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_INFO("not enough stolen space for compressed buffer (need %zd bytes), disabling\n", intel_fb->obj->base.size);
485 DRM_INFO("hint: you may be able to increase stolen memory size in the BIOS to avoid this\n");
486 DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
487 dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
491 /* If the scanout has not changed, don't modify the FBC settings.
492 * Note that we make the fundamental assumption that the fb->obj
493 * cannot be unpinned (and have its GTT offset and fence revoked)
494 * without first being decoupled from the scanout and FBC disabled.
496 if (dev_priv->cfb_plane == intel_crtc->plane &&
497 dev_priv->cfb_fb == fb->base.id &&
498 dev_priv->cfb_y == crtc->y)
501 if (intel_fbc_enabled(dev)) {
502 /* We update FBC along two paths, after changing fb/crtc
503 * configuration (modeswitching) and after page-flipping
504 * finishes. For the latter, we know that not only did
505 * we disable the FBC at the start of the page-flip
506 * sequence, but also more than one vblank has passed.
508 * For the former case of modeswitching, it is possible
509 * to switch between two FBC valid configurations
510 * instantaneously so we do need to disable the FBC
511 * before we can modify its control registers. We also
512 * have to wait for the next vblank for that to take
513 * effect. However, since we delay enabling FBC we can
514 * assume that a vblank has passed since disabling and
515 * that we can safely alter the registers in the deferred
518 * In the scenario that we go from a valid to invalid
519 * and then back to valid FBC configuration we have
520 * no strict enforcement that a vblank occurred since
521 * disabling the FBC. However, along all current pipe
522 * disabling paths we do need to wait for a vblank at
523 * some point. And we wait before enabling FBC anyway.
525 DRM_DEBUG_KMS("disabling active FBC for update\n");
526 intel_disable_fbc(dev);
529 intel_enable_fbc(crtc, 500);
533 /* Multiple disables should be harmless */
534 if (intel_fbc_enabled(dev)) {
535 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
536 intel_disable_fbc(dev);
538 i915_gem_stolen_cleanup_compression(dev);
541 static void i915_pineview_get_mem_freq(struct drm_device *dev)
543 drm_i915_private_t *dev_priv = dev->dev_private;
546 tmp = I915_READ(CLKCFG);
548 switch (tmp & CLKCFG_FSB_MASK) {
550 dev_priv->fsb_freq = 533; /* 133*4 */
553 dev_priv->fsb_freq = 800; /* 200*4 */
556 dev_priv->fsb_freq = 667; /* 167*4 */
559 dev_priv->fsb_freq = 400; /* 100*4 */
563 switch (tmp & CLKCFG_MEM_MASK) {
565 dev_priv->mem_freq = 533;
568 dev_priv->mem_freq = 667;
571 dev_priv->mem_freq = 800;
575 /* detect pineview DDR3 setting */
576 tmp = I915_READ(CSHRDDR3CTL);
577 dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
580 static void i915_ironlake_get_mem_freq(struct drm_device *dev)
582 drm_i915_private_t *dev_priv = dev->dev_private;
585 ddrpll = I915_READ16(DDRMPLL1);
586 csipll = I915_READ16(CSIPLL0);
588 switch (ddrpll & 0xff) {
590 dev_priv->mem_freq = 800;
593 dev_priv->mem_freq = 1066;
596 dev_priv->mem_freq = 1333;
599 dev_priv->mem_freq = 1600;
602 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
604 dev_priv->mem_freq = 0;
608 dev_priv->ips.r_t = dev_priv->mem_freq;
610 switch (csipll & 0x3ff) {
612 dev_priv->fsb_freq = 3200;
615 dev_priv->fsb_freq = 3733;
618 dev_priv->fsb_freq = 4266;
621 dev_priv->fsb_freq = 4800;
624 dev_priv->fsb_freq = 5333;
627 dev_priv->fsb_freq = 5866;
630 dev_priv->fsb_freq = 6400;
633 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
635 dev_priv->fsb_freq = 0;
639 if (dev_priv->fsb_freq == 3200) {
640 dev_priv->ips.c_m = 0;
641 } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
642 dev_priv->ips.c_m = 1;
644 dev_priv->ips.c_m = 2;
648 static const struct cxsr_latency cxsr_latency_table[] = {
649 {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */
650 {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */
651 {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */
652 {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */
653 {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */
655 {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */
656 {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */
657 {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */
658 {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */
659 {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */
661 {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */
662 {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */
663 {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */
664 {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */
665 {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */
667 {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */
668 {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */
669 {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */
670 {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */
671 {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */
673 {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */
674 {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */
675 {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */
676 {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */
677 {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */
679 {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */
680 {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */
681 {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */
682 {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */
683 {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */
686 static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
691 const struct cxsr_latency *latency;
694 if (fsb == 0 || mem == 0)
697 for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
698 latency = &cxsr_latency_table[i];
699 if (is_desktop == latency->is_desktop &&
700 is_ddr3 == latency->is_ddr3 &&
701 fsb == latency->fsb_freq && mem == latency->mem_freq)
705 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
710 static void pineview_disable_cxsr(struct drm_device *dev)
712 struct drm_i915_private *dev_priv = dev->dev_private;
714 /* deactivate cxsr */
715 I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
719 * Latency for FIFO fetches is dependent on several factors:
720 * - memory configuration (speed, channels)
722 * - current MCH state
723 * It can be fairly high in some situations, so here we assume a fairly
724 * pessimal value. It's a tradeoff between extra memory fetches (if we
725 * set this value too high, the FIFO will fetch frequently to stay full)
726 * and power consumption (set it too low to save power and we might see
727 * FIFO underruns and display "flicker").
729 * A value of 5us seems to be a good balance; safe for very low end
730 * platforms but not overly aggressive on lower latency configs.
732 static const int latency_ns = 5000;
734 static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
736 struct drm_i915_private *dev_priv = dev->dev_private;
737 uint32_t dsparb = I915_READ(DSPARB);
740 size = dsparb & 0x7f;
742 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
744 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
745 plane ? "B" : "A", size);
750 static int i85x_get_fifo_size(struct drm_device *dev, int plane)
752 struct drm_i915_private *dev_priv = dev->dev_private;
753 uint32_t dsparb = I915_READ(DSPARB);
756 size = dsparb & 0x1ff;
758 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
759 size >>= 1; /* Convert to cachelines */
761 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
762 plane ? "B" : "A", size);
767 static int i845_get_fifo_size(struct drm_device *dev, int plane)
769 struct drm_i915_private *dev_priv = dev->dev_private;
770 uint32_t dsparb = I915_READ(DSPARB);
773 size = dsparb & 0x7f;
774 size >>= 2; /* Convert to cachelines */
776 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
783 static int i830_get_fifo_size(struct drm_device *dev, int plane)
785 struct drm_i915_private *dev_priv = dev->dev_private;
786 uint32_t dsparb = I915_READ(DSPARB);
789 size = dsparb & 0x7f;
790 size >>= 1; /* Convert to cachelines */
792 DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
793 plane ? "B" : "A", size);
798 /* Pineview has different values for various configs */
799 static const struct intel_watermark_params pineview_display_wm = {
800 PINEVIEW_DISPLAY_FIFO,
804 PINEVIEW_FIFO_LINE_SIZE
806 static const struct intel_watermark_params pineview_display_hplloff_wm = {
807 PINEVIEW_DISPLAY_FIFO,
809 PINEVIEW_DFT_HPLLOFF_WM,
811 PINEVIEW_FIFO_LINE_SIZE
813 static const struct intel_watermark_params pineview_cursor_wm = {
814 PINEVIEW_CURSOR_FIFO,
815 PINEVIEW_CURSOR_MAX_WM,
816 PINEVIEW_CURSOR_DFT_WM,
817 PINEVIEW_CURSOR_GUARD_WM,
818 PINEVIEW_FIFO_LINE_SIZE,
820 static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
821 PINEVIEW_CURSOR_FIFO,
822 PINEVIEW_CURSOR_MAX_WM,
823 PINEVIEW_CURSOR_DFT_WM,
824 PINEVIEW_CURSOR_GUARD_WM,
825 PINEVIEW_FIFO_LINE_SIZE
827 static const struct intel_watermark_params g4x_wm_info = {
834 static const struct intel_watermark_params g4x_cursor_wm_info = {
841 static const struct intel_watermark_params valleyview_wm_info = {
842 VALLEYVIEW_FIFO_SIZE,
848 static const struct intel_watermark_params valleyview_cursor_wm_info = {
850 VALLEYVIEW_CURSOR_MAX_WM,
855 static const struct intel_watermark_params i965_cursor_wm_info = {
862 static const struct intel_watermark_params i945_wm_info = {
869 static const struct intel_watermark_params i915_wm_info = {
876 static const struct intel_watermark_params i855_wm_info = {
883 static const struct intel_watermark_params i830_wm_info = {
891 static const struct intel_watermark_params ironlake_display_wm_info = {
898 static const struct intel_watermark_params ironlake_cursor_wm_info = {
905 static const struct intel_watermark_params ironlake_display_srwm_info = {
907 ILK_DISPLAY_MAX_SRWM,
908 ILK_DISPLAY_DFT_SRWM,
912 static const struct intel_watermark_params ironlake_cursor_srwm_info = {
920 static const struct intel_watermark_params sandybridge_display_wm_info = {
927 static const struct intel_watermark_params sandybridge_cursor_wm_info = {
934 static const struct intel_watermark_params sandybridge_display_srwm_info = {
936 SNB_DISPLAY_MAX_SRWM,
937 SNB_DISPLAY_DFT_SRWM,
941 static const struct intel_watermark_params sandybridge_cursor_srwm_info = {
951 * intel_calculate_wm - calculate watermark level
952 * @clock_in_khz: pixel clock
953 * @wm: chip FIFO params
954 * @pixel_size: display pixel size
955 * @latency_ns: memory latency for the platform
957 * Calculate the watermark level (the level at which the display plane will
958 * start fetching from memory again). Each chip has a different display
959 * FIFO size and allocation, so the caller needs to figure that out and pass
960 * in the correct intel_watermark_params structure.
