2 * Copyright © 2012 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Eugeni Dodonov <eugeni.dodonov@intel.com>
28 #include <linux/cpufreq.h>
30 #include "intel_drv.h"
31 #include "../../../platform/x86/intel_ips.h"
32 #include <linux/module.h>
34 #define FORCEWAKE_ACK_TIMEOUT_MS 2
36 /* FBC, or Frame Buffer Compression, is a technique employed to compress the
37 * framebuffer contents in-memory, aiming at reducing the required bandwidth
38 * during in-memory transfers and, therefore, reduce the power packet.
40 * The benefits of FBC are mostly visible with solid backgrounds and
41 * variation-less patterns.
43 * FBC-related functionality can be enabled by the means of the
44 * i915.i915_enable_fbc parameter
47 static bool intel_crtc_active(struct drm_crtc *crtc)
49 /* Be paranoid as we can arrive here with only partial
50 * state retrieved from the hardware during setup.
52 return to_intel_crtc(crtc)->active && crtc->fb && crtc->mode.clock;
55 static void i8xx_disable_fbc(struct drm_device *dev)
57 struct drm_i915_private *dev_priv = dev->dev_private;
60 /* Disable compression */
61 fbc_ctl = I915_READ(FBC_CONTROL);
62 if ((fbc_ctl & FBC_CTL_EN) == 0)
65 fbc_ctl &= ~FBC_CTL_EN;
66 I915_WRITE(FBC_CONTROL, fbc_ctl);
68 /* Wait for compressing bit to clear */
69 if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) {
70 DRM_DEBUG_KMS("FBC idle timed out\n");
74 DRM_DEBUG_KMS("disabled FBC\n");
77 static void i8xx_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
79 struct drm_device *dev = crtc->dev;
80 struct drm_i915_private *dev_priv = dev->dev_private;
81 struct drm_framebuffer *fb = crtc->fb;
82 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
83 struct drm_i915_gem_object *obj = intel_fb->obj;
84 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
87 u32 fbc_ctl, fbc_ctl2;
89 cfb_pitch = dev_priv->cfb_size / FBC_LL_SIZE;
90 if (fb->pitches[0] < cfb_pitch)
91 cfb_pitch = fb->pitches[0];
93 /* FBC_CTL wants 64B units */
94 cfb_pitch = (cfb_pitch / 64) - 1;
95 plane = intel_crtc->plane == 0 ? FBC_CTL_PLANEA : FBC_CTL_PLANEB;
98 for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++)
99 I915_WRITE(FBC_TAG + (i * 4), 0);
102 fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE;
104 I915_WRITE(FBC_CONTROL2, fbc_ctl2);
105 I915_WRITE(FBC_FENCE_OFF, crtc->y);
108 fbc_ctl = FBC_CTL_EN | FBC_CTL_PERIODIC;
110 fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */
111 fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT;
112 fbc_ctl |= (interval & 0x2fff) << FBC_CTL_INTERVAL_SHIFT;
113 fbc_ctl |= obj->fence_reg;
114 I915_WRITE(FBC_CONTROL, fbc_ctl);
116 DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c, ",
117 cfb_pitch, crtc->y, plane_name(intel_crtc->plane));
120 static bool i8xx_fbc_enabled(struct drm_device *dev)
122 struct drm_i915_private *dev_priv = dev->dev_private;
124 return I915_READ(FBC_CONTROL) & FBC_CTL_EN;
127 static void g4x_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
129 struct drm_device *dev = crtc->dev;
130 struct drm_i915_private *dev_priv = dev->dev_private;
131 struct drm_framebuffer *fb = crtc->fb;
132 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
133 struct drm_i915_gem_object *obj = intel_fb->obj;
134 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
135 int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
136 unsigned long stall_watermark = 200;
139 dpfc_ctl = plane | DPFC_SR_EN | DPFC_CTL_LIMIT_1X;
140 dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg;
141 I915_WRITE(DPFC_CHICKEN, DPFC_HT_MODIFY);
143 I915_WRITE(DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
144 (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
145 (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
146 I915_WRITE(DPFC_FENCE_YOFF, crtc->y);
149 I915_WRITE(DPFC_CONTROL, I915_READ(DPFC_CONTROL) | DPFC_CTL_EN);
151 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
154 static void g4x_disable_fbc(struct drm_device *dev)
156 struct drm_i915_private *dev_priv = dev->dev_private;
159 /* Disable compression */
160 dpfc_ctl = I915_READ(DPFC_CONTROL);
161 if (dpfc_ctl & DPFC_CTL_EN) {
162 dpfc_ctl &= ~DPFC_CTL_EN;
163 I915_WRITE(DPFC_CONTROL, dpfc_ctl);
165 DRM_DEBUG_KMS("disabled FBC\n");
169 static bool g4x_fbc_enabled(struct drm_device *dev)
171 struct drm_i915_private *dev_priv = dev->dev_private;
173 return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN;
176 static void sandybridge_blit_fbc_update(struct drm_device *dev)
178 struct drm_i915_private *dev_priv = dev->dev_private;
181 /* Make sure blitter notifies FBC of writes */
182 gen6_gt_force_wake_get(dev_priv);
183 blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD);
184 blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY <<
185 GEN6_BLITTER_LOCK_SHIFT;
186 I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
187 blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY;
188 I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
189 blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY <<
190 GEN6_BLITTER_LOCK_SHIFT);
191 I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
192 POSTING_READ(GEN6_BLITTER_ECOSKPD);
193 gen6_gt_force_wake_put(dev_priv);
196 static void ironlake_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
198 struct drm_device *dev = crtc->dev;
199 struct drm_i915_private *dev_priv = dev->dev_private;
200 struct drm_framebuffer *fb = crtc->fb;
201 struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
202 struct drm_i915_gem_object *obj = intel_fb->obj;
203 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
204 int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
205 unsigned long stall_watermark = 200;
208 dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
209 dpfc_ctl &= DPFC_RESERVED;
210 dpfc_ctl |= (plane | DPFC_CTL_LIMIT_1X);
211 /* Set persistent mode for front-buffer rendering, ala X. */
212 dpfc_ctl |= DPFC_CTL_PERSISTENT_MODE;
213 dpfc_ctl |= (DPFC_CTL_FENCE_EN | obj->fence_reg);
214 I915_WRITE(ILK_DPFC_CHICKEN, DPFC_HT_MODIFY);
216 I915_WRITE(ILK_DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
217 (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
218 (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
219 I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
220 I915_WRITE(ILK_FBC_RT_BASE, obj->gtt_offset | ILK_FBC_RT_VALID);
222 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
225 I915_WRITE(SNB_DPFC_CTL_SA,
226 SNB_CPU_FENCE_ENABLE | obj->fence_reg);
227 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
228 sandybridge_blit_fbc_update(dev);
231 DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
234 static void ironlake_disable_fbc(struct drm_device *dev)
236 struct drm_i915_private *dev_priv = dev->dev_private;
239 /* Disable compression */
240 dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
241 if (dpfc_ctl & DPFC_CTL_EN) {
242 dpfc_ctl &= ~DPFC_CTL_EN;
243 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl);
245 DRM_DEBUG_KMS("disabled FBC\n");
249 static bool ironlake_fbc_enabled(struct drm_device *dev)
251 struct drm_i915_private *dev_priv = dev->dev_private;
253 return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
256 bool intel_fbc_enabled(struct drm_device *dev)
258 struct drm_i915_private *dev_priv = dev->dev_private;
260 if (!dev_priv->display.fbc_enabled)
263 return dev_priv->display.fbc_enabled(dev);
266 static void intel_fbc_work_fn(struct work_struct *__work)
268 struct intel_fbc_work *work =
269 container_of(to_delayed_work(__work),
270 struct intel_fbc_work, work);
271 struct drm_device *dev = work->crtc->dev;
272 struct drm_i915_private *dev_priv = dev->dev_private;
274 mutex_lock(&dev->struct_mutex);
275 if (work == dev_priv->fbc_work) {
276 /* Double check that we haven't switched fb without cancelling
279 if (work->crtc->fb == work->fb) {
280 dev_priv->display.enable_fbc(work->crtc,
283 dev_priv->cfb_plane = to_intel_crtc(work->crtc)->plane;
284 dev_priv->cfb_fb = work->crtc->fb->base.id;
285 dev_priv->cfb_y = work->crtc->y;
288 dev_priv->fbc_work = NULL;
290 mutex_unlock(&dev->struct_mutex);
295 static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
297 if (dev_priv->fbc_work == NULL)
300 DRM_DEBUG_KMS("cancelling pending FBC enable\n");
302 /* Synchronisation is provided by struct_mutex and checking of
303 * dev_priv->fbc_work, so we can perform the cancellation
304 * entirely asynchronously.
306 if (cancel_delayed_work(&dev_priv->fbc_work->work))
307 /* tasklet was killed before being run, clean up */
308 kfree(dev_priv->fbc_work);
310 /* Mark the work as no longer wanted so that if it does
311 * wake-up (because the work was already running and waiting
312 * for our mutex), it will discover that is no longer
315 dev_priv->fbc_work = NULL;
318 void intel_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
320 struct intel_fbc_work *work;
321 struct drm_device *dev = crtc->dev;
322 struct drm_i915_private *dev_priv = dev->dev_private;
324 if (!dev_priv->display.enable_fbc)
327 intel_cancel_fbc_work(dev_priv);
329 work = kzalloc(sizeof *work, GFP_KERNEL);
331 dev_priv->display.enable_fbc(crtc, interval);
337 work->interval = interval;
338 INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
340 dev_priv->fbc_work = work;
342 DRM_DEBUG_KMS("scheduling delayed FBC enable\n");
344 /* Delay the actual enabling to let pageflipping cease and the
345 * display to settle before starting the compression. Note that
346 * this delay also serves a second purpose: it allows for a
347 * vblank to pass after disabling the FBC before we attempt
348 * to modify the control registers.
350 * A more complicated solution would involve tracking vblanks
351 * following the termination of the page-flipping sequence
352 * and indeed performing the enable as a co-routine and not
353 * waiting synchronously upon the vblank.
355 schedule_delayed_work(&work->work, msecs_to_jiffies(50));
358 void intel_disable_fbc(struct drm_device *dev)
360 struct drm_i915_private *dev_priv = dev->dev_private;
362 intel_cancel_fbc_work(dev_priv);
364 if (!dev_priv->display.disable_fbc)
367 dev_priv->display.disable_fbc(dev);
368 dev_priv->cfb_plane = -1;
372 * intel_update_fbc - enable/disable FBC as needed
373 * @dev: the drm_device
375 * Set up the framebuffer compression hardware at mode set time. We
376 * enable it if possible:
377 * - plane A only (on pre-965)
378 * - no pixel mulitply/line duplication
379 * - no alpha buffer discard
381 * - framebuffer <= 2048 in width, 1536 in height
383 * We can't assume that any compression will take place (worst case),
384 * so the compressed buffer has to be the same size as the uncompressed
385 * one. It also must reside (along with the line length buffer) in
388 * We need to enable/disable FBC on a global basis.
390 void intel_update_fbc(struct drm_device *dev)
392 struct drm_i915_private *dev_priv = dev->dev_private;
393 struct drm_crtc *crtc = NULL, *tmp_crtc;
394 struct intel_crtc *intel_crtc;
395 struct drm_framebuffer *fb;
396 struct intel_framebuffer *intel_fb;
397 struct drm_i915_gem_object *obj;
403 if (!I915_HAS_FBC(dev))
407 * If FBC is already on, we just have to verify that we can
408 * keep it that way...
409 * Need to disable if:
410 * - more than one pipe is active
411 * - changing FBC params (stride, fence, mode)
412 * - new fb is too large to fit in compressed buffer
413 * - going to an unsupported config (interlace, pixel multiply, etc.)
415 list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) {
416 if (intel_crtc_active(tmp_crtc) &&
417 !to_intel_crtc(tmp_crtc)->primary_disabled) {
419 DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
420 dev_priv->no_fbc_reason = FBC_MULTIPLE_PIPES;
427 if (!crtc || crtc->fb == NULL) {
428 DRM_DEBUG_KMS("no output, disabling\n");
429 dev_priv->no_fbc_reason = FBC_NO_OUTPUT;
433 intel_crtc = to_intel_crtc(crtc);
435 intel_fb = to_intel_framebuffer(fb);
438 enable_fbc = i915_enable_fbc;
439 if (enable_fbc < 0) {
440 DRM_DEBUG_KMS("fbc set to per-chip default\n");
442 if (INTEL_INFO(dev)->gen <= 6)
446 DRM_DEBUG_KMS("fbc disabled per module param\n");
447 dev_priv->no_fbc_reason = FBC_MODULE_PARAM;
450 if ((crtc->mode.flags & DRM_MODE_FLAG_INTERLACE) ||
451 (crtc->mode.flags & DRM_MODE_FLAG_DBLSCAN)) {
452 DRM_DEBUG_KMS("mode incompatible with compression, "
454 dev_priv->no_fbc_reason = FBC_UNSUPPORTED_MODE;
457 if ((crtc->mode.hdisplay > 2048) ||
458 (crtc->mode.vdisplay > 1536)) {
459 DRM_DEBUG_KMS("mode too large for compression, disabling\n");
460 dev_priv->no_fbc_reason = FBC_MODE_TOO_LARGE;
463 if ((IS_I915GM(dev) || IS_I945GM(dev)) && intel_crtc->plane != 0) {
464 DRM_DEBUG_KMS("plane not 0, disabling compression\n");
465 dev_priv->no_fbc_reason = FBC_BAD_PLANE;
469 /* The use of a CPU fence is mandatory in order to detect writes
470 * by the CPU to the scanout and trigger updates to the FBC.
472 if (obj->tiling_mode != I915_TILING_X ||
473 obj->fence_reg == I915_FENCE_REG_NONE) {
474 DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
475 dev_priv->no_fbc_reason = FBC_NOT_TILED;
479 /* If the kernel debugger is active, always disable compression */
483 if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) {
484 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
485 dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
489 /* If the scanout has not changed, don't modify the FBC settings.
490 * Note that we make the fundamental assumption that the fb->obj
491 * cannot be unpinned (and have its GTT offset and fence revoked)
492 * without first being decoupled from the scanout and FBC disabled.
494 if (dev_priv->cfb_plane == intel_crtc->plane &&
495 dev_priv->cfb_fb == fb->base.id &&
496 dev_priv->cfb_y == crtc->y)
499 if (intel_fbc_enabled(dev)) {
500 /* We update FBC along two paths, after changing fb/crtc
501 * configuration (modeswitching) and after page-flipping
502 * finishes. For the latter, we know that not only did
503 * we disable the FBC at the start of the page-flip
504 * sequence, but also more than one vblank has passed.
506 * For the former case of modeswitching, it is possible
507 * to switch between two FBC valid configurations
508 * instantaneously so we do need to disable the FBC
509 * before we can modify its control registers. We also
510 * have to wait for the next vblank for that to take
511 * effect. However, since we delay enabling FBC we can
512 * assume that a vblank has passed since disabling and
513 * that we can safely alter the registers in the deferred
516 * In the scenario that we go from a valid to invalid
517 * and then back to valid FBC configuration we have
518 * no strict enforcement that a vblank occurred since
519 * disabling the FBC. However, along all current pipe
520 * disabling paths we do need to wait for a vblank at
521 * some point. And we wait before enabling FBC anyway.
