projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
drm/i915: add audio power domain
[deliverable/linux.git] / drivers / gpu / drm / i915 / intel_pm.c
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
28 #include <linux/cpufreq.h>
29 #include "i915_drv.h"
30 #include "intel_drv.h"
31 #include "../../../platform/x86/intel_ips.h"
32 #include <linux/module.h>
33 #include <drm/i915_powerwell.h>
34
35 /**
36 * RC6 is a special power stage which allows the GPU to enter an very
37 * low-voltage mode when idle, using down to 0V while at this stage. This
38 * stage is entered automatically when the GPU is idle when RC6 support is
39 * enabled, and as soon as new workload arises GPU wakes up automatically as well.
40 *
41 * There are different RC6 modes available in Intel GPU, which differentiate
42 * among each other with the latency required to enter and leave RC6 and
43 * voltage consumed by the GPU in different states.
44 *
45 * The combination of the following flags define which states GPU is allowed
46 * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
47 * RC6pp is deepest RC6. Their support by hardware varies according to the
48 * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
49 * which brings the most power savings; deeper states save more power, but
50 * require higher latency to switch to and wake up.
51 */
52 #define INTEL_RC6_ENABLE (1<<0)
53 #define INTEL_RC6p_ENABLE (1<<1)
54 #define INTEL_RC6pp_ENABLE (1<<2)
55
56 /* FBC, or Frame Buffer Compression, is a technique employed to compress the
57 * framebuffer contents in-memory, aiming at reducing the required bandwidth
58 * during in-memory transfers and, therefore, reduce the power packet.
59 *
60 * The benefits of FBC are mostly visible with solid backgrounds and
61 * variation-less patterns.
62 *
63 * FBC-related functionality can be enabled by the means of the
64 * i915.i915_enable_fbc parameter
65 */
66
67 static void i8xx_disable_fbc(struct drm_device *dev)
68 {
69 struct drm_i915_private *dev_priv = dev->dev_private;
70 u32 fbc_ctl;
71
72 /* Disable compression */
73 fbc_ctl = I915_READ(FBC_CONTROL);
74 if ((fbc_ctl & FBC_CTL_EN) == 0)
75 return;
76
77 fbc_ctl &= ~FBC_CTL_EN;
78 I915_WRITE(FBC_CONTROL, fbc_ctl);
79
80 /* Wait for compressing bit to clear */
81 if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) {
82 DRM_DEBUG_KMS("FBC idle timed out\n");
83 return;
84 }
85
86 DRM_DEBUG_KMS("disabled FBC\n");
87 }
88
89 static void i8xx_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
90 {
91 struct drm_device *dev = crtc->dev;
92 struct drm_i915_private *dev_priv = dev->dev_private;
93 struct drm_framebuffer *fb = crtc->fb;
94 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
95 struct drm_i915_gem_object *obj = intel_fb->obj;
96 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
97 int cfb_pitch;
98 int plane, i;
99 u32 fbc_ctl, fbc_ctl2;
100
101 cfb_pitch = dev_priv->fbc.size / FBC_LL_SIZE;
102 if (fb->pitches[0] < cfb_pitch)
103 cfb_pitch = fb->pitches[0];
104
105 /* FBC_CTL wants 64B units */
106 cfb_pitch = (cfb_pitch / 64) - 1;
107 plane = intel_crtc->plane == 0 ? FBC_CTL_PLANEA : FBC_CTL_PLANEB;
108
109 /* Clear old tags */
110 for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++)
111 I915_WRITE(FBC_TAG + (i * 4), 0);
112
113 /* Set it up... */
114 fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE;
115 fbc_ctl2 |= plane;
116 I915_WRITE(FBC_CONTROL2, fbc_ctl2);
117 I915_WRITE(FBC_FENCE_OFF, crtc->y);
118
119 /* enable it... */
120 fbc_ctl = FBC_CTL_EN | FBC_CTL_PERIODIC;
121 if (IS_I945GM(dev))
122 fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */
123 fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT;
124 fbc_ctl |= (interval & 0x2fff) << FBC_CTL_INTERVAL_SHIFT;
125 fbc_ctl |= obj->fence_reg;
126 I915_WRITE(FBC_CONTROL, fbc_ctl);
127
128 DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c, ",
129 cfb_pitch, crtc->y, plane_name(intel_crtc->plane));
130 }
131
132 static bool i8xx_fbc_enabled(struct drm_device *dev)
133 {
134 struct drm_i915_private *dev_priv = dev->dev_private;
135
136 return I915_READ(FBC_CONTROL) & FBC_CTL_EN;
137 }
138
139 static void g4x_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
140 {
141 struct drm_device *dev = crtc->dev;
142 struct drm_i915_private *dev_priv = dev->dev_private;
143 struct drm_framebuffer *fb = crtc->fb;
144 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
145 struct drm_i915_gem_object *obj = intel_fb->obj;
146 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
147 int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
148 unsigned long stall_watermark = 200;
149 u32 dpfc_ctl;
150
151 dpfc_ctl = plane | DPFC_SR_EN | DPFC_CTL_LIMIT_1X;
152 dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg;
153 I915_WRITE(DPFC_CHICKEN, DPFC_HT_MODIFY);
154
155 I915_WRITE(DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
156 (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
157 (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
158 I915_WRITE(DPFC_FENCE_YOFF, crtc->y);
159
160 /* enable it... */
161 I915_WRITE(DPFC_CONTROL, I915_READ(DPFC_CONTROL) | DPFC_CTL_EN);
162
163 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
164 }
165
166 static void g4x_disable_fbc(struct drm_device *dev)
167 {
168 struct drm_i915_private *dev_priv = dev->dev_private;
169 u32 dpfc_ctl;
170
171 /* Disable compression */
172 dpfc_ctl = I915_READ(DPFC_CONTROL);
173 if (dpfc_ctl & DPFC_CTL_EN) {
174 dpfc_ctl &= ~DPFC_CTL_EN;
175 I915_WRITE(DPFC_CONTROL, dpfc_ctl);
176
177 DRM_DEBUG_KMS("disabled FBC\n");
178 }
179 }
180
181 static bool g4x_fbc_enabled(struct drm_device *dev)
182 {
183 struct drm_i915_private *dev_priv = dev->dev_private;
184
185 return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN;
186 }
187
188 static void sandybridge_blit_fbc_update(struct drm_device *dev)
189 {
190 struct drm_i915_private *dev_priv = dev->dev_private;
191 u32 blt_ecoskpd;
192
193 /* Make sure blitter notifies FBC of writes */
194 gen6_gt_force_wake_get(dev_priv);
195 blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD);
196 blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY <<
197 GEN6_BLITTER_LOCK_SHIFT;
198 I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
199 blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY;
200 I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
201 blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY <<
202 GEN6_BLITTER_LOCK_SHIFT);
203 I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
204 POSTING_READ(GEN6_BLITTER_ECOSKPD);
205 gen6_gt_force_wake_put(dev_priv);
206 }
207
208 static void ironlake_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
209 {
210 struct drm_device *dev = crtc->dev;
211 struct drm_i915_private *dev_priv = dev->dev_private;
212 struct drm_framebuffer *fb = crtc->fb;
213 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
214 struct drm_i915_gem_object *obj = intel_fb->obj;
215 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
216 int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
217 unsigned long stall_watermark = 200;
218 u32 dpfc_ctl;
219
220 dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
221 dpfc_ctl &= DPFC_RESERVED;
222 dpfc_ctl |= (plane | DPFC_CTL_LIMIT_1X);
223 /* Set persistent mode for front-buffer rendering, ala X. */
224 dpfc_ctl |= DPFC_CTL_PERSISTENT_MODE;
225 dpfc_ctl |= DPFC_CTL_FENCE_EN;
226 if (IS_GEN5(dev))
227 dpfc_ctl |= obj->fence_reg;
228 I915_WRITE(ILK_DPFC_CHICKEN, DPFC_HT_MODIFY);
229
230 I915_WRITE(ILK_DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
231 (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
232 (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
233 I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
234 I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID);
235 /* enable it... */
236 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
237
238 if (IS_GEN6(dev)) {
239 I915_WRITE(SNB_DPFC_CTL_SA,
240 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
241 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
242 sandybridge_blit_fbc_update(dev);
243 }
244
245 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
246 }
247
248 static void ironlake_disable_fbc(struct drm_device *dev)
249 {
250 struct drm_i915_private *dev_priv = dev->dev_private;
251 u32 dpfc_ctl;
252
253 /* Disable compression */
254 dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
255 if (dpfc_ctl & DPFC_CTL_EN) {
256 dpfc_ctl &= ~DPFC_CTL_EN;
257 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl);
258
259 DRM_DEBUG_KMS("disabled FBC\n");
260 }
261 }
262
263 static bool ironlake_fbc_enabled(struct drm_device *dev)
264 {
265 struct drm_i915_private *dev_priv = dev->dev_private;
266
267 return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
268 }
269
270 static void gen7_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
271 {
272 struct drm_device *dev = crtc->dev;
273 struct drm_i915_private *dev_priv = dev->dev_private;
274 struct drm_framebuffer *fb = crtc->fb;
275 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
276 struct drm_i915_gem_object *obj = intel_fb->obj;
277 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
278
279 I915_WRITE(IVB_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj));
280
281 I915_WRITE(ILK_DPFC_CONTROL, DPFC_CTL_EN | DPFC_CTL_LIMIT_1X |
282 IVB_DPFC_CTL_FENCE_EN |
283 intel_crtc->plane << IVB_DPFC_CTL_PLANE_SHIFT);
284
285 if (IS_IVYBRIDGE(dev)) {
286 /* WaFbcAsynchFlipDisableFbcQueue:ivb */
287 I915_WRITE(ILK_DISPLAY_CHICKEN1, ILK_FBCQ_DIS);
288 } else {
289 /* WaFbcAsynchFlipDisableFbcQueue:hsw */
290 I915_WRITE(HSW_PIPE_SLICE_CHICKEN_1(intel_crtc->pipe),
291 HSW_BYPASS_FBC_QUEUE);
292 }
293
294 I915_WRITE(SNB_DPFC_CTL_SA,
295 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
296 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
297
298 sandybridge_blit_fbc_update(dev);
299
300 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
301 }
302
303 bool intel_fbc_enabled(struct drm_device *dev)
304 {
305 struct drm_i915_private *dev_priv = dev->dev_private;
306
307 if (!dev_priv->display.fbc_enabled)
308 return false;
309
310 return dev_priv->display.fbc_enabled(dev);
311 }
312
313 static void intel_fbc_work_fn(struct work_struct *__work)
314 {
315 struct intel_fbc_work *work =
316 container_of(to_delayed_work(__work),
317 struct intel_fbc_work, work);
318 struct drm_device *dev = work->crtc->dev;
319 struct drm_i915_private *dev_priv = dev->dev_private;
320
321 mutex_lock(&dev->struct_mutex);
322 if (work == dev_priv->fbc.fbc_work) {
323 /* Double check that we haven't switched fb without cancelling
324 * the prior work.
325 */
326 if (work->crtc->fb == work->fb) {
327 dev_priv->display.enable_fbc(work->crtc,
328 work->interval);
329
330 dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane;
331 dev_priv->fbc.fb_id = work->crtc->fb->base.id;
332 dev_priv->fbc.y = work->crtc->y;
333 }
334
335 dev_priv->fbc.fbc_work = NULL;
336 }
337 mutex_unlock(&dev->struct_mutex);
338
339 kfree(work);
340 }
341
342 static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
343 {
344 if (dev_priv->fbc.fbc_work == NULL)
345 return;
346
347 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
348
349 /* Synchronisation is provided by struct_mutex and checking of
350 * dev_priv->fbc.fbc_work, so we can perform the cancellation
351 * entirely asynchronously.
352 */
353 if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work))
354 /* tasklet was killed before being run, clean up */
355 kfree(dev_priv->fbc.fbc_work);
356
357 /* Mark the work as no longer wanted so that if it does
358 * wake-up (because the work was already running and waiting
359 * for our mutex), it will discover that is no longer
360 * necessary to run.
361 */
362 dev_priv->fbc.fbc_work = NULL;
363 }
364
365 static void intel_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
366 {
367 struct intel_fbc_work *work;
368 struct drm_device *dev = crtc->dev;
369 struct drm_i915_private *dev_priv = dev->dev_private;
370
371 if (!dev_priv->display.enable_fbc)
372 return;
373
374 intel_cancel_fbc_work(dev_priv);
375
376 work = kzalloc(sizeof(*work), GFP_KERNEL);
377 if (work == NULL) {
378 DRM_ERROR("Failed to allocate FBC work structure\n");
379 dev_priv->display.enable_fbc(crtc, interval);
380 return;
381 }
382
383 work->crtc = crtc;
384 work->fb = crtc->fb;
385 work->interval = interval;
386 INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
387
388 dev_priv->fbc.fbc_work = work;
389
390 /* Delay the actual enabling to let pageflipping cease and the
391 * display to settle before starting the compression. Note that
392 * this delay also serves a second purpose: it allows for a
393 * vblank to pass after disabling the FBC before we attempt
394 * to modify the control registers.
395 *
396 * A more complicated solution would involve tracking vblanks
397 * following the termination of the page-flipping sequence
398 * and indeed performing the enable as a co-routine and not
399 * waiting synchronously upon the vblank.
400 *
401 * WaFbcWaitForVBlankBeforeEnable:ilk,snb
402 */
403 schedule_delayed_work(&work->work, msecs_to_jiffies(50));
404 }
405
406 void intel_disable_fbc(struct drm_device *dev)
407 {
408 struct drm_i915_private *dev_priv = dev->dev_private;
409
410 intel_cancel_fbc_work(dev_priv);
411
412 if (!dev_priv->display.disable_fbc)
413 return;
414
415 dev_priv->display.disable_fbc(dev);
416 dev_priv->fbc.plane = -1;
417 }
418
419 static bool set_no_fbc_reason(struct drm_i915_private *dev_priv,
420 enum no_fbc_reason reason)
421 {
422 if (dev_priv->fbc.no_fbc_reason == reason)
423 return false;
424
425 dev_priv->fbc.no_fbc_reason = reason;
426 return true;
427 }
428
429 /**
430 * intel_update_fbc - enable/disable FBC as needed
431 * @dev: the drm_device
432 *
433 * Set up the framebuffer compression hardware at mode set time. We
434 * enable it if possible:
435 * - plane A only (on pre-965)
436 * - no pixel mulitply/line duplication
437 * - no alpha buffer discard
438 * - no dual wide
439 * - framebuffer <= max_hdisplay in width, max_vdisplay in height
440 *
441 * We can't assume that any compression will take place (worst case),
442 * so the compressed buffer has to be the same size as the uncompressed
443 * one. It also must reside (along with the line length buffer) in
444 * stolen memory.
445 *
446 * We need to enable/disable FBC on a global basis.
447 */
448 void intel_update_fbc(struct drm_device *dev)
449 {
450 struct drm_i915_private *dev_priv = dev->dev_private;
451 struct drm_crtc *crtc = NULL, *tmp_crtc;
452 struct intel_crtc *intel_crtc;
453 struct drm_framebuffer *fb;
454 struct intel_framebuffer *intel_fb;
455 struct drm_i915_gem_object *obj;
456 const struct drm_display_mode *adjusted_mode;
457 unsigned int max_width, max_height;
458
459 if (!I915_HAS_FBC(dev)) {
460 set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED);
461 return;
462 }
463
464 if (!i915_powersave) {
465 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
466 DRM_DEBUG_KMS("fbc disabled per module param\n");
467 return;
468 }
469
470 /*
471 * If FBC is already on, we just have to verify that we can
472 * keep it that way...
473 * Need to disable if:
474 * - more than one pipe is active
475 * - changing FBC params (stride, fence, mode)
476 * - new fb is too large to fit in compressed buffer
477 * - going to an unsupported config (interlace, pixel multiply, etc.)
478 */
479 list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) {
480 if (intel_crtc_active(tmp_crtc) &&
481 to_intel_crtc(tmp_crtc)->primary_enabled) {
482 if (crtc) {
483 if (set_no_fbc_reason(dev_priv, FBC_MULTIPLE_PIPES))
484 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
485 goto out_disable;
486 }
487 crtc = tmp_crtc;
488 }
489 }
490
491 if (!crtc || crtc->fb == NULL) {
492 if (set_no_fbc_reason(dev_priv, FBC_NO_OUTPUT))
493 DRM_DEBUG_KMS("no output, disabling\n");
494 goto out_disable;
495 }
496
497 intel_crtc = to_intel_crtc(crtc);
498 fb = crtc->fb;
499 intel_fb = to_intel_framebuffer(fb);
500 obj = intel_fb->obj;
501 adjusted_mode = &intel_crtc->config.adjusted_mode;
502
503 if (i915_enable_fbc < 0 &&
504 INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev)) {
505 if (set_no_fbc_reason(dev_priv, FBC_CHIP_DEFAULT))
506 DRM_DEBUG_KMS("disabled per chip default\n");
507 goto out_disable;
508 }
509 if (!i915_enable_fbc) {
510 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
511 DRM_DEBUG_KMS("fbc disabled per module param\n");
512 goto out_disable;
513 }
514 if ((adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) ||
515 (adjusted_mode->flags & DRM_MODE_FLAG_DBLSCAN)) {
516 if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE))
517 DRM_DEBUG_KMS("mode incompatible with compression, "
518 "disabling\n");
519 goto out_disable;
520 }
521
522 if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) {
523 max_width = 4096;
524 max_height = 2048;
525 } else {
526 max_width = 2048;
527 max_height = 1536;
528 }
529 if (intel_crtc->config.pipe_src_w > max_width ||
530 intel_crtc->config.pipe_src_h > max_height) {
531 if (set_no_fbc_reason(dev_priv, FBC_MODE_TOO_LARGE))
532 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
533 goto out_disable;
534 }
535 if ((IS_I915GM(dev) || IS_I945GM(dev) || IS_HASWELL(dev)) &&
536 intel_crtc->plane != 0) {
537 if (set_no_fbc_reason(dev_priv, FBC_BAD_PLANE))
538 DRM_DEBUG_KMS("plane not 0, disabling compression\n");
539 goto out_disable;
540 }
541
542 /* The use of a CPU fence is mandatory in order to detect writes
543 * by the CPU to the scanout and trigger updates to the FBC.
544 */
545 if (obj->tiling_mode != I915_TILING_X ||
546 obj->fence_reg == I915_FENCE_REG_NONE) {
547 if (set_no_fbc_reason(dev_priv, FBC_NOT_TILED))
548 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
549 goto out_disable;
550 }
551
552 /* If the kernel debugger is active, always disable compression */
553 if (in_dbg_master())
554 goto out_disable;
555
556 if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) {
557 if (set_no_fbc_reason(dev_priv, FBC_STOLEN_TOO_SMALL))
558 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
559 goto out_disable;
560 }
561
562 /* If the scanout has not changed, don't modify the FBC settings.
563 * Note that we make the fundamental assumption that the fb->obj
564 * cannot be unpinned (and have its GTT offset and fence revoked)
565 * without first being decoupled from the scanout and FBC disabled.
566 */
567 if (dev_priv->fbc.plane == intel_crtc->plane &&
568 dev_priv->fbc.fb_id == fb->base.id &&
569 dev_priv->fbc.y == crtc->y)
570 return;
571
572 if (intel_fbc_enabled(dev)) {
573 /* We update FBC along two paths, after changing fb/crtc
574 * configuration (modeswitching) and after page-flipping
575 * finishes. For the latter, we know that not only did
576 * we disable the FBC at the start of the page-flip
577 * sequence, but also more than one vblank has passed.
578 *
579 * For the former case of modeswitching, it is possible
580 * to switch between two FBC valid configurations
581 * instantaneously so we do need to disable the FBC
582 * before we can modify its control registers. We also
583 * have to wait for the next vblank for that to take
584 * effect. However, since we delay enabling FBC we can
585 * assume that a vblank has passed since disabling and
586 * that we can safely alter the registers in the deferred
587 * callback.
588 *
589 * In the scenario that we go from a valid to invalid
590 * and then back to valid FBC configuration we have
591 * no strict enforcement that a vblank occurred since
592 * disabling the FBC. However, along all current pipe
593 * disabling paths we do need to wait for a vblank at
594 * some point. And we wait before enabling FBC anyway.
595 */
596 DRM_DEBUG_KMS("disabling active FBC for update\n");
597 intel_disable_fbc(dev);
598 }
599
600 intel_enable_fbc(crtc, 500);
601 dev_priv->fbc.no_fbc_reason = FBC_OK;
602 return;
603
604 out_disable:
605 /* Multiple disables should be harmless */
606 if (intel_fbc_enabled(dev)) {
607 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
608 intel_disable_fbc(dev);
609 }
610 i915_gem_stolen_cleanup_compression(dev);
611 }
612
613 static void i915_pineview_get_mem_freq(struct drm_device *dev)
614 {
615 drm_i915_private_t *dev_priv = dev->dev_private;
616 u32 tmp;
617
618 tmp = I915_READ(CLKCFG);
619
620 switch (tmp & CLKCFG_FSB_MASK) {
621 case CLKCFG_FSB_533:
622 dev_priv->fsb_freq = 533; /* 133*4 */
623 break;
624 case CLKCFG_FSB_800:
625 dev_priv->fsb_freq = 800; /* 200*4 */
626 break;
627 case CLKCFG_FSB_667:
628 dev_priv->fsb_freq = 667; /* 167*4 */
629 break;
630 case CLKCFG_FSB_400:
631 dev_priv->fsb_freq = 400; /* 100*4 */
632 break;
633 }
634
635 switch (tmp & CLKCFG_MEM_MASK) {
636 case CLKCFG_MEM_533:
637 dev_priv->mem_freq = 533;
638 break;
639 case CLKCFG_MEM_667:
640 dev_priv->mem_freq = 667;
641 break;
642 case CLKCFG_MEM_800:
643 dev_priv->mem_freq = 800;
644 break;
645 }
646
647 /* detect pineview DDR3 setting */
648 tmp = I915_READ(CSHRDDR3CTL);
649 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
650 }
651
652 static void i915_ironlake_get_mem_freq(struct drm_device *dev)
653 {
654 drm_i915_private_t *dev_priv = dev->dev_private;
655 u16 ddrpll, csipll;
656
657 ddrpll = I915_READ16(DDRMPLL1);
658 csipll = I915_READ16(CSIPLL0);
659
660 switch (ddrpll & 0xff) {
661 case 0xc:
662 dev_priv->mem_freq = 800;
663 break;
664 case 0x10:
665 dev_priv->mem_freq = 1066;
666 break;
667 case 0x14:
668 dev_priv->mem_freq = 1333;
669 break;
670 case 0x18:
671 dev_priv->mem_freq = 1600;
672 break;
673 default:
674 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
675 ddrpll & 0xff);
676 dev_priv->mem_freq = 0;
677 break;
678 }
679
680 dev_priv->ips.r_t = dev_priv->mem_freq;
681
682 switch (csipll & 0x3ff) {
683 case 0x00c:
684 dev_priv->fsb_freq = 3200;
685 break;
686 case 0x00e:
687 dev_priv->fsb_freq = 3733;
688 break;
689 case 0x010:
690 dev_priv->fsb_freq = 4266;
691 break;
692 case 0x012:
693 dev_priv->fsb_freq = 4800;
694 break;
695 case 0x014:
696 dev_priv->fsb_freq = 5333;
697 break;
698 case 0x016:
699 dev_priv->fsb_freq = 5866;
700 break;
701 case 0x018:
702 dev_priv->fsb_freq = 6400;
703 break;
704 default:
705 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
706 csipll & 0x3ff);
707 dev_priv->fsb_freq = 0;
708 break;
709 }
710
711 if (dev_priv->fsb_freq == 3200) {
712 dev_priv->ips.c_m = 0;
713 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
714 dev_priv->ips.c_m = 1;
715 } else {
716 dev_priv->ips.c_m = 2;
717 }
718 }
719
720 static const struct cxsr_latency cxsr_latency_table[] = {
721 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
722 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
723 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
724 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
725 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
726
727 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
728 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
729 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
730 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
731 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
732
733 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
734 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
735 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
736 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
737 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
738
739 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
740 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
741 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
742 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
743 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
744
745 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
746 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
747 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
748 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
749 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
750
751 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
752 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
753 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
754 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
755 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
756 };
757
758 static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
759 int is_ddr3,
760 int fsb,
761 int mem)
762 {
763 const struct cxsr_latency *latency;
764 int i;
765
766 if (fsb == 0 || mem == 0)
767 return NULL;
768
769 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
770 latency = &cxsr_latency_table[i];
771 if (is_desktop == latency->is_desktop &&
772 is_ddr3 == latency->is_ddr3 &&
773 fsb == latency->fsb_freq && mem == latency->mem_freq)
774 return latency;
775 }
776
777 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
778
779 return NULL;
780 }
781
782 static void pineview_disable_cxsr(struct drm_device *dev)
783 {
784 struct drm_i915_private *dev_priv = dev->dev_private;
785
786 /* deactivate cxsr */
787 I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
788 }
789
790 /*
791 * Latency for FIFO fetches is dependent on several factors:
792 * - memory configuration (speed, channels)
793 * - chipset
794 * - current MCH state
795 * It can be fairly high in some situations, so here we assume a fairly
796 * pessimal value. It's a tradeoff between extra memory fetches (if we
797 * set this value too high, the FIFO will fetch frequently to stay full)
798 * and power consumption (set it too low to save power and we might see
799 * FIFO underruns and display "flicker").
800 *
801 * A value of 5us seems to be a good balance; safe for very low end
802 * platforms but not overly aggressive on lower latency configs.