962 * As the pixel clock runs, the FIFO will be drained at a rate that depends
963 * on the pixel size. When it reaches the watermark level, it'll start
964 * fetching FIFO line sized based chunks from memory until the FIFO fills
965 * past the watermark point. If the FIFO drains completely, a FIFO underrun
966 * will occur, and a display engine hang could result.
968 static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
969 const struct intel_watermark_params *wm,
972 unsigned long latency_ns)
974 long entries_required, wm_size;
977 * Note: we need to make sure we don't overflow for various clock &
979 * clocks go from a few thousand to several hundred thousand.
980 * latency is usually a few thousand
982 entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
984 entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
986 DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
988 wm_size = fifo_size - (entries_required + wm->guard_size);
990 DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
992 /* Don't promote wm_size to unsigned... */
993 if (wm_size > (long)wm->max_wm)
994 wm_size = wm->max_wm;
996 wm_size = wm->default_wm;
1000 static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1002 struct drm_crtc *crtc, *enabled = NULL;
1004 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
1005 if (intel_crtc_active(crtc)) {
1015 static void pineview_update_wm(struct drm_device *dev)
1017 struct drm_i915_private *dev_priv = dev->dev_private;
1018 struct drm_crtc *crtc;
1019 const struct cxsr_latency *latency;
1023 latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1024 dev_priv->fsb_freq, dev_priv->mem_freq);
1026 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1027 pineview_disable_cxsr(dev);
1031 crtc = single_enabled_crtc(dev);
1033 int clock = crtc->mode.clock;
1034 int pixel_size = crtc->fb->bits_per_pixel / 8;
1037 wm = intel_calculate_wm(clock, &pineview_display_wm,
1038 pineview_display_wm.fifo_size,
1039 pixel_size, latency->display_sr);
1040 reg = I915_READ(DSPFW1);
1041 reg &= ~DSPFW_SR_MASK;
1042 reg |= wm << DSPFW_SR_SHIFT;
1043 I915_WRITE(DSPFW1, reg);
1044 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1047 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1048 pineview_display_wm.fifo_size,
1049 pixel_size, latency->cursor_sr);
1050 reg = I915_READ(DSPFW3);
1051 reg &= ~DSPFW_CURSOR_SR_MASK;
1052 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1053 I915_WRITE(DSPFW3, reg);
1055 /* Display HPLL off SR */
1056 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1057 pineview_display_hplloff_wm.fifo_size,
1058 pixel_size, latency->display_hpll_disable);
1059 reg = I915_READ(DSPFW3);
1060 reg &= ~DSPFW_HPLL_SR_MASK;
1061 reg |= wm & DSPFW_HPLL_SR_MASK;
1062 I915_WRITE(DSPFW3, reg);
1064 /* cursor HPLL off SR */
1065 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1066 pineview_display_hplloff_wm.fifo_size,
1067 pixel_size, latency->cursor_hpll_disable);
1068 reg = I915_READ(DSPFW3);
1069 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1070 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1071 I915_WRITE(DSPFW3, reg);
1072 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1076 I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1077 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1079 pineview_disable_cxsr(dev);
1080 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1084 static bool g4x_compute_wm0(struct drm_device *dev,
1086 const struct intel_watermark_params *display,
1087 int display_latency_ns,
1088 const struct intel_watermark_params *cursor,
1089 int cursor_latency_ns,
1093 struct drm_crtc *crtc;
1094 int htotal, hdisplay, clock, pixel_size;
1095 int line_time_us, line_count;
1096 int entries, tlb_miss;
1098 crtc = intel_get_crtc_for_plane(dev, plane);
1099 if (!intel_crtc_active(crtc)) {
1100 *cursor_wm = cursor->guard_size;
1101 *plane_wm = display->guard_size;
1105 htotal = crtc->mode.htotal;
1106 hdisplay = crtc->mode.hdisplay;
1107 clock = crtc->mode.clock;
1108 pixel_size = crtc->fb->bits_per_pixel / 8;
1110 /* Use the small buffer method to calculate plane watermark */
1111 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1112 tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1114 entries += tlb_miss;
1115 entries = DIV_ROUND_UP(entries, display->cacheline_size);
1116 *plane_wm = entries + display->guard_size;
1117 if (*plane_wm > (int)display->max_wm)
1118 *plane_wm = display->max_wm;
1120 /* Use the large buffer method to calculate cursor watermark */
1121 line_time_us = ((htotal * 1000) / clock);
1122 line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
1123 entries = line_count * 64 * pixel_size;
1124 tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1126 entries += tlb_miss;
1127 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1128 *cursor_wm = entries + cursor->guard_size;
1129 if (*cursor_wm > (int)cursor->max_wm)
1130 *cursor_wm = (int)cursor->max_wm;
1136 * Check the wm result.
1138 * If any calculated watermark values is larger than the maximum value that
1139 * can be programmed into the associated watermark register, that watermark
1142 static bool g4x_check_srwm(struct drm_device *dev,
1143 int display_wm, int cursor_wm,
1144 const struct intel_watermark_params *display,
1145 const struct intel_watermark_params *cursor)
1147 DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1148 display_wm, cursor_wm);
1150 if (display_wm > display->max_wm) {
1151 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1152 display_wm, display->max_wm);
1156 if (cursor_wm > cursor->max_wm) {
1157 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1158 cursor_wm, cursor->max_wm);
1162 if (!(display_wm || cursor_wm)) {
1163 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1170 static bool g4x_compute_srwm(struct drm_device *dev,
1173 const struct intel_watermark_params *display,
1174 const struct intel_watermark_params *cursor,
1175 int *display_wm, int *cursor_wm)
1177 struct drm_crtc *crtc;
1178 int hdisplay, htotal, pixel_size, clock;
1179 unsigned long line_time_us;
1180 int line_count, line_size;
1185 *display_wm = *cursor_wm = 0;
1189 crtc = intel_get_crtc_for_plane(dev, plane);
1190 hdisplay = crtc->mode.hdisplay;
1191 htotal = crtc->mode.htotal;
1192 clock = crtc->mode.clock;
1193 pixel_size = crtc->fb->bits_per_pixel / 8;
1195 line_time_us = (htotal * 1000) / clock;
1196 line_count = (latency_ns / line_time_us + 1000) / 1000;
1197 line_size = hdisplay * pixel_size;
1199 /* Use the minimum of the small and large buffer method for primary */
1200 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1201 large = line_count * line_size;
1203 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1204 *display_wm = entries + display->guard_size;
1206 /* calculate the self-refresh watermark for display cursor */
1207 entries = line_count * pixel_size * 64;
1208 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1209 *cursor_wm = entries + cursor->guard_size;
1211 return g4x_check_srwm(dev,
1212 *display_wm, *cursor_wm,
1216 static bool vlv_compute_drain_latency(struct drm_device *dev,
1218 int *plane_prec_mult,
1220 int *cursor_prec_mult,
1223 struct drm_crtc *crtc;
1224 int clock, pixel_size;
1227 crtc = intel_get_crtc_for_plane(dev, plane);
1228 if (!intel_crtc_active(crtc))
1231 clock = crtc->mode.clock; /* VESA DOT Clock */
1232 pixel_size = crtc->fb->bits_per_pixel / 8; /* BPP */
1234 entries = (clock / 1000) * pixel_size;
1235 *plane_prec_mult = (entries > 256) ?
1236 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1237 *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1240 entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */
1241 *cursor_prec_mult = (entries > 256) ?