523 DRM_DEBUG_KMS("disabling active FBC for update\n");
524 intel_disable_fbc(dev);
527 intel_enable_fbc(crtc, 500);
531 /* Multiple disables should be harmless */
532 if (intel_fbc_enabled(dev)) {
533 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
534 intel_disable_fbc(dev);
536 i915_gem_stolen_cleanup_compression(dev);
539 static void i915_pineview_get_mem_freq(struct drm_device *dev)
541 drm_i915_private_t *dev_priv = dev->dev_private;
544 tmp = I915_READ(CLKCFG);
546 switch (tmp & CLKCFG_FSB_MASK) {
548 dev_priv->fsb_freq = 533; /* 133*4 */
551 dev_priv->fsb_freq = 800; /* 200*4 */
554 dev_priv->fsb_freq = 667; /* 167*4 */
557 dev_priv->fsb_freq = 400; /* 100*4 */
561 switch (tmp & CLKCFG_MEM_MASK) {
563 dev_priv->mem_freq = 533;
566 dev_priv->mem_freq = 667;
569 dev_priv->mem_freq = 800;
573 /* detect pineview DDR3 setting */
574 tmp = I915_READ(CSHRDDR3CTL);
575 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
578 static void i915_ironlake_get_mem_freq(struct drm_device *dev)
580 drm_i915_private_t *dev_priv = dev->dev_private;
583 ddrpll = I915_READ16(DDRMPLL1);
584 csipll = I915_READ16(CSIPLL0);
586 switch (ddrpll & 0xff) {
588 dev_priv->mem_freq = 800;
591 dev_priv->mem_freq = 1066;
594 dev_priv->mem_freq = 1333;
597 dev_priv->mem_freq = 1600;
600 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
602 dev_priv->mem_freq = 0;
606 dev_priv->ips.r_t = dev_priv->mem_freq;
608 switch (csipll & 0x3ff) {
610 dev_priv->fsb_freq = 3200;
613 dev_priv->fsb_freq = 3733;
616 dev_priv->fsb_freq = 4266;
619 dev_priv->fsb_freq = 4800;
622 dev_priv->fsb_freq = 5333;
625 dev_priv->fsb_freq = 5866;
628 dev_priv->fsb_freq = 6400;
631 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
633 dev_priv->fsb_freq = 0;
637 if (dev_priv->fsb_freq == 3200) {
638 dev_priv->ips.c_m = 0;
639 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
640 dev_priv->ips.c_m = 1;
642 dev_priv->ips.c_m = 2;
646 static const struct cxsr_latency cxsr_latency_table[] = {
647 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
648 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
649 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
650 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
651 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
653 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
654 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
655 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
656 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
657 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
659 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
660 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
661 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
662 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
663 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
665 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
666 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
667 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
668 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
669 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
671 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
672 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
673 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
674 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
675 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
677 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
678 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
679 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
680 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
681 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
684 static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
689 const struct cxsr_latency *latency;
692 if (fsb == 0 || mem == 0)
695 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
696 latency = &cxsr_latency_table[i];
697 if (is_desktop == latency->is_desktop &&
698 is_ddr3 == latency->is_ddr3 &&
699 fsb == latency->fsb_freq && mem == latency->mem_freq)
703 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
708 static void pineview_disable_cxsr(struct drm_device *dev)
710 struct drm_i915_private *dev_priv = dev->dev_private;
712 /* deactivate cxsr */
713 I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
717 * Latency for FIFO fetches is dependent on several factors:
718 * - memory configuration (speed, channels)
720 * - current MCH state
721 * It can be fairly high in some situations, so here we assume a fairly
722 * pessimal value. It's a tradeoff between extra memory fetches (if we
723 * set this value too high, the FIFO will fetch frequently to stay full)
724 * and power consumption (set it too low to save power and we might see
725 * FIFO underruns and display "flicker").
727 * A value of 5us seems to be a good balance; safe for very low end
728 * platforms but not overly aggressive on lower latency configs.
730 static const int latency_ns = 5000;
732 static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
734 struct drm_i915_private *dev_priv = dev->dev_private;
735 uint32_t dsparb = I915_READ(DSPARB);
738 size = dsparb & 0x7f;
740 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
742 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
743 plane ? "B" : "A", size);
748 static int i85x_get_fifo_size(struct drm_device *dev, int plane)
750 struct drm_i915_private *dev_priv = dev->dev_private;
751 uint32_t dsparb = I915_READ(DSPARB);
754 size = dsparb & 0x1ff;
756 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
757 size >>= 1; /* Convert to cachelines */
759 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
760 plane ? "B" : "A", size);
765 static int i845_get_fifo_size(struct drm_device *dev, int plane)
767 struct drm_i915_private *dev_priv = dev->dev_private;
768 uint32_t dsparb = I915_READ(DSPARB);
771 size = dsparb & 0x7f;
772 size >>= 2; /* Convert to cachelines */
774 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
781 static int i830_get_fifo_size(struct drm_device *dev, int plane)
783 struct drm_i915_private *dev_priv = dev->dev_private;
784 uint32_t dsparb = I915_READ(DSPARB);
787 size = dsparb & 0x7f;
788 size >>= 1; /* Convert to cachelines */
790 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
791 plane ? "B" : "A", size);
796 /* Pineview has different values for various configs */
797 static const struct intel_watermark_params pineview_display_wm = {
798 PINEVIEW_DISPLAY_FIFO,
802 PINEVIEW_FIFO_LINE_SIZE
804 static const struct intel_watermark_params pineview_display_hplloff_wm = {
805 PINEVIEW_DISPLAY_FIFO,
807 PINEVIEW_DFT_HPLLOFF_WM,
809 PINEVIEW_FIFO_LINE_SIZE
811 static const struct intel_watermark_params pineview_cursor_wm = {
812 PINEVIEW_CURSOR_FIFO,
813 PINEVIEW_CURSOR_MAX_WM,
814 PINEVIEW_CURSOR_DFT_WM,
815 PINEVIEW_CURSOR_GUARD_WM,
816 PINEVIEW_FIFO_LINE_SIZE,
818 static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
819 PINEVIEW_CURSOR_FIFO,
820 PINEVIEW_CURSOR_MAX_WM,
821 PINEVIEW_CURSOR_DFT_WM,
822 PINEVIEW_CURSOR_GUARD_WM,
823 PINEVIEW_FIFO_LINE_SIZE
825 static const struct intel_watermark_params g4x_wm_info = {
832 static const struct intel_watermark_params g4x_cursor_wm_info = {
839 static const struct intel_watermark_params valleyview_wm_info = {
840 VALLEYVIEW_FIFO_SIZE,
846 static const struct intel_watermark_params valleyview_cursor_wm_info = {
848 VALLEYVIEW_CURSOR_MAX_WM,
853 static const struct intel_watermark_params i965_cursor_wm_info = {
860 static const struct intel_watermark_params i945_wm_info = {
867 static const struct intel_watermark_params i915_wm_info = {
874 static const struct intel_watermark_params i855_wm_info = {
881 static const struct intel_watermark_params i830_wm_info = {
889 static const struct intel_watermark_params ironlake_display_wm_info = {
896 static const struct intel_watermark_params ironlake_cursor_wm_info = {
903 static const struct intel_watermark_params ironlake_display_srwm_info = {
905 ILK_DISPLAY_MAX_SRWM,
906 ILK_DISPLAY_DFT_SRWM,
910 static const struct intel_watermark_params ironlake_cursor_srwm_info = {
918 static const struct intel_watermark_params sandybridge_display_wm_info = {
925 static const struct intel_watermark_params sandybridge_cursor_wm_info = {
932 static const struct intel_watermark_params sandybridge_display_srwm_info = {
934 SNB_DISPLAY_MAX_SRWM,
935 SNB_DISPLAY_DFT_SRWM,
939 static const struct intel_watermark_params sandybridge_cursor_srwm_info = {
949 * intel_calculate_wm - calculate watermark level
950 * @clock_in_khz: pixel clock
951 * @wm: chip FIFO params
952 * @pixel_size: display pixel size
953 * @latency_ns: memory latency for the platform
955 * Calculate the watermark level (the level at which the display plane will
956 * start fetching from memory again). Each chip has a different display
957 * FIFO size and allocation, so the caller needs to figure that out and pass
958 * in the correct intel_watermark_params structure.
960 * As the pixel clock runs, the FIFO will be drained at a rate that depends
961 * on the pixel size. When it reaches the watermark level, it'll start
962 * fetching FIFO line sized based chunks from memory until the FIFO fills
963 * past the watermark point. If the FIFO drains completely, a FIFO underrun
964 * will occur, and a display engine hang could result.
966 static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
967 const struct intel_watermark_params *wm,
970 unsigned long latency_ns)
972 long entries_required, wm_size;
975 * Note: we need to make sure we don't overflow for various clock &
977 * clocks go from a few thousand to several hundred thousand.
978 * latency is usually a few thousand
980 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
982 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
984 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
986 wm_size = fifo_size - (entries_required + wm->guard_size);
988 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
990 /* Don't promote wm_size to unsigned... */
991 if (wm_size > (long)wm->max_wm)
992 wm_size = wm->max_wm;
994 wm_size = wm->default_wm;
998 static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1000 struct drm_crtc *crtc, *enabled = NULL;
1002 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
1003 if (intel_crtc_active(crtc)) {
1013 static void pineview_update_wm(struct drm_device *dev)
1015 struct drm_i915_private *dev_priv = dev->dev_private;
1016 struct drm_crtc *crtc;
1017 const struct cxsr_latency *latency;
1021 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1022 dev_priv->fsb_freq, dev_priv->mem_freq);
1024 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1025 pineview_disable_cxsr(dev);
1029 crtc = single_enabled_crtc(dev);
1031 int clock = crtc->mode.clock;
1032 int pixel_size = crtc->fb->bits_per_pixel / 8;
1035 wm = intel_calculate_wm(clock, &pineview_display_wm,
1036 pineview_display_wm.fifo_size,
1037 pixel_size, latency->display_sr);
1038 reg = I915_READ(DSPFW1);
1039 reg &= ~DSPFW_SR_MASK;
1040 reg |= wm << DSPFW_SR_SHIFT;
1041 I915_WRITE(DSPFW1, reg);
1042 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1045 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1046 pineview_display_wm.fifo_size,
1047 pixel_size, latency->cursor_sr);
1048 reg = I915_READ(DSPFW3);
1049 reg &= ~DSPFW_CURSOR_SR_MASK;
1050 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1051 I915_WRITE(DSPFW3, reg);
1053 /* Display HPLL off SR */
1054 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1055 pineview_display_hplloff_wm.fifo_size,
1056 pixel_size, latency->display_hpll_disable);
1057 reg = I915_READ(DSPFW3);
1058 reg &= ~DSPFW_HPLL_SR_MASK;
1059 reg |= wm & DSPFW_HPLL_SR_MASK;
1060 I915_WRITE(DSPFW3, reg);
1062 /* cursor HPLL off SR */
1063 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1064 pineview_display_hplloff_wm.fifo_size,
1065 pixel_size, latency->cursor_hpll_disable);
1066 reg = I915_READ(DSPFW3);
1067 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1068 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1069 I915_WRITE(DSPFW3, reg);
1070 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1074 I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1075 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1077 pineview_disable_cxsr(dev);
1078 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1082 static bool g4x_compute_wm0(struct drm_device *dev,
1084 const struct intel_watermark_params *display,
1085 int display_latency_ns,
1086 const struct intel_watermark_params *cursor,
1087 int cursor_latency_ns,
1091 struct drm_crtc *crtc;
1092 int htotal, hdisplay, clock, pixel_size;
1093 int line_time_us, line_count;
1094 int entries, tlb_miss;
1096 crtc = intel_get_crtc_for_plane(dev, plane);
1097 if (!intel_crtc_active(crtc)) {
1098 *cursor_wm = cursor->guard_size;
1099 *plane_wm = display->guard_size;
1103 htotal = crtc->mode.htotal;
1104 hdisplay = crtc->mode.hdisplay;
1105 clock = crtc->mode.clock;
1106 pixel_size = crtc->fb->bits_per_pixel / 8;
1108 /* Use the small buffer method to calculate plane watermark */
1109 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1110 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1112 entries += tlb_miss;
1113 entries = DIV_ROUND_UP(entries, display->cacheline_size);
1114 *plane_wm = entries + display->guard_size;
1115 if (*plane_wm > (int)display->max_wm)
1116 *plane_wm = display->max_wm;
1118 /* Use the large buffer method to calculate cursor watermark */
1119 line_time_us = ((htotal * 1000) / clock);
1120 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
1121 entries = line_count * 64 * pixel_size;
1122 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1124 entries += tlb_miss;
1125 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1126 *cursor_wm = entries + cursor->guard_size;
1127 if (*cursor_wm > (int)cursor->max_wm)
1128 *cursor_wm = (int)cursor->max_wm;
1134 * Check the wm result.
1136 * If any calculated watermark values is larger than the maximum value that
1137 * can be programmed into the associated watermark register, that watermark
1140 static bool g4x_check_srwm(struct drm_device *dev,
1141 int display_wm, int cursor_wm,
1142 const struct intel_watermark_params *display,
1143 const struct intel_watermark_params *cursor)
1145 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1146 display_wm, cursor_wm);
1148 if (display_wm > display->max_wm) {
1149 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1150 display_wm, display->max_wm);
1154 if (cursor_wm > cursor->max_wm) {
1155 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1156 cursor_wm, cursor->max_wm);
1160 if (!(display_wm || cursor_wm)) {
1161 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1168 static bool g4x_compute_srwm(struct drm_device *dev,
1171 const struct intel_watermark_params *display,
1172 const struct intel_watermark_params *cursor,
1173 int *display_wm, int *cursor_wm)
1175 struct drm_crtc *crtc;
1176 int hdisplay, htotal, pixel_size, clock;
1177 unsigned long line_time_us;
1178 int line_count, line_size;
1183 *display_wm = *cursor_wm = 0;
1187 crtc = intel_get_crtc_for_plane(dev, plane);
1188 hdisplay = crtc->mode.hdisplay;
1189 htotal = crtc->mode.htotal;
1190 clock = crtc->mode.clock;
1191 pixel_size = crtc->fb->bits_per_pixel / 8;
1193 line_time_us = (htotal * 1000) / clock;
1194 line_count = (latency_ns / line_time_us + 1000) / 1000;
1195 line_size = hdisplay * pixel_size;
1197 /* Use the minimum of the small and large buffer method for primary */
1198 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1199 large = line_count * line_size;
1201 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1202 *display_wm = entries + display->guard_size;
1204 /* calculate the self-refresh watermark for display cursor */
1205 entries = line_count * pixel_size * 64;
1206 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1207 *cursor_wm = entries + cursor->guard_size;
1209 return g4x_check_srwm(dev,
1210 *display_wm, *cursor_wm,
1214 static bool vlv_compute_drain_latency(struct drm_device *dev,
1216 int *plane_prec_mult,
1218 int *cursor_prec_mult,
1221 struct drm_crtc *crtc;
1222 int clock, pixel_size;
1225 crtc = intel_get_crtc_for_plane(dev, plane);
1226 if (!intel_crtc_active(crtc))
1229 clock = crtc->mode.clock; /* VESA DOT Clock */
1230 pixel_size = crtc->fb->bits_per_pixel / 8; /* BPP */
1232 entries = (clock / 1000) * pixel_size;
1233 *plane_prec_mult = (entries > 256) ?
1234 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1235 *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1238 entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */
1239 *cursor_prec_mult = (entries > 256) ?
