WaDisableSDEUnitClockGating:bdw */
5418 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
5419 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
5d708680
DL		 5420
5421 /* Wa4x4STCOptimizationDisable:bdw */
5422 I915_WRITE(CACHE_MODE_1,
5423 _MASKED_BIT_ENABLE(GEN8_4x4_STC_OPTIMIZATION_DISABLE));
1020a5c2
BW		 5424}
5425
cad2a2d7
ED		 5426static void haswell_init_clock_gating(struct drm_device *dev)
5427{
5428 struct drm_i915_private *dev_priv = dev->dev_private;
cad2a2d7 5429
017636cc 5430 ilk_init_lp_watermarks(dev);
cad2a2d7 5431
f3fc4884
FJ		 5432 /* L3 caching of data atomics doesn't work -- disable it. */
5433 I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
5434 I915_WRITE(HSW_ROW_CHICKEN3,
5435 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
5436
ecdb4eb7 5437 /* This is required by WaCatErrorRejectionIssue:hsw */
cad2a2d7
ED		 5438 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5439 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5440 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5441
e36ea7ff
VS		 5442 /* WaVSRefCountFullforceMissDisable:hsw */
5443 I915_WRITE(GEN7_FF_THREAD_MODE,
5444 I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);
cad2a2d7 5445
4e04632e
AG		 5446 /* WaDisable_RenderCache_OperationalFlush:hsw */
5447 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
5448
fe27c606
CW		 5449 /* enable HiZ Raw Stall Optimization */
5450 I915_WRITE(CACHE_MODE_0_GEN7,
5451 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
5452
ecdb4eb7 5453 /* WaDisable4x2SubspanOptimization:hsw */
cad2a2d7
ED		 5454 I915_WRITE(CACHE_MODE_1,
5455 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
1544d9d5 5456
a12c4967
VS		 5457 /*
5458 * BSpec recommends 8x4 when MSAA is used,
5459 * however in practice 16x4 seems fastest.
c5c98a58
VS		 5460 *
5461 * Note that PS/WM thread counts depend on the WIZ hashing
5462 * disable bit, which we don't touch here, but it's good
5463 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a12c4967
VS		 5464 */
5465 I915_WRITE(GEN7_GT_MODE,
5466 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
5467
ecdb4eb7 5468 /* WaSwitchSolVfFArbitrationPriority:hsw */
e3dff585
BW		 5469 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
5470
90a88643
PZ		 5471 /* WaRsPkgCStateDisplayPMReq:hsw */
5472 I915_WRITE(CHICKEN_PAR1_1,
5473 I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
1544d9d5 5474
17a303ec 5475 lpt_init_clock_gating(dev);
cad2a2d7
ED		 5476}
5477
1fa61106 5478static void ivybridge_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5479{
5480 struct drm_i915_private *dev_priv = dev->dev_private;
20848223 5481 uint32_t snpcr;
6f1d69b0 5482
017636cc 5483 ilk_init_lp_watermarks(dev);
6f1d69b0 5484
231e54f6 5485 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 5486
ecdb4eb7 5487 /* WaDisableEarlyCull:ivb */
87f8020e
JB		 5488 I915_WRITE(_3D_CHICKEN3,
5489 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5490
ecdb4eb7 5491 /* WaDisableBackToBackFlipFix:ivb */
6f1d69b0
ED		 5492 I915_WRITE(IVB_CHICKEN3,
5493 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5494 CHICKEN3_DGMG_DONE_FIX_DISABLE);
5495
ecdb4eb7 5496 /* WaDisablePSDDualDispatchEnable:ivb */
12f3382b
JB		 5497 if (IS_IVB_GT1(dev))
5498 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5499 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 5500
4e04632e
AG		 5501 /* WaDisable_RenderCache_OperationalFlush:ivb */
5502 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
5503
ecdb4eb7 5504 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6f1d69b0
ED		 5505 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5506 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5507
ecdb4eb7 5508 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6f1d69b0
ED		 5509 I915_WRITE(GEN7_L3CNTLREG1,
5510 GEN7_WA_FOR_GEN7_L3_CONTROL);
5511 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
8ab43976
JB		 5512 GEN7_WA_L3_CHICKEN_MODE);
5513 if (IS_IVB_GT1(dev))
5514 I915_WRITE(GEN7_ROW_CHICKEN2,
5515 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2
VS		 5516 else {
5517 /* must write both registers */
5518 I915_WRITE(GEN7_ROW_CHICKEN2,
5519 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
8ab43976
JB		 5520 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
5521 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2 5522 }
6f1d69b0 5523
ecdb4eb7 5524 /* WaForceL3Serialization:ivb */
61939d97
JB		 5525 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5526 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5527
1b80a19a 5528 /*
0f846f81 5529 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 5530 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
0f846f81
JB		 5531 */
5532 I915_WRITE(GEN6_UCGCTL2,
28acf3b2 5533 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 5534
ecdb4eb7 5535 /* This is required by WaCatErrorRejectionIssue:ivb */
6f1d69b0
ED		 5536 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5537 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5538 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5539
0e088b8f 5540 g4x_disable_trickle_feed(dev);
6f1d69b0
ED		 5541
5542 gen7_setup_fixed_func_scheduler(dev_priv);
97e1930f 5543
22721343
CW		 5544 if (0) { /* causes HiZ corruption on ivb:gt1 */
5545 /* enable HiZ Raw Stall Optimization */
5546 I915_WRITE(CACHE_MODE_0_GEN7,
5547 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
5548 }
116f2b6d 5549
ecdb4eb7 5550 /* WaDisable4x2SubspanOptimization:ivb */
97e1930f
DV		 5551 I915_WRITE(CACHE_MODE_1,
5552 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
20848223 5553
a607c1a4
VS		 5554 /*
5555 * BSpec recommends 8x4 when MSAA is used,
5556 * however in practice 16x4 seems fastest.
c5c98a58
VS		 5557 *
5558 * Note that PS/WM thread counts depend on the WIZ hashing
5559 * disable bit, which we don't touch here, but it's good
5560 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
a607c1a4
VS		 5561 */
5562 I915_WRITE(GEN7_GT_MODE,
5563 GEN6_WIZ_HASHING_MASK | GEN6_WIZ_HASHING_16x4);
5564
20848223
BW		 5565 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
5566 snpcr &= ~GEN6_MBC_SNPCR_MASK;
5567 snpcr |= GEN6_MBC_SNPCR_MED;
5568 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
3107bd48 5569
ab5c608b
BW		 5570 if (!HAS_PCH_NOP(dev))
5571 cpt_init_clock_gating(dev);
1d7aaa0c
DV		 5572
5573 gen6_check_mch_setup(dev);
6f1d69b0
ED		 5574}
5575
1fa61106 5576static void valleyview_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5577{
5578 struct drm_i915_private *dev_priv = dev->dev_private;
85b1d7b3
JB		 5579 u32 val;
5580
5581 mutex_lock(&dev_priv->rps.hw_lock);
5582 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5583 mutex_unlock(&dev_priv->rps.hw_lock);
5584 switch ((val >> 6) & 3) {
5585 case 0:
f64a28a7 5586 case 1:
f6d51948 5587 dev_priv->mem_freq = 800;
85b1d7b3 5588 break;
f64a28a7 5589 case 2:
f6d51948 5590 dev_priv->mem_freq = 1066;
85b1d7b3 5591 break;
f64a28a7 5592 case 3:
2325991e 5593 dev_priv->mem_freq = 1333;
f64a28a7 5594 break;
85b1d7b3
JB		 5595 }
5596 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
6f1d69b0 5597
d7fe0cc0 5598 I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 5599
ecdb4eb7 5600 /* WaDisableEarlyCull:vlv */
87f8020e
JB		 5601 I915_WRITE(_3D_CHICKEN3,
5602 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5603
ecdb4eb7 5604 /* WaDisableBackToBackFlipFix:vlv */
6f1d69b0
ED		 5605 I915_WRITE(IVB_CHICKEN3,
5606 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5607 CHICKEN3_DGMG_DONE_FIX_DISABLE);
5608
fad7d36e 5609 /* WaPsdDispatchEnable:vlv */
ecdb4eb7 5610 /* WaDisablePSDDualDispatchEnable:vlv */
12f3382b 5611 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
d3bc0303
JB		 5612 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
5613 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 5614
4e04632e
AG		 5615 /* WaDisable_RenderCache_OperationalFlush:vlv */
5616 I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
5617
ecdb4eb7 5618 /* WaForceL3Serialization:vlv */
61939d97
JB		 5619 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5620 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5621
ecdb4eb7 5622 /* WaDisableDopClockGating:vlv */
8ab43976
JB		 5623 I915_WRITE(GEN7_ROW_CHICKEN2,
5624 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5625
ecdb4eb7 5626 /* This is required by WaCatErrorRejectionIssue:vlv */
6f1d69b0
ED		 5627 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5628 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5629 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5630
46680e0a
VS		 5631 gen7_setup_fixed_func_scheduler(dev_priv);
5632
3c0edaeb 5633 /*
0f846f81 5634 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 5635 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
0f846f81
JB		 5636 */
5637 I915_WRITE(GEN6_UCGCTL2,
3c0edaeb 5638 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 5639
c98f5062
AG		 5640 /* WaDisableL3Bank2xClockGate:vlv
5641 * Disabling L3 clock gating- MMIO 940c[25] = 1
5642 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
5643 I915_WRITE(GEN7_UCGCTL4,
5644 I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
e3f33d46 5645
e0d8d59b 5646 I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
6f1d69b0 5647
afd58e79
VS		 5648 /*
5649 * BSpec says this must be set, even though
5650 * WaDisable4x2SubspanOptimization isn't listed for VLV.