803 */
804 static const int latency_ns = 5000;
805
806 static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
807 {
808 struct drm_i915_private *dev_priv = dev->dev_private;
809 uint32_t dsparb = I915_READ(DSPARB);
810 int size;
811
812 size = dsparb & 0x7f;
813 if (plane)
814 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
815
816 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
817 plane ? "B" : "A", size);
818
819 return size;
820 }
821
822 static int i85x_get_fifo_size(struct drm_device *dev, int plane)
823 {
824 struct drm_i915_private *dev_priv = dev->dev_private;
825 uint32_t dsparb = I915_READ(DSPARB);
826 int size;
827
828 size = dsparb & 0x1ff;
829 if (plane)
830 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
831 size >>= 1; /* Convert to cachelines */
832
833 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
834 plane ? "B" : "A", size);
835
836 return size;
837 }
838
839 static int i845_get_fifo_size(struct drm_device *dev, int plane)
840 {
841 struct drm_i915_private *dev_priv = dev->dev_private;
842 uint32_t dsparb = I915_READ(DSPARB);
843 int size;
844
845 size = dsparb & 0x7f;
846 size >>= 2; /* Convert to cachelines */
847
848 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
849 plane ? "B" : "A",
850 size);
851
852 return size;
853 }
854
855 static int i830_get_fifo_size(struct drm_device *dev, int plane)
856 {
857 struct drm_i915_private *dev_priv = dev->dev_private;
858 uint32_t dsparb = I915_READ(DSPARB);
859 int size;
860
861 size = dsparb & 0x7f;
862 size >>= 1; /* Convert to cachelines */
863
864 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
865 plane ? "B" : "A", size);
866
867 return size;
868 }
869
870 /* Pineview has different values for various configs */
871 static const struct intel_watermark_params pineview_display_wm = {
872 PINEVIEW_DISPLAY_FIFO,
873 PINEVIEW_MAX_WM,
874 PINEVIEW_DFT_WM,
875 PINEVIEW_GUARD_WM,
876 PINEVIEW_FIFO_LINE_SIZE
877 };
878 static const struct intel_watermark_params pineview_display_hplloff_wm = {
879 PINEVIEW_DISPLAY_FIFO,
880 PINEVIEW_MAX_WM,
881 PINEVIEW_DFT_HPLLOFF_WM,
882 PINEVIEW_GUARD_WM,
883 PINEVIEW_FIFO_LINE_SIZE
884 };
885 static const struct intel_watermark_params pineview_cursor_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 };
892 static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
893 PINEVIEW_CURSOR_FIFO,
894 PINEVIEW_CURSOR_MAX_WM,
895 PINEVIEW_CURSOR_DFT_WM,
896 PINEVIEW_CURSOR_GUARD_WM,
897 PINEVIEW_FIFO_LINE_SIZE
898 };
899 static const struct intel_watermark_params g4x_wm_info = {
900 G4X_FIFO_SIZE,
901 G4X_MAX_WM,
902 G4X_MAX_WM,
903 2,
904 G4X_FIFO_LINE_SIZE,
905 };
906 static const struct intel_watermark_params g4x_cursor_wm_info = {
907 I965_CURSOR_FIFO,
908 I965_CURSOR_MAX_WM,
909 I965_CURSOR_DFT_WM,
910 2,
911 G4X_FIFO_LINE_SIZE,
912 };
913 static const struct intel_watermark_params valleyview_wm_info = {
914 VALLEYVIEW_FIFO_SIZE,
915 VALLEYVIEW_MAX_WM,
916 VALLEYVIEW_MAX_WM,
917 2,
918 G4X_FIFO_LINE_SIZE,
919 };
920 static const struct intel_watermark_params valleyview_cursor_wm_info = {
921 I965_CURSOR_FIFO,
922 VALLEYVIEW_CURSOR_MAX_WM,
923 I965_CURSOR_DFT_WM,
924 2,
925 G4X_FIFO_LINE_SIZE,
926 };
927 static const struct intel_watermark_params i965_cursor_wm_info = {
928 I965_CURSOR_FIFO,
929 I965_CURSOR_MAX_WM,
930 I965_CURSOR_DFT_WM,
931 2,
932 I915_FIFO_LINE_SIZE,
933 };
934 static const struct intel_watermark_params i945_wm_info = {
935 I945_FIFO_SIZE,
936 I915_MAX_WM,
937 1,
938 2,
939 I915_FIFO_LINE_SIZE
940 };
941 static const struct intel_watermark_params i915_wm_info = {
942 I915_FIFO_SIZE,
943 I915_MAX_WM,
944 1,
945 2,
946 I915_FIFO_LINE_SIZE
947 };
948 static const struct intel_watermark_params i855_wm_info = {
949 I855GM_FIFO_SIZE,
950 I915_MAX_WM,
951 1,
952 2,
953 I830_FIFO_LINE_SIZE
954 };
955 static const struct intel_watermark_params i830_wm_info = {
956 I830_FIFO_SIZE,
957 I915_MAX_WM,
958 1,
959 2,
960 I830_FIFO_LINE_SIZE
961 };
962
963 static const struct intel_watermark_params ironlake_display_wm_info = {
964 ILK_DISPLAY_FIFO,
965 ILK_DISPLAY_MAXWM,
966 ILK_DISPLAY_DFTWM,
967 2,
968 ILK_FIFO_LINE_SIZE
969 };
970 static const struct intel_watermark_params ironlake_cursor_wm_info = {
971 ILK_CURSOR_FIFO,
972 ILK_CURSOR_MAXWM,
973 ILK_CURSOR_DFTWM,
974 2,
975 ILK_FIFO_LINE_SIZE
976 };
977 static const struct intel_watermark_params ironlake_display_srwm_info = {
978 ILK_DISPLAY_SR_FIFO,
979 ILK_DISPLAY_MAX_SRWM,
980 ILK_DISPLAY_DFT_SRWM,
981 2,
982 ILK_FIFO_LINE_SIZE
983 };
984 static const struct intel_watermark_params ironlake_cursor_srwm_info = {
985 ILK_CURSOR_SR_FIFO,
986 ILK_CURSOR_MAX_SRWM,
987 ILK_CURSOR_DFT_SRWM,
988 2,
989 ILK_FIFO_LINE_SIZE
990 };
991
992 static const struct intel_watermark_params sandybridge_display_wm_info = {
993 SNB_DISPLAY_FIFO,
994 SNB_DISPLAY_MAXWM,
995 SNB_DISPLAY_DFTWM,
996 2,
997 SNB_FIFO_LINE_SIZE
998 };
999 static const struct intel_watermark_params sandybridge_cursor_wm_info = {
1000 SNB_CURSOR_FIFO,
1001 SNB_CURSOR_MAXWM,
1002 SNB_CURSOR_DFTWM,
1003 2,
1004 SNB_FIFO_LINE_SIZE
1005 };
1006 static const struct intel_watermark_params sandybridge_display_srwm_info = {
1007 SNB_DISPLAY_SR_FIFO,
1008 SNB_DISPLAY_MAX_SRWM,
1009 SNB_DISPLAY_DFT_SRWM,
1010 2,
1011 SNB_FIFO_LINE_SIZE
1012 };
1013 static const struct intel_watermark_params sandybridge_cursor_srwm_info = {
1014 SNB_CURSOR_SR_FIFO,
1015 SNB_CURSOR_MAX_SRWM,
1016 SNB_CURSOR_DFT_SRWM,
1017 2,
1018 SNB_FIFO_LINE_SIZE
1019 };
1020
1021
1022 /**
1023 * intel_calculate_wm - calculate watermark level
1024 * @clock_in_khz: pixel clock
1025 * @wm: chip FIFO params
1026 * @pixel_size: display pixel size
1027 * @latency_ns: memory latency for the platform
1028 *
1029 * Calculate the watermark level (the level at which the display plane will
1030 * start fetching from memory again). Each chip has a different display
1031 * FIFO size and allocation, so the caller needs to figure that out and pass
1032 * in the correct intel_watermark_params structure.
1033 *
1034 * As the pixel clock runs, the FIFO will be drained at a rate that depends
1035 * on the pixel size. When it reaches the watermark level, it'll start
1036 * fetching FIFO line sized based chunks from memory until the FIFO fills
1037 * past the watermark point. If the FIFO drains completely, a FIFO underrun
1038 * will occur, and a display engine hang could result.
1039 */
1040 static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
1041 const struct intel_watermark_params *wm,
1042 int fifo_size,
1043 int pixel_size,
1044 unsigned long latency_ns)
1045 {
1046 long entries_required, wm_size;
1047
1048 /*
1049 * Note: we need to make sure we don't overflow for various clock &
1050 * latency values.
1051 * clocks go from a few thousand to several hundred thousand.
1052 * latency is usually a few thousand
1053 */
1054 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
1055 1000;
1056 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
1057
1058 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
1059
1060 wm_size = fifo_size - (entries_required + wm->guard_size);
1061
1062 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
1063
1064 /* Don't promote wm_size to unsigned... */
1065 if (wm_size > (long)wm->max_wm)
1066 wm_size = wm->max_wm;
1067 if (wm_size <= 0)
1068 wm_size = wm->default_wm;
1069 return wm_size;
1070 }
1071
1072 static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1073 {
1074 struct drm_crtc *crtc, *enabled = NULL;
1075
1076 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
1077 if (intel_crtc_active(crtc)) {
1078 if (enabled)
1079 return NULL;
1080 enabled = crtc;
1081 }
1082 }
1083
1084 return enabled;
1085 }
1086
1087 static void pineview_update_wm(struct drm_crtc *unused_crtc)
1088 {
1089 struct drm_device *dev = unused_crtc->dev;
1090 struct drm_i915_private *dev_priv = dev->dev_private;
1091 struct drm_crtc *crtc;
1092 const struct cxsr_latency *latency;
1093 u32 reg;
1094 unsigned long wm;
1095
1096 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1097 dev_priv->fsb_freq, dev_priv->mem_freq);
1098 if (!latency) {
1099 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1100 pineview_disable_cxsr(dev);
1101 return;
1102 }
1103
1104 crtc = single_enabled_crtc(dev);
1105 if (crtc) {
1106 const struct drm_display_mode *adjusted_mode;
1107 int pixel_size = crtc->fb->bits_per_pixel / 8;
1108 int clock;
1109
1110 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1111 clock = adjusted_mode->crtc_clock;
1112
1113 /* Display SR */
1114 wm = intel_calculate_wm(clock, &pineview_display_wm,
1115 pineview_display_wm.fifo_size,
1116 pixel_size, latency->display_sr);
1117 reg = I915_READ(DSPFW1);
1118 reg &= ~DSPFW_SR_MASK;
1119 reg |= wm << DSPFW_SR_SHIFT;
1120 I915_WRITE(DSPFW1, reg);
1121 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1122
1123 /* cursor SR */
1124 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1125 pineview_display_wm.fifo_size,
1126 pixel_size, latency->cursor_sr);
1127 reg = I915_READ(DSPFW3);
1128 reg &= ~DSPFW_CURSOR_SR_MASK;
1129 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1130 I915_WRITE(DSPFW3, reg);
1131
1132 /* Display HPLL off SR */
1133 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1134 pineview_display_hplloff_wm.fifo_size,
1135 pixel_size, latency->display_hpll_disable);
1136 reg = I915_READ(DSPFW3);
1137 reg &= ~DSPFW_HPLL_SR_MASK;
1138 reg |= wm & DSPFW_HPLL_SR_MASK;
1139 I915_WRITE(DSPFW3, reg);
1140
1141 /* cursor HPLL off SR */
1142 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1143 pineview_display_hplloff_wm.fifo_size,
1144 pixel_size, latency->cursor_hpll_disable);
1145 reg = I915_READ(DSPFW3);
1146 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1147 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1148 I915_WRITE(DSPFW3, reg);
1149 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1150
1151 /* activate cxsr */
1152 I915_WRITE(DSPFW3,
1153 I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1154 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1155 } else {
1156 pineview_disable_cxsr(dev);
1157 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1158 }
1159 }
1160
1161 static bool g4x_compute_wm0(struct drm_device *dev,
1162 int plane,
1163 const struct intel_watermark_params *display,
1164 int display_latency_ns,
1165 const struct intel_watermark_params *cursor,
1166 int cursor_latency_ns,
1167 int *plane_wm,
1168 int *cursor_wm)
1169 {
1170 struct drm_crtc *crtc;
1171 const struct drm_display_mode *adjusted_mode;
1172 int htotal, hdisplay, clock, pixel_size;
1173 int line_time_us, line_count;
1174 int entries, tlb_miss;
1175
1176 crtc = intel_get_crtc_for_plane(dev, plane);
1177 if (!intel_crtc_active(crtc)) {
1178 *cursor_wm = cursor->guard_size;
1179 *plane_wm = display->guard_size;
1180 return false;
1181 }
1182
1183 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1184 clock = adjusted_mode->crtc_clock;
1185 htotal = adjusted_mode->htotal;
1186 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1187 pixel_size = crtc->fb->bits_per_pixel / 8;
1188
1189 /* Use the small buffer method to calculate plane watermark */
1190 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1191 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1192 if (tlb_miss > 0)
1193 entries += tlb_miss;
1194 entries = DIV_ROUND_UP(entries, display->cacheline_size);
1195 *plane_wm = entries + display->guard_size;
1196 if (*plane_wm > (int)display->max_wm)
1197 *plane_wm = display->max_wm;
1198
1199 /* Use the large buffer method to calculate cursor watermark */
1200 line_time_us = ((htotal * 1000) / clock);
1201 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
1202 entries = line_count * 64 * pixel_size;
1203 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1204 if (tlb_miss > 0)
1205 entries += tlb_miss;
1206 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1207 *cursor_wm = entries + cursor->guard_size;
1208 if (*cursor_wm > (int)cursor->max_wm)
1209 *cursor_wm = (int)cursor->max_wm;
1210
1211 return true;
1212 }
1213
1214 /*
1215 * Check the wm result.
1216 *
1217 * If any calculated watermark values is larger than the maximum value that
1218 * can be programmed into the associated watermark register, that watermark
1219 * must be disabled.
1220 */
1221 static bool g4x_check_srwm(struct drm_device *dev,
1222 int display_wm, int cursor_wm,
1223 const struct intel_watermark_params *display,
1224 const struct intel_watermark_params *cursor)
1225 {
1226 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1227 display_wm, cursor_wm);
1228
1229 if (display_wm > display->max_wm) {
1230 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1231 display_wm, display->max_wm);
1232 return false;
1233 }
1234
1235 if (cursor_wm > cursor->max_wm) {
1236 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1237 cursor_wm, cursor->max_wm);
1238 return false;
1239 }
1240
1241 if (!(display_wm || cursor_wm)) {
1242 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1243 return false;
1244 }
1245
1246 return true;
1247 }
1248
1249 static bool g4x_compute_srwm(struct drm_device *dev,
1250 int plane,
1251 int latency_ns,
1252 const struct intel_watermark_params *display,
1253 const struct intel_watermark_params *cursor,
1254 int *display_wm, int *cursor_wm)
1255 {
1256 struct drm_crtc *crtc;
1257 const struct drm_display_mode *adjusted_mode;
1258 int hdisplay, htotal, pixel_size, clock;
1259 unsigned long line_time_us;
1260 int line_count, line_size;
1261 int small, large;
1262 int entries;
1263
1264 if (!latency_ns) {
1265 *display_wm = *cursor_wm = 0;
1266 return false;
1267 }
1268
1269 crtc = intel_get_crtc_for_plane(dev, plane);
1270 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1271 clock = adjusted_mode->crtc_clock;
1272 htotal = adjusted_mode->htotal;
1273 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1274 pixel_size = crtc->fb->bits_per_pixel / 8;
1275
1276 line_time_us = (htotal * 1000) / clock;
1277 line_count = (latency_ns / line_time_us + 1000) / 1000;
1278 line_size = hdisplay * pixel_size;
1279
1280 /* Use the minimum of the small and large buffer method for primary */
1281 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1282 large = line_count * line_size;
1283
1284 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1285 *display_wm = entries + display->guard_size;
1286
1287 /* calculate the self-refresh watermark for display cursor */
1288 entries = line_count * pixel_size * 64;
1289 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1290 *cursor_wm = entries + cursor->guard_size;
1291
1292 return g4x_check_srwm(dev,
1293 *display_wm, *cursor_wm,
1294 display, cursor);
1295 }
1296
1297 static bool vlv_compute_drain_latency(struct drm_device *dev,
1298 int plane,
1299 int *plane_prec_mult,
1300 int *plane_dl,
1301 int *cursor_prec_mult,
1302 int *cursor_dl)
1303 {
1304 struct drm_crtc *crtc;
1305 int clock, pixel_size;
1306 int entries;
1307
1308 crtc = intel_get_crtc_for_plane(dev, plane);
1309 if (!intel_crtc_active(crtc))
1310 return false;
1311
1312 clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
1313 pixel_size = crtc->fb->bits_per_pixel / 8; /* BPP */
1314
1315 entries = (clock / 1000) * pixel_size;
1316 *plane_prec_mult = (entries > 256) ?
1317 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1318 *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1319 pixel_size);
1320
1321 entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */
1322 *cursor_prec_mult = (entries > 256) ?
1323 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1324 *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1325
1326 return true;
1327 }
1328
1329 /*
1330 * Update drain latency registers of memory arbiter
1331 *
1332 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1333 * to be programmed. Each plane has a drain latency multiplier and a drain
1334 * latency value.
1335 */
1336
1337 static void vlv_update_drain_latency(struct drm_device *dev)
1338 {
1339 struct drm_i915_private *dev_priv = dev->dev_private;
1340 int planea_prec, planea_dl, planeb_prec, planeb_dl;
1341 int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1342 int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1343 either 16 or 32 */
1344
1345 /* For plane A, Cursor A */
1346 if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1347 &cursor_prec_mult, &cursora_dl)) {
1348 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1349 DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1350 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1351 DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1352
1353 I915_WRITE(VLV_DDL1, cursora_prec |
1354 (cursora_dl << DDL_CURSORA_SHIFT) |
1355 planea_prec | planea_dl);
1356 }
1357
1358 /* For plane B, Cursor B */
1359 if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1360 &cursor_prec_mult, &cursorb_dl)) {
1361 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1362 DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1363 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1364 DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1365
1366 I915_WRITE(VLV_DDL2, cursorb_prec |
1367 (cursorb_dl << DDL_CURSORB_SHIFT) |
1368 planeb_prec | planeb_dl);
1369 }
1370 }
1371
1372 #define single_plane_enabled(mask) is_power_of_2(mask)
1373
1374 static void valleyview_update_wm(struct drm_crtc *crtc)
1375 {
1376 struct drm_device *dev = crtc->dev;
1377 static const int sr_latency_ns = 12000;
1378 struct drm_i915_private *dev_priv = dev->dev_private;
1379 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1380 int plane_sr, cursor_sr;
1381 int ignore_plane_sr, ignore_cursor_sr;
1382 unsigned int enabled = 0;
1383
1384 vlv_update_drain_latency(dev);
1385
1386 if (g4x_compute_wm0(dev, PIPE_A,
1387 &valleyview_wm_info, latency_ns,
1388 &valleyview_cursor_wm_info, latency_ns,
1389 &planea_wm, &cursora_wm))
1390 enabled |= 1 << PIPE_A;
1391
1392 if (g4x_compute_wm0(dev, PIPE_B,
1393 &valleyview_wm_info, latency_ns,
1394 &valleyview_cursor_wm_info, latency_ns,
1395 &planeb_wm, &cursorb_wm))
1396 enabled |= 1 << PIPE_B;
1397
1398 if (single_plane_enabled(enabled) &&
1399 g4x_compute_srwm(dev, ffs(enabled) - 1,
1400 sr_latency_ns,
1401 &valleyview_wm_info,
1402 &valleyview_cursor_wm_info,
1403 &plane_sr, &ignore_cursor_sr) &&
1404 g4x_compute_srwm(dev, ffs(enabled) - 1,
1405 2*sr_latency_ns,
1406 &valleyview_wm_info,
1407 &valleyview_cursor_wm_info,
1408 &ignore_plane_sr, &cursor_sr)) {
1409 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
1410 } else {
1411 I915_WRITE(FW_BLC_SELF_VLV,
1412 I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
1413 plane_sr = cursor_sr = 0;
1414 }
1415
1416 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1417 planea_wm, cursora_wm,
1418 planeb_wm, cursorb_wm,
1419 plane_sr, cursor_sr);
1420
1421 I915_WRITE(DSPFW1,
1422 (plane_sr << DSPFW_SR_SHIFT) |
1423 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1424 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1425 planea_wm);
1426 I915_WRITE(DSPFW2,
1427 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1428 (cursora_wm << DSPFW_CURSORA_SHIFT));
1429 I915_WRITE(DSPFW3,
1430 (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1431 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1432 }
1433
1434 static void g4x_update_wm(struct drm_crtc *crtc)
1435 {
1436 struct drm_device *dev = crtc->dev;
1437 static const int sr_latency_ns = 12000;
1438 struct drm_i915_private *dev_priv = dev->dev_private;
1439 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1440 int plane_sr, cursor_sr;
1441 unsigned int enabled = 0;
1442
1443 if (g4x_compute_wm0(dev, PIPE_A,
1444 &g4x_wm_info, latency_ns,
1445 &g4x_cursor_wm_info, latency_ns,
1446 &planea_wm, &cursora_wm))
1447 enabled |= 1 << PIPE_A;
1448
1449 if (g4x_compute_wm0(dev, PIPE_B,
1450 &g4x_wm_info, latency_ns,
1451 &g4x_cursor_wm_info, latency_ns,
1452 &planeb_wm, &cursorb_wm))
1453 enabled |= 1 << PIPE_B;
1454
1455 if (single_plane_enabled(enabled) &&
1456 g4x_compute_srwm(dev, ffs(enabled) - 1,
1457 sr_latency_ns,
1458 &g4x_wm_info,
1459 &g4x_cursor_wm_info,
1460 &plane_sr, &cursor_sr)) {
1461 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1462 } else {
1463 I915_WRITE(FW_BLC_SELF,
1464 I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
1465 plane_sr = cursor_sr = 0;
1466 }
1467
1468 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1469 planea_wm, cursora_wm,
1470 planeb_wm, cursorb_wm,
1471 plane_sr, cursor_sr);
1472
1473 I915_WRITE(DSPFW1,
1474 (plane_sr << DSPFW_SR_SHIFT) |
1475 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1476 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1477 planea_wm);
1478 I915_WRITE(DSPFW2,
1479 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1480 (cursora_wm << DSPFW_CURSORA_SHIFT));
1481 /* HPLL off in SR has some issues on G4x... disable it */
1482 I915_WRITE(DSPFW3,
1483 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
1484 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1485 }
1486
1487 static void i965_update_wm(struct drm_crtc *unused_crtc)
1488 {
1489 struct drm_device *dev = unused_crtc->dev;
1490 struct drm_i915_private *dev_priv = dev->dev_private;
1491 struct drm_crtc *crtc;
1492 int srwm = 1;
1493 int cursor_sr = 16;
1494
1495 /* Calc sr entries for one plane configs */
1496 crtc = single_enabled_crtc(dev);
1497 if (crtc) {
1498 /* self-refresh has much higher latency */
1499 static const int sr_latency_ns = 12000;
1500 const struct drm_display_mode *adjusted_mode =
1501 &to_intel_crtc(crtc)->config.adjusted_mode;
1502 int clock = adjusted_mode->crtc_clock;
1503 int htotal = adjusted_mode->htotal;
1504 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1505 int pixel_size = crtc->fb->bits_per_pixel / 8;
1506 unsigned long line_time_us;
1507 int entries;
1508
1509 line_time_us = ((htotal * 1000) / clock);
1510
1511 /* Use ns/us then divide to preserve precision */
1512 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1513 pixel_size * hdisplay;
1514 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1515 srwm = I965_FIFO_SIZE - entries;
1516 if (srwm < 0)
1517 srwm = 1;
1518 srwm &= 0x1ff;
1519 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1520 entries, srwm);
1521
1522 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1523 pixel_size * 64;
1524 entries = DIV_ROUND_UP(entries,
1525 i965_cursor_wm_info.cacheline_size);
1526 cursor_sr = i965_cursor_wm_info.fifo_size -
1527 (entries + i965_cursor_wm_info.guard_size);
1528
1529 if (cursor_sr > i965_cursor_wm_info.max_wm)
1530 cursor_sr = i965_cursor_wm_info.max_wm;
1531
1532 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1533 "cursor %d\n", srwm, cursor_sr);
1534
1535 if (IS_CRESTLINE(dev))
1536 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1537 } else {
1538 /* Turn off self refresh if both pipes are enabled */
1539 if (IS_CRESTLINE(dev))
1540 I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1541 & ~FW_BLC_SELF_EN);
1542 }
1543
1544 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1545 srwm);
1546
1547 /* 965 has limitations... */
1548 I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1549 (8 << 16) | (8 << 8) | (8 << 0));
1550 I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1551 /* update cursor SR watermark */
1552 I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1553 }
1554
1555 static void i9xx_update_wm(struct drm_crtc *unused_crtc)
1556 {
1557 struct drm_device *dev = unused_crtc->dev;
1558 struct drm_i915_private *dev_priv = dev->dev_private;
1559 const struct intel_watermark_params *wm_info;
1560 uint32_t fwater_lo;
1561 uint32_t fwater_hi;
1562 int cwm, srwm = 1;
1563 int fifo_size;
1564 int planea_wm, planeb_wm;
1565 struct drm_crtc *crtc, *enabled = NULL;
1566
1567 if (IS_I945GM(dev))
1568 wm_info = &i945_wm_info;
1569 else if (!IS_GEN2(dev))
1570 wm_info = &i915_wm_info;
1571 else
1572 wm_info = &i855_wm_info;
1573
1574 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1575 crtc = intel_get_crtc_for_plane(dev, 0);
1576 if (intel_crtc_active(crtc)) {
1577 const struct drm_display_mode *adjusted_mode;
1578 int cpp = crtc->fb->bits_per_pixel / 8;
1579 if (IS_GEN2(dev))
1580 cpp = 4;
1581
1582 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1583 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
1584 wm_info, fifo_size, cpp,
1585 latency_ns);
1586 enabled = crtc;
1587 } else
1588 planea_wm = fifo_size - wm_info->guard_size;
1589
1590 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1591 crtc = intel_get_crtc_for_plane(dev, 1);
1592 if (intel_crtc_active(crtc)) {
1593 const struct drm_display_mode *adjusted_mode;
1594 int cpp = crtc->fb->bits_per_pixel / 8;
1595 if (IS_GEN2(dev))
1596 cpp = 4;
1597
1598 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1599 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
1600 wm_info, fifo_size, cpp,
1601 latency_ns);
1602 if (enabled == NULL)
1603 enabled = crtc;
1604 else
1605 enabled = NULL;
1606 } else
1607 planeb_wm = fifo_size - wm_info->guard_size;
1608
1609 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1610
1611 /*
1612 * Overlay gets an aggressive default since video jitter is bad.