1240 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1241 *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1247 * Update drain latency registers of memory arbiter
1249 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1250 * to be programmed. Each plane has a drain latency multiplier and a drain
1254 static void vlv_update_drain_latency(struct drm_device *dev)
1256 struct drm_i915_private *dev_priv = dev->dev_private;
1257 int planea_prec, planea_dl, planeb_prec, planeb_dl;
1258 int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1259 int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1262 /* For plane A, Cursor A */
1263 if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1264 &cursor_prec_mult, &cursora_dl)) {
1265 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1266 DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1267 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1268 DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1270 I915_WRITE(VLV_DDL1, cursora_prec |
1271 (cursora_dl << DDL_CURSORA_SHIFT) |
1272 planea_prec | planea_dl);
1275 /* For plane B, Cursor B */
1276 if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1277 &cursor_prec_mult, &cursorb_dl)) {
1278 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1279 DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1280 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1281 DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1283 I915_WRITE(VLV_DDL2, cursorb_prec |
1284 (cursorb_dl << DDL_CURSORB_SHIFT) |
1285 planeb_prec | planeb_dl);
1289 #define single_plane_enabled(mask) is_power_of_2(mask)
1291 static void valleyview_update_wm(struct drm_device *dev)
1293 static const int sr_latency_ns = 12000;
1294 struct drm_i915_private *dev_priv = dev->dev_private;
1295 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1296 int plane_sr, cursor_sr;
1297 int ignore_plane_sr, ignore_cursor_sr;
1298 unsigned int enabled = 0;
1300 vlv_update_drain_latency(dev);
1302 if (g4x_compute_wm0(dev, 0,
1303 &valleyview_wm_info, latency_ns,
1304 &valleyview_cursor_wm_info, latency_ns,
1305 &planea_wm, &cursora_wm))
1308 if (g4x_compute_wm0(dev, 1,
1309 &valleyview_wm_info, latency_ns,
1310 &valleyview_cursor_wm_info, latency_ns,
1311 &planeb_wm, &cursorb_wm))
1314 if (single_plane_enabled(enabled) &&
1315 g4x_compute_srwm(dev, ffs(enabled) - 1,
1317 &valleyview_wm_info,
1318 &valleyview_cursor_wm_info,
1319 &plane_sr, &ignore_cursor_sr) &&
1320 g4x_compute_srwm(dev, ffs(enabled) - 1,
1322 &valleyview_wm_info,
1323 &valleyview_cursor_wm_info,
1324 &ignore_plane_sr, &cursor_sr)) {
1325 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
1327 I915_WRITE(FW_BLC_SELF_VLV,
1328 I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
1329 plane_sr = cursor_sr = 0;
1332 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1333 planea_wm, cursora_wm,
1334 planeb_wm, cursorb_wm,
1335 plane_sr, cursor_sr);
1338 (plane_sr << DSPFW_SR_SHIFT) |
1339 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1340 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1343 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1344 (cursora_wm << DSPFW_CURSORA_SHIFT));
1346 (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1347 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1350 static void g4x_update_wm(struct drm_device *dev)
1352 static const int sr_latency_ns = 12000;
1353 struct drm_i915_private *dev_priv = dev->dev_private;
1354 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1355 int plane_sr, cursor_sr;
1356 unsigned int enabled = 0;
1358 if (g4x_compute_wm0(dev, 0,
1359 &g4x_wm_info, latency_ns,
1360 &g4x_cursor_wm_info, latency_ns,
1361 &planea_wm, &cursora_wm))
1364 if (g4x_compute_wm0(dev, 1,
1365 &g4x_wm_info, latency_ns,
1366 &g4x_cursor_wm_info, latency_ns,
1367 &planeb_wm, &cursorb_wm))
1370 if (single_plane_enabled(enabled) &&
1371 g4x_compute_srwm(dev, ffs(enabled) - 1,
1374 &g4x_cursor_wm_info,
1375 &plane_sr, &cursor_sr)) {
1376 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1378 I915_WRITE(FW_BLC_SELF,
1379 I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
1380 plane_sr = cursor_sr = 0;
1383 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1384 planea_wm, cursora_wm,
1385 planeb_wm, cursorb_wm,
1386 plane_sr, cursor_sr);
1389 (plane_sr << DSPFW_SR_SHIFT) |
1390 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1391 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1394 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1395 (cursora_wm << DSPFW_CURSORA_SHIFT));
1396 /* HPLL off in SR has some issues on G4x... disable it */
1398 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
1399 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1402 static void i965_update_wm(struct drm_device *dev)
1404 struct drm_i915_private *dev_priv = dev->dev_private;
1405 struct drm_crtc *crtc;
1409 /* Calc sr entries for one plane configs */
1410 crtc = single_enabled_crtc(dev);
1412 /* self-refresh has much higher latency */
1413 static const int sr_latency_ns = 12000;
1414 int clock = crtc->mode.clock;
1415 int htotal = crtc->mode.htotal;
1416 int hdisplay = crtc->mode.hdisplay;
1417 int pixel_size = crtc->fb->bits_per_pixel / 8;
1418 unsigned long line_time_us;
1421 line_time_us = ((htotal * 1000) / clock);
1423 /* Use ns/us then divide to preserve precision */
1424 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1425 pixel_size * hdisplay;
1426 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1427 srwm = I965_FIFO_SIZE - entries;
1431 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1434 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1436 entries = DIV_ROUND_UP(entries,
1437 i965_cursor_wm_info.cacheline_size);
1438 cursor_sr = i965_cursor_wm_info.fifo_size -
1439 (entries + i965_cursor_wm_info.guard_size);
1441 if (cursor_sr > i965_cursor_wm_info.max_wm)
1442 cursor_sr = i965_cursor_wm_info.max_wm;
1444 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1445 "cursor %d\n", srwm, cursor_sr);
1447 if (IS_CRESTLINE(dev))
1448 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1450 /* Turn off self refresh if both pipes are enabled */
1451 if (IS_CRESTLINE(dev))
1452 I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1456 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1459 /* 965 has limitations... */
1460 I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1461 (8 << 16) | (8 << 8) | (8 << 0));
1462 I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1463 /* update cursor SR watermark */
1464 I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1467 static void i9xx_update_wm(struct drm_device *dev)
1469 struct drm_i915_private *dev_priv = dev->dev_private;
1470 const struct intel_watermark_params *wm_info;
1475 int planea_wm, planeb_wm;
1476 struct drm_crtc *crtc, *enabled = NULL;
1479 wm_info = &i945_wm_info;
1480 else if (!IS_GEN2(dev))
1481 wm_info = &i915_wm_info;
1483 wm_info = &i855_wm_info;
1485 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1486 crtc = intel_get_crtc_for_plane(dev, 0);
1487 if (intel_crtc_active(crtc)) {
1488 int cpp = crtc->fb->bits_per_pixel / 8;
1492 planea_wm = intel_calculate_wm(crtc->mode.clock,
1493 wm_info, fifo_size, cpp,
1497 planea_wm = fifo_size - wm_info->guard_size;
1499 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1500 crtc = intel_get_crtc_for_plane(dev, 1);
1501 if (intel_crtc_active(crtc)) {
1502 int cpp = crtc->fb->bits_per_pixel / 8;
1506 planeb_wm = intel_calculate_wm(crtc->mode.clock,
1507 wm_info, fifo_size, cpp,
1509 if (enabled == NULL)
1514 planeb_wm = fifo_size - wm_info->guard_size;
1516 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1519 * Overlay gets an aggressive default since video jitter is bad.
1523 /* Play safe and disable self-refresh before adjusting watermarks. */
1524 if (IS_I945G(dev) || IS_I945GM(dev))
1525 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1526 else if (IS_I915GM(dev))
1527 I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN);
1529 /* Calc sr entries for one plane configs */
1530 if (HAS_FW_BLC(dev) && enabled) {
1531 /* self-refresh has much higher latency */
1532 static const int sr_latency_ns = 6000;
1533 int clock = enabled->mode.clock;
1534 int htotal = enabled->mode.htotal;
1535 int hdisplay = enabled->mode.hdisplay;
1536 int pixel_size = enabled->fb->bits_per_pixel / 8;
1537 unsigned long line_time_us;
1540 line_time_us = (htotal * 1000) / clock;
1542 /* Use ns/us then divide to preserve precision */
1543 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1544 pixel_size * hdisplay;
1545 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1546 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1547 srwm = wm_info->fifo_size - entries;
1551 if (IS_I945G(dev) || IS_I945GM(dev))
1552 I915_WRITE(FW_BLC_SELF,
1553 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1554 else if (IS_I915GM(dev))
1555 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1558 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1559 planea_wm, planeb_wm, cwm, srwm);
1561 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1562 fwater_hi = (cwm & 0x1f);
1564 /* Set request length to 8 cachelines per fetch */
1565 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1566 fwater_hi = fwater_hi | (1 << 8);
1568 I915_WRITE(FW_BLC, fwater_lo);
1569 I915_WRITE(FW_BLC2, fwater_hi);
1571 if (HAS_FW_BLC(dev)) {
1573 if (IS_I945G(dev) || IS_I945GM(dev))
1574 I915_WRITE(FW_BLC_SELF,
1575 FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1576 else if (IS_I915GM(dev))
1577 I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN);
1578 DRM_DEBUG_KMS("memory self refresh enabled\n");
1580 DRM_DEBUG_KMS("memory self refresh disabled\n");
1584 static void i830_update_wm(struct drm_device *dev)
1586 struct drm_i915_private *dev_priv = dev->dev_private;
1587 struct drm_crtc *crtc;
1591 crtc = single_enabled_crtc(dev);
1595 planea_wm = intel_calculate_wm(crtc->mode.clock, &i830_wm_info,
1596 dev_priv->display.get_fifo_size(dev, 0),
1598 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1599 fwater_lo |= (3<<8) | planea_wm;
1601 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1603 I915_WRITE(FW_BLC, fwater_lo);
1606 #define ILK_LP0_PLANE_LATENCY 700
1607 #define ILK_LP0_CURSOR_LATENCY 1300
1610 * Check the wm result.
1612 * If any calculated watermark values is larger than the maximum value that
1613 * can be programmed into the associated watermark register, that watermark
1616 static bool ironlake_check_srwm(struct drm_device *dev, int level,
1617 int fbc_wm, int display_wm, int cursor_wm,
1618 const struct intel_watermark_params *display,
1619 const struct intel_watermark_params *cursor)
1621 struct drm_i915_private *dev_priv = dev->dev_private;
1623 DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d,"
1624 " cursor %d\n", level, display_wm, fbc_wm, cursor_wm);
1626 if (fbc_wm > SNB_FBC_MAX_SRWM) {
1627 DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n",
1628 fbc_wm, SNB_FBC_MAX_SRWM, level);
1630 /* fbc has it's own way to disable FBC WM */
1631 I915_WRITE(DISP_ARB_CTL,
1632 I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS);
1636 if (display_wm > display->max_wm) {
1637 DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n",
1638 display_wm, SNB_DISPLAY_MAX_SRWM, level);
1642 if (cursor_wm > cursor->max_wm) {
1643 DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n",
1644 cursor_wm, SNB_CURSOR_MAX_SRWM, level);
1648 if (!(fbc_wm || display_wm || cursor_wm)) {
1649 DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level);
1657 * Compute watermark values of WM[1-3],
1659 static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane,
1661 const struct intel_watermark_params *display,
1662 const struct intel_watermark_params *cursor,
1663 int *fbc_wm, int *display_wm, int *cursor_wm)
1665 struct drm_crtc *crtc;
1666 unsigned long line_time_us;
1667 int hdisplay, htotal, pixel_size, clock;
1668 int line_count, line_size;
1673 *fbc_wm = *display_wm = *cursor_wm = 0;
1677 crtc = intel_get_crtc_for_plane(dev, plane);
1678 hdisplay = crtc->mode.hdisplay;
1679 htotal = crtc->mode.htotal;
1680 clock = crtc->mode.clock;
1681 pixel_size = crtc->fb->bits_per_pixel / 8;
1683 line_time_us = (htotal * 1000) / clock;
1684 line_count = (latency_ns / line_time_us + 1000) / 1000;
1685 line_size = hdisplay * pixel_size;
1687 /* Use the minimum of the small and large buffer method for primary */
1688 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1689 large = line_count * line_size;
1691 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1692 *display_wm = entries + display->guard_size;
1696 * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2
1698 *fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2;
1700 /* calculate the self-refresh watermark for display cursor */
1701 entries = line_count * pixel_size * 64;
1702 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1703 *cursor_wm = entries + cursor->guard_size;
1705 return ironlake_check_srwm(dev, level,
1706 *fbc_wm, *display_wm, *cursor_wm,
1710 static void ironlake_update_wm(struct drm_device *dev)
1712 struct drm_i915_private *dev_priv = dev->dev_private;
1713 int fbc_wm, plane_wm, cursor_wm;
1714 unsigned int enabled;
1717 if (g4x_compute_wm0(dev, 0,
1718 &ironlake_display_wm_info,
1719 ILK_LP0_PLANE_LATENCY,
1720 &ironlake_cursor_wm_info,
1721 ILK_LP0_CURSOR_LATENCY,
1722 &plane_wm, &cursor_wm)) {
1723 I915_WRITE(WM0_PIPEA_ILK,
1724 (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1725 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1726 " plane %d, " "cursor: %d\n",
1727 plane_wm, cursor_wm);
1731 if (g4x_compute_wm0(dev, 1,
1732 &ironlake_display_wm_info,
1733 ILK_LP0_PLANE_LATENCY,
1734 &ironlake_cursor_wm_info,
1735 ILK_LP0_CURSOR_LATENCY,
1736 &plane_wm, &cursor_wm)) {
1737 I915_WRITE(WM0_PIPEB_ILK,
1738 (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1739 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1740 " plane %d, cursor: %d\n",
1741 plane_wm, cursor_wm);
1746 * Calculate and update the self-refresh watermark only when one
1747 * display plane is used.
1749 I915_WRITE(WM3_LP_ILK, 0);
1750 I915_WRITE(WM2_LP_ILK, 0);
1751 I915_WRITE(WM1_LP_ILK, 0);
1753 if (!single_plane_enabled(enabled))
1755 enabled = ffs(enabled) - 1;
1758 if (!ironlake_compute_srwm(dev, 1, enabled,
1759 ILK_READ_WM1_LATENCY() * 500,
1760 &ironlake_display_srwm_info,
1761 &ironlake_cursor_srwm_info,
1762 &fbc_wm, &plane_wm, &cursor_wm))
1765 I915_WRITE(WM1_LP_ILK,
1767 (ILK_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1768 (fbc_wm << WM1_LP_FBC_SHIFT) |
1769 (plane_wm << WM1_LP_SR_SHIFT) |
1773 if (!ironlake_compute_srwm(dev, 2, enabled,
1774 ILK_READ_WM2_LATENCY() * 500,
1775 &ironlake_display_srwm_info,
1776 &ironlake_cursor_srwm_info,
1777 &fbc_wm, &plane_wm, &cursor_wm))
1780 I915_WRITE(WM2_LP_ILK,
1782 (ILK_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1783 (fbc_wm << WM1_LP_FBC_SHIFT) |
1784 (plane_wm << WM1_LP_SR_SHIFT) |
1788 * WM3 is unsupported on ILK, probably because we don't have latency
1789 * data for that power state
1793 static void sandybridge_update_wm(struct drm_device *dev)
1795 struct drm_i915_private *dev_priv = dev->dev_private;
1796 int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
1798 int fbc_wm, plane_wm, cursor_wm;
1799 unsigned int enabled;
1802 if (g4x_compute_wm0(dev, 0,
1803 &sandybridge_display_wm_info, latency,
1804 &sandybridge_cursor_wm_info, latency,
1805 &plane_wm, &cursor_wm)) {
1806 val = I915_READ(WM0_PIPEA_ILK);
1807 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1808 I915_WRITE(WM0_PIPEA_ILK, val |
1809 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1810 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1811 " plane %d, " "cursor: %d\n",
1812 plane_wm, cursor_wm);
1816 if (g4x_compute_wm0(dev, 1,
1817 &sandybridge_display_wm_info, latency,
1818 &sandybridge_cursor_wm_info, latency,
1819 &plane_wm, &cursor_wm)) {
1820 val = I915_READ(WM0_PIPEB_ILK);
1821 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1822 I915_WRITE(WM0_PIPEB_ILK, val |
1823 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1824 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1825 " plane %d, cursor: %d\n",
1826 plane_wm, cursor_wm);
1831 * Calculate and update the self-refresh watermark only when one
1832 * display plane is used.
1834 * SNB support 3 levels of watermark.