5651 */
6b26c86d
DV		 5652 I915_WRITE(CACHE_MODE_1,
5653 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
7983117f 5654
031994ee
VS		 5655 /*
5656 * WaIncreaseL3CreditsForVLVB0:vlv
5657 * This is the hardware default actually.
5658 */
5659 I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);
5660
2d809570 5661 /*
ecdb4eb7 5662 * WaDisableVLVClockGating_VBIIssue:vlv
2d809570
JB		 5663 * Disable clock gating on th GCFG unit to prevent a delay
5664 * in the reporting of vblank events.
5665 */
7a0d1eed 5666 I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
6f1d69b0
ED		 5667}
5668
a4565da8
VS		 5669static void cherryview_init_clock_gating(struct drm_device *dev)
5670{
5671 struct drm_i915_private *dev_priv = dev->dev_private;
5672
5673 I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
5674
5675 I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
dd811e70
VS		 5676
5677 /* WaDisablePartialInstShootdown:chv */
5678 I915_WRITE(GEN8_ROW_CHICKEN,
5679 _MASKED_BIT_ENABLE(PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE));
a7068025
VS		 5680
5681 /* WaDisableThreadStallDopClockGating:chv */
5682 I915_WRITE(GEN8_ROW_CHICKEN,
5683 _MASKED_BIT_ENABLE(STALL_DOP_GATING_DISABLE));
232ce337
VS		 5684
5685 /* WaVSRefCountFullforceMissDisable:chv */
5686 /* WaDSRefCountFullforceMissDisable:chv */
5687 I915_WRITE(GEN7_FF_THREAD_MODE,
5688 I915_READ(GEN7_FF_THREAD_MODE) &
5689 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
acea6f95
VS		 5690
5691 /* WaDisableSemaphoreAndSyncFlipWait:chv */
5692 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
5693 _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
0846697c
VS		 5694
5695 /* WaDisableCSUnitClockGating:chv */
5696 I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
5697 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
c631780f
VS		 5698
5699 /* WaDisableSDEUnitClockGating:chv */
5700 I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
5701 GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
e0d34ce7
RB		 5702
5703 /* WaDisableSamplerPowerBypass:chv (pre-production hw) */
5704 I915_WRITE(HALF_SLICE_CHICKEN3,
5705 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS));
e4443e45
VS		 5706
5707 /* WaDisableGunitClockGating:chv (pre-production hw) */
5708 I915_WRITE(VLV_GUNIT_CLOCK_GATE, I915_READ(VLV_GUNIT_CLOCK_GATE) |
5709 GINT_DIS);
5710
5711 /* WaDisableFfDopClockGating:chv (pre-production hw) */
5712 I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
5713 _MASKED_BIT_ENABLE(GEN8_FF_DOP_CLOCK_GATE_DISABLE));
5714
5715 /* WaDisableDopClockGating:chv (pre-production hw) */
5716 I915_WRITE(GEN7_ROW_CHICKEN2,
5717 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5718 I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
5719 GEN6_EU_TCUNIT_CLOCK_GATE_DISABLE);
a4565da8
VS		 5720}
5721
1fa61106 5722static void g4x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5723{
5724 struct drm_i915_private *dev_priv = dev->dev_private;
5725 uint32_t dspclk_gate;
5726
5727 I915_WRITE(RENCLK_GATE_D1, 0);
5728 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
5729 GS_UNIT_CLOCK_GATE_DISABLE |
5730 CL_UNIT_CLOCK_GATE_DISABLE);
5731 I915_WRITE(RAMCLK_GATE_D, 0);
5732 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
5733 OVRUNIT_CLOCK_GATE_DISABLE |
5734 OVCUNIT_CLOCK_GATE_DISABLE;
5735 if (IS_GM45(dev))
5736 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
5737 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4358a374
DV		 5738
5739 /* WaDisableRenderCachePipelinedFlush */
5740 I915_WRITE(CACHE_MODE_0,
5741 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
de1aa629 5742
4e04632e
AG		 5743 /* WaDisable_RenderCache_OperationalFlush:g4x */
5744 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
5745
0e088b8f 5746 g4x_disable_trickle_feed(dev);
6f1d69b0
ED		 5747}
5748
1fa61106 5749static void crestline_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5750{
5751 struct drm_i915_private *dev_priv = dev->dev_private;
5752
5753 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
5754 I915_WRITE(RENCLK_GATE_D2, 0);
5755 I915_WRITE(DSPCLK_GATE_D, 0);
5756 I915_WRITE(RAMCLK_GATE_D, 0);
5757 I915_WRITE16(DEUC, 0);
20f94967
VS		 5758 I915_WRITE(MI_ARB_STATE,
5759 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
4e04632e
AG		 5760
5761 /* WaDisable_RenderCache_OperationalFlush:gen4 */
5762 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6f1d69b0
ED		 5763}
5764
1fa61106 5765static void broadwater_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5766{
5767 struct drm_i915_private *dev_priv = dev->dev_private;
5768
5769 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
5770 I965_RCC_CLOCK_GATE_DISABLE |
5771 I965_RCPB_CLOCK_GATE_DISABLE |
5772 I965_ISC_CLOCK_GATE_DISABLE |
5773 I965_FBC_CLOCK_GATE_DISABLE);
5774 I915_WRITE(RENCLK_GATE_D2, 0);
20f94967
VS		 5775 I915_WRITE(MI_ARB_STATE,
5776 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
4e04632e
AG		 5777
5778 /* WaDisable_RenderCache_OperationalFlush:gen4 */
5779 I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
6f1d69b0
ED		 5780}
5781
1fa61106 5782static void gen3_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5783{
5784 struct drm_i915_private *dev_priv = dev->dev_private;
5785 u32 dstate = I915_READ(D_STATE);
5786
5787 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
5788 DSTATE_DOT_CLOCK_GATING;
5789 I915_WRITE(D_STATE, dstate);
13a86b85
CW		 5790
5791 if (IS_PINEVIEW(dev))
5792 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
974a3b0f
DV		 5793
5794 /* IIR "flip pending" means done if this bit is set */
5795 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
12fabbcb
VS		 5796
5797 /* interrupts should cause a wake up from C3 */
3299254f 5798 I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN));
dbb42748
VS		 5799
5800 /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
5801 I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE));
6f1d69b0
ED		 5802}
5803
1fa61106 5804static void i85x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5805{
5806 struct drm_i915_private *dev_priv = dev->dev_private;
5807
5808 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
54e472ae
VS		 5809
5810 /* interrupts should cause a wake up from C3 */
5811 I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) |
5812 _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE));
6f1d69b0
ED		 5813}
5814
1fa61106 5815static void i830_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED		 5816{
5817 struct drm_i915_private *dev_priv = dev->dev_private;
5818
5819 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
5820}
5821
6f1d69b0
ED		 5822void intel_init_clock_gating(struct drm_device *dev)
5823{
5824 struct drm_i915_private *dev_priv = dev->dev_private;
5825
5826 dev_priv->display.init_clock_gating(dev);
6f1d69b0
ED		 5827}
5828
7d708ee4
ID		 5829void intel_suspend_hw(struct drm_device *dev)
5830{
5831 if (HAS_PCH_LPT(dev))
5832 lpt_suspend_hw(dev);
5833}
5834
c1ca727f
ID		 5835#define for_each_power_well(i, power_well, domain_mask, power_domains) \
5836 for (i = 0; \
5837 i < (power_domains)->power_well_count && \
5838 ((power_well) = &(power_domains)->power_wells[i]); \
5839 i++) \
5840 if ((power_well)->domains & (domain_mask))
5841
5842#define for_each_power_well_rev(i, power_well, domain_mask, power_domains) \
5843 for (i = (power_domains)->power_well_count - 1; \
5844 i >= 0 && ((power_well) = &(power_domains)->power_wells[i]);\
5845 i--) \
5846 if ((power_well)->domains & (domain_mask))
5847
15d199ea
PZ		 5848/**
5849 * We should only use the power well if we explicitly asked the hardware to
5850 * enable it, so check if it's enabled and also check if we've requested it to
5851 * be enabled.
