]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/gpu/drm/i915/intel_pm.c
drm/i915: Update rps interrupt limits
[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;
95 struct drm_framebuffer *fb = crtc->fb;
96 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
97 struct drm_i915_gem_object *obj = intel_fb->obj;
98 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
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;
152 struct drm_framebuffer *fb = crtc->fb;
153 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
154 struct drm_i915_gem_object *obj = intel_fb->obj;
155 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
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;
224 struct drm_framebuffer *fb = crtc->fb;
225 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
226 struct drm_i915_gem_object *obj = intel_fb->obj;
227 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
85208be0
ED
228 u32 dpfc_ctl;
229
46f3dab9 230 dpfc_ctl = DPFC_CTL_PLANE(intel_crtc->plane);
3fa2e0ee
VS
231 if (drm_format_plane_cpp(fb->pixel_format, 0) == 2)
232 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
233 else
234 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
d629336b
VS
235 dpfc_ctl |= DPFC_CTL_FENCE_EN;
236 if (IS_GEN5(dev))
237 dpfc_ctl |= obj->fence_reg;
85208be0 238
85208be0 239 I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
f343c5f6 240 I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID);
85208be0
ED
241 /* enable it... */
242 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
243
244 if (IS_GEN6(dev)) {
245 I915_WRITE(SNB_DPFC_CTL_SA,
246 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
247 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
248 sandybridge_blit_fbc_update(dev);
249 }
250
84f44ce7 251 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
85208be0
ED
252}
253
1fa61106 254static void ironlake_disable_fbc(struct drm_device *dev)
85208be0
ED
255{
256 struct drm_i915_private *dev_priv = dev->dev_private;
257 u32 dpfc_ctl;
258
259 /* Disable compression */
260 dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
261 if (dpfc_ctl & DPFC_CTL_EN) {
262 dpfc_ctl &= ~DPFC_CTL_EN;
263 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl);
264
265 DRM_DEBUG_KMS("disabled FBC\n");
266 }
267}
268
1fa61106 269static bool ironlake_fbc_enabled(struct drm_device *dev)
85208be0
ED
270{
271 struct drm_i915_private *dev_priv = dev->dev_private;
272
273 return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
274}
275
993495ae 276static void gen7_enable_fbc(struct drm_crtc *crtc)
abe959c7
RV
277{
278 struct drm_device *dev = crtc->dev;
279 struct drm_i915_private *dev_priv = dev->dev_private;
280 struct drm_framebuffer *fb = crtc->fb;
281 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
282 struct drm_i915_gem_object *obj = intel_fb->obj;
283 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3fa2e0ee 284 u32 dpfc_ctl;
abe959c7 285
3fa2e0ee
VS
286 dpfc_ctl = IVB_DPFC_CTL_PLANE(intel_crtc->plane);
287 if (drm_format_plane_cpp(fb->pixel_format, 0) == 2)
288 dpfc_ctl |= DPFC_CTL_LIMIT_2X;
289 else
290 dpfc_ctl |= DPFC_CTL_LIMIT_1X;
291 dpfc_ctl |= IVB_DPFC_CTL_FENCE_EN;
292
293 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
abe959c7 294
891348b2 295 if (IS_IVYBRIDGE(dev)) {
7dd23ba0 296 /* WaFbcAsynchFlipDisableFbcQueue:ivb */
891348b2 297 I915_WRITE(ILK_DISPLAY_CHICKEN1, ILK_FBCQ_DIS);
28554164 298 } else {
7dd23ba0 299 /* WaFbcAsynchFlipDisableFbcQueue:hsw */
28554164
RV
300 I915_WRITE(HSW_PIPE_SLICE_CHICKEN_1(intel_crtc->pipe),
301 HSW_BYPASS_FBC_QUEUE);
891348b2 302 }
b74ea102 303
abe959c7
RV
304 I915_WRITE(SNB_DPFC_CTL_SA,
305 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
306 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
307
308 sandybridge_blit_fbc_update(dev);
309
b19870ee 310 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
abe959c7
RV
311}
312
85208be0
ED
313bool intel_fbc_enabled(struct drm_device *dev)
314{
315 struct drm_i915_private *dev_priv = dev->dev_private;
316
317 if (!dev_priv->display.fbc_enabled)
318 return false;
319
320 return dev_priv->display.fbc_enabled(dev);
321}
322
323static void intel_fbc_work_fn(struct work_struct *__work)
324{
325 struct intel_fbc_work *work =
326 container_of(to_delayed_work(__work),
327 struct intel_fbc_work, work);
328 struct drm_device *dev = work->crtc->dev;
329 struct drm_i915_private *dev_priv = dev->dev_private;
330
331 mutex_lock(&dev->struct_mutex);
5c3fe8b0 332 if (work == dev_priv->fbc.fbc_work) {
85208be0
ED
333 /* Double check that we haven't switched fb without cancelling
334 * the prior work.
335 */
336 if (work->crtc->fb == work->fb) {
993495ae 337 dev_priv->display.enable_fbc(work->crtc);
85208be0 338
5c3fe8b0
BW
339 dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane;
340 dev_priv->fbc.fb_id = work->crtc->fb->base.id;
341 dev_priv->fbc.y = work->crtc->y;
85208be0
ED
342 }
343
5c3fe8b0 344 dev_priv->fbc.fbc_work = NULL;
85208be0
ED
345 }
346 mutex_unlock(&dev->struct_mutex);
347
348 kfree(work);
349}
350
351static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
352{
5c3fe8b0 353 if (dev_priv->fbc.fbc_work == NULL)
85208be0
ED
354 return;
355
356 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
357
358 /* Synchronisation is provided by struct_mutex and checking of
5c3fe8b0 359 * dev_priv->fbc.fbc_work, so we can perform the cancellation
85208be0
ED
360 * entirely asynchronously.
361 */
5c3fe8b0 362 if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work))
85208be0 363 /* tasklet was killed before being run, clean up */
5c3fe8b0 364 kfree(dev_priv->fbc.fbc_work);
85208be0
ED
365
366 /* Mark the work as no longer wanted so that if it does
367 * wake-up (because the work was already running and waiting
368 * for our mutex), it will discover that is no longer
369 * necessary to run.
370 */
5c3fe8b0 371 dev_priv->fbc.fbc_work = NULL;
85208be0
ED
372}
373
993495ae 374static void intel_enable_fbc(struct drm_crtc *crtc)
85208be0
ED
375{
376 struct intel_fbc_work *work;
377 struct drm_device *dev = crtc->dev;
378 struct drm_i915_private *dev_priv = dev->dev_private;
379
380 if (!dev_priv->display.enable_fbc)
381 return;
382
383 intel_cancel_fbc_work(dev_priv);
384
b14c5679 385 work = kzalloc(sizeof(*work), GFP_KERNEL);
85208be0 386 if (work == NULL) {
6cdcb5e7 387 DRM_ERROR("Failed to allocate FBC work structure\n");
993495ae 388 dev_priv->display.enable_fbc(crtc);
85208be0
ED
389 return;
390 }
391
392 work->crtc = crtc;
393 work->fb = crtc->fb;
85208be0
ED
394 INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
395
5c3fe8b0 396 dev_priv->fbc.fbc_work = work;
85208be0 397
85208be0
ED
398 /* Delay the actual enabling to let pageflipping cease and the
399 * display to settle before starting the compression. Note that
400 * this delay also serves a second purpose: it allows for a
401 * vblank to pass after disabling the FBC before we attempt
402 * to modify the control registers.
403 *
404 * A more complicated solution would involve tracking vblanks
405 * following the termination of the page-flipping sequence
406 * and indeed performing the enable as a co-routine and not
407 * waiting synchronously upon the vblank.
7457d617
DL
408 *
409 * WaFbcWaitForVBlankBeforeEnable:ilk,snb
85208be0
ED
410 */
411 schedule_delayed_work(&work->work, msecs_to_jiffies(50));
412}
413
414void intel_disable_fbc(struct drm_device *dev)
415{
416 struct drm_i915_private *dev_priv = dev->dev_private;
417
418 intel_cancel_fbc_work(dev_priv);
419
420 if (!dev_priv->display.disable_fbc)
421 return;
422
423 dev_priv->display.disable_fbc(dev);
5c3fe8b0 424 dev_priv->fbc.plane = -1;
85208be0
ED
425}
426
29ebf90f
CW
427static bool set_no_fbc_reason(struct drm_i915_private *dev_priv,
428 enum no_fbc_reason reason)
429{
430 if (dev_priv->fbc.no_fbc_reason == reason)
431 return false;
432
433 dev_priv->fbc.no_fbc_reason = reason;
434 return true;
435}
436
85208be0
ED
437/**
438 * intel_update_fbc - enable/disable FBC as needed
439 * @dev: the drm_device
440 *
441 * Set up the framebuffer compression hardware at mode set time. We
442 * enable it if possible:
443 * - plane A only (on pre-965)
444 * - no pixel mulitply/line duplication
445 * - no alpha buffer discard
446 * - no dual wide
f85da868 447 * - framebuffer <= max_hdisplay in width, max_vdisplay in height
85208be0
ED
448 *
449 * We can't assume that any compression will take place (worst case),
450 * so the compressed buffer has to be the same size as the uncompressed
451 * one. It also must reside (along with the line length buffer) in
452 * stolen memory.
453 *
454 * We need to enable/disable FBC on a global basis.
455 */
456void intel_update_fbc(struct drm_device *dev)
457{
458 struct drm_i915_private *dev_priv = dev->dev_private;
459 struct drm_crtc *crtc = NULL, *tmp_crtc;
460 struct intel_crtc *intel_crtc;
461 struct drm_framebuffer *fb;
462 struct intel_framebuffer *intel_fb;
463 struct drm_i915_gem_object *obj;
ef644fda 464 const struct drm_display_mode *adjusted_mode;
37327abd 465 unsigned int max_width, max_height;
85208be0 466
3a77c4c4 467 if (!HAS_FBC(dev)) {
29ebf90f 468 set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED);
85208be0 469 return;
29ebf90f 470 }
85208be0 471
d330a953 472 if (!i915.powersave) {
29ebf90f
CW
473 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
474 DRM_DEBUG_KMS("fbc disabled per module param\n");
85208be0 475 return;
29ebf90f 476 }
85208be0
ED
477
478 /*
479 * If FBC is already on, we just have to verify that we can
480 * keep it that way...
481 * Need to disable if:
482 * - more than one pipe is active
483 * - changing FBC params (stride, fence, mode)
484 * - new fb is too large to fit in compressed buffer
485 * - going to an unsupported config (interlace, pixel multiply, etc.)
486 */
487 list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) {
3490ea5d 488 if (intel_crtc_active(tmp_crtc) &&
4c445e0e 489 to_intel_crtc(tmp_crtc)->primary_enabled) {
85208be0 490 if (crtc) {
29ebf90f
CW
491 if (set_no_fbc_reason(dev_priv, FBC_MULTIPLE_PIPES))
492 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
85208be0
ED
493 goto out_disable;
494 }
495 crtc = tmp_crtc;
496 }
497 }
498
499 if (!crtc || crtc->fb == NULL) {
29ebf90f
CW
500 if (set_no_fbc_reason(dev_priv, FBC_NO_OUTPUT))
501 DRM_DEBUG_KMS("no output, disabling\n");
85208be0
ED
502 goto out_disable;
503 }
504
505 intel_crtc = to_intel_crtc(crtc);
506 fb = crtc->fb;
507 intel_fb = to_intel_framebuffer(fb);
508 obj = intel_fb->obj;
ef644fda 509 adjusted_mode = &intel_crtc->config.adjusted_mode;
85208be0 510
d330a953 511 if (i915.enable_fbc < 0 &&
8a5729a3 512 INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev)) {
29ebf90f
CW
513 if (set_no_fbc_reason(dev_priv, FBC_CHIP_DEFAULT))
514 DRM_DEBUG_KMS("disabled per chip default\n");
8a5729a3 515 goto out_disable;
85208be0 516 }
d330a953 517 if (!i915.enable_fbc) {
29ebf90f
CW
518 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
519 DRM_DEBUG_KMS("fbc disabled per module param\n");
85208be0
ED
520 goto out_disable;
521 }
ef644fda
VS
522 if ((adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) ||
523 (adjusted_mode->flags & DRM_MODE_FLAG_DBLSCAN)) {
29ebf90f
CW
524 if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE))
525 DRM_DEBUG_KMS("mode incompatible with compression, "
526 "disabling\n");
85208be0
ED
527 goto out_disable;
528 }
f85da868
PZ
529
530 if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) {
37327abd
VS
531 max_width = 4096;
532 max_height = 2048;
f85da868 533 } else {
37327abd
VS
534 max_width = 2048;
535 max_height = 1536;
f85da868 536 }
37327abd
VS
537 if (intel_crtc->config.pipe_src_w > max_width ||
538 intel_crtc->config.pipe_src_h > max_height) {
29ebf90f
CW
539 if (set_no_fbc_reason(dev_priv, FBC_MODE_TOO_LARGE))
540 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
85208be0
ED
541 goto out_disable;
542 }
c5a44aa0
VS
543 if ((INTEL_INFO(dev)->gen < 4 || IS_HASWELL(dev)) &&
544 intel_crtc->plane != PLANE_A) {
29ebf90f 545 if (set_no_fbc_reason(dev_priv, FBC_BAD_PLANE))
c5a44aa0 546 DRM_DEBUG_KMS("plane not A, disabling compression\n");
85208be0
ED
547 goto out_disable;
548 }
549
550 /* The use of a CPU fence is mandatory in order to detect writes
551 * by the CPU to the scanout and trigger updates to the FBC.
552 */
553 if (obj->tiling_mode != I915_TILING_X ||
554 obj->fence_reg == I915_FENCE_REG_NONE) {
29ebf90f
CW
555 if (set_no_fbc_reason(dev_priv, FBC_NOT_TILED))
556 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
85208be0
ED
557 goto out_disable;
558 }
559
560 /* If the kernel debugger is active, always disable compression */
561 if (in_dbg_master())
562 goto out_disable;
563
11be49eb 564 if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) {
29ebf90f
CW
565 if (set_no_fbc_reason(dev_priv, FBC_STOLEN_TOO_SMALL))
566 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
11be49eb
CW
567 goto out_disable;
568 }
569
85208be0
ED
570 /* If the scanout has not changed, don't modify the FBC settings.
571 * Note that we make the fundamental assumption that the fb->obj
572 * cannot be unpinned (and have its GTT offset and fence revoked)
573 * without first being decoupled from the scanout and FBC disabled.
574 */
5c3fe8b0
BW
575 if (dev_priv->fbc.plane == intel_crtc->plane &&
576 dev_priv->fbc.fb_id == fb->base.id &&
577 dev_priv->fbc.y == crtc->y)
85208be0
ED
578 return;
579
580 if (intel_fbc_enabled(dev)) {
581 /* We update FBC along two paths, after changing fb/crtc
582 * configuration (modeswitching) and after page-flipping
583 * finishes. For the latter, we know that not only did
584 * we disable the FBC at the start of the page-flip
585 * sequence, but also more than one vblank has passed.
586 *
587 * For the former case of modeswitching, it is possible
588 * to switch between two FBC valid configurations
589 * instantaneously so we do need to disable the FBC
590 * before we can modify its control registers. We also
591 * have to wait for the next vblank for that to take
592 * effect. However, since we delay enabling FBC we can
593 * assume that a vblank has passed since disabling and
594 * that we can safely alter the registers in the deferred
595 * callback.
596 *
597 * In the scenario that we go from a valid to invalid
598 * and then back to valid FBC configuration we have
599 * no strict enforcement that a vblank occurred since
600 * disabling the FBC. However, along all current pipe
601 * disabling paths we do need to wait for a vblank at
602 * some point. And we wait before enabling FBC anyway.
603 */
604 DRM_DEBUG_KMS("disabling active FBC for update\n");
605 intel_disable_fbc(dev);
606 }
607
993495ae 608 intel_enable_fbc(crtc);
29ebf90f 609 dev_priv->fbc.no_fbc_reason = FBC_OK;
85208be0
ED
610 return;
611
612out_disable:
613 /* Multiple disables should be harmless */
614 if (intel_fbc_enabled(dev)) {
615 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
616 intel_disable_fbc(dev);
617 }
11be49eb 618 i915_gem_stolen_cleanup_compression(dev);
85208be0
ED
619}
620
c921aba8
DV
621static void i915_pineview_get_mem_freq(struct drm_device *dev)
622{
623 drm_i915_private_t *dev_priv = dev->dev_private;
624 u32 tmp;
625
626 tmp = I915_READ(CLKCFG);
627
628 switch (tmp & CLKCFG_FSB_MASK) {
629 case CLKCFG_FSB_533:
630 dev_priv->fsb_freq = 533; /* 133*4 */
631 break;
632 case CLKCFG_FSB_800:
633 dev_priv->fsb_freq = 800; /* 200*4 */
634 break;
635 case CLKCFG_FSB_667:
636 dev_priv->fsb_freq = 667; /* 167*4 */
637 break;
638 case CLKCFG_FSB_400:
639 dev_priv->fsb_freq = 400; /* 100*4 */
640 break;
641 }
642
643 switch (tmp & CLKCFG_MEM_MASK) {
644 case CLKCFG_MEM_533:
645 dev_priv->mem_freq = 533;
646 break;
647 case CLKCFG_MEM_667:
648 dev_priv->mem_freq = 667;
649 break;
650 case CLKCFG_MEM_800:
651 dev_priv->mem_freq = 800;
652 break;
653 }
654
655 /* detect pineview DDR3 setting */
656 tmp = I915_READ(CSHRDDR3CTL);
657 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
658}
659
660static void i915_ironlake_get_mem_freq(struct drm_device *dev)
661{
662 drm_i915_private_t *dev_priv = dev->dev_private;
663 u16 ddrpll, csipll;
664
665 ddrpll = I915_READ16(DDRMPLL1);
666 csipll = I915_READ16(CSIPLL0);
667
668 switch (ddrpll & 0xff) {
669 case 0xc:
670 dev_priv->mem_freq = 800;
671 break;
672 case 0x10:
673 dev_priv->mem_freq = 1066;
674 break;
675 case 0x14:
676 dev_priv->mem_freq = 1333;
677 break;
678 case 0x18:
679 dev_priv->mem_freq = 1600;
680 break;
681 default:
682 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
683 ddrpll & 0xff);
684 dev_priv->mem_freq = 0;
685 break;
686 }
687
20e4d407 688 dev_priv->ips.r_t = dev_priv->mem_freq;
c921aba8
DV
689
690 switch (csipll & 0x3ff) {
691 case 0x00c:
692 dev_priv->fsb_freq = 3200;
693 break;
694 case 0x00e:
695 dev_priv->fsb_freq = 3733;
696 break;
697 case 0x010:
698 dev_priv->fsb_freq = 4266;
699 break;
700 case 0x012:
701 dev_priv->fsb_freq = 4800;
702 break;
703 case 0x014:
704 dev_priv->fsb_freq = 5333;
705 break;
706 case 0x016:
707 dev_priv->fsb_freq = 5866;
708 break;
709 case 0x018:
710 dev_priv->fsb_freq = 6400;
711 break;
712 default:
713 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
714 csipll & 0x3ff);
715 dev_priv->fsb_freq = 0;
716 break;
717 }
718
719 if (dev_priv->fsb_freq == 3200) {
20e4d407 720 dev_priv->ips.c_m = 0;
c921aba8 721 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
20e4d407 722 dev_priv->ips.c_m = 1;
c921aba8 723 } else {
20e4d407 724 dev_priv->ips.c_m = 2;
c921aba8
DV
725 }
726}
727
b445e3b0
ED
728static const struct cxsr_latency cxsr_latency_table[] = {
729 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
730 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
731 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
732 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
733 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
734
735 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
736 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
737 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
738 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
739 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
740
741 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
742 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
743 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
744 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
745 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
746
747 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
748 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
749 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
750 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
751 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
752
753 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
754 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
755 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
756 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
757 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
758
759 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
760 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
761 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
762 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
763 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
764};
765
63c62275 766static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
b445e3b0
ED
767 int is_ddr3,
768 int fsb,
769 int mem)
770{
771 const struct cxsr_latency *latency;
772 int i;
773
774 if (fsb == 0 || mem == 0)
775 return NULL;
776
777 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
778 latency = &cxsr_latency_table[i];
779 if (is_desktop == latency->is_desktop &&
780 is_ddr3 == latency->is_ddr3 &&
781 fsb == latency->fsb_freq && mem == latency->mem_freq)
782 return latency;
783 }
784
785 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
786
787 return NULL;
788}
789
1fa61106 790static void pineview_disable_cxsr(struct drm_device *dev)
b445e3b0
ED
791{
792 struct drm_i915_private *dev_priv = dev->dev_private;
793
794 /* deactivate cxsr */
795 I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
796}
797
798/*
799 * Latency for FIFO fetches is dependent on several factors:
800 * - memory configuration (speed, channels)
801 * - chipset
802 * - current MCH state
803 * It can be fairly high in some situations, so here we assume a fairly
804 * pessimal value. It's a tradeoff between extra memory fetches (if we
805 * set this value too high, the FIFO will fetch frequently to stay full)
806 * and power consumption (set it too low to save power and we might see
807 * FIFO underruns and display "flicker").
808 *
809 * A value of 5us seems to be a good balance; safe for very low end
810 * platforms but not overly aggressive on lower latency configs.
811 */
812static const int latency_ns = 5000;
813
1fa61106 814static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED
815{
816 struct drm_i915_private *dev_priv = dev->dev_private;
817 uint32_t dsparb = I915_READ(DSPARB);
818 int size;
819
820 size = dsparb & 0x7f;
821 if (plane)
822 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
823
824 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
825 plane ? "B" : "A", size);
826
827 return size;
828}
829
feb56b93 830static int i830_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED
831{
832 struct drm_i915_private *dev_priv = dev->dev_private;
833 uint32_t dsparb = I915_READ(DSPARB);
834 int size;
835
836 size = dsparb & 0x1ff;
837 if (plane)
838 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
839 size >>= 1; /* Convert to cachelines */
840
841 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
842 plane ? "B" : "A", size);
843
844 return size;
845}
846
1fa61106 847static int i845_get_fifo_size(struct drm_device *dev, int plane)
b445e3b0
ED
848{
849 struct drm_i915_private *dev_priv = dev->dev_private;
850 uint32_t dsparb = I915_READ(DSPARB);
851 int size;
852
853 size = dsparb & 0x7f;
854 size >>= 2; /* Convert to cachelines */
855
856 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
857 plane ? "B" : "A",
858 size);
859
860 return size;
861}
862
b445e3b0
ED
863/* Pineview has different values for various configs */
864static const struct intel_watermark_params pineview_display_wm = {
865 PINEVIEW_DISPLAY_FIFO,
866 PINEVIEW_MAX_WM,
867 PINEVIEW_DFT_WM,
868 PINEVIEW_GUARD_WM,
869 PINEVIEW_FIFO_LINE_SIZE
870};
871static const struct intel_watermark_params pineview_display_hplloff_wm = {
872 PINEVIEW_DISPLAY_FIFO,
873 PINEVIEW_MAX_WM,
874 PINEVIEW_DFT_HPLLOFF_WM,
875 PINEVIEW_GUARD_WM,
876 PINEVIEW_FIFO_LINE_SIZE
877};
878static const struct intel_watermark_params pineview_cursor_wm = {
879 PINEVIEW_CURSOR_FIFO,
880 PINEVIEW_CURSOR_MAX_WM,
881 PINEVIEW_CURSOR_DFT_WM,
882 PINEVIEW_CURSOR_GUARD_WM,
883 PINEVIEW_FIFO_LINE_SIZE,
884};
885static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
886 PINEVIEW_CURSOR_FIFO,
887 PINEVIEW_CURSOR_MAX_WM,
888 PINEVIEW_CURSOR_DFT_WM,
889 PINEVIEW_CURSOR_GUARD_WM,
890 PINEVIEW_FIFO_LINE_SIZE
891};
892static const struct intel_watermark_params g4x_wm_info = {
893 G4X_FIFO_SIZE,
894 G4X_MAX_WM,
895 G4X_MAX_WM,
896 2,
897 G4X_FIFO_LINE_SIZE,
898};
899static const struct intel_watermark_params g4x_cursor_wm_info = {
900 I965_CURSOR_FIFO,
901 I965_CURSOR_MAX_WM,
902 I965_CURSOR_DFT_WM,
903 2,
904 G4X_FIFO_LINE_SIZE,
905};
906static const struct intel_watermark_params valleyview_wm_info = {
907 VALLEYVIEW_FIFO_SIZE,
908 VALLEYVIEW_MAX_WM,
909 VALLEYVIEW_MAX_WM,
910 2,
911 G4X_FIFO_LINE_SIZE,
912};
913static const struct intel_watermark_params valleyview_cursor_wm_info = {
914 I965_CURSOR_FIFO,
915 VALLEYVIEW_CURSOR_MAX_WM,
916 I965_CURSOR_DFT_WM,
917 2,
918 G4X_FIFO_LINE_SIZE,
919};
920static const struct intel_watermark_params i965_cursor_wm_info = {
921 I965_CURSOR_FIFO,
922 I965_CURSOR_MAX_WM,
923 I965_CURSOR_DFT_WM,
924 2,
925 I915_FIFO_LINE_SIZE,
926};
927static const struct intel_watermark_params i945_wm_info = {
928 I945_FIFO_SIZE,
929 I915_MAX_WM,
930 1,
931 2,
932 I915_FIFO_LINE_SIZE
933};
934static const struct intel_watermark_params i915_wm_info = {
935 I915_FIFO_SIZE,
936 I915_MAX_WM,
937 1,
938 2,
939 I915_FIFO_LINE_SIZE
940};
feb56b93 941static const struct intel_watermark_params i830_wm_info = {
b445e3b0
ED
942 I855GM_FIFO_SIZE,
943 I915_MAX_WM,
944 1,
945 2,
946 I830_FIFO_LINE_SIZE
947};
feb56b93 948static const struct intel_watermark_params i845_wm_info = {
b445e3b0
ED
949 I830_FIFO_SIZE,
950 I915_MAX_WM,
951 1,
952 2,
953 I830_FIFO_LINE_SIZE
954};
955
b445e3b0
ED
956/**
957 * intel_calculate_wm - calculate watermark level
958 * @clock_in_khz: pixel clock
959 * @wm: chip FIFO params
960 * @pixel_size: display pixel size
961 * @latency_ns: memory latency for the platform
962 *
963 * Calculate the watermark level (the level at which the display plane will
964 * start fetching from memory again). Each chip has a different display
965 * FIFO size and allocation, so the caller needs to figure that out and pass
966 * in the correct intel_watermark_params structure.