1242 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1243 *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1249 * Update drain latency registers of memory arbiter
1251 * Valleyview SoC has a new memory arbiter and needs drain latency registers
1252 * to be programmed. Each plane has a drain latency multiplier and a drain
1256 static void vlv_update_drain_latency(struct drm_device *dev)
1258 struct drm_i915_private *dev_priv = dev->dev_private;
1259 int planea_prec, planea_dl, planeb_prec, planeb_dl;
1260 int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1261 int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1264 /* For plane A, Cursor A */
1265 if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1266 &cursor_prec_mult, &cursora_dl)) {
1267 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1268 DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1269 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1270 DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1272 I915_WRITE(VLV_DDL1, cursora_prec |
1273 (cursora_dl << DDL_CURSORA_SHIFT) |
1274 planea_prec | planea_dl);
1277 /* For plane B, Cursor B */
1278 if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1279 &cursor_prec_mult, &cursorb_dl)) {
1280 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1281 DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1282 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1283 DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1285 I915_WRITE(VLV_DDL2, cursorb_prec |
1286 (cursorb_dl << DDL_CURSORB_SHIFT) |
1287 planeb_prec | planeb_dl);
1291 #define single_plane_enabled(mask) is_power_of_2(mask)
1293 static void valleyview_update_wm(struct drm_device *dev)
1295 static const int sr_latency_ns = 12000;
1296 struct drm_i915_private *dev_priv = dev->dev_private;
1297 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1298 int plane_sr, cursor_sr;
1299 int ignore_plane_sr, ignore_cursor_sr;
1300 unsigned int enabled = 0;
1302 vlv_update_drain_latency(dev);
1304 if (g4x_compute_wm0(dev, 0,
1305 &valleyview_wm_info, latency_ns,
1306 &valleyview_cursor_wm_info, latency_ns,
1307 &planea_wm, &cursora_wm))
1310 if (g4x_compute_wm0(dev, 1,
1311 &valleyview_wm_info, latency_ns,
1312 &valleyview_cursor_wm_info, latency_ns,
1313 &planeb_wm, &cursorb_wm))
1316 if (single_plane_enabled(enabled) &&
1317 g4x_compute_srwm(dev, ffs(enabled) - 1,
1319 &valleyview_wm_info,
1320 &valleyview_cursor_wm_info,
1321 &plane_sr, &ignore_cursor_sr) &&
1322 g4x_compute_srwm(dev, ffs(enabled) - 1,
1324 &valleyview_wm_info,
1325 &valleyview_cursor_wm_info,
1326 &ignore_plane_sr, &cursor_sr)) {
1327 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
1329 I915_WRITE(FW_BLC_SELF_VLV,
1330 I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
1331 plane_sr = cursor_sr = 0;
1334 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1335 planea_wm, cursora_wm,
1336 planeb_wm, cursorb_wm,
1337 plane_sr, cursor_sr);
1340 (plane_sr << DSPFW_SR_SHIFT) |
1341 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1342 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1345 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1346 (cursora_wm << DSPFW_CURSORA_SHIFT));
1348 (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1349 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1352 static void g4x_update_wm(struct drm_device *dev)
1354 static const int sr_latency_ns = 12000;
1355 struct drm_i915_private *dev_priv = dev->dev_private;
1356 int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1357 int plane_sr, cursor_sr;
1358 unsigned int enabled = 0;
1360 if (g4x_compute_wm0(dev, 0,
1361 &g4x_wm_info, latency_ns,
1362 &g4x_cursor_wm_info, latency_ns,
1363 &planea_wm, &cursora_wm))
1366 if (g4x_compute_wm0(dev, 1,
1367 &g4x_wm_info, latency_ns,
1368 &g4x_cursor_wm_info, latency_ns,
1369 &planeb_wm, &cursorb_wm))
1372 if (single_plane_enabled(enabled) &&
1373 g4x_compute_srwm(dev, ffs(enabled) - 1,
1376 &g4x_cursor_wm_info,
1377 &plane_sr, &cursor_sr)) {
1378 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1380 I915_WRITE(FW_BLC_SELF,
1381 I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
1382 plane_sr = cursor_sr = 0;
1385 DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1386 planea_wm, cursora_wm,
1387 planeb_wm, cursorb_wm,
1388 plane_sr, cursor_sr);
1391 (plane_sr << DSPFW_SR_SHIFT) |
1392 (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1393 (planeb_wm << DSPFW_PLANEB_SHIFT) |
1396 (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1397 (cursora_wm << DSPFW_CURSORA_SHIFT));
1398 /* HPLL off in SR has some issues on G4x... disable it */
1400 (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
1401 (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1404 static void i965_update_wm(struct drm_device *dev)
1406 struct drm_i915_private *dev_priv = dev->dev_private;
1407 struct drm_crtc *crtc;
1411 /* Calc sr entries for one plane configs */
1412 crtc = single_enabled_crtc(dev);
1414 /* self-refresh has much higher latency */
1415 static const int sr_latency_ns = 12000;
1416 int clock = crtc->mode.clock;
1417 int htotal = crtc->mode.htotal;
1418 int hdisplay = crtc->mode.hdisplay;
1419 int pixel_size = crtc->fb->bits_per_pixel / 8;
1420 unsigned long line_time_us;
1423 line_time_us = ((htotal * 1000) / clock);
1425 /* Use ns/us then divide to preserve precision */
1426 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1427 pixel_size * hdisplay;
1428 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1429 srwm = I965_FIFO_SIZE - entries;
1433 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1436 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1438 entries = DIV_ROUND_UP(entries,
1439 i965_cursor_wm_info.cacheline_size);
1440 cursor_sr = i965_cursor_wm_info.fifo_size -
1441 (entries + i965_cursor_wm_info.guard_size);
1443 if (cursor_sr > i965_cursor_wm_info.max_wm)
1444 cursor_sr = i965_cursor_wm_info.max_wm;
1446 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1447 "cursor %d\n", srwm, cursor_sr);
1449 if (IS_CRESTLINE(dev))
1450 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1452 /* Turn off self refresh if both pipes are enabled */
1453 if (IS_CRESTLINE(dev))
1454 I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1458 DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1461 /* 965 has limitations... */
1462 I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1463 (8 << 16) | (8 << 8) | (8 << 0));
1464 I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1465 /* update cursor SR watermark */
1466 I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1469 static void i9xx_update_wm(struct drm_device *dev)
1471 struct drm_i915_private *dev_priv = dev->dev_private;
1472 const struct intel_watermark_params *wm_info;
1477 int planea_wm, planeb_wm;
1478 struct drm_crtc *crtc, *enabled = NULL;
1481 wm_info = &i945_wm_info;
1482 else if (!IS_GEN2(dev))
1483 wm_info = &i915_wm_info;
1485 wm_info = &i855_wm_info;
1487 fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1488 crtc = intel_get_crtc_for_plane(dev, 0);
1489 if (intel_crtc_active(crtc)) {
1490 int cpp = crtc->fb->bits_per_pixel / 8;
1494 planea_wm = intel_calculate_wm(crtc->mode.clock,
1495 wm_info, fifo_size, cpp,
1499 planea_wm = fifo_size - wm_info->guard_size;
1501 fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1502 crtc = intel_get_crtc_for_plane(dev, 1);
1503 if (intel_crtc_active(crtc)) {
1504 int cpp = crtc->fb->bits_per_pixel / 8;
1508 planeb_wm = intel_calculate_wm(crtc->mode.clock,
1509 wm_info, fifo_size, cpp,
1511 if (enabled == NULL)
1516 planeb_wm = fifo_size - wm_info->guard_size;
1518 DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1521 * Overlay gets an aggressive default since video jitter is bad.
1525 /* Play safe and disable self-refresh before adjusting watermarks. */
1526 if (IS_I945G(dev) || IS_I945GM(dev))
1527 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1528 else if (IS_I915GM(dev))
1529 I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN);
1531 /* Calc sr entries for one plane configs */
1532 if (HAS_FW_BLC(dev) && enabled) {
1533 /* self-refresh has much higher latency */
1534 static const int sr_latency_ns = 6000;
1535 int clock = enabled->mode.clock;
1536 int htotal = enabled->mode.htotal;
1537 int hdisplay = enabled->mode.hdisplay;
1538 int pixel_size = enabled->fb->bits_per_pixel / 8;
1539 unsigned long line_time_us;
1542 line_time_us = (htotal * 1000) / clock;
1544 /* Use ns/us then divide to preserve precision */
1545 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1546 pixel_size * hdisplay;
1547 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1548 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1549 srwm = wm_info->fifo_size - entries;
1553 if (IS_I945G(dev) || IS_I945GM(dev))
1554 I915_WRITE(FW_BLC_SELF,
1555 FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1556 else if (IS_I915GM(dev))
1557 I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1560 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1561 planea_wm, planeb_wm, cwm, srwm);
1563 fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1564 fwater_hi = (cwm & 0x1f);
1566 /* Set request length to 8 cachelines per fetch */
1567 fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1568 fwater_hi = fwater_hi | (1 << 8);
1570 I915_WRITE(FW_BLC, fwater_lo);
1571 I915_WRITE(FW_BLC2, fwater_hi);
1573 if (HAS_FW_BLC(dev)) {
1575 if (IS_I945G(dev) || IS_I945GM(dev))
1576 I915_WRITE(FW_BLC_SELF,
1577 FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1578 else if (IS_I915GM(dev))
1579 I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN);
1580 DRM_DEBUG_KMS("memory self refresh enabled\n");
1582 DRM_DEBUG_KMS("memory self refresh disabled\n");
1586 static void i830_update_wm(struct drm_device *dev)
1588 struct drm_i915_private *dev_priv = dev->dev_private;
1589 struct drm_crtc *crtc;
1593 crtc = single_enabled_crtc(dev);
1597 planea_wm = intel_calculate_wm(crtc->mode.clock, &i830_wm_info,
1598 dev_priv->display.get_fifo_size(dev, 0),
1600 fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1601 fwater_lo |= (3<<8) | planea_wm;
1603 DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1605 I915_WRITE(FW_BLC, fwater_lo);
1608 #define ILK_LP0_PLANE_LATENCY 700
1609 #define ILK_LP0_CURSOR_LATENCY 1300
1612 * Check the wm result.
1614 * If any calculated watermark values is larger than the maximum value that
1615 * can be programmed into the associated watermark register, that watermark
1618 static bool ironlake_check_srwm(struct drm_device *dev, int level,
1619 int fbc_wm, int display_wm, int cursor_wm,
1620 const struct intel_watermark_params *display,
1621 const struct intel_watermark_params *cursor)
1623 struct drm_i915_private *dev_priv = dev->dev_private;
1625 DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d,"
1626 " cursor %d\n", level, display_wm, fbc_wm, cursor_wm);
1628 if (fbc_wm > SNB_FBC_MAX_SRWM) {
1629 DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n",
1630 fbc_wm, SNB_FBC_MAX_SRWM, level);
1632 /* fbc has it's own way to disable FBC WM */
1633 I915_WRITE(DISP_ARB_CTL,
1634 I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS);
1638 if (display_wm > display->max_wm) {
1639 DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n",
1640 display_wm, SNB_DISPLAY_MAX_SRWM, level);
1644 if (cursor_wm > cursor->max_wm) {
1645 DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n",
1646 cursor_wm, SNB_CURSOR_MAX_SRWM, level);
1650 if (!(fbc_wm || display_wm || cursor_wm)) {
1651 DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level);
1659 * Compute watermark values of WM[1-3],
1661 static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane,
1663 const struct intel_watermark_params *display,
1664 const struct intel_watermark_params *cursor,
1665 int *fbc_wm, int *display_wm, int *cursor_wm)
1667 struct drm_crtc *crtc;
1668 unsigned long line_time_us;
1669 int hdisplay, htotal, pixel_size, clock;
1670 int line_count, line_size;
1675 *fbc_wm = *display_wm = *cursor_wm = 0;
1679 crtc = intel_get_crtc_for_plane(dev, plane);
1680 hdisplay = crtc->mode.hdisplay;
1681 htotal = crtc->mode.htotal;
1682 clock = crtc->mode.clock;
1683 pixel_size = crtc->fb->bits_per_pixel / 8;
1685 line_time_us = (htotal * 1000) / clock;
1686 line_count = (latency_ns / line_time_us + 1000) / 1000;
1687 line_size = hdisplay * pixel_size;
1689 /* Use the minimum of the small and large buffer method for primary */
1690 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1691 large = line_count * line_size;
1693 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1694 *display_wm = entries + display->guard_size;
1698 * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2
1700 *fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2;
1702 /* calculate the self-refresh watermark for display cursor */
1703 entries = line_count * pixel_size * 64;
1704 entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1705 *cursor_wm = entries + cursor->guard_size;
1707 return ironlake_check_srwm(dev, level,
1708 *fbc_wm, *display_wm, *cursor_wm,
1712 static void ironlake_update_wm(struct drm_device *dev)
1714 struct drm_i915_private *dev_priv = dev->dev_private;
1715 int fbc_wm, plane_wm, cursor_wm;
1716 unsigned int enabled;
1719 if (g4x_compute_wm0(dev, 0,
1720 &ironlake_display_wm_info,
1721 ILK_LP0_PLANE_LATENCY,
1722 &ironlake_cursor_wm_info,
1723 ILK_LP0_CURSOR_LATENCY,
1724 &plane_wm, &cursor_wm)) {
1725 I915_WRITE(WM0_PIPEA_ILK,
1726 (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1727 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1728 " plane %d, " "cursor: %d\n",
1729 plane_wm, cursor_wm);
1733 if (g4x_compute_wm0(dev, 1,
1734 &ironlake_display_wm_info,
1735 ILK_LP0_PLANE_LATENCY,
1736 &ironlake_cursor_wm_info,
1737 ILK_LP0_CURSOR_LATENCY,
1738 &plane_wm, &cursor_wm)) {
1739 I915_WRITE(WM0_PIPEB_ILK,
1740 (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1741 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1742 " plane %d, cursor: %d\n",
1743 plane_wm, cursor_wm);
1748 * Calculate and update the self-refresh watermark only when one
1749 * display plane is used.