5852 */
da7e29bd 5853static bool hsw_power_well_enabled(struct drm_i915_private *dev_priv,
c1ca727f
ID		 5854 struct i915_power_well *power_well)
5855{
c1ca727f
ID		 5856 return I915_READ(HSW_PWR_WELL_DRIVER) ==
5857 (HSW_PWR_WELL_ENABLE_REQUEST | HSW_PWR_WELL_STATE_ENABLED);
5858}
5859
bfafe93a
ID		 5860bool intel_display_power_enabled_unlocked(struct drm_i915_private *dev_priv,
5861 enum intel_display_power_domain domain)
ddf9c536 5862{
ddf9c536 5863 struct i915_power_domains *power_domains;
b8c000d9
ID		 5864 struct i915_power_well *power_well;
5865 bool is_enabled;
5866 int i;
5867
5868 if (dev_priv->pm.suspended)
5869 return false;
ddf9c536
ID		 5870
5871 power_domains = &dev_priv->power_domains;
bfafe93a 5872
b8c000d9 5873 is_enabled = true;
bfafe93a 5874
b8c000d9
ID		 5875 for_each_power_well_rev(i, power_well, BIT(domain), power_domains) {
5876 if (power_well->always_on)
5877 continue;
ddf9c536 5878
bfafe93a 5879 if (!power_well->hw_enabled) {
b8c000d9
ID		 5880 is_enabled = false;
5881 break;
5882 }
5883 }
bfafe93a 5884
b8c000d9 5885 return is_enabled;
ddf9c536
ID		 5886}
5887
da7e29bd 5888bool intel_display_power_enabled(struct drm_i915_private *dev_priv,
b97186f0 5889 enum intel_display_power_domain domain)
15d199ea 5890{
c1ca727f 5891 struct i915_power_domains *power_domains;
bfafe93a 5892 bool ret;
882244a3 5893
c1ca727f
ID		 5894 power_domains = &dev_priv->power_domains;
5895
c1ca727f 5896 mutex_lock(&power_domains->lock);
bfafe93a 5897 ret = intel_display_power_enabled_unlocked(dev_priv, domain);
c1ca727f
ID		 5898 mutex_unlock(&power_domains->lock);
5899
bfafe93a 5900 return ret;
15d199ea
PZ		 5901}
5902
93c73e8c
ID		 5903/*
5904 * Starting with Haswell, we have a "Power Down Well" that can be turned off
5905 * when not needed anymore. We have 4 registers that can request the power well
5906 * to be enabled, and it will only be disabled if none of the registers is
5907 * requesting it to be enabled.
5908 */
d5e8fdc8
PZ		 5909static void hsw_power_well_post_enable(struct drm_i915_private *dev_priv)
5910{
5911 struct drm_device *dev = dev_priv->dev;
5912 unsigned long irqflags;
5913
f9dcb0df
PZ		 5914 /*
5915 * After we re-enable the power well, if we touch VGA register 0x3d5
5916 * we'll get unclaimed register interrupts. This stops after we write
5917 * anything to the VGA MSR register. The vgacon module uses this
5918 * register all the time, so if we unbind our driver and, as a
5919 * consequence, bind vgacon, we'll get stuck in an infinite loop at
5920 * console_unlock(). So make here we touch the VGA MSR register, making
5921 * sure vgacon can keep working normally without triggering interrupts
5922 * and error messages.
5923 */
5924 vga_get_uninterruptible(dev->pdev, VGA_RSRC_LEGACY_IO);
5925 outb(inb(VGA_MSR_READ), VGA_MSR_WRITE);
5926 vga_put(dev->pdev, VGA_RSRC_LEGACY_IO);
5927
d5e8fdc8
PZ		 5928 if (IS_BROADWELL(dev)) {
5929 spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
5930 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_B),
5931 dev_priv->de_irq_mask[PIPE_B]);
5932 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_B),
5933 ~dev_priv->de_irq_mask[PIPE_B] |
5934 GEN8_PIPE_VBLANK);
5935 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_C),
5936 dev_priv->de_irq_mask[PIPE_C]);
5937 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_C),
5938 ~dev_priv->de_irq_mask[PIPE_C] |
5939 GEN8_PIPE_VBLANK);
5940 POSTING_READ(GEN8_DE_PIPE_IER(PIPE_C));
5941 spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
5942 }
5943}
5944
da7e29bd 5945static void hsw_set_power_well(struct drm_i915_private *dev_priv,
c1ca727f 5946 struct i915_power_well *power_well, bool enable)
d0d3e513 5947{
fa42e23c
PZ		 5948 bool is_enabled, enable_requested;
5949 uint32_t tmp;
d0d3e513 5950
fa42e23c 5951 tmp = I915_READ(HSW_PWR_WELL_DRIVER);
6aedd1f5
PZ		 5952 is_enabled = tmp & HSW_PWR_WELL_STATE_ENABLED;
5953 enable_requested = tmp & HSW_PWR_WELL_ENABLE_REQUEST;
d0d3e513 5954
fa42e23c
PZ		 5955 if (enable) {
5956 if (!enable_requested)
6aedd1f5
PZ		 5957 I915_WRITE(HSW_PWR_WELL_DRIVER,
5958 HSW_PWR_WELL_ENABLE_REQUEST);
d0d3e513 5959
fa42e23c
PZ		 5960 if (!is_enabled) {
5961 DRM_DEBUG_KMS("Enabling power well\n");
5962 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
6aedd1f5 5963 HSW_PWR_WELL_STATE_ENABLED), 20))
fa42e23c
PZ		 5964 DRM_ERROR("Timeout enabling power well\n");
5965 }
596cc11e 5966
d5e8fdc8 5967 hsw_power_well_post_enable(dev_priv);
fa42e23c
PZ		 5968 } else {
5969 if (enable_requested) {
5970 I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
9dbd8feb 5971 POSTING_READ(HSW_PWR_WELL_DRIVER);
fa42e23c 5972 DRM_DEBUG_KMS("Requesting to disable the power well\n");
d0d3e513
ED		 5973 }
5974 }
fa42e23c 5975}
d0d3e513 5976
c6cb582e
ID		 5977static void hsw_power_well_sync_hw(struct drm_i915_private *dev_priv,
5978 struct i915_power_well *power_well)
5979{
5980 hsw_set_power_well(dev_priv, power_well, power_well->count > 0);
5981
5982 /*
5983 * We're taking over the BIOS, so clear any requests made by it since
5984 * the driver is in charge now.