967 *
968 * As the pixel clock runs, the FIFO will be drained at a rate that depends
969 * on the pixel size. When it reaches the watermark level, it'll start
970 * fetching FIFO line sized based chunks from memory until the FIFO fills
971 * past the watermark point. If the FIFO drains completely, a FIFO underrun
972 * will occur, and a display engine hang could result.
973 */
974static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
975 const struct intel_watermark_params *wm,
976 int fifo_size,
977 int pixel_size,
978 unsigned long latency_ns)
979{
980 long entries_required, wm_size;
981
982 /*
983 * Note: we need to make sure we don't overflow for various clock &
984 * latency values.
985 * clocks go from a few thousand to several hundred thousand.
986 * latency is usually a few thousand
987 */
988 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
989 1000;
990 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
991
992 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
993
994 wm_size = fifo_size - (entries_required + wm->guard_size);
995
996 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
997
998 /* Don't promote wm_size to unsigned... */
999 if (wm_size > (long)wm->max_wm)
1000 wm_size = wm->max_wm;
1001 if (wm_size <= 0)
1002 wm_size = wm->default_wm;
1003 return wm_size;
1004}
1005
1006static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1007{
1008 struct drm_crtc *crtc, *enabled = NULL;
1009
1010 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
3490ea5d 1011 if (intel_crtc_active(crtc)) {
b445e3b0
ED
1012 if (enabled)
1013 return NULL;
1014 enabled = crtc;
1015 }
1016 }
1017
1018 return enabled;
1019}
1020
46ba614c 1021static void pineview_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1022{
46ba614c 1023 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1024 struct drm_i915_private *dev_priv = dev->dev_private;
1025 struct drm_crtc *crtc;
1026 const struct cxsr_latency *latency;
1027 u32 reg;
1028 unsigned long wm;
1029
1030 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1031 dev_priv->fsb_freq, dev_priv->mem_freq);
1032 if (!latency) {
1033 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1034 pineview_disable_cxsr(dev);
1035 return;
1036 }
1037
1038 crtc = single_enabled_crtc(dev);
1039 if (crtc) {
241bfc38 1040 const struct drm_display_mode *adjusted_mode;
b445e3b0 1041 int pixel_size = crtc->fb->bits_per_pixel / 8;
241bfc38
DL
1042 int clock;
1043
1044 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1045 clock = adjusted_mode->crtc_clock;
b445e3b0
ED
1046
1047 /* Display SR */
1048 wm = intel_calculate_wm(clock, &pineview_display_wm,
1049 pineview_display_wm.fifo_size,
1050 pixel_size, latency->display_sr);
1051 reg = I915_READ(DSPFW1);
1052 reg &= ~DSPFW_SR_MASK;
1053 reg |= wm << DSPFW_SR_SHIFT;
1054 I915_WRITE(DSPFW1, reg);
1055 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1056
1057 /* cursor SR */
1058 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1059 pineview_display_wm.fifo_size,
1060 pixel_size, latency->cursor_sr);
1061 reg = I915_READ(DSPFW3);
1062 reg &= ~DSPFW_CURSOR_SR_MASK;
1063 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1064 I915_WRITE(DSPFW3, reg);
1065
1066 /* Display HPLL off SR */
1067 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1068 pineview_display_hplloff_wm.fifo_size,
1069 pixel_size, latency->display_hpll_disable);
1070 reg = I915_READ(DSPFW3);
1071 reg &= ~DSPFW_HPLL_SR_MASK;
1072 reg |= wm & DSPFW_HPLL_SR_MASK;
1073 I915_WRITE(DSPFW3, reg);
1074
1075 /* cursor HPLL off SR */
1076 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1077 pineview_display_hplloff_wm.fifo_size,
1078 pixel_size, latency->cursor_hpll_disable);
1079 reg = I915_READ(DSPFW3);
1080 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1081 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1082 I915_WRITE(DSPFW3, reg);
1083 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1084
1085 /* activate cxsr */
1086 I915_WRITE(DSPFW3,
1087 I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1088 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1089 } else {
1090 pineview_disable_cxsr(dev);
1091 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1092 }
1093}
1094
1095static bool g4x_compute_wm0(struct drm_device *dev,
1096 int plane,
1097 const struct intel_watermark_params *display,
1098 int display_latency_ns,
1099 const struct intel_watermark_params *cursor,
1100 int cursor_latency_ns,
1101 int *plane_wm,
1102 int *cursor_wm)
1103{
1104 struct drm_crtc *crtc;
4fe8590a 1105 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED
1106 int htotal, hdisplay, clock, pixel_size;
1107 int line_time_us, line_count;
1108 int entries, tlb_miss;
1109
1110 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 1111 if (!intel_crtc_active(crtc)) {
b445e3b0
ED
1112 *cursor_wm = cursor->guard_size;
1113 *plane_wm = display->guard_size;
1114 return false;
1115 }
1116
4fe8590a 1117 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1118 clock = adjusted_mode->crtc_clock;
fec8cba3 1119 htotal = adjusted_mode->crtc_htotal;
37327abd 1120 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
b445e3b0
ED
1121 pixel_size = crtc->fb->bits_per_pixel / 8;
1122
1123 /* Use the small buffer method to calculate plane watermark */
1124 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1125 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1126 if (tlb_miss > 0)
1127 entries += tlb_miss;
1128 entries = DIV_ROUND_UP(entries, display->cacheline_size);
1129 *plane_wm = entries + display->guard_size;
1130 if (*plane_wm > (int)display->max_wm)
1131 *plane_wm = display->max_wm;
1132
1133 /* Use the large buffer method to calculate cursor watermark */
1134 line_time_us = ((htotal * 1000) / clock);
1135 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
1136 entries = line_count * 64 * pixel_size;
1137 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1138 if (tlb_miss > 0)
1139 entries += tlb_miss;
1140 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1141 *cursor_wm = entries + cursor->guard_size;
1142 if (*cursor_wm > (int)cursor->max_wm)
1143 *cursor_wm = (int)cursor->max_wm;
1144
1145 return true;
1146}
1147
1148/*
1149 * Check the wm result.
1150 *
1151 * If any calculated watermark values is larger than the maximum value that
1152 * can be programmed into the associated watermark register, that watermark
1153 * must be disabled.
1154 */
1155static bool g4x_check_srwm(struct drm_device *dev,
1156 int display_wm, int cursor_wm,
1157 const struct intel_watermark_params *display,
1158 const struct intel_watermark_params *cursor)
1159{
1160 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1161 display_wm, cursor_wm);
1162
1163 if (display_wm > display->max_wm) {
1164 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1165 display_wm, display->max_wm);
1166 return false;
1167 }
1168
1169 if (cursor_wm > cursor->max_wm) {
1170 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1171 cursor_wm, cursor->max_wm);
1172 return false;
1173 }
1174
1175 if (!(display_wm || cursor_wm)) {
1176 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1177 return false;
1178 }
1179
1180 return true;
1181}
1182
1183static bool g4x_compute_srwm(struct drm_device *dev,
1184 int plane,
1185 int latency_ns,
1186 const struct intel_watermark_params *display,
1187 const struct intel_watermark_params *cursor,
1188 int *display_wm, int *cursor_wm)
1189{
1190 struct drm_crtc *crtc;
4fe8590a 1191 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED
1192 int hdisplay, htotal, pixel_size, clock;
1193 unsigned long line_time_us;
1194 int line_count, line_size;
1195 int small, large;
1196 int entries;
1197
1198 if (!latency_ns) {
1199 *display_wm = *cursor_wm = 0;
1200 return false;
1201 }
1202
1203 crtc = intel_get_crtc_for_plane(dev, plane);
4fe8590a 1204 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1205 clock = adjusted_mode->crtc_clock;
fec8cba3 1206 htotal = adjusted_mode->crtc_htotal;
37327abd 1207 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
b445e3b0
ED
1208 pixel_size = crtc->fb->bits_per_pixel / 8;
1209
1210 line_time_us = (htotal * 1000) / clock;
1211 line_count = (latency_ns / line_time_us + 1000) / 1000;
1212 line_size = hdisplay * pixel_size;
1213
1214 /* Use the minimum of the small and large buffer method for primary */
1215 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1216 large = line_count * line_size;
1217
1218 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1219 *display_wm = entries + display->guard_size;
1220
1221 /* calculate the self-refresh watermark for display cursor */
1222 entries = line_count * pixel_size * 64;
1223 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1224 *cursor_wm = entries + cursor->guard_size;
1225
1226 return g4x_check_srwm(dev,
1227 *display_wm, *cursor_wm,
1228 display, cursor);
1229}
1230
1231static bool vlv_compute_drain_latency(struct drm_device *dev,
1232 int plane,
1233 int *plane_prec_mult,
1234 int *plane_dl,
1235 int *cursor_prec_mult,
1236 int *cursor_dl)
1237{
1238 struct drm_crtc *crtc;
1239 int clock, pixel_size;
1240 int entries;
1241
1242 crtc = intel_get_crtc_for_plane(dev, plane);
3490ea5d 1243 if (!intel_crtc_active(crtc))
b445e3b0
ED
1244 return false;
1245
241bfc38 1246 clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
b445e3b0
ED
1247 pixel_size = crtc->fb->bits_per_pixel / 8; /* BPP */
1248
1249 entries = (clock / 1000) * pixel_size;
1250 *plane_prec_mult = (entries > 256) ?
1251 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1252 *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1253 pixel_size);
1254
1255 entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */
1256 *cursor_prec_mult = (entries > 256) ?
1257 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1258 *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1259
1260 return true;
1261}
1262
1263/*
1264 * Update drain latency registers of memory arbiter
1265 *
1266 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1267 * to be programmed. Each plane has a drain latency multiplier and a drain
1268 * latency value.
1269 */
1270
1271static void vlv_update_drain_latency(struct drm_device *dev)
1272{
1273 struct drm_i915_private *dev_priv = dev->dev_private;
1274 int planea_prec, planea_dl, planeb_prec, planeb_dl;
1275 int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1276 int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1277 either 16 or 32 */
1278
1279 /* For plane A, Cursor A */
1280 if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1281 &cursor_prec_mult, &cursora_dl)) {
1282 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1283 DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1284 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1285 DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1286
1287 I915_WRITE(VLV_DDL1, cursora_prec |
1288 (cursora_dl << DDL_CURSORA_SHIFT) |
1289 planea_prec | planea_dl);
1290 }
1291
1292 /* For plane B, Cursor B */
1293 if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1294 &cursor_prec_mult, &cursorb_dl)) {
1295 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1296 DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1297 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1298 DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1299
1300 I915_WRITE(VLV_DDL2, cursorb_prec |
1301 (cursorb_dl << DDL_CURSORB_SHIFT) |
1302 planeb_prec | planeb_dl);
1303 }
1304}
1305
1306#define single_plane_enabled(mask) is_power_of_2(mask)
1307
46ba614c 1308static void valleyview_update_wm(struct drm_crtc *crtc)
b445e3b0 1309{
46ba614c 1310 struct drm_device *dev = crtc->dev;
b445e3b0
ED
1311 static const int sr_latency_ns = 12000;
1312 struct drm_i915_private *dev_priv = dev->dev_private;
1313 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1314 int plane_sr, cursor_sr;
af6c4575 1315 int ignore_plane_sr, ignore_cursor_sr;
b445e3b0
ED
1316 unsigned int enabled = 0;
1317
1318 vlv_update_drain_latency(dev);
1319
51cea1f4 1320 if (g4x_compute_wm0(dev, PIPE_A,
b445e3b0
ED
1321 &valleyview_wm_info, latency_ns,
1322 &valleyview_cursor_wm_info, latency_ns,
1323 &planea_wm, &cursora_wm))
51cea1f4 1324 enabled |= 1 << PIPE_A;
b445e3b0 1325
51cea1f4 1326 if (g4x_compute_wm0(dev, PIPE_B,
b445e3b0
ED
1327 &valleyview_wm_info, latency_ns,
1328 &valleyview_cursor_wm_info, latency_ns,
1329 &planeb_wm, &cursorb_wm))
51cea1f4 1330 enabled |= 1 << PIPE_B;
b445e3b0 1331
b445e3b0
ED
1332 if (single_plane_enabled(enabled) &&
1333 g4x_compute_srwm(dev, ffs(enabled) - 1,
1334 sr_latency_ns,
1335 &valleyview_wm_info,
1336 &valleyview_cursor_wm_info,
af6c4575
CW
1337 &plane_sr, &ignore_cursor_sr) &&
1338 g4x_compute_srwm(dev, ffs(enabled) - 1,
1339 2*sr_latency_ns,
1340 &valleyview_wm_info,
1341 &valleyview_cursor_wm_info,
52bd02d8 1342 &ignore_plane_sr, &cursor_sr)) {
b445e3b0 1343 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
52bd02d8 1344 } else {
b445e3b0
ED
1345 I915_WRITE(FW_BLC_SELF_VLV,
1346 I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
52bd02d8
CW
1347 plane_sr = cursor_sr = 0;
1348 }
b445e3b0
ED
1349
1350 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1351 planea_wm, cursora_wm,
1352 planeb_wm, cursorb_wm,
1353 plane_sr, cursor_sr);
1354
1355 I915_WRITE(DSPFW1,
1356 (plane_sr << DSPFW_SR_SHIFT) |
1357 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1358 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1359 planea_wm);
1360 I915_WRITE(DSPFW2,
8c919b28 1361 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
b445e3b0
ED
1362 (cursora_wm << DSPFW_CURSORA_SHIFT));
1363 I915_WRITE(DSPFW3,
8c919b28
CW
1364 (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1365 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
b445e3b0
ED
1366}
1367
46ba614c 1368static void g4x_update_wm(struct drm_crtc *crtc)
b445e3b0 1369{
46ba614c 1370 struct drm_device *dev = crtc->dev;
b445e3b0
ED
1371 static const int sr_latency_ns = 12000;
1372 struct drm_i915_private *dev_priv = dev->dev_private;
1373 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1374 int plane_sr, cursor_sr;
1375 unsigned int enabled = 0;
1376
51cea1f4 1377 if (g4x_compute_wm0(dev, PIPE_A,
b445e3b0
ED
1378 &g4x_wm_info, latency_ns,
1379 &g4x_cursor_wm_info, latency_ns,
1380 &planea_wm, &cursora_wm))
51cea1f4 1381 enabled |= 1 << PIPE_A;
b445e3b0 1382
51cea1f4 1383 if (g4x_compute_wm0(dev, PIPE_B,
b445e3b0
ED
1384 &g4x_wm_info, latency_ns,
1385 &g4x_cursor_wm_info, latency_ns,
1386 &planeb_wm, &cursorb_wm))
51cea1f4 1387 enabled |= 1 << PIPE_B;
b445e3b0 1388
b445e3b0
ED
1389 if (single_plane_enabled(enabled) &&
1390 g4x_compute_srwm(dev, ffs(enabled) - 1,
1391 sr_latency_ns,
1392 &g4x_wm_info,
1393 &g4x_cursor_wm_info,
52bd02d8 1394 &plane_sr, &cursor_sr)) {
b445e3b0 1395 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
52bd02d8 1396 } else {
b445e3b0
ED
1397 I915_WRITE(FW_BLC_SELF,
1398 I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
52bd02d8
CW
1399 plane_sr = cursor_sr = 0;
1400 }
b445e3b0
ED
1401
1402 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1403 planea_wm, cursora_wm,
1404 planeb_wm, cursorb_wm,
1405 plane_sr, cursor_sr);
1406
1407 I915_WRITE(DSPFW1,
1408 (plane_sr << DSPFW_SR_SHIFT) |
1409 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1410 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1411 planea_wm);
1412 I915_WRITE(DSPFW2,
8c919b28 1413 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
b445e3b0
ED
1414 (cursora_wm << DSPFW_CURSORA_SHIFT));
1415 /* HPLL off in SR has some issues on G4x... disable it */
1416 I915_WRITE(DSPFW3,
8c919b28 1417 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
b445e3b0
ED
1418 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1419}
1420
46ba614c 1421static void i965_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1422{
46ba614c 1423 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1424 struct drm_i915_private *dev_priv = dev->dev_private;
1425 struct drm_crtc *crtc;
1426 int srwm = 1;
1427 int cursor_sr = 16;
1428
1429 /* Calc sr entries for one plane configs */
1430 crtc = single_enabled_crtc(dev);
1431 if (crtc) {
1432 /* self-refresh has much higher latency */
1433 static const int sr_latency_ns = 12000;
4fe8590a
VS
1434 const struct drm_display_mode *adjusted_mode =
1435 &to_intel_crtc(crtc)->config.adjusted_mode;
241bfc38 1436 int clock = adjusted_mode->crtc_clock;
fec8cba3 1437 int htotal = adjusted_mode->crtc_htotal;
37327abd 1438 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
b445e3b0
ED
1439 int pixel_size = crtc->fb->bits_per_pixel / 8;
1440 unsigned long line_time_us;
1441 int entries;
1442
1443 line_time_us = ((htotal * 1000) / clock);
1444
1445 /* Use ns/us then divide to preserve precision */
1446 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1447 pixel_size * hdisplay;
1448 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1449 srwm = I965_FIFO_SIZE - entries;
1450 if (srwm < 0)
1451 srwm = 1;
1452 srwm &= 0x1ff;
1453 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1454 entries, srwm);
1455
1456 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1457 pixel_size * 64;
1458 entries = DIV_ROUND_UP(entries,
1459 i965_cursor_wm_info.cacheline_size);
1460 cursor_sr = i965_cursor_wm_info.fifo_size -
1461 (entries + i965_cursor_wm_info.guard_size);
1462
1463 if (cursor_sr > i965_cursor_wm_info.max_wm)
1464 cursor_sr = i965_cursor_wm_info.max_wm;
1465
1466 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1467 "cursor %d\n", srwm, cursor_sr);
1468
1469 if (IS_CRESTLINE(dev))
1470 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1471 } else {
1472 /* Turn off self refresh if both pipes are enabled */
1473 if (IS_CRESTLINE(dev))
1474 I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1475 & ~FW_BLC_SELF_EN);
1476 }
1477
1478 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1479 srwm);
1480
1481 /* 965 has limitations... */
1482 I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1483 (8 << 16) | (8 << 8) | (8 << 0));
1484 I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1485 /* update cursor SR watermark */
1486 I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1487}
1488
46ba614c 1489static void i9xx_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1490{
46ba614c 1491 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1492 struct drm_i915_private *dev_priv = dev->dev_private;
1493 const struct intel_watermark_params *wm_info;
1494 uint32_t fwater_lo;
1495 uint32_t fwater_hi;
1496 int cwm, srwm = 1;
1497 int fifo_size;
1498 int planea_wm, planeb_wm;
1499 struct drm_crtc *crtc, *enabled = NULL;
1500
1501 if (IS_I945GM(dev))
1502 wm_info = &i945_wm_info;
1503 else if (!IS_GEN2(dev))
1504 wm_info = &i915_wm_info;
1505 else
feb56b93 1506 wm_info = &i830_wm_info;
b445e3b0
ED
1507
1508 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1509 crtc = intel_get_crtc_for_plane(dev, 0);
3490ea5d 1510 if (intel_crtc_active(crtc)) {
241bfc38 1511 const struct drm_display_mode *adjusted_mode;
b9e0bda3
CW
1512 int cpp = crtc->fb->bits_per_pixel / 8;
1513 if (IS_GEN2(dev))
1514 cpp = 4;
1515
241bfc38
DL
1516 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1517 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1518 wm_info, fifo_size, cpp,
b445e3b0
ED
1519 latency_ns);
1520 enabled = crtc;
1521 } else
1522 planea_wm = fifo_size - wm_info->guard_size;
1523
1524 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1525 crtc = intel_get_crtc_for_plane(dev, 1);
3490ea5d 1526 if (intel_crtc_active(crtc)) {
241bfc38 1527 const struct drm_display_mode *adjusted_mode;
b9e0bda3
CW
1528 int cpp = crtc->fb->bits_per_pixel / 8;
1529 if (IS_GEN2(dev))
1530 cpp = 4;
1531
241bfc38
DL
1532 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1533 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
b9e0bda3 1534 wm_info, fifo_size, cpp,
b445e3b0
ED
1535 latency_ns);
1536 if (enabled == NULL)
1537 enabled = crtc;
1538 else
1539 enabled = NULL;
1540 } else
1541 planeb_wm = fifo_size - wm_info->guard_size;
1542
1543 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1544
1545 /*
1546 * Overlay gets an aggressive default since video jitter is bad.
1547 */
1548 cwm = 2;
1549
1550 /* Play safe and disable self-refresh before adjusting watermarks. */
1551 if (IS_I945G(dev) || IS_I945GM(dev))
1552 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1553 else if (IS_I915GM(dev))
3f2dc5ac 1554 I915_WRITE(INSTPM, _MASKED_BIT_DISABLE(INSTPM_SELF_EN));
b445e3b0
ED
1555
1556 /* Calc sr entries for one plane configs */
1557 if (HAS_FW_BLC(dev) && enabled) {
1558 /* self-refresh has much higher latency */
1559 static const int sr_latency_ns = 6000;
4fe8590a
VS
1560 const struct drm_display_mode *adjusted_mode =
1561 &to_intel_crtc(enabled)->config.adjusted_mode;
241bfc38 1562 int clock = adjusted_mode->crtc_clock;
fec8cba3 1563 int htotal = adjusted_mode->crtc_htotal;
f727b490 1564 int hdisplay = to_intel_crtc(enabled)->config.pipe_src_w;
b445e3b0
ED
1565 int pixel_size = enabled->fb->bits_per_pixel / 8;
1566 unsigned long line_time_us;
1567 int entries;
1568
1569 line_time_us = (htotal * 1000) / clock;
1570
1571 /* Use ns/us then divide to preserve precision */
1572 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1573 pixel_size * hdisplay;
1574 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1575 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1576 srwm = wm_info->fifo_size - entries;
1577 if (srwm < 0)
1578 srwm = 1;
1579
1580 if (IS_I945G(dev) || IS_I945GM(dev))
1581 I915_WRITE(FW_BLC_SELF,
1582 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1583 else if (IS_I915GM(dev))
1584 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1585 }
1586
1587 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1588 planea_wm, planeb_wm, cwm, srwm);
1589
1590 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1591 fwater_hi = (cwm & 0x1f);
1592
1593 /* Set request length to 8 cachelines per fetch */
1594 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1595 fwater_hi = fwater_hi | (1 << 8);
1596
1597 I915_WRITE(FW_BLC, fwater_lo);
1598 I915_WRITE(FW_BLC2, fwater_hi);
1599
1600 if (HAS_FW_BLC(dev)) {
1601 if (enabled) {
1602 if (IS_I945G(dev) || IS_I945GM(dev))
1603 I915_WRITE(FW_BLC_SELF,
1604 FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1605 else if (IS_I915GM(dev))
3f2dc5ac 1606 I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_SELF_EN));
b445e3b0
ED
1607 DRM_DEBUG_KMS("memory self refresh enabled\n");
1608 } else
1609 DRM_DEBUG_KMS("memory self refresh disabled\n");
1610 }
1611}
1612
feb56b93 1613static void i845_update_wm(struct drm_crtc *unused_crtc)
b445e3b0 1614{
46ba614c 1615 struct drm_device *dev = unused_crtc->dev;
b445e3b0
ED
1616 struct drm_i915_private *dev_priv = dev->dev_private;
1617 struct drm_crtc *crtc;
241bfc38 1618 const struct drm_display_mode *adjusted_mode;
b445e3b0
ED
1619 uint32_t fwater_lo;
1620 int planea_wm;
1621
1622 crtc = single_enabled_crtc(dev);
1623 if (crtc == NULL)
1624 return;
1625
241bfc38
DL
1626 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1627 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
feb56b93 1628 &i845_wm_info,
b445e3b0 1629 dev_priv->display.get_fifo_size(dev, 0),
b9e0bda3 1630 4, latency_ns);
b445e3b0
ED
1631 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1632 fwater_lo |= (3<<8) | planea_wm;
1633
1634 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1635
1636 I915_WRITE(FW_BLC, fwater_lo);
1637}
1638
3658729a
VS
1639static uint32_t ilk_pipe_pixel_rate(struct drm_device *dev,
1640 struct drm_crtc *crtc)
801bcfff
PZ
1641{
1642 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
fd4daa9c 1643 uint32_t pixel_rate;
801bcfff 1644
241bfc38 1645 pixel_rate = intel_crtc->config.adjusted_mode.crtc_clock;
801bcfff
PZ
1646
1647 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
1648 * adjust the pixel_rate here. */
1649
fd4daa9c 1650 if (intel_crtc->config.pch_pfit.enabled) {
801bcfff 1651 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
fd4daa9c 1652 uint32_t pfit_size = intel_crtc->config.pch_pfit.size;
801bcfff 1653
37327abd
VS
1654 pipe_w = intel_crtc->config.pipe_src_w;
1655 pipe_h = intel_crtc->config.pipe_src_h;
801bcfff
PZ
1656 pfit_w = (pfit_size >> 16) & 0xFFFF;
1657 pfit_h = pfit_size & 0xFFFF;
1658 if (pipe_w < pfit_w)
1659 pipe_w = pfit_w;
1660 if (pipe_h < pfit_h)
1661 pipe_h = pfit_h;
1662
1663 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
1664 pfit_w * pfit_h);
1665 }
1666
1667 return pixel_rate;
1668}
1669
37126462 1670/* latency must be in 0.1us units. */
23297044 1671static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
801bcfff
PZ
1672 uint32_t latency)
1673{
1674 uint64_t ret;
1675
3312ba65
VS
1676 if (WARN(latency == 0, "Latency value missing\n"))
1677 return UINT_MAX;
1678
801bcfff
PZ
1679 ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
1680 ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
1681
1682 return ret;
1683}
1684
37126462 1685/* latency must be in 0.1us units. */
23297044 1686static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
801bcfff
PZ
1687 uint32_t horiz_pixels, uint8_t bytes_per_pixel,
1688 uint32_t latency)
1689{
1690 uint32_t ret;
1691
3312ba65
VS
1692 if (WARN(latency == 0, "Latency value missing\n"))
1693 return UINT_MAX;
1694
801bcfff
PZ
1695 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
1696 ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
1697 ret = DIV_ROUND_UP(ret, 64) + 2;
1698 return ret;
1699}
1700
23297044 1701static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
cca32e9a
PZ
1702 uint8_t bytes_per_pixel)
1703{
1704 return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
1705}
1706
820c1980 1707struct ilk_pipe_wm_parameters {
801bcfff 1708 bool active;
801bcfff
PZ
1709 uint32_t pipe_htotal;
1710 uint32_t pixel_rate;
c35426d2
VS
1711 struct intel_plane_wm_parameters pri;
1712 struct intel_plane_wm_parameters spr;
1713 struct intel_plane_wm_parameters cur;
801bcfff
PZ
1714};
1715
820c1980 1716struct ilk_wm_maximums {
cca32e9a
PZ
1717 uint16_t pri;
1718 uint16_t spr;
1719 uint16_t cur;
1720 uint16_t fbc;
1721};
1722
240264f4
VS
1723/* used in computing the new watermarks state */
1724struct intel_wm_config {
1725 unsigned int num_pipes_active;
1726 bool sprites_enabled;
1727 bool sprites_scaled;
240264f4
VS
1728};
1729
37126462
VS
1730/*
1731 * For both WM_PIPE and WM_LP.