1836 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1837 * and disabled in the descending order
1840 I915_WRITE(WM3_LP_ILK, 0);
1841 I915_WRITE(WM2_LP_ILK, 0);
1842 I915_WRITE(WM1_LP_ILK, 0);
1844 if (!single_plane_enabled(enabled) ||
1845 dev_priv->sprite_scaling_enabled)
1847 enabled = ffs(enabled) - 1;
1850 if (!ironlake_compute_srwm(dev, 1, enabled,
1851 SNB_READ_WM1_LATENCY() * 500,
1852 &sandybridge_display_srwm_info,
1853 &sandybridge_cursor_srwm_info,
1854 &fbc_wm, &plane_wm, &cursor_wm))
1857 I915_WRITE(WM1_LP_ILK,
1859 (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1860 (fbc_wm << WM1_LP_FBC_SHIFT) |
1861 (plane_wm << WM1_LP_SR_SHIFT) |
1865 if (!ironlake_compute_srwm(dev, 2, enabled,
1866 SNB_READ_WM2_LATENCY() * 500,
1867 &sandybridge_display_srwm_info,
1868 &sandybridge_cursor_srwm_info,
1869 &fbc_wm, &plane_wm, &cursor_wm))
1872 I915_WRITE(WM2_LP_ILK,
1874 (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1875 (fbc_wm << WM1_LP_FBC_SHIFT) |
1876 (plane_wm << WM1_LP_SR_SHIFT) |
1880 if (!ironlake_compute_srwm(dev, 3, enabled,
1881 SNB_READ_WM3_LATENCY() * 500,
1882 &sandybridge_display_srwm_info,
1883 &sandybridge_cursor_srwm_info,
1884 &fbc_wm, &plane_wm, &cursor_wm))
1887 I915_WRITE(WM3_LP_ILK,
1889 (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1890 (fbc_wm << WM1_LP_FBC_SHIFT) |
1891 (plane_wm << WM1_LP_SR_SHIFT) |
1895 static void ivybridge_update_wm(struct drm_device *dev)
1897 struct drm_i915_private *dev_priv = dev->dev_private;
1898 int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
1900 int fbc_wm, plane_wm, cursor_wm;
1901 int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm;
1902 unsigned int enabled;
1905 if (g4x_compute_wm0(dev, 0,
1906 &sandybridge_display_wm_info, latency,
1907 &sandybridge_cursor_wm_info, latency,
1908 &plane_wm, &cursor_wm)) {
1909 val = I915_READ(WM0_PIPEA_ILK);
1910 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1911 I915_WRITE(WM0_PIPEA_ILK, val |
1912 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1913 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1914 " plane %d, " "cursor: %d\n",
1915 plane_wm, cursor_wm);
1919 if (g4x_compute_wm0(dev, 1,
1920 &sandybridge_display_wm_info, latency,
1921 &sandybridge_cursor_wm_info, latency,
1922 &plane_wm, &cursor_wm)) {
1923 val = I915_READ(WM0_PIPEB_ILK);
1924 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1925 I915_WRITE(WM0_PIPEB_ILK, val |
1926 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1927 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1928 " plane %d, cursor: %d\n",
1929 plane_wm, cursor_wm);
1933 if (g4x_compute_wm0(dev, 2,
1934 &sandybridge_display_wm_info, latency,
1935 &sandybridge_cursor_wm_info, latency,
1936 &plane_wm, &cursor_wm)) {
1937 val = I915_READ(WM0_PIPEC_IVB);
1938 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1939 I915_WRITE(WM0_PIPEC_IVB, val |
1940 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1941 DRM_DEBUG_KMS("FIFO watermarks For pipe C -"
1942 " plane %d, cursor: %d\n",
1943 plane_wm, cursor_wm);
1948 * Calculate and update the self-refresh watermark only when one
1949 * display plane is used.
1951 * SNB support 3 levels of watermark.
1953 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1954 * and disabled in the descending order
1957 I915_WRITE(WM3_LP_ILK, 0);
1958 I915_WRITE(WM2_LP_ILK, 0);
1959 I915_WRITE(WM1_LP_ILK, 0);
1961 if (!single_plane_enabled(enabled) ||
1962 dev_priv->sprite_scaling_enabled)
1964 enabled = ffs(enabled) - 1;
1967 if (!ironlake_compute_srwm(dev, 1, enabled,
1968 SNB_READ_WM1_LATENCY() * 500,
1969 &sandybridge_display_srwm_info,
1970 &sandybridge_cursor_srwm_info,
1971 &fbc_wm, &plane_wm, &cursor_wm))
1974 I915_WRITE(WM1_LP_ILK,
1976 (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1977 (fbc_wm << WM1_LP_FBC_SHIFT) |
1978 (plane_wm << WM1_LP_SR_SHIFT) |
1982 if (!ironlake_compute_srwm(dev, 2, enabled,
1983 SNB_READ_WM2_LATENCY() * 500,
1984 &sandybridge_display_srwm_info,
1985 &sandybridge_cursor_srwm_info,
1986 &fbc_wm, &plane_wm, &cursor_wm))
1989 I915_WRITE(WM2_LP_ILK,
1991 (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1992 (fbc_wm << WM1_LP_FBC_SHIFT) |
1993 (plane_wm << WM1_LP_SR_SHIFT) |
1996 /* WM3, note we have to correct the cursor latency */
1997 if (!ironlake_compute_srwm(dev, 3, enabled,
1998 SNB_READ_WM3_LATENCY() * 500,
1999 &sandybridge_display_srwm_info,
2000 &sandybridge_cursor_srwm_info,
2001 &fbc_wm, &plane_wm, &ignore_cursor_wm) ||
2002 !ironlake_compute_srwm(dev, 3, enabled,
2003 2 * SNB_READ_WM3_LATENCY() * 500,
2004 &sandybridge_display_srwm_info,
2005 &sandybridge_cursor_srwm_info,
2006 &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm))
2009 I915_WRITE(WM3_LP_ILK,
2011 (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) |
2012 (fbc_wm << WM1_LP_FBC_SHIFT) |
2013 (plane_wm << WM1_LP_SR_SHIFT) |
2018 haswell_update_linetime_wm(struct drm_device *dev, int pipe,
2019 struct drm_display_mode *mode)
2021 struct drm_i915_private *dev_priv = dev->dev_private;
2024 temp = I915_READ(PIPE_WM_LINETIME(pipe));
2025 temp &= ~PIPE_WM_LINETIME_MASK;
2027 /* The WM are computed with base on how long it takes to fill a single
2028 * row at the given clock rate, multiplied by 8.
2030 temp |= PIPE_WM_LINETIME_TIME(
2031 ((mode->crtc_hdisplay * 1000) / mode->clock) * 8);
2033 /* IPS watermarks are only used by pipe A, and are ignored by
2034 * pipes B and C. They are calculated similarly to the common
2035 * linetime values, except that we are using CD clock frequency
2036 * in MHz instead of pixel rate for the division.
2038 * This is a placeholder for the IPS watermark calculation code.
2041 I915_WRITE(PIPE_WM_LINETIME(pipe), temp);
2045 sandybridge_compute_sprite_wm(struct drm_device *dev, int plane,
2046 uint32_t sprite_width, int pixel_size,
2047 const struct intel_watermark_params *display,
2048 int display_latency_ns, int *sprite_wm)
2050 struct drm_crtc *crtc;
2052 int entries, tlb_miss;
2054 crtc = intel_get_crtc_for_plane(dev, plane);
2055 if (!intel_crtc_active(crtc)) {
2056 *sprite_wm = display->guard_size;
2060 clock = crtc->mode.clock;
2062 /* Use the small buffer method to calculate the sprite watermark */
2063 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
2064 tlb_miss = display->fifo_size*display->cacheline_size -
2067 entries += tlb_miss;
2068 entries = DIV_ROUND_UP(entries, display->cacheline_size);
2069 *sprite_wm = entries + display->guard_size;
2070 if (*sprite_wm > (int)display->max_wm)
2071 *sprite_wm = display->max_wm;
2077 sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane,
2078 uint32_t sprite_width, int pixel_size,
2079 const struct intel_watermark_params *display,
2080 int latency_ns, int *sprite_wm)
2082 struct drm_crtc *crtc;
2083 unsigned long line_time_us;
2085 int line_count, line_size;
2094 crtc = intel_get_crtc_for_plane(dev, plane);
2095 clock = crtc->mode.clock;
2101 line_time_us = (sprite_width * 1000) / clock;
2102 if (!line_time_us) {
2107 line_count = (latency_ns / line_time_us + 1000) / 1000;
2108 line_size = sprite_width * pixel_size;
2110 /* Use the minimum of the small and large buffer method for primary */
2111 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
2112 large = line_count * line_size;
2114 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
2115 *sprite_wm = entries + display->guard_size;
2117 return *sprite_wm > 0x3ff ? false : true;
2120 static void sandybridge_update_sprite_wm(struct drm_device *dev, int pipe,
2121 uint32_t sprite_width, int pixel_size)
2123 struct drm_i915_private *dev_priv = dev->dev_private;
2124 int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
2131 reg = WM0_PIPEA_ILK;
2134 reg = WM0_PIPEB_ILK;
2137 reg = WM0_PIPEC_IVB;
2140 return; /* bad pipe */
2143 ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size,
2144 &sandybridge_display_wm_info,
2145 latency, &sprite_wm);
2147 DRM_DEBUG_KMS("failed to compute sprite wm for pipe %c\n",
2152 val = I915_READ(reg);
2153 val &= ~WM0_PIPE_SPRITE_MASK;
2154 I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT));
2155 DRM_DEBUG_KMS("sprite watermarks For pipe %c - %d\n", pipe_name(pipe), sprite_wm);
2158 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2160 &sandybridge_display_srwm_info,
2161 SNB_READ_WM1_LATENCY() * 500,
2164 DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %c\n",
2168 I915_WRITE(WM1S_LP_ILK, sprite_wm);
2170 /* Only IVB has two more LP watermarks for sprite */
2171 if (!IS_IVYBRIDGE(dev))
2174 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2176 &sandybridge_display_srwm_info,
2177 SNB_READ_WM2_LATENCY() * 500,
2180 DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %c\n",
2184 I915_WRITE(WM2S_LP_IVB, sprite_wm);
2186 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2188 &sandybridge_display_srwm_info,
2189 SNB_READ_WM3_LATENCY() * 500,
2192 DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %c\n",
2196 I915_WRITE(WM3S_LP_IVB, sprite_wm);
2200 * intel_update_watermarks - update FIFO watermark values based on current modes
2202 * Calculate watermark values for the various WM regs based on current mode
2203 * and plane configuration.
2205 * There are several cases to deal with here:
2206 * - normal (i.e. non-self-refresh)
2207 * - self-refresh (SR) mode
2208 * - lines are large relative to FIFO size (buffer can hold up to 2)
2209 * - lines are small relative to FIFO size (buffer can hold more than 2
2210 * lines), so need to account for TLB latency
2212 * The normal calculation is:
2213 * watermark = dotclock * bytes per pixel * latency
2214 * where latency is platform & configuration dependent (we assume pessimal
2217 * The SR calculation is:
2218 * watermark = (trunc(latency/line time)+1) * surface width *
2221 * line time = htotal / dotclock
2222 * surface width = hdisplay for normal plane and 64 for cursor
2223 * and latency is assumed to be high, as above.
2225 * The final value programmed to the register should always be rounded up,
2226 * and include an extra 2 entries to account for clock crossings.
2228 * We don't use the sprite, so we can ignore that. And on Crestline we have
2229 * to set the non-SR watermarks to 8.
2231 void intel_update_watermarks(struct drm_device *dev)
2233 struct drm_i915_private *dev_priv = dev->dev_private;
2235 if (dev_priv->display.update_wm)
2236 dev_priv->display.update_wm(dev);
2239 void intel_update_linetime_watermarks(struct drm_device *dev,
2240 int pipe, struct drm_display_mode *mode)
2242 struct drm_i915_private *dev_priv = dev->dev_private;
2244 if (dev_priv->display.update_linetime_wm)
2245 dev_priv->display.update_linetime_wm(dev, pipe, mode);
2248 void intel_update_sprite_watermarks(struct drm_device *dev, int pipe,
2249 uint32_t sprite_width, int pixel_size)
2251 struct drm_i915_private *dev_priv = dev->dev_private;
2253 if (dev_priv->display.update_sprite_wm)
2254 dev_priv->display.update_sprite_wm(dev, pipe, sprite_width,
2258 static struct drm_i915_gem_object *
2259 intel_alloc_context_page(struct drm_device *dev)
2261 struct drm_i915_gem_object *ctx;
2264 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2266 ctx = i915_gem_alloc_object(dev, 4096);
2268 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2272 ret = i915_gem_object_pin(ctx, 4096, true, false);
2274 DRM_ERROR("failed to pin power context: %d\n", ret);
2278 ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
2280 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
2287 i915_gem_object_unpin(ctx);
2289 drm_gem_object_unreference(&ctx->base);
2294 * Lock protecting IPS related data structures
2296 DEFINE_SPINLOCK(mchdev_lock);
2298 /* Global for IPS driver to get at the current i915 device. Protected by
2300 static struct drm_i915_private *i915_mch_dev;
2302 bool ironlake_set_drps(struct drm_device *dev, u8 val)
2304 struct drm_i915_private *dev_priv = dev->dev_private;
2307 assert_spin_locked(&mchdev_lock);
2309 rgvswctl = I915_READ16(MEMSWCTL);
2310 if (rgvswctl & MEMCTL_CMD_STS) {
2311 DRM_DEBUG("gpu busy, RCS change rejected\n");
2312 return false; /* still busy with another command */
2315 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
2316 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
2317 I915_WRITE16(MEMSWCTL, rgvswctl);
2318 POSTING_READ16(MEMSWCTL);
2320 rgvswctl |= MEMCTL_CMD_STS;
2321 I915_WRITE16(MEMSWCTL, rgvswctl);
2326 static void ironlake_enable_drps(struct drm_device *dev)
2328 struct drm_i915_private *dev_priv = dev->dev_private;
2329 u32 rgvmodectl = I915_READ(MEMMODECTL);
2330 u8 fmax, fmin, fstart, vstart;
2332 spin_lock_irq(&mchdev_lock);
2334 /* Enable temp reporting */
2335 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
2336 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
2338 /* 100ms RC evaluation intervals */
2339 I915_WRITE(RCUPEI, 100000);
2340 I915_WRITE(RCDNEI, 100000);
2342 /* Set max/min thresholds to 90ms and 80ms respectively */
2343 I915_WRITE(RCBMAXAVG, 90000);
2344 I915_WRITE(RCBMINAVG, 80000);
2346 I915_WRITE(MEMIHYST, 1);
2348 /* Set up min, max, and cur for interrupt handling */
2349 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
2350 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
2351 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
2352 MEMMODE_FSTART_SHIFT;
2354 vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
2357 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
2358 dev_priv->ips.fstart = fstart;
2360 dev_priv->ips.max_delay = fstart;
2361 dev_priv->ips.min_delay = fmin;
2362 dev_priv->ips.cur_delay = fstart;
2364 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2365 fmax, fmin, fstart);
2367 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
2370 * Interrupts will be enabled in ironlake_irq_postinstall
2373 I915_WRITE(VIDSTART, vstart);
2374 POSTING_READ(VIDSTART);
2376 rgvmodectl |= MEMMODE_SWMODE_EN;
2377 I915_WRITE(MEMMODECTL, rgvmodectl);
2379 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2380 DRM_ERROR("stuck trying to change perf mode\n");
2383 ironlake_set_drps(dev, fstart);
2385 dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
2387 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
2388 dev_priv->ips.last_count2 = I915_READ(0x112f4);
2389 getrawmonotonic(&dev_priv->ips.last_time2);
2391 spin_unlock_irq(&mchdev_lock);
2394 static void ironlake_disable_drps(struct drm_device *dev)
2396 struct drm_i915_private *dev_priv = dev->dev_private;
2399 spin_lock_irq(&mchdev_lock);
2401 rgvswctl = I915_READ16(MEMSWCTL);
2403 /* Ack interrupts, disable EFC interrupt */
2404 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
2405 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
2406 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
2407 I915_WRITE(DEIIR, DE_PCU_EVENT);
2408 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
2410 /* Go back to the starting frequency */
2411 ironlake_set_drps(dev, dev_priv->ips.fstart);
2413 rgvswctl |= MEMCTL_CMD_STS;
2414 I915_WRITE(MEMSWCTL, rgvswctl);
2417 spin_unlock_irq(&mchdev_lock);
2420 /* There's a funny hw issue where the hw returns all 0 when reading from
2421 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
2422 * ourselves, instead of doing a rmw cycle (which might result in us clearing
2423 * all limits and the gpu stuck at whatever frequency it is at atm).
2425 static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 *val)
2431 if (*val >= dev_priv->rps.max_delay)
2432 *val = dev_priv->rps.max_delay;
2433 limits |= dev_priv->rps.max_delay << 24;
2435 /* Only set the down limit when we've reached the lowest level to avoid
2436 * getting more interrupts, otherwise leave this clear. This prevents a
2437 * race in the hw when coming out of rc6: There's a tiny window where
2438 * the hw runs at the minimal clock before selecting the desired
2439 * frequency, if the down threshold expires in that window we will not
2440 * receive a down interrupt. */
2441 if (*val <= dev_priv->rps.min_delay) {
2442 *val = dev_priv->rps.min_delay;
2443 limits |= dev_priv->rps.min_delay << 16;
2449 void gen6_set_rps(struct drm_device *dev, u8 val)
2451 struct drm_i915_private *dev_priv = dev->dev_private;
2452 u32 limits = gen6_rps_limits(dev_priv, &val);
2454 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2455 WARN_ON(val > dev_priv->rps.max_delay);
2456 WARN_ON(val < dev_priv->rps.min_delay);
2458 if (val == dev_priv->rps.cur_delay)
2461 if (IS_HASWELL(dev))
2462 I915_WRITE(GEN6_RPNSWREQ,
2463 HSW_FREQUENCY(val));
2465 I915_WRITE(GEN6_RPNSWREQ,
2466 GEN6_FREQUENCY(val) |
2468 GEN6_AGGRESSIVE_TURBO);
2470 /* Make sure we continue to get interrupts
2471 * until we hit the minimum or maximum frequencies.