5985 */
5986 if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE_REQUEST)
5987 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
5988}
5989
5990static void hsw_power_well_enable(struct drm_i915_private *dev_priv,
5991 struct i915_power_well *power_well)
5992{
c6cb582e
ID		 5993 hsw_set_power_well(dev_priv, power_well, true);
5994}
5995
5996static void hsw_power_well_disable(struct drm_i915_private *dev_priv,
5997 struct i915_power_well *power_well)
5998{
5999 hsw_set_power_well(dev_priv, power_well, false);
c6cb582e
ID		 6000}
6001
a45f4466
ID		 6002static void i9xx_always_on_power_well_noop(struct drm_i915_private *dev_priv,
6003 struct i915_power_well *power_well)
6004{
6005}
6006
6007static bool i9xx_always_on_power_well_enabled(struct drm_i915_private *dev_priv,
6008 struct i915_power_well *power_well)
6009{
6010 return true;
6011}
6012
d2011dc8
VS		 6013static void vlv_set_power_well(struct drm_i915_private *dev_priv,
6014 struct i915_power_well *power_well, bool enable)
77961eb9 6015{
d2011dc8 6016 enum punit_power_well power_well_id = power_well->data;
77961eb9
ID		 6017 u32 mask;
6018 u32 state;
6019 u32 ctrl;
6020
6021 mask = PUNIT_PWRGT_MASK(power_well_id);
6022 state = enable ? PUNIT_PWRGT_PWR_ON(power_well_id) :
6023 PUNIT_PWRGT_PWR_GATE(power_well_id);
6024
6025 mutex_lock(&dev_priv->rps.hw_lock);
6026
6027#define COND \
6028 ((vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_STATUS) & mask) == state)
6029
6030 if (COND)
6031 goto out;
6032
6033 ctrl = vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_CTRL);
6034 ctrl &= ~mask;
6035 ctrl |= state;
6036 vlv_punit_write(dev_priv, PUNIT_REG_PWRGT_CTRL, ctrl);
6037
6038 if (wait_for(COND, 100))
6039 DRM_ERROR("timout setting power well state %08x (%08x)\n",
6040 state,
6041 vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_CTRL));
6042
6043#undef COND
6044
6045out:
6046 mutex_unlock(&dev_priv->rps.hw_lock);
6047}
6048
6049static void vlv_power_well_sync_hw(struct drm_i915_private *dev_priv,
6050 struct i915_power_well *power_well)
6051{
6052 vlv_set_power_well(dev_priv, power_well, power_well->count > 0);
6053}
6054
6055static void vlv_power_well_enable(struct drm_i915_private *dev_priv,
6056 struct i915_power_well *power_well)
6057{
6058 vlv_set_power_well(dev_priv, power_well, true);
6059}
6060
6061static void vlv_power_well_disable(struct drm_i915_private *dev_priv,
6062 struct i915_power_well *power_well)
6063{
6064 vlv_set_power_well(dev_priv, power_well, false);
6065}
6066
6067static bool vlv_power_well_enabled(struct drm_i915_private *dev_priv,
6068 struct i915_power_well *power_well)
6069{
6070 int power_well_id = power_well->data;
6071 bool enabled = false;
6072 u32 mask;
6073 u32 state;
6074 u32 ctrl;
6075
6076 mask = PUNIT_PWRGT_MASK(power_well_id);
6077 ctrl = PUNIT_PWRGT_PWR_ON(power_well_id);
6078
6079 mutex_lock(&dev_priv->rps.hw_lock);
6080
6081 state = vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_STATUS) & mask;
6082 /*
6083 * We only ever set the power-on and power-gate states, anything
6084 * else is unexpected.
6085 */
6086 WARN_ON(state != PUNIT_PWRGT_PWR_ON(power_well_id) &&
6087 state != PUNIT_PWRGT_PWR_GATE(power_well_id));
6088 if (state == ctrl)
6089 enabled = true;
6090
6091 /*
6092 * A transient state at this point would mean some unexpected party
6093 * is poking at the power controls too.
6094 */
6095 ctrl = vlv_punit_read(dev_priv, PUNIT_REG_PWRGT_CTRL) & mask;
6096 WARN_ON(ctrl != state);
6097
6098 mutex_unlock(&dev_priv->rps.hw_lock);
6099
6100 return enabled;
6101}
6102
6103static void vlv_display_power_well_enable(struct drm_i915_private *dev_priv,
6104 struct i915_power_well *power_well)
6105{
6106 WARN_ON_ONCE(power_well->data != PUNIT_POWER_WELL_DISP2D);
6107
6108 vlv_set_power_well(dev_priv, power_well, true);
6109
6110 spin_lock_irq(&dev_priv->irq_lock);
6111 valleyview_enable_display_irqs(dev_priv);
6112 spin_unlock_irq(&dev_priv->irq_lock);
6113
6114 /*
0d116a29
ID		 6115 * During driver initialization/resume we can avoid restoring the
6116 * part of the HW/SW state that will be inited anyway explicitly.
77961eb9 6117 */
0d116a29
ID		 6118 if (dev_priv->power_domains.initializing)
6119 return;
6120
6121 intel_hpd_init(dev_priv->dev);
77961eb9
ID		 6122
6123 i915_redisable_vga_power_on(dev_priv->dev);
6124}
6125
6126static void vlv_display_power_well_disable(struct drm_i915_private *dev_priv,
6127 struct i915_power_well *power_well)
6128{
77961eb9
ID		 6129 WARN_ON_ONCE(power_well->data != PUNIT_POWER_WELL_DISP2D);
6130
6131 spin_lock_irq(&dev_priv->irq_lock);
77961eb9
ID		 6132 valleyview_disable_display_irqs(dev_priv);
6133 spin_unlock_irq(&dev_priv->irq_lock);
6134
77961eb9
ID		 6135 vlv_set_power_well(dev_priv, power_well, false);
6136}
6137
aa519f23
VS		 6138static void vlv_dpio_cmn_power_well_enable(struct drm_i915_private *dev_priv,
6139 struct i915_power_well *power_well)
6140{
6141 WARN_ON_ONCE(power_well->data != PUNIT_POWER_WELL_DPIO_CMN_BC);
6142
6143 /*
6144 * Enable the CRI clock source so we can get at the
6145 * display and the reference clock for VGA
6146 * hotplug / manual detection.
6147 */
6148 I915_WRITE(DPLL(PIPE_B), I915_READ(DPLL(PIPE_B)) |
6149 DPLL_REFA_CLK_ENABLE_VLV | DPLL_INTEGRATED_CRI_CLK_VLV);
6150 udelay(1); /* >10ns for cmnreset, >0ns for sidereset */
6151
6152 vlv_set_power_well(dev_priv, power_well, true);
6153
6154 /*
6155 * From VLV2A0_DP_eDP_DPIO_driver_vbios_notes_10.docx -
6156 * 6. De-assert cmn_reset/side_reset. Same as VLV X0.
6157 * a. GUnit 0x2110 bit[0] set to 1 (def 0)
6158 * b. The other bits such as sfr settings / modesel may all
6159 * be set to 0.
6160 *
6161 * This should only be done on init and resume from S3 with
6162 * both PLLs disabled, or we risk losing DPIO and PLL
6163 * synchronization.