1751 I915_WRITE(WM3_LP_ILK, 0);
1752 I915_WRITE(WM2_LP_ILK, 0);
1753 I915_WRITE(WM1_LP_ILK, 0);
1755 if (!single_plane_enabled(enabled))
1757 enabled = ffs(enabled) - 1;
1760 if (!ironlake_compute_srwm(dev, 1, enabled,
1761 ILK_READ_WM1_LATENCY() * 500,
1762 &ironlake_display_srwm_info,
1763 &ironlake_cursor_srwm_info,
1764 &fbc_wm, &plane_wm, &cursor_wm))
1767 I915_WRITE(WM1_LP_ILK,
1769 (ILK_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1770 (fbc_wm << WM1_LP_FBC_SHIFT) |
1771 (plane_wm << WM1_LP_SR_SHIFT) |
1775 if (!ironlake_compute_srwm(dev, 2, enabled,
1776 ILK_READ_WM2_LATENCY() * 500,
1777 &ironlake_display_srwm_info,
1778 &ironlake_cursor_srwm_info,
1779 &fbc_wm, &plane_wm, &cursor_wm))
1782 I915_WRITE(WM2_LP_ILK,
1784 (ILK_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1785 (fbc_wm << WM1_LP_FBC_SHIFT) |
1786 (plane_wm << WM1_LP_SR_SHIFT) |
1790 * WM3 is unsupported on ILK, probably because we don't have latency
1791 * data for that power state
1795 static void sandybridge_update_wm(struct drm_device *dev)
1797 struct drm_i915_private *dev_priv = dev->dev_private;
1798 int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
1800 int fbc_wm, plane_wm, cursor_wm;
1801 unsigned int enabled;
1804 if (g4x_compute_wm0(dev, 0,
1805 &sandybridge_display_wm_info, latency,
1806 &sandybridge_cursor_wm_info, latency,
1807 &plane_wm, &cursor_wm)) {
1808 val = I915_READ(WM0_PIPEA_ILK);
1809 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1810 I915_WRITE(WM0_PIPEA_ILK, val |
1811 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1812 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1813 " plane %d, " "cursor: %d\n",
1814 plane_wm, cursor_wm);
1818 if (g4x_compute_wm0(dev, 1,
1819 &sandybridge_display_wm_info, latency,
1820 &sandybridge_cursor_wm_info, latency,
1821 &plane_wm, &cursor_wm)) {
1822 val = I915_READ(WM0_PIPEB_ILK);
1823 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1824 I915_WRITE(WM0_PIPEB_ILK, val |
1825 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1826 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1827 " plane %d, cursor: %d\n",
1828 plane_wm, cursor_wm);
1833 * Calculate and update the self-refresh watermark only when one
1834 * display plane is used.
1836 * SNB support 3 levels of watermark.
1838 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1839 * and disabled in the descending order
1842 I915_WRITE(WM3_LP_ILK, 0);
1843 I915_WRITE(WM2_LP_ILK, 0);
1844 I915_WRITE(WM1_LP_ILK, 0);
1846 if (!single_plane_enabled(enabled) ||
1847 dev_priv->sprite_scaling_enabled)
1849 enabled = ffs(enabled) - 1;
1852 if (!ironlake_compute_srwm(dev, 1, enabled,
1853 SNB_READ_WM1_LATENCY() * 500,
1854 &sandybridge_display_srwm_info,
1855 &sandybridge_cursor_srwm_info,
1856 &fbc_wm, &plane_wm, &cursor_wm))
1859 I915_WRITE(WM1_LP_ILK,
1861 (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1862 (fbc_wm << WM1_LP_FBC_SHIFT) |
1863 (plane_wm << WM1_LP_SR_SHIFT) |
1867 if (!ironlake_compute_srwm(dev, 2, enabled,
1868 SNB_READ_WM2_LATENCY() * 500,
1869 &sandybridge_display_srwm_info,
1870 &sandybridge_cursor_srwm_info,
1871 &fbc_wm, &plane_wm, &cursor_wm))
1874 I915_WRITE(WM2_LP_ILK,
1876 (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1877 (fbc_wm << WM1_LP_FBC_SHIFT) |
1878 (plane_wm << WM1_LP_SR_SHIFT) |
1882 if (!ironlake_compute_srwm(dev, 3, enabled,
1883 SNB_READ_WM3_LATENCY() * 500,
1884 &sandybridge_display_srwm_info,
1885 &sandybridge_cursor_srwm_info,
1886 &fbc_wm, &plane_wm, &cursor_wm))
1889 I915_WRITE(WM3_LP_ILK,
1891 (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1892 (fbc_wm << WM1_LP_FBC_SHIFT) |
1893 (plane_wm << WM1_LP_SR_SHIFT) |
1897 static void ivybridge_update_wm(struct drm_device *dev)
1899 struct drm_i915_private *dev_priv = dev->dev_private;
1900 int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
1902 int fbc_wm, plane_wm, cursor_wm;
1903 int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm;
1904 unsigned int enabled;
1907 if (g4x_compute_wm0(dev, 0,
1908 &sandybridge_display_wm_info, latency,
1909 &sandybridge_cursor_wm_info, latency,
1910 &plane_wm, &cursor_wm)) {
1911 val = I915_READ(WM0_PIPEA_ILK);
1912 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1913 I915_WRITE(WM0_PIPEA_ILK, val |
1914 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1915 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1916 " plane %d, " "cursor: %d\n",
1917 plane_wm, cursor_wm);
1921 if (g4x_compute_wm0(dev, 1,
1922 &sandybridge_display_wm_info, latency,
1923 &sandybridge_cursor_wm_info, latency,
1924 &plane_wm, &cursor_wm)) {
1925 val = I915_READ(WM0_PIPEB_ILK);
1926 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1927 I915_WRITE(WM0_PIPEB_ILK, val |
1928 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1929 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1930 " plane %d, cursor: %d\n",
1931 plane_wm, cursor_wm);
1935 if (g4x_compute_wm0(dev, 2,
1936 &sandybridge_display_wm_info, latency,
1937 &sandybridge_cursor_wm_info, latency,
1938 &plane_wm, &cursor_wm)) {
1939 val = I915_READ(WM0_PIPEC_IVB);
1940 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1941 I915_WRITE(WM0_PIPEC_IVB, val |
1942 ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1943 DRM_DEBUG_KMS("FIFO watermarks For pipe C -"
1944 " plane %d, cursor: %d\n",
1945 plane_wm, cursor_wm);
1950 * Calculate and update the self-refresh watermark only when one
1951 * display plane is used.
1953 * SNB support 3 levels of watermark.
1955 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1956 * and disabled in the descending order
1959 I915_WRITE(WM3_LP_ILK, 0);
1960 I915_WRITE(WM2_LP_ILK, 0);
1961 I915_WRITE(WM1_LP_ILK, 0);
1963 if (!single_plane_enabled(enabled) ||
1964 dev_priv->sprite_scaling_enabled)
1966 enabled = ffs(enabled) - 1;
1969 if (!ironlake_compute_srwm(dev, 1, enabled,
1970 SNB_READ_WM1_LATENCY() * 500,
1971 &sandybridge_display_srwm_info,
1972 &sandybridge_cursor_srwm_info,
1973 &fbc_wm, &plane_wm, &cursor_wm))
1976 I915_WRITE(WM1_LP_ILK,
1978 (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1979 (fbc_wm << WM1_LP_FBC_SHIFT) |
1980 (plane_wm << WM1_LP_SR_SHIFT) |
1984 if (!ironlake_compute_srwm(dev, 2, enabled,
1985 SNB_READ_WM2_LATENCY() * 500,
1986 &sandybridge_display_srwm_info,
1987 &sandybridge_cursor_srwm_info,
1988 &fbc_wm, &plane_wm, &cursor_wm))
1991 I915_WRITE(WM2_LP_ILK,
1993 (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1994 (fbc_wm << WM1_LP_FBC_SHIFT) |
1995 (plane_wm << WM1_LP_SR_SHIFT) |
1998 /* WM3, note we have to correct the cursor latency */
1999 if (!ironlake_compute_srwm(dev, 3, enabled,
2000 SNB_READ_WM3_LATENCY() * 500,
2001 &sandybridge_display_srwm_info,
2002 &sandybridge_cursor_srwm_info,
2003 &fbc_wm, &plane_wm, &ignore_cursor_wm) ||
2004 !ironlake_compute_srwm(dev, 3, enabled,
2005 2 * SNB_READ_WM3_LATENCY() * 500,
2006 &sandybridge_display_srwm_info,
2007 &sandybridge_cursor_srwm_info,
2008 &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm))
2011 I915_WRITE(WM3_LP_ILK,
2013 (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) |
2014 (fbc_wm << WM1_LP_FBC_SHIFT) |
2015 (plane_wm << WM1_LP_SR_SHIFT) |
2020 haswell_update_linetime_wm(struct drm_device *dev, int pipe,
2021 struct drm_display_mode *mode)
2023 struct drm_i915_private *dev_priv = dev->dev_private;
2026 temp = I915_READ(PIPE_WM_LINETIME(pipe));
2027 temp &= ~PIPE_WM_LINETIME_MASK;
2029 /* The WM are computed with base on how long it takes to fill a single
2030 * row at the given clock rate, multiplied by 8.