1732 * mem_value must be in 0.1us units.
1733 */
820c1980 1734static uint32_t ilk_compute_pri_wm(const struct ilk_pipe_wm_parameters *params,
cca32e9a
PZ
1735 uint32_t mem_value,
1736 bool is_lp)
801bcfff 1737{
cca32e9a
PZ
1738 uint32_t method1, method2;
1739
c35426d2 1740 if (!params->active || !params->pri.enabled)
801bcfff
PZ
1741 return 0;
1742
23297044 1743 method1 = ilk_wm_method1(params->pixel_rate,
c35426d2 1744 params->pri.bytes_per_pixel,
cca32e9a
PZ
1745 mem_value);
1746
1747 if (!is_lp)
1748 return method1;
1749
23297044 1750 method2 = ilk_wm_method2(params->pixel_rate,
cca32e9a 1751 params->pipe_htotal,
c35426d2
VS
1752 params->pri.horiz_pixels,
1753 params->pri.bytes_per_pixel,
cca32e9a
PZ
1754 mem_value);
1755
1756 return min(method1, method2);
801bcfff
PZ
1757}
1758
37126462
VS
1759/*
1760 * For both WM_PIPE and WM_LP.
1761 * mem_value must be in 0.1us units.
1762 */
820c1980 1763static uint32_t ilk_compute_spr_wm(const struct ilk_pipe_wm_parameters *params,
801bcfff
PZ
1764 uint32_t mem_value)
1765{
1766 uint32_t method1, method2;
1767
c35426d2 1768 if (!params->active || !params->spr.enabled)
801bcfff
PZ
1769 return 0;
1770
23297044 1771 method1 = ilk_wm_method1(params->pixel_rate,
c35426d2 1772 params->spr.bytes_per_pixel,
801bcfff 1773 mem_value);
23297044 1774 method2 = ilk_wm_method2(params->pixel_rate,
801bcfff 1775 params->pipe_htotal,
c35426d2
VS
1776 params->spr.horiz_pixels,
1777 params->spr.bytes_per_pixel,
801bcfff
PZ
1778 mem_value);
1779 return min(method1, method2);
1780}
1781
37126462
VS
1782/*
1783 * For both WM_PIPE and WM_LP.
1784 * mem_value must be in 0.1us units.
1785 */
820c1980 1786static uint32_t ilk_compute_cur_wm(const struct ilk_pipe_wm_parameters *params,
801bcfff
PZ
1787 uint32_t mem_value)
1788{
c35426d2 1789 if (!params->active || !params->cur.enabled)
801bcfff
PZ
1790 return 0;
1791
23297044 1792 return ilk_wm_method2(params->pixel_rate,
801bcfff 1793 params->pipe_htotal,
c35426d2
VS
1794 params->cur.horiz_pixels,
1795 params->cur.bytes_per_pixel,
801bcfff
PZ
1796 mem_value);
1797}
1798
cca32e9a 1799/* Only for WM_LP. */
820c1980 1800static uint32_t ilk_compute_fbc_wm(const struct ilk_pipe_wm_parameters *params,
1fda9882 1801 uint32_t pri_val)
cca32e9a 1802{
c35426d2 1803 if (!params->active || !params->pri.enabled)
cca32e9a
PZ
1804 return 0;
1805
23297044 1806 return ilk_wm_fbc(pri_val,
c35426d2
VS
1807 params->pri.horiz_pixels,
1808 params->pri.bytes_per_pixel);
cca32e9a
PZ
1809}
1810
158ae64f
VS
1811static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
1812{
416f4727
VS
1813 if (INTEL_INFO(dev)->gen >= 8)
1814 return 3072;
1815 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS
1816 return 768;
1817 else
1818 return 512;
1819}
1820
1821/* Calculate the maximum primary/sprite plane watermark */
1822static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
1823 int level,
240264f4 1824 const struct intel_wm_config *config,
158ae64f
VS
1825 enum intel_ddb_partitioning ddb_partitioning,
1826 bool is_sprite)
1827{
1828 unsigned int fifo_size = ilk_display_fifo_size(dev);
1829 unsigned int max;
1830
1831 /* if sprites aren't enabled, sprites get nothing */
240264f4 1832 if (is_sprite && !config->sprites_enabled)
158ae64f
VS
1833 return 0;
1834
1835 /* HSW allows LP1+ watermarks even with multiple pipes */
240264f4 1836 if (level == 0 || config->num_pipes_active > 1) {
158ae64f
VS
1837 fifo_size /= INTEL_INFO(dev)->num_pipes;
1838
1839 /*
1840 * For some reason the non self refresh
1841 * FIFO size is only half of the self
1842 * refresh FIFO size on ILK/SNB.
1843 */
1844 if (INTEL_INFO(dev)->gen <= 6)
1845 fifo_size /= 2;
1846 }
1847
240264f4 1848 if (config->sprites_enabled) {
158ae64f
VS
1849 /* level 0 is always calculated with 1:1 split */
1850 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
1851 if (is_sprite)
1852 fifo_size *= 5;
1853 fifo_size /= 6;
1854 } else {
1855 fifo_size /= 2;
1856 }
1857 }
1858
1859 /* clamp to max that the registers can hold */
416f4727
VS
1860 if (INTEL_INFO(dev)->gen >= 8)
1861 max = level == 0 ? 255 : 2047;
1862 else if (INTEL_INFO(dev)->gen >= 7)
158ae64f
VS
1863 /* IVB/HSW primary/sprite plane watermarks */
1864 max = level == 0 ? 127 : 1023;
1865 else if (!is_sprite)
1866 /* ILK/SNB primary plane watermarks */
1867 max = level == 0 ? 127 : 511;
1868 else
1869 /* ILK/SNB sprite plane watermarks */
1870 max = level == 0 ? 63 : 255;
1871
1872 return min(fifo_size, max);
1873}
1874
1875/* Calculate the maximum cursor plane watermark */
1876static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
240264f4
VS
1877 int level,
1878 const struct intel_wm_config *config)
158ae64f
VS
1879{
1880 /* HSW LP1+ watermarks w/ multiple pipes */
240264f4 1881 if (level > 0 && config->num_pipes_active > 1)
158ae64f
VS
1882 return 64;
1883
1884 /* otherwise just report max that registers can hold */
1885 if (INTEL_INFO(dev)->gen >= 7)
1886 return level == 0 ? 63 : 255;
1887 else
1888 return level == 0 ? 31 : 63;
1889}
1890
1891/* Calculate the maximum FBC watermark */
d34ff9c6 1892static unsigned int ilk_fbc_wm_max(const struct drm_device *dev)
158ae64f
VS
1893{
1894 /* max that registers can hold */
416f4727
VS
1895 if (INTEL_INFO(dev)->gen >= 8)
1896 return 31;
1897 else
1898 return 15;
158ae64f
VS
1899}
1900
d34ff9c6 1901static void ilk_compute_wm_maximums(const struct drm_device *dev,
34982fe1
VS
1902 int level,
1903 const struct intel_wm_config *config,
1904 enum intel_ddb_partitioning ddb_partitioning,
820c1980 1905 struct ilk_wm_maximums *max)
158ae64f 1906{
240264f4
VS
1907 max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
1908 max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
1909 max->cur = ilk_cursor_wm_max(dev, level, config);
416f4727 1910 max->fbc = ilk_fbc_wm_max(dev);
158ae64f
VS
1911}
1912
d9395655 1913static bool ilk_validate_wm_level(int level,
820c1980 1914 const struct ilk_wm_maximums *max,
d9395655 1915 struct intel_wm_level *result)
a9786a11
VS
1916{
1917 bool ret;
1918
1919 /* already determined to be invalid? */
1920 if (!result->enable)
1921 return false;
1922
1923 result->enable = result->pri_val <= max->pri &&
1924 result->spr_val <= max->spr &&
1925 result->cur_val <= max->cur;
1926
1927 ret = result->enable;
1928
1929 /*
1930 * HACK until we can pre-compute everything,
1931 * and thus fail gracefully if LP0 watermarks
1932 * are exceeded...
1933 */
1934 if (level == 0 && !result->enable) {
1935 if (result->pri_val > max->pri)
1936 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
1937 level, result->pri_val, max->pri);
1938 if (result->spr_val > max->spr)
1939 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
1940 level, result->spr_val, max->spr);
1941 if (result->cur_val > max->cur)
1942 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
1943 level, result->cur_val, max->cur);
1944
1945 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
1946 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
1947 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
1948 result->enable = true;
1949 }
1950
a9786a11
VS
1951 return ret;
1952}
1953
d34ff9c6 1954static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
6f5ddd17 1955 int level,
820c1980 1956 const struct ilk_pipe_wm_parameters *p,
1fd527cc 1957 struct intel_wm_level *result)
6f5ddd17
VS
1958{
1959 uint16_t pri_latency = dev_priv->wm.pri_latency[level];
1960 uint16_t spr_latency = dev_priv->wm.spr_latency[level];
1961 uint16_t cur_latency = dev_priv->wm.cur_latency[level];
1962
1963 /* WM1+ latency values stored in 0.5us units */
1964 if (level > 0) {
1965 pri_latency *= 5;
1966 spr_latency *= 5;
1967 cur_latency *= 5;
1968 }
1969
1970 result->pri_val = ilk_compute_pri_wm(p, pri_latency, level);
1971 result->spr_val = ilk_compute_spr_wm(p, spr_latency);
1972 result->cur_val = ilk_compute_cur_wm(p, cur_latency);
1973 result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val);
1974 result->enable = true;
1975}
1976
801bcfff
PZ
1977static uint32_t
1978hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
1f8eeabf
ED
1979{
1980 struct drm_i915_private *dev_priv = dev->dev_private;
1011d8c4 1981 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
1011d8c4 1982 struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode;
85a02deb 1983 u32 linetime, ips_linetime;
1f8eeabf 1984
801bcfff
PZ
1985 if (!intel_crtc_active(crtc))
1986 return 0;
1011d8c4 1987
1f8eeabf
ED
1988 /* The WM are computed with base on how long it takes to fill a single
1989 * row at the given clock rate, multiplied by 8.
1990 * */
fec8cba3
JB
1991 linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8,
1992 mode->crtc_clock);
1993 ips_linetime = DIV_ROUND_CLOSEST(mode->crtc_htotal * 1000 * 8,
85a02deb 1994 intel_ddi_get_cdclk_freq(dev_priv));
1f8eeabf 1995
801bcfff
PZ
1996 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
1997 PIPE_WM_LINETIME_TIME(linetime);
1f8eeabf
ED
1998}
1999
12b134df
VS
2000static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[5])
2001{
2002 struct drm_i915_private *dev_priv = dev->dev_private;
2003
a42a5719 2004 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
12b134df
VS
2005 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2006
2007 wm[0] = (sskpd >> 56) & 0xFF;
2008 if (wm[0] == 0)
2009 wm[0] = sskpd & 0xF;
e5d5019e
VS
2010 wm[1] = (sskpd >> 4) & 0xFF;
2011 wm[2] = (sskpd >> 12) & 0xFF;
2012 wm[3] = (sskpd >> 20) & 0x1FF;
2013 wm[4] = (sskpd >> 32) & 0x1FF;
63cf9a13
VS
2014 } else if (INTEL_INFO(dev)->gen >= 6) {
2015 uint32_t sskpd = I915_READ(MCH_SSKPD);
2016
2017 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2018 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2019 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2020 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
3a88d0ac
VS
2021 } else if (INTEL_INFO(dev)->gen >= 5) {
2022 uint32_t mltr = I915_READ(MLTR_ILK);
2023
2024 /* ILK primary LP0 latency is 700 ns */
2025 wm[0] = 7;
2026 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2027 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
12b134df
VS
2028 }
2029}
2030
53615a5e
VS
2031static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2032{
2033 /* ILK sprite LP0 latency is 1300 ns */
2034 if (INTEL_INFO(dev)->gen == 5)
2035 wm[0] = 13;
2036}
2037
2038static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2039{
2040 /* ILK cursor LP0 latency is 1300 ns */
2041 if (INTEL_INFO(dev)->gen == 5)
2042 wm[0] = 13;
2043
2044 /* WaDoubleCursorLP3Latency:ivb */
2045 if (IS_IVYBRIDGE(dev))
2046 wm[3] *= 2;
2047}
2048
ad0d6dc4 2049static int ilk_wm_max_level(const struct drm_device *dev)
26ec971e 2050{
26ec971e 2051 /* how many WM levels are we expecting */
a42a5719 2052 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ad0d6dc4 2053 return 4;
26ec971e 2054 else if (INTEL_INFO(dev)->gen >= 6)
ad0d6dc4 2055 return 3;
26ec971e 2056 else
ad0d6dc4
VS
2057 return 2;
2058}
2059
2060static void intel_print_wm_latency(struct drm_device *dev,
2061 const char *name,
2062 const uint16_t wm[5])
2063{
2064 int level, max_level = ilk_wm_max_level(dev);
26ec971e
VS
2065
2066 for (level = 0; level <= max_level; level++) {
2067 unsigned int latency = wm[level];
2068
2069 if (latency == 0) {
2070 DRM_ERROR("%s WM%d latency not provided\n",
2071 name, level);
2072 continue;
2073 }
2074
2075 /* WM1+ latency values in 0.5us units */
2076 if (level > 0)
2077 latency *= 5;
2078
2079 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2080 name, level, wm[level],
2081 latency / 10, latency % 10);
2082 }
2083}
2084
53615a5e
VS
2085static void intel_setup_wm_latency(struct drm_device *dev)
2086{
2087 struct drm_i915_private *dev_priv = dev->dev_private;
2088
2089 intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2090
2091 memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2092 sizeof(dev_priv->wm.pri_latency));
2093 memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2094 sizeof(dev_priv->wm.pri_latency));
2095
2096 intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2097 intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
26ec971e
VS
2098
2099 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2100 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2101 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
53615a5e
VS
2102}
2103
820c1980
ID
2104static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
2105 struct ilk_pipe_wm_parameters *p,
a485bfb8 2106 struct intel_wm_config *config)
1011d8c4 2107{
7c4a395f
VS
2108 struct drm_device *dev = crtc->dev;
2109 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2110 enum pipe pipe = intel_crtc->pipe;
7c4a395f 2111 struct drm_plane *plane;
1011d8c4 2112
7c4a395f
VS
2113 p->active = intel_crtc_active(crtc);
2114 if (p->active) {
576b259e 2115 p->pipe_htotal = intel_crtc->config.adjusted_mode.crtc_htotal;
3658729a 2116 p->pixel_rate = ilk_pipe_pixel_rate(dev, crtc);
c35426d2
VS
2117 p->pri.bytes_per_pixel = crtc->fb->bits_per_pixel / 8;
2118 p->cur.bytes_per_pixel = 4;
37327abd 2119 p->pri.horiz_pixels = intel_crtc->config.pipe_src_w;
c35426d2
VS
2120 p->cur.horiz_pixels = 64;
2121 /* TODO: for now, assume primary and cursor planes are always enabled. */
2122 p->pri.enabled = true;
2123 p->cur.enabled = true;
801bcfff
PZ
2124 }
2125
7c4a395f 2126 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
a485bfb8 2127 config->num_pipes_active += intel_crtc_active(crtc);
7c4a395f 2128
801bcfff
PZ
2129 list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
2130 struct intel_plane *intel_plane = to_intel_plane(plane);
801bcfff 2131
7c4a395f
VS
2132 if (intel_plane->pipe == pipe)
2133 p->spr = intel_plane->wm;
cca32e9a 2134
a485bfb8
VS
2135 config->sprites_enabled |= intel_plane->wm.enabled;
2136 config->sprites_scaled |= intel_plane->wm.scaled;
cca32e9a 2137 }
801bcfff
PZ
2138}
2139
0b2ae6d7
VS
2140/* Compute new watermarks for the pipe */
2141static bool intel_compute_pipe_wm(struct drm_crtc *crtc,
820c1980 2142 const struct ilk_pipe_wm_parameters *params,
0b2ae6d7
VS
2143 struct intel_pipe_wm *pipe_wm)
2144{
2145 struct drm_device *dev = crtc->dev;
d34ff9c6 2146 const struct drm_i915_private *dev_priv = dev->dev_private;
0b2ae6d7
VS
2147 int level, max_level = ilk_wm_max_level(dev);
2148 /* LP0 watermark maximums depend on this pipe alone */
2149 struct intel_wm_config config = {
2150 .num_pipes_active = 1,
2151 .sprites_enabled = params->spr.enabled,
2152 .sprites_scaled = params->spr.scaled,
2153 };
820c1980 2154 struct ilk_wm_maximums max;
0b2ae6d7 2155
0b2ae6d7 2156 /* LP0 watermarks always use 1/2 DDB partitioning */
34982fe1 2157 ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
0b2ae6d7 2158
7b39a0b7
VS
2159 /* ILK/SNB: LP2+ watermarks only w/o sprites */
2160 if (INTEL_INFO(dev)->gen <= 6 && params->spr.enabled)
2161 max_level = 1;
2162
2163 /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
2164 if (params->spr.scaled)
2165 max_level = 0;
2166
0b2ae6d7
VS
2167 for (level = 0; level <= max_level; level++)
2168 ilk_compute_wm_level(dev_priv, level, params,
2169 &pipe_wm->wm[level]);
2170
a42a5719 2171 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 2172 pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
0b2ae6d7
VS
2173
2174 /* At least LP0 must be valid */
d9395655 2175 return ilk_validate_wm_level(0, &max, &pipe_wm->wm[0]);
0b2ae6d7
VS
2176}
2177
2178/*
2179 * Merge the watermarks from all active pipes for a specific level.
2180 */
2181static void ilk_merge_wm_level(struct drm_device *dev,
2182 int level,
2183 struct intel_wm_level *ret_wm)
2184{
2185 const struct intel_crtc *intel_crtc;
2186
2187 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2188 const struct intel_wm_level *wm =
2189 &intel_crtc->wm.active.wm[level];
2190
2191 if (!wm->enable)
2192 return;
2193
2194 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2195 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2196 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2197 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2198 }
2199
2200 ret_wm->enable = true;
2201}
2202
2203/*
2204 * Merge all low power watermarks for all active pipes.
2205 */
2206static void ilk_wm_merge(struct drm_device *dev,
0ba22e26 2207 const struct intel_wm_config *config,
820c1980 2208 const struct ilk_wm_maximums *max,
0b2ae6d7
VS
2209 struct intel_pipe_wm *merged)
2210{
2211 int level, max_level = ilk_wm_max_level(dev);
2212
0ba22e26
VS
2213 /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
2214 if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) &&
2215 config->num_pipes_active > 1)
2216 return;
2217
6c8b6c28
VS
2218 /* ILK: FBC WM must be disabled always */
2219 merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6;
0b2ae6d7
VS
2220
2221 /* merge each WM1+ level */
2222 for (level = 1; level <= max_level; level++) {
2223 struct intel_wm_level *wm = &merged->wm[level];
2224
2225 ilk_merge_wm_level(dev, level, wm);
2226
d9395655 2227 if (!ilk_validate_wm_level(level, max, wm))
0b2ae6d7
VS
2228 break;
2229
2230 /*
2231 * The spec says it is preferred to disable
2232 * FBC WMs instead of disabling a WM level.
2233 */
2234 if (wm->fbc_val > max->fbc) {
2235 merged->fbc_wm_enabled = false;
2236 wm->fbc_val = 0;
2237 }
2238 }
6c8b6c28
VS
2239
2240 /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
2241 /*
2242 * FIXME this is racy. FBC might get enabled later.
2243 * What we should check here is whether FBC can be
2244 * enabled sometime later.
2245 */
2246 if (IS_GEN5(dev) && !merged->fbc_wm_enabled && intel_fbc_enabled(dev)) {
2247 for (level = 2; level <= max_level; level++) {
2248 struct intel_wm_level *wm = &merged->wm[level];
2249
2250 wm->enable = false;
2251 }
2252 }
0b2ae6d7
VS
2253}
2254
b380ca3c
VS
2255static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2256{
2257 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2258 return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2259}
2260
a68d68ee
VS
2261/* The value we need to program into the WM_LPx latency field */
2262static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level)
2263{
2264 struct drm_i915_private *dev_priv = dev->dev_private;
2265
a42a5719 2266 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
a68d68ee
VS
2267 return 2 * level;
2268 else
2269 return dev_priv->wm.pri_latency[level];
2270}
2271
820c1980 2272static void ilk_compute_wm_results(struct drm_device *dev,
0362c781 2273 const struct intel_pipe_wm *merged,
609cedef 2274 enum intel_ddb_partitioning partitioning,
820c1980 2275 struct ilk_wm_values *results)
801bcfff 2276{
0b2ae6d7
VS
2277 struct intel_crtc *intel_crtc;
2278 int level, wm_lp;
cca32e9a 2279
0362c781 2280 results->enable_fbc_wm = merged->fbc_wm_enabled;
609cedef 2281 results->partitioning = partitioning;
cca32e9a 2282
0b2ae6d7 2283 /* LP1+ register values */
cca32e9a 2284 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
1fd527cc 2285 const struct intel_wm_level *r;
801bcfff 2286
b380ca3c 2287 level = ilk_wm_lp_to_level(wm_lp, merged);
0b2ae6d7 2288
0362c781 2289 r = &merged->wm[level];
0b2ae6d7 2290 if (!r->enable)
cca32e9a
PZ
2291 break;
2292
416f4727 2293 results->wm_lp[wm_lp - 1] = WM3_LP_EN |
a68d68ee 2294 (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) |
416f4727
VS
2295 (r->pri_val << WM1_LP_SR_SHIFT) |
2296 r->cur_val;
2297
2298 if (INTEL_INFO(dev)->gen >= 8)
2299 results->wm_lp[wm_lp - 1] |=
2300 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
2301 else
2302 results->wm_lp[wm_lp - 1] |=
2303 r->fbc_val << WM1_LP_FBC_SHIFT;
2304
6cef2b8a
VS
2305 if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) {
2306 WARN_ON(wm_lp != 1);
2307 results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
2308 } else
2309 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
cca32e9a 2310 }
801bcfff 2311
0b2ae6d7
VS
2312 /* LP0 register values */
2313 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2314 enum pipe pipe = intel_crtc->pipe;
2315 const struct intel_wm_level *r =
2316 &intel_crtc->wm.active.wm[0];
2317
2318 if (WARN_ON(!r->enable))
2319 continue;
2320
2321 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
1011d8c4 2322
0b2ae6d7
VS
2323 results->wm_pipe[pipe] =
2324 (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2325 (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2326 r->cur_val;
801bcfff
PZ
2327 }
2328}
2329
861f3389
PZ
2330/* Find the result with the highest level enabled. Check for enable_fbc_wm in
2331 * case both are at the same level. Prefer r1 in case they're the same. */
820c1980 2332static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev,
198a1e9b
VS
2333 struct intel_pipe_wm *r1,
2334 struct intel_pipe_wm *r2)
861f3389 2335{
198a1e9b
VS
2336 int level, max_level = ilk_wm_max_level(dev);
2337 int level1 = 0, level2 = 0;
861f3389 2338
198a1e9b
VS
2339 for (level = 1; level <= max_level; level++) {
2340 if (r1->wm[level].enable)
2341 level1 = level;
2342 if (r2->wm[level].enable)
2343 level2 = level;
861f3389
PZ
2344 }
2345
198a1e9b
VS
2346 if (level1 == level2) {
2347 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
861f3389
PZ
2348 return r2;
2349 else
2350 return r1;
198a1e9b 2351 } else if (level1 > level2) {
861f3389
PZ
2352 return r1;
2353 } else {
2354 return r2;
2355 }
2356}
2357
49a687c4
VS
2358/* dirty bits used to track which watermarks need changes */
2359#define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2360#define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2361#define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2362#define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2363#define WM_DIRTY_FBC (1 << 24)
2364#define WM_DIRTY_DDB (1 << 25)
2365
2366static unsigned int ilk_compute_wm_dirty(struct drm_device *dev,
820c1980
ID
2367 const struct ilk_wm_values *old,
2368 const struct ilk_wm_values *new)
49a687c4
VS
2369{
2370 unsigned int dirty = 0;
2371 enum pipe pipe;
2372 int wm_lp;
2373
2374 for_each_pipe(pipe) {
2375 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
2376 dirty |= WM_DIRTY_LINETIME(pipe);
2377 /* Must disable LP1+ watermarks too */
2378 dirty |= WM_DIRTY_LP_ALL;
2379 }
2380
2381 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
2382 dirty |= WM_DIRTY_PIPE(pipe);
2383 /* Must disable LP1+ watermarks too */
2384 dirty |= WM_DIRTY_LP_ALL;
2385 }
2386 }
2387
2388 if (old->enable_fbc_wm != new->enable_fbc_wm) {
2389 dirty |= WM_DIRTY_FBC;
2390 /* Must disable LP1+ watermarks too */
2391 dirty |= WM_DIRTY_LP_ALL;
2392 }
2393
2394 if (old->partitioning != new->partitioning) {
2395 dirty |= WM_DIRTY_DDB;
2396 /* Must disable LP1+ watermarks too */
2397 dirty |= WM_DIRTY_LP_ALL;
2398 }
2399
2400 /* LP1+ watermarks already deemed dirty, no need to continue */
2401 if (dirty & WM_DIRTY_LP_ALL)
2402 return dirty;
2403
2404 /* Find the lowest numbered LP1+ watermark in need of an update... */
2405 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2406 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
2407 old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
2408 break;
2409 }
2410
2411 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2412 for (; wm_lp <= 3; wm_lp++)
2413 dirty |= WM_DIRTY_LP(wm_lp);
2414
2415 return dirty;
2416}
2417
8553c18e
VS
2418static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
2419 unsigned int dirty)
801bcfff 2420{
820c1980 2421 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e 2422 bool changed = false;
801bcfff 2423
facd619b
VS
2424 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
2425 previous->wm_lp[2] &= ~WM1_LP_SR_EN;
2426 I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
8553c18e 2427 changed = true;
facd619b
VS
2428 }
2429 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
2430 previous->wm_lp[1] &= ~WM1_LP_SR_EN;
2431 I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
8553c18e 2432 changed = true;
facd619b
VS
2433 }
2434 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
2435 previous->wm_lp[0] &= ~WM1_LP_SR_EN;
2436 I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
8553c18e 2437 changed = true;
facd619b 2438 }
801bcfff 2439
facd619b
VS
2440 /*
2441 * Don't touch WM1S_LP_EN here.