2473 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
2475 POSTING_READ(GEN6_RPNSWREQ);
2477 dev_priv->rps.cur_delay = val;
2479 trace_intel_gpu_freq_change(val * 50);
2482 void valleyview_set_rps(struct drm_device *dev, u8 val)
2484 struct drm_i915_private *dev_priv = dev->dev_private;
2485 unsigned long timeout = jiffies + msecs_to_jiffies(10);
2486 u32 limits = gen6_rps_limits(dev_priv, &val);
2489 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2490 WARN_ON(val > dev_priv->rps.max_delay);
2491 WARN_ON(val < dev_priv->rps.min_delay);
2493 DRM_DEBUG_DRIVER("gpu freq request from %d to %d\n",
2494 vlv_gpu_freq(dev_priv->mem_freq,
2495 dev_priv->rps.cur_delay),
2496 vlv_gpu_freq(dev_priv->mem_freq, val));
2498 if (val == dev_priv->rps.cur_delay)
2501 valleyview_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
2504 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS, &pval);
2505 if (time_after(jiffies, timeout)) {
2506 DRM_DEBUG_DRIVER("timed out waiting for Punit\n");
2512 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS, &pval);
2513 if ((pval >> 8) != val)
2514 DRM_DEBUG_DRIVER("punit overrode freq: %d requested, but got %d\n",
2517 /* Make sure we continue to get interrupts
2518 * until we hit the minimum or maximum frequencies.
2520 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
2522 dev_priv->rps.cur_delay = pval >> 8;
2524 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv->mem_freq, val));
2528 static void gen6_disable_rps(struct drm_device *dev)
2530 struct drm_i915_private *dev_priv = dev->dev_private;
2532 I915_WRITE(GEN6_RC_CONTROL, 0);
2533 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
2534 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
2535 I915_WRITE(GEN6_PMIER, 0);
2536 /* Complete PM interrupt masking here doesn't race with the rps work
2537 * item again unmasking PM interrupts because that is using a different
2538 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
2539 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
2541 spin_lock_irq(&dev_priv->rps.lock);
2542 dev_priv->rps.pm_iir = 0;
2543 spin_unlock_irq(&dev_priv->rps.lock);
2545 I915_WRITE(GEN6_PMIIR, I915_READ(GEN6_PMIIR));
2548 static void valleyview_disable_rps(struct drm_device *dev)
2550 struct drm_i915_private *dev_priv = dev->dev_private;
2552 I915_WRITE(GEN6_RC_CONTROL, 0);
2553 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
2554 I915_WRITE(GEN6_PMIER, 0);
2555 /* Complete PM interrupt masking here doesn't race with the rps work
2556 * item again unmasking PM interrupts because that is using a different
2557 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
2558 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
2560 spin_lock_irq(&dev_priv->rps.lock);
2561 dev_priv->rps.pm_iir = 0;
2562 spin_unlock_irq(&dev_priv->rps.lock);
2564 I915_WRITE(GEN6_PMIIR, I915_READ(GEN6_PMIIR));
2567 int intel_enable_rc6(const struct drm_device *dev)
2569 /* Respect the kernel parameter if it is set */
2570 if (i915_enable_rc6 >= 0)
2571 return i915_enable_rc6;
2573 /* Disable RC6 on Ironlake */
2574 if (INTEL_INFO(dev)->gen == 5)
2577 if (IS_HASWELL(dev)) {
2578 DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
2579 return INTEL_RC6_ENABLE;
2582 /* snb/ivb have more than one rc6 state. */
2583 if (INTEL_INFO(dev)->gen == 6) {
2584 DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
2585 return INTEL_RC6_ENABLE;
2588 DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n");
2589 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
2592 static void gen6_enable_rps(struct drm_device *dev)
2594 struct drm_i915_private *dev_priv = dev->dev_private;
2595 struct intel_ring_buffer *ring;
2598 u32 rc6vids, pcu_mbox, rc6_mask = 0;
2603 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2605 /* Here begins a magic sequence of register writes to enable
2606 * auto-downclocking.
2608 * Perhaps there might be some value in exposing these to
2611 I915_WRITE(GEN6_RC_STATE, 0);
2613 /* Clear the DBG now so we don't confuse earlier errors */
2614 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
2615 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
2616 I915_WRITE(GTFIFODBG, gtfifodbg);
2619 gen6_gt_force_wake_get(dev_priv);
2621 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
2622 gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
2624 /* In units of 50MHz */
2625 dev_priv->rps.hw_max = dev_priv->rps.max_delay = rp_state_cap & 0xff;
2626 dev_priv->rps.min_delay = (rp_state_cap & 0xff0000) >> 16;
2627 dev_priv->rps.cur_delay = 0;
2629 /* disable the counters and set deterministic thresholds */
2630 I915_WRITE(GEN6_RC_CONTROL, 0);
2632 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
2633 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
2634 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
2635 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
2636 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
2638 for_each_ring(ring, dev_priv, i)
2639 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2641 I915_WRITE(GEN6_RC_SLEEP, 0);
2642 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
2643 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
2644 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2645 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
2647 /* Check if we are enabling RC6 */
2648 rc6_mode = intel_enable_rc6(dev_priv->dev);
2649 if (rc6_mode & INTEL_RC6_ENABLE)
2650 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
2652 /* We don't use those on Haswell */
2653 if (!IS_HASWELL(dev)) {
2654 if (rc6_mode & INTEL_RC6p_ENABLE)
2655 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2657 if (rc6_mode & INTEL_RC6pp_ENABLE)
2658 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
2661 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
2662 (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
2663 (rc6_mask & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
2664 (rc6_mask & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
2666 I915_WRITE(GEN6_RC_CONTROL,
2668 GEN6_RC_CTL_EI_MODE(1) |
2669 GEN6_RC_CTL_HW_ENABLE);
2671 if (IS_HASWELL(dev)) {
2672 I915_WRITE(GEN6_RPNSWREQ,
2674 I915_WRITE(GEN6_RC_VIDEO_FREQ,
2677 I915_WRITE(GEN6_RPNSWREQ,
2678 GEN6_FREQUENCY(10) |
2680 GEN6_AGGRESSIVE_TURBO);
2681 I915_WRITE(GEN6_RC_VIDEO_FREQ,
2682 GEN6_FREQUENCY(12));
2685 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
2686 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
2687 dev_priv->rps.max_delay << 24 |
2688 dev_priv->rps.min_delay << 16);
2690 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
2691 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
2692 I915_WRITE(GEN6_RP_UP_EI, 66000);
2693 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
2695 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2696 I915_WRITE(GEN6_RP_CONTROL,
2697 GEN6_RP_MEDIA_TURBO |
2698 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2699 GEN6_RP_MEDIA_IS_GFX |
2701 GEN6_RP_UP_BUSY_AVG |
2702 (IS_HASWELL(dev) ? GEN7_RP_DOWN_IDLE_AVG : GEN6_RP_DOWN_IDLE_CONT));
2704 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
2705 if (!ret && (IS_GEN6(dev) || IS_IVYBRIDGE(dev))) {
2707 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
2708 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
2709 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
2710 (dev_priv->rps.max_delay & 0xff) * 50,
2711 (pcu_mbox & 0xff) * 50);
2712 dev_priv->rps.hw_max = pcu_mbox & 0xff;
2715 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
2718 gen6_set_rps(dev_priv->dev, (gt_perf_status & 0xff00) >> 8);
2720 /* requires MSI enabled */
2721 I915_WRITE(GEN6_PMIER, GEN6_PM_DEFERRED_EVENTS);
2722 spin_lock_irq(&dev_priv->rps.lock);
2723 WARN_ON(dev_priv->rps.pm_iir != 0);
2724 I915_WRITE(GEN6_PMIMR, 0);
2725 spin_unlock_irq(&dev_priv->rps.lock);
2726 /* enable all PM interrupts */
2727 I915_WRITE(GEN6_PMINTRMSK, 0);
2730 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
2731 if (IS_GEN6(dev) && ret) {
2732 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
2733 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
2734 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
2735 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
2736 rc6vids &= 0xffff00;
2737 rc6vids |= GEN6_ENCODE_RC6_VID(450);
2738 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
2740 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
2743 gen6_gt_force_wake_put(dev_priv);
2746 static void gen6_update_ring_freq(struct drm_device *dev)
2748 struct drm_i915_private *dev_priv = dev->dev_private;
2750 unsigned int gpu_freq;
2751 unsigned int max_ia_freq, min_ring_freq;
2752 int scaling_factor = 180;
2754 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2756 max_ia_freq = cpufreq_quick_get_max(0);
2758 * Default to measured freq if none found, PCU will ensure we don't go
2762 max_ia_freq = tsc_khz;
2764 /* Convert from kHz to MHz */
2765 max_ia_freq /= 1000;
2767 min_ring_freq = I915_READ(MCHBAR_MIRROR_BASE_SNB + DCLK);
2768 /* convert DDR frequency from units of 133.3MHz to bandwidth */
2769 min_ring_freq = (2 * 4 * min_ring_freq + 2) / 3;
2772 * For each potential GPU frequency, load a ring frequency we'd like
2773 * to use for memory access. We do this by specifying the IA frequency
2774 * the PCU should use as a reference to determine the ring frequency.
2776 for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay;
2778 int diff = dev_priv->rps.max_delay - gpu_freq;
2779 unsigned int ia_freq = 0, ring_freq = 0;
2781 if (IS_HASWELL(dev)) {
2782 ring_freq = (gpu_freq * 5 + 3) / 4;
2783 ring_freq = max(min_ring_freq, ring_freq);
2784 /* leave ia_freq as the default, chosen by cpufreq */
2786 /* On older processors, there is no separate ring
2787 * clock domain, so in order to boost the bandwidth
2788 * of the ring, we need to upclock the CPU (ia_freq).
2790 * For GPU frequencies less than 750MHz,
2791 * just use the lowest ring freq.
2793 if (gpu_freq < min_freq)
2796 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
2797 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
2800 sandybridge_pcode_write(dev_priv,
2801 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
2802 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
2803 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
2808 int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
2812 valleyview_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE, &val);
2814 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
2816 rp0 = min_t(u32, rp0, 0xea);
2821 static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
2825 valleyview_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO, &val);
2826 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
2827 valleyview_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI, &val);
2828 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
2833 int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
2837 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_LFM, &val);
2842 static void vlv_rps_timer_work(struct work_struct *work)
2844 drm_i915_private_t *dev_priv = container_of(work, drm_i915_private_t,
2848 * Timer fired, we must be idle. Drop to min voltage state.
2849 * Note: we use RPe here since it should match the
2850 * Vmin we were shooting for. That should give us better
2851 * perf when we come back out of RC6 than if we used the
2852 * min freq available.
2854 mutex_lock(&dev_priv->rps.hw_lock);
2855 valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay);
2856 mutex_unlock(&dev_priv->rps.hw_lock);
2859 static void valleyview_enable_rps(struct drm_device *dev)
2861 struct drm_i915_private *dev_priv = dev->dev_private;
2862 struct intel_ring_buffer *ring;
2863 u32 gtfifodbg, val, rpe;
2866 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2868 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
2869 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
2870 I915_WRITE(GTFIFODBG, gtfifodbg);
2873 gen6_gt_force_wake_get(dev_priv);
2875 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
2876 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
2877 I915_WRITE(GEN6_RP_UP_EI, 66000);
2878 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
2880 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2882 I915_WRITE(GEN6_RP_CONTROL,
2883 GEN6_RP_MEDIA_TURBO |
2884 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2885 GEN6_RP_MEDIA_IS_GFX |
2887 GEN6_RP_UP_BUSY_AVG |
2888 GEN6_RP_DOWN_IDLE_CONT);
2890 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
2891 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
2892 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
2894 for_each_ring(ring, dev_priv, i)
2895 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2897 I915_WRITE(GEN6_RC6_THRESHOLD, 0xc350);
2899 /* allows RC6 residency counter to work */
2900 I915_WRITE(0x138104, _MASKED_BIT_ENABLE(0x3));
2901 I915_WRITE(GEN6_RC_CONTROL,
2902 GEN7_RC_CTL_TO_MODE);
2904 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS, &val);
2905 switch ((val >> 6) & 3) {
2908 dev_priv->mem_freq = 800;
2911 dev_priv->mem_freq = 1066;
2914 dev_priv->mem_freq = 1333;
2917 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
2919 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
2920 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
2922 DRM_DEBUG_DRIVER("current GPU freq: %d\n",
2923 vlv_gpu_freq(dev_priv->mem_freq, (val >> 8) & 0xff));
2924 dev_priv->rps.cur_delay = (val >> 8) & 0xff;
2926 dev_priv->rps.max_delay = valleyview_rps_max_freq(dev_priv);
2927 dev_priv->rps.hw_max = dev_priv->rps.max_delay;
2928 DRM_DEBUG_DRIVER("max GPU freq: %d\n", vlv_gpu_freq(dev_priv->mem_freq,
2929 dev_priv->rps.max_delay));
2931 rpe = valleyview_rps_rpe_freq(dev_priv);
2932 DRM_DEBUG_DRIVER("RPe GPU freq: %d\n",
2933 vlv_gpu_freq(dev_priv->mem_freq, rpe));
2934 dev_priv->rps.rpe_delay = rpe;
2936 val = valleyview_rps_min_freq(dev_priv);
2937 DRM_DEBUG_DRIVER("min GPU freq: %d\n", vlv_gpu_freq(dev_priv->mem_freq,
2939 dev_priv->rps.min_delay = val;
2941 DRM_DEBUG_DRIVER("setting GPU freq to %d\n",
2942 vlv_gpu_freq(dev_priv->mem_freq, rpe));
2944 INIT_DELAYED_WORK(&dev_priv->rps.vlv_work, vlv_rps_timer_work);
2946 valleyview_set_rps(dev_priv->dev, rpe);
2948 /* requires MSI enabled */
2949 I915_WRITE(GEN6_PMIER, GEN6_PM_DEFERRED_EVENTS);
2950 spin_lock_irq(&dev_priv->rps.lock);
2951 WARN_ON(dev_priv->rps.pm_iir != 0);
2952 I915_WRITE(GEN6_PMIMR, 0);
2953 spin_unlock_irq(&dev_priv->rps.lock);
2954 /* enable all PM interrupts */
2955 I915_WRITE(GEN6_PMINTRMSK, 0);
2957 gen6_gt_force_wake_put(dev_priv);
2960 void ironlake_teardown_rc6(struct drm_device *dev)
2962 struct drm_i915_private *dev_priv = dev->dev_private;
2964 if (dev_priv->ips.renderctx) {
2965 i915_gem_object_unpin(dev_priv->ips.renderctx);
2966 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
2967 dev_priv->ips.renderctx = NULL;
2970 if (dev_priv->ips.pwrctx) {
2971 i915_gem_object_unpin(dev_priv->ips.pwrctx);
2972 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
2973 dev_priv->ips.pwrctx = NULL;
2977 static void ironlake_disable_rc6(struct drm_device *dev)
2979 struct drm_i915_private *dev_priv = dev->dev_private;
2981 if (I915_READ(PWRCTXA)) {
2982 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
2983 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
2984 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
2987 I915_WRITE(PWRCTXA, 0);
2988 POSTING_READ(PWRCTXA);
2990 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
2991 POSTING_READ(RSTDBYCTL);
2995 static int ironlake_setup_rc6(struct drm_device *dev)
2997 struct drm_i915_private *dev_priv = dev->dev_private;
2999 if (dev_priv->ips.renderctx == NULL)
3000 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
3001 if (!dev_priv->ips.renderctx)
3004 if (dev_priv->ips.pwrctx == NULL)
3005 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
3006 if (!dev_priv->ips.pwrctx) {
3007 ironlake_teardown_rc6(dev);
3014 static void ironlake_enable_rc6(struct drm_device *dev)
3016 struct drm_i915_private *dev_priv = dev->dev_private;
3017 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
3018 bool was_interruptible;
3021 /* rc6 disabled by default due to repeated reports of hanging during
3024 if (!intel_enable_rc6(dev))
3027 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3029 ret = ironlake_setup_rc6(dev);
3033 was_interruptible = dev_priv->mm.interruptible;
3034 dev_priv->mm.interruptible = false;
3037 * GPU can automatically power down the render unit if given a page
3040 ret = intel_ring_begin(ring, 6);
3042 ironlake_teardown_rc6(dev);
3043 dev_priv->mm.interruptible = was_interruptible;
3047 intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
3048 intel_ring_emit(ring, MI_SET_CONTEXT);
3049 intel_ring_emit(ring, dev_priv->ips.renderctx->gtt_offset |
3051 MI_SAVE_EXT_STATE_EN |
3052 MI_RESTORE_EXT_STATE_EN |
3053 MI_RESTORE_INHIBIT);
3054 intel_ring_emit(ring, MI_SUSPEND_FLUSH);
3055 intel_ring_emit(ring, MI_NOOP);
3056 intel_ring_emit(ring, MI_FLUSH);
3057 intel_ring_advance(ring);
3060 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
3061 * does an implicit flush, combined with MI_FLUSH above, it should be
3062 * safe to assume that renderctx is valid
3064 ret = intel_ring_idle(ring);
3065 dev_priv->mm.interruptible = was_interruptible;
3067 DRM_ERROR("failed to enable ironlake power savings\n");
3068 ironlake_teardown_rc6(dev);
3072 I915_WRITE(PWRCTXA, dev_priv->ips.pwrctx->gtt_offset | PWRCTX_EN);
3073 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
3076 static unsigned long intel_pxfreq(u32 vidfreq)
3079 int div = (vidfreq & 0x3f0000) >> 16;
3080 int post = (vidfreq & 0x3000) >> 12;
3081 int pre = (vidfreq & 0x7);
3086 freq = ((div * 133333) / ((1<<post) * pre));
3091 static const struct cparams {
3097 { 1, 1333, 301, 28664 },
3098 { 1, 1066, 294, 24460 },
3099 { 1, 800, 294, 25192 },
3100 { 0, 1333, 276, 27605 },
3101 { 0, 1066, 276, 27605 },
3102 { 0, 800, 231, 23784 },
3105 static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
3107 u64 total_count, diff, ret;
3108 u32 count1, count2, count3, m = 0, c = 0;
3109 unsigned long now = jiffies_to_msecs(jiffies), diff1;
3112 assert_spin_locked(&mchdev_lock);
3114 diff1 = now - dev_priv->ips.last_time1;
3116 /* Prevent division-by-zero if we are asking too fast.