2032 temp |= PIPE_WM_LINETIME_TIME(
2033 ((mode->crtc_hdisplay * 1000) / mode->clock) * 8);
2035 /* IPS watermarks are only used by pipe A, and are ignored by
2036 * pipes B and C. They are calculated similarly to the common
2037 * linetime values, except that we are using CD clock frequency
2038 * in MHz instead of pixel rate for the division.
2040 * This is a placeholder for the IPS watermark calculation code.
2043 I915_WRITE(PIPE_WM_LINETIME(pipe), temp);
2047 sandybridge_compute_sprite_wm(struct drm_device *dev, int plane,
2048 uint32_t sprite_width, int pixel_size,
2049 const struct intel_watermark_params *display,
2050 int display_latency_ns, int *sprite_wm)
2052 struct drm_crtc *crtc;
2054 int entries, tlb_miss;
2056 crtc = intel_get_crtc_for_plane(dev, plane);
2057 if (!intel_crtc_active(crtc)) {
2058 *sprite_wm = display->guard_size;
2062 clock = crtc->mode.clock;
2064 /* Use the small buffer method to calculate the sprite watermark */
2065 entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
2066 tlb_miss = display->fifo_size*display->cacheline_size -
2069 entries += tlb_miss;
2070 entries = DIV_ROUND_UP(entries, display->cacheline_size);
2071 *sprite_wm = entries + display->guard_size;
2072 if (*sprite_wm > (int)display->max_wm)
2073 *sprite_wm = display->max_wm;
2079 sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane,
2080 uint32_t sprite_width, int pixel_size,
2081 const struct intel_watermark_params *display,
2082 int latency_ns, int *sprite_wm)
2084 struct drm_crtc *crtc;
2085 unsigned long line_time_us;
2087 int line_count, line_size;
2096 crtc = intel_get_crtc_for_plane(dev, plane);
2097 clock = crtc->mode.clock;
2103 line_time_us = (sprite_width * 1000) / clock;
2104 if (!line_time_us) {
2109 line_count = (latency_ns / line_time_us + 1000) / 1000;
2110 line_size = sprite_width * pixel_size;
2112 /* Use the minimum of the small and large buffer method for primary */
2113 small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
2114 large = line_count * line_size;
2116 entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
2117 *sprite_wm = entries + display->guard_size;
2119 return *sprite_wm > 0x3ff ? false : true;
2122 static void sandybridge_update_sprite_wm(struct drm_device *dev, int pipe,
2123 uint32_t sprite_width, int pixel_size)
2125 struct drm_i915_private *dev_priv = dev->dev_private;
2126 int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */
2133 reg = WM0_PIPEA_ILK;
2136 reg = WM0_PIPEB_ILK;
2139 reg = WM0_PIPEC_IVB;
2142 return; /* bad pipe */
2145 ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size,
2146 &sandybridge_display_wm_info,
2147 latency, &sprite_wm);
2149 DRM_DEBUG_KMS("failed to compute sprite wm for pipe %c\n",
2154 val = I915_READ(reg);
2155 val &= ~WM0_PIPE_SPRITE_MASK;
2156 I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT));
2157 DRM_DEBUG_KMS("sprite watermarks For pipe %c - %d\n", pipe_name(pipe), sprite_wm);
2160 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2162 &sandybridge_display_srwm_info,
2163 SNB_READ_WM1_LATENCY() * 500,
2166 DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %c\n",
2170 I915_WRITE(WM1S_LP_ILK, sprite_wm);
2172 /* Only IVB has two more LP watermarks for sprite */
2173 if (!IS_IVYBRIDGE(dev))
2176 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2178 &sandybridge_display_srwm_info,
2179 SNB_READ_WM2_LATENCY() * 500,
2182 DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %c\n",
2186 I915_WRITE(WM2S_LP_IVB, sprite_wm);
2188 ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2190 &sandybridge_display_srwm_info,
2191 SNB_READ_WM3_LATENCY() * 500,
2194 DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %c\n",
2198 I915_WRITE(WM3S_LP_IVB, sprite_wm);
2202 * intel_update_watermarks - update FIFO watermark values based on current modes
2204 * Calculate watermark values for the various WM regs based on current mode
2205 * and plane configuration.
2207 * There are several cases to deal with here:
2208 * - normal (i.e. non-self-refresh)
2209 * - self-refresh (SR) mode
2210 * - lines are large relative to FIFO size (buffer can hold up to 2)
2211 * - lines are small relative to FIFO size (buffer can hold more than 2
2212 * lines), so need to account for TLB latency
2214 * The normal calculation is:
2215 * watermark = dotclock * bytes per pixel * latency
2216 * where latency is platform & configuration dependent (we assume pessimal
2219 * The SR calculation is:
2220 * watermark = (trunc(latency/line time)+1) * surface width *
2223 * line time = htotal / dotclock
2224 * surface width = hdisplay for normal plane and 64 for cursor
2225 * and latency is assumed to be high, as above.
2227 * The final value programmed to the register should always be rounded up,
2228 * and include an extra 2 entries to account for clock crossings.
2230 * We don't use the sprite, so we can ignore that. And on Crestline we have
2231 * to set the non-SR watermarks to 8.
2233 void intel_update_watermarks(struct drm_device *dev)
2235 struct drm_i915_private *dev_priv = dev->dev_private;
2237 if (dev_priv->display.update_wm)
2238 dev_priv->display.update_wm(dev);
2241 void intel_update_linetime_watermarks(struct drm_device *dev,
2242 int pipe, struct drm_display_mode *mode)
2244 struct drm_i915_private *dev_priv = dev->dev_private;
2246 if (dev_priv->display.update_linetime_wm)
2247 dev_priv->display.update_linetime_wm(dev, pipe, mode);
2250 void intel_update_sprite_watermarks(struct drm_device *dev, int pipe,
2251 uint32_t sprite_width, int pixel_size)
2253 struct drm_i915_private *dev_priv = dev->dev_private;
2255 if (dev_priv->display.update_sprite_wm)
2256 dev_priv->display.update_sprite_wm(dev, pipe, sprite_width,
2260 static struct drm_i915_gem_object *
2261 intel_alloc_context_page(struct drm_device *dev)
2263 struct drm_i915_gem_object *ctx;
2266 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2268 ctx = i915_gem_alloc_object(dev, 4096);
2270 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2274 ret = i915_gem_object_pin(ctx, 4096, true, false);
2276 DRM_ERROR("failed to pin power context: %d\n", ret);
2280 ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
2282 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
2289 i915_gem_object_unpin(ctx);
2291 drm_gem_object_unreference(&ctx->base);
2296 * Lock protecting IPS related data structures
2298 DEFINE_SPINLOCK(mchdev_lock);
2300 /* Global for IPS driver to get at the current i915 device. Protected by
2302 static struct drm_i915_private *i915_mch_dev;
2304 bool ironlake_set_drps(struct drm_device *dev, u8 val)
2306 struct drm_i915_private *dev_priv = dev->dev_private;
2309 assert_spin_locked(&mchdev_lock);
2311 rgvswctl = I915_READ16(MEMSWCTL);
2312 if (rgvswctl & MEMCTL_CMD_STS) {
2313 DRM_DEBUG("gpu busy, RCS change rejected\n");
2314 return false; /* still busy with another command */
2317 rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
2318 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
2319 I915_WRITE16(MEMSWCTL, rgvswctl);
2320 POSTING_READ16(MEMSWCTL);
2322 rgvswctl |= MEMCTL_CMD_STS;
2323 I915_WRITE16(MEMSWCTL, rgvswctl);
2328 static void ironlake_enable_drps(struct drm_device *dev)
2330 struct drm_i915_private *dev_priv = dev->dev_private;
2331 u32 rgvmodectl = I915_READ(MEMMODECTL);
2332 u8 fmax, fmin, fstart, vstart;
2334 spin_lock_irq(&mchdev_lock);
2336 /* Enable temp reporting */
2337 I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
2338 I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
2340 /* 100ms RC evaluation intervals */
2341 I915_WRITE(RCUPEI, 100000);
2342 I915_WRITE(RCDNEI, 100000);
2344 /* Set max/min thresholds to 90ms and 80ms respectively */
2345 I915_WRITE(RCBMAXAVG, 90000);
2346 I915_WRITE(RCBMINAVG, 80000);
2348 I915_WRITE(MEMIHYST, 1);
2350 /* Set up min, max, and cur for interrupt handling */
2351 fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
2352 fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
2353 fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
2354 MEMMODE_FSTART_SHIFT;
2356 vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
2359 dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
2360 dev_priv->ips.fstart = fstart;
2362 dev_priv->ips.max_delay = fstart;
2363 dev_priv->ips.min_delay = fmin;
2364 dev_priv->ips.cur_delay = fstart;
2366 DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2367 fmax, fmin, fstart);
2369 I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
2372 * Interrupts will be enabled in ironlake_irq_postinstall
2375 I915_WRITE(VIDSTART, vstart);
2376 POSTING_READ(VIDSTART);
2378 rgvmodectl |= MEMMODE_SWMODE_EN;
2379 I915_WRITE(MEMMODECTL, rgvmodectl);
2381 if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2382 DRM_ERROR("stuck trying to change perf mode\n");
2385 ironlake_set_drps(dev, fstart);
2387 dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
2389 dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
2390 dev_priv->ips.last_count2 = I915_READ(0x112f4);
2391 getrawmonotonic(&dev_priv->ips.last_time2);
2393 spin_unlock_irq(&mchdev_lock);
2396 static void ironlake_disable_drps(struct drm_device *dev)
2398 struct drm_i915_private *dev_priv = dev->dev_private;
2401 spin_lock_irq(&mchdev_lock);
2403 rgvswctl = I915_READ16(MEMSWCTL);
2405 /* Ack interrupts, disable EFC interrupt */
2406 I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
2407 I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
2408 I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
2409 I915_WRITE(DEIIR, DE_PCU_EVENT);
2410 I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
2412 /* Go back to the starting frequency */
2413 ironlake_set_drps(dev, dev_priv->ips.fstart);
2415 rgvswctl |= MEMCTL_CMD_STS;
2416 I915_WRITE(MEMSWCTL, rgvswctl);
2419 spin_unlock_irq(&mchdev_lock);
2422 /* There's a funny hw issue where the hw returns all 0 when reading from
2423 * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
2424 * ourselves, instead of doing a rmw cycle (which might result in us clearing
2425 * all limits and the gpu stuck at whatever frequency it is at atm).