2442 * Doing so could cause underruns.
2443 */
6cef2b8a 2444
8553c18e
VS
2445 return changed;
2446}
2447
2448/*
2449 * The spec says we shouldn't write when we don't need, because every write
2450 * causes WMs to be re-evaluated, expending some power.
2451 */
820c1980
ID
2452static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
2453 struct ilk_wm_values *results)
8553c18e
VS
2454{
2455 struct drm_device *dev = dev_priv->dev;
820c1980 2456 struct ilk_wm_values *previous = &dev_priv->wm.hw;
8553c18e
VS
2457 unsigned int dirty;
2458 uint32_t val;
2459
2460 dirty = ilk_compute_wm_dirty(dev, previous, results);
2461 if (!dirty)
2462 return;
2463
2464 _ilk_disable_lp_wm(dev_priv, dirty);
2465
49a687c4 2466 if (dirty & WM_DIRTY_PIPE(PIPE_A))
801bcfff 2467 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
49a687c4 2468 if (dirty & WM_DIRTY_PIPE(PIPE_B))
801bcfff 2469 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
49a687c4 2470 if (dirty & WM_DIRTY_PIPE(PIPE_C))
801bcfff
PZ
2471 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2472
49a687c4 2473 if (dirty & WM_DIRTY_LINETIME(PIPE_A))
801bcfff 2474 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
49a687c4 2475 if (dirty & WM_DIRTY_LINETIME(PIPE_B))
801bcfff 2476 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
49a687c4 2477 if (dirty & WM_DIRTY_LINETIME(PIPE_C))
801bcfff
PZ
2478 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2479
49a687c4 2480 if (dirty & WM_DIRTY_DDB) {
a42a5719 2481 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
ac9545fd
VS
2482 val = I915_READ(WM_MISC);
2483 if (results->partitioning == INTEL_DDB_PART_1_2)
2484 val &= ~WM_MISC_DATA_PARTITION_5_6;
2485 else
2486 val |= WM_MISC_DATA_PARTITION_5_6;
2487 I915_WRITE(WM_MISC, val);
2488 } else {
2489 val = I915_READ(DISP_ARB_CTL2);
2490 if (results->partitioning == INTEL_DDB_PART_1_2)
2491 val &= ~DISP_DATA_PARTITION_5_6;
2492 else
2493 val |= DISP_DATA_PARTITION_5_6;
2494 I915_WRITE(DISP_ARB_CTL2, val);
2495 }
1011d8c4
PZ
2496 }
2497
49a687c4 2498 if (dirty & WM_DIRTY_FBC) {
cca32e9a
PZ
2499 val = I915_READ(DISP_ARB_CTL);
2500 if (results->enable_fbc_wm)
2501 val &= ~DISP_FBC_WM_DIS;
2502 else
2503 val |= DISP_FBC_WM_DIS;
2504 I915_WRITE(DISP_ARB_CTL, val);
2505 }
2506
954911eb
ID
2507 if (dirty & WM_DIRTY_LP(1) &&
2508 previous->wm_lp_spr[0] != results->wm_lp_spr[0])
2509 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2510
2511 if (INTEL_INFO(dev)->gen >= 7) {
6cef2b8a
VS
2512 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
2513 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2514 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
2515 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2516 }
801bcfff 2517
facd619b 2518 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
801bcfff 2519 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
facd619b 2520 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
801bcfff 2521 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
facd619b 2522 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
801bcfff 2523 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
609cedef
VS
2524
2525 dev_priv->wm.hw = *results;
801bcfff
PZ
2526}
2527
8553c18e
VS
2528static bool ilk_disable_lp_wm(struct drm_device *dev)
2529{
2530 struct drm_i915_private *dev_priv = dev->dev_private;
2531
2532 return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
2533}
2534
820c1980 2535static void ilk_update_wm(struct drm_crtc *crtc)
801bcfff 2536{
7c4a395f 2537 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
46ba614c 2538 struct drm_device *dev = crtc->dev;
801bcfff 2539 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980
ID
2540 struct ilk_wm_maximums max;
2541 struct ilk_pipe_wm_parameters params = {};
2542 struct ilk_wm_values results = {};
77c122bc 2543 enum intel_ddb_partitioning partitioning;
7c4a395f 2544 struct intel_pipe_wm pipe_wm = {};
198a1e9b 2545 struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
a485bfb8 2546 struct intel_wm_config config = {};
7c4a395f 2547
820c1980 2548 ilk_compute_wm_parameters(crtc, &params, &config);
7c4a395f
VS
2549
2550 intel_compute_pipe_wm(crtc, &params, &pipe_wm);
2551
2552 if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
2553 return;
861f3389 2554
7c4a395f 2555 intel_crtc->wm.active = pipe_wm;
861f3389 2556
34982fe1 2557 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
0ba22e26 2558 ilk_wm_merge(dev, &config, &max, &lp_wm_1_2);
a485bfb8
VS
2559
2560 /* 5/6 split only in single pipe config on IVB+ */
ec98c8d1
VS
2561 if (INTEL_INFO(dev)->gen >= 7 &&
2562 config.num_pipes_active == 1 && config.sprites_enabled) {
34982fe1 2563 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
0ba22e26 2564 ilk_wm_merge(dev, &config, &max, &lp_wm_5_6);
0362c781 2565
820c1980 2566 best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
861f3389 2567 } else {
198a1e9b 2568 best_lp_wm = &lp_wm_1_2;
861f3389
PZ
2569 }
2570
198a1e9b 2571 partitioning = (best_lp_wm == &lp_wm_1_2) ?
77c122bc 2572 INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
801bcfff 2573
820c1980 2574 ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results);
609cedef 2575
820c1980 2576 ilk_write_wm_values(dev_priv, &results);
1011d8c4
PZ
2577}
2578
820c1980 2579static void ilk_update_sprite_wm(struct drm_plane *plane,
adf3d35e 2580 struct drm_crtc *crtc,
526682e9 2581 uint32_t sprite_width, int pixel_size,
bdd57d03 2582 bool enabled, bool scaled)
526682e9 2583{
8553c18e 2584 struct drm_device *dev = plane->dev;
adf3d35e 2585 struct intel_plane *intel_plane = to_intel_plane(plane);
526682e9 2586
adf3d35e
VS
2587 intel_plane->wm.enabled = enabled;
2588 intel_plane->wm.scaled = scaled;
2589 intel_plane->wm.horiz_pixels = sprite_width;
2590 intel_plane->wm.bytes_per_pixel = pixel_size;
526682e9 2591
8553c18e
VS
2592 /*
2593 * IVB workaround: must disable low power watermarks for at least
2594 * one frame before enabling scaling. LP watermarks can be re-enabled
2595 * when scaling is disabled.
2596 *
2597 * WaCxSRDisabledForSpriteScaling:ivb
2598 */
2599 if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev))
2600 intel_wait_for_vblank(dev, intel_plane->pipe);
2601
820c1980 2602 ilk_update_wm(crtc);
526682e9
PZ
2603}
2604
243e6a44
VS
2605static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
2606{
2607 struct drm_device *dev = crtc->dev;
2608 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 2609 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS
2610 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2611 struct intel_pipe_wm *active = &intel_crtc->wm.active;
2612 enum pipe pipe = intel_crtc->pipe;
2613 static const unsigned int wm0_pipe_reg[] = {
2614 [PIPE_A] = WM0_PIPEA_ILK,
2615 [PIPE_B] = WM0_PIPEB_ILK,
2616 [PIPE_C] = WM0_PIPEC_IVB,
2617 };
2618
2619 hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
a42a5719 2620 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ce0e0713 2621 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
243e6a44
VS
2622
2623 if (intel_crtc_active(crtc)) {
2624 u32 tmp = hw->wm_pipe[pipe];
2625
2626 /*
2627 * For active pipes LP0 watermark is marked as
2628 * enabled, and LP1+ watermaks as disabled since
2629 * we can't really reverse compute them in case
2630 * multiple pipes are active.
2631 */
2632 active->wm[0].enable = true;
2633 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
2634 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
2635 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
2636 active->linetime = hw->wm_linetime[pipe];
2637 } else {
2638 int level, max_level = ilk_wm_max_level(dev);
2639
2640 /*
2641 * For inactive pipes, all watermark levels
2642 * should be marked as enabled but zeroed,
2643 * which is what we'd compute them to.
2644 */
2645 for (level = 0; level <= max_level; level++)
2646 active->wm[level].enable = true;
2647 }
2648}
2649
2650void ilk_wm_get_hw_state(struct drm_device *dev)
2651{
2652 struct drm_i915_private *dev_priv = dev->dev_private;
820c1980 2653 struct ilk_wm_values *hw = &dev_priv->wm.hw;
243e6a44
VS
2654 struct drm_crtc *crtc;
2655
2656 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
2657 ilk_pipe_wm_get_hw_state(crtc);
2658
2659 hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
2660 hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
2661 hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
2662
2663 hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
2664 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
2665 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
2666
a42a5719 2667 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
ac9545fd
VS
2668 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
2669 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
2670 else if (IS_IVYBRIDGE(dev))
2671 hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
2672 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
243e6a44
VS
2673
2674 hw->enable_fbc_wm =
2675 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
2676}
2677
b445e3b0
ED
2678/**
2679 * intel_update_watermarks - update FIFO watermark values based on current modes
2680 *
2681 * Calculate watermark values for the various WM regs based on current mode
2682 * and plane configuration.
2683 *
2684 * There are several cases to deal with here:
2685 * - normal (i.e. non-self-refresh)
2686 * - self-refresh (SR) mode
2687 * - lines are large relative to FIFO size (buffer can hold up to 2)
2688 * - lines are small relative to FIFO size (buffer can hold more than 2
2689 * lines), so need to account for TLB latency
2690 *
2691 * The normal calculation is:
2692 * watermark = dotclock * bytes per pixel * latency
2693 * where latency is platform & configuration dependent (we assume pessimal
2694 * values here).
2695 *
2696 * The SR calculation is:
2697 * watermark = (trunc(latency/line time)+1) * surface width *
2698 * bytes per pixel
2699 * where
2700 * line time = htotal / dotclock
2701 * surface width = hdisplay for normal plane and 64 for cursor
2702 * and latency is assumed to be high, as above.
2703 *
2704 * The final value programmed to the register should always be rounded up,
2705 * and include an extra 2 entries to account for clock crossings.
2706 *
2707 * We don't use the sprite, so we can ignore that. And on Crestline we have
2708 * to set the non-SR watermarks to 8.
2709 */
46ba614c 2710void intel_update_watermarks(struct drm_crtc *crtc)
b445e3b0 2711{
46ba614c 2712 struct drm_i915_private *dev_priv = crtc->dev->dev_private;
b445e3b0
ED
2713
2714 if (dev_priv->display.update_wm)
46ba614c 2715 dev_priv->display.update_wm(crtc);
b445e3b0
ED
2716}
2717
adf3d35e
VS
2718void intel_update_sprite_watermarks(struct drm_plane *plane,
2719 struct drm_crtc *crtc,
4c4ff43a 2720 uint32_t sprite_width, int pixel_size,
39db4a4d 2721 bool enabled, bool scaled)
b445e3b0 2722{
adf3d35e 2723 struct drm_i915_private *dev_priv = plane->dev->dev_private;
b445e3b0
ED
2724
2725 if (dev_priv->display.update_sprite_wm)
adf3d35e 2726 dev_priv->display.update_sprite_wm(plane, crtc, sprite_width,
39db4a4d 2727 pixel_size, enabled, scaled);
b445e3b0
ED
2728}
2729
2b4e57bd
ED
2730static struct drm_i915_gem_object *
2731intel_alloc_context_page(struct drm_device *dev)
2732{
2733 struct drm_i915_gem_object *ctx;
2734 int ret;
2735
2736 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2737
2738 ctx = i915_gem_alloc_object(dev, 4096);
2739 if (!ctx) {
2740 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2741 return NULL;
2742 }
2743
c37e2204 2744 ret = i915_gem_obj_ggtt_pin(ctx, 4096, true, false);
2b4e57bd
ED
2745 if (ret) {
2746 DRM_ERROR("failed to pin power context: %d\n", ret);
2747 goto err_unref;
2748 }
2749
2750 ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
2751 if (ret) {
2752 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
2753 goto err_unpin;
2754 }
2755
2756 return ctx;
2757
2758err_unpin:
d7f46fc4 2759 i915_gem_object_ggtt_unpin(ctx);
2b4e57bd
ED
2760err_unref:
2761 drm_gem_object_unreference(&ctx->base);
2b4e57bd
ED
2762 return NULL;
2763}
2764
9270388e
DV
2765/**
2766 * Lock protecting IPS related data structures
9270388e
DV
2767 */
2768DEFINE_SPINLOCK(mchdev_lock);
2769
2770/* Global for IPS driver to get at the current i915 device. Protected by
2771 * mchdev_lock. */
2772static struct drm_i915_private *i915_mch_dev;
2773
2b4e57bd
ED
2774bool ironlake_set_drps(struct drm_device *dev, u8 val)
2775{
2776 struct drm_i915_private *dev_priv = dev->dev_private;
2777 u16 rgvswctl;
2778
9270388e
DV
2779 assert_spin_locked(&mchdev_lock);
2780
2b4e57bd
ED
2781 rgvswctl = I915_READ16(MEMSWCTL);
2782 if (rgvswctl & MEMCTL_CMD_STS) {
2783 DRM_DEBUG("gpu busy, RCS change rejected\n");
2784 return false; /* still busy with another command */
2785 }
2786
2787 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
2788 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
2789 I915_WRITE16(MEMSWCTL, rgvswctl);
2790 POSTING_READ16(MEMSWCTL);
2791
2792 rgvswctl |= MEMCTL_CMD_STS;
2793 I915_WRITE16(MEMSWCTL, rgvswctl);
2794
2795 return true;
2796}
2797
8090c6b9 2798static void ironlake_enable_drps(struct drm_device *dev)
2b4e57bd
ED
2799{
2800 struct drm_i915_private *dev_priv = dev->dev_private;
2801 u32 rgvmodectl = I915_READ(MEMMODECTL);
2802 u8 fmax, fmin, fstart, vstart;
2803
9270388e
DV
2804 spin_lock_irq(&mchdev_lock);
2805
2b4e57bd
ED
2806 /* Enable temp reporting */
2807 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
2808 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
2809
2810 /* 100ms RC evaluation intervals */
2811 I915_WRITE(RCUPEI, 100000);
2812 I915_WRITE(RCDNEI, 100000);
2813
2814 /* Set max/min thresholds to 90ms and 80ms respectively */
2815 I915_WRITE(RCBMAXAVG, 90000);
2816 I915_WRITE(RCBMINAVG, 80000);
2817
2818 I915_WRITE(MEMIHYST, 1);
2819
2820 /* Set up min, max, and cur for interrupt handling */
2821 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
2822 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
2823 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
2824 MEMMODE_FSTART_SHIFT;
2825
2826 vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
2827 PXVFREQ_PX_SHIFT;
2828
20e4d407
DV
2829 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
2830 dev_priv->ips.fstart = fstart;
2b4e57bd 2831
20e4d407
DV
2832 dev_priv->ips.max_delay = fstart;
2833 dev_priv->ips.min_delay = fmin;
2834 dev_priv->ips.cur_delay = fstart;
2b4e57bd
ED
2835
2836 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2837 fmax, fmin, fstart);
2838
2839 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
2840
2841 /*
2842 * Interrupts will be enabled in ironlake_irq_postinstall
2843 */
2844
2845 I915_WRITE(VIDSTART, vstart);
2846 POSTING_READ(VIDSTART);
2847
2848 rgvmodectl |= MEMMODE_SWMODE_EN;
2849 I915_WRITE(MEMMODECTL, rgvmodectl);
2850
9270388e 2851 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2b4e57bd 2852 DRM_ERROR("stuck trying to change perf mode\n");
9270388e 2853 mdelay(1);
2b4e57bd
ED
2854
2855 ironlake_set_drps(dev, fstart);
2856
20e4d407 2857 dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
2b4e57bd 2858 I915_READ(0x112e0);
20e4d407
DV
2859 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
2860 dev_priv->ips.last_count2 = I915_READ(0x112f4);
2861 getrawmonotonic(&dev_priv->ips.last_time2);
9270388e
DV
2862
2863 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED
2864}
2865
8090c6b9 2866static void ironlake_disable_drps(struct drm_device *dev)
2b4e57bd
ED
2867{
2868 struct drm_i915_private *dev_priv = dev->dev_private;
9270388e
DV
2869 u16 rgvswctl;
2870
2871 spin_lock_irq(&mchdev_lock);
2872
2873 rgvswctl = I915_READ16(MEMSWCTL);
2b4e57bd
ED
2874
2875 /* Ack interrupts, disable EFC interrupt */
2876 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
2877 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
2878 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
2879 I915_WRITE(DEIIR, DE_PCU_EVENT);
2880 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
2881
2882 /* Go back to the starting frequency */
20e4d407 2883 ironlake_set_drps(dev, dev_priv->ips.fstart);
9270388e 2884 mdelay(1);
2b4e57bd
ED
2885 rgvswctl |= MEMCTL_CMD_STS;
2886 I915_WRITE(MEMSWCTL, rgvswctl);
9270388e 2887 mdelay(1);
2b4e57bd 2888
9270388e 2889 spin_unlock_irq(&mchdev_lock);
2b4e57bd
ED
2890}
2891
acbe9475
DV
2892/* There's a funny hw issue where the hw returns all 0 when reading from
2893 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
2894 * ourselves, instead of doing a rmw cycle (which might result in us clearing
2895 * all limits and the gpu stuck at whatever frequency it is at atm).
2896 */
6917c7b9 2897static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 val)
2b4e57bd 2898{
7b9e0ae6 2899 u32 limits;
2b4e57bd 2900
20b46e59
DV
2901 /* Only set the down limit when we've reached the lowest level to avoid
2902 * getting more interrupts, otherwise leave this clear. This prevents a
2903 * race in the hw when coming out of rc6: There's a tiny window where
2904 * the hw runs at the minimal clock before selecting the desired
2905 * frequency, if the down threshold expires in that window we will not
2906 * receive a down interrupt. */
6917c7b9
CW
2907 limits = dev_priv->rps.max_delay << 24;
2908 if (val <= dev_priv->rps.min_delay)
c6a828d3 2909 limits |= dev_priv->rps.min_delay << 16;
20b46e59
DV
2910
2911 return limits;
2912}
2913
dd75fdc8
CW
2914static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
2915{
2916 int new_power;
2917
2918 new_power = dev_priv->rps.power;
2919 switch (dev_priv->rps.power) {
2920 case LOW_POWER:
2921 if (val > dev_priv->rps.rpe_delay + 1 && val > dev_priv->rps.cur_delay)
2922 new_power = BETWEEN;
2923 break;
2924
2925 case BETWEEN:
2926 if (val <= dev_priv->rps.rpe_delay && val < dev_priv->rps.cur_delay)
2927 new_power = LOW_POWER;
2928 else if (val >= dev_priv->rps.rp0_delay && val > dev_priv->rps.cur_delay)
2929 new_power = HIGH_POWER;
2930 break;
2931
2932 case HIGH_POWER:
2933 if (val < (dev_priv->rps.rp1_delay + dev_priv->rps.rp0_delay) >> 1 && val < dev_priv->rps.cur_delay)
2934 new_power = BETWEEN;
2935 break;
2936 }
2937 /* Max/min bins are special */
2938 if (val == dev_priv->rps.min_delay)
2939 new_power = LOW_POWER;
2940 if (val == dev_priv->rps.max_delay)
2941 new_power = HIGH_POWER;
2942 if (new_power == dev_priv->rps.power)
2943 return;
2944
2945 /* Note the units here are not exactly 1us, but 1280ns. */
2946 switch (new_power) {
2947 case LOW_POWER:
2948 /* Upclock if more than 95% busy over 16ms */
2949 I915_WRITE(GEN6_RP_UP_EI, 12500);
2950 I915_WRITE(GEN6_RP_UP_THRESHOLD, 11800);
2951
2952 /* Downclock if less than 85% busy over 32ms */
2953 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
2954 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 21250);
2955
2956 I915_WRITE(GEN6_RP_CONTROL,
2957 GEN6_RP_MEDIA_TURBO |
2958 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2959 GEN6_RP_MEDIA_IS_GFX |
2960 GEN6_RP_ENABLE |
2961 GEN6_RP_UP_BUSY_AVG |
2962 GEN6_RP_DOWN_IDLE_AVG);
2963 break;
2964
2965 case BETWEEN:
2966 /* Upclock if more than 90% busy over 13ms */
2967 I915_WRITE(GEN6_RP_UP_EI, 10250);
2968 I915_WRITE(GEN6_RP_UP_THRESHOLD, 9225);
2969
2970 /* Downclock if less than 75% busy over 32ms */
2971 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
2972 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 18750);
2973
2974 I915_WRITE(GEN6_RP_CONTROL,
2975 GEN6_RP_MEDIA_TURBO |
2976 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2977 GEN6_RP_MEDIA_IS_GFX |
2978 GEN6_RP_ENABLE |
2979 GEN6_RP_UP_BUSY_AVG |
2980 GEN6_RP_DOWN_IDLE_AVG);
2981 break;
2982
2983 case HIGH_POWER:
2984 /* Upclock if more than 85% busy over 10ms */
2985 I915_WRITE(GEN6_RP_UP_EI, 8000);
2986 I915_WRITE(GEN6_RP_UP_THRESHOLD, 6800);
2987
2988 /* Downclock if less than 60% busy over 32ms */
2989 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
2990 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 15000);
2991
2992 I915_WRITE(GEN6_RP_CONTROL,
2993 GEN6_RP_MEDIA_TURBO |
2994 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2995 GEN6_RP_MEDIA_IS_GFX |
2996 GEN6_RP_ENABLE |
2997 GEN6_RP_UP_BUSY_AVG |
2998 GEN6_RP_DOWN_IDLE_AVG);
2999 break;
3000 }
3001
3002 dev_priv->rps.power = new_power;
3003 dev_priv->rps.last_adj = 0;
3004}
3005
b8a5ff8d
JM
3006/* gen6_set_rps is called to update the frequency request, but should also be
3007 * called when the range (min_delay and max_delay) is modified so that we can
3008 * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
20b46e59
DV
3009void gen6_set_rps(struct drm_device *dev, u8 val)
3010{
3011 struct drm_i915_private *dev_priv = dev->dev_private;
7b9e0ae6 3012
4fc688ce 3013 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79249636
BW
3014 WARN_ON(val > dev_priv->rps.max_delay);
3015 WARN_ON(val < dev_priv->rps.min_delay);
004777cb 3016
b8a5ff8d
JM
3017 if (val == dev_priv->rps.cur_delay) {
3018 /* min/max delay may still have been modified so be sure to
3019 * write the limits value */
3020 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
3021 gen6_rps_limits(dev_priv, val));
3022
7b9e0ae6 3023 return;
b8a5ff8d 3024 }
7b9e0ae6 3025
dd75fdc8
CW
3026 gen6_set_rps_thresholds(dev_priv, val);
3027
92bd1bf0
RV
3028 if (IS_HASWELL(dev))
3029 I915_WRITE(GEN6_RPNSWREQ,
3030 HSW_FREQUENCY(val));
3031 else
3032 I915_WRITE(GEN6_RPNSWREQ,
3033 GEN6_FREQUENCY(val) |
3034 GEN6_OFFSET(0) |
3035 GEN6_AGGRESSIVE_TURBO);
7b9e0ae6
CW
3036
3037 /* Make sure we continue to get interrupts
3038 * until we hit the minimum or maximum frequencies.
3039 */
6917c7b9
CW
3040 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
3041 gen6_rps_limits(dev_priv, val));
7b9e0ae6 3042
d5570a72
BW
3043 POSTING_READ(GEN6_RPNSWREQ);
3044
c6a828d3 3045 dev_priv->rps.cur_delay = val;
be2cde9a
DV
3046
3047 trace_intel_gpu_freq_change(val * 50);
2b4e57bd
ED
3048}
3049
76c3552f
D
3050/* vlv_set_rps_idle: Set the frequency to Rpn if Gfx clocks are down
3051 *
3052 * * If Gfx is Idle, then
3053 * 1. Mask Turbo interrupts
3054 * 2. Bring up Gfx clock
3055 * 3. Change the freq to Rpn and wait till P-Unit updates freq
3056 * 4. Clear the Force GFX CLK ON bit so that Gfx can down
3057 * 5. Unmask Turbo interrupts
3058*/
3059static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
3060{
3061 /*
3062 * When we are idle. Drop to min voltage state.