3117 * Also, we don't get interesting results if we are polling
3118 * faster than once in 10ms, so just return the saved value
3122 return dev_priv->ips.chipset_power;
3124 count1 = I915_READ(DMIEC);
3125 count2 = I915_READ(DDREC);
3126 count3 = I915_READ(CSIEC);
3128 total_count = count1 + count2 + count3;
3130 /* FIXME: handle per-counter overflow */
3131 if (total_count < dev_priv->ips.last_count1) {
3132 diff = ~0UL - dev_priv->ips.last_count1;
3133 diff += total_count;
3135 diff = total_count - dev_priv->ips.last_count1;
3138 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
3139 if (cparams[i].i == dev_priv->ips.c_m &&
3140 cparams[i].t == dev_priv->ips.r_t) {
3147 diff = div_u64(diff, diff1);
3148 ret = ((m * diff) + c);
3149 ret = div_u64(ret, 10);
3151 dev_priv->ips.last_count1 = total_count;
3152 dev_priv->ips.last_time1 = now;
3154 dev_priv->ips.chipset_power = ret;
3159 unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
3163 if (dev_priv->info->gen != 5)
3166 spin_lock_irq(&mchdev_lock);
3168 val = __i915_chipset_val(dev_priv);
3170 spin_unlock_irq(&mchdev_lock);
3175 unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
3177 unsigned long m, x, b;
3180 tsfs = I915_READ(TSFS);
3182 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
3183 x = I915_READ8(TR1);
3185 b = tsfs & TSFS_INTR_MASK;
3187 return ((m * x) / 127) - b;
3190 static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
3192 static const struct v_table {
3193 u16 vd; /* in .1 mil */
3194 u16 vm; /* in .1 mil */
3325 if (dev_priv->info->is_mobile)
3326 return v_table[pxvid].vm;
3328 return v_table[pxvid].vd;
3331 static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
3333 struct timespec now, diff1;
3335 unsigned long diffms;
3338 assert_spin_locked(&mchdev_lock);
3340 getrawmonotonic(&now);
3341 diff1 = timespec_sub(now, dev_priv->ips.last_time2);
3343 /* Don't divide by 0 */
3344 diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
3348 count = I915_READ(GFXEC);
3350 if (count < dev_priv->ips.last_count2) {
3351 diff = ~0UL - dev_priv->ips.last_count2;
3354 diff = count - dev_priv->ips.last_count2;
3357 dev_priv->ips.last_count2 = count;
3358 dev_priv->ips.last_time2 = now;
3360 /* More magic constants... */
3362 diff = div_u64(diff, diffms * 10);
3363 dev_priv->ips.gfx_power = diff;
3366 void i915_update_gfx_val(struct drm_i915_private *dev_priv)
3368 if (dev_priv->info->gen != 5)
3371 spin_lock_irq(&mchdev_lock);
3373 __i915_update_gfx_val(dev_priv);
3375 spin_unlock_irq(&mchdev_lock);
3378 static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
3380 unsigned long t, corr, state1, corr2, state2;
3383 assert_spin_locked(&mchdev_lock);
3385 pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4));
3386 pxvid = (pxvid >> 24) & 0x7f;
3387 ext_v = pvid_to_extvid(dev_priv, pxvid);
3391 t = i915_mch_val(dev_priv);
3393 /* Revel in the empirically derived constants */
3395 /* Correction factor in 1/100000 units */
3397 corr = ((t * 2349) + 135940);
3399 corr = ((t * 964) + 29317);
3401 corr = ((t * 301) + 1004);
3403 corr = corr * ((150142 * state1) / 10000 - 78642);
3405 corr2 = (corr * dev_priv->ips.corr);
3407 state2 = (corr2 * state1) / 10000;
3408 state2 /= 100; /* convert to mW */
3410 __i915_update_gfx_val(dev_priv);
3412 return dev_priv->ips.gfx_power + state2;
3415 unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
3419 if (dev_priv->info->gen != 5)
3422 spin_lock_irq(&mchdev_lock);
3424 val = __i915_gfx_val(dev_priv);
3426 spin_unlock_irq(&mchdev_lock);
3432 * i915_read_mch_val - return value for IPS use
3434 * Calculate and return a value for the IPS driver to use when deciding whether
3435 * we have thermal and power headroom to increase CPU or GPU power budget.
3437 unsigned long i915_read_mch_val(void)
3439 struct drm_i915_private *dev_priv;
3440 unsigned long chipset_val, graphics_val, ret = 0;
3442 spin_lock_irq(&mchdev_lock);
3445 dev_priv = i915_mch_dev;
3447 chipset_val = __i915_chipset_val(dev_priv);
3448 graphics_val = __i915_gfx_val(dev_priv);
3450 ret = chipset_val + graphics_val;
3453 spin_unlock_irq(&mchdev_lock);
3457 EXPORT_SYMBOL_GPL(i915_read_mch_val);
3460 * i915_gpu_raise - raise GPU frequency limit
3462 * Raise the limit; IPS indicates we have thermal headroom.
3464 bool i915_gpu_raise(void)
3466 struct drm_i915_private *dev_priv;
3469 spin_lock_irq(&mchdev_lock);
3470 if (!i915_mch_dev) {
3474 dev_priv = i915_mch_dev;
3476 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
3477 dev_priv->ips.max_delay--;
3480 spin_unlock_irq(&mchdev_lock);
3484 EXPORT_SYMBOL_GPL(i915_gpu_raise);
3487 * i915_gpu_lower - lower GPU frequency limit
3489 * IPS indicates we're close to a thermal limit, so throttle back the GPU
3490 * frequency maximum.
3492 bool i915_gpu_lower(void)
3494 struct drm_i915_private *dev_priv;
3497 spin_lock_irq(&mchdev_lock);
3498 if (!i915_mch_dev) {
3502 dev_priv = i915_mch_dev;
3504 if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
3505 dev_priv->ips.max_delay++;
3508 spin_unlock_irq(&mchdev_lock);
3512 EXPORT_SYMBOL_GPL(i915_gpu_lower);
3515 * i915_gpu_busy - indicate GPU business to IPS
3517 * Tell the IPS driver whether or not the GPU is busy.
3519 bool i915_gpu_busy(void)
3521 struct drm_i915_private *dev_priv;
3522 struct intel_ring_buffer *ring;
3526 spin_lock_irq(&mchdev_lock);
3529 dev_priv = i915_mch_dev;
3531 for_each_ring(ring, dev_priv, i)
3532 ret |= !list_empty(&ring->request_list);
3535 spin_unlock_irq(&mchdev_lock);
3539 EXPORT_SYMBOL_GPL(i915_gpu_busy);
3542 * i915_gpu_turbo_disable - disable graphics turbo
3544 * Disable graphics turbo by resetting the max frequency and setting the
3545 * current frequency to the default.
3547 bool i915_gpu_turbo_disable(void)
3549 struct drm_i915_private *dev_priv;
3552 spin_lock_irq(&mchdev_lock);
3553 if (!i915_mch_dev) {
3557 dev_priv = i915_mch_dev;
3559 dev_priv->ips.max_delay = dev_priv->ips.fstart;
3561 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
3565 spin_unlock_irq(&mchdev_lock);
3569 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
3572 * Tells the intel_ips driver that the i915 driver is now loaded, if
3573 * IPS got loaded first.
3575 * This awkward dance is so that neither module has to depend on the
3576 * other in order for IPS to do the appropriate communication of
3577 * GPU turbo limits to i915.
3580 ips_ping_for_i915_load(void)
3584 link = symbol_get(ips_link_to_i915_driver);
3587 symbol_put(ips_link_to_i915_driver);
3591 void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
3593 /* We only register the i915 ips part with intel-ips once everything is
3594 * set up, to avoid intel-ips sneaking in and reading bogus values. */
3595 spin_lock_irq(&mchdev_lock);
3596 i915_mch_dev = dev_priv;
3597 spin_unlock_irq(&mchdev_lock);
3599 ips_ping_for_i915_load();
3602 void intel_gpu_ips_teardown(void)
3604 spin_lock_irq(&mchdev_lock);
3605 i915_mch_dev = NULL;
3606 spin_unlock_irq(&mchdev_lock);
3608 static void intel_init_emon(struct drm_device *dev)
3610 struct drm_i915_private *dev_priv = dev->dev_private;
3615 /* Disable to program */
3619 /* Program energy weights for various events */
3620 I915_WRITE(SDEW, 0x15040d00);
3621 I915_WRITE(CSIEW0, 0x007f0000);
3622 I915_WRITE(CSIEW1, 0x1e220004);
3623 I915_WRITE(CSIEW2, 0x04000004);
3625 for (i = 0; i < 5; i++)
3626 I915_WRITE(PEW + (i * 4), 0);
3627 for (i = 0; i < 3; i++)
3628 I915_WRITE(DEW + (i * 4), 0);
3630 /* Program P-state weights to account for frequency power adjustment */
3631 for (i = 0; i < 16; i++) {
3632 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
3633 unsigned long freq = intel_pxfreq(pxvidfreq);
3634 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
3639 val *= (freq / 1000);
3641 val /= (127*127*900);
3643 DRM_ERROR("bad pxval: %ld\n", val);
3646 /* Render standby states get 0 weight */
3650 for (i = 0; i < 4; i++) {
3651 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
3652 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
3653 I915_WRITE(PXW + (i * 4), val);
3656 /* Adjust magic regs to magic values (more experimental results) */
3657 I915_WRITE(OGW0, 0);
3658 I915_WRITE(OGW1, 0);
3659 I915_WRITE(EG0, 0x00007f00);
3660 I915_WRITE(EG1, 0x0000000e);
3661 I915_WRITE(EG2, 0x000e0000);
3662 I915_WRITE(EG3, 0x68000300);
3663 I915_WRITE(EG4, 0x42000000);
3664 I915_WRITE(EG5, 0x00140031);
3668 for (i = 0; i < 8; i++)
3669 I915_WRITE(PXWL + (i * 4), 0);
3671 /* Enable PMON + select events */
3672 I915_WRITE(ECR, 0x80000019);
3674 lcfuse = I915_READ(LCFUSE02);
3676 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
3679 void intel_disable_gt_powersave(struct drm_device *dev)
3681 struct drm_i915_private *dev_priv = dev->dev_private;
3683 /* Interrupts should be disabled already to avoid re-arming. */
3684 WARN_ON(dev->irq_enabled);
3686 if (IS_IRONLAKE_M(dev)) {
3687 ironlake_disable_drps(dev);
3688 ironlake_disable_rc6(dev);
3689 } else if (INTEL_INFO(dev)->gen >= 6) {
3690 cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
3691 cancel_work_sync(&dev_priv->rps.work);
3692 if (IS_VALLEYVIEW(dev))
3693 cancel_delayed_work_sync(&dev_priv->rps.vlv_work);
3694 mutex_lock(&dev_priv->rps.hw_lock);
3695 if (IS_VALLEYVIEW(dev))
3696 valleyview_disable_rps(dev);
3698 gen6_disable_rps(dev);
3699 mutex_unlock(&dev_priv->rps.hw_lock);
3703 static void intel_gen6_powersave_work(struct work_struct *work)
3705 struct drm_i915_private *dev_priv =
3706 container_of(work, struct drm_i915_private,
3707 rps.delayed_resume_work.work);
3708 struct drm_device *dev = dev_priv->dev;
3710 mutex_lock(&dev_priv->rps.hw_lock);
3712 if (IS_VALLEYVIEW(dev)) {
3713 valleyview_enable_rps(dev);
3715 gen6_enable_rps(dev);
3716 gen6_update_ring_freq(dev);
3718 mutex_unlock(&dev_priv->rps.hw_lock);
3721 void intel_enable_gt_powersave(struct drm_device *dev)
3723 struct drm_i915_private *dev_priv = dev->dev_private;
3725 if (IS_IRONLAKE_M(dev)) {
3726 ironlake_enable_drps(dev);
3727 ironlake_enable_rc6(dev);
3728 intel_init_emon(dev);
3729 } else if (IS_GEN6(dev) || IS_GEN7(dev)) {
3731 * PCU communication is slow and this doesn't need to be
3732 * done at any specific time, so do this out of our fast path
3733 * to make resume and init faster.
3735 schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
3736 round_jiffies_up_relative(HZ));
3740 static void ibx_init_clock_gating(struct drm_device *dev)
3742 struct drm_i915_private *dev_priv = dev->dev_private;
3745 * On Ibex Peak and Cougar Point, we need to disable clock
3746 * gating for the panel power sequencer or it will fail to
3747 * start up when no ports are active.
3749 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
3752 static void ironlake_init_clock_gating(struct drm_device *dev)
3754 struct drm_i915_private *dev_priv = dev->dev_private;
3755 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
3757 /* Required for FBC */
3758 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
3759 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
3760 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
3762 I915_WRITE(PCH_3DCGDIS0,
3763 MARIUNIT_CLOCK_GATE_DISABLE |
3764 SVSMUNIT_CLOCK_GATE_DISABLE);
3765 I915_WRITE(PCH_3DCGDIS1,
3766 VFMUNIT_CLOCK_GATE_DISABLE);
3769 * According to the spec the following bits should be set in
3770 * order to enable memory self-refresh
3771 * The bit 22/21 of 0x42004
3772 * The bit 5 of 0x42020
3773 * The bit 15 of 0x45000
3775 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3776 (I915_READ(ILK_DISPLAY_CHICKEN2) |
3777 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
3778 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
3779 I915_WRITE(DISP_ARB_CTL,
3780 (I915_READ(DISP_ARB_CTL) |
3782 I915_WRITE(WM3_LP_ILK, 0);
3783 I915_WRITE(WM2_LP_ILK, 0);
3784 I915_WRITE(WM1_LP_ILK, 0);
3787 * Based on the document from hardware guys the following bits
3788 * should be set unconditionally in order to enable FBC.
3789 * The bit 22 of 0x42000
3790 * The bit 22 of 0x42004
3791 * The bit 7,8,9 of 0x42020.