6164 */
6165 I915_WRITE(DPIO_CTL, I915_READ(DPIO_CTL) | DPIO_CMNRST);
6166}
6167
6168static void vlv_dpio_cmn_power_well_disable(struct drm_i915_private *dev_priv,
6169 struct i915_power_well *power_well)
6170{
6171 struct drm_device *dev = dev_priv->dev;
6172 enum pipe pipe;
6173
6174 WARN_ON_ONCE(power_well->data != PUNIT_POWER_WELL_DPIO_CMN_BC);
6175
6176 for_each_pipe(pipe)
6177 assert_pll_disabled(dev_priv, pipe);
6178
6179 /* Assert common reset */
6180 I915_WRITE(DPIO_CTL, I915_READ(DPIO_CTL) & ~DPIO_CMNRST);
6181
6182 vlv_set_power_well(dev_priv, power_well, false);
6183}
6184
25eaa003
ID		 6185static void check_power_well_state(struct drm_i915_private *dev_priv,
6186 struct i915_power_well *power_well)
6187{
6188 bool enabled = power_well->ops->is_enabled(dev_priv, power_well);
6189
6190 if (power_well->always_on || !i915.disable_power_well) {
6191 if (!enabled)
6192 goto mismatch;
6193
6194 return;
6195 }
6196
6197 if (enabled != (power_well->count > 0))
6198 goto mismatch;
6199
6200 return;
6201
6202mismatch:
6203 WARN(1, "state mismatch for '%s' (always_on %d hw state %d use-count %d disable_power_well %d\n",
6204 power_well->name, power_well->always_on, enabled,
6205 power_well->count, i915.disable_power_well);
6206}
6207
da7e29bd 6208void intel_display_power_get(struct drm_i915_private *dev_priv,
6765625e
VS		 6209 enum intel_display_power_domain domain)
6210{
83c00f55 6211 struct i915_power_domains *power_domains;
c1ca727f
ID		 6212 struct i915_power_well *power_well;
6213 int i;
6765625e 6214
9e6ea71a
PZ		 6215 intel_runtime_pm_get(dev_priv);
6216
83c00f55
ID		 6217 power_domains = &dev_priv->power_domains;
6218
6219 mutex_lock(&power_domains->lock);
1da51581 6220
25eaa003
ID		 6221 for_each_power_well(i, power_well, BIT(domain), power_domains) {
6222 if (!power_well->count++) {
6223 DRM_DEBUG_KMS("enabling %s\n", power_well->name);
c6cb582e 6224 power_well->ops->enable(dev_priv, power_well);
bfafe93a 6225 power_well->hw_enabled = true;
25eaa003
ID		 6226 }
6227
6228 check_power_well_state(dev_priv, power_well);
6229 }
1da51581 6230
ddf9c536
ID		 6231 power_domains->domain_use_count[domain]++;
6232
83c00f55 6233 mutex_unlock(&power_domains->lock);
6765625e
VS		 6234}
6235
da7e29bd 6236void intel_display_power_put(struct drm_i915_private *dev_priv,
6765625e
VS		 6237 enum intel_display_power_domain domain)
6238{
83c00f55 6239 struct i915_power_domains *power_domains;
c1ca727f
ID		 6240 struct i915_power_well *power_well;
6241 int i;
6765625e 6242
83c00f55
ID		 6243 power_domains = &dev_priv->power_domains;
6244
6245 mutex_lock(&power_domains->lock);
1da51581 6246
1da51581
ID		 6247 WARN_ON(!power_domains->domain_use_count[domain]);
6248 power_domains->domain_use_count[domain]--;
ddf9c536 6249
70bf407c
ID		 6250 for_each_power_well_rev(i, power_well, BIT(domain), power_domains) {
6251 WARN_ON(!power_well->count);
6252
25eaa003
ID		 6253 if (!--power_well->count && i915.disable_power_well) {
6254 DRM_DEBUG_KMS("disabling %s\n", power_well->name);
bfafe93a 6255 power_well->hw_enabled = false;
c6cb582e 6256 power_well->ops->disable(dev_priv, power_well);
25eaa003
ID		 6257 }
6258
6259 check_power_well_state(dev_priv, power_well);
70bf407c 6260 }
1da51581 6261
83c00f55 6262 mutex_unlock(&power_domains->lock);
9e6ea71a
PZ		 6263
6264 intel_runtime_pm_put(dev_priv);
6765625e
VS		 6265}
6266
83c00f55 6267static struct i915_power_domains *hsw_pwr;
a38911a3
WX		 6268
6269/* Display audio driver power well request */
74b0c2d7 6270int i915_request_power_well(void)
a38911a3 6271{
b4ed4484
ID		 6272 struct drm_i915_private *dev_priv;
6273
74b0c2d7
TI		 6274 if (!hsw_pwr)
6275 return -ENODEV;
a38911a3 6276
b4ed4484
ID		 6277 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
6278 power_domains);
da7e29bd 6279 intel_display_power_get(dev_priv, POWER_DOMAIN_AUDIO);
74b0c2d7 6280 return 0;
a38911a3
WX		 6281}
6282EXPORT_SYMBOL_GPL(i915_request_power_well);
6283
6284/* Display audio driver power well release */
74b0c2d7 6285int i915_release_power_well(void)
a38911a3 6286{
b4ed4484
ID		 6287 struct drm_i915_private *dev_priv;
6288
74b0c2d7
TI		 6289 if (!hsw_pwr)
6290 return -ENODEV;
a38911a3 6291
b4ed4484
ID		 6292 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
6293 power_domains);
da7e29bd 6294 intel_display_power_put(dev_priv, POWER_DOMAIN_AUDIO);
74b0c2d7 6295 return 0;
a38911a3
WX		 6296}
6297EXPORT_SYMBOL_GPL(i915_release_power_well);
6298
c149dcb5
JN		 6299/*
6300 * Private interface for the audio driver to get CDCLK in kHz.
6301 *
6302 * Caller must request power well using i915_request_power_well() prior to
6303 * making the call.
6304 */
6305int i915_get_cdclk_freq(void)
6306{
6307 struct drm_i915_private *dev_priv;
6308
6309 if (!hsw_pwr)
6310 return -ENODEV;
6311
6312 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
6313 power_domains);
6314
6315 return intel_ddi_get_cdclk_freq(dev_priv);
6316}
6317EXPORT_SYMBOL_GPL(i915_get_cdclk_freq);
6318
6319
efcad917
ID		 6320#define POWER_DOMAIN_MASK (BIT(POWER_DOMAIN_NUM) - 1)
6321
6322#define HSW_ALWAYS_ON_POWER_DOMAINS ( \
6323 BIT(POWER_DOMAIN_PIPE_A) | \
f5938f36 6324 BIT(POWER_DOMAIN_TRANSCODER_EDP) | \
319be8ae
ID		 6325 BIT(POWER_DOMAIN_PORT_DDI_A_2_LANES) | \
6326 BIT(POWER_DOMAIN_PORT_DDI_A_4_LANES) | \
6327 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
6328 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6329 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
6330 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6331 BIT(POWER_DOMAIN_PORT_DDI_D_2_LANES) | \
6332 BIT(POWER_DOMAIN_PORT_DDI_D_4_LANES) | \
6333 BIT(POWER_DOMAIN_PORT_CRT) | \
f5938f36 6334 BIT(POWER_DOMAIN_INIT))
efcad917
ID		 6335#define HSW_DISPLAY_POWER_DOMAINS ( \
6336 (POWER_DOMAIN_MASK & ~HSW_ALWAYS_ON_POWER_DOMAINS) | \
6337 BIT(POWER_DOMAIN_INIT))
6338
6339#define BDW_ALWAYS_ON_POWER_DOMAINS ( \
6340 HSW_ALWAYS_ON_POWER_DOMAINS | \
6341 BIT(POWER_DOMAIN_PIPE_A_PANEL_FITTER))
6342#define BDW_DISPLAY_POWER_DOMAINS ( \
6343 (POWER_DOMAIN_MASK & ~BDW_ALWAYS_ON_POWER_DOMAINS) | \
6344 BIT(POWER_DOMAIN_INIT))
6345
77961eb9
ID		 6346#define VLV_ALWAYS_ON_POWER_DOMAINS BIT(POWER_DOMAIN_INIT)
6347#define VLV_DISPLAY_POWER_DOMAINS POWER_DOMAIN_MASK
6348
6349#define VLV_DPIO_CMN_BC_POWER_DOMAINS ( \
6350 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
6351 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6352 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