3063 */
3064
3065 if (dev_priv->rps.cur_delay <= dev_priv->rps.min_delay)
3066 return;
3067
3068 /* Mask turbo interrupt so that they will not come in between */
3069 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
3070
3071 /* Bring up the Gfx clock */
3072 I915_WRITE(VLV_GTLC_SURVIVABILITY_REG,
3073 I915_READ(VLV_GTLC_SURVIVABILITY_REG) |
3074 VLV_GFX_CLK_FORCE_ON_BIT);
3075
3076 if (wait_for(((VLV_GFX_CLK_STATUS_BIT &
3077 I915_READ(VLV_GTLC_SURVIVABILITY_REG)) != 0), 5)) {
3078 DRM_ERROR("GFX_CLK_ON request timed out\n");
3079 return;
3080 }
3081
3082 dev_priv->rps.cur_delay = dev_priv->rps.min_delay;
3083
3084 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ,
3085 dev_priv->rps.min_delay);
3086
3087 if (wait_for(((vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS))
3088 & GENFREQSTATUS) == 0, 5))
3089 DRM_ERROR("timed out waiting for Punit\n");
3090
3091 /* Release the Gfx clock */
3092 I915_WRITE(VLV_GTLC_SURVIVABILITY_REG,
3093 I915_READ(VLV_GTLC_SURVIVABILITY_REG) &
3094 ~VLV_GFX_CLK_FORCE_ON_BIT);
3095
3096 /* Unmask Up interrupts */
3097 dev_priv->rps.rp_up_masked = true;
3098 gen6_set_pm_mask(dev_priv, GEN6_PM_RP_DOWN_THRESHOLD,
3099 dev_priv->rps.min_delay);
3100}
3101
b29c19b6
CW
3102void gen6_rps_idle(struct drm_i915_private *dev_priv)
3103{
691bb717
DL
3104 struct drm_device *dev = dev_priv->dev;
3105
b29c19b6 3106 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 3107 if (dev_priv->rps.enabled) {
691bb717 3108 if (IS_VALLEYVIEW(dev))
76c3552f 3109 vlv_set_rps_idle(dev_priv);
c0951f0c
CW
3110 else
3111 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3112 dev_priv->rps.last_adj = 0;
3113 }
b29c19b6
CW
3114 mutex_unlock(&dev_priv->rps.hw_lock);
3115}
3116
3117void gen6_rps_boost(struct drm_i915_private *dev_priv)
3118{
691bb717
DL
3119 struct drm_device *dev = dev_priv->dev;
3120
b29c19b6 3121 mutex_lock(&dev_priv->rps.hw_lock);
c0951f0c 3122 if (dev_priv->rps.enabled) {
691bb717 3123 if (IS_VALLEYVIEW(dev))
c0951f0c
CW
3124 valleyview_set_rps(dev_priv->dev, dev_priv->rps.max_delay);
3125 else
3126 gen6_set_rps(dev_priv->dev, dev_priv->rps.max_delay);
3127 dev_priv->rps.last_adj = 0;
3128 }
b29c19b6
CW
3129 mutex_unlock(&dev_priv->rps.hw_lock);
3130}
3131
0a073b84
JB
3132void valleyview_set_rps(struct drm_device *dev, u8 val)
3133{
3134 struct drm_i915_private *dev_priv = dev->dev_private;
7a67092a 3135
0a073b84
JB
3136 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3137 WARN_ON(val > dev_priv->rps.max_delay);
3138 WARN_ON(val < dev_priv->rps.min_delay);
3139
73008b98 3140 DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
2ec3815f 3141 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_delay),
73008b98 3142 dev_priv->rps.cur_delay,
2ec3815f 3143 vlv_gpu_freq(dev_priv, val), val);
0a073b84
JB
3144
3145 if (val == dev_priv->rps.cur_delay)
3146 return;
3147
ae99258f 3148 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
0a073b84 3149
80814ae4 3150 dev_priv->rps.cur_delay = val;
0a073b84 3151
2ec3815f 3152 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv, val));
0a073b84
JB
3153}
3154
44fc7d5c 3155static void gen6_disable_rps_interrupts(struct drm_device *dev)
2b4e57bd
ED
3156{
3157 struct drm_i915_private *dev_priv = dev->dev_private;
3158
2b4e57bd 3159 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
4848405c 3160 I915_WRITE(GEN6_PMIER, I915_READ(GEN6_PMIER) & ~GEN6_PM_RPS_EVENTS);
2b4e57bd
ED
3161 /* Complete PM interrupt masking here doesn't race with the rps work
3162 * item again unmasking PM interrupts because that is using a different
3163 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
3164 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
3165
59cdb63d 3166 spin_lock_irq(&dev_priv->irq_lock);
c6a828d3 3167 dev_priv->rps.pm_iir = 0;
59cdb63d 3168 spin_unlock_irq(&dev_priv->irq_lock);
2b4e57bd 3169
4848405c 3170 I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS);
2b4e57bd
ED
3171}
3172
44fc7d5c 3173static void gen6_disable_rps(struct drm_device *dev)
d20d4f0c
JB
3174{
3175 struct drm_i915_private *dev_priv = dev->dev_private;
3176
3177 I915_WRITE(GEN6_RC_CONTROL, 0);
44fc7d5c 3178 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
d20d4f0c 3179
44fc7d5c
DV
3180 gen6_disable_rps_interrupts(dev);
3181}
3182
3183static void valleyview_disable_rps(struct drm_device *dev)
3184{
3185 struct drm_i915_private *dev_priv = dev->dev_private;
3186
3187 I915_WRITE(GEN6_RC_CONTROL, 0);
d20d4f0c 3188
44fc7d5c 3189 gen6_disable_rps_interrupts(dev);
c9cddffc
JB
3190
3191 if (dev_priv->vlv_pctx) {
3192 drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
3193 dev_priv->vlv_pctx = NULL;
3194 }
d20d4f0c
JB
3195}
3196
dc39fff7
BW
3197static void intel_print_rc6_info(struct drm_device *dev, u32 mode)
3198{
3199 if (IS_GEN6(dev))
3200 DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
3201
3202 if (IS_HASWELL(dev))
3203 DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
3204
3205 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
3206 (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
3207 (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
3208 (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
3209}
3210
2b4e57bd
ED
3211int intel_enable_rc6(const struct drm_device *dev)
3212{
eb4926e4
DL
3213 /* No RC6 before Ironlake */
3214 if (INTEL_INFO(dev)->gen < 5)
3215 return 0;
3216
456470eb 3217 /* Respect the kernel parameter if it is set */
d330a953
JN
3218 if (i915.enable_rc6 >= 0)
3219 return i915.enable_rc6;
2b4e57bd 3220
6567d748
CW
3221 /* Disable RC6 on Ironlake */
3222 if (INTEL_INFO(dev)->gen == 5)
3223 return 0;
2b4e57bd 3224
dc39fff7 3225 if (IS_HASWELL(dev))
4a637c2c 3226 return INTEL_RC6_ENABLE;
2b4e57bd 3227
456470eb 3228 /* snb/ivb have more than one rc6 state. */
dc39fff7 3229 if (INTEL_INFO(dev)->gen == 6)
2b4e57bd 3230 return INTEL_RC6_ENABLE;
456470eb 3231
2b4e57bd
ED
3232 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
3233}
3234
44fc7d5c
DV
3235static void gen6_enable_rps_interrupts(struct drm_device *dev)
3236{
3237 struct drm_i915_private *dev_priv = dev->dev_private;
a9c1f90c 3238 u32 enabled_intrs;
44fc7d5c
DV
3239
3240 spin_lock_irq(&dev_priv->irq_lock);
a0b3335a 3241 WARN_ON(dev_priv->rps.pm_iir);
edbfdb45 3242 snb_enable_pm_irq(dev_priv, GEN6_PM_RPS_EVENTS);
44fc7d5c
DV
3243 I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS);
3244 spin_unlock_irq(&dev_priv->irq_lock);
a9c1f90c 3245
fd547d25 3246 /* only unmask PM interrupts we need. Mask all others. */
a9c1f90c
MK
3247 enabled_intrs = GEN6_PM_RPS_EVENTS;
3248
3249 /* IVB and SNB hard hangs on looping batchbuffer
3250 * if GEN6_PM_UP_EI_EXPIRED is masked.
3251 */
3252 if (INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev))
3253 enabled_intrs |= GEN6_PM_RP_UP_EI_EXPIRED;
3254
3255 I915_WRITE(GEN6_PMINTRMSK, ~enabled_intrs);
44fc7d5c
DV
3256}
3257
6edee7f3
BW
3258static void gen8_enable_rps(struct drm_device *dev)
3259{
3260 struct drm_i915_private *dev_priv = dev->dev_private;
3261 struct intel_ring_buffer *ring;
3262 uint32_t rc6_mask = 0, rp_state_cap;
3263 int unused;
3264
3265 /* 1a: Software RC state - RC0 */
3266 I915_WRITE(GEN6_RC_STATE, 0);
3267
3268 /* 1c & 1d: Get forcewake during program sequence. Although the driver
3269 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
c8d9a590 3270 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW
3271
3272 /* 2a: Disable RC states. */
3273 I915_WRITE(GEN6_RC_CONTROL, 0);
3274
3275 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3276
3277 /* 2b: Program RC6 thresholds.*/
3278 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
3279 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
3280 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
3281 for_each_ring(ring, dev_priv, unused)
3282 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3283 I915_WRITE(GEN6_RC_SLEEP, 0);
3284 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
3285
3286 /* 3: Enable RC6 */
3287 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
3288 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
3289 DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off");
3290 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
3291 GEN6_RC_CTL_EI_MODE(1) |
3292 rc6_mask);
3293
3294 /* 4 Program defaults and thresholds for RPS*/
3295 I915_WRITE(GEN6_RPNSWREQ, HSW_FREQUENCY(10)); /* Request 500 MHz */
3296 I915_WRITE(GEN6_RC_VIDEO_FREQ, HSW_FREQUENCY(12)); /* Request 600 MHz */
3297 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
3298 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
3299
3300 /* Docs recommend 900MHz, and 300 MHz respectively */
3301 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
3302 dev_priv->rps.max_delay << 24 |
3303 dev_priv->rps.min_delay << 16);
3304
3305 I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
3306 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
3307 I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
3308 I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
3309
3310 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3311
3312 /* 5: Enable RPS */
3313 I915_WRITE(GEN6_RP_CONTROL,
3314 GEN6_RP_MEDIA_TURBO |
3315 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3316 GEN6_RP_MEDIA_IS_GFX |
3317 GEN6_RP_ENABLE |
3318 GEN6_RP_UP_BUSY_AVG |
3319 GEN6_RP_DOWN_IDLE_AVG);
3320
3321 /* 6: Ring frequency + overclocking (our driver does this later */
3322
3323 gen6_set_rps(dev, (I915_READ(GEN6_GT_PERF_STATUS) & 0xff00) >> 8);
3324
3325 gen6_enable_rps_interrupts(dev);
3326
c8d9a590 3327 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
6edee7f3
BW
3328}
3329
79f5b2c7 3330static void gen6_enable_rps(struct drm_device *dev)
2b4e57bd 3331{
79f5b2c7 3332 struct drm_i915_private *dev_priv = dev->dev_private;
b4519513 3333 struct intel_ring_buffer *ring;
dd0a1aa1 3334 u32 rp_state_cap, hw_max, hw_min;
7b9e0ae6 3335 u32 gt_perf_status;
31643d54 3336 u32 rc6vids, pcu_mbox, rc6_mask = 0;
2b4e57bd 3337 u32 gtfifodbg;
2b4e57bd 3338 int rc6_mode;
42c0526c 3339 int i, ret;
2b4e57bd 3340
4fc688ce 3341 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 3342
2b4e57bd
ED
3343 /* Here begins a magic sequence of register writes to enable
3344 * auto-downclocking.
3345 *
3346 * Perhaps there might be some value in exposing these to
3347 * userspace...
3348 */
3349 I915_WRITE(GEN6_RC_STATE, 0);
2b4e57bd
ED
3350
3351 /* Clear the DBG now so we don't confuse earlier errors */
3352 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
3353 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
3354 I915_WRITE(GTFIFODBG, gtfifodbg);
3355 }
3356
c8d9a590 3357 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
2b4e57bd 3358
7b9e0ae6
CW
3359 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3360 gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
3361
31c77388 3362 /* In units of 50MHz */
dd0a1aa1
JM
3363 dev_priv->rps.hw_max = hw_max = rp_state_cap & 0xff;
3364 hw_min = (rp_state_cap >> 16) & 0xff;
dd75fdc8
CW
3365 dev_priv->rps.rp1_delay = (rp_state_cap >> 8) & 0xff;
3366 dev_priv->rps.rp0_delay = (rp_state_cap >> 0) & 0xff;
3367 dev_priv->rps.rpe_delay = dev_priv->rps.rp1_delay;
c6a828d3 3368 dev_priv->rps.cur_delay = 0;
7b9e0ae6 3369
dd0a1aa1
JM
3370 /* Preserve min/max settings in case of re-init */
3371 if (dev_priv->rps.max_delay == 0)
3372 dev_priv->rps.max_delay = hw_max;
3373
3374 if (dev_priv->rps.min_delay == 0)
3375 dev_priv->rps.min_delay = hw_min;
3376
2b4e57bd
ED
3377 /* disable the counters and set deterministic thresholds */
3378 I915_WRITE(GEN6_RC_CONTROL, 0);
3379
3380 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
3381 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
3382 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
3383 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3384 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3385
b4519513
CW
3386 for_each_ring(ring, dev_priv, i)
3387 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2b4e57bd
ED
3388
3389 I915_WRITE(GEN6_RC_SLEEP, 0);
3390 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
29c78f60 3391 if (IS_IVYBRIDGE(dev))
351aa566
SM
3392 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
3393 else
3394 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
0920a487 3395 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2b4e57bd
ED
3396 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
3397
5a7dc92a 3398 /* Check if we are enabling RC6 */
2b4e57bd
ED
3399 rc6_mode = intel_enable_rc6(dev_priv->dev);
3400 if (rc6_mode & INTEL_RC6_ENABLE)
3401 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
3402
5a7dc92a
ED
3403 /* We don't use those on Haswell */
3404 if (!IS_HASWELL(dev)) {
3405 if (rc6_mode & INTEL_RC6p_ENABLE)
3406 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2b4e57bd 3407
5a7dc92a
ED
3408 if (rc6_mode & INTEL_RC6pp_ENABLE)
3409 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
3410 }
2b4e57bd 3411
dc39fff7 3412 intel_print_rc6_info(dev, rc6_mask);
2b4e57bd
ED
3413
3414 I915_WRITE(GEN6_RC_CONTROL,
3415 rc6_mask |
3416 GEN6_RC_CTL_EI_MODE(1) |
3417 GEN6_RC_CTL_HW_ENABLE);
3418
dd75fdc8
CW
3419 /* Power down if completely idle for over 50ms */
3420 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
2b4e57bd 3421 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2b4e57bd 3422
42c0526c 3423 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
988b36e5 3424 if (!ret) {
42c0526c
BW
3425 pcu_mbox = 0;
3426 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
a2b3fc01 3427 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
10e08497 3428 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
a2b3fc01
BW
3429 (dev_priv->rps.max_delay & 0xff) * 50,
3430 (pcu_mbox & 0xff) * 50);
31c77388 3431 dev_priv->rps.hw_max = pcu_mbox & 0xff;
42c0526c
BW
3432 }
3433 } else {
3434 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
2b4e57bd
ED
3435 }
3436
dd75fdc8
CW
3437 dev_priv->rps.power = HIGH_POWER; /* force a reset */
3438 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
2b4e57bd 3439
44fc7d5c 3440 gen6_enable_rps_interrupts(dev);
2b4e57bd 3441
31643d54
BW
3442 rc6vids = 0;
3443 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
3444 if (IS_GEN6(dev) && ret) {
3445 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
3446 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
3447 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
3448 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
3449 rc6vids &= 0xffff00;
3450 rc6vids |= GEN6_ENCODE_RC6_VID(450);
3451 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
3452 if (ret)
3453 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
3454 }
3455
c8d9a590 3456 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
2b4e57bd
ED
3457}
3458
c67a470b 3459void gen6_update_ring_freq(struct drm_device *dev)
2b4e57bd 3460{
79f5b2c7 3461 struct drm_i915_private *dev_priv = dev->dev_private;
2b4e57bd 3462 int min_freq = 15;
3ebecd07
CW
3463 unsigned int gpu_freq;
3464 unsigned int max_ia_freq, min_ring_freq;
2b4e57bd 3465 int scaling_factor = 180;
eda79642 3466 struct cpufreq_policy *policy;
2b4e57bd 3467
4fc688ce 3468 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
79f5b2c7 3469
eda79642
BW
3470 policy = cpufreq_cpu_get(0);
3471 if (policy) {
3472 max_ia_freq = policy->cpuinfo.max_freq;
3473 cpufreq_cpu_put(policy);
3474 } else {
3475 /*
3476 * Default to measured freq if none found, PCU will ensure we
3477 * don't go over
3478 */
2b4e57bd 3479 max_ia_freq = tsc_khz;
eda79642 3480 }
2b4e57bd
ED
3481
3482 /* Convert from kHz to MHz */
3483 max_ia_freq /= 1000;
3484
153b4b95 3485 min_ring_freq = I915_READ(DCLK) & 0xf;
f6aca45c
BW
3486 /* convert DDR frequency from units of 266.6MHz to bandwidth */
3487 min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3ebecd07 3488
2b4e57bd
ED
3489 /*
3490 * For each potential GPU frequency, load a ring frequency we'd like
3491 * to use for memory access. We do this by specifying the IA frequency
3492 * the PCU should use as a reference to determine the ring frequency.
3493 */
c6a828d3 3494 for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay;
2b4e57bd 3495 gpu_freq--) {
c6a828d3 3496 int diff = dev_priv->rps.max_delay - gpu_freq;
3ebecd07
CW
3497 unsigned int ia_freq = 0, ring_freq = 0;
3498
46c764d4
BW
3499 if (INTEL_INFO(dev)->gen >= 8) {
3500 /* max(2 * GT, DDR). NB: GT is 50MHz units */
3501 ring_freq = max(min_ring_freq, gpu_freq);
3502 } else if (IS_HASWELL(dev)) {
f6aca45c 3503 ring_freq = mult_frac(gpu_freq, 5, 4);
3ebecd07
CW
3504 ring_freq = max(min_ring_freq, ring_freq);
3505 /* leave ia_freq as the default, chosen by cpufreq */
3506 } else {
3507 /* On older processors, there is no separate ring
3508 * clock domain, so in order to boost the bandwidth
3509 * of the ring, we need to upclock the CPU (ia_freq).
3510 *
3511 * For GPU frequencies less than 750MHz,
3512 * just use the lowest ring freq.
3513 */
3514 if (gpu_freq < min_freq)
3515 ia_freq = 800;
3516 else
3517 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
3518 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
3519 }
2b4e57bd 3520
42c0526c
BW
3521 sandybridge_pcode_write(dev_priv,
3522 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3ebecd07
CW
3523 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
3524 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
3525 gpu_freq);
2b4e57bd 3526 }
2b4e57bd
ED
3527}
3528
0a073b84
JB
3529int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
3530{
3531 u32 val, rp0;
3532
64936258 3533 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
0a073b84
JB
3534
3535 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
3536 /* Clamp to max */
3537 rp0 = min_t(u32, rp0, 0xea);
3538
3539 return rp0;
3540}
3541
3542static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
3543{
3544 u32 val, rpe;
3545
64936258 3546 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
0a073b84 3547 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
64936258 3548 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
0a073b84
JB
3549 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
3550
3551 return rpe;
3552}
3553
3554int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
3555{
64936258 3556 return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
0a073b84
JB
3557}
3558
c9cddffc
JB
3559static void valleyview_setup_pctx(struct drm_device *dev)
3560{
3561 struct drm_i915_private *dev_priv = dev->dev_private;
3562 struct drm_i915_gem_object *pctx;
3563 unsigned long pctx_paddr;
3564 u32 pcbr;
3565 int pctx_size = 24*1024;
3566
3567 pcbr = I915_READ(VLV_PCBR);
3568 if (pcbr) {
3569 /* BIOS set it up already, grab the pre-alloc'd space */
3570 int pcbr_offset;
3571
3572 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
3573 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
3574 pcbr_offset,
190d6cd5 3575 I915_GTT_OFFSET_NONE,
c9cddffc
JB
3576 pctx_size);
3577 goto out;
3578 }
3579
3580 /*
3581 * From the Gunit register HAS:
3582 * The Gfx driver is expected to program this register and ensure
3583 * proper allocation within Gfx stolen memory. For example, this
3584 * register should be programmed such than the PCBR range does not
3585 * overlap with other ranges, such as the frame buffer, protected
3586 * memory, or any other relevant ranges.
3587 */
3588 pctx = i915_gem_object_create_stolen(dev, pctx_size);
3589 if (!pctx) {
3590 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
3591 return;
3592 }
3593
3594 pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
3595 I915_WRITE(VLV_PCBR, pctx_paddr);
3596
3597out:
3598 dev_priv->vlv_pctx = pctx;
3599}
3600
0a073b84
JB
3601static void valleyview_enable_rps(struct drm_device *dev)
3602{
3603 struct drm_i915_private *dev_priv = dev->dev_private;
3604 struct intel_ring_buffer *ring;
dd0a1aa1 3605 u32 gtfifodbg, val, hw_max, hw_min, rc6_mode = 0;
0a073b84
JB
3606 int i;
3607
3608 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3609
3610 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
f7d85c1e
JB
3611 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
3612 gtfifodbg);
0a073b84
JB
3613 I915_WRITE(GTFIFODBG, gtfifodbg);
3614 }
3615
c9cddffc
JB
3616 valleyview_setup_pctx(dev);
3617
c8d9a590
D
3618 /* If VLV, Forcewake all wells, else re-direct to regular path */
3619 gen6_gt_force_wake_get(dev_priv, FORCEWAKE_ALL);
0a073b84
JB
3620
3621 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
3622 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
3623 I915_WRITE(GEN6_RP_UP_EI, 66000);
3624 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
3625
3626 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3627
3628 I915_WRITE(GEN6_RP_CONTROL,
3629 GEN6_RP_MEDIA_TURBO |
3630 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3631 GEN6_RP_MEDIA_IS_GFX |
3632 GEN6_RP_ENABLE |
3633 GEN6_RP_UP_BUSY_AVG |
3634 GEN6_RP_DOWN_IDLE_CONT);
3635
3636 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
3637 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3638 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3639
3640 for_each_ring(ring, dev_priv, i)
3641 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3642
2f0aa304 3643 I915_WRITE(GEN6_RC6_THRESHOLD, 0x557);
0a073b84
JB
3644
3645 /* allows RC6 residency counter to work */
49798eb2
JB
3646 I915_WRITE(VLV_COUNTER_CONTROL,
3647 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
3648 VLV_MEDIA_RC6_COUNT_EN |
3649 VLV_RENDER_RC6_COUNT_EN));
a2b23fe0 3650 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
6b88f295 3651 rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
dc39fff7
BW
3652
3653 intel_print_rc6_info(dev, rc6_mode);
3654
a2b23fe0 3655 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
0a073b84 3656
64936258 3657 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
0a073b84
JB
3658
3659 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
3660 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
3661
0a073b84 3662 dev_priv->rps.cur_delay = (val >> 8) & 0xff;
73008b98 3663 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
2ec3815f 3664 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_delay),
73008b98 3665 dev_priv->rps.cur_delay);
0a073b84 3666
dd0a1aa1 3667 dev_priv->rps.hw_max = hw_max = valleyview_rps_max_freq(dev_priv);
73008b98 3668 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
dd0a1aa1
JM
3669 vlv_gpu_freq(dev_priv, hw_max),
3670 hw_max);
0a073b84 3671
73008b98
VS
3672 dev_priv->rps.rpe_delay = valleyview_rps_rpe_freq(dev_priv);
3673 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
2ec3815f 3674 vlv_gpu_freq(dev_priv, dev_priv->rps.rpe_delay),
73008b98 3675 dev_priv->rps.rpe_delay);
0a073b84 3676
dd0a1aa1 3677 hw_min = valleyview_rps_min_freq(dev_priv);
73008b98 3678 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
dd0a1aa1
JM
3679 vlv_gpu_freq(dev_priv, hw_min),
3680 hw_min);
3681
3682 /* Preserve min/max settings in case of re-init */
3683 if (dev_priv->rps.max_delay == 0)
3684 dev_priv->rps.max_delay = hw_max;
3685
3686 if (dev_priv->rps.min_delay == 0)
3687 dev_priv->rps.min_delay = hw_min;
0a073b84 3688
73008b98 3689 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
2ec3815f 3690 vlv_gpu_freq(dev_priv, dev_priv->rps.rpe_delay),
73008b98 3691 dev_priv->rps.rpe_delay);
0a073b84 3692
73008b98 3693 valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay);
0a073b84 3694
27544369
D
3695 dev_priv->rps.rp_up_masked = false;
3696 dev_priv->rps.rp_down_masked = false;
3697
44fc7d5c 3698 gen6_enable_rps_interrupts(dev);
0a073b84 3699
c8d9a590 3700 gen6_gt_force_wake_put(dev_priv, FORCEWAKE_ALL);
0a073b84
JB
3701}
3702
930ebb46 3703void ironlake_teardown_rc6(struct drm_device *dev)
2b4e57bd
ED
3704{
3705 struct drm_i915_private *dev_priv = dev->dev_private;
3706
3e373948 3707 if (dev_priv->ips.renderctx) {
d7f46fc4 3708 i915_gem_object_ggtt_unpin(dev_priv->ips.renderctx);
3e373948
DV
3709 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
3710 dev_priv->ips.renderctx = NULL;
2b4e57bd
ED
3711 }
3712
3e373948 3713 if (dev_priv->ips.pwrctx) {
d7f46fc4 3714 i915_gem_object_ggtt_unpin(dev_priv->ips.pwrctx);
3e373948
DV
3715 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
3716 dev_priv->ips.pwrctx = NULL;
2b4e57bd
ED
3717 }
3718}
3719
930ebb46 3720static void ironlake_disable_rc6(struct drm_device *dev)
2b4e57bd
ED
3721{
3722 struct drm_i915_private *dev_priv = dev->dev_private;
3723
3724 if (I915_READ(PWRCTXA)) {
3725 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
3726 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
3727 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
3728 50);
3729
3730 I915_WRITE(PWRCTXA, 0);
3731 POSTING_READ(PWRCTXA);
3732
3733 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
3734 POSTING_READ(RSTDBYCTL);
3735 }
2b4e57bd
ED
3736}
3737
3738static int ironlake_setup_rc6(struct drm_device *dev)
3739{
3740 struct drm_i915_private *dev_priv = dev->dev_private;
3741
3e373948
DV
3742 if (dev_priv->ips.renderctx == NULL)
3743 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
3744 if (!dev_priv->ips.renderctx)
2b4e57bd
ED
3745 return -ENOMEM;
3746
3e373948
DV
3747 if (dev_priv->ips.pwrctx == NULL)
3748 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
3749 if (!dev_priv->ips.pwrctx) {
2b4e57bd
ED
3750 ironlake_teardown_rc6(dev);
3751 return -ENOMEM;
3752 }
3753
3754 return 0;
3755}
3756
930ebb46 3757static void ironlake_enable_rc6(struct drm_device *dev)
2b4e57bd
ED
3758{
3759 struct drm_i915_private *dev_priv = dev->dev_private;
6d90c952 3760 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
3e960501 3761 bool was_interruptible;
2b4e57bd
ED
3762 int ret;
3763
3764 /* rc6 disabled by default due to repeated reports of hanging during
3765 * boot and resume.