1613 */
1614 cwm = 2;
1615
1616 /* Play safe and disable self-refresh before adjusting watermarks. */
1617 if (IS_I945G(dev) || IS_I945GM(dev))
1618 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1619 else if (IS_I915GM(dev))
1620 I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN);
1621
1622 /* Calc sr entries for one plane configs */
1623 if (HAS_FW_BLC(dev) && enabled) {
1624 /* self-refresh has much higher latency */
1625 static const int sr_latency_ns = 6000;
1626 const struct drm_display_mode *adjusted_mode =
1627 &to_intel_crtc(enabled)->config.adjusted_mode;
1628 int clock = adjusted_mode->crtc_clock;
1629 int htotal = adjusted_mode->htotal;
1630 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1631 int pixel_size = enabled->fb->bits_per_pixel / 8;
1632 unsigned long line_time_us;
1633 int entries;
1634
1635 line_time_us = (htotal * 1000) / clock;
1636
1637 /* Use ns/us then divide to preserve precision */
1638 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1639 pixel_size * hdisplay;
1640 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1641 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1642 srwm = wm_info->fifo_size - entries;
1643 if (srwm < 0)
1644 srwm = 1;
1645
1646 if (IS_I945G(dev) || IS_I945GM(dev))
1647 I915_WRITE(FW_BLC_SELF,
1648 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1649 else if (IS_I915GM(dev))
1650 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1651 }
1652
1653 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1654 planea_wm, planeb_wm, cwm, srwm);
1655
1656 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1657 fwater_hi = (cwm & 0x1f);
1658
1659 /* Set request length to 8 cachelines per fetch */
1660 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1661 fwater_hi = fwater_hi | (1 << 8);
1662
1663 I915_WRITE(FW_BLC, fwater_lo);
1664 I915_WRITE(FW_BLC2, fwater_hi);
1665
1666 if (HAS_FW_BLC(dev)) {
1667 if (enabled) {
1668 if (IS_I945G(dev) || IS_I945GM(dev))
1669 I915_WRITE(FW_BLC_SELF,
1670 FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1671 else if (IS_I915GM(dev))
1672 I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN);
1673 DRM_DEBUG_KMS("memory self refresh enabled\n");
1674 } else
1675 DRM_DEBUG_KMS("memory self refresh disabled\n");
1676 }
1677 }
1678
1679 static void i830_update_wm(struct drm_crtc *unused_crtc)
1680 {
1681 struct drm_device *dev = unused_crtc->dev;
1682 struct drm_i915_private *dev_priv = dev->dev_private;
1683 struct drm_crtc *crtc;
1684 const struct drm_display_mode *adjusted_mode;
1685 uint32_t fwater_lo;
1686 int planea_wm;
1687
1688 crtc = single_enabled_crtc(dev);
1689 if (crtc == NULL)
1690 return;
1691
1692 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1693 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
1694 &i830_wm_info,
1695 dev_priv->display.get_fifo_size(dev, 0),
1696 4, latency_ns);
1697 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1698 fwater_lo |= (3<<8) | planea_wm;
1699
1700 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1701
1702 I915_WRITE(FW_BLC, fwater_lo);
1703 }
1704
1705 /*
1706 * Check the wm result.
1707 *
1708 * If any calculated watermark values is larger than the maximum value that
1709 * can be programmed into the associated watermark register, that watermark
1710 * must be disabled.
1711 */
1712 static bool ironlake_check_srwm(struct drm_device *dev, int level,
1713 int fbc_wm, int display_wm, int cursor_wm,
1714 const struct intel_watermark_params *display,
1715 const struct intel_watermark_params *cursor)
1716 {
1717 struct drm_i915_private *dev_priv = dev->dev_private;
1718
1719 DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d,"
1720 " cursor %d\n", level, display_wm, fbc_wm, cursor_wm);
1721
1722 if (fbc_wm > SNB_FBC_MAX_SRWM) {
1723 DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n",
1724 fbc_wm, SNB_FBC_MAX_SRWM, level);
1725
1726 /* fbc has it's own way to disable FBC WM */
1727 I915_WRITE(DISP_ARB_CTL,
1728 I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS);
1729 return false;
1730 } else if (INTEL_INFO(dev)->gen >= 6) {
1731 /* enable FBC WM (except on ILK, where it must remain off) */
1732 I915_WRITE(DISP_ARB_CTL,
1733 I915_READ(DISP_ARB_CTL) & ~DISP_FBC_WM_DIS);
1734 }
1735
1736 if (display_wm > display->max_wm) {
1737 DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n",
1738 display_wm, SNB_DISPLAY_MAX_SRWM, level);
1739 return false;
1740 }
1741
1742 if (cursor_wm > cursor->max_wm) {
1743 DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n",
1744 cursor_wm, SNB_CURSOR_MAX_SRWM, level);
1745 return false;
1746 }
1747
1748 if (!(fbc_wm || display_wm || cursor_wm)) {
1749 DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level);
1750 return false;
1751 }
1752
1753 return true;
1754 }
1755
1756 /*
1757 * Compute watermark values of WM[1-3],
1758 */
1759 static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane,
1760 int latency_ns,
1761 const struct intel_watermark_params *display,
1762 const struct intel_watermark_params *cursor,
1763 int *fbc_wm, int *display_wm, int *cursor_wm)
1764 {
1765 struct drm_crtc *crtc;
1766 const struct drm_display_mode *adjusted_mode;
1767 unsigned long line_time_us;
1768 int hdisplay, htotal, pixel_size, clock;
1769 int line_count, line_size;
1770 int small, large;
1771 int entries;
1772
1773 if (!latency_ns) {
1774 *fbc_wm = *display_wm = *cursor_wm = 0;
1775 return false;
1776 }
1777
1778 crtc = intel_get_crtc_for_plane(dev, plane);
1779 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1780 clock = adjusted_mode->crtc_clock;
1781 htotal = adjusted_mode->htotal;
1782 hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1783 pixel_size = crtc->fb->bits_per_pixel / 8;
1784
1785 line_time_us = (htotal * 1000) / clock;
1786 line_count = (latency_ns / line_time_us + 1000) / 1000;
1787 line_size = hdisplay * pixel_size;
1788
1789 /* Use the minimum of the small and large buffer method for primary */
1790 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1791 large = line_count * line_size;
1792
1793 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1794 *display_wm = entries + display->guard_size;
1795
1796 /*
1797 * Spec says:
1798 * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2
1799 */
1800 *fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2;
1801
1802 /* calculate the self-refresh watermark for display cursor */
1803 entries = line_count * pixel_size * 64;
1804 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1805 *cursor_wm = entries + cursor->guard_size;
1806
1807 return ironlake_check_srwm(dev, level,
1808 *fbc_wm, *display_wm, *cursor_wm,
1809 display, cursor);
1810 }
1811
1812 static void ironlake_update_wm(struct drm_crtc *crtc)
1813 {
1814 struct drm_device *dev = crtc->dev;
1815 struct drm_i915_private *dev_priv = dev->dev_private;
1816 int fbc_wm, plane_wm, cursor_wm;
1817 unsigned int enabled;
1818
1819 enabled = 0;
1820 if (g4x_compute_wm0(dev, PIPE_A,
1821 &ironlake_display_wm_info,
1822 dev_priv->wm.pri_latency[0] * 100,
1823 &ironlake_cursor_wm_info,
1824 dev_priv->wm.cur_latency[0] * 100,
1825 &plane_wm, &cursor_wm)) {
1826 I915_WRITE(WM0_PIPEA_ILK,
1827 (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1828 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1829 " plane %d, " "cursor: %d\n",
1830 plane_wm, cursor_wm);
1831 enabled |= 1 << PIPE_A;
1832 }
1833
1834 if (g4x_compute_wm0(dev, PIPE_B,
1835 &ironlake_display_wm_info,
1836 dev_priv->wm.pri_latency[0] * 100,
1837 &ironlake_cursor_wm_info,
1838 dev_priv->wm.cur_latency[0] * 100,
1839 &plane_wm, &cursor_wm)) {
1840 I915_WRITE(WM0_PIPEB_ILK,
1841 (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1842 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1843 " plane %d, cursor: %d\n",
1844 plane_wm, cursor_wm);
1845 enabled |= 1 << PIPE_B;
1846 }
1847
1848 /*
1849 * Calculate and update the self-refresh watermark only when one
1850 * display plane is used.
1851 */
1852 I915_WRITE(WM3_LP_ILK, 0);
1853 I915_WRITE(WM2_LP_ILK, 0);
1854 I915_WRITE(WM1_LP_ILK, 0);
1855
1856 if (!single_plane_enabled(enabled))
1857 return;
1858 enabled = ffs(enabled) - 1;
1859
1860 /* WM1 */
1861 if (!ironlake_compute_srwm(dev, 1, enabled,
1862 dev_priv->wm.pri_latency[1] * 500,
1863 &ironlake_display_srwm_info,
1864 &ironlake_cursor_srwm_info,
1865 &fbc_wm, &plane_wm, &cursor_wm))
1866 return;
1867
1868 I915_WRITE(WM1_LP_ILK,
1869 WM1_LP_SR_EN |
1870 (dev_priv->wm.pri_latency[1] << WM1_LP_LATENCY_SHIFT) |
1871 (fbc_wm << WM1_LP_FBC_SHIFT) |
1872 (plane_wm << WM1_LP_SR_SHIFT) |
1873 cursor_wm);
1874
1875 /* WM2 */
1876 if (!ironlake_compute_srwm(dev, 2, enabled,
1877 dev_priv->wm.pri_latency[2] * 500,
1878 &ironlake_display_srwm_info,
1879 &ironlake_cursor_srwm_info,
1880 &fbc_wm, &plane_wm, &cursor_wm))
1881 return;
1882
1883 I915_WRITE(WM2_LP_ILK,
1884 WM2_LP_EN |
1885 (dev_priv->wm.pri_latency[2] << WM1_LP_LATENCY_SHIFT) |
1886 (fbc_wm << WM1_LP_FBC_SHIFT) |
1887 (plane_wm << WM1_LP_SR_SHIFT) |
1888 cursor_wm);
1889
1890 /*
1891 * WM3 is unsupported on ILK, probably because we don't have latency
1892 * data for that power state
1893 */
1894 }
1895
1896 static void sandybridge_update_wm(struct drm_crtc *crtc)
1897 {
1898 struct drm_device *dev = crtc->dev;
1899 struct drm_i915_private *dev_priv = dev->dev_private;
1900 int latency = dev_priv->wm.pri_latency[0] * 100; /* In unit 0.1us */
1901 u32 val;
1902 int fbc_wm, plane_wm, cursor_wm;
1903 unsigned int enabled;
1904
1905 enabled = 0;
1906 if (g4x_compute_wm0(dev, PIPE_A,
1907 &sandybridge_display_wm_info, latency,
1908 &sandybridge_cursor_wm_info, latency,
1909 &plane_wm, &cursor_wm)) {
1910 val = I915_READ(WM0_PIPEA_ILK);
1911 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1912 I915_WRITE(WM0_PIPEA_ILK, val |
1913 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1914 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1915 " plane %d, " "cursor: %d\n",
1916 plane_wm, cursor_wm);
1917 enabled |= 1 << PIPE_A;
1918 }
1919
1920 if (g4x_compute_wm0(dev, PIPE_B,
1921 &sandybridge_display_wm_info, latency,
1922 &sandybridge_cursor_wm_info, latency,
1923 &plane_wm, &cursor_wm)) {
1924 val = I915_READ(WM0_PIPEB_ILK);
1925 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1926 I915_WRITE(WM0_PIPEB_ILK, val |
1927 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1928 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1929 " plane %d, cursor: %d\n",
1930 plane_wm, cursor_wm);
1931 enabled |= 1 << PIPE_B;
1932 }
1933
1934 /*
1935 * Calculate and update the self-refresh watermark only when one
1936 * display plane is used.
1937 *
1938 * SNB support 3 levels of watermark.
1939 *
1940 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1941 * and disabled in the descending order
1942 *
1943 */
1944 I915_WRITE(WM3_LP_ILK, 0);
1945 I915_WRITE(WM2_LP_ILK, 0);
1946 I915_WRITE(WM1_LP_ILK, 0);
1947
1948 if (!single_plane_enabled(enabled) ||
1949 dev_priv->sprite_scaling_enabled)
1950 return;
1951 enabled = ffs(enabled) - 1;
1952
1953 /* WM1 */
1954 if (!ironlake_compute_srwm(dev, 1, enabled,
1955 dev_priv->wm.pri_latency[1] * 500,
1956 &sandybridge_display_srwm_info,
1957 &sandybridge_cursor_srwm_info,
1958 &fbc_wm, &plane_wm, &cursor_wm))
1959 return;
1960
1961 I915_WRITE(WM1_LP_ILK,
1962 WM1_LP_SR_EN |
1963 (dev_priv->wm.pri_latency[1] << WM1_LP_LATENCY_SHIFT) |
1964 (fbc_wm << WM1_LP_FBC_SHIFT) |
1965 (plane_wm << WM1_LP_SR_SHIFT) |
1966 cursor_wm);
1967
1968 /* WM2 */
1969 if (!ironlake_compute_srwm(dev, 2, enabled,
1970 dev_priv->wm.pri_latency[2] * 500,
1971 &sandybridge_display_srwm_info,
1972 &sandybridge_cursor_srwm_info,
1973 &fbc_wm, &plane_wm, &cursor_wm))
1974 return;
1975
1976 I915_WRITE(WM2_LP_ILK,
1977 WM2_LP_EN |
1978 (dev_priv->wm.pri_latency[2] << WM1_LP_LATENCY_SHIFT) |
1979 (fbc_wm << WM1_LP_FBC_SHIFT) |
1980 (plane_wm << WM1_LP_SR_SHIFT) |
1981 cursor_wm);
1982
1983 /* WM3 */
1984 if (!ironlake_compute_srwm(dev, 3, enabled,
1985 dev_priv->wm.pri_latency[3] * 500,
1986 &sandybridge_display_srwm_info,
1987 &sandybridge_cursor_srwm_info,
1988 &fbc_wm, &plane_wm, &cursor_wm))
1989 return;
1990
1991 I915_WRITE(WM3_LP_ILK,
1992 WM3_LP_EN |
1993 (dev_priv->wm.pri_latency[3] << WM1_LP_LATENCY_SHIFT) |
1994 (fbc_wm << WM1_LP_FBC_SHIFT) |
1995 (plane_wm << WM1_LP_SR_SHIFT) |
1996 cursor_wm);
1997 }
1998
1999 static void ivybridge_update_wm(struct drm_crtc *crtc)
2000 {
2001 struct drm_device *dev = crtc->dev;
2002 struct drm_i915_private *dev_priv = dev->dev_private;
2003 int latency = dev_priv->wm.pri_latency[0] * 100; /* In unit 0.1us */
2004 u32 val;
2005 int fbc_wm, plane_wm, cursor_wm;
2006 int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm;
2007 unsigned int enabled;
2008
2009 enabled = 0;
2010 if (g4x_compute_wm0(dev, PIPE_A,
2011 &sandybridge_display_wm_info, latency,
2012 &sandybridge_cursor_wm_info, latency,
2013 &plane_wm, &cursor_wm)) {
2014 val = I915_READ(WM0_PIPEA_ILK);
2015 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
2016 I915_WRITE(WM0_PIPEA_ILK, val |
2017 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
2018 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
2019 " plane %d, " "cursor: %d\n",
2020 plane_wm, cursor_wm);
2021 enabled |= 1 << PIPE_A;
2022 }
2023
2024 if (g4x_compute_wm0(dev, PIPE_B,
2025 &sandybridge_display_wm_info, latency,
2026 &sandybridge_cursor_wm_info, latency,
2027 &plane_wm, &cursor_wm)) {
2028 val = I915_READ(WM0_PIPEB_ILK);
2029 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
2030 I915_WRITE(WM0_PIPEB_ILK, val |
2031 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
2032 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
2033 " plane %d, cursor: %d\n",
2034 plane_wm, cursor_wm);
2035 enabled |= 1 << PIPE_B;
2036 }
2037
2038 if (g4x_compute_wm0(dev, PIPE_C,
2039 &sandybridge_display_wm_info, latency,
2040 &sandybridge_cursor_wm_info, latency,
2041 &plane_wm, &cursor_wm)) {
2042 val = I915_READ(WM0_PIPEC_IVB);
2043 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
2044 I915_WRITE(WM0_PIPEC_IVB, val |
2045 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
2046 DRM_DEBUG_KMS("FIFO watermarks For pipe C -"
2047 " plane %d, cursor: %d\n",
2048 plane_wm, cursor_wm);
2049 enabled |= 1 << PIPE_C;
2050 }
2051
2052 /*
2053 * Calculate and update the self-refresh watermark only when one
2054 * display plane is used.
2055 *
2056 * SNB support 3 levels of watermark.
2057 *
2058 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
2059 * and disabled in the descending order
2060 *
2061 */
2062 I915_WRITE(WM3_LP_ILK, 0);
2063 I915_WRITE(WM2_LP_ILK, 0);
2064 I915_WRITE(WM1_LP_ILK, 0);
2065
2066 if (!single_plane_enabled(enabled) ||
2067 dev_priv->sprite_scaling_enabled)
2068 return;
2069 enabled = ffs(enabled) - 1;
2070
2071 /* WM1 */
2072 if (!ironlake_compute_srwm(dev, 1, enabled,
2073 dev_priv->wm.pri_latency[1] * 500,
2074 &sandybridge_display_srwm_info,
2075 &sandybridge_cursor_srwm_info,
2076 &fbc_wm, &plane_wm, &cursor_wm))
2077 return;
2078
2079 I915_WRITE(WM1_LP_ILK,
2080 WM1_LP_SR_EN |
2081 (dev_priv->wm.pri_latency[1] << WM1_LP_LATENCY_SHIFT) |
2082 (fbc_wm << WM1_LP_FBC_SHIFT) |
2083 (plane_wm << WM1_LP_SR_SHIFT) |
2084 cursor_wm);
2085
2086 /* WM2 */
2087 if (!ironlake_compute_srwm(dev, 2, enabled,
2088 dev_priv->wm.pri_latency[2] * 500,
2089 &sandybridge_display_srwm_info,
2090 &sandybridge_cursor_srwm_info,
2091 &fbc_wm, &plane_wm, &cursor_wm))
2092 return;
2093
2094 I915_WRITE(WM2_LP_ILK,
2095 WM2_LP_EN |
2096 (dev_priv->wm.pri_latency[2] << WM1_LP_LATENCY_SHIFT) |
2097 (fbc_wm << WM1_LP_FBC_SHIFT) |
2098 (plane_wm << WM1_LP_SR_SHIFT) |
2099 cursor_wm);
2100
2101 /* WM3, note we have to correct the cursor latency */
2102 if (!ironlake_compute_srwm(dev, 3, enabled,
2103 dev_priv->wm.pri_latency[3] * 500,
2104 &sandybridge_display_srwm_info,
2105 &sandybridge_cursor_srwm_info,
2106 &fbc_wm, &plane_wm, &ignore_cursor_wm) ||
2107 !ironlake_compute_srwm(dev, 3, enabled,
2108 dev_priv->wm.cur_latency[3] * 500,
2109 &sandybridge_display_srwm_info,
2110 &sandybridge_cursor_srwm_info,
2111 &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm))
2112 return;
2113
2114 I915_WRITE(WM3_LP_ILK,
2115 WM3_LP_EN |
2116 (dev_priv->wm.pri_latency[3] << WM1_LP_LATENCY_SHIFT) |
2117 (fbc_wm << WM1_LP_FBC_SHIFT) |
2118 (plane_wm << WM1_LP_SR_SHIFT) |
2119 cursor_wm);
2120 }
2121
2122 static uint32_t ilk_pipe_pixel_rate(struct drm_device *dev,
2123 struct drm_crtc *crtc)
2124 {
2125 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2126 uint32_t pixel_rate;
2127
2128 pixel_rate = intel_crtc->config.adjusted_mode.crtc_clock;
2129
2130 /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
2131 * adjust the pixel_rate here. */
2132
2133 if (intel_crtc->config.pch_pfit.enabled) {
2134 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
2135 uint32_t pfit_size = intel_crtc->config.pch_pfit.size;
2136
2137 pipe_w = intel_crtc->config.pipe_src_w;
2138 pipe_h = intel_crtc->config.pipe_src_h;
2139 pfit_w = (pfit_size >> 16) & 0xFFFF;
2140 pfit_h = pfit_size & 0xFFFF;
2141 if (pipe_w < pfit_w)
2142 pipe_w = pfit_w;
2143 if (pipe_h < pfit_h)
2144 pipe_h = pfit_h;
2145
2146 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
2147 pfit_w * pfit_h);
2148 }
2149
2150 return pixel_rate;
2151 }
2152
2153 /* latency must be in 0.1us units. */
2154 static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
2155 uint32_t latency)
2156 {
2157 uint64_t ret;
2158
2159 if (WARN(latency == 0, "Latency value missing\n"))
2160 return UINT_MAX;
2161
2162 ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
2163 ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
2164
2165 return ret;
2166 }
2167
2168 /* latency must be in 0.1us units. */
2169 static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
2170 uint32_t horiz_pixels, uint8_t bytes_per_pixel,
2171 uint32_t latency)
2172 {
2173 uint32_t ret;
2174
2175 if (WARN(latency == 0, "Latency value missing\n"))
2176 return UINT_MAX;
2177
2178 ret = (latency * pixel_rate) / (pipe_htotal * 10000);
2179 ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
2180 ret = DIV_ROUND_UP(ret, 64) + 2;
2181 return ret;
2182 }
2183
2184 static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
2185 uint8_t bytes_per_pixel)
2186 {
2187 return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
2188 }
2189
2190 struct hsw_pipe_wm_parameters {
2191 bool active;
2192 uint32_t pipe_htotal;
2193 uint32_t pixel_rate;
2194 struct intel_plane_wm_parameters pri;
2195 struct intel_plane_wm_parameters spr;
2196 struct intel_plane_wm_parameters cur;
2197 };
2198
2199 struct hsw_wm_maximums {
2200 uint16_t pri;
2201 uint16_t spr;
2202 uint16_t cur;
2203 uint16_t fbc;
2204 };
2205
2206 /* used in computing the new watermarks state */
2207 struct intel_wm_config {
2208 unsigned int num_pipes_active;
2209 bool sprites_enabled;
2210 bool sprites_scaled;
2211 };
2212
2213 /*
2214 * For both WM_PIPE and WM_LP.
2215 * mem_value must be in 0.1us units.
2216 */
2217 static uint32_t ilk_compute_pri_wm(const struct hsw_pipe_wm_parameters *params,
2218 uint32_t mem_value,
2219 bool is_lp)
2220 {
2221 uint32_t method1, method2;
2222
2223 if (!params->active || !params->pri.enabled)
2224 return 0;
2225
2226 method1 = ilk_wm_method1(params->pixel_rate,
2227 params->pri.bytes_per_pixel,
2228 mem_value);
2229
2230 if (!is_lp)
2231 return method1;
2232
2233 method2 = ilk_wm_method2(params->pixel_rate,
2234 params->pipe_htotal,
2235 params->pri.horiz_pixels,
2236 params->pri.bytes_per_pixel,
2237 mem_value);
2238
2239 return min(method1, method2);
2240 }
2241
2242 /*
2243 * For both WM_PIPE and WM_LP.
2244 * mem_value must be in 0.1us units.
2245 */
2246 static uint32_t ilk_compute_spr_wm(const struct hsw_pipe_wm_parameters *params,
2247 uint32_t mem_value)
2248 {
2249 uint32_t method1, method2;
2250
2251 if (!params->active || !params->spr.enabled)
2252 return 0;
2253
2254 method1 = ilk_wm_method1(params->pixel_rate,
2255 params->spr.bytes_per_pixel,
2256 mem_value);
2257 method2 = ilk_wm_method2(params->pixel_rate,
2258 params->pipe_htotal,
2259 params->spr.horiz_pixels,
2260 params->spr.bytes_per_pixel,
2261 mem_value);
2262 return min(method1, method2);
2263 }
2264
2265 /*
2266 * For both WM_PIPE and WM_LP.
2267 * mem_value must be in 0.1us units.