6353 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6354 BIT(POWER_DOMAIN_PORT_CRT) | \
6355 BIT(POWER_DOMAIN_INIT))
6356
6357#define VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS ( \
6358 BIT(POWER_DOMAIN_PORT_DDI_B_2_LANES) | \
6359 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6360 BIT(POWER_DOMAIN_INIT))
6361
6362#define VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS ( \
6363 BIT(POWER_DOMAIN_PORT_DDI_B_4_LANES) | \
6364 BIT(POWER_DOMAIN_INIT))
6365
6366#define VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS ( \
6367 BIT(POWER_DOMAIN_PORT_DDI_C_2_LANES) | \
6368 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6369 BIT(POWER_DOMAIN_INIT))
6370
6371#define VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS ( \
6372 BIT(POWER_DOMAIN_PORT_DDI_C_4_LANES) | \
6373 BIT(POWER_DOMAIN_INIT))
6374
a45f4466
ID		 6375static const struct i915_power_well_ops i9xx_always_on_power_well_ops = {
6376 .sync_hw = i9xx_always_on_power_well_noop,
6377 .enable = i9xx_always_on_power_well_noop,
6378 .disable = i9xx_always_on_power_well_noop,
6379 .is_enabled = i9xx_always_on_power_well_enabled,
6380};
c6cb582e 6381
1c2256df
ID		 6382static struct i915_power_well i9xx_always_on_power_well[] = {
6383 {
6384 .name = "always-on",
6385 .always_on = 1,
6386 .domains = POWER_DOMAIN_MASK,
c6cb582e 6387 .ops = &i9xx_always_on_power_well_ops,
1c2256df
ID		 6388 },
6389};
6390
c6cb582e
ID		 6391static const struct i915_power_well_ops hsw_power_well_ops = {
6392 .sync_hw = hsw_power_well_sync_hw,
6393 .enable = hsw_power_well_enable,
6394 .disable = hsw_power_well_disable,
6395 .is_enabled = hsw_power_well_enabled,
6396};
6397
c1ca727f 6398static struct i915_power_well hsw_power_wells[] = {
6f3ef5dd
ID		 6399 {
6400 .name = "always-on",
6401 .always_on = 1,
6402 .domains = HSW_ALWAYS_ON_POWER_DOMAINS,
c6cb582e 6403 .ops = &i9xx_always_on_power_well_ops,
6f3ef5dd 6404 },
c1ca727f
ID		 6405 {
6406 .name = "display",
efcad917 6407 .domains = HSW_DISPLAY_POWER_DOMAINS,
c6cb582e 6408 .ops = &hsw_power_well_ops,
c1ca727f
ID		 6409 },
6410};
6411
6412static struct i915_power_well bdw_power_wells[] = {
6f3ef5dd
ID		 6413 {
6414 .name = "always-on",
6415 .always_on = 1,
6416 .domains = BDW_ALWAYS_ON_POWER_DOMAINS,
c6cb582e 6417 .ops = &i9xx_always_on_power_well_ops,
6f3ef5dd 6418 },
c1ca727f
ID		 6419 {
6420 .name = "display",
efcad917 6421 .domains = BDW_DISPLAY_POWER_DOMAINS,
c6cb582e 6422 .ops = &hsw_power_well_ops,
c1ca727f
ID		 6423 },
6424};
6425
77961eb9
ID		 6426static const struct i915_power_well_ops vlv_display_power_well_ops = {
6427 .sync_hw = vlv_power_well_sync_hw,
6428 .enable = vlv_display_power_well_enable,
6429 .disable = vlv_display_power_well_disable,
6430 .is_enabled = vlv_power_well_enabled,
6431};
6432
aa519f23
VS		 6433static const struct i915_power_well_ops vlv_dpio_cmn_power_well_ops = {
6434 .sync_hw = vlv_power_well_sync_hw,
6435 .enable = vlv_dpio_cmn_power_well_enable,
6436 .disable = vlv_dpio_cmn_power_well_disable,
6437 .is_enabled = vlv_power_well_enabled,
6438};
6439
77961eb9
ID		 6440static const struct i915_power_well_ops vlv_dpio_power_well_ops = {
6441 .sync_hw = vlv_power_well_sync_hw,
6442 .enable = vlv_power_well_enable,
6443 .disable = vlv_power_well_disable,
6444 .is_enabled = vlv_power_well_enabled,
6445};
6446
6447static struct i915_power_well vlv_power_wells[] = {
6448 {
6449 .name = "always-on",
6450 .always_on = 1,
6451 .domains = VLV_ALWAYS_ON_POWER_DOMAINS,
6452 .ops = &i9xx_always_on_power_well_ops,
6453 },
6454 {
6455 .name = "display",
6456 .domains = VLV_DISPLAY_POWER_DOMAINS,
6457 .data = PUNIT_POWER_WELL_DISP2D,
6458 .ops = &vlv_display_power_well_ops,
6459 },
77961eb9
ID		 6460 {
6461 .name = "dpio-tx-b-01",
6462 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
6463 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
6464 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
6465 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
6466 .ops = &vlv_dpio_power_well_ops,
6467 .data = PUNIT_POWER_WELL_DPIO_TX_B_LANES_01,
6468 },
6469 {
6470 .name = "dpio-tx-b-23",
6471 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
6472 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
6473 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
6474 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
6475 .ops = &vlv_dpio_power_well_ops,
6476 .data = PUNIT_POWER_WELL_DPIO_TX_B_LANES_23,
6477 },
6478 {
6479 .name = "dpio-tx-c-01",
6480 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
6481 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
6482 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
6483 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
6484 .ops = &vlv_dpio_power_well_ops,
6485 .data = PUNIT_POWER_WELL_DPIO_TX_C_LANES_01,
6486 },
6487 {
6488 .name = "dpio-tx-c-23",
6489 .domains = VLV_DPIO_TX_B_LANES_01_POWER_DOMAINS |
6490 VLV_DPIO_TX_B_LANES_23_POWER_DOMAINS |
6491 VLV_DPIO_TX_C_LANES_01_POWER_DOMAINS |
6492 VLV_DPIO_TX_C_LANES_23_POWER_DOMAINS,
6493 .ops = &vlv_dpio_power_well_ops,
6494 .data = PUNIT_POWER_WELL_DPIO_TX_C_LANES_23,
6495 },
f099a3c6
JB		 6496 {
6497 .name = "dpio-common",
6498 .domains = VLV_DPIO_CMN_BC_POWER_DOMAINS,
6499 .data = PUNIT_POWER_WELL_DPIO_CMN_BC,
aa519f23 6500 .ops = &vlv_dpio_cmn_power_well_ops,
f099a3c6 6501 },
77961eb9
ID		 6502};
6503
d2011dc8
VS		 6504static struct i915_power_well *lookup_power_well(struct drm_i915_private *dev_priv,
6505 enum punit_power_well power_well_id)
6506{
6507 struct i915_power_domains *power_domains = &dev_priv->power_domains;
6508 struct i915_power_well *power_well;
6509 int i;
6510
6511 for_each_power_well(i, power_well, POWER_DOMAIN_MASK, power_domains) {
6512 if (power_well->data == power_well_id)
6513 return power_well;
6514 }
6515
6516 return NULL;
6517}
6518
c1ca727f
ID		 6519#define set_power_wells(power_domains, __power_wells) ({ \
6520 (power_domains)->power_wells = (__power_wells); \
6521 (power_domains)->power_well_count = ARRAY_SIZE(__power_wells); \
6522})
6523
da7e29bd 6524int intel_power_domains_init(struct drm_i915_private *dev_priv)
a38911a3 6525{
83c00f55 6526 struct i915_power_domains *power_domains = &dev_priv->power_domains;
c1ca727f 6527
83c00f55 6528 mutex_init(&power_domains->lock);
a38911a3 6529
c1ca727f
ID		 6530 /*
6531 * The enabling order will be from lower to higher indexed wells,
6532 * the disabling order is reversed.