2427 static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 *val)
2433 if (*val >= dev_priv->rps.max_delay)
2434 *val = dev_priv->rps.max_delay;
2435 limits |= dev_priv->rps.max_delay << 24;
2437 /* Only set the down limit when we've reached the lowest level to avoid
2438 * getting more interrupts, otherwise leave this clear. This prevents a
2439 * race in the hw when coming out of rc6: There's a tiny window where
2440 * the hw runs at the minimal clock before selecting the desired
2441 * frequency, if the down threshold expires in that window we will not
2442 * receive a down interrupt. */
2443 if (*val <= dev_priv->rps.min_delay) {
2444 *val = dev_priv->rps.min_delay;
2445 limits |= dev_priv->rps.min_delay << 16;
2451 void gen6_set_rps(struct drm_device *dev, u8 val)
2453 struct drm_i915_private *dev_priv = dev->dev_private;
2454 u32 limits = gen6_rps_limits(dev_priv, &val);
2456 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2457 WARN_ON(val > dev_priv->rps.max_delay);
2458 WARN_ON(val < dev_priv->rps.min_delay);
2460 if (val == dev_priv->rps.cur_delay)
2463 if (IS_HASWELL(dev))
2464 I915_WRITE(GEN6_RPNSWREQ,
2465 HSW_FREQUENCY(val));
2467 I915_WRITE(GEN6_RPNSWREQ,
2468 GEN6_FREQUENCY(val) |
2470 GEN6_AGGRESSIVE_TURBO);
2472 /* Make sure we continue to get interrupts
2473 * until we hit the minimum or maximum frequencies.
2475 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
2477 POSTING_READ(GEN6_RPNSWREQ);
2479 dev_priv->rps.cur_delay = val;
2481 trace_intel_gpu_freq_change(val * 50);
2484 void valleyview_set_rps(struct drm_device *dev, u8 val)
2486 struct drm_i915_private *dev_priv = dev->dev_private;
2487 unsigned long timeout = jiffies + msecs_to_jiffies(10);
2488 u32 limits = gen6_rps_limits(dev_priv, &val);
2491 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2492 WARN_ON(val > dev_priv->rps.max_delay);
2493 WARN_ON(val < dev_priv->rps.min_delay);
2495 DRM_DEBUG_DRIVER("gpu freq request from %d to %d\n",
2496 vlv_gpu_freq(dev_priv->mem_freq,
2497 dev_priv->rps.cur_delay),
2498 vlv_gpu_freq(dev_priv->mem_freq, val));
2500 if (val == dev_priv->rps.cur_delay)
2503 valleyview_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
2506 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS, &pval);
2507 if (time_after(jiffies, timeout)) {
2508 DRM_DEBUG_DRIVER("timed out waiting for Punit\n");
2514 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS, &pval);
2515 if ((pval >> 8) != val)
2516 DRM_DEBUG_DRIVER("punit overrode freq: %d requested, but got %d\n",
2519 /* Make sure we continue to get interrupts
2520 * until we hit the minimum or maximum frequencies.
2522 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
2524 dev_priv->rps.cur_delay = pval >> 8;
2526 trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv->mem_freq, val));
2530 static void gen6_disable_rps(struct drm_device *dev)
2532 struct drm_i915_private *dev_priv = dev->dev_private;
2534 I915_WRITE(GEN6_RC_CONTROL, 0);
2535 I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
2536 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
2537 I915_WRITE(GEN6_PMIER, 0);
2538 /* Complete PM interrupt masking here doesn't race with the rps work
2539 * item again unmasking PM interrupts because that is using a different
2540 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
2541 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
2543 spin_lock_irq(&dev_priv->rps.lock);
2544 dev_priv->rps.pm_iir = 0;
2545 spin_unlock_irq(&dev_priv->rps.lock);
2547 I915_WRITE(GEN6_PMIIR, I915_READ(GEN6_PMIIR));
2550 int intel_enable_rc6(const struct drm_device *dev)
2552 /* Respect the kernel parameter if it is set */
2553 if (i915_enable_rc6 >= 0)
2554 return i915_enable_rc6;
2556 /* Disable RC6 on Ironlake */
2557 if (INTEL_INFO(dev)->gen == 5)
2560 if (IS_HASWELL(dev)) {
2561 DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
2562 return INTEL_RC6_ENABLE;
2565 /* snb/ivb have more than one rc6 state. */
2566 if (INTEL_INFO(dev)->gen == 6) {
2567 DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
2568 return INTEL_RC6_ENABLE;
2571 DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n");
2572 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
2575 static void gen6_enable_rps(struct drm_device *dev)
2577 struct drm_i915_private *dev_priv = dev->dev_private;
2578 struct intel_ring_buffer *ring;
2581 u32 rc6vids, pcu_mbox, rc6_mask = 0;
2586 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2588 /* Here begins a magic sequence of register writes to enable
2589 * auto-downclocking.
2591 * Perhaps there might be some value in exposing these to
2594 I915_WRITE(GEN6_RC_STATE, 0);
2596 /* Clear the DBG now so we don't confuse earlier errors */
2597 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
2598 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
2599 I915_WRITE(GTFIFODBG, gtfifodbg);
2602 gen6_gt_force_wake_get(dev_priv);
2604 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
2605 gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
2607 /* In units of 50MHz */
2608 dev_priv->rps.hw_max = dev_priv->rps.max_delay = rp_state_cap & 0xff;
2609 dev_priv->rps.min_delay = (rp_state_cap & 0xff0000) >> 16;
2610 dev_priv->rps.cur_delay = 0;
2612 /* disable the counters and set deterministic thresholds */
2613 I915_WRITE(GEN6_RC_CONTROL, 0);
2615 I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
2616 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
2617 I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
2618 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
2619 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
2621 for_each_ring(ring, dev_priv, i)
2622 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2624 I915_WRITE(GEN6_RC_SLEEP, 0);
2625 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
2626 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
2627 I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
2628 I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
2630 /* Check if we are enabling RC6 */
2631 rc6_mode = intel_enable_rc6(dev_priv->dev);
2632 if (rc6_mode & INTEL_RC6_ENABLE)
2633 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
2635 /* We don't use those on Haswell */
2636 if (!IS_HASWELL(dev)) {
2637 if (rc6_mode & INTEL_RC6p_ENABLE)
2638 rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2640 if (rc6_mode & INTEL_RC6pp_ENABLE)
2641 rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
2644 DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
2645 (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
2646 (rc6_mask & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
2647 (rc6_mask & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
2649 I915_WRITE(GEN6_RC_CONTROL,
2651 GEN6_RC_CTL_EI_MODE(1) |
2652 GEN6_RC_CTL_HW_ENABLE);
2654 if (IS_HASWELL(dev)) {
2655 I915_WRITE(GEN6_RPNSWREQ,
2657 I915_WRITE(GEN6_RC_VIDEO_FREQ,
2660 I915_WRITE(GEN6_RPNSWREQ,
2661 GEN6_FREQUENCY(10) |
2663 GEN6_AGGRESSIVE_TURBO);
2664 I915_WRITE(GEN6_RC_VIDEO_FREQ,
2665 GEN6_FREQUENCY(12));
2668 I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
2669 I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
2670 dev_priv->rps.max_delay << 24 |
2671 dev_priv->rps.min_delay << 16);
2673 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
2674 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
2675 I915_WRITE(GEN6_RP_UP_EI, 66000);
2676 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
2678 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2679 I915_WRITE(GEN6_RP_CONTROL,
2680 GEN6_RP_MEDIA_TURBO |
2681 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2682 GEN6_RP_MEDIA_IS_GFX |
2684 GEN6_RP_UP_BUSY_AVG |
2685 (IS_HASWELL(dev) ? GEN7_RP_DOWN_IDLE_AVG : GEN6_RP_DOWN_IDLE_CONT));
2687 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
2688 if (!ret && (IS_GEN6(dev) || IS_IVYBRIDGE(dev))) {
2690 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
2691 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
2692 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
2693 (dev_priv->rps.max_delay & 0xff) * 50,
2694 (pcu_mbox & 0xff) * 50);
2695 dev_priv->rps.hw_max = pcu_mbox & 0xff;
2698 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
2701 gen6_set_rps(dev_priv->dev, (gt_perf_status & 0xff00) >> 8);
2703 /* requires MSI enabled */
2704 I915_WRITE(GEN6_PMIER, GEN6_PM_DEFERRED_EVENTS);
2705 spin_lock_irq(&dev_priv->rps.lock);
2706 WARN_ON(dev_priv->rps.pm_iir != 0);
2707 I915_WRITE(GEN6_PMIMR, 0);
2708 spin_unlock_irq(&dev_priv->rps.lock);
2709 /* enable all PM interrupts */
2710 I915_WRITE(GEN6_PMINTRMSK, 0);
2713 ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
2714 if (IS_GEN6(dev) && ret) {
2715 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
2716 } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
2717 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
2718 GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
2719 rc6vids &= 0xffff00;
2720 rc6vids |= GEN6_ENCODE_RC6_VID(450);
2721 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
2723 DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
2726 gen6_gt_force_wake_put(dev_priv);
2729 static void gen6_update_ring_freq(struct drm_device *dev)
2731 struct drm_i915_private *dev_priv = dev->dev_private;
2733 unsigned int gpu_freq;
2734 unsigned int max_ia_freq, min_ring_freq;
2735 int scaling_factor = 180;
2737 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2739 max_ia_freq = cpufreq_quick_get_max(0);
2741 * Default to measured freq if none found, PCU will ensure we don't go
2745 max_ia_freq = tsc_khz;
2747 /* Convert from kHz to MHz */
2748 max_ia_freq /= 1000;
2750 min_ring_freq = I915_READ(MCHBAR_MIRROR_BASE_SNB + DCLK);
2751 /* convert DDR frequency from units of 133.3MHz to bandwidth */
2752 min_ring_freq = (2 * 4 * min_ring_freq + 2) / 3;
2755 * For each potential GPU frequency, load a ring frequency we'd like
2756 * to use for memory access. We do this by specifying the IA frequency
2757 * the PCU should use as a reference to determine the ring frequency.