3793 if (IS_IRONLAKE_M(dev)) {
3794 I915_WRITE(ILK_DISPLAY_CHICKEN1,
3795 I915_READ(ILK_DISPLAY_CHICKEN1) |
3797 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3798 I915_READ(ILK_DISPLAY_CHICKEN2) |
3802 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
3804 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3805 I915_READ(ILK_DISPLAY_CHICKEN2) |
3806 ILK_ELPIN_409_SELECT);
3807 I915_WRITE(_3D_CHICKEN2,
3808 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
3809 _3D_CHICKEN2_WM_READ_PIPELINED);
3811 /* WaDisableRenderCachePipelinedFlush */
3812 I915_WRITE(CACHE_MODE_0,
3813 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3815 ibx_init_clock_gating(dev);
3818 static void cpt_init_clock_gating(struct drm_device *dev)
3820 struct drm_i915_private *dev_priv = dev->dev_private;
3825 * On Ibex Peak and Cougar Point, we need to disable clock
3826 * gating for the panel power sequencer or it will fail to
3827 * start up when no ports are active.
3829 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
3830 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
3831 DPLS_EDP_PPS_FIX_DIS);
3832 /* The below fixes the weird display corruption, a few pixels shifted
3833 * downward, on (only) LVDS of some HP laptops with IVY.
3835 for_each_pipe(pipe) {
3836 val = I915_READ(TRANS_CHICKEN2(pipe));
3837 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
3838 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
3839 if (dev_priv->fdi_rx_polarity_inverted)
3840 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
3841 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
3842 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
3843 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3844 I915_WRITE(TRANS_CHICKEN2(pipe), val);
3846 /* WADP0ClockGatingDisable */
3847 for_each_pipe(pipe) {
3848 I915_WRITE(TRANS_CHICKEN1(pipe),
3849 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
3853 static void gen6_check_mch_setup(struct drm_device *dev)
3855 struct drm_i915_private *dev_priv = dev->dev_private;
3858 tmp = I915_READ(MCH_SSKPD);
3859 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
3860 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
3861 DRM_INFO("This can cause pipe underruns and display issues.\n");
3862 DRM_INFO("Please upgrade your BIOS to fix this.\n");
3866 static void gen6_init_clock_gating(struct drm_device *dev)
3868 struct drm_i915_private *dev_priv = dev->dev_private;
3870 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
3872 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
3874 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3875 I915_READ(ILK_DISPLAY_CHICKEN2) |
3876 ILK_ELPIN_409_SELECT);
3878 /* WaDisableHiZPlanesWhenMSAAEnabled */
3879 I915_WRITE(_3D_CHICKEN,
3880 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
3882 /* WaSetupGtModeTdRowDispatch */
3883 if (IS_SNB_GT1(dev))
3884 I915_WRITE(GEN6_GT_MODE,
3885 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
3887 I915_WRITE(WM3_LP_ILK, 0);
3888 I915_WRITE(WM2_LP_ILK, 0);
3889 I915_WRITE(WM1_LP_ILK, 0);
3891 I915_WRITE(CACHE_MODE_0,
3892 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
3894 I915_WRITE(GEN6_UCGCTL1,
3895 I915_READ(GEN6_UCGCTL1) |
3896 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
3897 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
3899 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3900 * gating disable must be set. Failure to set it results in
3901 * flickering pixels due to Z write ordering failures after
3902 * some amount of runtime in the Mesa "fire" demo, and Unigine
3903 * Sanctuary and Tropics, and apparently anything else with
3904 * alpha test or pixel discard.
3906 * According to the spec, bit 11 (RCCUNIT) must also be set,
3907 * but we didn't debug actual testcases to find it out.
3909 * Also apply WaDisableVDSUnitClockGating and
3910 * WaDisableRCPBUnitClockGating.
3912 I915_WRITE(GEN6_UCGCTL2,
3913 GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
3914 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
3915 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
3917 /* Bspec says we need to always set all mask bits. */
3918 I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) |
3919 _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL);
3922 * According to the spec the following bits should be
3923 * set in order to enable memory self-refresh and fbc:
3924 * The bit21 and bit22 of 0x42000
3925 * The bit21 and bit22 of 0x42004
3926 * The bit5 and bit7 of 0x42020
3927 * The bit14 of 0x70180
3928 * The bit14 of 0x71180
3930 I915_WRITE(ILK_DISPLAY_CHICKEN1,
3931 I915_READ(ILK_DISPLAY_CHICKEN1) |
3932 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
3933 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3934 I915_READ(ILK_DISPLAY_CHICKEN2) |
3935 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
3936 I915_WRITE(ILK_DSPCLK_GATE_D,
3937 I915_READ(ILK_DSPCLK_GATE_D) |
3938 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
3939 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
3941 /* WaMbcDriverBootEnable */
3942 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
3943 GEN6_MBCTL_ENABLE_BOOT_FETCH);
3945 for_each_pipe(pipe) {
3946 I915_WRITE(DSPCNTR(pipe),
3947 I915_READ(DSPCNTR(pipe)) |
3948 DISPPLANE_TRICKLE_FEED_DISABLE);
3949 intel_flush_display_plane(dev_priv, pipe);
3952 /* The default value should be 0x200 according to docs, but the two
3953 * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */
3954 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff));
3955 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI));
3957 cpt_init_clock_gating(dev);
3959 gen6_check_mch_setup(dev);
3962 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
3964 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
3966 reg &= ~GEN7_FF_SCHED_MASK;
3967 reg |= GEN7_FF_TS_SCHED_HW;
3968 reg |= GEN7_FF_VS_SCHED_HW;
3969 reg |= GEN7_FF_DS_SCHED_HW;
3971 /* WaVSRefCountFullforceMissDisable */
3972 if (IS_HASWELL(dev_priv->dev))
3973 reg &= ~GEN7_FF_VS_REF_CNT_FFME;
3975 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
3978 static void lpt_init_clock_gating(struct drm_device *dev)
3980 struct drm_i915_private *dev_priv = dev->dev_private;
3983 * TODO: this bit should only be enabled when really needed, then
3984 * disabled when not needed anymore in order to save power.
3986 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
3987 I915_WRITE(SOUTH_DSPCLK_GATE_D,
3988 I915_READ(SOUTH_DSPCLK_GATE_D) |
3989 PCH_LP_PARTITION_LEVEL_DISABLE);
3992 static void haswell_init_clock_gating(struct drm_device *dev)
3994 struct drm_i915_private *dev_priv = dev->dev_private;
3997 I915_WRITE(WM3_LP_ILK, 0);
3998 I915_WRITE(WM2_LP_ILK, 0);
3999 I915_WRITE(WM1_LP_ILK, 0);
4001 /* According to the spec, bit 13 (RCZUNIT) must be set on IVB.
4002 * This implements the WaDisableRCZUnitClockGating workaround.
4004 I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
4006 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
4007 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4008 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4010 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
4011 I915_WRITE(GEN7_L3CNTLREG1,
4012 GEN7_WA_FOR_GEN7_L3_CONTROL);
4013 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
4014 GEN7_WA_L3_CHICKEN_MODE);
4016 /* This is required by WaCatErrorRejectionIssue */
4017 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4018 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4019 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4021 for_each_pipe(pipe) {
4022 I915_WRITE(DSPCNTR(pipe),
4023 I915_READ(DSPCNTR(pipe)) |
4024 DISPPLANE_TRICKLE_FEED_DISABLE);
4025 intel_flush_display_plane(dev_priv, pipe);
4028 gen7_setup_fixed_func_scheduler(dev_priv);
4030 /* WaDisable4x2SubspanOptimization */
4031 I915_WRITE(CACHE_MODE_1,
4032 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
4034 /* WaMbcDriverBootEnable */
4035 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
4036 GEN6_MBCTL_ENABLE_BOOT_FETCH);
4038 /* WaSwitchSolVfFArbitrationPriority */
4039 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
4041 /* XXX: This is a workaround for early silicon revisions and should be
4046 WM_DBG_DISALLOW_MULTIPLE_LP |
4047 WM_DBG_DISALLOW_SPRITE |
4048 WM_DBG_DISALLOW_MAXFIFO);
4050 lpt_init_clock_gating(dev);
4053 static void ivybridge_init_clock_gating(struct drm_device *dev)
4055 struct drm_i915_private *dev_priv = dev->dev_private;
4059 I915_WRITE(WM3_LP_ILK, 0);
4060 I915_WRITE(WM2_LP_ILK, 0);
4061 I915_WRITE(WM1_LP_ILK, 0);
4063 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
4065 /* WaDisableEarlyCull */
4066 I915_WRITE(_3D_CHICKEN3,
4067 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4069 /* WaDisableBackToBackFlipFix */
4070 I915_WRITE(IVB_CHICKEN3,
4071 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4072 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4074 /* WaDisablePSDDualDispatchEnable */
4075 if (IS_IVB_GT1(dev))
4076 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4077 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
4079 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2,
4080 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
4082 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
4083 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4084 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4086 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
4087 I915_WRITE(GEN7_L3CNTLREG1,
4088 GEN7_WA_FOR_GEN7_L3_CONTROL);
4089 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
4090 GEN7_WA_L3_CHICKEN_MODE);
4091 if (IS_IVB_GT1(dev))
4092 I915_WRITE(GEN7_ROW_CHICKEN2,
4093 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
4095 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
4096 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
4099 /* WaForceL3Serialization */
4100 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4101 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4103 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4104 * gating disable must be set. Failure to set it results in
4105 * flickering pixels due to Z write ordering failures after
4106 * some amount of runtime in the Mesa "fire" demo, and Unigine
4107 * Sanctuary and Tropics, and apparently anything else with
4108 * alpha test or pixel discard.
4110 * According to the spec, bit 11 (RCCUNIT) must also be set,
4111 * but we didn't debug actual testcases to find it out.
4113 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
4114 * This implements the WaDisableRCZUnitClockGating workaround.
4116 I915_WRITE(GEN6_UCGCTL2,
4117 GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
4118 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4120 /* This is required by WaCatErrorRejectionIssue */
4121 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4122 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4123 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4125 for_each_pipe(pipe) {
4126 I915_WRITE(DSPCNTR(pipe),
4127 I915_READ(DSPCNTR(pipe)) |
4128 DISPPLANE_TRICKLE_FEED_DISABLE);
4129 intel_flush_display_plane(dev_priv, pipe);
4132 /* WaMbcDriverBootEnable */
4133 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
4134 GEN6_MBCTL_ENABLE_BOOT_FETCH);
4136 gen7_setup_fixed_func_scheduler(dev_priv);
4138 /* WaDisable4x2SubspanOptimization */
4139 I915_WRITE(CACHE_MODE_1,
4140 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
4142 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
4143 snpcr &= ~GEN6_MBC_SNPCR_MASK;
4144 snpcr |= GEN6_MBC_SNPCR_MED;
4145 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
4147 if (!HAS_PCH_NOP(dev))
4148 cpt_init_clock_gating(dev);
4150 gen6_check_mch_setup(dev);
4153 static void valleyview_init_clock_gating(struct drm_device *dev)
4155 struct drm_i915_private *dev_priv = dev->dev_private;
4158 I915_WRITE(WM3_LP_ILK, 0);
4159 I915_WRITE(WM2_LP_ILK, 0);
4160 I915_WRITE(WM1_LP_ILK, 0);
4162 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
4164 /* WaDisableEarlyCull */
4165 I915_WRITE(_3D_CHICKEN3,
4166 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4168 /* WaDisableBackToBackFlipFix */
4169 I915_WRITE(IVB_CHICKEN3,
4170 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4171 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4173 /* WaDisablePSDDualDispatchEnable */
4174 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4175 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
4176 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
4178 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
4179 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4180 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4182 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
4183 I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS);
4184 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE);
4186 /* WaForceL3Serialization */
4187 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4188 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4190 /* WaDisableDopClockGating */
4191 I915_WRITE(GEN7_ROW_CHICKEN2,
4192 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
4194 /* WaForceL3Serialization */
4195 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4196 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4198 /* This is required by WaCatErrorRejectionIssue */
4199 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4200 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4201 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4203 /* WaMbcDriverBootEnable */
4204 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
4205 GEN6_MBCTL_ENABLE_BOOT_FETCH);
4208 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4209 * gating disable must be set. Failure to set it results in
4210 * flickering pixels due to Z write ordering failures after
4211 * some amount of runtime in the Mesa "fire" demo, and Unigine
4212 * Sanctuary and Tropics, and apparently anything else with
4213 * alpha test or pixel discard.
4215 * According to the spec, bit 11 (RCCUNIT) must also be set,
4216 * but we didn't debug actual testcases to find it out.
4218 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
4219 * This implements the WaDisableRCZUnitClockGating workaround.
4221 * Also apply WaDisableVDSUnitClockGating and
4222 * WaDisableRCPBUnitClockGating.
4224 I915_WRITE(GEN6_UCGCTL2,
4225 GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
4226 GEN7_TDLUNIT_CLOCK_GATE_DISABLE |
4227 GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
4228 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
4229 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4231 I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
4233 for_each_pipe(pipe) {
4234 I915_WRITE(DSPCNTR(pipe),
4235 I915_READ(DSPCNTR(pipe)) |
4236 DISPPLANE_TRICKLE_FEED_DISABLE);
4237 intel_flush_display_plane(dev_priv, pipe);
4240 I915_WRITE(CACHE_MODE_1,
4241 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
4244 * WaDisableVLVClockGating_VBIIssue
4245 * Disable clock gating on th GCFG unit to prevent a delay
4246 * in the reporting of vblank events.
4248 I915_WRITE(VLV_GUNIT_CLOCK_GATE, 0xffffffff);
4250 /* Conservative clock gating settings for now */
4251 I915_WRITE(0x9400, 0xffffffff);
4252 I915_WRITE(0x9404, 0xffffffff);
4253 I915_WRITE(0x9408, 0xffffffff);
4254 I915_WRITE(0x940c, 0xffffffff);
4255 I915_WRITE(0x9410, 0xffffffff);
4256 I915_WRITE(0x9414, 0xffffffff);
4257 I915_WRITE(0x9418, 0xffffffff);
4260 static void g4x_init_clock_gating(struct drm_device *dev)
4262 struct drm_i915_private *dev_priv = dev->dev_private;
4263 uint32_t dspclk_gate;
4265 I915_WRITE(RENCLK_GATE_D1, 0);
4266 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
4267 GS_UNIT_CLOCK_GATE_DISABLE |
4268 CL_UNIT_CLOCK_GATE_DISABLE);
4269 I915_WRITE(RAMCLK_GATE_D, 0);
4270 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
4271 OVRUNIT_CLOCK_GATE_DISABLE |
4272 OVCUNIT_CLOCK_GATE_DISABLE;
4274 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
4275 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4277 /* WaDisableRenderCachePipelinedFlush */
4278 I915_WRITE(CACHE_MODE_0,
4279 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
4282 static void crestline_init_clock_gating(struct drm_device *dev)
4284 struct drm_i915_private *dev_priv = dev->dev_private;
4286 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
4287 I915_WRITE(RENCLK_GATE_D2, 0);
4288 I915_WRITE(DSPCLK_GATE_D, 0);
4289 I915_WRITE(RAMCLK_GATE_D, 0);
4290 I915_WRITE16(DEUC, 0);
4293 static void broadwater_init_clock_gating(struct drm_device *dev)
4295 struct drm_i915_private *dev_priv = dev->dev_private;
4297 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
4298 I965_RCC_CLOCK_GATE_DISABLE |
4299 I965_RCPB_CLOCK_GATE_DISABLE |
4300 I965_ISC_CLOCK_GATE_DISABLE |
4301 I965_FBC_CLOCK_GATE_DISABLE);
4302 I915_WRITE(RENCLK_GATE_D2, 0);
4305 static void gen3_init_clock_gating(struct drm_device *dev)
4307 struct drm_i915_private *dev_priv = dev->dev_private;
4308 u32 dstate = I915_READ(D_STATE);
4310 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
4311 DSTATE_DOT_CLOCK_GATING;
4312 I915_WRITE(D_STATE, dstate);
4314 if (IS_PINEVIEW(dev))
4315 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
4317 /* IIR "flip pending" means done if this bit is set */
4318 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
4321 static void i85x_init_clock_gating(struct drm_device *dev)
4323 struct drm_i915_private *dev_priv = dev->dev_private;
4325 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
4328 static void i830_init_clock_gating(struct drm_device *dev)
4330 struct drm_i915_private *dev_priv = dev->dev_private;
4332 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
4335 void intel_init_clock_gating(struct drm_device *dev)
4337 struct drm_i915_private *dev_priv = dev->dev_private;
4339 dev_priv->display.init_clock_gating(dev);
4343 * We should only use the power well if we explicitly asked the hardware to
4344 * enable it, so check if it's enabled and also check if we've requested it to
4347 bool intel_using_power_well(struct drm_device *dev)
4349 struct drm_i915_private *dev_priv = dev->dev_private;
4351 if (IS_HASWELL(dev))
4352 return I915_READ(HSW_PWR_WELL_DRIVER) ==
4353 (HSW_PWR_WELL_ENABLE | HSW_PWR_WELL_STATE);
4358 void intel_set_power_well(struct drm_device *dev, bool enable)
4360 struct drm_i915_private *dev_priv = dev->dev_private;
4361 bool is_enabled, enable_requested;
4364 if (!HAS_POWER_WELL(dev))
4367 if (!i915_disable_power_well && !enable)
4370 tmp = I915_READ(HSW_PWR_WELL_DRIVER);
4371 is_enabled = tmp & HSW_PWR_WELL_STATE;
4372 enable_requested = tmp & HSW_PWR_WELL_ENABLE;
4375 if (!enable_requested)
4376 I915_WRITE(HSW_PWR_WELL_DRIVER, HSW_PWR_WELL_ENABLE);
4379 DRM_DEBUG_KMS("Enabling power well\n");
4380 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
4381 HSW_PWR_WELL_STATE), 20))
4382 DRM_ERROR("Timeout enabling power well\n");
4385 if (enable_requested) {
4386 I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
4387 DRM_DEBUG_KMS("Requesting to disable the power well\n");
4393 * Starting with Haswell, we have a "Power Down Well" that can be turned off
4394 * when not needed anymore. We have 4 registers that can request the power well
4395 * to be enabled, and it will only be disabled if none of the registers is
4396 * requesting it to be enabled.