6533 */
da7e29bd 6534 if (IS_HASWELL(dev_priv->dev)) {
c1ca727f
ID		 6535 set_power_wells(power_domains, hsw_power_wells);
6536 hsw_pwr = power_domains;
da7e29bd 6537 } else if (IS_BROADWELL(dev_priv->dev)) {
c1ca727f
ID		 6538 set_power_wells(power_domains, bdw_power_wells);
6539 hsw_pwr = power_domains;
77961eb9
ID		 6540 } else if (IS_VALLEYVIEW(dev_priv->dev)) {
6541 set_power_wells(power_domains, vlv_power_wells);
c1ca727f 6542 } else {
1c2256df 6543 set_power_wells(power_domains, i9xx_always_on_power_well);
c1ca727f 6544 }
a38911a3
WX		 6545
6546 return 0;
6547}
6548
da7e29bd 6549void intel_power_domains_remove(struct drm_i915_private *dev_priv)
a38911a3
WX		 6550{
6551 hsw_pwr = NULL;
6552}
6553
da7e29bd 6554static void intel_power_domains_resume(struct drm_i915_private *dev_priv)
9cdb826c 6555{
83c00f55
ID		 6556 struct i915_power_domains *power_domains = &dev_priv->power_domains;
6557 struct i915_power_well *power_well;
c1ca727f 6558 int i;
9cdb826c 6559
83c00f55 6560 mutex_lock(&power_domains->lock);
bfafe93a 6561 for_each_power_well(i, power_well, POWER_DOMAIN_MASK, power_domains) {
a45f4466 6562 power_well->ops->sync_hw(dev_priv, power_well);
bfafe93a
ID		 6563 power_well->hw_enabled = power_well->ops->is_enabled(dev_priv,
6564 power_well);
6565 }
83c00f55 6566 mutex_unlock(&power_domains->lock);
a38911a3
WX		 6567}
6568
d2011dc8
VS		 6569static void vlv_cmnlane_wa(struct drm_i915_private *dev_priv)
6570{
6571 struct i915_power_well *cmn =
6572 lookup_power_well(dev_priv, PUNIT_POWER_WELL_DPIO_CMN_BC);
6573 struct i915_power_well *disp2d =
6574 lookup_power_well(dev_priv, PUNIT_POWER_WELL_DISP2D);
6575
6576 /* nothing to do if common lane is already off */
6577 if (!cmn->ops->is_enabled(dev_priv, cmn))
6578 return;
6579
6580 /* If the display might be already active skip this */
6581 if (disp2d->ops->is_enabled(dev_priv, disp2d) &&
6582 I915_READ(DPIO_CTL) & DPIO_CMNRST)
6583 return;
6584
6585 DRM_DEBUG_KMS("toggling display PHY side reset\n");
6586
6587 /* cmnlane needs DPLL registers */
6588 disp2d->ops->enable(dev_priv, disp2d);
6589
6590 /*
6591 * From VLV2A0_DP_eDP_HDMI_DPIO_driver_vbios_notes_11.docx:
6592 * Need to assert and de-assert PHY SB reset by gating the
6593 * common lane power, then un-gating it.
6594 * Simply ungating isn't enough to reset the PHY enough to get
6595 * ports and lanes running.
6596 */
6597 cmn->ops->disable(dev_priv, cmn);
6598}
6599
da7e29bd 6600void intel_power_domains_init_hw(struct drm_i915_private *dev_priv)
d0d3e513 6601{
d2011dc8 6602 struct drm_device *dev = dev_priv->dev;
0d116a29
ID		 6603 struct i915_power_domains *power_domains = &dev_priv->power_domains;
6604
6605 power_domains->initializing = true;
d2011dc8
VS		 6606
6607 if (IS_VALLEYVIEW(dev) && !IS_CHERRYVIEW(dev)) {
6608 mutex_lock(&power_domains->lock);
6609 vlv_cmnlane_wa(dev_priv);
6610 mutex_unlock(&power_domains->lock);
6611 }
6612
fa42e23c 6613 /* For now, we need the power well to be always enabled. */
da7e29bd
ID		 6614 intel_display_set_init_power(dev_priv, true);
6615 intel_power_domains_resume(dev_priv);
0d116a29 6616 power_domains->initializing = false;
d0d3e513
ED		 6617}
6618
c67a470b
PZ		 6619void intel_aux_display_runtime_get(struct drm_i915_private *dev_priv)
6620{
d361ae26 6621 intel_runtime_pm_get(dev_priv);
c67a470b
PZ		 6622}
6623
6624void intel_aux_display_runtime_put(struct drm_i915_private *dev_priv)
6625{
d361ae26 6626 intel_runtime_pm_put(dev_priv);
c67a470b
PZ		 6627}
6628
8a187455
PZ		 6629void intel_runtime_pm_get(struct drm_i915_private *dev_priv)
6630{
6631 struct drm_device *dev = dev_priv->dev;
6632 struct device *device = &dev->pdev->dev;
6633
6634 if (!HAS_RUNTIME_PM(dev))
6635 return;
6636
6637 pm_runtime_get_sync(device);
6638 WARN(dev_priv->pm.suspended, "Device still suspended.\n");
6639}
6640
c6df39b5
ID		 6641void intel_runtime_pm_get_noresume(struct drm_i915_private *dev_priv)
6642{
6643 struct drm_device *dev = dev_priv->dev;
6644 struct device *device = &dev->pdev->dev;
6645
6646 if (!HAS_RUNTIME_PM(dev))
6647 return;
6648
6649 WARN(dev_priv->pm.suspended, "Getting nosync-ref while suspended.\n");
6650 pm_runtime_get_noresume(device);
6651}
6652
8a187455
PZ		 6653void intel_runtime_pm_put(struct drm_i915_private *dev_priv)
6654{
6655 struct drm_device *dev = dev_priv->dev;
6656 struct device *device = &dev->pdev->dev;
6657
6658 if (!HAS_RUNTIME_PM(dev))
6659 return;
6660
6661 pm_runtime_mark_last_busy(device);
6662 pm_runtime_put_autosuspend(device);
6663}
6664
6665void intel_init_runtime_pm(struct drm_i915_private *dev_priv)
6666{
6667 struct drm_device *dev = dev_priv->dev;
6668 struct device *device = &dev->pdev->dev;
6669
8a187455
PZ		 6670 if (!HAS_RUNTIME_PM(dev))
6671 return;
6672
6673 pm_runtime_set_active(device);
6674
aeab0b5a
ID		 6675 /*
6676 * RPM depends on RC6 to save restore the GT HW context, so make RC6 a
6677 * requirement.