3766 */
3767 if (!intel_enable_rc6(dev))
3768 return;
3769
79f5b2c7
DV
3770 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3771
2b4e57bd 3772 ret = ironlake_setup_rc6(dev);
79f5b2c7 3773 if (ret)
2b4e57bd 3774 return;
2b4e57bd 3775
3e960501
CW
3776 was_interruptible = dev_priv->mm.interruptible;
3777 dev_priv->mm.interruptible = false;
3778
2b4e57bd
ED
3779 /*
3780 * GPU can automatically power down the render unit if given a page
3781 * to save state.
3782 */
6d90c952 3783 ret = intel_ring_begin(ring, 6);
2b4e57bd
ED
3784 if (ret) {
3785 ironlake_teardown_rc6(dev);
3e960501 3786 dev_priv->mm.interruptible = was_interruptible;
2b4e57bd
ED
3787 return;
3788 }
3789
6d90c952
DV
3790 intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
3791 intel_ring_emit(ring, MI_SET_CONTEXT);
f343c5f6 3792 intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) |
6d90c952
DV
3793 MI_MM_SPACE_GTT |
3794 MI_SAVE_EXT_STATE_EN |
3795 MI_RESTORE_EXT_STATE_EN |
3796 MI_RESTORE_INHIBIT);
3797 intel_ring_emit(ring, MI_SUSPEND_FLUSH);
3798 intel_ring_emit(ring, MI_NOOP);
3799 intel_ring_emit(ring, MI_FLUSH);
3800 intel_ring_advance(ring);
2b4e57bd
ED
3801
3802 /*
3803 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
3804 * does an implicit flush, combined with MI_FLUSH above, it should be
3805 * safe to assume that renderctx is valid
3806 */
3e960501
CW
3807 ret = intel_ring_idle(ring);
3808 dev_priv->mm.interruptible = was_interruptible;
2b4e57bd 3809 if (ret) {
def27a58 3810 DRM_ERROR("failed to enable ironlake power savings\n");
2b4e57bd 3811 ironlake_teardown_rc6(dev);
2b4e57bd
ED
3812 return;
3813 }
3814
f343c5f6 3815 I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN);
2b4e57bd 3816 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
dc39fff7
BW
3817
3818 intel_print_rc6_info(dev, INTEL_RC6_ENABLE);
2b4e57bd
ED
3819}
3820
dde18883
ED
3821static unsigned long intel_pxfreq(u32 vidfreq)
3822{
3823 unsigned long freq;
3824 int div = (vidfreq & 0x3f0000) >> 16;
3825 int post = (vidfreq & 0x3000) >> 12;
3826 int pre = (vidfreq & 0x7);
3827
3828 if (!pre)
3829 return 0;
3830
3831 freq = ((div * 133333) / ((1<<post) * pre));
3832
3833 return freq;
3834}
3835
eb48eb00
DV
3836static const struct cparams {
3837 u16 i;
3838 u16 t;
3839 u16 m;
3840 u16 c;
3841} cparams[] = {
3842 { 1, 1333, 301, 28664 },
3843 { 1, 1066, 294, 24460 },
3844 { 1, 800, 294, 25192 },
3845 { 0, 1333, 276, 27605 },
3846 { 0, 1066, 276, 27605 },
3847 { 0, 800, 231, 23784 },
3848};
3849
f531dcb2 3850static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
eb48eb00
DV
3851{
3852 u64 total_count, diff, ret;
3853 u32 count1, count2, count3, m = 0, c = 0;
3854 unsigned long now = jiffies_to_msecs(jiffies), diff1;
3855 int i;
3856
02d71956
DV
3857 assert_spin_locked(&mchdev_lock);
3858
20e4d407 3859 diff1 = now - dev_priv->ips.last_time1;
eb48eb00
DV
3860
3861 /* Prevent division-by-zero if we are asking too fast.
3862 * Also, we don't get interesting results if we are polling
3863 * faster than once in 10ms, so just return the saved value
3864 * in such cases.
3865 */
3866 if (diff1 <= 10)
20e4d407 3867 return dev_priv->ips.chipset_power;
eb48eb00
DV
3868
3869 count1 = I915_READ(DMIEC);
3870 count2 = I915_READ(DDREC);
3871 count3 = I915_READ(CSIEC);
3872
3873 total_count = count1 + count2 + count3;
3874
3875 /* FIXME: handle per-counter overflow */
20e4d407
DV
3876 if (total_count < dev_priv->ips.last_count1) {
3877 diff = ~0UL - dev_priv->ips.last_count1;
eb48eb00
DV
3878 diff += total_count;
3879 } else {
20e4d407 3880 diff = total_count - dev_priv->ips.last_count1;
eb48eb00
DV
3881 }
3882
3883 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
20e4d407
DV
3884 if (cparams[i].i == dev_priv->ips.c_m &&
3885 cparams[i].t == dev_priv->ips.r_t) {
eb48eb00
DV
3886 m = cparams[i].m;
3887 c = cparams[i].c;
3888 break;
3889 }
3890 }
3891
3892 diff = div_u64(diff, diff1);
3893 ret = ((m * diff) + c);
3894 ret = div_u64(ret, 10);
3895
20e4d407
DV
3896 dev_priv->ips.last_count1 = total_count;
3897 dev_priv->ips.last_time1 = now;
eb48eb00 3898
20e4d407 3899 dev_priv->ips.chipset_power = ret;
eb48eb00
DV
3900
3901 return ret;
3902}
3903
f531dcb2
CW
3904unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
3905{
3906 unsigned long val;
3907
3908 if (dev_priv->info->gen != 5)
3909 return 0;
3910
3911 spin_lock_irq(&mchdev_lock);
3912
3913 val = __i915_chipset_val(dev_priv);
3914
3915 spin_unlock_irq(&mchdev_lock);
3916
3917 return val;
3918}
3919
eb48eb00
DV
3920unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
3921{
3922 unsigned long m, x, b;
3923 u32 tsfs;
3924
3925 tsfs = I915_READ(TSFS);
3926
3927 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
3928 x = I915_READ8(TR1);
3929
3930 b = tsfs & TSFS_INTR_MASK;
3931
3932 return ((m * x) / 127) - b;
3933}
3934
3935static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
3936{
3937 static const struct v_table {
3938 u16 vd; /* in .1 mil */
3939 u16 vm; /* in .1 mil */
3940 } v_table[] = {
3941 { 0, 0, },
3942 { 375, 0, },
3943 { 500, 0, },
3944 { 625, 0, },
3945 { 750, 0, },
3946 { 875, 0, },
3947 { 1000, 0, },
3948 { 1125, 0, },
3949 { 4125, 3000, },
3950 { 4125, 3000, },
3951 { 4125, 3000, },
3952 { 4125, 3000, },
3953 { 4125, 3000, },
3954 { 4125, 3000, },
3955 { 4125, 3000, },
3956 { 4125, 3000, },
3957 { 4125, 3000, },
3958 { 4125, 3000, },
3959 { 4125, 3000, },
3960 { 4125, 3000, },
3961 { 4125, 3000, },
3962 { 4125, 3000, },
3963 { 4125, 3000, },
3964 { 4125, 3000, },
3965 { 4125, 3000, },
3966 { 4125, 3000, },
3967 { 4125, 3000, },
3968 { 4125, 3000, },
3969 { 4125, 3000, },
3970 { 4125, 3000, },
3971 { 4125, 3000, },
3972 { 4125, 3000, },
3973 { 4250, 3125, },
3974 { 4375, 3250, },
3975 { 4500, 3375, },
3976 { 4625, 3500, },
3977 { 4750, 3625, },
3978 { 4875, 3750, },
3979 { 5000, 3875, },
3980 { 5125, 4000, },
3981 { 5250, 4125, },
3982 { 5375, 4250, },
3983 { 5500, 4375, },
3984 { 5625, 4500, },
3985 { 5750, 4625, },
3986 { 5875, 4750, },
3987 { 6000, 4875, },
3988 { 6125, 5000, },
3989 { 6250, 5125, },
3990 { 6375, 5250, },
3991 { 6500, 5375, },
3992 { 6625, 5500, },
3993 { 6750, 5625, },
3994 { 6875, 5750, },
3995 { 7000, 5875, },
3996 { 7125, 6000, },
3997 { 7250, 6125, },
3998 { 7375, 6250, },
3999 { 7500, 6375, },
4000 { 7625, 6500, },
4001 { 7750, 6625, },
4002 { 7875, 6750, },
4003 { 8000, 6875, },
4004 { 8125, 7000, },
4005 { 8250, 7125, },
4006 { 8375, 7250, },
4007 { 8500, 7375, },
4008 { 8625, 7500, },
4009 { 8750, 7625, },
4010 { 8875, 7750, },
4011 { 9000, 7875, },
4012 { 9125, 8000, },
4013 { 9250, 8125, },
4014 { 9375, 8250, },
4015 { 9500, 8375, },
4016 { 9625, 8500, },
4017 { 9750, 8625, },
4018 { 9875, 8750, },
4019 { 10000, 8875, },
4020 { 10125, 9000, },
4021 { 10250, 9125, },
4022 { 10375, 9250, },
4023 { 10500, 9375, },
4024 { 10625, 9500, },
4025 { 10750, 9625, },
4026 { 10875, 9750, },
4027 { 11000, 9875, },
4028 { 11125, 10000, },
4029 { 11250, 10125, },
4030 { 11375, 10250, },
4031 { 11500, 10375, },
4032 { 11625, 10500, },
4033 { 11750, 10625, },
4034 { 11875, 10750, },
4035 { 12000, 10875, },
4036 { 12125, 11000, },
4037 { 12250, 11125, },
4038 { 12375, 11250, },
4039 { 12500, 11375, },
4040 { 12625, 11500, },
4041 { 12750, 11625, },
4042 { 12875, 11750, },
4043 { 13000, 11875, },
4044 { 13125, 12000, },
4045 { 13250, 12125, },
4046 { 13375, 12250, },
4047 { 13500, 12375, },
4048 { 13625, 12500, },
4049 { 13750, 12625, },
4050 { 13875, 12750, },
4051 { 14000, 12875, },
4052 { 14125, 13000, },
4053 { 14250, 13125, },
4054 { 14375, 13250, },
4055 { 14500, 13375, },
4056 { 14625, 13500, },
4057 { 14750, 13625, },
4058 { 14875, 13750, },
4059 { 15000, 13875, },
4060 { 15125, 14000, },
4061 { 15250, 14125, },
4062 { 15375, 14250, },
4063 { 15500, 14375, },
4064 { 15625, 14500, },
4065 { 15750, 14625, },
4066 { 15875, 14750, },
4067 { 16000, 14875, },
4068 { 16125, 15000, },
4069 };
4070 if (dev_priv->info->is_mobile)
4071 return v_table[pxvid].vm;
4072 else
4073 return v_table[pxvid].vd;
4074}
4075
02d71956 4076static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV
4077{
4078 struct timespec now, diff1;
4079 u64 diff;
4080 unsigned long diffms;
4081 u32 count;
4082
02d71956 4083 assert_spin_locked(&mchdev_lock);
eb48eb00
DV
4084
4085 getrawmonotonic(&now);
20e4d407 4086 diff1 = timespec_sub(now, dev_priv->ips.last_time2);
eb48eb00
DV
4087
4088 /* Don't divide by 0 */
4089 diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
4090 if (!diffms)
4091 return;
4092
4093 count = I915_READ(GFXEC);
4094
20e4d407
DV
4095 if (count < dev_priv->ips.last_count2) {
4096 diff = ~0UL - dev_priv->ips.last_count2;
eb48eb00
DV
4097 diff += count;
4098 } else {
20e4d407 4099 diff = count - dev_priv->ips.last_count2;
eb48eb00
DV
4100 }
4101
20e4d407
DV
4102 dev_priv->ips.last_count2 = count;
4103 dev_priv->ips.last_time2 = now;
eb48eb00
DV
4104
4105 /* More magic constants... */
4106 diff = diff * 1181;
4107 diff = div_u64(diff, diffms * 10);
20e4d407 4108 dev_priv->ips.gfx_power = diff;
eb48eb00
DV
4109}
4110
02d71956
DV
4111void i915_update_gfx_val(struct drm_i915_private *dev_priv)
4112{
4113 if (dev_priv->info->gen != 5)
4114 return;
4115
9270388e 4116 spin_lock_irq(&mchdev_lock);
02d71956
DV
4117
4118 __i915_update_gfx_val(dev_priv);
4119
9270388e 4120 spin_unlock_irq(&mchdev_lock);
02d71956
DV
4121}
4122
f531dcb2 4123static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
eb48eb00
DV
4124{
4125 unsigned long t, corr, state1, corr2, state2;
4126 u32 pxvid, ext_v;
4127
02d71956
DV
4128 assert_spin_locked(&mchdev_lock);
4129
c6a828d3 4130 pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4));
eb48eb00
DV
4131 pxvid = (pxvid >> 24) & 0x7f;
4132 ext_v = pvid_to_extvid(dev_priv, pxvid);
4133
4134 state1 = ext_v;
4135
4136 t = i915_mch_val(dev_priv);
4137
4138 /* Revel in the empirically derived constants */
4139
4140 /* Correction factor in 1/100000 units */
4141 if (t > 80)
4142 corr = ((t * 2349) + 135940);
4143 else if (t >= 50)
4144 corr = ((t * 964) + 29317);
4145 else /* < 50 */
4146 corr = ((t * 301) + 1004);
4147
4148 corr = corr * ((150142 * state1) / 10000 - 78642);
4149 corr /= 100000;
20e4d407 4150 corr2 = (corr * dev_priv->ips.corr);
eb48eb00
DV
4151
4152 state2 = (corr2 * state1) / 10000;
4153 state2 /= 100; /* convert to mW */
4154
02d71956 4155 __i915_update_gfx_val(dev_priv);
eb48eb00 4156
20e4d407 4157 return dev_priv->ips.gfx_power + state2;
eb48eb00
DV
4158}
4159
f531dcb2
CW
4160unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
4161{
4162 unsigned long val;
4163
4164 if (dev_priv->info->gen != 5)
4165 return 0;
4166
4167 spin_lock_irq(&mchdev_lock);
4168
4169 val = __i915_gfx_val(dev_priv);
4170
4171 spin_unlock_irq(&mchdev_lock);
4172
4173 return val;
4174}
4175
eb48eb00
DV
4176/**
4177 * i915_read_mch_val - return value for IPS use
4178 *
4179 * Calculate and return a value for the IPS driver to use when deciding whether
4180 * we have thermal and power headroom to increase CPU or GPU power budget.
4181 */
4182unsigned long i915_read_mch_val(void)
4183{
4184 struct drm_i915_private *dev_priv;
4185 unsigned long chipset_val, graphics_val, ret = 0;
4186
9270388e 4187 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
4188 if (!i915_mch_dev)
4189 goto out_unlock;
4190 dev_priv = i915_mch_dev;
4191
f531dcb2
CW
4192 chipset_val = __i915_chipset_val(dev_priv);
4193 graphics_val = __i915_gfx_val(dev_priv);
eb48eb00
DV
4194
4195 ret = chipset_val + graphics_val;
4196
4197out_unlock:
9270388e 4198 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
4199
4200 return ret;
4201}
4202EXPORT_SYMBOL_GPL(i915_read_mch_val);
4203
4204/**
4205 * i915_gpu_raise - raise GPU frequency limit
4206 *
4207 * Raise the limit; IPS indicates we have thermal headroom.
4208 */
4209bool i915_gpu_raise(void)
4210{
4211 struct drm_i915_private *dev_priv;
4212 bool ret = true;
4213
9270388e 4214 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
4215 if (!i915_mch_dev) {
4216 ret = false;
4217 goto out_unlock;
4218 }
4219 dev_priv = i915_mch_dev;
4220
20e4d407
DV
4221 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
4222 dev_priv->ips.max_delay--;
eb48eb00
DV
4223
4224out_unlock:
9270388e 4225 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
4226
4227 return ret;
4228}
4229EXPORT_SYMBOL_GPL(i915_gpu_raise);
4230
4231/**
4232 * i915_gpu_lower - lower GPU frequency limit
4233 *
4234 * IPS indicates we're close to a thermal limit, so throttle back the GPU
4235 * frequency maximum.
4236 */
4237bool i915_gpu_lower(void)
4238{
4239 struct drm_i915_private *dev_priv;
4240 bool ret = true;
4241
9270388e 4242 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
4243 if (!i915_mch_dev) {
4244 ret = false;
4245 goto out_unlock;
4246 }
4247 dev_priv = i915_mch_dev;
4248
20e4d407
DV
4249 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
4250 dev_priv->ips.max_delay++;
eb48eb00
DV
4251
4252out_unlock:
9270388e 4253 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
4254
4255 return ret;
4256}
4257EXPORT_SYMBOL_GPL(i915_gpu_lower);
4258
4259/**
4260 * i915_gpu_busy - indicate GPU business to IPS
4261 *
4262 * Tell the IPS driver whether or not the GPU is busy.
4263 */
4264bool i915_gpu_busy(void)
4265{
4266 struct drm_i915_private *dev_priv;
f047e395 4267 struct intel_ring_buffer *ring;
eb48eb00 4268 bool ret = false;
f047e395 4269 int i;
eb48eb00 4270
9270388e 4271 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
4272 if (!i915_mch_dev)
4273 goto out_unlock;
4274 dev_priv = i915_mch_dev;
4275
f047e395
CW
4276 for_each_ring(ring, dev_priv, i)
4277 ret |= !list_empty(&ring->request_list);
eb48eb00
DV
4278
4279out_unlock:
9270388e 4280 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
4281
4282 return ret;
4283}
4284EXPORT_SYMBOL_GPL(i915_gpu_busy);
4285
4286/**
4287 * i915_gpu_turbo_disable - disable graphics turbo
4288 *
4289 * Disable graphics turbo by resetting the max frequency and setting the
4290 * current frequency to the default.
4291 */
4292bool i915_gpu_turbo_disable(void)
4293{
4294 struct drm_i915_private *dev_priv;
4295 bool ret = true;
4296
9270388e 4297 spin_lock_irq(&mchdev_lock);
eb48eb00
DV
4298 if (!i915_mch_dev) {
4299 ret = false;
4300 goto out_unlock;
4301 }
4302 dev_priv = i915_mch_dev;
4303
20e4d407 4304 dev_priv->ips.max_delay = dev_priv->ips.fstart;
eb48eb00 4305
20e4d407 4306 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
eb48eb00
DV
4307 ret = false;
4308
4309out_unlock:
9270388e 4310 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
4311
4312 return ret;
4313}
4314EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
4315
4316/**
4317 * Tells the intel_ips driver that the i915 driver is now loaded, if
4318 * IPS got loaded first.
4319 *
4320 * This awkward dance is so that neither module has to depend on the
4321 * other in order for IPS to do the appropriate communication of
4322 * GPU turbo limits to i915.
4323 */
4324static void
4325ips_ping_for_i915_load(void)
4326{
4327 void (*link)(void);
4328
4329 link = symbol_get(ips_link_to_i915_driver);
4330 if (link) {
4331 link();
4332 symbol_put(ips_link_to_i915_driver);
4333 }
4334}
4335
4336void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
4337{
02d71956
DV
4338 /* We only register the i915 ips part with intel-ips once everything is
4339 * set up, to avoid intel-ips sneaking in and reading bogus values. */
9270388e 4340 spin_lock_irq(&mchdev_lock);
eb48eb00 4341 i915_mch_dev = dev_priv;
9270388e 4342 spin_unlock_irq(&mchdev_lock);
eb48eb00
DV
4343
4344 ips_ping_for_i915_load();
4345}
4346
4347void intel_gpu_ips_teardown(void)
4348{
9270388e 4349 spin_lock_irq(&mchdev_lock);
eb48eb00 4350 i915_mch_dev = NULL;
9270388e 4351 spin_unlock_irq(&mchdev_lock);
eb48eb00 4352}
76c3552f 4353
8090c6b9 4354static void intel_init_emon(struct drm_device *dev)
dde18883
ED
4355{
4356 struct drm_i915_private *dev_priv = dev->dev_private;
4357 u32 lcfuse;
4358 u8 pxw[16];
4359 int i;
4360
4361 /* Disable to program */
4362 I915_WRITE(ECR, 0);
4363 POSTING_READ(ECR);
4364
4365 /* Program energy weights for various events */
4366 I915_WRITE(SDEW, 0x15040d00);
4367 I915_WRITE(CSIEW0, 0x007f0000);
4368 I915_WRITE(CSIEW1, 0x1e220004);
4369 I915_WRITE(CSIEW2, 0x04000004);
4370
4371 for (i = 0; i < 5; i++)
4372 I915_WRITE(PEW + (i * 4), 0);
4373 for (i = 0; i < 3; i++)
4374 I915_WRITE(DEW + (i * 4), 0);
4375
4376 /* Program P-state weights to account for frequency power adjustment */
4377 for (i = 0; i < 16; i++) {
4378 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
4379 unsigned long freq = intel_pxfreq(pxvidfreq);
4380 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
4381 PXVFREQ_PX_SHIFT;
4382 unsigned long val;
4383
4384 val = vid * vid;
4385 val *= (freq / 1000);
4386 val *= 255;
4387 val /= (127*127*900);
4388 if (val > 0xff)
4389 DRM_ERROR("bad pxval: %ld\n", val);
4390 pxw[i] = val;
4391 }
4392 /* Render standby states get 0 weight */
4393 pxw[14] = 0;
4394 pxw[15] = 0;
4395
4396 for (i = 0; i < 4; i++) {
4397 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
4398 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
4399 I915_WRITE(PXW + (i * 4), val);
4400 }
4401
4402 /* Adjust magic regs to magic values (more experimental results) */
4403 I915_WRITE(OGW0, 0);
4404 I915_WRITE(OGW1, 0);
4405 I915_WRITE(EG0, 0x00007f00);
4406 I915_WRITE(EG1, 0x0000000e);
4407 I915_WRITE(EG2, 0x000e0000);
4408 I915_WRITE(EG3, 0x68000300);
4409 I915_WRITE(EG4, 0x42000000);
4410 I915_WRITE(EG5, 0x00140031);
4411 I915_WRITE(EG6, 0);
4412 I915_WRITE(EG7, 0);
4413
4414 for (i = 0; i < 8; i++)
4415 I915_WRITE(PXWL + (i * 4), 0);
4416
4417 /* Enable PMON + select events */
4418 I915_WRITE(ECR, 0x80000019);
4419
4420 lcfuse = I915_READ(LCFUSE02);
4421
20e4d407 4422 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
dde18883
ED
4423}
4424
8090c6b9
DV
4425void intel_disable_gt_powersave(struct drm_device *dev)
4426{
1a01ab3b
JB
4427 struct drm_i915_private *dev_priv = dev->dev_private;
4428
fd0c0642
DV
4429 /* Interrupts should be disabled already to avoid re-arming. */
4430 WARN_ON(dev->irq_enabled);
4431
930ebb46 4432 if (IS_IRONLAKE_M(dev)) {
8090c6b9 4433 ironlake_disable_drps(dev);
930ebb46 4434 ironlake_disable_rc6(dev);
0a073b84 4435 } else if (INTEL_INFO(dev)->gen >= 6) {
1a01ab3b 4436 cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
250848ca 4437 cancel_work_sync(&dev_priv->rps.work);
4fc688ce 4438 mutex_lock(&dev_priv->rps.hw_lock);
d20d4f0c
JB
4439 if (IS_VALLEYVIEW(dev))
4440 valleyview_disable_rps(dev);
4441 else
4442 gen6_disable_rps(dev);
c0951f0c 4443 dev_priv->rps.enabled = false;
4fc688ce 4444 mutex_unlock(&dev_priv->rps.hw_lock);
930ebb46 4445 }
8090c6b9
DV
4446}
4447
1a01ab3b
JB
4448static void intel_gen6_powersave_work(struct work_struct *work)
4449{
4450 struct drm_i915_private *dev_priv =
4451 container_of(work, struct drm_i915_private,
4452 rps.delayed_resume_work.work);
4453 struct drm_device *dev = dev_priv->dev;
4454
4fc688ce 4455 mutex_lock(&dev_priv->rps.hw_lock);
0a073b84
JB
4456
4457 if (IS_VALLEYVIEW(dev)) {
4458 valleyview_enable_rps(dev);
6edee7f3
BW
4459 } else if (IS_BROADWELL(dev)) {
4460 gen8_enable_rps(dev);
4461 gen6_update_ring_freq(dev);
0a073b84
JB
4462 } else {
4463 gen6_enable_rps(dev);
4464 gen6_update_ring_freq(dev);
4465 }
c0951f0c 4466 dev_priv->rps.enabled = true;
4fc688ce 4467 mutex_unlock(&dev_priv->rps.hw_lock);
1a01ab3b
JB
4468}
4469
8090c6b9
DV
4470void intel_enable_gt_powersave(struct drm_device *dev)
4471{
1a01ab3b
JB
4472 struct drm_i915_private *dev_priv = dev->dev_private;
4473
8090c6b9
DV
4474 if (IS_IRONLAKE_M(dev)) {
4475 ironlake_enable_drps(dev);
4476 ironlake_enable_rc6(dev);
4477 intel_init_emon(dev);
0a073b84 4478 } else if (IS_GEN6(dev) || IS_GEN7(dev)) {
1a01ab3b
JB
4479 /*
4480 * PCU communication is slow and this doesn't need to be
4481 * done at any specific time, so do this out of our fast path
4482 * to make resume and init faster.
4483 */
4484 schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
4485 round_jiffies_up_relative(HZ));
8090c6b9
DV
4486 }
4487}
4488
3107bd48
DV
4489static void ibx_init_clock_gating(struct drm_device *dev)
4490{
4491 struct drm_i915_private *dev_priv = dev->dev_private;
4492
4493 /*
4494 * On Ibex Peak and Cougar Point, we need to disable clock
4495 * gating for the panel power sequencer or it will fail to
4496 * start up when no ports are active.