2759 for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay;
2761 int diff = dev_priv->rps.max_delay - gpu_freq;
2762 unsigned int ia_freq = 0, ring_freq = 0;
2764 if (IS_HASWELL(dev)) {
2765 ring_freq = (gpu_freq * 5 + 3) / 4;
2766 ring_freq = max(min_ring_freq, ring_freq);
2767 /* leave ia_freq as the default, chosen by cpufreq */
2769 /* On older processors, there is no separate ring
2770 * clock domain, so in order to boost the bandwidth
2771 * of the ring, we need to upclock the CPU (ia_freq).
2773 * For GPU frequencies less than 750MHz,
2774 * just use the lowest ring freq.
2776 if (gpu_freq < min_freq)
2779 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
2780 ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
2783 sandybridge_pcode_write(dev_priv,
2784 GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
2785 ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
2786 ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
2791 int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
2795 valleyview_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE, &val);
2797 rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
2799 rp0 = min_t(u32, rp0, 0xea);
2804 static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
2808 valleyview_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO, &val);
2809 rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
2810 valleyview_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI, &val);
2811 rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
2816 int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
2820 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_LFM, &val);
2825 static void vlv_rps_timer_work(struct work_struct *work)
2827 drm_i915_private_t *dev_priv = container_of(work, drm_i915_private_t,
2831 * Timer fired, we must be idle. Drop to min voltage state.
2832 * Note: we use RPe here since it should match the
2833 * Vmin we were shooting for. That should give us better
2834 * perf when we come back out of RC6 than if we used the
2835 * min freq available.
2837 mutex_lock(&dev_priv->rps.hw_lock);
2838 valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay);
2839 mutex_unlock(&dev_priv->rps.hw_lock);
2842 static void valleyview_enable_rps(struct drm_device *dev)
2844 struct drm_i915_private *dev_priv = dev->dev_private;
2845 struct intel_ring_buffer *ring;
2846 u32 gtfifodbg, val, rpe;
2849 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2851 if ((gtfifodbg = I915_READ(GTFIFODBG))) {
2852 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
2853 I915_WRITE(GTFIFODBG, gtfifodbg);
2856 gen6_gt_force_wake_get(dev_priv);
2858 I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
2859 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
2860 I915_WRITE(GEN6_RP_UP_EI, 66000);
2861 I915_WRITE(GEN6_RP_DOWN_EI, 350000);
2863 I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2865 I915_WRITE(GEN6_RP_CONTROL,
2866 GEN6_RP_MEDIA_TURBO |
2867 GEN6_RP_MEDIA_HW_NORMAL_MODE |
2868 GEN6_RP_MEDIA_IS_GFX |
2870 GEN6_RP_UP_BUSY_AVG |
2871 GEN6_RP_DOWN_IDLE_CONT);
2873 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
2874 I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
2875 I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
2877 for_each_ring(ring, dev_priv, i)
2878 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2880 I915_WRITE(GEN6_RC6_THRESHOLD, 0xc350);
2882 /* allows RC6 residency counter to work */
2883 I915_WRITE(0x138104, _MASKED_BIT_ENABLE(0x3));
2884 I915_WRITE(GEN6_RC_CONTROL,
2885 GEN7_RC_CTL_TO_MODE);
2887 valleyview_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS, &val);
2888 dev_priv->mem_freq = 800 + (266 * (val >> 6) & 3);
2889 DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
2891 DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
2892 DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
2894 DRM_DEBUG_DRIVER("current GPU freq: %d\n",
2895 vlv_gpu_freq(dev_priv->mem_freq, (val >> 8) & 0xff));
2896 dev_priv->rps.cur_delay = (val >> 8) & 0xff;
2898 dev_priv->rps.max_delay = valleyview_rps_max_freq(dev_priv);
2899 dev_priv->rps.hw_max = dev_priv->rps.max_delay;
2900 DRM_DEBUG_DRIVER("max GPU freq: %d\n", vlv_gpu_freq(dev_priv->mem_freq,
2901 dev_priv->rps.max_delay));
2903 rpe = valleyview_rps_rpe_freq(dev_priv);
2904 DRM_DEBUG_DRIVER("RPe GPU freq: %d\n",
2905 vlv_gpu_freq(dev_priv->mem_freq, rpe));
2906 dev_priv->rps.rpe_delay = rpe;
2908 val = valleyview_rps_min_freq(dev_priv);
2909 DRM_DEBUG_DRIVER("min GPU freq: %d\n", vlv_gpu_freq(dev_priv->mem_freq,
2911 dev_priv->rps.min_delay = val;
2913 DRM_DEBUG_DRIVER("setting GPU freq to %d\n",
2914 vlv_gpu_freq(dev_priv->mem_freq, rpe));
2916 INIT_DELAYED_WORK(&dev_priv->rps.vlv_work, vlv_rps_timer_work);
2918 valleyview_set_rps(dev_priv->dev, rpe);
2920 /* requires MSI enabled */
2921 I915_WRITE(GEN6_PMIER, GEN6_PM_DEFERRED_EVENTS);
2922 spin_lock_irq(&dev_priv->rps.lock);
2923 WARN_ON(dev_priv->rps.pm_iir != 0);
2924 I915_WRITE(GEN6_PMIMR, 0);
2925 spin_unlock_irq(&dev_priv->rps.lock);
2926 /* enable all PM interrupts */
2927 I915_WRITE(GEN6_PMINTRMSK, 0);
2929 gen6_gt_force_wake_put(dev_priv);
2932 void ironlake_teardown_rc6(struct drm_device *dev)
2934 struct drm_i915_private *dev_priv = dev->dev_private;
2936 if (dev_priv->ips.renderctx) {
2937 i915_gem_object_unpin(dev_priv->ips.renderctx);
2938 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
2939 dev_priv->ips.renderctx = NULL;
2942 if (dev_priv->ips.pwrctx) {
2943 i915_gem_object_unpin(dev_priv->ips.pwrctx);
2944 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
2945 dev_priv->ips.pwrctx = NULL;
2949 static void ironlake_disable_rc6(struct drm_device *dev)
2951 struct drm_i915_private *dev_priv = dev->dev_private;
2953 if (I915_READ(PWRCTXA)) {
2954 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
2955 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
2956 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
2959 I915_WRITE(PWRCTXA, 0);
2960 POSTING_READ(PWRCTXA);
2962 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
2963 POSTING_READ(RSTDBYCTL);
2967 static int ironlake_setup_rc6(struct drm_device *dev)
2969 struct drm_i915_private *dev_priv = dev->dev_private;
2971 if (dev_priv->ips.renderctx == NULL)
2972 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
2973 if (!dev_priv->ips.renderctx)
2976 if (dev_priv->ips.pwrctx == NULL)
2977 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
2978 if (!dev_priv->ips.pwrctx) {
2979 ironlake_teardown_rc6(dev);
2986 static void ironlake_enable_rc6(struct drm_device *dev)
2988 struct drm_i915_private *dev_priv = dev->dev_private;
2989 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
2990 bool was_interruptible;
2993 /* rc6 disabled by default due to repeated reports of hanging during
2996 if (!intel_enable_rc6(dev))
2999 WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3001 ret = ironlake_setup_rc6(dev);
3005 was_interruptible = dev_priv->mm.interruptible;
3006 dev_priv->mm.interruptible = false;
3009 * GPU can automatically power down the render unit if given a page
3012 ret = intel_ring_begin(ring, 6);
3014 ironlake_teardown_rc6(dev);
3015 dev_priv->mm.interruptible = was_interruptible;
3019 intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
3020 intel_ring_emit(ring, MI_SET_CONTEXT);
3021 intel_ring_emit(ring, dev_priv->ips.renderctx->gtt_offset |
3023 MI_SAVE_EXT_STATE_EN |
3024 MI_RESTORE_EXT_STATE_EN |
3025 MI_RESTORE_INHIBIT);
3026 intel_ring_emit(ring, MI_SUSPEND_FLUSH);
3027 intel_ring_emit(ring, MI_NOOP);
3028 intel_ring_emit(ring, MI_FLUSH);
3029 intel_ring_advance(ring);
3032 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
3033 * does an implicit flush, combined with MI_FLUSH above, it should be
3034 * safe to assume that renderctx is valid
3036 ret = intel_ring_idle(ring);
3037 dev_priv->mm.interruptible = was_interruptible;
3039 DRM_ERROR("failed to enable ironlake power savings\n");
3040 ironlake_teardown_rc6(dev);
3044 I915_WRITE(PWRCTXA, dev_priv->ips.pwrctx->gtt_offset | PWRCTX_EN);
3045 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
3048 static unsigned long intel_pxfreq(u32 vidfreq)
3051 int div = (vidfreq & 0x3f0000) >> 16;
3052 int post = (vidfreq & 0x3000) >> 12;
3053 int pre = (vidfreq & 0x7);
3058 freq = ((div * 133333) / ((1<<post) * pre));
3063 static const struct cparams {
3069 { 1, 1333, 301, 28664 },
3070 { 1, 1066, 294, 24460 },
3071 { 1, 800, 294, 25192 },
3072 { 0, 1333, 276, 27605 },
3073 { 0, 1066, 276, 27605 },
3074 { 0, 800, 231, 23784 },
3077 static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
3079 u64 total_count, diff, ret;
3080 u32 count1, count2, count3, m = 0, c = 0;
3081 unsigned long now = jiffies_to_msecs(jiffies), diff1;
3084 assert_spin_locked(&mchdev_lock);
3086 diff1 = now - dev_priv->ips.last_time1;
3088 /* Prevent division-by-zero if we are asking too fast.