2268 */
2269 static uint32_t ilk_compute_cur_wm(const struct hsw_pipe_wm_parameters *params,
2270 uint32_t mem_value)
2271 {
2272 if (!params->active || !params->cur.enabled)
2273 return 0;
2274
2275 return ilk_wm_method2(params->pixel_rate,
2276 params->pipe_htotal,
2277 params->cur.horiz_pixels,
2278 params->cur.bytes_per_pixel,
2279 mem_value);
2280 }
2281
2282 /* Only for WM_LP. */
2283 static uint32_t ilk_compute_fbc_wm(const struct hsw_pipe_wm_parameters *params,
2284 uint32_t pri_val)
2285 {
2286 if (!params->active || !params->pri.enabled)
2287 return 0;
2288
2289 return ilk_wm_fbc(pri_val,
2290 params->pri.horiz_pixels,
2291 params->pri.bytes_per_pixel);
2292 }
2293
2294 static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
2295 {
2296 if (INTEL_INFO(dev)->gen >= 8)
2297 return 3072;
2298 else if (INTEL_INFO(dev)->gen >= 7)
2299 return 768;
2300 else
2301 return 512;
2302 }
2303
2304 /* Calculate the maximum primary/sprite plane watermark */
2305 static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
2306 int level,
2307 const struct intel_wm_config *config,
2308 enum intel_ddb_partitioning ddb_partitioning,
2309 bool is_sprite)
2310 {
2311 unsigned int fifo_size = ilk_display_fifo_size(dev);
2312 unsigned int max;
2313
2314 /* if sprites aren't enabled, sprites get nothing */
2315 if (is_sprite && !config->sprites_enabled)
2316 return 0;
2317
2318 /* HSW allows LP1+ watermarks even with multiple pipes */
2319 if (level == 0 || config->num_pipes_active > 1) {
2320 fifo_size /= INTEL_INFO(dev)->num_pipes;
2321
2322 /*
2323 * For some reason the non self refresh
2324 * FIFO size is only half of the self
2325 * refresh FIFO size on ILK/SNB.
2326 */
2327 if (INTEL_INFO(dev)->gen <= 6)
2328 fifo_size /= 2;
2329 }
2330
2331 if (config->sprites_enabled) {
2332 /* level 0 is always calculated with 1:1 split */
2333 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
2334 if (is_sprite)
2335 fifo_size *= 5;
2336 fifo_size /= 6;
2337 } else {
2338 fifo_size /= 2;
2339 }
2340 }
2341
2342 /* clamp to max that the registers can hold */
2343 if (INTEL_INFO(dev)->gen >= 8)
2344 max = level == 0 ? 255 : 2047;
2345 else if (INTEL_INFO(dev)->gen >= 7)
2346 /* IVB/HSW primary/sprite plane watermarks */
2347 max = level == 0 ? 127 : 1023;
2348 else if (!is_sprite)
2349 /* ILK/SNB primary plane watermarks */
2350 max = level == 0 ? 127 : 511;
2351 else
2352 /* ILK/SNB sprite plane watermarks */
2353 max = level == 0 ? 63 : 255;
2354
2355 return min(fifo_size, max);
2356 }
2357
2358 /* Calculate the maximum cursor plane watermark */
2359 static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
2360 int level,
2361 const struct intel_wm_config *config)
2362 {
2363 /* HSW LP1+ watermarks w/ multiple pipes */
2364 if (level > 0 && config->num_pipes_active > 1)
2365 return 64;
2366
2367 /* otherwise just report max that registers can hold */
2368 if (INTEL_INFO(dev)->gen >= 7)
2369 return level == 0 ? 63 : 255;
2370 else
2371 return level == 0 ? 31 : 63;
2372 }
2373
2374 /* Calculate the maximum FBC watermark */
2375 static unsigned int ilk_fbc_wm_max(struct drm_device *dev)
2376 {
2377 /* max that registers can hold */
2378 if (INTEL_INFO(dev)->gen >= 8)
2379 return 31;
2380 else
2381 return 15;
2382 }
2383
2384 static void ilk_compute_wm_maximums(struct drm_device *dev,
2385 int level,
2386 const struct intel_wm_config *config,
2387 enum intel_ddb_partitioning ddb_partitioning,
2388 struct hsw_wm_maximums *max)
2389 {
2390 max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
2391 max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
2392 max->cur = ilk_cursor_wm_max(dev, level, config);
2393 max->fbc = ilk_fbc_wm_max(dev);
2394 }
2395
2396 static bool ilk_validate_wm_level(int level,
2397 const struct hsw_wm_maximums *max,
2398 struct intel_wm_level *result)
2399 {
2400 bool ret;
2401
2402 /* already determined to be invalid? */
2403 if (!result->enable)
2404 return false;
2405
2406 result->enable = result->pri_val <= max->pri &&
2407 result->spr_val <= max->spr &&
2408 result->cur_val <= max->cur;
2409
2410 ret = result->enable;
2411
2412 /*
2413 * HACK until we can pre-compute everything,
2414 * and thus fail gracefully if LP0 watermarks
2415 * are exceeded...
2416 */
2417 if (level == 0 && !result->enable) {
2418 if (result->pri_val > max->pri)
2419 DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
2420 level, result->pri_val, max->pri);
2421 if (result->spr_val > max->spr)
2422 DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
2423 level, result->spr_val, max->spr);
2424 if (result->cur_val > max->cur)
2425 DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
2426 level, result->cur_val, max->cur);
2427
2428 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
2429 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
2430 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
2431 result->enable = true;
2432 }
2433
2434 return ret;
2435 }
2436
2437 static void ilk_compute_wm_level(struct drm_i915_private *dev_priv,
2438 int level,
2439 const struct hsw_pipe_wm_parameters *p,
2440 struct intel_wm_level *result)
2441 {
2442 uint16_t pri_latency = dev_priv->wm.pri_latency[level];
2443 uint16_t spr_latency = dev_priv->wm.spr_latency[level];
2444 uint16_t cur_latency = dev_priv->wm.cur_latency[level];
2445
2446 /* WM1+ latency values stored in 0.5us units */
2447 if (level > 0) {
2448 pri_latency *= 5;
2449 spr_latency *= 5;
2450 cur_latency *= 5;
2451 }
2452
2453 result->pri_val = ilk_compute_pri_wm(p, pri_latency, level);
2454 result->spr_val = ilk_compute_spr_wm(p, spr_latency);
2455 result->cur_val = ilk_compute_cur_wm(p, cur_latency);
2456 result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val);
2457 result->enable = true;
2458 }
2459
2460 static uint32_t
2461 hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
2462 {
2463 struct drm_i915_private *dev_priv = dev->dev_private;
2464 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2465 struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode;
2466 u32 linetime, ips_linetime;
2467
2468 if (!intel_crtc_active(crtc))
2469 return 0;
2470
2471 /* The WM are computed with base on how long it takes to fill a single
2472 * row at the given clock rate, multiplied by 8.
2473 * */
2474 linetime = DIV_ROUND_CLOSEST(mode->htotal * 1000 * 8, mode->clock);
2475 ips_linetime = DIV_ROUND_CLOSEST(mode->htotal * 1000 * 8,
2476 intel_ddi_get_cdclk_freq(dev_priv));
2477
2478 return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
2479 PIPE_WM_LINETIME_TIME(linetime);
2480 }
2481
2482 static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[5])
2483 {
2484 struct drm_i915_private *dev_priv = dev->dev_private;
2485
2486 if (IS_HASWELL(dev)) {
2487 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2488
2489 wm[0] = (sskpd >> 56) & 0xFF;
2490 if (wm[0] == 0)
2491 wm[0] = sskpd & 0xF;
2492 wm[1] = (sskpd >> 4) & 0xFF;
2493 wm[2] = (sskpd >> 12) & 0xFF;
2494 wm[3] = (sskpd >> 20) & 0x1FF;
2495 wm[4] = (sskpd >> 32) & 0x1FF;
2496 } else if (INTEL_INFO(dev)->gen >= 6) {
2497 uint32_t sskpd = I915_READ(MCH_SSKPD);
2498
2499 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2500 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2501 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2502 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
2503 } else if (INTEL_INFO(dev)->gen >= 5) {
2504 uint32_t mltr = I915_READ(MLTR_ILK);
2505
2506 /* ILK primary LP0 latency is 700 ns */
2507 wm[0] = 7;
2508 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2509 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
2510 }
2511 }
2512
2513 static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2514 {
2515 /* ILK sprite LP0 latency is 1300 ns */
2516 if (INTEL_INFO(dev)->gen == 5)
2517 wm[0] = 13;
2518 }
2519
2520 static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2521 {
2522 /* ILK cursor LP0 latency is 1300 ns */
2523 if (INTEL_INFO(dev)->gen == 5)
2524 wm[0] = 13;
2525
2526 /* WaDoubleCursorLP3Latency:ivb */
2527 if (IS_IVYBRIDGE(dev))
2528 wm[3] *= 2;
2529 }
2530
2531 static int ilk_wm_max_level(const struct drm_device *dev)
2532 {
2533 /* how many WM levels are we expecting */
2534 if (IS_HASWELL(dev))
2535 return 4;
2536 else if (INTEL_INFO(dev)->gen >= 6)
2537 return 3;
2538 else
2539 return 2;
2540 }
2541
2542 static void intel_print_wm_latency(struct drm_device *dev,
2543 const char *name,
2544 const uint16_t wm[5])
2545 {
2546 int level, max_level = ilk_wm_max_level(dev);
2547
2548 for (level = 0; level <= max_level; level++) {
2549 unsigned int latency = wm[level];
2550
2551 if (latency == 0) {
2552 DRM_ERROR("%s WM%d latency not provided\n",
2553 name, level);
2554 continue;
2555 }
2556
2557 /* WM1+ latency values in 0.5us units */
2558 if (level > 0)
2559 latency *= 5;
2560
2561 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2562 name, level, wm[level],
2563 latency / 10, latency % 10);
2564 }
2565 }
2566
2567 static void intel_setup_wm_latency(struct drm_device *dev)
2568 {
2569 struct drm_i915_private *dev_priv = dev->dev_private;
2570
2571 intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2572
2573 memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2574 sizeof(dev_priv->wm.pri_latency));
2575 memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2576 sizeof(dev_priv->wm.pri_latency));
2577
2578 intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2579 intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
2580
2581 intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2582 intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2583 intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
2584 }
2585
2586 static void hsw_compute_wm_parameters(struct drm_crtc *crtc,
2587 struct hsw_pipe_wm_parameters *p,
2588 struct intel_wm_config *config)
2589 {
2590 struct drm_device *dev = crtc->dev;
2591 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2592 enum pipe pipe = intel_crtc->pipe;
2593 struct drm_plane *plane;
2594
2595 p->active = intel_crtc_active(crtc);
2596 if (p->active) {
2597 p->pipe_htotal = intel_crtc->config.adjusted_mode.htotal;
2598 p->pixel_rate = ilk_pipe_pixel_rate(dev, crtc);
2599 p->pri.bytes_per_pixel = crtc->fb->bits_per_pixel / 8;
2600 p->cur.bytes_per_pixel = 4;
2601 p->pri.horiz_pixels = intel_crtc->config.pipe_src_w;
2602 p->cur.horiz_pixels = 64;
2603 /* TODO: for now, assume primary and cursor planes are always enabled. */
2604 p->pri.enabled = true;
2605 p->cur.enabled = true;
2606 }
2607
2608 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
2609 config->num_pipes_active += intel_crtc_active(crtc);
2610
2611 list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
2612 struct intel_plane *intel_plane = to_intel_plane(plane);
2613
2614 if (intel_plane->pipe == pipe)
2615 p->spr = intel_plane->wm;
2616
2617 config->sprites_enabled |= intel_plane->wm.enabled;
2618 config->sprites_scaled |= intel_plane->wm.scaled;
2619 }
2620 }
2621
2622 /* Compute new watermarks for the pipe */
2623 static bool intel_compute_pipe_wm(struct drm_crtc *crtc,
2624 const struct hsw_pipe_wm_parameters *params,
2625 struct intel_pipe_wm *pipe_wm)
2626 {
2627 struct drm_device *dev = crtc->dev;
2628 struct drm_i915_private *dev_priv = dev->dev_private;
2629 int level, max_level = ilk_wm_max_level(dev);
2630 /* LP0 watermark maximums depend on this pipe alone */
2631 struct intel_wm_config config = {
2632 .num_pipes_active = 1,
2633 .sprites_enabled = params->spr.enabled,
2634 .sprites_scaled = params->spr.scaled,
2635 };
2636 struct hsw_wm_maximums max;
2637
2638 /* LP0 watermarks always use 1/2 DDB partitioning */
2639 ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
2640
2641 for (level = 0; level <= max_level; level++)
2642 ilk_compute_wm_level(dev_priv, level, params,
2643 &pipe_wm->wm[level]);
2644
2645 pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
2646
2647 /* At least LP0 must be valid */
2648 return ilk_validate_wm_level(0, &max, &pipe_wm->wm[0]);
2649 }
2650
2651 /*
2652 * Merge the watermarks from all active pipes for a specific level.
2653 */
2654 static void ilk_merge_wm_level(struct drm_device *dev,
2655 int level,
2656 struct intel_wm_level *ret_wm)
2657 {
2658 const struct intel_crtc *intel_crtc;
2659
2660 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2661 const struct intel_wm_level *wm =
2662 &intel_crtc->wm.active.wm[level];
2663
2664 if (!wm->enable)
2665 return;
2666
2667 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2668 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2669 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2670 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2671 }
2672
2673 ret_wm->enable = true;
2674 }
2675
2676 /*
2677 * Merge all low power watermarks for all active pipes.
2678 */
2679 static void ilk_wm_merge(struct drm_device *dev,
2680 const struct hsw_wm_maximums *max,
2681 struct intel_pipe_wm *merged)
2682 {
2683 int level, max_level = ilk_wm_max_level(dev);
2684
2685 merged->fbc_wm_enabled = true;
2686
2687 /* merge each WM1+ level */
2688 for (level = 1; level <= max_level; level++) {
2689 struct intel_wm_level *wm = &merged->wm[level];
2690
2691 ilk_merge_wm_level(dev, level, wm);
2692
2693 if (!ilk_validate_wm_level(level, max, wm))
2694 break;
2695
2696 /*
2697 * The spec says it is preferred to disable
2698 * FBC WMs instead of disabling a WM level.
2699 */
2700 if (wm->fbc_val > max->fbc) {
2701 merged->fbc_wm_enabled = false;
2702 wm->fbc_val = 0;
2703 }
2704 }
2705 }
2706
2707 static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2708 {
2709 /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2710 return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2711 }
2712
2713 static void hsw_compute_wm_results(struct drm_device *dev,
2714 const struct intel_pipe_wm *merged,
2715 enum intel_ddb_partitioning partitioning,
2716 struct hsw_wm_values *results)
2717 {
2718 struct intel_crtc *intel_crtc;
2719 int level, wm_lp;
2720
2721 results->enable_fbc_wm = merged->fbc_wm_enabled;
2722 results->partitioning = partitioning;
2723
2724 /* LP1+ register values */
2725 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2726 const struct intel_wm_level *r;
2727
2728 level = ilk_wm_lp_to_level(wm_lp, merged);
2729
2730 r = &merged->wm[level];
2731 if (!r->enable)
2732 break;
2733
2734 results->wm_lp[wm_lp - 1] = WM3_LP_EN |
2735 ((level * 2) << WM1_LP_LATENCY_SHIFT) |
2736 (r->pri_val << WM1_LP_SR_SHIFT) |
2737 r->cur_val;
2738
2739 if (INTEL_INFO(dev)->gen >= 8)
2740 results->wm_lp[wm_lp - 1] |=
2741 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
2742 else
2743 results->wm_lp[wm_lp - 1] |=
2744 r->fbc_val << WM1_LP_FBC_SHIFT;
2745
2746 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
2747 }
2748
2749 /* LP0 register values */
2750 list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2751 enum pipe pipe = intel_crtc->pipe;
2752 const struct intel_wm_level *r =
2753 &intel_crtc->wm.active.wm[0];
2754
2755 if (WARN_ON(!r->enable))
2756 continue;
2757
2758 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
2759
2760 results->wm_pipe[pipe] =
2761 (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2762 (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2763 r->cur_val;
2764 }
2765 }
2766
2767 /* Find the result with the highest level enabled. Check for enable_fbc_wm in
2768 * case both are at the same level. Prefer r1 in case they're the same. */
2769 static struct intel_pipe_wm *hsw_find_best_result(struct drm_device *dev,
2770 struct intel_pipe_wm *r1,
2771 struct intel_pipe_wm *r2)
2772 {
2773 int level, max_level = ilk_wm_max_level(dev);
2774 int level1 = 0, level2 = 0;
2775
2776 for (level = 1; level <= max_level; level++) {
2777 if (r1->wm[level].enable)
2778 level1 = level;
2779 if (r2->wm[level].enable)
2780 level2 = level;
2781 }
2782
2783 if (level1 == level2) {
2784 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
2785 return r2;
2786 else
2787 return r1;
2788 } else if (level1 > level2) {
2789 return r1;
2790 } else {
2791 return r2;
2792 }
2793 }
2794
2795 /* dirty bits used to track which watermarks need changes */
2796 #define WM_DIRTY_PIPE(pipe) (1 << (pipe))
2797 #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
2798 #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
2799 #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
2800 #define WM_DIRTY_FBC (1 << 24)
2801 #define WM_DIRTY_DDB (1 << 25)
2802
2803 static unsigned int ilk_compute_wm_dirty(struct drm_device *dev,
2804 const struct hsw_wm_values *old,
2805 const struct hsw_wm_values *new)
2806 {
2807 unsigned int dirty = 0;
2808 enum pipe pipe;
2809 int wm_lp;
2810
2811 for_each_pipe(pipe) {
2812 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
2813 dirty |= WM_DIRTY_LINETIME(pipe);
2814 /* Must disable LP1+ watermarks too */
2815 dirty |= WM_DIRTY_LP_ALL;
2816 }
2817
2818 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
2819 dirty |= WM_DIRTY_PIPE(pipe);
2820 /* Must disable LP1+ watermarks too */
2821 dirty |= WM_DIRTY_LP_ALL;
2822 }
2823 }
2824
2825 if (old->enable_fbc_wm != new->enable_fbc_wm) {
2826 dirty |= WM_DIRTY_FBC;
2827 /* Must disable LP1+ watermarks too */
2828 dirty |= WM_DIRTY_LP_ALL;
2829 }
2830
2831 if (old->partitioning != new->partitioning) {
2832 dirty |= WM_DIRTY_DDB;
2833 /* Must disable LP1+ watermarks too */
2834 dirty |= WM_DIRTY_LP_ALL;
2835 }
2836
2837 /* LP1+ watermarks already deemed dirty, no need to continue */
2838 if (dirty & WM_DIRTY_LP_ALL)
2839 return dirty;
2840
2841 /* Find the lowest numbered LP1+ watermark in need of an update... */
2842 for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2843 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
2844 old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
2845 break;
2846 }
2847
2848 /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
2849 for (; wm_lp <= 3; wm_lp++)
2850 dirty |= WM_DIRTY_LP(wm_lp);
2851
2852 return dirty;
2853 }
2854
2855 /*
2856 * The spec says we shouldn't write when we don't need, because every write
2857 * causes WMs to be re-evaluated, expending some power.
2858 */
2859 static void hsw_write_wm_values(struct drm_i915_private *dev_priv,
2860 struct hsw_wm_values *results)
2861 {
2862 struct hsw_wm_values *previous = &dev_priv->wm.hw;
2863 unsigned int dirty;
2864 uint32_t val;
2865
2866 dirty = ilk_compute_wm_dirty(dev_priv->dev, previous, results);
2867 if (!dirty)
2868 return;
2869
2870 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != 0)
2871 I915_WRITE(WM3_LP_ILK, 0);
2872 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != 0)
2873 I915_WRITE(WM2_LP_ILK, 0);
2874 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != 0)
2875 I915_WRITE(WM1_LP_ILK, 0);
2876
2877 if (dirty & WM_DIRTY_PIPE(PIPE_A))
2878 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
2879 if (dirty & WM_DIRTY_PIPE(PIPE_B))
2880 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
2881 if (dirty & WM_DIRTY_PIPE(PIPE_C))
2882 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2883
2884 if (dirty & WM_DIRTY_LINETIME(PIPE_A))
2885 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
2886 if (dirty & WM_DIRTY_LINETIME(PIPE_B))
2887 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
2888 if (dirty & WM_DIRTY_LINETIME(PIPE_C))
2889 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2890
2891 if (dirty & WM_DIRTY_DDB) {
2892 val = I915_READ(WM_MISC);
2893 if (results->partitioning == INTEL_DDB_PART_1_2)
2894 val &= ~WM_MISC_DATA_PARTITION_5_6;
2895 else
2896 val |= WM_MISC_DATA_PARTITION_5_6;
2897 I915_WRITE(WM_MISC, val);
2898 }
2899
2900 if (dirty & WM_DIRTY_FBC) {
2901 val = I915_READ(DISP_ARB_CTL);
2902 if (results->enable_fbc_wm)
2903 val &= ~DISP_FBC_WM_DIS;
2904 else
2905 val |= DISP_FBC_WM_DIS;
2906 I915_WRITE(DISP_ARB_CTL, val);
2907 }
2908
2909 if (dirty & WM_DIRTY_LP(1) && previous->wm_lp_spr[0] != results->wm_lp_spr[0])
2910 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2911 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
2912 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2913 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
2914 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2915
2916 if (dirty & WM_DIRTY_LP(1) && results->wm_lp[0] != 0)
2917 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
2918 if (dirty & WM_DIRTY_LP(2) && results->wm_lp[1] != 0)
2919 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
2920 if (dirty & WM_DIRTY_LP(3) && results->wm_lp[2] != 0)
2921 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
2922
2923 dev_priv->wm.hw = *results;
2924 }
2925
2926 static void haswell_update_wm(struct drm_crtc *crtc)
2927 {
2928 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2929 struct drm_device *dev = crtc->dev;
2930 struct drm_i915_private *dev_priv = dev->dev_private;
2931 struct hsw_wm_maximums max;
2932 struct hsw_pipe_wm_parameters params = {};
2933 struct hsw_wm_values results = {};
2934 enum intel_ddb_partitioning partitioning;
2935 struct intel_pipe_wm pipe_wm = {};
2936 struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
2937 struct intel_wm_config config = {};
2938
2939 hsw_compute_wm_parameters(crtc, &params, &config);
2940
2941 intel_compute_pipe_wm(crtc, &params, &pipe_wm);
2942
2943 if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
2944 return;
2945
2946 intel_crtc->wm.active = pipe_wm;
2947
2948 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
2949 ilk_wm_merge(dev, &max, &lp_wm_1_2);
2950
2951 /* 5/6 split only in single pipe config on IVB+ */
2952 if (INTEL_INFO(dev)->gen >= 7 &&
2953 config.num_pipes_active == 1 && config.sprites_enabled) {
2954 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
2955 ilk_wm_merge(dev, &max, &lp_wm_5_6);
2956
2957 best_lp_wm = hsw_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
2958 } else {
2959 best_lp_wm = &lp_wm_1_2;
2960 }
2961
2962 partitioning = (best_lp_wm == &lp_wm_1_2) ?