4398 void intel_init_power_well(struct drm_device *dev)
4400 struct drm_i915_private *dev_priv = dev->dev_private;
4402 if (!HAS_POWER_WELL(dev))
4405 /* For now, we need the power well to be always enabled. */
4406 intel_set_power_well(dev, true);
4408 /* We're taking over the BIOS, so clear any requests made by it since
4409 * the driver is in charge now. */
4410 if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE)
4411 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
4414 /* Set up chip specific power management-related functions */
4415 void intel_init_pm(struct drm_device *dev)
4417 struct drm_i915_private *dev_priv = dev->dev_private;
4419 if (I915_HAS_FBC(dev)) {
4420 if (HAS_PCH_SPLIT(dev)) {
4421 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
4422 dev_priv->display.enable_fbc = ironlake_enable_fbc;
4423 dev_priv->display.disable_fbc = ironlake_disable_fbc;
4424 } else if (IS_GM45(dev)) {
4425 dev_priv->display.fbc_enabled = g4x_fbc_enabled;
4426 dev_priv->display.enable_fbc = g4x_enable_fbc;
4427 dev_priv->display.disable_fbc = g4x_disable_fbc;
4428 } else if (IS_CRESTLINE(dev)) {
4429 dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
4430 dev_priv->display.enable_fbc = i8xx_enable_fbc;
4431 dev_priv->display.disable_fbc = i8xx_disable_fbc;
4433 /* 855GM needs testing */
4437 if (IS_PINEVIEW(dev))
4438 i915_pineview_get_mem_freq(dev);
4439 else if (IS_GEN5(dev))
4440 i915_ironlake_get_mem_freq(dev);
4442 /* For FIFO watermark updates */
4443 if (HAS_PCH_SPLIT(dev)) {
4445 if (I915_READ(MLTR_ILK) & ILK_SRLT_MASK)
4446 dev_priv->display.update_wm = ironlake_update_wm;
4448 DRM_DEBUG_KMS("Failed to get proper latency. "
4450 dev_priv->display.update_wm = NULL;
4452 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
4453 } else if (IS_GEN6(dev)) {
4454 if (SNB_READ_WM0_LATENCY()) {
4455 dev_priv->display.update_wm = sandybridge_update_wm;
4456 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4458 DRM_DEBUG_KMS("Failed to read display plane latency. "
4460 dev_priv->display.update_wm = NULL;
4462 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
4463 } else if (IS_IVYBRIDGE(dev)) {
4464 if (SNB_READ_WM0_LATENCY()) {
4465 dev_priv->display.update_wm = ivybridge_update_wm;
4466 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4468 DRM_DEBUG_KMS("Failed to read display plane latency. "
4470 dev_priv->display.update_wm = NULL;
4472 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
4473 } else if (IS_HASWELL(dev)) {
4474 if (SNB_READ_WM0_LATENCY()) {
4475 dev_priv->display.update_wm = sandybridge_update_wm;
4476 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4477 dev_priv->display.update_linetime_wm = haswell_update_linetime_wm;
4479 DRM_DEBUG_KMS("Failed to read display plane latency. "
4481 dev_priv->display.update_wm = NULL;
4483 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
4485 dev_priv->display.update_wm = NULL;
4486 } else if (IS_VALLEYVIEW(dev)) {
4487 dev_priv->display.update_wm = valleyview_update_wm;
4488 dev_priv->display.init_clock_gating =
4489 valleyview_init_clock_gating;
4490 } else if (IS_PINEVIEW(dev)) {
4491 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
4494 dev_priv->mem_freq)) {
4495 DRM_INFO("failed to find known CxSR latency "
4496 "(found ddr%s fsb freq %d, mem freq %d), "
4498 (dev_priv->is_ddr3 == 1) ? "3" : "2",
4499 dev_priv->fsb_freq, dev_priv->mem_freq);
4500 /* Disable CxSR and never update its watermark again */
4501 pineview_disable_cxsr(dev);
4502 dev_priv->display.update_wm = NULL;
4504 dev_priv->display.update_wm = pineview_update_wm;
4505 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
4506 } else if (IS_G4X(dev)) {
4507 dev_priv->display.update_wm = g4x_update_wm;
4508 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
4509 } else if (IS_GEN4(dev)) {
4510 dev_priv->display.update_wm = i965_update_wm;
4511 if (IS_CRESTLINE(dev))
4512 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
4513 else if (IS_BROADWATER(dev))
4514 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
4515 } else if (IS_GEN3(dev)) {
4516 dev_priv->display.update_wm = i9xx_update_wm;
4517 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
4518 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
4519 } else if (IS_I865G(dev)) {
4520 dev_priv->display.update_wm = i830_update_wm;
4521 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
4522 dev_priv->display.get_fifo_size = i830_get_fifo_size;
4523 } else if (IS_I85X(dev)) {
4524 dev_priv->display.update_wm = i9xx_update_wm;
4525 dev_priv->display.get_fifo_size = i85x_get_fifo_size;
4526 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
4528 dev_priv->display.update_wm = i830_update_wm;
4529 dev_priv->display.init_clock_gating = i830_init_clock_gating;
4531 dev_priv->display.get_fifo_size = i845_get_fifo_size;
4533 dev_priv->display.get_fifo_size = i830_get_fifo_size;
4537 static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv)
4539 u32 gt_thread_status_mask;
4541 if (IS_HASWELL(dev_priv->dev))
4542 gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK_HSW;
4544 gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK;
4546 /* w/a for a sporadic read returning 0 by waiting for the GT
4547 * thread to wake up.
4549 if (wait_for_atomic_us((I915_READ_NOTRACE(GEN6_GT_THREAD_STATUS_REG) & gt_thread_status_mask) == 0, 500))
4550 DRM_ERROR("GT thread status wait timed out\n");
4553 static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv)
4555 I915_WRITE_NOTRACE(FORCEWAKE, 0);
4556 POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */
4559 static void __gen6_gt_force_wake_get(struct drm_i915_private *dev_priv)
4561 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1) == 0,
4562 FORCEWAKE_ACK_TIMEOUT_MS))
4563 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4565 I915_WRITE_NOTRACE(FORCEWAKE, 1);
4566 POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */
4568 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1),
4569 FORCEWAKE_ACK_TIMEOUT_MS))
4570 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4572 __gen6_gt_wait_for_thread_c0(dev_priv);
4575 static void __gen6_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv)
4577 I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(0xffff));
4578 /* something from same cacheline, but !FORCEWAKE_MT */
4579 POSTING_READ(ECOBUS);
4582 static void __gen6_gt_force_wake_mt_get(struct drm_i915_private *dev_priv)
4586 if (IS_HASWELL(dev_priv->dev))
4587 forcewake_ack = FORCEWAKE_ACK_HSW;
4589 forcewake_ack = FORCEWAKE_MT_ACK;
4591 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL) == 0,
4592 FORCEWAKE_ACK_TIMEOUT_MS))
4593 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4595 I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4596 /* something from same cacheline, but !FORCEWAKE_MT */
4597 POSTING_READ(ECOBUS);
4599 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL),
4600 FORCEWAKE_ACK_TIMEOUT_MS))
4601 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4603 __gen6_gt_wait_for_thread_c0(dev_priv);
4607 * Generally this is called implicitly by the register read function. However,
4608 * if some sequence requires the GT to not power down then this function should
4609 * be called at the beginning of the sequence followed by a call to
4610 * gen6_gt_force_wake_put() at the end of the sequence.
4612 void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv)
4614 unsigned long irqflags;
4616 spin_lock_irqsave(&dev_priv->gt_lock, irqflags);
4617 if (dev_priv->forcewake_count++ == 0)
4618 dev_priv->gt.force_wake_get(dev_priv);
4619 spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags);
4622 void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv)
4625 gtfifodbg = I915_READ_NOTRACE(GTFIFODBG);
4626 if (WARN(gtfifodbg & GT_FIFO_CPU_ERROR_MASK,
4627 "MMIO read or write has been dropped %x\n", gtfifodbg))
4628 I915_WRITE_NOTRACE(GTFIFODBG, GT_FIFO_CPU_ERROR_MASK);
4631 static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv)
4633 I915_WRITE_NOTRACE(FORCEWAKE, 0);
4634 /* something from same cacheline, but !FORCEWAKE */
4635 POSTING_READ(ECOBUS);
4636 gen6_gt_check_fifodbg(dev_priv);
4639 static void __gen6_gt_force_wake_mt_put(struct drm_i915_private *dev_priv)
4641 I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4642 /* something from same cacheline, but !FORCEWAKE_MT */
4643 POSTING_READ(ECOBUS);
4644 gen6_gt_check_fifodbg(dev_priv);
4648 * see gen6_gt_force_wake_get()
4650 void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv)
4652 unsigned long irqflags;
4654 spin_lock_irqsave(&dev_priv->gt_lock, irqflags);
4655 if (--dev_priv->forcewake_count == 0)
4656 dev_priv->gt.force_wake_put(dev_priv);
4657 spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags);
4660 int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv)
4664 if (dev_priv->gt_fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) {
4666 u32 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES);
4667 while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) {
4669 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES);
4671 if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES))
4673 dev_priv->gt_fifo_count = fifo;
4675 dev_priv->gt_fifo_count--;
4680 static void vlv_force_wake_reset(struct drm_i915_private *dev_priv)
4682 I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(0xffff));
4683 /* something from same cacheline, but !FORCEWAKE_VLV */
4684 POSTING_READ(FORCEWAKE_ACK_VLV);
4687 static void vlv_force_wake_get(struct drm_i915_private *dev_priv)
4689 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL) == 0,
4690 FORCEWAKE_ACK_TIMEOUT_MS))
4691 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4693 I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4694 I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV,
4695 _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4697 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL),
4698 FORCEWAKE_ACK_TIMEOUT_MS))
4699 DRM_ERROR("Timed out waiting for GT to ack forcewake request.\n");
4701 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_MEDIA_VLV) &
4703 FORCEWAKE_ACK_TIMEOUT_MS))
4704 DRM_ERROR("Timed out waiting for media to ack forcewake request.\n");
4706 __gen6_gt_wait_for_thread_c0(dev_priv);
4709 static void vlv_force_wake_put(struct drm_i915_private *dev_priv)
4711 I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4712 I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV,
4713 _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4714 /* The below doubles as a POSTING_READ */
4715 gen6_gt_check_fifodbg(dev_priv);
4718 void intel_gt_reset(struct drm_device *dev)
4720 struct drm_i915_private *dev_priv = dev->dev_private;
4722 if (IS_VALLEYVIEW(dev)) {
4723 vlv_force_wake_reset(dev_priv);
4724 } else if (INTEL_INFO(dev)->gen >= 6) {
4725 __gen6_gt_force_wake_reset(dev_priv);
4726 if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
4727 __gen6_gt_force_wake_mt_reset(dev_priv);
4731 void intel_gt_init(struct drm_device *dev)
4733 struct drm_i915_private *dev_priv = dev->dev_private;
4735 spin_lock_init(&dev_priv->gt_lock);
4737 intel_gt_reset(dev);
4739 if (IS_VALLEYVIEW(dev)) {
4740 dev_priv->gt.force_wake_get = vlv_force_wake_get;
4741 dev_priv->gt.force_wake_put = vlv_force_wake_put;
4742 } else if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) {
4743 dev_priv->gt.force_wake_get = __gen6_gt_force_wake_mt_get;
4744 dev_priv->gt.force_wake_put = __gen6_gt_force_wake_mt_put;
4745 } else if (IS_GEN6(dev)) {
4746 dev_priv->gt.force_wake_get = __gen6_gt_force_wake_get;
4747 dev_priv->gt.force_wake_put = __gen6_gt_force_wake_put;
4749 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
4750 intel_gen6_powersave_work);
4753 int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
4755 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4757 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
4758 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
4762 I915_WRITE(GEN6_PCODE_DATA, *val);
4763 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
4765 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
4767 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
4771 *val = I915_READ(GEN6_PCODE_DATA);
4772 I915_WRITE(GEN6_PCODE_DATA, 0);
4777 int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
4779 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4781 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
4782 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
4786 I915_WRITE(GEN6_PCODE_DATA, val);
4787 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
4789 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
4791 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
4795 I915_WRITE(GEN6_PCODE_DATA, 0);
4800 static int vlv_punit_rw(struct drm_i915_private *dev_priv, u32 port, u8 opcode,
4803 u32 cmd, devfn, be, bar;
4807 devfn = PCI_DEVFN(2, 0);
4809 cmd = (devfn << IOSF_DEVFN_SHIFT) | (opcode << IOSF_OPCODE_SHIFT) |
4810 (port << IOSF_PORT_SHIFT) | (be << IOSF_BYTE_ENABLES_SHIFT) |
4811 (bar << IOSF_BAR_SHIFT);
4813 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4815 if (I915_READ(VLV_IOSF_DOORBELL_REQ) & IOSF_SB_BUSY) {
4816 DRM_DEBUG_DRIVER("warning: pcode (%s) mailbox access failed\n",
4817 opcode == PUNIT_OPCODE_REG_READ ?
4822 I915_WRITE(VLV_IOSF_ADDR, addr);
4823 if (opcode == PUNIT_OPCODE_REG_WRITE)
4824 I915_WRITE(VLV_IOSF_DATA, *val);
4825 I915_WRITE(VLV_IOSF_DOORBELL_REQ, cmd);
4827 if (wait_for((I915_READ(VLV_IOSF_DOORBELL_REQ) & IOSF_SB_BUSY) == 0,
4829 DRM_ERROR("timeout waiting for pcode %s (%d) to finish\n",
4830 opcode == PUNIT_OPCODE_REG_READ ? "read" : "write",
4835 if (opcode == PUNIT_OPCODE_REG_READ)
4836 *val = I915_READ(VLV_IOSF_DATA);
4837 I915_WRITE(VLV_IOSF_DATA, 0);
4842 int valleyview_punit_read(struct drm_i915_private *dev_priv, u8 addr, u32 *val)
4844 return vlv_punit_rw(dev_priv, IOSF_PORT_PUNIT, PUNIT_OPCODE_REG_READ,
4848 int valleyview_punit_write(struct drm_i915_private *dev_priv, u8 addr, u32 val)
4850 return vlv_punit_rw(dev_priv, IOSF_PORT_PUNIT, PUNIT_OPCODE_REG_WRITE,
4854 int valleyview_nc_read(struct drm_i915_private *dev_priv, u8 addr, u32 *val)
4856 return vlv_punit_rw(dev_priv, IOSF_PORT_NC, PUNIT_OPCODE_REG_READ,
4860 int vlv_gpu_freq(int ddr_freq, int val)
4881 return ((val - 0xbd) * mult) + base;
4884 int vlv_freq_opcode(int ddr_freq, int val)