6678 */
6679 if (!intel_enable_rc6(dev)) {
6680 DRM_INFO("RC6 disabled, disabling runtime PM support\n");
6681 return;
6682 }
6683
8a187455
PZ		 6684 pm_runtime_set_autosuspend_delay(device, 10000); /* 10s */
6685 pm_runtime_mark_last_busy(device);
6686 pm_runtime_use_autosuspend(device);
ba0239e0
PZ		 6687
6688 pm_runtime_put_autosuspend(device);
8a187455
PZ		 6689}
6690
6691void intel_fini_runtime_pm(struct drm_i915_private *dev_priv)
6692{
6693 struct drm_device *dev = dev_priv->dev;
6694 struct device *device = &dev->pdev->dev;
6695
6696 if (!HAS_RUNTIME_PM(dev))
6697 return;
6698
aeab0b5a
ID		 6699 if (!intel_enable_rc6(dev))
6700 return;
6701
8a187455
PZ		 6702 /* Make sure we're not suspended first. */
6703 pm_runtime_get_sync(device);
6704 pm_runtime_disable(device);
6705}
6706
1fa61106
ED		 6707/* Set up chip specific power management-related functions */
6708void intel_init_pm(struct drm_device *dev)
6709{
6710 struct drm_i915_private *dev_priv = dev->dev_private;
6711
3a77c4c4 6712 if (HAS_FBC(dev)) {
40045465 6713 if (INTEL_INFO(dev)->gen >= 7) {
1fa61106 6714 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
40045465
VS		 6715 dev_priv->display.enable_fbc = gen7_enable_fbc;
6716 dev_priv->display.disable_fbc = ironlake_disable_fbc;
6717 } else if (INTEL_INFO(dev)->gen >= 5) {
6718 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
6719 dev_priv->display.enable_fbc = ironlake_enable_fbc;
1fa61106
ED		 6720 dev_priv->display.disable_fbc = ironlake_disable_fbc;
6721 } else if (IS_GM45(dev)) {
6722 dev_priv->display.fbc_enabled = g4x_fbc_enabled;
6723 dev_priv->display.enable_fbc = g4x_enable_fbc;
6724 dev_priv->display.disable_fbc = g4x_disable_fbc;
40045465 6725 } else {
1fa61106
ED		 6726 dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
6727 dev_priv->display.enable_fbc = i8xx_enable_fbc;
6728 dev_priv->display.disable_fbc = i8xx_disable_fbc;
993495ae
VS		 6729
6730 /* This value was pulled out of someone's hat */
6731 I915_WRITE(FBC_CONTROL, 500 << FBC_CTL_INTERVAL_SHIFT);
1fa61106 6732 }
1fa61106
ED		 6733 }
6734
c921aba8
DV		 6735 /* For cxsr */
6736 if (IS_PINEVIEW(dev))
6737 i915_pineview_get_mem_freq(dev);
6738 else if (IS_GEN5(dev))
6739 i915_ironlake_get_mem_freq(dev);
6740
1fa61106
ED		 6741 /* For FIFO watermark updates */
6742 if (HAS_PCH_SPLIT(dev)) {
fa50ad61 6743 ilk_setup_wm_latency(dev);
53615a5e 6744
bd602544
VS		 6745 if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] &&
6746 dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
6747 (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] &&
6748 dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
6749 dev_priv->display.update_wm = ilk_update_wm;
6750 dev_priv->display.update_sprite_wm = ilk_update_sprite_wm;
6751 } else {
6752 DRM_DEBUG_KMS("Failed to read display plane latency. "
6753 "Disable CxSR\n");
6754 }
6755
6756 if (IS_GEN5(dev))
1fa61106 6757 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
bd602544 6758 else if (IS_GEN6(dev))
1fa61106 6759 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
bd602544 6760 else if (IS_IVYBRIDGE(dev))
1fa61106 6761 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
bd602544 6762 else if (IS_HASWELL(dev))
cad2a2d7 6763 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
bd602544 6764 else if (INTEL_INFO(dev)->gen == 8)
1020a5c2 6765 dev_priv->display.init_clock_gating = gen8_init_clock_gating;
a4565da8
VS		 6766 } else if (IS_CHERRYVIEW(dev)) {
6767 dev_priv->display.update_wm = valleyview_update_wm;
6768 dev_priv->display.init_clock_gating =
6769 cherryview_init_clock_gating;
1fa61106
ED		 6770 } else if (IS_VALLEYVIEW(dev)) {
6771 dev_priv->display.update_wm = valleyview_update_wm;
6772 dev_priv->display.init_clock_gating =
6773 valleyview_init_clock_gating;
1fa61106
ED		 6774 } else if (IS_PINEVIEW(dev)) {
6775 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
6776 dev_priv->is_ddr3,
6777 dev_priv->fsb_freq,
6778 dev_priv->mem_freq)) {
6779 DRM_INFO("failed to find known CxSR latency "
6780 "(found ddr%s fsb freq %d, mem freq %d), "
6781 "disabling CxSR\n",
6782 (dev_priv->is_ddr3 == 1) ? "3" : "2",
6783 dev_priv->fsb_freq, dev_priv->mem_freq);
6784 /* Disable CxSR and never update its watermark again */
6785 pineview_disable_cxsr(dev);
6786 dev_priv->display.update_wm = NULL;
6787 } else
6788 dev_priv->display.update_wm = pineview_update_wm;
6789 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
6790 } else if (IS_G4X(dev)) {
6791 dev_priv->display.update_wm = g4x_update_wm;
6792 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
6793 } else if (IS_GEN4(dev)) {
6794 dev_priv->display.update_wm = i965_update_wm;
6795 if (IS_CRESTLINE(dev))
6796 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
6797 else if (IS_BROADWATER(dev))
6798 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
6799 } else if (IS_GEN3(dev)) {
6800 dev_priv->display.update_wm = i9xx_update_wm;
6801 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
6802 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
feb56b93
DV		 6803 } else if (IS_GEN2(dev)) {
6804 if (INTEL_INFO(dev)->num_pipes == 1) {
6805 dev_priv->display.update_wm = i845_update_wm;
1fa61106 6806 dev_priv->display.get_fifo_size = i845_get_fifo_size;
feb56b93
DV		 6807 } else {
6808 dev_priv->display.update_wm = i9xx_update_wm;
1fa61106 6809 dev_priv->display.get_fifo_size = i830_get_fifo_size;
feb56b93
DV		 6810 }
6811
6812 if (IS_I85X(dev) || IS_I865G(dev))
6813 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
6814 else
6815 dev_priv->display.init_clock_gating = i830_init_clock_gating;
6816 } else {
6817 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
1fa61106
ED		 6818 }
6819}
6820
42c0526c
BW		 6821int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
6822{
4fc688ce 6823 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW		 6824
6825 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
6826 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
6827 return -EAGAIN;
6828 }
6829
6830 I915_WRITE(GEN6_PCODE_DATA, *val);
6831 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
6832
6833 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
6834 500)) {
6835 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
6836 return -ETIMEDOUT;
6837 }
6838
6839 *val = I915_READ(GEN6_PCODE_DATA);
6840 I915_WRITE(GEN6_PCODE_DATA, 0);
6841
6842 return 0;
6843}
6844
6845int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
6846{
4fc688ce 6847 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW		 6848
6849 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
6850 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
6851 return -EAGAIN;
6852 }
6853
6854 I915_WRITE(GEN6_PCODE_DATA, val);
6855 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
6856
6857 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
6858 500)) {
6859 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
6860 return -ETIMEDOUT;
6861 }
6862
6863 I915_WRITE(GEN6_PCODE_DATA, 0);
6864
6865 return 0;
6866}
a0e4e199 6867
2ec3815f 6868int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val)
855ba3be 6869{
07ab118b 6870 int div;
855ba3be 6871
07ab118b 6872 /* 4 x czclk */
2ec3815f 6873 switch (dev_priv->mem_freq) {
855ba3be 6874 case 800:
07ab118b 6875 div = 10;
855ba3be
JB		 6876 break;
6877 case 1066:
07ab118b 6878 div = 12;
855ba3be
JB		 6879 break;
6880 case 1333:
07ab118b 6881 div = 16;
855ba3be
JB		 6882 break;
6883 default:
6884 return -1;
6885 }
6886
2ec3815f 6887 return DIV_ROUND_CLOSEST(dev_priv->mem_freq * (val + 6 - 0xbd), 4 * div);
855ba3be
JB		 6888}
6889
2ec3815f 6890int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val)
855ba3be 6891{
07ab118b 6892 int mul;
855ba3be 6893
07ab118b 6894 /* 4 x czclk */
2ec3815f 6895 switch (dev_priv->mem_freq) {
855ba3be 6896 case 800:
07ab118b 6897 mul = 10;
855ba3be
JB		 6898 break;
6899 case 1066:
07ab118b 6900 mul = 12;
855ba3be
JB		 6901 break;
6902 case 1333:
07ab118b 6903 mul = 16;
855ba3be
JB		 6904 break;
6905 default:
6906 return -1;
6907 }
6908
2ec3815f 6909 return DIV_ROUND_CLOSEST(4 * mul * val, dev_priv->mem_freq) + 0xbd - 6;
855ba3be
JB		 6910}
6911
f742a552 6912void intel_pm_setup(struct drm_device *dev)
907b28c5
CW		 6913{
6914 struct drm_i915_private *dev_priv = dev->dev_private;
6915
f742a552
DV		 6916 mutex_init(&dev_priv->rps.hw_lock);
6917
907b28c5
CW		 6918 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
6919 intel_gen6_powersave_work);
5d584b2e 6920
33688d95 6921 dev_priv->pm.suspended = false;
5d584b2e 6922 dev_priv->pm.irqs_disabled = false;
907b28c5 6923}
ubuntu-bionic-kernel mirror