2963 INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
2964
2965 hsw_compute_wm_results(dev, best_lp_wm, partitioning, &results);
2966
2967 hsw_write_wm_values(dev_priv, &results);
2968 }
2969
2970 static void haswell_update_sprite_wm(struct drm_plane *plane,
2971 struct drm_crtc *crtc,
2972 uint32_t sprite_width, int pixel_size,
2973 bool enabled, bool scaled)
2974 {
2975 struct intel_plane *intel_plane = to_intel_plane(plane);
2976
2977 intel_plane->wm.enabled = enabled;
2978 intel_plane->wm.scaled = scaled;
2979 intel_plane->wm.horiz_pixels = sprite_width;
2980 intel_plane->wm.bytes_per_pixel = pixel_size;
2981
2982 haswell_update_wm(crtc);
2983 }
2984
2985 static bool
2986 sandybridge_compute_sprite_wm(struct drm_device *dev, int plane,
2987 uint32_t sprite_width, int pixel_size,
2988 const struct intel_watermark_params *display,
2989 int display_latency_ns, int *sprite_wm)
2990 {
2991 struct drm_crtc *crtc;
2992 int clock;
2993 int entries, tlb_miss;
2994
2995 crtc = intel_get_crtc_for_plane(dev, plane);
2996 if (!intel_crtc_active(crtc)) {
2997 *sprite_wm = display->guard_size;
2998 return false;
2999 }
3000
3001 clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
3002
3003 /* Use the small buffer method to calculate the sprite watermark */
3004 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
3005 tlb_miss = display->fifo_size*display->cacheline_size -
3006 sprite_width * 8;
3007 if (tlb_miss > 0)
3008 entries += tlb_miss;
3009 entries = DIV_ROUND_UP(entries, display->cacheline_size);
3010 *sprite_wm = entries + display->guard_size;
3011 if (*sprite_wm > (int)display->max_wm)
3012 *sprite_wm = display->max_wm;
3013
3014 return true;
3015 }
3016
3017 static bool
3018 sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane,
3019 uint32_t sprite_width, int pixel_size,
3020 const struct intel_watermark_params *display,
3021 int latency_ns, int *sprite_wm)
3022 {
3023 struct drm_crtc *crtc;
3024 unsigned long line_time_us;
3025 int clock;
3026 int line_count, line_size;
3027 int small, large;
3028 int entries;
3029
3030 if (!latency_ns) {
3031 *sprite_wm = 0;
3032 return false;
3033 }
3034
3035 crtc = intel_get_crtc_for_plane(dev, plane);
3036 clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
3037 if (!clock) {
3038 *sprite_wm = 0;
3039 return false;
3040 }
3041
3042 line_time_us = (sprite_width * 1000) / clock;
3043 if (!line_time_us) {
3044 *sprite_wm = 0;
3045 return false;
3046 }
3047
3048 line_count = (latency_ns / line_time_us + 1000) / 1000;
3049 line_size = sprite_width * pixel_size;
3050
3051 /* Use the minimum of the small and large buffer method for primary */
3052 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
3053 large = line_count * line_size;
3054
3055 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
3056 *sprite_wm = entries + display->guard_size;
3057
3058 return *sprite_wm > 0x3ff ? false : true;
3059 }
3060
3061 static void sandybridge_update_sprite_wm(struct drm_plane *plane,
3062 struct drm_crtc *crtc,
3063 uint32_t sprite_width, int pixel_size,
3064 bool enabled, bool scaled)
3065 {
3066 struct drm_device *dev = plane->dev;
3067 struct drm_i915_private *dev_priv = dev->dev_private;
3068 int pipe = to_intel_plane(plane)->pipe;
3069 int latency = dev_priv->wm.spr_latency[0] * 100; /* In unit 0.1us */
3070 u32 val;
3071 int sprite_wm, reg;
3072 int ret;
3073
3074 if (!enabled)
3075 return;
3076
3077 switch (pipe) {
3078 case 0:
3079 reg = WM0_PIPEA_ILK;
3080 break;
3081 case 1:
3082 reg = WM0_PIPEB_ILK;
3083 break;
3084 case 2:
3085 reg = WM0_PIPEC_IVB;
3086 break;
3087 default:
3088 return; /* bad pipe */
3089 }
3090
3091 ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size,
3092 &sandybridge_display_wm_info,
3093 latency, &sprite_wm);
3094 if (!ret) {
3095 DRM_DEBUG_KMS("failed to compute sprite wm for pipe %c\n",
3096 pipe_name(pipe));
3097 return;
3098 }
3099
3100 val = I915_READ(reg);
3101 val &= ~WM0_PIPE_SPRITE_MASK;
3102 I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT));
3103 DRM_DEBUG_KMS("sprite watermarks For pipe %c - %d\n", pipe_name(pipe), sprite_wm);
3104
3105
3106 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
3107 pixel_size,
3108 &sandybridge_display_srwm_info,
3109 dev_priv->wm.spr_latency[1] * 500,
3110 &sprite_wm);
3111 if (!ret) {
3112 DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %c\n",
3113 pipe_name(pipe));
3114 return;
3115 }
3116 I915_WRITE(WM1S_LP_ILK, sprite_wm);
3117
3118 /* Only IVB has two more LP watermarks for sprite */
3119 if (!IS_IVYBRIDGE(dev))
3120 return;
3121
3122 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
3123 pixel_size,
3124 &sandybridge_display_srwm_info,
3125 dev_priv->wm.spr_latency[2] * 500,
3126 &sprite_wm);
3127 if (!ret) {
3128 DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %c\n",
3129 pipe_name(pipe));
3130 return;
3131 }
3132 I915_WRITE(WM2S_LP_IVB, sprite_wm);
3133
3134 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
3135 pixel_size,
3136 &sandybridge_display_srwm_info,
3137 dev_priv->wm.spr_latency[3] * 500,
3138 &sprite_wm);
3139 if (!ret) {
3140 DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %c\n",
3141 pipe_name(pipe));
3142 return;
3143 }
3144 I915_WRITE(WM3S_LP_IVB, sprite_wm);
3145 }
3146
3147 static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
3148 {
3149 struct drm_device *dev = crtc->dev;
3150 struct drm_i915_private *dev_priv = dev->dev_private;
3151 struct hsw_wm_values *hw = &dev_priv->wm.hw;
3152 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
3153 struct intel_pipe_wm *active = &intel_crtc->wm.active;
3154 enum pipe pipe = intel_crtc->pipe;
3155 static const unsigned int wm0_pipe_reg[] = {
3156 [PIPE_A] = WM0_PIPEA_ILK,
3157 [PIPE_B] = WM0_PIPEB_ILK,
3158 [PIPE_C] = WM0_PIPEC_IVB,
3159 };
3160
3161 hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
3162 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
3163
3164 if (intel_crtc_active(crtc)) {
3165 u32 tmp = hw->wm_pipe[pipe];
3166
3167 /*
3168 * For active pipes LP0 watermark is marked as
3169 * enabled, and LP1+ watermaks as disabled since
3170 * we can't really reverse compute them in case
3171 * multiple pipes are active.
3172 */
3173 active->wm[0].enable = true;
3174 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
3175 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
3176 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
3177 active->linetime = hw->wm_linetime[pipe];
3178 } else {
3179 int level, max_level = ilk_wm_max_level(dev);
3180
3181 /*
3182 * For inactive pipes, all watermark levels
3183 * should be marked as enabled but zeroed,
3184 * which is what we'd compute them to.
3185 */
3186 for (level = 0; level <= max_level; level++)
3187 active->wm[level].enable = true;
3188 }
3189 }
3190
3191 void ilk_wm_get_hw_state(struct drm_device *dev)
3192 {
3193 struct drm_i915_private *dev_priv = dev->dev_private;
3194 struct hsw_wm_values *hw = &dev_priv->wm.hw;
3195 struct drm_crtc *crtc;
3196
3197 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
3198 ilk_pipe_wm_get_hw_state(crtc);
3199
3200 hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
3201 hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
3202 hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
3203
3204 hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
3205 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
3206 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
3207
3208 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
3209 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
3210
3211 hw->enable_fbc_wm =
3212 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
3213 }
3214
3215 /**
3216 * intel_update_watermarks - update FIFO watermark values based on current modes
3217 *
3218 * Calculate watermark values for the various WM regs based on current mode
3219 * and plane configuration.
3220 *
3221 * There are several cases to deal with here:
3222 * - normal (i.e. non-self-refresh)
3223 * - self-refresh (SR) mode
3224 * - lines are large relative to FIFO size (buffer can hold up to 2)
3225 * - lines are small relative to FIFO size (buffer can hold more than 2
3226 * lines), so need to account for TLB latency
3227 *
3228 * The normal calculation is:
3229 * watermark = dotclock * bytes per pixel * latency
3230 * where latency is platform & configuration dependent (we assume pessimal
3231 * values here).
3232 *
3233 * The SR calculation is:
3234 * watermark = (trunc(latency/line time)+1) * surface width *
3235 * bytes per pixel
3236 * where
3237 * line time = htotal / dotclock
3238 * surface width = hdisplay for normal plane and 64 for cursor
3239 * and latency is assumed to be high, as above.
3240 *
3241 * The final value programmed to the register should always be rounded up,
3242 * and include an extra 2 entries to account for clock crossings.
3243 *
3244 * We don't use the sprite, so we can ignore that. And on Crestline we have
3245 * to set the non-SR watermarks to 8.
3246 */
3247 void intel_update_watermarks(struct drm_crtc *crtc)
3248 {
3249 struct drm_i915_private *dev_priv = crtc->dev->dev_private;
3250
3251 if (dev_priv->display.update_wm)
3252 dev_priv->display.update_wm(crtc);
3253 }
3254
3255 void intel_update_sprite_watermarks(struct drm_plane *plane,
3256 struct drm_crtc *crtc,
3257 uint32_t sprite_width, int pixel_size,
3258 bool enabled, bool scaled)
3259 {
3260 struct drm_i915_private *dev_priv = plane->dev->dev_private;
3261
3262 if (dev_priv->display.update_sprite_wm)
3263 dev_priv->display.update_sprite_wm(plane, crtc, sprite_width,
3264 pixel_size, enabled, scaled);
3265 }
3266
3267 static struct drm_i915_gem_object *
3268 intel_alloc_context_page(struct drm_device *dev)
3269 {
3270 struct drm_i915_gem_object *ctx;
3271 int ret;
3272
3273 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3274
3275 ctx = i915_gem_alloc_object(dev, 4096);
3276 if (!ctx) {
3277 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
3278 return NULL;
3279 }
3280
3281 ret = i915_gem_obj_ggtt_pin(ctx, 4096, true, false);
3282 if (ret) {
3283 DRM_ERROR("failed to pin power context: %d\n", ret);
3284 goto err_unref;
3285 }
3286
3287 ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
3288 if (ret) {
3289 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
3290 goto err_unpin;
3291 }
3292
3293 return ctx;
3294
3295 err_unpin:
3296 i915_gem_object_unpin(ctx);
3297 err_unref:
3298 drm_gem_object_unreference(&ctx->base);
3299 return NULL;
3300 }
3301
3302 /**
3303 * Lock protecting IPS related data structures
3304 */
3305 DEFINE_SPINLOCK(mchdev_lock);
3306
3307 /* Global for IPS driver to get at the current i915 device. Protected by
3308 * mchdev_lock. */
3309 static struct drm_i915_private *i915_mch_dev;
3310
3311 bool ironlake_set_drps(struct drm_device *dev, u8 val)
3312 {
3313 struct drm_i915_private *dev_priv = dev->dev_private;
3314 u16 rgvswctl;
3315
3316 assert_spin_locked(&mchdev_lock);
3317
3318 rgvswctl = I915_READ16(MEMSWCTL);
3319 if (rgvswctl & MEMCTL_CMD_STS) {
3320 DRM_DEBUG("gpu busy, RCS change rejected\n");
3321 return false; /* still busy with another command */
3322 }
3323
3324 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
3325 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
3326 I915_WRITE16(MEMSWCTL, rgvswctl);
3327 POSTING_READ16(MEMSWCTL);
3328
3329 rgvswctl |= MEMCTL_CMD_STS;
3330 I915_WRITE16(MEMSWCTL, rgvswctl);
3331
3332 return true;
3333 }
3334
3335 static void ironlake_enable_drps(struct drm_device *dev)
3336 {
3337 struct drm_i915_private *dev_priv = dev->dev_private;
3338 u32 rgvmodectl = I915_READ(MEMMODECTL);
3339 u8 fmax, fmin, fstart, vstart;
3340
3341 spin_lock_irq(&mchdev_lock);
3342
3343 /* Enable temp reporting */
3344 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
3345 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
3346
3347 /* 100ms RC evaluation intervals */
3348 I915_WRITE(RCUPEI, 100000);
3349 I915_WRITE(RCDNEI, 100000);
3350
3351 /* Set max/min thresholds to 90ms and 80ms respectively */
3352 I915_WRITE(RCBMAXAVG, 90000);
3353 I915_WRITE(RCBMINAVG, 80000);
3354
3355 I915_WRITE(MEMIHYST, 1);
3356
3357 /* Set up min, max, and cur for interrupt handling */
3358 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
3359 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
3360 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
3361 MEMMODE_FSTART_SHIFT;
3362
3363 vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
3364 PXVFREQ_PX_SHIFT;
3365
3366 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
3367 dev_priv->ips.fstart = fstart;
3368
3369 dev_priv->ips.max_delay = fstart;
3370 dev_priv->ips.min_delay = fmin;
3371 dev_priv->ips.cur_delay = fstart;
3372
3373 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
3374 fmax, fmin, fstart);
3375
3376 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
3377
3378 /*
3379 * Interrupts will be enabled in ironlake_irq_postinstall
3380 */
3381
3382 I915_WRITE(VIDSTART, vstart);
3383 POSTING_READ(VIDSTART);
3384
3385 rgvmodectl |= MEMMODE_SWMODE_EN;
3386 I915_WRITE(MEMMODECTL, rgvmodectl);
3387
3388 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
3389 DRM_ERROR("stuck trying to change perf mode\n");
3390 mdelay(1);
3391
3392 ironlake_set_drps(dev, fstart);
3393
3394 dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
3395 I915_READ(0x112e0);
3396 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
3397 dev_priv->ips.last_count2 = I915_READ(0x112f4);
3398 getrawmonotonic(&dev_priv->ips.last_time2);
3399
3400 spin_unlock_irq(&mchdev_lock);
3401 }
3402
3403 static void ironlake_disable_drps(struct drm_device *dev)
3404 {
3405 struct drm_i915_private *dev_priv = dev->dev_private;
3406 u16 rgvswctl;
3407
3408 spin_lock_irq(&mchdev_lock);
3409
3410 rgvswctl = I915_READ16(MEMSWCTL);
3411
3412 /* Ack interrupts, disable EFC interrupt */
3413 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
3414 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
3415 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
3416 I915_WRITE(DEIIR, DE_PCU_EVENT);
3417 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
3418
3419 /* Go back to the starting frequency */
3420 ironlake_set_drps(dev, dev_priv->ips.fstart);
3421 mdelay(1);
3422 rgvswctl |= MEMCTL_CMD_STS;
3423 I915_WRITE(MEMSWCTL, rgvswctl);
3424 mdelay(1);
3425
3426 spin_unlock_irq(&mchdev_lock);
3427 }
3428
3429 /* There's a funny hw issue where the hw returns all 0 when reading from
3430 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
3431 * ourselves, instead of doing a rmw cycle (which might result in us clearing
3432 * all limits and the gpu stuck at whatever frequency it is at atm).
3433 */
3434 static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 val)
3435 {
3436 u32 limits;
3437
3438 /* Only set the down limit when we've reached the lowest level to avoid
3439 * getting more interrupts, otherwise leave this clear. This prevents a
3440 * race in the hw when coming out of rc6: There's a tiny window where
3441 * the hw runs at the minimal clock before selecting the desired
3442 * frequency, if the down threshold expires in that window we will not
3443 * receive a down interrupt. */
3444 limits = dev_priv->rps.max_delay << 24;
3445 if (val <= dev_priv->rps.min_delay)
3446 limits |= dev_priv->rps.min_delay << 16;
3447
3448 return limits;
3449 }
3450
3451 static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
3452 {
3453 int new_power;
3454
3455 new_power = dev_priv->rps.power;
3456 switch (dev_priv->rps.power) {
3457 case LOW_POWER:
3458 if (val > dev_priv->rps.rpe_delay + 1 && val > dev_priv->rps.cur_delay)
3459 new_power = BETWEEN;
3460 break;
3461
3462 case BETWEEN:
3463 if (val <= dev_priv->rps.rpe_delay && val < dev_priv->rps.cur_delay)
3464 new_power = LOW_POWER;
3465 else if (val >= dev_priv->rps.rp0_delay && val > dev_priv->rps.cur_delay)
3466 new_power = HIGH_POWER;
3467 break;
3468
3469 case HIGH_POWER:
3470 if (val < (dev_priv->rps.rp1_delay + dev_priv->rps.rp0_delay) >> 1 && val < dev_priv->rps.cur_delay)
3471 new_power = BETWEEN;
3472 break;
3473 }
3474 /* Max/min bins are special */
3475 if (val == dev_priv->rps.min_delay)
3476 new_power = LOW_POWER;
3477 if (val == dev_priv->rps.max_delay)
3478 new_power = HIGH_POWER;
3479 if (new_power == dev_priv->rps.power)
3480 return;
3481
3482 /* Note the units here are not exactly 1us, but 1280ns. */
3483 switch (new_power) {
3484 case LOW_POWER:
3485 /* Upclock if more than 95% busy over 16ms */
3486 I915_WRITE(GEN6_RP_UP_EI, 12500);
3487 I915_WRITE(GEN6_RP_UP_THRESHOLD, 11800);
3488
3489 /* Downclock if less than 85% busy over 32ms */
3490 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3491 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 21250);
3492
3493 I915_WRITE(GEN6_RP_CONTROL,
3494 GEN6_RP_MEDIA_TURBO |
3495 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3496 GEN6_RP_MEDIA_IS_GFX |
3497 GEN6_RP_ENABLE |
3498 GEN6_RP_UP_BUSY_AVG |
3499 GEN6_RP_DOWN_IDLE_AVG);
3500 break;
3501
3502 case BETWEEN:
3503 /* Upclock if more than 90% busy over 13ms */
3504 I915_WRITE(GEN6_RP_UP_EI, 10250);
3505 I915_WRITE(GEN6_RP_UP_THRESHOLD, 9225);
3506
3507 /* Downclock if less than 75% busy over 32ms */
3508 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3509 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 18750);
3510
3511 I915_WRITE(GEN6_RP_CONTROL,
3512 GEN6_RP_MEDIA_TURBO |
3513 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3514 GEN6_RP_MEDIA_IS_GFX |
3515 GEN6_RP_ENABLE |
3516 GEN6_RP_UP_BUSY_AVG |
3517 GEN6_RP_DOWN_IDLE_AVG);
3518 break;
3519
3520 case HIGH_POWER:
3521 /* Upclock if more than 85% busy over 10ms */
3522 I915_WRITE(GEN6_RP_UP_EI, 8000);
3523 I915_WRITE(GEN6_RP_UP_THRESHOLD, 6800);
3524
3525 /* Downclock if less than 60% busy over 32ms */
3526 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3527 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 15000);
3528
3529 I915_WRITE(GEN6_RP_CONTROL,
3530 GEN6_RP_MEDIA_TURBO |
3531 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3532 GEN6_RP_MEDIA_IS_GFX |
3533 GEN6_RP_ENABLE |
3534 GEN6_RP_UP_BUSY_AVG |
3535 GEN6_RP_DOWN_IDLE_AVG);
3536 break;
3537 }
3538
3539 dev_priv->rps.power = new_power;
3540 dev_priv->rps.last_adj = 0;
3541 }
3542
3543 void gen6_set_rps(struct drm_device *dev, u8 val)
3544 {
3545 struct drm_i915_private *dev_priv = dev->dev_private;
3546
3547 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3548 WARN_ON(val > dev_priv->rps.max_delay);
3549 WARN_ON(val < dev_priv->rps.min_delay);
3550
3551 if (val == dev_priv->rps.cur_delay)
3552 return;
3553
3554 gen6_set_rps_thresholds(dev_priv, val);
3555
3556 if (IS_HASWELL(dev))
3557 I915_WRITE(GEN6_RPNSWREQ,
3558 HSW_FREQUENCY(val));
3559 else
3560 I915_WRITE(GEN6_RPNSWREQ,
3561 GEN6_FREQUENCY(val) |
3562 GEN6_OFFSET(0) |
3563 GEN6_AGGRESSIVE_TURBO);
3564
3565 /* Make sure we continue to get interrupts
3566 * until we hit the minimum or maximum frequencies.
3567 */
3568 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
3569 gen6_rps_limits(dev_priv, val));
3570
3571 POSTING_READ(GEN6_RPNSWREQ);
3572
3573 dev_priv->rps.cur_delay = val;
3574
3575 trace_intel_gpu_freq_change(val * 50);
3576 }
3577
3578 void gen6_rps_idle(struct drm_i915_private *dev_priv)
3579 {
3580 mutex_lock(&dev_priv->rps.hw_lock);
3581 if (dev_priv->rps.enabled) {
3582 if (dev_priv->info->is_valleyview)
3583 valleyview_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3584 else
3585 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3586 dev_priv->rps.last_adj = 0;
3587 }
3588 mutex_unlock(&dev_priv->rps.hw_lock);
3589 }
3590
3591 void gen6_rps_boost(struct drm_i915_private *dev_priv)
3592 {
3593 mutex_lock(&dev_priv->rps.hw_lock);
3594 if (dev_priv->rps.enabled) {
3595 if (dev_priv->info->is_valleyview)
3596 valleyview_set_rps(dev_priv->dev, dev_priv->rps.max_delay);
3597 else
3598 gen6_set_rps(dev_priv->dev, dev_priv->rps.max_delay);
3599 dev_priv->rps.last_adj = 0;
3600 }
3601 mutex_unlock(&dev_priv->rps.hw_lock);
3602 }
3603
3604 void valleyview_set_rps(struct drm_device *dev, u8 val)
3605 {
3606 struct drm_i915_private *dev_priv = dev->dev_private;
3607
3608 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3609 WARN_ON(val > dev_priv->rps.max_delay);
3610 WARN_ON(val < dev_priv->rps.min_delay);
3611
3612 DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
3613 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_delay),
3614 dev_priv->rps.cur_delay,
3615 vlv_gpu_freq(dev_priv, val), val);
3616
3617 if (val == dev_priv->rps.cur_delay)
3618 return;
3619
3620 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
3621
3622 dev_priv->rps.cur_delay = val;
3623
3624 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv, val));
3625 }
3626
3627 static void gen6_disable_rps_interrupts(struct drm_device *dev)
3628 {
3629 struct drm_i915_private *dev_priv = dev->dev_private;
3630
3631 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
3632 I915_WRITE(GEN6_PMIER, I915_READ(GEN6_PMIER) & ~GEN6_PM_RPS_EVENTS);
3633 /* Complete PM interrupt masking here doesn't race with the rps work
3634 * item again unmasking PM interrupts because that is using a different
3635 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
3636 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
3637
3638 spin_lock_irq(&dev_priv->irq_lock);
3639 dev_priv->rps.pm_iir = 0;
3640 spin_unlock_irq(&dev_priv->irq_lock);
3641
3642 I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS);
3643 }
3644
3645 static void gen6_disable_rps(struct drm_device *dev)
3646 {
3647 struct drm_i915_private *dev_priv = dev->dev_private;
3648
3649 I915_WRITE(GEN6_RC_CONTROL, 0);
3650 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
3651
3652 gen6_disable_rps_interrupts(dev);
3653 }
3654
3655 static void valleyview_disable_rps(struct drm_device *dev)
3656 {
3657 struct drm_i915_private *dev_priv = dev->dev_private;
3658
3659 I915_WRITE(GEN6_RC_CONTROL, 0);
3660
3661 gen6_disable_rps_interrupts(dev);
3662
3663 if (dev_priv->vlv_pctx) {
3664 drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
3665 dev_priv->vlv_pctx = NULL;
3666 }
3667 }
3668
3669 static void intel_print_rc6_info(struct drm_device *dev, u32 mode)
3670 {
3671 if (IS_GEN6(dev))
3672 DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
3673
3674 if (IS_HASWELL(dev))
3675 DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
3676
3677 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
3678 (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
3679 (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
3680 (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
3681 }
3682
3683 int intel_enable_rc6(const struct drm_device *dev)
3684 {
3685 /* No RC6 before Ironlake */
3686 if (INTEL_INFO(dev)->gen < 5)
3687 return 0;
3688
3689 /* Respect the kernel parameter if it is set */
3690 if (i915_enable_rc6 >= 0)
3691 return i915_enable_rc6;
3692
3693 /* Disable RC6 on Ironlake */
3694 if (INTEL_INFO(dev)->gen == 5)
3695 return 0;
3696
3697 if (IS_HASWELL(dev))
3698 return INTEL_RC6_ENABLE;
3699
3700 /* snb/ivb have more than one rc6 state. */
3701 if (INTEL_INFO(dev)->gen == 6)
3702 return INTEL_RC6_ENABLE;
3703
3704 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
3705 }
3706
3707 static void gen6_enable_rps_interrupts(struct drm_device *dev)
3708 {
3709 struct drm_i915_private *dev_priv = dev->dev_private;
3710 u32 enabled_intrs;
3711
3712 spin_lock_irq(&dev_priv->irq_lock);
3713 WARN_ON(dev_priv->rps.pm_iir);
3714 snb_enable_pm_irq(dev_priv, GEN6_PM_RPS_EVENTS);
3715 I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS);
3716 spin_unlock_irq(&dev_priv->irq_lock);
3717
3718 /* only unmask PM interrupts we need. Mask all others. */
3719 enabled_intrs = GEN6_PM_RPS_EVENTS;
3720
3721 /* IVB and SNB hard hangs on looping batchbuffer
3722 * if GEN6_PM_UP_EI_EXPIRED is masked.
3723 */
3724 if (INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev))
3725 enabled_intrs |= GEN6_PM_RP_UP_EI_EXPIRED;
3726
3727 I915_WRITE(GEN6_PMINTRMSK, ~enabled_intrs);
3728 }
3729
3730 static void gen8_enable_rps(struct drm_device *dev)
3731 {
3732 struct drm_i915_private *dev_priv = dev->dev_private;
3733 struct intel_ring_buffer *ring;
3734 uint32_t rc6_mask = 0, rp_state_cap;
3735 int unused;
3736
3737 /* 1a: Software RC state - RC0 */
3738 I915_WRITE(GEN6_RC_STATE, 0);
3739
3740 /* 1c & 1d: Get forcewake during program sequence. Although the driver
3741 * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
3742 gen6_gt_force_wake_get(dev_priv);
3743
3744 /* 2a: Disable RC states. */
3745 I915_WRITE(GEN6_RC_CONTROL, 0);
3746
3747 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3748
3749 /* 2b: Program RC6 thresholds.*/
3750 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
3751 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
3752 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
3753 for_each_ring(ring, dev_priv, unused)
3754 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3755 I915_WRITE(GEN6_RC_SLEEP, 0);
3756 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
3757
3758 /* 3: Enable RC6 */
3759 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
3760 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
3761 DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off");
3762 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
3763 GEN6_RC_CTL_EI_MODE(1) |
3764 rc6_mask);
3765
3766 /* 4 Program defaults and thresholds for RPS*/
3767 I915_WRITE(GEN6_RPNSWREQ, HSW_FREQUENCY(10)); /* Request 500 MHz */
3768 I915_WRITE(GEN6_RC_VIDEO_FREQ, HSW_FREQUENCY(12)); /* Request 600 MHz */
3769 /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
3770 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
3771
3772 /* Docs recommend 900MHz, and 300 MHz respectively */
3773 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
3774 dev_priv->rps.max_delay << 24 |
3775 dev_priv->rps.min_delay << 16);
3776
3777 I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
3778 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
3779 I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
3780 I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
3781
3782 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3783
3784 /* 5: Enable RPS */
3785 I915_WRITE(GEN6_RP_CONTROL,
3786 GEN6_RP_MEDIA_TURBO |
3787 GEN6_RP_MEDIA_HW_NORMAL_MODE |
3788 GEN6_RP_MEDIA_IS_GFX |
3789 GEN6_RP_ENABLE |
3790 GEN6_RP_UP_BUSY_AVG |
3791 GEN6_RP_DOWN_IDLE_AVG);
3792
3793 /* 6: Ring frequency + overclocking (our driver does this later */
3794
3795 gen6_set_rps(dev, (I915_READ(GEN6_GT_PERF_STATUS) & 0xff00) >> 8);
3796
3797 gen6_enable_rps_interrupts(dev);
3798
3799 gen6_gt_force_wake_put(dev_priv);
3800 }
3801
3802 static void gen6_enable_rps(struct drm_device *dev)
3803 {
3804 struct drm_i915_private *dev_priv = dev->dev_private;
3805 struct intel_ring_buffer *ring;
3806 u32 rp_state_cap;
3807 u32 gt_perf_status;
3808 u32 rc6vids, pcu_mbox, rc6_mask = 0;
3809 u32 gtfifodbg;
3810 int rc6_mode;
3811 int i, ret;
3812
3813 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3814
3815 /* Here begins a magic sequence of register writes to enable
3816 * auto-downclocking.