4497 */
4498 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
4499}
4500
0e088b8f
VS
4501static void g4x_disable_trickle_feed(struct drm_device *dev)
4502{
4503 struct drm_i915_private *dev_priv = dev->dev_private;
4504 int pipe;
4505
4506 for_each_pipe(pipe) {
4507 I915_WRITE(DSPCNTR(pipe),
4508 I915_READ(DSPCNTR(pipe)) |
4509 DISPPLANE_TRICKLE_FEED_DISABLE);
1dba99f4 4510 intel_flush_primary_plane(dev_priv, pipe);
0e088b8f
VS
4511 }
4512}
4513
017636cc
VS
4514static void ilk_init_lp_watermarks(struct drm_device *dev)
4515{
4516 struct drm_i915_private *dev_priv = dev->dev_private;
4517
4518 I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
4519 I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
4520 I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
4521
4522 /*
4523 * Don't touch WM1S_LP_EN here.
4524 * Doing so could cause underruns.
4525 */
4526}
4527
1fa61106 4528static void ironlake_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
4529{
4530 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 4531 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 4532
f1e8fa56
DL
4533 /*
4534 * Required for FBC
4535 * WaFbcDisableDpfcClockGating:ilk
4536 */
4d47e4f5
DL
4537 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
4538 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
4539 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED
4540
4541 I915_WRITE(PCH_3DCGDIS0,
4542 MARIUNIT_CLOCK_GATE_DISABLE |
4543 SVSMUNIT_CLOCK_GATE_DISABLE);
4544 I915_WRITE(PCH_3DCGDIS1,
4545 VFMUNIT_CLOCK_GATE_DISABLE);
4546
6f1d69b0
ED
4547 /*
4548 * According to the spec the following bits should be set in
4549 * order to enable memory self-refresh
4550 * The bit 22/21 of 0x42004
4551 * The bit 5 of 0x42020
4552 * The bit 15 of 0x45000
4553 */
4554 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4555 (I915_READ(ILK_DISPLAY_CHICKEN2) |
4556 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4d47e4f5 4557 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
6f1d69b0
ED
4558 I915_WRITE(DISP_ARB_CTL,
4559 (I915_READ(DISP_ARB_CTL) |
4560 DISP_FBC_WM_DIS));
017636cc
VS
4561
4562 ilk_init_lp_watermarks(dev);
6f1d69b0
ED
4563
4564 /*
4565 * Based on the document from hardware guys the following bits
4566 * should be set unconditionally in order to enable FBC.
4567 * The bit 22 of 0x42000
4568 * The bit 22 of 0x42004
4569 * The bit 7,8,9 of 0x42020.
4570 */
4571 if (IS_IRONLAKE_M(dev)) {
4bb35334 4572 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
6f1d69b0
ED
4573 I915_WRITE(ILK_DISPLAY_CHICKEN1,
4574 I915_READ(ILK_DISPLAY_CHICKEN1) |
4575 ILK_FBCQ_DIS);
4576 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4577 I915_READ(ILK_DISPLAY_CHICKEN2) |
4578 ILK_DPARB_GATE);
6f1d69b0
ED
4579 }
4580
4d47e4f5
DL
4581 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
4582
6f1d69b0
ED
4583 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4584 I915_READ(ILK_DISPLAY_CHICKEN2) |
4585 ILK_ELPIN_409_SELECT);
4586 I915_WRITE(_3D_CHICKEN2,
4587 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
4588 _3D_CHICKEN2_WM_READ_PIPELINED);
4358a374 4589
ecdb4eb7 4590 /* WaDisableRenderCachePipelinedFlush:ilk */
4358a374
DV
4591 I915_WRITE(CACHE_MODE_0,
4592 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3107bd48 4593
0e088b8f 4594 g4x_disable_trickle_feed(dev);
bdad2b2f 4595
3107bd48
DV
4596 ibx_init_clock_gating(dev);
4597}
4598
4599static void cpt_init_clock_gating(struct drm_device *dev)
4600{
4601 struct drm_i915_private *dev_priv = dev->dev_private;
4602 int pipe;
3f704fa2 4603 uint32_t val;
3107bd48
DV
4604
4605 /*
4606 * On Ibex Peak and Cougar Point, we need to disable clock
4607 * gating for the panel power sequencer or it will fail to
4608 * start up when no ports are active.
4609 */
cd664078
JB
4610 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
4611 PCH_DPLUNIT_CLOCK_GATE_DISABLE |
4612 PCH_CPUNIT_CLOCK_GATE_DISABLE);
3107bd48
DV
4613 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
4614 DPLS_EDP_PPS_FIX_DIS);
335c07b7
TI
4615 /* The below fixes the weird display corruption, a few pixels shifted
4616 * downward, on (only) LVDS of some HP laptops with IVY.
4617 */
3f704fa2 4618 for_each_pipe(pipe) {
dc4bd2d1
PZ
4619 val = I915_READ(TRANS_CHICKEN2(pipe));
4620 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
4621 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
41aa3448 4622 if (dev_priv->vbt.fdi_rx_polarity_inverted)
3f704fa2 4623 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
dc4bd2d1
PZ
4624 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
4625 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
4626 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3f704fa2
PZ
4627 I915_WRITE(TRANS_CHICKEN2(pipe), val);
4628 }
3107bd48
DV
4629 /* WADP0ClockGatingDisable */
4630 for_each_pipe(pipe) {
4631 I915_WRITE(TRANS_CHICKEN1(pipe),
4632 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
4633 }
6f1d69b0
ED
4634}
4635
1d7aaa0c
DV
4636static void gen6_check_mch_setup(struct drm_device *dev)
4637{
4638 struct drm_i915_private *dev_priv = dev->dev_private;
4639 uint32_t tmp;
4640
4641 tmp = I915_READ(MCH_SSKPD);
4642 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
4643 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
4644 DRM_INFO("This can cause pipe underruns and display issues.\n");
4645 DRM_INFO("Please upgrade your BIOS to fix this.\n");
4646 }
4647}
4648
1fa61106 4649static void gen6_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
4650{
4651 struct drm_i915_private *dev_priv = dev->dev_private;
231e54f6 4652 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
6f1d69b0 4653
231e54f6 4654 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
6f1d69b0
ED
4655
4656 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4657 I915_READ(ILK_DISPLAY_CHICKEN2) |
4658 ILK_ELPIN_409_SELECT);
4659
ecdb4eb7 4660 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4283908e
DV
4661 I915_WRITE(_3D_CHICKEN,
4662 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
4663
ecdb4eb7 4664 /* WaSetupGtModeTdRowDispatch:snb */
6547fbdb
DV
4665 if (IS_SNB_GT1(dev))
4666 I915_WRITE(GEN6_GT_MODE,
4667 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
4668
017636cc 4669 ilk_init_lp_watermarks(dev);
6f1d69b0 4670
6f1d69b0 4671 I915_WRITE(CACHE_MODE_0,
50743298 4672 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
6f1d69b0
ED
4673
4674 I915_WRITE(GEN6_UCGCTL1,
4675 I915_READ(GEN6_UCGCTL1) |
4676 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
4677 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
4678
4679 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4680 * gating disable must be set. Failure to set it results in
4681 * flickering pixels due to Z write ordering failures after
4682 * some amount of runtime in the Mesa "fire" demo, and Unigine
4683 * Sanctuary and Tropics, and apparently anything else with
4684 * alpha test or pixel discard.
4685 *
4686 * According to the spec, bit 11 (RCCUNIT) must also be set,
4687 * but we didn't debug actual testcases to find it out.
0f846f81 4688 *
ef59318c
VS
4689 * WaDisableRCCUnitClockGating:snb
4690 * WaDisableRCPBUnitClockGating:snb
6f1d69b0
ED
4691 */
4692 I915_WRITE(GEN6_UCGCTL2,
4693 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
4694 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4695
4696 /* Bspec says we need to always set all mask bits. */
26b6e44a
KG
4697 I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) |
4698 _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL);
6f1d69b0
ED
4699
4700 /*
4701 * According to the spec the following bits should be
4702 * set in order to enable memory self-refresh and fbc:
4703 * The bit21 and bit22 of 0x42000
4704 * The bit21 and bit22 of 0x42004
4705 * The bit5 and bit7 of 0x42020
4706 * The bit14 of 0x70180
4707 * The bit14 of 0x71180
4bb35334
DL
4708 *
4709 * WaFbcAsynchFlipDisableFbcQueue:snb
6f1d69b0
ED
4710 */
4711 I915_WRITE(ILK_DISPLAY_CHICKEN1,
4712 I915_READ(ILK_DISPLAY_CHICKEN1) |
4713 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
4714 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4715 I915_READ(ILK_DISPLAY_CHICKEN2) |
4716 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
231e54f6
DL
4717 I915_WRITE(ILK_DSPCLK_GATE_D,
4718 I915_READ(ILK_DSPCLK_GATE_D) |
4719 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
4720 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
6f1d69b0 4721
0e088b8f 4722 g4x_disable_trickle_feed(dev);
f8f2ac9a
BW
4723
4724 /* The default value should be 0x200 according to docs, but the two
4725 * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */
4726 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff));
4727 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI));
3107bd48
DV
4728
4729 cpt_init_clock_gating(dev);
1d7aaa0c
DV
4730
4731 gen6_check_mch_setup(dev);
6f1d69b0
ED
4732}
4733
4734static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
4735{
4736 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
4737
3aad9059 4738 /*
46680e0a 4739 * WaVSThreadDispatchOverride:ivb,vlv
3aad9059
VS
4740 *
4741 * This actually overrides the dispatch
4742 * mode for all thread types.
4743 */
6f1d69b0
ED
4744 reg &= ~GEN7_FF_SCHED_MASK;
4745 reg |= GEN7_FF_TS_SCHED_HW;
4746 reg |= GEN7_FF_VS_SCHED_HW;
4747 reg |= GEN7_FF_DS_SCHED_HW;
4748
4749 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
4750}
4751
17a303ec
PZ
4752static void lpt_init_clock_gating(struct drm_device *dev)
4753{
4754 struct drm_i915_private *dev_priv = dev->dev_private;
4755
4756 /*
4757 * TODO: this bit should only be enabled when really needed, then
4758 * disabled when not needed anymore in order to save power.
4759 */
4760 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
4761 I915_WRITE(SOUTH_DSPCLK_GATE_D,
4762 I915_READ(SOUTH_DSPCLK_GATE_D) |
4763 PCH_LP_PARTITION_LEVEL_DISABLE);
0a790cdb
PZ
4764
4765 /* WADPOClockGatingDisable:hsw */
4766 I915_WRITE(_TRANSA_CHICKEN1,
4767 I915_READ(_TRANSA_CHICKEN1) |
4768 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
17a303ec
PZ
4769}
4770
7d708ee4
ID
4771static void lpt_suspend_hw(struct drm_device *dev)
4772{
4773 struct drm_i915_private *dev_priv = dev->dev_private;
4774
4775 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) {
4776 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
4777
4778 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
4779 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
4780 }
4781}
4782
1020a5c2
BW
4783static void gen8_init_clock_gating(struct drm_device *dev)
4784{
4785 struct drm_i915_private *dev_priv = dev->dev_private;
fe4ab3ce 4786 enum pipe i;
1020a5c2
BW
4787
4788 I915_WRITE(WM3_LP_ILK, 0);
4789 I915_WRITE(WM2_LP_ILK, 0);
4790 I915_WRITE(WM1_LP_ILK, 0);
50ed5fbd
BW
4791
4792 /* FIXME(BDW): Check all the w/a, some might only apply to
4793 * pre-production hw. */
4794
4167e32c
DL
4795 /*
4796 * This GEN8_CENTROID_PIXEL_OPT_DIS W/A is only needed for
4797 * pre-production hardware
4798 */
fd392b60
BW
4799 I915_WRITE(HALF_SLICE_CHICKEN3,
4800 _MASKED_BIT_ENABLE(GEN8_CENTROID_PIXEL_OPT_DIS));
bf66347c
BW
4801 I915_WRITE(HALF_SLICE_CHICKEN3,
4802 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS));
4afe8d33
BW
4803 I915_WRITE(GAMTARBMODE, _MASKED_BIT_ENABLE(ARB_MODE_BWGTLB_DISABLE));
4804
7f88da0c
BW
4805 I915_WRITE(_3D_CHICKEN3,
4806 _3D_CHICKEN_SDE_LIMIT_FIFO_POLY_DEPTH(2));
4807
a75f3628
BW
4808 I915_WRITE(COMMON_SLICE_CHICKEN2,
4809 _MASKED_BIT_ENABLE(GEN8_CSC2_SBE_VUE_CACHE_CONSERVATIVE));
4810
4c2e7a5f
BW
4811 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4812 _MASKED_BIT_ENABLE(GEN7_SINGLE_SUBSCAN_DISPATCH_ENABLE));
4813
ab57fff1 4814 /* WaSwitchSolVfFArbitrationPriority:bdw */
50ed5fbd 4815 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
fe4ab3ce 4816
ab57fff1 4817 /* WaPsrDPAMaskVBlankInSRD:bdw */
fe4ab3ce
BW
4818 I915_WRITE(CHICKEN_PAR1_1,
4819 I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
4820
ab57fff1 4821 /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
fe4ab3ce
BW
4822 for_each_pipe(i) {
4823 I915_WRITE(CHICKEN_PIPESL_1(i),
4824 I915_READ(CHICKEN_PIPESL_1(i) |
4825 DPRS_MASK_VBLANK_SRD));
4826 }
63801f21
BW
4827
4828 /* Use Force Non-Coherent whenever executing a 3D context. This is a
4829 * workaround for for a possible hang in the unlikely event a TLB
4830 * invalidation occurs during a PSD flush.
4831 */
4832 I915_WRITE(HDC_CHICKEN0,
4833 I915_READ(HDC_CHICKEN0) |
4834 _MASKED_BIT_ENABLE(HDC_FORCE_NON_COHERENT));
ab57fff1
BW
4835
4836 /* WaVSRefCountFullforceMissDisable:bdw */
4837 /* WaDSRefCountFullforceMissDisable:bdw */
4838 I915_WRITE(GEN7_FF_THREAD_MODE,
4839 I915_READ(GEN7_FF_THREAD_MODE) &
4840 ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
1020a5c2
BW
4841}
4842
cad2a2d7
ED
4843static void haswell_init_clock_gating(struct drm_device *dev)
4844{
4845 struct drm_i915_private *dev_priv = dev->dev_private;
cad2a2d7 4846
017636cc 4847 ilk_init_lp_watermarks(dev);
cad2a2d7 4848
f3fc4884
FJ
4849 /* L3 caching of data atomics doesn't work -- disable it. */
4850 I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
4851 I915_WRITE(HSW_ROW_CHICKEN3,
4852 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
4853
ecdb4eb7 4854 /* This is required by WaCatErrorRejectionIssue:hsw */
cad2a2d7
ED
4855 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4856 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4857 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4858
e36ea7ff
VS
4859 /* WaVSRefCountFullforceMissDisable:hsw */
4860 I915_WRITE(GEN7_FF_THREAD_MODE,
4861 I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);
cad2a2d7 4862
fe27c606
CW
4863 /* enable HiZ Raw Stall Optimization */
4864 I915_WRITE(CACHE_MODE_0_GEN7,
4865 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
4866
ecdb4eb7 4867 /* WaDisable4x2SubspanOptimization:hsw */
cad2a2d7
ED
4868 I915_WRITE(CACHE_MODE_1,
4869 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
1544d9d5 4870
ecdb4eb7 4871 /* WaSwitchSolVfFArbitrationPriority:hsw */
e3dff585
BW
4872 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
4873
90a88643
PZ
4874 /* WaRsPkgCStateDisplayPMReq:hsw */
4875 I915_WRITE(CHICKEN_PAR1_1,
4876 I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
1544d9d5 4877
17a303ec 4878 lpt_init_clock_gating(dev);
cad2a2d7
ED
4879}
4880
1fa61106 4881static void ivybridge_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
4882{
4883 struct drm_i915_private *dev_priv = dev->dev_private;
20848223 4884 uint32_t snpcr;
6f1d69b0 4885
017636cc 4886 ilk_init_lp_watermarks(dev);
6f1d69b0 4887
231e54f6 4888 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 4889
ecdb4eb7 4890 /* WaDisableEarlyCull:ivb */
87f8020e
JB
4891 I915_WRITE(_3D_CHICKEN3,
4892 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4893
ecdb4eb7 4894 /* WaDisableBackToBackFlipFix:ivb */
6f1d69b0
ED
4895 I915_WRITE(IVB_CHICKEN3,
4896 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4897 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4898
ecdb4eb7 4899 /* WaDisablePSDDualDispatchEnable:ivb */
12f3382b
JB
4900 if (IS_IVB_GT1(dev))
4901 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4902 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 4903
ecdb4eb7 4904 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
6f1d69b0
ED
4905 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4906 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4907
ecdb4eb7 4908 /* WaApplyL3ControlAndL3ChickenMode:ivb */
6f1d69b0
ED
4909 I915_WRITE(GEN7_L3CNTLREG1,
4910 GEN7_WA_FOR_GEN7_L3_CONTROL);
4911 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
8ab43976
JB
4912 GEN7_WA_L3_CHICKEN_MODE);
4913 if (IS_IVB_GT1(dev))
4914 I915_WRITE(GEN7_ROW_CHICKEN2,
4915 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2
VS
4916 else {
4917 /* must write both registers */
4918 I915_WRITE(GEN7_ROW_CHICKEN2,
4919 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
8ab43976
JB
4920 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
4921 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
412236c2 4922 }
6f1d69b0 4923
ecdb4eb7 4924 /* WaForceL3Serialization:ivb */
61939d97
JB
4925 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4926 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4927
1b80a19a 4928 /*
0f846f81 4929 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 4930 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
0f846f81
JB
4931 */
4932 I915_WRITE(GEN6_UCGCTL2,
28acf3b2 4933 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 4934
ecdb4eb7 4935 /* This is required by WaCatErrorRejectionIssue:ivb */
6f1d69b0
ED
4936 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4937 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4938 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4939
0e088b8f 4940 g4x_disable_trickle_feed(dev);
6f1d69b0
ED
4941
4942 gen7_setup_fixed_func_scheduler(dev_priv);
97e1930f 4943
116f2b6d
CW
4944 /* enable HiZ Raw Stall Optimization */
4945 I915_WRITE(CACHE_MODE_0_GEN7,
4946 _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
4947
ecdb4eb7 4948 /* WaDisable4x2SubspanOptimization:ivb */
97e1930f
DV
4949 I915_WRITE(CACHE_MODE_1,
4950 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
20848223
BW
4951
4952 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
4953 snpcr &= ~GEN6_MBC_SNPCR_MASK;
4954 snpcr |= GEN6_MBC_SNPCR_MED;
4955 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
3107bd48 4956
ab5c608b
BW
4957 if (!HAS_PCH_NOP(dev))
4958 cpt_init_clock_gating(dev);
1d7aaa0c
DV
4959
4960 gen6_check_mch_setup(dev);
6f1d69b0
ED
4961}
4962
1fa61106 4963static void valleyview_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
4964{
4965 struct drm_i915_private *dev_priv = dev->dev_private;
85b1d7b3
JB
4966 u32 val;
4967
4968 mutex_lock(&dev_priv->rps.hw_lock);
4969 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
4970 mutex_unlock(&dev_priv->rps.hw_lock);
4971 switch ((val >> 6) & 3) {
4972 case 0:
85b1d7b3
JB
4973 dev_priv->mem_freq = 800;
4974 break;
f64a28a7 4975 case 1:
85b1d7b3
JB
4976 dev_priv->mem_freq = 1066;
4977 break;
f64a28a7 4978 case 2:
85b1d7b3
JB
4979 dev_priv->mem_freq = 1333;
4980 break;
f64a28a7 4981 case 3:
2325991e 4982 dev_priv->mem_freq = 1333;
f64a28a7 4983 break;
85b1d7b3
JB
4984 }
4985 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
6f1d69b0 4986
d7fe0cc0 4987 I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
6f1d69b0 4988
ecdb4eb7 4989 /* WaDisableEarlyCull:vlv */
87f8020e
JB
4990 I915_WRITE(_3D_CHICKEN3,
4991 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4992
ecdb4eb7 4993 /* WaDisableBackToBackFlipFix:vlv */
6f1d69b0
ED
4994 I915_WRITE(IVB_CHICKEN3,
4995 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4996 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4997
fad7d36e 4998 /* WaPsdDispatchEnable:vlv */
ecdb4eb7 4999 /* WaDisablePSDDualDispatchEnable:vlv */
12f3382b 5000 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
d3bc0303
JB
5001 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
5002 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
12f3382b 5003
d50764a9 5004 /* WaDisableL3CacheAging:vlv */
d0cf5ead 5005 I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS);
6f1d69b0 5006
ecdb4eb7 5007 /* WaForceL3Serialization:vlv */
61939d97
JB
5008 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5009 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5010
ecdb4eb7 5011 /* WaDisableDopClockGating:vlv */
8ab43976
JB
5012 I915_WRITE(GEN7_ROW_CHICKEN2,
5013 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5014
ecdb4eb7 5015 /* This is required by WaCatErrorRejectionIssue:vlv */
6f1d69b0
ED
5016 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5017 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5018 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5019
46680e0a
VS
5020 gen7_setup_fixed_func_scheduler(dev_priv);
5021
3c0edaeb 5022 /*
0f846f81 5023 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
ecdb4eb7 5024 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
0f846f81
JB
5025 */
5026 I915_WRITE(GEN6_UCGCTL2,
3c0edaeb 5027 GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
0f846f81 5028
c5c32cda 5029 /* WaDisableL3Bank2xClockGate:vlv */
e3f33d46
JB
5030 I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
5031
e0d8d59b 5032 I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
6f1d69b0 5033
afd58e79
VS
5034 /*
5035 * BSpec says this must be set, even though
5036 * WaDisable4x2SubspanOptimization isn't listed for VLV.
5037 */
6b26c86d
DV
5038 I915_WRITE(CACHE_MODE_1,
5039 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
7983117f 5040
031994ee
VS
5041 /*
5042 * WaIncreaseL3CreditsForVLVB0:vlv
5043 * This is the hardware default actually.
5044 */
5045 I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);
5046
2d809570 5047 /*
ecdb4eb7 5048 * WaDisableVLVClockGating_VBIIssue:vlv
2d809570
JB
5049 * Disable clock gating on th GCFG unit to prevent a delay
5050 * in the reporting of vblank events.
5051 */
7a0d1eed 5052 I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
6f1d69b0
ED
5053}
5054
1fa61106 5055static void g4x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
5056{
5057 struct drm_i915_private *dev_priv = dev->dev_private;
5058 uint32_t dspclk_gate;
5059
5060 I915_WRITE(RENCLK_GATE_D1, 0);
5061 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
5062 GS_UNIT_CLOCK_GATE_DISABLE |
5063 CL_UNIT_CLOCK_GATE_DISABLE);
5064 I915_WRITE(RAMCLK_GATE_D, 0);
5065 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
5066 OVRUNIT_CLOCK_GATE_DISABLE |
5067 OVCUNIT_CLOCK_GATE_DISABLE;
5068 if (IS_GM45(dev))
5069 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
5070 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4358a374
DV
5071
5072 /* WaDisableRenderCachePipelinedFlush */
5073 I915_WRITE(CACHE_MODE_0,
5074 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
de1aa629 5075
0e088b8f 5076 g4x_disable_trickle_feed(dev);
6f1d69b0
ED
5077}
5078
1fa61106 5079static void crestline_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
5080{
5081 struct drm_i915_private *dev_priv = dev->dev_private;
5082
5083 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
5084 I915_WRITE(RENCLK_GATE_D2, 0);
5085 I915_WRITE(DSPCLK_GATE_D, 0);
5086 I915_WRITE(RAMCLK_GATE_D, 0);
5087 I915_WRITE16(DEUC, 0);
20f94967
VS
5088 I915_WRITE(MI_ARB_STATE,
5089 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED
5090}
5091
1fa61106 5092static void broadwater_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
5093{
5094 struct drm_i915_private *dev_priv = dev->dev_private;
5095
5096 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
5097 I965_RCC_CLOCK_GATE_DISABLE |
5098 I965_RCPB_CLOCK_GATE_DISABLE |
5099 I965_ISC_CLOCK_GATE_DISABLE |
5100 I965_FBC_CLOCK_GATE_DISABLE);
5101 I915_WRITE(RENCLK_GATE_D2, 0);
20f94967
VS
5102 I915_WRITE(MI_ARB_STATE,
5103 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
6f1d69b0
ED
5104}
5105
1fa61106 5106static void gen3_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
5107{
5108 struct drm_i915_private *dev_priv = dev->dev_private;
5109 u32 dstate = I915_READ(D_STATE);
5110
5111 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
5112 DSTATE_DOT_CLOCK_GATING;
5113 I915_WRITE(D_STATE, dstate);
13a86b85
CW
5114
5115 if (IS_PINEVIEW(dev))
5116 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
974a3b0f
DV
5117
5118 /* IIR "flip pending" means done if this bit is set */
5119 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
6f1d69b0
ED
5120}
5121
1fa61106 5122static void i85x_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
5123{
5124 struct drm_i915_private *dev_priv = dev->dev_private;
5125
5126 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
5127}
5128
1fa61106 5129static void i830_init_clock_gating(struct drm_device *dev)
6f1d69b0
ED
5130{
5131 struct drm_i915_private *dev_priv = dev->dev_private;
5132
5133 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
5134}
5135
6f1d69b0
ED
5136void intel_init_clock_gating(struct drm_device *dev)
5137{
5138 struct drm_i915_private *dev_priv = dev->dev_private;
5139
5140 dev_priv->display.init_clock_gating(dev);
6f1d69b0
ED
5141}
5142
7d708ee4
ID
5143void intel_suspend_hw(struct drm_device *dev)
5144{
5145 if (HAS_PCH_LPT(dev))
5146 lpt_suspend_hw(dev);
5147}
5148
c1ca727f
ID
5149#define for_each_power_well(i, power_well, domain_mask, power_domains) \
5150 for (i = 0; \
5151 i < (power_domains)->power_well_count && \
5152 ((power_well) = &(power_domains)->power_wells[i]); \
5153 i++) \
5154 if ((power_well)->domains & (domain_mask))
5155
5156#define for_each_power_well_rev(i, power_well, domain_mask, power_domains) \
5157 for (i = (power_domains)->power_well_count - 1; \
5158 i >= 0 && ((power_well) = &(power_domains)->power_wells[i]);\
5159 i--) \
5160 if ((power_well)->domains & (domain_mask))
5161
15d199ea
PZ
5162/**
5163 * We should only use the power well if we explicitly asked the hardware to
5164 * enable it, so check if it's enabled and also check if we've requested it to
5165 * be enabled.