3817 *
3818 * Perhaps there might be some value in exposing these to
3819 * userspace...
3820 */
3821 I915_WRITE(GEN6_RC_STATE, 0);
3822
3823 /* Clear the DBG now so we don't confuse earlier errors */
3824 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
3825 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
3826 I915_WRITE(GTFIFODBG, gtfifodbg);
3827 }
3828
3829 gen6_gt_force_wake_get(dev_priv);
3830
3831 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3832 gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
3833
3834 /* In units of 50MHz */
3835 dev_priv->rps.hw_max = dev_priv->rps.max_delay = rp_state_cap & 0xff;
3836 dev_priv->rps.min_delay = (rp_state_cap >> 16) & 0xff;
3837 dev_priv->rps.rp1_delay = (rp_state_cap >> 8) & 0xff;
3838 dev_priv->rps.rp0_delay = (rp_state_cap >> 0) & 0xff;
3839 dev_priv->rps.rpe_delay = dev_priv->rps.rp1_delay;
3840 dev_priv->rps.cur_delay = 0;
3841
3842 /* disable the counters and set deterministic thresholds */
3843 I915_WRITE(GEN6_RC_CONTROL, 0);
3844
3845 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
3846 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
3847 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
3848 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3849 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3850
3851 for_each_ring(ring, dev_priv, i)
3852 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3853
3854 I915_WRITE(GEN6_RC_SLEEP, 0);
3855 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
3856 if (INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev))
3857 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
3858 else
3859 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
3860 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
3861 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
3862
3863 /* Check if we are enabling RC6 */
3864 rc6_mode = intel_enable_rc6(dev_priv->dev);
3865 if (rc6_mode & INTEL_RC6_ENABLE)
3866 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
3867
3868 /* We don't use those on Haswell */
3869 if (!IS_HASWELL(dev)) {
3870 if (rc6_mode & INTEL_RC6p_ENABLE)
3871 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
3872
3873 if (rc6_mode & INTEL_RC6pp_ENABLE)
3874 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
3875 }
3876
3877 intel_print_rc6_info(dev, rc6_mask);
3878
3879 I915_WRITE(GEN6_RC_CONTROL,
3880 rc6_mask |
3881 GEN6_RC_CTL_EI_MODE(1) |
3882 GEN6_RC_CTL_HW_ENABLE);
3883
3884 /* Power down if completely idle for over 50ms */
3885 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
3886 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3887
3888 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
3889 if (!ret) {
3890 pcu_mbox = 0;
3891 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
3892 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
3893 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
3894 (dev_priv->rps.max_delay & 0xff) * 50,
3895 (pcu_mbox & 0xff) * 50);
3896 dev_priv->rps.hw_max = pcu_mbox & 0xff;
3897 }
3898 } else {
3899 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
3900 }
3901
3902 dev_priv->rps.power = HIGH_POWER; /* force a reset */
3903 gen6_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3904
3905 gen6_enable_rps_interrupts(dev);
3906
3907 rc6vids = 0;
3908 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
3909 if (IS_GEN6(dev) && ret) {
3910 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
3911 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
3912 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
3913 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
3914 rc6vids &= 0xffff00;
3915 rc6vids |= GEN6_ENCODE_RC6_VID(450);
3916 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
3917 if (ret)
3918 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
3919 }
3920
3921 gen6_gt_force_wake_put(dev_priv);
3922 }
3923
3924 void gen6_update_ring_freq(struct drm_device *dev)
3925 {
3926 struct drm_i915_private *dev_priv = dev->dev_private;
3927 int min_freq = 15;
3928 unsigned int gpu_freq;
3929 unsigned int max_ia_freq, min_ring_freq;
3930 int scaling_factor = 180;
3931 struct cpufreq_policy *policy;
3932
3933 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3934
3935 policy = cpufreq_cpu_get(0);
3936 if (policy) {
3937 max_ia_freq = policy->cpuinfo.max_freq;
3938 cpufreq_cpu_put(policy);
3939 } else {
3940 /*
3941 * Default to measured freq if none found, PCU will ensure we
3942 * don't go over
3943 */
3944 max_ia_freq = tsc_khz;
3945 }
3946
3947 /* Convert from kHz to MHz */
3948 max_ia_freq /= 1000;
3949
3950 min_ring_freq = I915_READ(DCLK) & 0xf;
3951 /* convert DDR frequency from units of 266.6MHz to bandwidth */
3952 min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3953
3954 /*
3955 * For each potential GPU frequency, load a ring frequency we'd like
3956 * to use for memory access. We do this by specifying the IA frequency
3957 * the PCU should use as a reference to determine the ring frequency.
3958 */
3959 for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay;
3960 gpu_freq--) {
3961 int diff = dev_priv->rps.max_delay - gpu_freq;
3962 unsigned int ia_freq = 0, ring_freq = 0;
3963
3964 if (INTEL_INFO(dev)->gen >= 8) {
3965 /* max(2 * GT, DDR). NB: GT is 50MHz units */
3966 ring_freq = max(min_ring_freq, gpu_freq);
3967 } else if (IS_HASWELL(dev)) {
3968 ring_freq = mult_frac(gpu_freq, 5, 4);
3969 ring_freq = max(min_ring_freq, ring_freq);
3970 /* leave ia_freq as the default, chosen by cpufreq */
3971 } else {
3972 /* On older processors, there is no separate ring
3973 * clock domain, so in order to boost the bandwidth
3974 * of the ring, we need to upclock the CPU (ia_freq).
3975 *
3976 * For GPU frequencies less than 750MHz,
3977 * just use the lowest ring freq.
3978 */
3979 if (gpu_freq < min_freq)
3980 ia_freq = 800;
3981 else
3982 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
3983 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
3984 }
3985
3986 sandybridge_pcode_write(dev_priv,
3987 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3988 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
3989 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
3990 gpu_freq);
3991 }
3992 }
3993
3994 int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
3995 {
3996 u32 val, rp0;
3997
3998 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
3999
4000 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
4001 /* Clamp to max */
4002 rp0 = min_t(u32, rp0, 0xea);
4003
4004 return rp0;
4005 }
4006
4007 static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
4008 {
4009 u32 val, rpe;
4010
4011 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
4012 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
4013 val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
4014 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
4015
4016 return rpe;
4017 }
4018
4019 int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
4020 {
4021 return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
4022 }
4023
4024 static void valleyview_setup_pctx(struct drm_device *dev)
4025 {
4026 struct drm_i915_private *dev_priv = dev->dev_private;
4027 struct drm_i915_gem_object *pctx;
4028 unsigned long pctx_paddr;
4029 u32 pcbr;
4030 int pctx_size = 24*1024;
4031
4032 pcbr = I915_READ(VLV_PCBR);
4033 if (pcbr) {
4034 /* BIOS set it up already, grab the pre-alloc'd space */
4035 int pcbr_offset;
4036
4037 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
4038 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
4039 pcbr_offset,
4040 I915_GTT_OFFSET_NONE,
4041 pctx_size);
4042 goto out;
4043 }
4044
4045 /*
4046 * From the Gunit register HAS:
4047 * The Gfx driver is expected to program this register and ensure
4048 * proper allocation within Gfx stolen memory. For example, this
4049 * register should be programmed such than the PCBR range does not
4050 * overlap with other ranges, such as the frame buffer, protected
4051 * memory, or any other relevant ranges.
4052 */
4053 pctx = i915_gem_object_create_stolen(dev, pctx_size);
4054 if (!pctx) {
4055 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
4056 return;
4057 }
4058
4059 pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
4060 I915_WRITE(VLV_PCBR, pctx_paddr);
4061
4062 out:
4063 dev_priv->vlv_pctx = pctx;
4064 }
4065
4066 static void valleyview_enable_rps(struct drm_device *dev)
4067 {
4068 struct drm_i915_private *dev_priv = dev->dev_private;
4069 struct intel_ring_buffer *ring;
4070 u32 gtfifodbg, val, rc6_mode = 0;
4071 int i;
4072
4073 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4074
4075 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
4076 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
4077 gtfifodbg);
4078 I915_WRITE(GTFIFODBG, gtfifodbg);
4079 }
4080
4081 valleyview_setup_pctx(dev);
4082
4083 gen6_gt_force_wake_get(dev_priv);
4084
4085 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
4086 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
4087 I915_WRITE(GEN6_RP_UP_EI, 66000);
4088 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
4089
4090 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
4091
4092 I915_WRITE(GEN6_RP_CONTROL,
4093 GEN6_RP_MEDIA_TURBO |
4094 GEN6_RP_MEDIA_HW_NORMAL_MODE |
4095 GEN6_RP_MEDIA_IS_GFX |
4096 GEN6_RP_ENABLE |
4097 GEN6_RP_UP_BUSY_AVG |
4098 GEN6_RP_DOWN_IDLE_CONT);
4099
4100 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
4101 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
4102 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
4103
4104 for_each_ring(ring, dev_priv, i)
4105 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
4106
4107 I915_WRITE(GEN6_RC6_THRESHOLD, 0xc350);
4108
4109 /* allows RC6 residency counter to work */
4110 I915_WRITE(VLV_COUNTER_CONTROL,
4111 _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
4112 VLV_MEDIA_RC6_COUNT_EN |
4113 VLV_RENDER_RC6_COUNT_EN));
4114 if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
4115 rc6_mode = GEN7_RC_CTL_TO_MODE;
4116
4117 intel_print_rc6_info(dev, rc6_mode);
4118
4119 I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
4120
4121 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
4122
4123 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
4124 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
4125
4126 dev_priv->rps.cur_delay = (val >> 8) & 0xff;
4127 DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
4128 vlv_gpu_freq(dev_priv, dev_priv->rps.cur_delay),
4129 dev_priv->rps.cur_delay);
4130
4131 dev_priv->rps.max_delay = valleyview_rps_max_freq(dev_priv);
4132 dev_priv->rps.hw_max = dev_priv->rps.max_delay;
4133 DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
4134 vlv_gpu_freq(dev_priv, dev_priv->rps.max_delay),
4135 dev_priv->rps.max_delay);
4136
4137 dev_priv->rps.rpe_delay = valleyview_rps_rpe_freq(dev_priv);
4138 DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
4139 vlv_gpu_freq(dev_priv, dev_priv->rps.rpe_delay),
4140 dev_priv->rps.rpe_delay);
4141
4142 dev_priv->rps.min_delay = valleyview_rps_min_freq(dev_priv);
4143 DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
4144 vlv_gpu_freq(dev_priv, dev_priv->rps.min_delay),
4145 dev_priv->rps.min_delay);
4146
4147 DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
4148 vlv_gpu_freq(dev_priv, dev_priv->rps.rpe_delay),
4149 dev_priv->rps.rpe_delay);
4150
4151 valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay);
4152
4153 gen6_enable_rps_interrupts(dev);
4154
4155 gen6_gt_force_wake_put(dev_priv);
4156 }
4157
4158 void ironlake_teardown_rc6(struct drm_device *dev)
4159 {
4160 struct drm_i915_private *dev_priv = dev->dev_private;
4161
4162 if (dev_priv->ips.renderctx) {
4163 i915_gem_object_unpin(dev_priv->ips.renderctx);
4164 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
4165 dev_priv->ips.renderctx = NULL;
4166 }
4167
4168 if (dev_priv->ips.pwrctx) {
4169 i915_gem_object_unpin(dev_priv->ips.pwrctx);
4170 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
4171 dev_priv->ips.pwrctx = NULL;
4172 }
4173 }
4174
4175 static void ironlake_disable_rc6(struct drm_device *dev)
4176 {
4177 struct drm_i915_private *dev_priv = dev->dev_private;
4178
4179 if (I915_READ(PWRCTXA)) {
4180 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
4181 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
4182 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
4183 50);
4184
4185 I915_WRITE(PWRCTXA, 0);
4186 POSTING_READ(PWRCTXA);
4187
4188 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
4189 POSTING_READ(RSTDBYCTL);
4190 }
4191 }
4192
4193 static int ironlake_setup_rc6(struct drm_device *dev)
4194 {
4195 struct drm_i915_private *dev_priv = dev->dev_private;
4196
4197 if (dev_priv->ips.renderctx == NULL)
4198 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
4199 if (!dev_priv->ips.renderctx)
4200 return -ENOMEM;
4201
4202 if (dev_priv->ips.pwrctx == NULL)
4203 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
4204 if (!dev_priv->ips.pwrctx) {
4205 ironlake_teardown_rc6(dev);
4206 return -ENOMEM;
4207 }
4208
4209 return 0;
4210 }
4211
4212 static void ironlake_enable_rc6(struct drm_device *dev)
4213 {
4214 struct drm_i915_private *dev_priv = dev->dev_private;
4215 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
4216 bool was_interruptible;
4217 int ret;
4218
4219 /* rc6 disabled by default due to repeated reports of hanging during
4220 * boot and resume.
4221 */
4222 if (!intel_enable_rc6(dev))
4223 return;
4224
4225 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
4226
4227 ret = ironlake_setup_rc6(dev);
4228 if (ret)
4229 return;
4230
4231 was_interruptible = dev_priv->mm.interruptible;
4232 dev_priv->mm.interruptible = false;
4233
4234 /*
4235 * GPU can automatically power down the render unit if given a page
4236 * to save state.
4237 */
4238 ret = intel_ring_begin(ring, 6);
4239 if (ret) {
4240 ironlake_teardown_rc6(dev);
4241 dev_priv->mm.interruptible = was_interruptible;
4242 return;
4243 }
4244
4245 intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
4246 intel_ring_emit(ring, MI_SET_CONTEXT);
4247 intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) |
4248 MI_MM_SPACE_GTT |
4249 MI_SAVE_EXT_STATE_EN |
4250 MI_RESTORE_EXT_STATE_EN |
4251 MI_RESTORE_INHIBIT);
4252 intel_ring_emit(ring, MI_SUSPEND_FLUSH);
4253 intel_ring_emit(ring, MI_NOOP);
4254 intel_ring_emit(ring, MI_FLUSH);
4255 intel_ring_advance(ring);
4256
4257 /*
4258 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
4259 * does an implicit flush, combined with MI_FLUSH above, it should be
4260 * safe to assume that renderctx is valid
4261 */
4262 ret = intel_ring_idle(ring);
4263 dev_priv->mm.interruptible = was_interruptible;
4264 if (ret) {
4265 DRM_ERROR("failed to enable ironlake power savings\n");
4266 ironlake_teardown_rc6(dev);
4267 return;
4268 }
4269
4270 I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN);
4271 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
4272
4273 intel_print_rc6_info(dev, INTEL_RC6_ENABLE);
4274 }
4275
4276 static unsigned long intel_pxfreq(u32 vidfreq)
4277 {
4278 unsigned long freq;
4279 int div = (vidfreq & 0x3f0000) >> 16;
4280 int post = (vidfreq & 0x3000) >> 12;
4281 int pre = (vidfreq & 0x7);
4282
4283 if (!pre)
4284 return 0;
4285
4286 freq = ((div * 133333) / ((1<<post) * pre));
4287
4288 return freq;
4289 }
4290
4291 static const struct cparams {
4292 u16 i;
4293 u16 t;
4294 u16 m;
4295 u16 c;
4296 } cparams[] = {
4297 { 1, 1333, 301, 28664 },
4298 { 1, 1066, 294, 24460 },
4299 { 1, 800, 294, 25192 },
4300 { 0, 1333, 276, 27605 },
4301 { 0, 1066, 276, 27605 },
4302 { 0, 800, 231, 23784 },
4303 };
4304
4305 static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
4306 {
4307 u64 total_count, diff, ret;
4308 u32 count1, count2, count3, m = 0, c = 0;
4309 unsigned long now = jiffies_to_msecs(jiffies), diff1;
4310 int i;
4311
4312 assert_spin_locked(&mchdev_lock);
4313
4314 diff1 = now - dev_priv->ips.last_time1;
4315
4316 /* Prevent division-by-zero if we are asking too fast.
4317 * Also, we don't get interesting results if we are polling
4318 * faster than once in 10ms, so just return the saved value
4319 * in such cases.
4320 */
4321 if (diff1 <= 10)
4322 return dev_priv->ips.chipset_power;
4323
4324 count1 = I915_READ(DMIEC);
4325 count2 = I915_READ(DDREC);
4326 count3 = I915_READ(CSIEC);
4327
4328 total_count = count1 + count2 + count3;
4329
4330 /* FIXME: handle per-counter overflow */
4331 if (total_count < dev_priv->ips.last_count1) {
4332 diff = ~0UL - dev_priv->ips.last_count1;
4333 diff += total_count;
4334 } else {
4335 diff = total_count - dev_priv->ips.last_count1;
4336 }
4337
4338 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
4339 if (cparams[i].i == dev_priv->ips.c_m &&
4340 cparams[i].t == dev_priv->ips.r_t) {
4341 m = cparams[i].m;
4342 c = cparams[i].c;
4343 break;
4344 }
4345 }
4346
4347 diff = div_u64(diff, diff1);
4348 ret = ((m * diff) + c);
4349 ret = div_u64(ret, 10);
4350
4351 dev_priv->ips.last_count1 = total_count;
4352 dev_priv->ips.last_time1 = now;
4353
4354 dev_priv->ips.chipset_power = ret;
4355
4356 return ret;
4357 }
4358
4359 unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
4360 {
4361 unsigned long val;
4362
4363 if (dev_priv->info->gen != 5)
4364 return 0;
4365
4366 spin_lock_irq(&mchdev_lock);
4367
4368 val = __i915_chipset_val(dev_priv);
4369
4370 spin_unlock_irq(&mchdev_lock);
4371
4372 return val;
4373 }
4374
4375 unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
4376 {
4377 unsigned long m, x, b;
4378 u32 tsfs;
4379
4380 tsfs = I915_READ(TSFS);
4381
4382 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
4383 x = I915_READ8(TR1);
4384
4385 b = tsfs & TSFS_INTR_MASK;
4386
4387 return ((m * x) / 127) - b;
4388 }
4389
4390 static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
4391 {
4392 static const struct v_table {
4393 u16 vd; /* in .1 mil */
4394 u16 vm; /* in .1 mil */
4395 } v_table[] = {
4396 { 0, 0, },
4397 { 375, 0, },
4398 { 500, 0, },
4399 { 625, 0, },
4400 { 750, 0, },
4401 { 875, 0, },
4402 { 1000, 0, },
4403 { 1125, 0, },
4404 { 4125, 3000, },
4405 { 4125, 3000, },
4406 { 4125, 3000, },
4407 { 4125, 3000, },
4408 { 4125, 3000, },
4409 { 4125, 3000, },
4410 { 4125, 3000, },
4411 { 4125, 3000, },
4412 { 4125, 3000, },
4413 { 4125, 3000, },
4414 { 4125, 3000, },
4415 { 4125, 3000, },
4416 { 4125, 3000, },
4417 { 4125, 3000, },
4418 { 4125, 3000, },
4419 { 4125, 3000, },
4420 { 4125, 3000, },
4421 { 4125, 3000, },
4422 { 4125, 3000, },
4423 { 4125, 3000, },
4424 { 4125, 3000, },
4425 { 4125, 3000, },
4426 { 4125, 3000, },
4427 { 4125, 3000, },
4428 { 4250, 3125, },
4429 { 4375, 3250, },
4430 { 4500, 3375, },
4431 { 4625, 3500, },
4432 { 4750, 3625, },
4433 { 4875, 3750, },
4434 { 5000, 3875, },
4435 { 5125, 4000, },
4436 { 5250, 4125, },
4437 { 5375, 4250, },
4438 { 5500, 4375, },
4439 { 5625, 4500, },
4440 { 5750, 4625, },
4441 { 5875, 4750, },
4442 { 6000, 4875, },
4443 { 6125, 5000, },
4444 { 6250, 5125, },
4445 { 6375, 5250, },
4446 { 6500, 5375, },
4447 { 6625, 5500, },
4448 { 6750, 5625, },
4449 { 6875, 5750, },
4450 { 7000, 5875, },
4451 { 7125, 6000, },
4452 { 7250, 6125, },
4453 { 7375, 6250, },
4454 { 7500, 6375, },
4455 { 7625, 6500, },
4456 { 7750, 6625, },
4457 { 7875, 6750, },
4458 { 8000, 6875, },
4459 { 8125, 7000, },
4460 { 8250, 7125, },
4461 { 8375, 7250, },
4462 { 8500, 7375, },
4463 { 8625, 7500, },
4464 { 8750, 7625, },
4465 { 8875, 7750, },
4466 { 9000, 7875, },
4467 { 9125, 8000, },
4468 { 9250, 8125, },
4469 { 9375, 8250, },
4470 { 9500, 8375, },
4471 { 9625, 8500, },
4472 { 9750, 8625, },
4473 { 9875, 8750, },
4474 { 10000, 8875, },
4475 { 10125, 9000, },
4476 { 10250, 9125, },
4477 { 10375, 9250, },
4478 { 10500, 9375, },
4479 { 10625, 9500, },
4480 { 10750, 9625, },
4481 { 10875, 9750, },
4482 { 11000, 9875, },
4483 { 11125, 10000, },
4484 { 11250, 10125, },
4485 { 11375, 10250, },
4486 { 11500, 10375, },
4487 { 11625, 10500, },
4488 { 11750, 10625, },
4489 { 11875, 10750, },
4490 { 12000, 10875, },
4491 { 12125, 11000, },
4492 { 12250, 11125, },
4493 { 12375, 11250, },
4494 { 12500, 11375, },
4495 { 12625, 11500, },
4496 { 12750, 11625, },
4497 { 12875, 11750, },
4498 { 13000, 11875, },
4499 { 13125, 12000, },
4500 { 13250, 12125, },
4501 { 13375, 12250, },
4502 { 13500, 12375, },
4503 { 13625, 12500, },
4504 { 13750, 12625, },
4505 { 13875, 12750, },
4506 { 14000, 12875, },
4507 { 14125, 13000, },
4508 { 14250, 13125, },
4509 { 14375, 13250, },
4510 { 14500, 13375, },
4511 { 14625, 13500, },
4512 { 14750, 13625, },
4513 { 14875, 13750, },
4514 { 15000, 13875, },
4515 { 15125, 14000, },
4516 { 15250, 14125, },
4517 { 15375, 14250, },
4518 { 15500, 14375, },
4519 { 15625, 14500, },
4520 { 15750, 14625, },
4521 { 15875, 14750, },
4522 { 16000, 14875, },
4523 { 16125, 15000, },
4524 };
4525 if (dev_priv->info->is_mobile)
4526 return v_table[pxvid].vm;
4527 else
4528 return v_table[pxvid].vd;
4529 }
4530
4531 static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
4532 {
4533 struct timespec now, diff1;
4534 u64 diff;
4535 unsigned long diffms;
4536 u32 count;
4537
4538 assert_spin_locked(&mchdev_lock);
4539
4540 getrawmonotonic(&now);
4541 diff1 = timespec_sub(now, dev_priv->ips.last_time2);
4542
4543 /* Don't divide by 0 */
4544 diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
4545 if (!diffms)
4546 return;
4547
4548 count = I915_READ(GFXEC);
4549
4550 if (count < dev_priv->ips.last_count2) {
4551 diff = ~0UL - dev_priv->ips.last_count2;
4552 diff += count;
4553 } else {
4554 diff = count - dev_priv->ips.last_count2;
4555 }
4556
4557 dev_priv->ips.last_count2 = count;
4558 dev_priv->ips.last_time2 = now;
4559
4560 /* More magic constants... */
4561 diff = diff * 1181;
4562 diff = div_u64(diff, diffms * 10);
4563 dev_priv->ips.gfx_power = diff;
4564 }
4565
4566 void i915_update_gfx_val(struct drm_i915_private *dev_priv)
4567 {
4568 if (dev_priv->info->gen != 5)
4569 return;
4570
4571 spin_lock_irq(&mchdev_lock);
4572
4573 __i915_update_gfx_val(dev_priv);
4574
4575 spin_unlock_irq(&mchdev_lock);
4576 }
4577
4578 static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
4579 {
4580 unsigned long t, corr, state1, corr2, state2;
4581 u32 pxvid, ext_v;
4582
4583 assert_spin_locked(&mchdev_lock);
4584
4585 pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4));
4586 pxvid = (pxvid >> 24) & 0x7f;
4587 ext_v = pvid_to_extvid(dev_priv, pxvid);
4588
4589 state1 = ext_v;
4590
4591 t = i915_mch_val(dev_priv);
4592
4593 /* Revel in the empirically derived constants */
4594
4595 /* Correction factor in 1/100000 units */
4596 if (t > 80)
4597 corr = ((t * 2349) + 135940);
4598 else if (t >= 50)
4599 corr = ((t * 964) + 29317);
4600 else /* < 50 */
4601 corr = ((t * 301) + 1004);
4602
4603 corr = corr * ((150142 * state1) / 10000 - 78642);
4604 corr /= 100000;
4605 corr2 = (corr * dev_priv->ips.corr);
4606
4607 state2 = (corr2 * state1) / 10000;
4608 state2 /= 100; /* convert to mW */
4609
4610 __i915_update_gfx_val(dev_priv);
4611
4612 return dev_priv->ips.gfx_power + state2;
4613 }
4614
4615 unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
4616 {
4617 unsigned long val;
4618
4619 if (dev_priv->info->gen != 5)
4620 return 0;
4621
4622 spin_lock_irq(&mchdev_lock);
4623
4624 val = __i915_gfx_val(dev_priv);
4625
4626 spin_unlock_irq(&mchdev_lock);
4627
4628 return val;
4629 }
4630
4631 /**
4632 * i915_read_mch_val - return value for IPS use
4633 *
4634 * Calculate and return a value for the IPS driver to use when deciding whether
4635 * we have thermal and power headroom to increase CPU or GPU power budget.
4636 */
4637 unsigned long i915_read_mch_val(void)
4638 {
4639 struct drm_i915_private *dev_priv;
4640 unsigned long chipset_val, graphics_val, ret = 0;
4641
4642 spin_lock_irq(&mchdev_lock);
4643 if (!i915_mch_dev)
4644 goto out_unlock;
4645 dev_priv = i915_mch_dev;
4646
4647 chipset_val = __i915_chipset_val(dev_priv);
4648 graphics_val = __i915_gfx_val(dev_priv);
4649
4650 ret = chipset_val + graphics_val;
4651
4652 out_unlock:
4653 spin_unlock_irq(&mchdev_lock);
4654
4655 return ret;
4656 }
4657 EXPORT_SYMBOL_GPL(i915_read_mch_val);
4658
4659 /**
4660 * i915_gpu_raise - raise GPU frequency limit
4661 *
4662 * Raise the limit; IPS indicates we have thermal headroom.
4663 */
4664 bool i915_gpu_raise(void)
4665 {
4666 struct drm_i915_private *dev_priv;
4667 bool ret = true;
4668
4669 spin_lock_irq(&mchdev_lock);
4670 if (!i915_mch_dev) {
4671 ret = false;
4672 goto out_unlock;
4673 }
4674 dev_priv = i915_mch_dev;
4675
4676 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
4677 dev_priv->ips.max_delay--;
4678
4679 out_unlock:
4680 spin_unlock_irq(&mchdev_lock);
4681
4682 return ret;
4683 }
4684 EXPORT_SYMBOL_GPL(i915_gpu_raise);
4685
4686 /**
4687 * i915_gpu_lower - lower GPU frequency limit
4688 *
4689 * IPS indicates we're close to a thermal limit, so throttle back the GPU
4690 * frequency maximum.
4691 */
4692 bool i915_gpu_lower(void)
4693 {
4694 struct drm_i915_private *dev_priv;
4695 bool ret = true;
4696
4697 spin_lock_irq(&mchdev_lock);
4698 if (!i915_mch_dev) {
4699 ret = false;
4700 goto out_unlock;
4701 }
4702 dev_priv = i915_mch_dev;
4703
4704 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
4705 dev_priv->ips.max_delay++;
4706
4707 out_unlock:
4708 spin_unlock_irq(&mchdev_lock);
4709
4710 return ret;
4711 }
4712 EXPORT_SYMBOL_GPL(i915_gpu_lower);
4713
4714 /**
4715 * i915_gpu_busy - indicate GPU business to IPS
4716 *
4717 * Tell the IPS driver whether or not the GPU is busy.
4718 */
4719 bool i915_gpu_busy(void)
4720 {
4721 struct drm_i915_private *dev_priv;
4722 struct intel_ring_buffer *ring;
4723 bool ret = false;
4724 int i;
4725
4726 spin_lock_irq(&mchdev_lock);
4727 if (!i915_mch_dev)
4728 goto out_unlock;
4729 dev_priv = i915_mch_dev;
4730
4731 for_each_ring(ring, dev_priv, i)
4732 ret |= !list_empty(&ring->request_list);
4733
4734 out_unlock:
4735 spin_unlock_irq(&mchdev_lock);
4736
4737 return ret;
4738 }
4739 EXPORT_SYMBOL_GPL(i915_gpu_busy);
4740
4741 /**
4742 * i915_gpu_turbo_disable - disable graphics turbo
4743 *
4744 * Disable graphics turbo by resetting the max frequency and setting the
4745 * current frequency to the default.
4746 */
4747 bool i915_gpu_turbo_disable(void)
4748 {
4749 struct drm_i915_private *dev_priv;
4750 bool ret = true;
4751
4752 spin_lock_irq(&mchdev_lock);
4753 if (!i915_mch_dev) {
4754 ret = false;
4755 goto out_unlock;
4756 }
4757 dev_priv = i915_mch_dev;
4758
4759 dev_priv->ips.max_delay = dev_priv->ips.fstart;
4760
4761 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
4762 ret = false;
4763
4764 out_unlock:
4765 spin_unlock_irq(&mchdev_lock);
4766
4767 return ret;
4768 }
4769 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
4770
4771 /**
4772 * Tells the intel_ips driver that the i915 driver is now loaded, if
4773 * IPS got loaded first.
4774 *
4775 * This awkward dance is so that neither module has to depend on the
4776 * other in order for IPS to do the appropriate communication of
4777 * GPU turbo limits to i915.
4778 */
4779 static void
4780 ips_ping_for_i915_load(void)
4781 {
4782 void (*link)(void);
4783
4784 link = symbol_get(ips_link_to_i915_driver);
4785 if (link) {
4786 link();
4787 symbol_put(ips_link_to_i915_driver);
4788 }
4789 }
4790
4791 void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
4792 {
4793 /* We only register the i915 ips part with intel-ips once everything is
4794 * set up, to avoid intel-ips sneaking in and reading bogus values. */
4795 spin_lock_irq(&mchdev_lock);
4796 i915_mch_dev = dev_priv;
4797 spin_unlock_irq(&mchdev_lock);
4798
4799 ips_ping_for_i915_load();
4800 }
4801
4802 void intel_gpu_ips_teardown(void)
4803 {
4804 spin_lock_irq(&mchdev_lock);
4805 i915_mch_dev = NULL;
4806 spin_unlock_irq(&mchdev_lock);
4807 }
4808 static void intel_init_emon(struct drm_device *dev)
4809 {
4810 struct drm_i915_private *dev_priv = dev->dev_private;
4811 u32 lcfuse;
4812 u8 pxw[16];
4813 int i;
4814
4815 /* Disable to program */
4816 I915_WRITE(ECR, 0);
4817 POSTING_READ(ECR);
4818
4819 /* Program energy weights for various events */
4820 I915_WRITE(SDEW, 0x15040d00);
4821 I915_WRITE(CSIEW0, 0x007f0000);
4822 I915_WRITE(CSIEW1, 0x1e220004);
4823 I915_WRITE(CSIEW2, 0x04000004);
4824
4825 for (i = 0; i < 5; i++)
4826 I915_WRITE(PEW + (i * 4), 0);
4827 for (i = 0; i < 3; i++)
4828 I915_WRITE(DEW + (i * 4), 0);
4829
4830 /* Program P-state weights to account for frequency power adjustment */
4831 for (i = 0; i < 16; i++) {
4832 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
4833 unsigned long freq = intel_pxfreq(pxvidfreq);
4834 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
4835 PXVFREQ_PX_SHIFT;
4836 unsigned long val;
4837
4838 val = vid * vid;
4839 val *= (freq / 1000);
4840 val *= 255;
4841 val /= (127*127*900);
4842 if (val > 0xff)
4843 DRM_ERROR("bad pxval: %ld\n", val);
4844 pxw[i] = val;
4845 }
4846 /* Render standby states get 0 weight */
4847 pxw[14] = 0;
4848 pxw[15] = 0;
4849
4850 for (i = 0; i < 4; i++) {
4851 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
4852 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
4853 I915_WRITE(PXW + (i * 4), val);
4854 }
4855
4856 /* Adjust magic regs to magic values (more experimental results) */
4857 I915_WRITE(OGW0, 0);
4858 I915_WRITE(OGW1, 0);
4859 I915_WRITE(EG0, 0x00007f00);
4860 I915_WRITE(EG1, 0x0000000e);
4861 I915_WRITE(EG2, 0x000e0000);
4862 I915_WRITE(EG3, 0x68000300);
4863 I915_WRITE(EG4, 0x42000000);
4864 I915_WRITE(EG5, 0x00140031);
4865 I915_WRITE(EG6, 0);
4866 I915_WRITE(EG7, 0);
4867
4868 for (i = 0; i < 8; i++)
4869 I915_WRITE(PXWL + (i * 4), 0);
4870
4871 /* Enable PMON + select events */
4872 I915_WRITE(ECR, 0x80000019);
4873
4874 lcfuse = I915_READ(LCFUSE02);
4875
4876 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
4877 }
4878
4879 void intel_disable_gt_powersave(struct drm_device *dev)
4880 {
4881 struct drm_i915_private *dev_priv = dev->dev_private;
4882
4883 /* Interrupts should be disabled already to avoid re-arming. */
4884 WARN_ON(dev->irq_enabled);
4885
4886 if (IS_IRONLAKE_M(dev)) {
4887 ironlake_disable_drps(dev);
4888 ironlake_disable_rc6(dev);
4889 } else if (INTEL_INFO(dev)->gen >= 6) {
4890 cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
4891 cancel_work_sync(&dev_priv->rps.work);
4892 mutex_lock(&dev_priv->rps.hw_lock);
4893 if (IS_VALLEYVIEW(dev))
4894 valleyview_disable_rps(dev);
4895 else
4896 gen6_disable_rps(dev);
4897 dev_priv->rps.enabled = false;
4898 mutex_unlock(&dev_priv->rps.hw_lock);
4899 }
4900 }
4901
4902 static void intel_gen6_powersave_work(struct work_struct *work)
4903 {
4904 struct drm_i915_private *dev_priv =
4905 container_of(work, struct drm_i915_private,
4906 rps.delayed_resume_work.work);
4907 struct drm_device *dev = dev_priv->dev;
4908
4909 mutex_lock(&dev_priv->rps.hw_lock);
4910
4911 if (IS_VALLEYVIEW(dev)) {
4912 valleyview_enable_rps(dev);
4913 } else if (IS_BROADWELL(dev)) {
4914 gen8_enable_rps(dev);
4915 gen6_update_ring_freq(dev);
4916 } else {
4917 gen6_enable_rps(dev);
4918 gen6_update_ring_freq(dev);
4919 }
4920 dev_priv->rps.enabled = true;
4921 mutex_unlock(&dev_priv->rps.hw_lock);
4922 }
4923
4924 void intel_enable_gt_powersave(struct drm_device *dev)
4925 {
4926 struct drm_i915_private *dev_priv = dev->dev_private;
4927
4928 if (IS_IRONLAKE_M(dev)) {
4929 ironlake_enable_drps(dev);
4930 ironlake_enable_rc6(dev);
4931 intel_init_emon(dev);
4932 } else if (IS_GEN6(dev) || IS_GEN7(dev)) {
4933 /*
4934 * PCU communication is slow and this doesn't need to be
4935 * done at any specific time, so do this out of our fast path
4936 * to make resume and init faster.
4937 */
4938 schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
4939 round_jiffies_up_relative(HZ));
4940 }
4941 }
4942
4943 static void ibx_init_clock_gating(struct drm_device *dev)
4944 {
4945 struct drm_i915_private *dev_priv = dev->dev_private;
4946
4947 /*
4948 * On Ibex Peak and Cougar Point, we need to disable clock
4949 * gating for the panel power sequencer or it will fail to
4950 * start up when no ports are active.
4951 */
4952 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
4953 }
4954
4955 static void g4x_disable_trickle_feed(struct drm_device *dev)
4956 {
4957 struct drm_i915_private *dev_priv = dev->dev_private;
4958 int pipe;
4959
4960 for_each_pipe(pipe) {
4961 I915_WRITE(DSPCNTR(pipe),
4962 I915_READ(DSPCNTR(pipe)) |
4963 DISPPLANE_TRICKLE_FEED_DISABLE);
4964 intel_flush_primary_plane(dev_priv, pipe);
4965 }
4966 }
4967
4968 static void ironlake_init_clock_gating(struct drm_device *dev)
4969 {
4970 struct drm_i915_private *dev_priv = dev->dev_private;
4971 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
4972
4973 /*
4974 * Required for FBC
4975 * WaFbcDisableDpfcClockGating:ilk
4976 */
4977 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
4978 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
4979 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
4980
4981 I915_WRITE(PCH_3DCGDIS0,
4982 MARIUNIT_CLOCK_GATE_DISABLE |
4983 SVSMUNIT_CLOCK_GATE_DISABLE);
4984 I915_WRITE(PCH_3DCGDIS1,
4985 VFMUNIT_CLOCK_GATE_DISABLE);
4986
4987 /*
4988 * According to the spec the following bits should be set in
4989 * order to enable memory self-refresh
4990 * The bit 22/21 of 0x42004
4991 * The bit 5 of 0x42020
4992 * The bit 15 of 0x45000
4993 */
4994 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4995 (I915_READ(ILK_DISPLAY_CHICKEN2) |
4996 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4997 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
4998 I915_WRITE(DISP_ARB_CTL,
4999 (I915_READ(DISP_ARB_CTL) |
5000 DISP_FBC_WM_DIS));
5001 I915_WRITE(WM3_LP_ILK, 0);
5002 I915_WRITE(WM2_LP_ILK, 0);
5003 I915_WRITE(WM1_LP_ILK, 0);
5004
5005 /*
5006 * Based on the document from hardware guys the following bits
5007 * should be set unconditionally in order to enable FBC.
5008 * The bit 22 of 0x42000
5009 * The bit 22 of 0x42004
5010 * The bit 7,8,9 of 0x42020.
5011 */
5012 if (IS_IRONLAKE_M(dev)) {
5013 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
5014 I915_WRITE(ILK_DISPLAY_CHICKEN1,
5015 I915_READ(ILK_DISPLAY_CHICKEN1) |
5016 ILK_FBCQ_DIS);
5017 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5018 I915_READ(ILK_DISPLAY_CHICKEN2) |
5019 ILK_DPARB_GATE);
5020 }
5021
5022 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
5023
5024 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5025 I915_READ(ILK_DISPLAY_CHICKEN2) |
5026 ILK_ELPIN_409_SELECT);
5027 I915_WRITE(_3D_CHICKEN2,
5028 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
5029 _3D_CHICKEN2_WM_READ_PIPELINED);
5030
5031 /* WaDisableRenderCachePipelinedFlush:ilk */
5032 I915_WRITE(CACHE_MODE_0,
5033 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
5034
5035 g4x_disable_trickle_feed(dev);
5036
5037 ibx_init_clock_gating(dev);
5038 }
5039
5040 static void cpt_init_clock_gating(struct drm_device *dev)
5041 {
5042 struct drm_i915_private *dev_priv = dev->dev_private;
5043 int pipe;
5044 uint32_t val;
5045
5046 /*
5047 * On Ibex Peak and Cougar Point, we need to disable clock
5048 * gating for the panel power sequencer or it will fail to
5049 * start up when no ports are active.
5050 */
5051 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
5052 PCH_DPLUNIT_CLOCK_GATE_DISABLE |
5053 PCH_CPUNIT_CLOCK_GATE_DISABLE);
5054 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
5055 DPLS_EDP_PPS_FIX_DIS);
5056 /* The below fixes the weird display corruption, a few pixels shifted
5057 * downward, on (only) LVDS of some HP laptops with IVY.
5058 */
5059 for_each_pipe(pipe) {
5060 val = I915_READ(TRANS_CHICKEN2(pipe));
5061 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
5062 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
5063 if (dev_priv->vbt.fdi_rx_polarity_inverted)
5064 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
5065 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
5066 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
5067 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
5068 I915_WRITE(TRANS_CHICKEN2(pipe), val);
5069 }
5070 /* WADP0ClockGatingDisable */
5071 for_each_pipe(pipe) {
5072 I915_WRITE(TRANS_CHICKEN1(pipe),
5073 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
5074 }
5075 }
5076
5077 static void gen6_check_mch_setup(struct drm_device *dev)
5078 {
5079 struct drm_i915_private *dev_priv = dev->dev_private;
5080 uint32_t tmp;
5081
5082 tmp = I915_READ(MCH_SSKPD);
5083 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
5084 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
5085 DRM_INFO("This can cause pipe underruns and display issues.\n");
5086 DRM_INFO("Please upgrade your BIOS to fix this.\n");
5087 }
5088 }
5089
5090 static void gen6_init_clock_gating(struct drm_device *dev)
5091 {
5092 struct drm_i915_private *dev_priv = dev->dev_private;
5093 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
5094
5095 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
5096
5097 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5098 I915_READ(ILK_DISPLAY_CHICKEN2) |
5099 ILK_ELPIN_409_SELECT);
5100
5101 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
5102 I915_WRITE(_3D_CHICKEN,
5103 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
5104
5105 /* WaSetupGtModeTdRowDispatch:snb */
5106 if (IS_SNB_GT1(dev))
5107 I915_WRITE(GEN6_GT_MODE,
5108 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
5109
5110 I915_WRITE(WM3_LP_ILK, 0);
5111 I915_WRITE(WM2_LP_ILK, 0);
5112 I915_WRITE(WM1_LP_ILK, 0);
5113
5114 I915_WRITE(CACHE_MODE_0,
5115 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
5116
5117 I915_WRITE(GEN6_UCGCTL1,
5118 I915_READ(GEN6_UCGCTL1) |
5119 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
5120 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
5121
5122 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
5123 * gating disable must be set. Failure to set it results in
5124 * flickering pixels due to Z write ordering failures after
5125 * some amount of runtime in the Mesa "fire" demo, and Unigine
5126 * Sanctuary and Tropics, and apparently anything else with
5127 * alpha test or pixel discard.
5128 *
5129 * According to the spec, bit 11 (RCCUNIT) must also be set,
5130 * but we didn't debug actual testcases to find it out.
5131 *
5132 * Also apply WaDisableVDSUnitClockGating:snb and
5133 * WaDisableRCPBUnitClockGating:snb.
5134 */
5135 I915_WRITE(GEN6_UCGCTL2,
5136 GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
5137 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
5138 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
5139
5140 /* Bspec says we need to always set all mask bits. */
5141 I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) |
5142 _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL);
5143
5144 /*
5145 * According to the spec the following bits should be
5146 * set in order to enable memory self-refresh and fbc:
5147 * The bit21 and bit22 of 0x42000
5148 * The bit21 and bit22 of 0x42004
5149 * The bit5 and bit7 of 0x42020
5150 * The bit14 of 0x70180
5151 * The bit14 of 0x71180
5152 *
5153 * WaFbcAsynchFlipDisableFbcQueue:snb
5154 */
5155 I915_WRITE(ILK_DISPLAY_CHICKEN1,
5156 I915_READ(ILK_DISPLAY_CHICKEN1) |
5157 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
5158 I915_WRITE(ILK_DISPLAY_CHICKEN2,
5159 I915_READ(ILK_DISPLAY_CHICKEN2) |
5160 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
5161 I915_WRITE(ILK_DSPCLK_GATE_D,
5162 I915_READ(ILK_DSPCLK_GATE_D) |
5163 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
5164 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
5165
5166 g4x_disable_trickle_feed(dev);
5167
5168 /* The default value should be 0x200 according to docs, but the two
5169 * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */
5170 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff));
5171 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI));
5172
5173 cpt_init_clock_gating(dev);
5174
5175 gen6_check_mch_setup(dev);
5176 }
5177
5178 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
5179 {
5180 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
5181
5182 reg &= ~GEN7_FF_SCHED_MASK;
5183 reg |= GEN7_FF_TS_SCHED_HW;
5184 reg |= GEN7_FF_VS_SCHED_HW;
5185 reg |= GEN7_FF_DS_SCHED_HW;
5186
5187 if (IS_HASWELL(dev_priv->dev))
5188 reg &= ~GEN7_FF_VS_REF_CNT_FFME;
5189
5190 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
5191 }
5192
5193 static void lpt_init_clock_gating(struct drm_device *dev)
5194 {
5195 struct drm_i915_private *dev_priv = dev->dev_private;
5196
5197 /*
5198 * TODO: this bit should only be enabled when really needed, then
5199 * disabled when not needed anymore in order to save power.
5200 */
5201 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
5202 I915_WRITE(SOUTH_DSPCLK_GATE_D,
5203 I915_READ(SOUTH_DSPCLK_GATE_D) |
5204 PCH_LP_PARTITION_LEVEL_DISABLE);
5205
5206 /* WADPOClockGatingDisable:hsw */
5207 I915_WRITE(_TRANSA_CHICKEN1,
5208 I915_READ(_TRANSA_CHICKEN1) |
5209 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
5210 }
5211
5212 static void lpt_suspend_hw(struct drm_device *dev)
5213 {
5214 struct drm_i915_private *dev_priv = dev->dev_private;
5215
5216 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) {
5217 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
5218
5219 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
5220 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
5221 }
5222 }
5223
5224 static void gen8_init_clock_gating(struct drm_device *dev)
5225 {
5226 struct drm_i915_private *dev_priv = dev->dev_private;
5227 enum pipe i;
5228
5229 I915_WRITE(WM3_LP_ILK, 0);
5230 I915_WRITE(WM2_LP_ILK, 0);
5231 I915_WRITE(WM1_LP_ILK, 0);
5232
5233 /* FIXME(BDW): Check all the w/a, some might only apply to
5234 * pre-production hw. */
5235
5236 WARN(!i915_preliminary_hw_support,
5237 "GEN8_CENTROID_PIXEL_OPT_DIS not be needed for production\n");
5238 I915_WRITE(HALF_SLICE_CHICKEN3,
5239 _MASKED_BIT_ENABLE(GEN8_CENTROID_PIXEL_OPT_DIS));
5240 I915_WRITE(HALF_SLICE_CHICKEN3,
5241 _MASKED_BIT_ENABLE(GEN8_SAMPLER_POWER_BYPASS_DIS));
5242 I915_WRITE(GAMTARBMODE, _MASKED_BIT_ENABLE(ARB_MODE_BWGTLB_DISABLE));
5243
5244 I915_WRITE(_3D_CHICKEN3,
5245 _3D_CHICKEN_SDE_LIMIT_FIFO_POLY_DEPTH(2));
5246
5247 I915_WRITE(COMMON_SLICE_CHICKEN2,
5248 _MASKED_BIT_ENABLE(GEN8_CSC2_SBE_VUE_CACHE_CONSERVATIVE));
5249
5250 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5251 _MASKED_BIT_ENABLE(GEN7_SINGLE_SUBSCAN_DISPATCH_ENABLE));
5252
5253 /* WaSwitchSolVfFArbitrationPriority */
5254 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
5255
5256 /* WaPsrDPAMaskVBlankInSRD */
5257 I915_WRITE(CHICKEN_PAR1_1,
5258 I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
5259
5260 /* WaPsrDPRSUnmaskVBlankInSRD */
5261 for_each_pipe(i) {
5262 I915_WRITE(CHICKEN_PIPESL_1(i),
5263 I915_READ(CHICKEN_PIPESL_1(i) |
5264 DPRS_MASK_VBLANK_SRD));
5265 }
5266 }
5267
5268 static void haswell_init_clock_gating(struct drm_device *dev)
5269 {
5270 struct drm_i915_private *dev_priv = dev->dev_private;
5271
5272 I915_WRITE(WM3_LP_ILK, 0);
5273 I915_WRITE(WM2_LP_ILK, 0);
5274 I915_WRITE(WM1_LP_ILK, 0);
5275
5276 /* According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5277 * This implements the WaDisableRCZUnitClockGating:hsw workaround.
5278 */
5279 I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
5280
5281 /* Apply the WaDisableRHWOOptimizationForRenderHang:hsw workaround. */
5282 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5283 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5284
5285 /* WaApplyL3ControlAndL3ChickenMode:hsw */
5286 I915_WRITE(GEN7_L3CNTLREG1,
5287 GEN7_WA_FOR_GEN7_L3_CONTROL);
5288 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
5289 GEN7_WA_L3_CHICKEN_MODE);
5290
5291 /* L3 caching of data atomics doesn't work -- disable it. */
5292 I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
5293 I915_WRITE(HSW_ROW_CHICKEN3,
5294 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
5295
5296 /* This is required by WaCatErrorRejectionIssue:hsw */
5297 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5298 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5299 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5300
5301 /* WaVSRefCountFullforceMissDisable:hsw */
5302 gen7_setup_fixed_func_scheduler(dev_priv);
5303
5304 /* WaDisable4x2SubspanOptimization:hsw */
5305 I915_WRITE(CACHE_MODE_1,
5306 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
5307
5308 /* WaSwitchSolVfFArbitrationPriority:hsw */
5309 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
5310
5311 /* WaRsPkgCStateDisplayPMReq:hsw */
5312 I915_WRITE(CHICKEN_PAR1_1,
5313 I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
5314
5315 lpt_init_clock_gating(dev);
5316 }
5317
5318 static void ivybridge_init_clock_gating(struct drm_device *dev)
5319 {
5320 struct drm_i915_private *dev_priv = dev->dev_private;
5321 uint32_t snpcr;
5322
5323 I915_WRITE(WM3_LP_ILK, 0);
5324 I915_WRITE(WM2_LP_ILK, 0);
5325 I915_WRITE(WM1_LP_ILK, 0);
5326
5327 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
5328
5329 /* WaDisableEarlyCull:ivb */
5330 I915_WRITE(_3D_CHICKEN3,
5331 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5332
5333 /* WaDisableBackToBackFlipFix:ivb */
5334 I915_WRITE(IVB_CHICKEN3,
5335 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5336 CHICKEN3_DGMG_DONE_FIX_DISABLE);
5337
5338 /* WaDisablePSDDualDispatchEnable:ivb */
5339 if (IS_IVB_GT1(dev))
5340 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5341 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
5342 else
5343 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2,
5344 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
5345
5346 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
5347 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5348 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5349
5350 /* WaApplyL3ControlAndL3ChickenMode:ivb */
5351 I915_WRITE(GEN7_L3CNTLREG1,
5352 GEN7_WA_FOR_GEN7_L3_CONTROL);
5353 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
5354 GEN7_WA_L3_CHICKEN_MODE);
5355 if (IS_IVB_GT1(dev))
5356 I915_WRITE(GEN7_ROW_CHICKEN2,
5357 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5358 else
5359 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
5360 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5361
5362
5363 /* WaForceL3Serialization:ivb */
5364 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5365 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5366
5367 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
5368 * gating disable must be set. Failure to set it results in
5369 * flickering pixels due to Z write ordering failures after
5370 * some amount of runtime in the Mesa "fire" demo, and Unigine
5371 * Sanctuary and Tropics, and apparently anything else with
5372 * alpha test or pixel discard.