5166 */
c1ca727f
ID
5167static bool hsw_power_well_enabled(struct drm_device *dev,
5168 struct i915_power_well *power_well)
5169{
5170 struct drm_i915_private *dev_priv = dev->dev_private;
5171
5172 return I915_READ(HSW_PWR_WELL_DRIVER) ==
5173 (HSW_PWR_WELL_ENABLE_REQUEST | HSW_PWR_WELL_STATE_ENABLED);
5174}
5175
ddf9c536
ID
5176bool intel_display_power_enabled_sw(struct drm_device *dev,
5177 enum intel_display_power_domain domain)
5178{
5179 struct drm_i915_private *dev_priv = dev->dev_private;
5180 struct i915_power_domains *power_domains;
5181
5182 power_domains = &dev_priv->power_domains;
5183
5184 return power_domains->domain_use_count[domain];
5185}
5186
b97186f0
PZ
5187bool intel_display_power_enabled(struct drm_device *dev,
5188 enum intel_display_power_domain domain)
15d199ea
PZ
5189{
5190 struct drm_i915_private *dev_priv = dev->dev_private;
c1ca727f
ID
5191 struct i915_power_domains *power_domains;
5192 struct i915_power_well *power_well;
5193 bool is_enabled;
5194 int i;
15d199ea 5195
c1ca727f
ID
5196 power_domains = &dev_priv->power_domains;
5197
5198 is_enabled = true;
5199
5200 mutex_lock(&power_domains->lock);
5201 for_each_power_well_rev(i, power_well, BIT(domain), power_domains) {
6f3ef5dd
ID
5202 if (power_well->always_on)
5203 continue;
5204
c1ca727f
ID
5205 if (!power_well->is_enabled(dev, power_well)) {
5206 is_enabled = false;
5207 break;
5208 }
5209 }
5210 mutex_unlock(&power_domains->lock);
5211
5212 return is_enabled;
15d199ea
PZ
5213}
5214
d5e8fdc8
PZ
5215static void hsw_power_well_post_enable(struct drm_i915_private *dev_priv)
5216{
5217 struct drm_device *dev = dev_priv->dev;
5218 unsigned long irqflags;
5219
f9dcb0df
PZ
5220 /*
5221 * After we re-enable the power well, if we touch VGA register 0x3d5
5222 * we'll get unclaimed register interrupts. This stops after we write
5223 * anything to the VGA MSR register. The vgacon module uses this
5224 * register all the time, so if we unbind our driver and, as a
5225 * consequence, bind vgacon, we'll get stuck in an infinite loop at
5226 * console_unlock(). So make here we touch the VGA MSR register, making
5227 * sure vgacon can keep working normally without triggering interrupts
5228 * and error messages.
5229 */
5230 vga_get_uninterruptible(dev->pdev, VGA_RSRC_LEGACY_IO);
5231 outb(inb(VGA_MSR_READ), VGA_MSR_WRITE);
5232 vga_put(dev->pdev, VGA_RSRC_LEGACY_IO);
5233
d5e8fdc8
PZ
5234 if (IS_BROADWELL(dev)) {
5235 spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
5236 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_B),
5237 dev_priv->de_irq_mask[PIPE_B]);
5238 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_B),
5239 ~dev_priv->de_irq_mask[PIPE_B] |
5240 GEN8_PIPE_VBLANK);
5241 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_C),
5242 dev_priv->de_irq_mask[PIPE_C]);
5243 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_C),
5244 ~dev_priv->de_irq_mask[PIPE_C] |
5245 GEN8_PIPE_VBLANK);
5246 POSTING_READ(GEN8_DE_PIPE_IER(PIPE_C));
5247 spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
5248 }
5249}
5250
5251static void hsw_power_well_post_disable(struct drm_i915_private *dev_priv)
5252{
5253 struct drm_device *dev = dev_priv->dev;
5254 enum pipe p;
5255 unsigned long irqflags;
5256
5257 /*
5258 * After this, the registers on the pipes that are part of the power
5259 * well will become zero, so we have to adjust our counters according to
5260 * that.
5261 *
5262 * FIXME: Should we do this in general in drm_vblank_post_modeset?
5263 */
5264 spin_lock_irqsave(&dev->vbl_lock, irqflags);
5265 for_each_pipe(p)
5266 if (p != PIPE_A)
5267 dev->vblank[p].last = 0;
5268 spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
5269}
5270
c1ca727f
ID
5271static void hsw_set_power_well(struct drm_device *dev,
5272 struct i915_power_well *power_well, bool enable)
d0d3e513
ED
5273{
5274 struct drm_i915_private *dev_priv = dev->dev_private;
fa42e23c
PZ
5275 bool is_enabled, enable_requested;
5276 uint32_t tmp;
d0d3e513 5277
d62292c8
PZ
5278 WARN_ON(dev_priv->pc8.enabled);
5279
fa42e23c 5280 tmp = I915_READ(HSW_PWR_WELL_DRIVER);
6aedd1f5
PZ
5281 is_enabled = tmp & HSW_PWR_WELL_STATE_ENABLED;
5282 enable_requested = tmp & HSW_PWR_WELL_ENABLE_REQUEST;
d0d3e513 5283
fa42e23c
PZ
5284 if (enable) {
5285 if (!enable_requested)
6aedd1f5
PZ
5286 I915_WRITE(HSW_PWR_WELL_DRIVER,
5287 HSW_PWR_WELL_ENABLE_REQUEST);
d0d3e513 5288
fa42e23c
PZ
5289 if (!is_enabled) {
5290 DRM_DEBUG_KMS("Enabling power well\n");
5291 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
6aedd1f5 5292 HSW_PWR_WELL_STATE_ENABLED), 20))
fa42e23c
PZ
5293 DRM_ERROR("Timeout enabling power well\n");
5294 }
596cc11e 5295
d5e8fdc8 5296 hsw_power_well_post_enable(dev_priv);
fa42e23c
PZ
5297 } else {
5298 if (enable_requested) {
5299 I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
9dbd8feb 5300 POSTING_READ(HSW_PWR_WELL_DRIVER);
fa42e23c 5301 DRM_DEBUG_KMS("Requesting to disable the power well\n");
9dbd8feb 5302
d5e8fdc8 5303 hsw_power_well_post_disable(dev_priv);
d0d3e513
ED
5304 }
5305 }
fa42e23c 5306}
d0d3e513 5307
b4ed4484
ID
5308static void __intel_power_well_get(struct drm_device *dev,
5309 struct i915_power_well *power_well)
2d66aef5 5310{
d62292c8
PZ
5311 struct drm_i915_private *dev_priv = dev->dev_private;
5312
5313 if (!power_well->count++ && power_well->set) {
5314 hsw_disable_package_c8(dev_priv);
c1ca727f 5315 power_well->set(dev, power_well, true);
d62292c8 5316 }
2d66aef5
VS
5317}
5318
b4ed4484
ID
5319static void __intel_power_well_put(struct drm_device *dev,
5320 struct i915_power_well *power_well)
2d66aef5 5321{
d62292c8
PZ
5322 struct drm_i915_private *dev_priv = dev->dev_private;
5323
2d66aef5 5324 WARN_ON(!power_well->count);
c1ca727f 5325
d62292c8 5326 if (!--power_well->count && power_well->set &&
d330a953 5327 i915.disable_power_well) {
c1ca727f 5328 power_well->set(dev, power_well, false);
d62292c8
PZ
5329 hsw_enable_package_c8(dev_priv);
5330 }
2d66aef5
VS
5331}
5332
6765625e
VS
5333void intel_display_power_get(struct drm_device *dev,
5334 enum intel_display_power_domain domain)
5335{
5336 struct drm_i915_private *dev_priv = dev->dev_private;
83c00f55 5337 struct i915_power_domains *power_domains;
c1ca727f
ID
5338 struct i915_power_well *power_well;
5339 int i;
6765625e 5340
83c00f55
ID
5341 power_domains = &dev_priv->power_domains;
5342
5343 mutex_lock(&power_domains->lock);
1da51581 5344
c1ca727f
ID
5345 for_each_power_well(i, power_well, BIT(domain), power_domains)
5346 __intel_power_well_get(dev, power_well);
1da51581 5347
ddf9c536
ID
5348 power_domains->domain_use_count[domain]++;
5349
83c00f55 5350 mutex_unlock(&power_domains->lock);
6765625e
VS
5351}
5352
5353void intel_display_power_put(struct drm_device *dev,
5354 enum intel_display_power_domain domain)
5355{
5356 struct drm_i915_private *dev_priv = dev->dev_private;
83c00f55 5357 struct i915_power_domains *power_domains;
c1ca727f
ID
5358 struct i915_power_well *power_well;
5359 int i;
6765625e 5360
83c00f55
ID
5361 power_domains = &dev_priv->power_domains;
5362
5363 mutex_lock(&power_domains->lock);
1da51581 5364
1da51581
ID
5365 WARN_ON(!power_domains->domain_use_count[domain]);
5366 power_domains->domain_use_count[domain]--;
ddf9c536
ID
5367
5368 for_each_power_well_rev(i, power_well, BIT(domain), power_domains)
5369 __intel_power_well_put(dev, power_well);
1da51581 5370
83c00f55 5371 mutex_unlock(&power_domains->lock);
6765625e
VS
5372}
5373
83c00f55 5374static struct i915_power_domains *hsw_pwr;
a38911a3
WX
5375
5376/* Display audio driver power well request */
5377void i915_request_power_well(void)
5378{
b4ed4484
ID
5379 struct drm_i915_private *dev_priv;
5380
a38911a3
WX
5381 if (WARN_ON(!hsw_pwr))
5382 return;
5383
b4ed4484
ID
5384 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
5385 power_domains);
fbeeaa23 5386 intel_display_power_get(dev_priv->dev, POWER_DOMAIN_AUDIO);
a38911a3
WX
5387}
5388EXPORT_SYMBOL_GPL(i915_request_power_well);
5389
5390/* Display audio driver power well release */
5391void i915_release_power_well(void)
5392{
b4ed4484
ID
5393 struct drm_i915_private *dev_priv;
5394
a38911a3
WX
5395 if (WARN_ON(!hsw_pwr))
5396 return;
5397
b4ed4484
ID
5398 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
5399 power_domains);
fbeeaa23 5400 intel_display_power_put(dev_priv->dev, POWER_DOMAIN_AUDIO);
a38911a3
WX
5401}
5402EXPORT_SYMBOL_GPL(i915_release_power_well);
5403
1c2256df
ID
5404static struct i915_power_well i9xx_always_on_power_well[] = {
5405 {
5406 .name = "always-on",
5407 .always_on = 1,
5408 .domains = POWER_DOMAIN_MASK,
5409 },
5410};
5411
c1ca727f 5412static struct i915_power_well hsw_power_wells[] = {
6f3ef5dd
ID
5413 {
5414 .name = "always-on",
5415 .always_on = 1,
5416 .domains = HSW_ALWAYS_ON_POWER_DOMAINS,
5417 },
c1ca727f
ID
5418 {
5419 .name = "display",
5420 .domains = POWER_DOMAIN_MASK & ~HSW_ALWAYS_ON_POWER_DOMAINS,
5421 .is_enabled = hsw_power_well_enabled,
5422 .set = hsw_set_power_well,
5423 },
5424};
5425
5426static struct i915_power_well bdw_power_wells[] = {
6f3ef5dd
ID
5427 {
5428 .name = "always-on",
5429 .always_on = 1,
5430 .domains = BDW_ALWAYS_ON_POWER_DOMAINS,
5431 },
c1ca727f
ID
5432 {
5433 .name = "display",
5434 .domains = POWER_DOMAIN_MASK & ~BDW_ALWAYS_ON_POWER_DOMAINS,
5435 .is_enabled = hsw_power_well_enabled,
5436 .set = hsw_set_power_well,
5437 },
5438};
5439
5440#define set_power_wells(power_domains, __power_wells) ({ \
5441 (power_domains)->power_wells = (__power_wells); \
5442 (power_domains)->power_well_count = ARRAY_SIZE(__power_wells); \
5443})
5444
ddb642fb 5445int intel_power_domains_init(struct drm_device *dev)
a38911a3
WX
5446{
5447 struct drm_i915_private *dev_priv = dev->dev_private;
83c00f55 5448 struct i915_power_domains *power_domains = &dev_priv->power_domains;
c1ca727f 5449
83c00f55 5450 mutex_init(&power_domains->lock);
a38911a3 5451
c1ca727f
ID
5452 /*
5453 * The enabling order will be from lower to higher indexed wells,
5454 * the disabling order is reversed.
5455 */
5456 if (IS_HASWELL(dev)) {
5457 set_power_wells(power_domains, hsw_power_wells);
5458 hsw_pwr = power_domains;
5459 } else if (IS_BROADWELL(dev)) {
5460 set_power_wells(power_domains, bdw_power_wells);
5461 hsw_pwr = power_domains;
5462 } else {
1c2256df 5463 set_power_wells(power_domains, i9xx_always_on_power_well);
c1ca727f 5464 }
a38911a3
WX
5465
5466 return 0;
5467}
5468
ddb642fb 5469void intel_power_domains_remove(struct drm_device *dev)
a38911a3
WX
5470{
5471 hsw_pwr = NULL;
5472}
5473
ddb642fb 5474static void intel_power_domains_resume(struct drm_device *dev)
9cdb826c
VS
5475{
5476 struct drm_i915_private *dev_priv = dev->dev_private;
83c00f55
ID
5477 struct i915_power_domains *power_domains = &dev_priv->power_domains;
5478 struct i915_power_well *power_well;
c1ca727f 5479 int i;
9cdb826c 5480
83c00f55 5481 mutex_lock(&power_domains->lock);
c1ca727f
ID
5482 for_each_power_well(i, power_well, POWER_DOMAIN_MASK, power_domains) {
5483 if (power_well->set)
5484 power_well->set(dev, power_well, power_well->count > 0);
5485 }
83c00f55 5486 mutex_unlock(&power_domains->lock);
a38911a3
WX
5487}
5488
fa42e23c
PZ
5489/*
5490 * Starting with Haswell, we have a "Power Down Well" that can be turned off
5491 * when not needed anymore. We have 4 registers that can request the power well
5492 * to be enabled, and it will only be disabled if none of the registers is
5493 * requesting it to be enabled.
d0d3e513 5494 */
ddb642fb 5495void intel_power_domains_init_hw(struct drm_device *dev)
d0d3e513
ED
5496{
5497 struct drm_i915_private *dev_priv = dev->dev_private;
d0d3e513 5498
fa42e23c 5499 /* For now, we need the power well to be always enabled. */
baa70707 5500 intel_display_set_init_power(dev, true);
ddb642fb 5501 intel_power_domains_resume(dev);
d0d3e513 5502
f7243ac9
ID
5503 if (!(IS_HASWELL(dev) || IS_BROADWELL(dev)))
5504 return;
5505
fa42e23c
PZ
5506 /* We're taking over the BIOS, so clear any requests made by it since
5507 * the driver is in charge now. */
6aedd1f5 5508 if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE_REQUEST)
fa42e23c 5509 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
d0d3e513
ED
5510}
5511
c67a470b
PZ
5512/* Disables PC8 so we can use the GMBUS and DP AUX interrupts. */
5513void intel_aux_display_runtime_get(struct drm_i915_private *dev_priv)
5514{
5515 hsw_disable_package_c8(dev_priv);
5516}
5517
5518void intel_aux_display_runtime_put(struct drm_i915_private *dev_priv)
5519{
5520 hsw_enable_package_c8(dev_priv);
5521}
5522
8a187455
PZ
5523void intel_runtime_pm_get(struct drm_i915_private *dev_priv)
5524{
5525 struct drm_device *dev = dev_priv->dev;
5526 struct device *device = &dev->pdev->dev;
5527
5528 if (!HAS_RUNTIME_PM(dev))
5529 return;
5530
5531 pm_runtime_get_sync(device);
5532 WARN(dev_priv->pm.suspended, "Device still suspended.\n");
5533}
5534
5535void intel_runtime_pm_put(struct drm_i915_private *dev_priv)
5536{
5537 struct drm_device *dev = dev_priv->dev;
5538 struct device *device = &dev->pdev->dev;
5539
5540 if (!HAS_RUNTIME_PM(dev))
5541 return;
5542
5543 pm_runtime_mark_last_busy(device);
5544 pm_runtime_put_autosuspend(device);
5545}
5546
5547void intel_init_runtime_pm(struct drm_i915_private *dev_priv)
5548{
5549 struct drm_device *dev = dev_priv->dev;
5550 struct device *device = &dev->pdev->dev;
5551
5552 dev_priv->pm.suspended = false;
5553
5554 if (!HAS_RUNTIME_PM(dev))
5555 return;
5556
5557 pm_runtime_set_active(device);
5558
5559 pm_runtime_set_autosuspend_delay(device, 10000); /* 10s */
5560 pm_runtime_mark_last_busy(device);
5561 pm_runtime_use_autosuspend(device);
5562}
5563
5564void intel_fini_runtime_pm(struct drm_i915_private *dev_priv)
5565{
5566 struct drm_device *dev = dev_priv->dev;
5567 struct device *device = &dev->pdev->dev;
5568
5569 if (!HAS_RUNTIME_PM(dev))
5570 return;
5571
5572 /* Make sure we're not suspended first. */
5573 pm_runtime_get_sync(device);
5574 pm_runtime_disable(device);
5575}
5576
1fa61106
ED
5577/* Set up chip specific power management-related functions */
5578void intel_init_pm(struct drm_device *dev)
5579{
5580 struct drm_i915_private *dev_priv = dev->dev_private;
5581
3a77c4c4 5582 if (HAS_FBC(dev)) {
40045465 5583 if (INTEL_INFO(dev)->gen >= 7) {
1fa61106 5584 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
40045465
VS
5585 dev_priv->display.enable_fbc = gen7_enable_fbc;
5586 dev_priv->display.disable_fbc = ironlake_disable_fbc;
5587 } else if (INTEL_INFO(dev)->gen >= 5) {
5588 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
5589 dev_priv->display.enable_fbc = ironlake_enable_fbc;
1fa61106
ED
5590 dev_priv->display.disable_fbc = ironlake_disable_fbc;
5591 } else if (IS_GM45(dev)) {
5592 dev_priv->display.fbc_enabled = g4x_fbc_enabled;
5593 dev_priv->display.enable_fbc = g4x_enable_fbc;
5594 dev_priv->display.disable_fbc = g4x_disable_fbc;
40045465 5595 } else {
1fa61106
ED
5596 dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
5597 dev_priv->display.enable_fbc = i8xx_enable_fbc;
5598 dev_priv->display.disable_fbc = i8xx_disable_fbc;
993495ae
VS
5599
5600 /* This value was pulled out of someone's hat */
5601 I915_WRITE(FBC_CONTROL, 500 << FBC_CTL_INTERVAL_SHIFT);
1fa61106 5602 }
1fa61106
ED
5603 }
5604
c921aba8
DV
5605 /* For cxsr */
5606 if (IS_PINEVIEW(dev))
5607 i915_pineview_get_mem_freq(dev);
5608 else if (IS_GEN5(dev))
5609 i915_ironlake_get_mem_freq(dev);
5610
1fa61106
ED
5611 /* For FIFO watermark updates */
5612 if (HAS_PCH_SPLIT(dev)) {
53615a5e
VS
5613 intel_setup_wm_latency(dev);
5614
bd602544
VS
5615 if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] &&
5616 dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
5617 (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] &&
5618 dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
5619 dev_priv->display.update_wm = ilk_update_wm;
5620 dev_priv->display.update_sprite_wm = ilk_update_sprite_wm;
5621 } else {
5622 DRM_DEBUG_KMS("Failed to read display plane latency. "
5623 "Disable CxSR\n");
5624 }
5625
5626 if (IS_GEN5(dev))
1fa61106 5627 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
bd602544 5628 else if (IS_GEN6(dev))
1fa61106 5629 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
bd602544 5630 else if (IS_IVYBRIDGE(dev))
1fa61106 5631 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
bd602544 5632 else if (IS_HASWELL(dev))
cad2a2d7 5633 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
bd602544 5634 else if (INTEL_INFO(dev)->gen == 8)
1020a5c2 5635 dev_priv->display.init_clock_gating = gen8_init_clock_gating;
1fa61106
ED
5636 } else if (IS_VALLEYVIEW(dev)) {
5637 dev_priv->display.update_wm = valleyview_update_wm;
5638 dev_priv->display.init_clock_gating =
5639 valleyview_init_clock_gating;
1fa61106
ED
5640 } else if (IS_PINEVIEW(dev)) {
5641 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
5642 dev_priv->is_ddr3,
5643 dev_priv->fsb_freq,
5644 dev_priv->mem_freq)) {
5645 DRM_INFO("failed to find known CxSR latency "
5646 "(found ddr%s fsb freq %d, mem freq %d), "
5647 "disabling CxSR\n",
5648 (dev_priv->is_ddr3 == 1) ? "3" : "2",
5649 dev_priv->fsb_freq, dev_priv->mem_freq);
5650 /* Disable CxSR and never update its watermark again */
5651 pineview_disable_cxsr(dev);
5652 dev_priv->display.update_wm = NULL;
5653 } else
5654 dev_priv->display.update_wm = pineview_update_wm;
5655 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
5656 } else if (IS_G4X(dev)) {
5657 dev_priv->display.update_wm = g4x_update_wm;
5658 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
5659 } else if (IS_GEN4(dev)) {
5660 dev_priv->display.update_wm = i965_update_wm;
5661 if (IS_CRESTLINE(dev))
5662 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
5663 else if (IS_BROADWATER(dev))
5664 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
5665 } else if (IS_GEN3(dev)) {
5666 dev_priv->display.update_wm = i9xx_update_wm;
5667 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
5668 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
feb56b93
DV
5669 } else if (IS_GEN2(dev)) {
5670 if (INTEL_INFO(dev)->num_pipes == 1) {
5671 dev_priv->display.update_wm = i845_update_wm;
1fa61106 5672 dev_priv->display.get_fifo_size = i845_get_fifo_size;
feb56b93
DV
5673 } else {
5674 dev_priv->display.update_wm = i9xx_update_wm;
1fa61106 5675 dev_priv->display.get_fifo_size = i830_get_fifo_size;
feb56b93
DV
5676 }
5677
5678 if (IS_I85X(dev) || IS_I865G(dev))
5679 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
5680 else
5681 dev_priv->display.init_clock_gating = i830_init_clock_gating;
5682 } else {
5683 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
1fa61106
ED
5684 }
5685}
5686
42c0526c
BW
5687int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
5688{
4fc688ce 5689 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW
5690
5691 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
5692 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
5693 return -EAGAIN;
5694 }
5695
5696 I915_WRITE(GEN6_PCODE_DATA, *val);
5697 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
5698
5699 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
5700 500)) {
5701 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
5702 return -ETIMEDOUT;
5703 }
5704
5705 *val = I915_READ(GEN6_PCODE_DATA);
5706 I915_WRITE(GEN6_PCODE_DATA, 0);
5707
5708 return 0;
5709}
5710
5711int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
5712{
4fc688ce 5713 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
42c0526c
BW
5714
5715 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
5716 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
5717 return -EAGAIN;
5718 }
5719
5720 I915_WRITE(GEN6_PCODE_DATA, val);
5721 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
5722
5723 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
5724 500)) {
5725 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
5726 return -ETIMEDOUT;
5727 }
5728
5729 I915_WRITE(GEN6_PCODE_DATA, 0);
5730
5731 return 0;
5732}
a0e4e199 5733
2ec3815f 5734int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val)
855ba3be 5735{
07ab118b 5736 int div;
855ba3be 5737
07ab118b 5738 /* 4 x czclk */
2ec3815f 5739 switch (dev_priv->mem_freq) {
855ba3be 5740 case 800:
07ab118b 5741 div = 10;
855ba3be
JB
5742 break;
5743 case 1066:
07ab118b 5744 div = 12;
855ba3be
JB
5745 break;
5746 case 1333:
07ab118b 5747 div = 16;
855ba3be
JB
5748 break;
5749 default:
5750 return -1;
5751 }
5752
2ec3815f 5753 return DIV_ROUND_CLOSEST(dev_priv->mem_freq * (val + 6 - 0xbd), 4 * div);
855ba3be
JB
5754}
5755
2ec3815f 5756int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val)
855ba3be 5757{
07ab118b 5758 int mul;
855ba3be 5759
07ab118b 5760 /* 4 x czclk */
2ec3815f 5761 switch (dev_priv->mem_freq) {
855ba3be 5762 case 800:
07ab118b 5763 mul = 10;
855ba3be
JB
5764 break;
5765 case 1066:
07ab118b 5766 mul = 12;
855ba3be
JB
5767 break;
5768 case 1333:
07ab118b 5769 mul = 16;
855ba3be
JB
5770 break;
5771 default:
5772 return -1;
5773 }
5774
2ec3815f 5775 return DIV_ROUND_CLOSEST(4 * mul * val, dev_priv->mem_freq) + 0xbd - 6;
855ba3be
JB
5776}
5777
f742a552 5778void intel_pm_setup(struct drm_device *dev)
907b28c5
CW
5779{
5780 struct drm_i915_private *dev_priv = dev->dev_private;
5781
f742a552
DV
5782 mutex_init(&dev_priv->rps.hw_lock);
5783
5784 mutex_init(&dev_priv->pc8.lock);
5785 dev_priv->pc8.requirements_met = false;
5786 dev_priv->pc8.gpu_idle = false;
5787 dev_priv->pc8.irqs_disabled = false;
5788 dev_priv->pc8.enabled = false;
5789 dev_priv->pc8.disable_count = 2; /* requirements_met + gpu_idle */
5790 INIT_DELAYED_WORK(&dev_priv->pc8.enable_work, hsw_enable_pc8_work);
907b28c5
CW
5791 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
5792 intel_gen6_powersave_work);
5793}