5373 *
5374 * According to the spec, bit 11 (RCCUNIT) must also be set,
5375 * but we didn't debug actual testcases to find it out.
5376 *
5377 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5378 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
5379 */
5380 I915_WRITE(GEN6_UCGCTL2,
5381 GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
5382 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
5383
5384 /* This is required by WaCatErrorRejectionIssue:ivb */
5385 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5386 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5387 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5388
5389 g4x_disable_trickle_feed(dev);
5390
5391 /* WaVSRefCountFullforceMissDisable:ivb */
5392 gen7_setup_fixed_func_scheduler(dev_priv);
5393
5394 /* WaDisable4x2SubspanOptimization:ivb */
5395 I915_WRITE(CACHE_MODE_1,
5396 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
5397
5398 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
5399 snpcr &= ~GEN6_MBC_SNPCR_MASK;
5400 snpcr |= GEN6_MBC_SNPCR_MED;
5401 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
5402
5403 if (!HAS_PCH_NOP(dev))
5404 cpt_init_clock_gating(dev);
5405
5406 gen6_check_mch_setup(dev);
5407 }
5408
5409 static void valleyview_init_clock_gating(struct drm_device *dev)
5410 {
5411 struct drm_i915_private *dev_priv = dev->dev_private;
5412 u32 val;
5413
5414 mutex_lock(&dev_priv->rps.hw_lock);
5415 val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
5416 mutex_unlock(&dev_priv->rps.hw_lock);
5417 switch ((val >> 6) & 3) {
5418 case 0:
5419 dev_priv->mem_freq = 800;
5420 break;
5421 case 1:
5422 dev_priv->mem_freq = 1066;
5423 break;
5424 case 2:
5425 dev_priv->mem_freq = 1333;
5426 break;
5427 case 3:
5428 dev_priv->mem_freq = 1333;
5429 break;
5430 }
5431 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
5432
5433 I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
5434
5435 /* WaDisableEarlyCull:vlv */
5436 I915_WRITE(_3D_CHICKEN3,
5437 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5438
5439 /* WaDisableBackToBackFlipFix:vlv */
5440 I915_WRITE(IVB_CHICKEN3,
5441 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5442 CHICKEN3_DGMG_DONE_FIX_DISABLE);
5443
5444 /* WaDisablePSDDualDispatchEnable:vlv */
5445 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5446 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
5447 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
5448
5449 /* Apply the WaDisableRHWOOptimizationForRenderHang:vlv workaround. */
5450 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5451 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5452
5453 /* WaApplyL3ControlAndL3ChickenMode:vlv */
5454 I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS);
5455 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE);
5456
5457 /* WaForceL3Serialization:vlv */
5458 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5459 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5460
5461 /* WaDisableDopClockGating:vlv */
5462 I915_WRITE(GEN7_ROW_CHICKEN2,
5463 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5464
5465 /* This is required by WaCatErrorRejectionIssue:vlv */
5466 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5467 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5468 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5469
5470 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
5471 * gating disable must be set. Failure to set it results in
5472 * flickering pixels due to Z write ordering failures after
5473 * some amount of runtime in the Mesa "fire" demo, and Unigine
5474 * Sanctuary and Tropics, and apparently anything else with
5475 * alpha test or pixel discard.
5476 *
5477 * According to the spec, bit 11 (RCCUNIT) must also be set,
5478 * but we didn't debug actual testcases to find it out.
5479 *
5480 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5481 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
5482 *
5483 * Also apply WaDisableVDSUnitClockGating:vlv and
5484 * WaDisableRCPBUnitClockGating:vlv.
5485 */
5486 I915_WRITE(GEN6_UCGCTL2,
5487 GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
5488 GEN7_TDLUNIT_CLOCK_GATE_DISABLE |
5489 GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
5490 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
5491 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
5492
5493 I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
5494
5495 I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
5496
5497 I915_WRITE(CACHE_MODE_1,
5498 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
5499
5500 /*
5501 * WaDisableVLVClockGating_VBIIssue:vlv
5502 * Disable clock gating on th GCFG unit to prevent a delay
5503 * in the reporting of vblank events.
5504 */
5505 I915_WRITE(VLV_GUNIT_CLOCK_GATE, 0xffffffff);
5506
5507 /* Conservative clock gating settings for now */
5508 I915_WRITE(0x9400, 0xffffffff);
5509 I915_WRITE(0x9404, 0xffffffff);
5510 I915_WRITE(0x9408, 0xffffffff);
5511 I915_WRITE(0x940c, 0xffffffff);
5512 I915_WRITE(0x9410, 0xffffffff);
5513 I915_WRITE(0x9414, 0xffffffff);
5514 I915_WRITE(0x9418, 0xffffffff);
5515 }
5516
5517 static void g4x_init_clock_gating(struct drm_device *dev)
5518 {
5519 struct drm_i915_private *dev_priv = dev->dev_private;
5520 uint32_t dspclk_gate;
5521
5522 I915_WRITE(RENCLK_GATE_D1, 0);
5523 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
5524 GS_UNIT_CLOCK_GATE_DISABLE |
5525 CL_UNIT_CLOCK_GATE_DISABLE);
5526 I915_WRITE(RAMCLK_GATE_D, 0);
5527 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
5528 OVRUNIT_CLOCK_GATE_DISABLE |
5529 OVCUNIT_CLOCK_GATE_DISABLE;
5530 if (IS_GM45(dev))
5531 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
5532 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
5533
5534 /* WaDisableRenderCachePipelinedFlush */
5535 I915_WRITE(CACHE_MODE_0,
5536 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
5537
5538 g4x_disable_trickle_feed(dev);
5539 }
5540
5541 static void crestline_init_clock_gating(struct drm_device *dev)
5542 {
5543 struct drm_i915_private *dev_priv = dev->dev_private;
5544
5545 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
5546 I915_WRITE(RENCLK_GATE_D2, 0);
5547 I915_WRITE(DSPCLK_GATE_D, 0);
5548 I915_WRITE(RAMCLK_GATE_D, 0);
5549 I915_WRITE16(DEUC, 0);
5550 I915_WRITE(MI_ARB_STATE,
5551 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
5552 }
5553
5554 static void broadwater_init_clock_gating(struct drm_device *dev)
5555 {
5556 struct drm_i915_private *dev_priv = dev->dev_private;
5557
5558 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
5559 I965_RCC_CLOCK_GATE_DISABLE |
5560 I965_RCPB_CLOCK_GATE_DISABLE |
5561 I965_ISC_CLOCK_GATE_DISABLE |
5562 I965_FBC_CLOCK_GATE_DISABLE);
5563 I915_WRITE(RENCLK_GATE_D2, 0);
5564 I915_WRITE(MI_ARB_STATE,
5565 _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
5566 }
5567
5568 static void gen3_init_clock_gating(struct drm_device *dev)
5569 {
5570 struct drm_i915_private *dev_priv = dev->dev_private;
5571 u32 dstate = I915_READ(D_STATE);
5572
5573 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
5574 DSTATE_DOT_CLOCK_GATING;
5575 I915_WRITE(D_STATE, dstate);
5576
5577 if (IS_PINEVIEW(dev))
5578 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
5579
5580 /* IIR "flip pending" means done if this bit is set */
5581 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
5582 }
5583
5584 static void i85x_init_clock_gating(struct drm_device *dev)
5585 {
5586 struct drm_i915_private *dev_priv = dev->dev_private;
5587
5588 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
5589 }
5590
5591 static void i830_init_clock_gating(struct drm_device *dev)
5592 {
5593 struct drm_i915_private *dev_priv = dev->dev_private;
5594
5595 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
5596 }
5597
5598 void intel_init_clock_gating(struct drm_device *dev)
5599 {
5600 struct drm_i915_private *dev_priv = dev->dev_private;
5601
5602 dev_priv->display.init_clock_gating(dev);
5603 }
5604
5605 void intel_suspend_hw(struct drm_device *dev)
5606 {
5607 if (HAS_PCH_LPT(dev))
5608 lpt_suspend_hw(dev);
5609 }
5610
5611 static bool is_always_on_power_domain(struct drm_device *dev,
5612 enum intel_display_power_domain domain)
5613 {
5614 unsigned long always_on_domains;
5615
5616 BUG_ON(BIT(domain) & ~POWER_DOMAIN_MASK);
5617
5618 if (IS_BROADWELL(dev)) {
5619 always_on_domains = BDW_ALWAYS_ON_POWER_DOMAINS;
5620 } else if (IS_HASWELL(dev)) {
5621 always_on_domains = HSW_ALWAYS_ON_POWER_DOMAINS;
5622 } else {
5623 WARN_ON(1);
5624 return true;
5625 }
5626
5627 return BIT(domain) & always_on_domains;
5628 }
5629
5630 /**
5631 * We should only use the power well if we explicitly asked the hardware to
5632 * enable it, so check if it's enabled and also check if we've requested it to
5633 * be enabled.
5634 */
5635 bool intel_display_power_enabled(struct drm_device *dev,
5636 enum intel_display_power_domain domain)
5637 {
5638 struct drm_i915_private *dev_priv = dev->dev_private;
5639
5640 if (!HAS_POWER_WELL(dev))
5641 return true;
5642
5643 if (is_always_on_power_domain(dev, domain))
5644 return true;
5645
5646 return I915_READ(HSW_PWR_WELL_DRIVER) ==
5647 (HSW_PWR_WELL_ENABLE_REQUEST | HSW_PWR_WELL_STATE_ENABLED);
5648 }
5649
5650 static void __intel_set_power_well(struct drm_device *dev, bool enable)
5651 {
5652 struct drm_i915_private *dev_priv = dev->dev_private;
5653 bool is_enabled, enable_requested;
5654 unsigned long irqflags;
5655 uint32_t tmp;
5656
5657 tmp = I915_READ(HSW_PWR_WELL_DRIVER);
5658 is_enabled = tmp & HSW_PWR_WELL_STATE_ENABLED;
5659 enable_requested = tmp & HSW_PWR_WELL_ENABLE_REQUEST;
5660
5661 if (enable) {
5662 if (!enable_requested)
5663 I915_WRITE(HSW_PWR_WELL_DRIVER,
5664 HSW_PWR_WELL_ENABLE_REQUEST);
5665
5666 if (!is_enabled) {
5667 DRM_DEBUG_KMS("Enabling power well\n");
5668 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
5669 HSW_PWR_WELL_STATE_ENABLED), 20))
5670 DRM_ERROR("Timeout enabling power well\n");
5671 }
5672
5673 if (IS_BROADWELL(dev)) {
5674 spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
5675 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_B),
5676 dev_priv->de_irq_mask[PIPE_B]);
5677 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_B),
5678 ~dev_priv->de_irq_mask[PIPE_B] |
5679 GEN8_PIPE_VBLANK);
5680 I915_WRITE(GEN8_DE_PIPE_IMR(PIPE_C),
5681 dev_priv->de_irq_mask[PIPE_C]);
5682 I915_WRITE(GEN8_DE_PIPE_IER(PIPE_C),
5683 ~dev_priv->de_irq_mask[PIPE_C] |
5684 GEN8_PIPE_VBLANK);
5685 POSTING_READ(GEN8_DE_PIPE_IER(PIPE_C));
5686 spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
5687 }
5688 } else {
5689 if (enable_requested) {
5690 enum pipe p;
5691
5692 I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
5693 POSTING_READ(HSW_PWR_WELL_DRIVER);
5694 DRM_DEBUG_KMS("Requesting to disable the power well\n");
5695
5696 /*
5697 * After this, the registers on the pipes that are part
5698 * of the power well will become zero, so we have to
5699 * adjust our counters according to that.
5700 *
5701 * FIXME: Should we do this in general in
5702 * drm_vblank_post_modeset?
5703 */
5704 spin_lock_irqsave(&dev->vbl_lock, irqflags);
5705 for_each_pipe(p)
5706 if (p != PIPE_A)
5707 dev->vblank[p].last = 0;
5708 spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
5709 }
5710 }
5711 }
5712
5713 static void __intel_power_well_get(struct drm_device *dev,
5714 struct i915_power_well *power_well)
5715 {
5716 if (!power_well->count++)
5717 __intel_set_power_well(dev, true);
5718 }
5719
5720 static void __intel_power_well_put(struct drm_device *dev,
5721 struct i915_power_well *power_well)
5722 {
5723 WARN_ON(!power_well->count);
5724 if (!--power_well->count && i915_disable_power_well)
5725 __intel_set_power_well(dev, false);
5726 }
5727
5728 void intel_display_power_get(struct drm_device *dev,
5729 enum intel_display_power_domain domain)
5730 {
5731 struct drm_i915_private *dev_priv = dev->dev_private;
5732 struct i915_power_domains *power_domains;
5733
5734 if (!HAS_POWER_WELL(dev))
5735 return;
5736
5737 if (is_always_on_power_domain(dev, domain))
5738 return;
5739
5740 power_domains = &dev_priv->power_domains;
5741
5742 mutex_lock(&power_domains->lock);
5743 __intel_power_well_get(dev, &power_domains->power_wells[0]);
5744 mutex_unlock(&power_domains->lock);
5745 }
5746
5747 void intel_display_power_put(struct drm_device *dev,
5748 enum intel_display_power_domain domain)
5749 {
5750 struct drm_i915_private *dev_priv = dev->dev_private;
5751 struct i915_power_domains *power_domains;
5752
5753 if (!HAS_POWER_WELL(dev))
5754 return;
5755
5756 if (is_always_on_power_domain(dev, domain))
5757 return;
5758
5759 power_domains = &dev_priv->power_domains;
5760
5761 mutex_lock(&power_domains->lock);
5762 __intel_power_well_put(dev, &power_domains->power_wells[0]);
5763 mutex_unlock(&power_domains->lock);
5764 }
5765
5766 static struct i915_power_domains *hsw_pwr;
5767
5768 /* Display audio driver power well request */
5769 void i915_request_power_well(void)
5770 {
5771 struct drm_i915_private *dev_priv;
5772
5773 if (WARN_ON(!hsw_pwr))
5774 return;
5775
5776 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
5777 power_domains);
5778 intel_display_power_get(dev_priv->dev, POWER_DOMAIN_AUDIO);
5779 }
5780 EXPORT_SYMBOL_GPL(i915_request_power_well);
5781
5782 /* Display audio driver power well release */
5783 void i915_release_power_well(void)
5784 {
5785 struct drm_i915_private *dev_priv;
5786
5787 if (WARN_ON(!hsw_pwr))
5788 return;
5789
5790 dev_priv = container_of(hsw_pwr, struct drm_i915_private,
5791 power_domains);
5792 intel_display_power_put(dev_priv->dev, POWER_DOMAIN_AUDIO);
5793 }
5794 EXPORT_SYMBOL_GPL(i915_release_power_well);
5795
5796 int intel_power_domains_init(struct drm_device *dev)
5797 {
5798 struct drm_i915_private *dev_priv = dev->dev_private;
5799 struct i915_power_domains *power_domains = &dev_priv->power_domains;
5800 struct i915_power_well *power_well;
5801
5802 mutex_init(&power_domains->lock);
5803 hsw_pwr = power_domains;
5804
5805 power_well = &power_domains->power_wells[0];
5806 power_well->count = 0;
5807
5808 return 0;
5809 }
5810
5811 void intel_power_domains_remove(struct drm_device *dev)
5812 {
5813 hsw_pwr = NULL;
5814 }
5815
5816 static void intel_power_domains_resume(struct drm_device *dev)
5817 {
5818 struct drm_i915_private *dev_priv = dev->dev_private;
5819 struct i915_power_domains *power_domains = &dev_priv->power_domains;
5820 struct i915_power_well *power_well;
5821
5822 if (!HAS_POWER_WELL(dev))
5823 return;
5824
5825 mutex_lock(&power_domains->lock);
5826
5827 power_well = &power_domains->power_wells[0];
5828 __intel_set_power_well(dev, power_well->count > 0);
5829
5830 mutex_unlock(&power_domains->lock);
5831 }
5832
5833 /*
5834 * Starting with Haswell, we have a "Power Down Well" that can be turned off
5835 * when not needed anymore. We have 4 registers that can request the power well
5836 * to be enabled, and it will only be disabled if none of the registers is
5837 * requesting it to be enabled.
5838 */
5839 void intel_power_domains_init_hw(struct drm_device *dev)
5840 {
5841 struct drm_i915_private *dev_priv = dev->dev_private;
5842
5843 if (!HAS_POWER_WELL(dev))
5844 return;
5845
5846 /* For now, we need the power well to be always enabled. */
5847 intel_display_set_init_power(dev, true);
5848 intel_power_domains_resume(dev);
5849
5850 /* We're taking over the BIOS, so clear any requests made by it since
5851 * the driver is in charge now. */
5852 if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE_REQUEST)
5853 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
5854 }
5855
5856 /* Disables PC8 so we can use the GMBUS and DP AUX interrupts. */
5857 void intel_aux_display_runtime_get(struct drm_i915_private *dev_priv)
5858 {
5859 hsw_disable_package_c8(dev_priv);
5860 }
5861
5862 void intel_aux_display_runtime_put(struct drm_i915_private *dev_priv)
5863 {
5864 hsw_enable_package_c8(dev_priv);
5865 }
5866
5867 /* Set up chip specific power management-related functions */
5868 void intel_init_pm(struct drm_device *dev)
5869 {
5870 struct drm_i915_private *dev_priv = dev->dev_private;
5871
5872 if (I915_HAS_FBC(dev)) {
5873 if (HAS_PCH_SPLIT(dev)) {
5874 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
5875 if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
5876 dev_priv->display.enable_fbc =
5877 gen7_enable_fbc;
5878 else
5879 dev_priv->display.enable_fbc =
5880 ironlake_enable_fbc;
5881 dev_priv->display.disable_fbc = ironlake_disable_fbc;
5882 } else if (IS_GM45(dev)) {
5883 dev_priv->display.fbc_enabled = g4x_fbc_enabled;
5884 dev_priv->display.enable_fbc = g4x_enable_fbc;
5885 dev_priv->display.disable_fbc = g4x_disable_fbc;
5886 } else if (IS_CRESTLINE(dev)) {
5887 dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
5888 dev_priv->display.enable_fbc = i8xx_enable_fbc;
5889 dev_priv->display.disable_fbc = i8xx_disable_fbc;
5890 }
5891 /* 855GM needs testing */
5892 }
5893
5894 /* For cxsr */
5895 if (IS_PINEVIEW(dev))
5896 i915_pineview_get_mem_freq(dev);
5897 else if (IS_GEN5(dev))
5898 i915_ironlake_get_mem_freq(dev);
5899
5900 /* For FIFO watermark updates */
5901 if (HAS_PCH_SPLIT(dev)) {
5902 intel_setup_wm_latency(dev);
5903
5904 if (IS_GEN5(dev)) {
5905 if (dev_priv->wm.pri_latency[1] &&
5906 dev_priv->wm.spr_latency[1] &&
5907 dev_priv->wm.cur_latency[1])
5908 dev_priv->display.update_wm = ironlake_update_wm;
5909 else {
5910 DRM_DEBUG_KMS("Failed to get proper latency. "
5911 "Disable CxSR\n");
5912 dev_priv->display.update_wm = NULL;
5913 }
5914 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
5915 } else if (IS_GEN6(dev)) {
5916 if (dev_priv->wm.pri_latency[0] &&
5917 dev_priv->wm.spr_latency[0] &&
5918 dev_priv->wm.cur_latency[0]) {
5919 dev_priv->display.update_wm = sandybridge_update_wm;
5920 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
5921 } else {
5922 DRM_DEBUG_KMS("Failed to read display plane latency. "
5923 "Disable CxSR\n");
5924 dev_priv->display.update_wm = NULL;
5925 }
5926 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
5927 } else if (IS_IVYBRIDGE(dev)) {
5928 if (dev_priv->wm.pri_latency[0] &&
5929 dev_priv->wm.spr_latency[0] &&
5930 dev_priv->wm.cur_latency[0]) {
5931 dev_priv->display.update_wm = ivybridge_update_wm;
5932 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
5933 } else {
5934 DRM_DEBUG_KMS("Failed to read display plane latency. "
5935 "Disable CxSR\n");
5936 dev_priv->display.update_wm = NULL;
5937 }
5938 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
5939 } else if (IS_HASWELL(dev)) {
5940 if (dev_priv->wm.pri_latency[0] &&
5941 dev_priv->wm.spr_latency[0] &&
5942 dev_priv->wm.cur_latency[0]) {
5943 dev_priv->display.update_wm = haswell_update_wm;
5944 dev_priv->display.update_sprite_wm =
5945 haswell_update_sprite_wm;
5946 } else {
5947 DRM_DEBUG_KMS("Failed to read display plane latency. "
5948 "Disable CxSR\n");
5949 dev_priv->display.update_wm = NULL;
5950 }
5951 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
5952 } else if (INTEL_INFO(dev)->gen == 8) {
5953 dev_priv->display.init_clock_gating = gen8_init_clock_gating;
5954 } else
5955 dev_priv->display.update_wm = NULL;
5956 } else if (IS_VALLEYVIEW(dev)) {
5957 dev_priv->display.update_wm = valleyview_update_wm;
5958 dev_priv->display.init_clock_gating =
5959 valleyview_init_clock_gating;
5960 } else if (IS_PINEVIEW(dev)) {
5961 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
5962 dev_priv->is_ddr3,
5963 dev_priv->fsb_freq,
5964 dev_priv->mem_freq)) {
5965 DRM_INFO("failed to find known CxSR latency "
5966 "(found ddr%s fsb freq %d, mem freq %d), "
5967 "disabling CxSR\n",
5968 (dev_priv->is_ddr3 == 1) ? "3" : "2",
5969 dev_priv->fsb_freq, dev_priv->mem_freq);
5970 /* Disable CxSR and never update its watermark again */
5971 pineview_disable_cxsr(dev);
5972 dev_priv->display.update_wm = NULL;
5973 } else
5974 dev_priv->display.update_wm = pineview_update_wm;
5975 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
5976 } else if (IS_G4X(dev)) {
5977 dev_priv->display.update_wm = g4x_update_wm;
5978 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
5979 } else if (IS_GEN4(dev)) {
5980 dev_priv->display.update_wm = i965_update_wm;
5981 if (IS_CRESTLINE(dev))
5982 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
5983 else if (IS_BROADWATER(dev))
5984 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
5985 } else if (IS_GEN3(dev)) {
5986 dev_priv->display.update_wm = i9xx_update_wm;
5987 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
5988 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
5989 } else if (IS_I865G(dev)) {
5990 dev_priv->display.update_wm = i830_update_wm;
5991 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
5992 dev_priv->display.get_fifo_size = i830_get_fifo_size;
5993 } else if (IS_I85X(dev)) {
5994 dev_priv->display.update_wm = i9xx_update_wm;
5995 dev_priv->display.get_fifo_size = i85x_get_fifo_size;
5996 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
5997 } else {
5998 dev_priv->display.update_wm = i830_update_wm;
5999 dev_priv->display.init_clock_gating = i830_init_clock_gating;
6000 if (IS_845G(dev))
6001 dev_priv->display.get_fifo_size = i845_get_fifo_size;
6002 else
6003 dev_priv->display.get_fifo_size = i830_get_fifo_size;
6004 }
6005 }
6006
6007 int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
6008 {
6009 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
6010
6011 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
6012 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
6013 return -EAGAIN;
6014 }
6015
6016 I915_WRITE(GEN6_PCODE_DATA, *val);
6017 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
6018
6019 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
6020 500)) {
6021 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
6022 return -ETIMEDOUT;
6023 }
6024
6025 *val = I915_READ(GEN6_PCODE_DATA);
6026 I915_WRITE(GEN6_PCODE_DATA, 0);
6027
6028 return 0;
6029 }
6030
6031 int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
6032 {
6033 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
6034
6035 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
6036 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
6037 return -EAGAIN;
6038 }
6039
6040 I915_WRITE(GEN6_PCODE_DATA, val);
6041 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
6042
6043 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
6044 500)) {
6045 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
6046 return -ETIMEDOUT;
6047 }
6048
6049 I915_WRITE(GEN6_PCODE_DATA, 0);
6050
6051 return 0;
6052 }
6053
6054 int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val)
6055 {
6056 int div;
6057
6058 /* 4 x czclk */
6059 switch (dev_priv->mem_freq) {
6060 case 800:
6061 div = 10;
6062 break;
6063 case 1066:
6064 div = 12;
6065 break;
6066 case 1333:
6067 div = 16;
6068 break;
6069 default:
6070 return -1;
6071 }
6072
6073 return DIV_ROUND_CLOSEST(dev_priv->mem_freq * (val + 6 - 0xbd), 4 * div);
6074 }
6075
6076 int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val)
6077 {
6078 int mul;
6079
6080 /* 4 x czclk */
6081 switch (dev_priv->mem_freq) {
6082 case 800:
6083 mul = 10;
6084 break;
6085 case 1066:
6086 mul = 12;
6087 break;
6088 case 1333:
6089 mul = 16;
6090 break;
6091 default:
6092 return -1;
6093 }
6094
6095 return DIV_ROUND_CLOSEST(4 * mul * val, dev_priv->mem_freq) + 0xbd - 6;
6096 }
6097
6098 void intel_pm_init(struct drm_device *dev)
6099 {
6100 struct drm_i915_private *dev_priv = dev->dev_private;
6101
6102 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
6103 intel_gen6_powersave_work);
6104 }
Linux kernel - Deliverables
This page took 0.272333 seconds and 6 git commands to generate.