3089 * Also, we don't get interesting results if we are polling
3090 * faster than once in 10ms, so just return the saved value
3094 return dev_priv->ips.chipset_power;
3096 count1 = I915_READ(DMIEC);
3097 count2 = I915_READ(DDREC);
3098 count3 = I915_READ(CSIEC);
3100 total_count = count1 + count2 + count3;
3102 /* FIXME: handle per-counter overflow */
3103 if (total_count < dev_priv->ips.last_count1) {
3104 diff = ~0UL - dev_priv->ips.last_count1;
3105 diff += total_count;
3107 diff = total_count - dev_priv->ips.last_count1;
3110 for (i = 0; i < ARRAY_SIZE(cparams); i++) {
3111 if (cparams[i].i == dev_priv->ips.c_m &&
3112 cparams[i].t == dev_priv->ips.r_t) {
3119 diff = div_u64(diff, diff1);
3120 ret = ((m * diff) + c);
3121 ret = div_u64(ret, 10);
3123 dev_priv->ips.last_count1 = total_count;
3124 dev_priv->ips.last_time1 = now;
3126 dev_priv->ips.chipset_power = ret;
3131 unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
3135 if (dev_priv->info->gen != 5)
3138 spin_lock_irq(&mchdev_lock);
3140 val = __i915_chipset_val(dev_priv);
3142 spin_unlock_irq(&mchdev_lock);
3147 unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
3149 unsigned long m, x, b;
3152 tsfs = I915_READ(TSFS);
3154 m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
3155 x = I915_READ8(TR1);
3157 b = tsfs & TSFS_INTR_MASK;
3159 return ((m * x) / 127) - b;
3162 static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
3164 static const struct v_table {
3165 u16 vd; /* in .1 mil */
3166 u16 vm; /* in .1 mil */
3297 if (dev_priv->info->is_mobile)
3298 return v_table[pxvid].vm;
3300 return v_table[pxvid].vd;
3303 static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
3305 struct timespec now, diff1;
3307 unsigned long diffms;
3310 assert_spin_locked(&mchdev_lock);
3312 getrawmonotonic(&now);
3313 diff1 = timespec_sub(now, dev_priv->ips.last_time2);
3315 /* Don't divide by 0 */
3316 diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
3320 count = I915_READ(GFXEC);
3322 if (count < dev_priv->ips.last_count2) {
3323 diff = ~0UL - dev_priv->ips.last_count2;
3326 diff = count - dev_priv->ips.last_count2;
3329 dev_priv->ips.last_count2 = count;
3330 dev_priv->ips.last_time2 = now;
3332 /* More magic constants... */
3334 diff = div_u64(diff, diffms * 10);
3335 dev_priv->ips.gfx_power = diff;
3338 void i915_update_gfx_val(struct drm_i915_private *dev_priv)
3340 if (dev_priv->info->gen != 5)
3343 spin_lock_irq(&mchdev_lock);
3345 __i915_update_gfx_val(dev_priv);
3347 spin_unlock_irq(&mchdev_lock);
3350 static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
3352 unsigned long t, corr, state1, corr2, state2;
3355 assert_spin_locked(&mchdev_lock);
3357 pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4));
3358 pxvid = (pxvid >> 24) & 0x7f;
3359 ext_v = pvid_to_extvid(dev_priv, pxvid);
3363 t = i915_mch_val(dev_priv);
3365 /* Revel in the empirically derived constants */
3367 /* Correction factor in 1/100000 units */
3369 corr = ((t * 2349) + 135940);
3371 corr = ((t * 964) + 29317);
3373 corr = ((t * 301) + 1004);
3375 corr = corr * ((150142 * state1) / 10000 - 78642);
3377 corr2 = (corr * dev_priv->ips.corr);
3379 state2 = (corr2 * state1) / 10000;
3380 state2 /= 100; /* convert to mW */
3382 __i915_update_gfx_val(dev_priv);
3384 return dev_priv->ips.gfx_power + state2;
3387 unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
3391 if (dev_priv->info->gen != 5)
3394 spin_lock_irq(&mchdev_lock);
3396 val = __i915_gfx_val(dev_priv);
3398 spin_unlock_irq(&mchdev_lock);
3404 * i915_read_mch_val - return value for IPS use
3406 * Calculate and return a value for the IPS driver to use when deciding whether
3407 * we have thermal and power headroom to increase CPU or GPU power budget.
3409 unsigned long i915_read_mch_val(void)
3411 struct drm_i915_private *dev_priv;
3412 unsigned long chipset_val, graphics_val, ret = 0;
3414 spin_lock_irq(&mchdev_lock);
3417 dev_priv = i915_mch_dev;
3419 chipset_val = __i915_chipset_val(dev_priv);
3420 graphics_val = __i915_gfx_val(dev_priv);
3422 ret = chipset_val + graphics_val;
3425 spin_unlock_irq(&mchdev_lock);
3429 EXPORT_SYMBOL_GPL(i915_read_mch_val);
3432 * i915_gpu_raise - raise GPU frequency limit
3434 * Raise the limit; IPS indicates we have thermal headroom.
3436 bool i915_gpu_raise(void)
3438 struct drm_i915_private *dev_priv;
3441 spin_lock_irq(&mchdev_lock);
3442 if (!i915_mch_dev) {
3446 dev_priv = i915_mch_dev;
3448 if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
3449 dev_priv->ips.max_delay--;
3452 spin_unlock_irq(&mchdev_lock);
3456 EXPORT_SYMBOL_GPL(i915_gpu_raise);
3459 * i915_gpu_lower - lower GPU frequency limit
3461 * IPS indicates we're close to a thermal limit, so throttle back the GPU
3462 * frequency maximum.
3464 bool i915_gpu_lower(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.min_delay)
3477 dev_priv->ips.max_delay++;
3480 spin_unlock_irq(&mchdev_lock);
3484 EXPORT_SYMBOL_GPL(i915_gpu_lower);
3487 * i915_gpu_busy - indicate GPU business to IPS
3489 * Tell the IPS driver whether or not the GPU is busy.
3491 bool i915_gpu_busy(void)
3493 struct drm_i915_private *dev_priv;
3494 struct intel_ring_buffer *ring;
3498 spin_lock_irq(&mchdev_lock);
3501 dev_priv = i915_mch_dev;
3503 for_each_ring(ring, dev_priv, i)
3504 ret |= !list_empty(&ring->request_list);
3507 spin_unlock_irq(&mchdev_lock);
3511 EXPORT_SYMBOL_GPL(i915_gpu_busy);
3514 * i915_gpu_turbo_disable - disable graphics turbo
3516 * Disable graphics turbo by resetting the max frequency and setting the
3517 * current frequency to the default.
3519 bool i915_gpu_turbo_disable(void)
3521 struct drm_i915_private *dev_priv;
3524 spin_lock_irq(&mchdev_lock);
3525 if (!i915_mch_dev) {
3529 dev_priv = i915_mch_dev;
3531 dev_priv->ips.max_delay = dev_priv->ips.fstart;
3533 if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
3537 spin_unlock_irq(&mchdev_lock);
3541 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
3544 * Tells the intel_ips driver that the i915 driver is now loaded, if
3545 * IPS got loaded first.
3547 * This awkward dance is so that neither module has to depend on the
3548 * other in order for IPS to do the appropriate communication of
3549 * GPU turbo limits to i915.
3552 ips_ping_for_i915_load(void)
3556 link = symbol_get(ips_link_to_i915_driver);
3559 symbol_put(ips_link_to_i915_driver);
3563 void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
3565 /* We only register the i915 ips part with intel-ips once everything is
3566 * set up, to avoid intel-ips sneaking in and reading bogus values. */
3567 spin_lock_irq(&mchdev_lock);
3568 i915_mch_dev = dev_priv;
3569 spin_unlock_irq(&mchdev_lock);
3571 ips_ping_for_i915_load();
3574 void intel_gpu_ips_teardown(void)
3576 spin_lock_irq(&mchdev_lock);
3577 i915_mch_dev = NULL;
3578 spin_unlock_irq(&mchdev_lock);
3580 static void intel_init_emon(struct drm_device *dev)
3582 struct drm_i915_private *dev_priv = dev->dev_private;
3587 /* Disable to program */
3591 /* Program energy weights for various events */
3592 I915_WRITE(SDEW, 0x15040d00);
3593 I915_WRITE(CSIEW0, 0x007f0000);
3594 I915_WRITE(CSIEW1, 0x1e220004);
3595 I915_WRITE(CSIEW2, 0x04000004);
3597 for (i = 0; i < 5; i++)
3598 I915_WRITE(PEW + (i * 4), 0);
3599 for (i = 0; i < 3; i++)
3600 I915_WRITE(DEW + (i * 4), 0);
3602 /* Program P-state weights to account for frequency power adjustment */
3603 for (i = 0; i < 16; i++) {
3604 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
3605 unsigned long freq = intel_pxfreq(pxvidfreq);
3606 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
3611 val *= (freq / 1000);
3613 val /= (127*127*900);
3615 DRM_ERROR("bad pxval: %ld\n", val);
3618 /* Render standby states get 0 weight */
3622 for (i = 0; i < 4; i++) {
3623 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
3624 (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
3625 I915_WRITE(PXW + (i * 4), val);
3628 /* Adjust magic regs to magic values (more experimental results) */
3629 I915_WRITE(OGW0, 0);
3630 I915_WRITE(OGW1, 0);
3631 I915_WRITE(EG0, 0x00007f00);
3632 I915_WRITE(EG1, 0x0000000e);
3633 I915_WRITE(EG2, 0x000e0000);
3634 I915_WRITE(EG3, 0x68000300);
3635 I915_WRITE(EG4, 0x42000000);
3636 I915_WRITE(EG5, 0x00140031);
3640 for (i = 0; i < 8; i++)
3641 I915_WRITE(PXWL + (i * 4), 0);
3643 /* Enable PMON + select events */
3644 I915_WRITE(ECR, 0x80000019);
3646 lcfuse = I915_READ(LCFUSE02);
3648 dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
3651 void intel_disable_gt_powersave(struct drm_device *dev)
3653 struct drm_i915_private *dev_priv = dev->dev_private;
3655 if (IS_IRONLAKE_M(dev)) {
3656 ironlake_disable_drps(dev);
3657 ironlake_disable_rc6(dev);
3658 } else if (INTEL_INFO(dev)->gen >= 6) {
3659 cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
3660 cancel_work_sync(&dev_priv->rps.work);
3661 if (IS_VALLEYVIEW(dev))
3662 cancel_delayed_work_sync(&dev_priv->rps.vlv_work);
3663 mutex_lock(&dev_priv->rps.hw_lock);
3664 gen6_disable_rps(dev);
3665 mutex_unlock(&dev_priv->rps.hw_lock);
3669 static void intel_gen6_powersave_work(struct work_struct *work)
3671 struct drm_i915_private *dev_priv =
3672 container_of(work, struct drm_i915_private,
3673 rps.delayed_resume_work.work);
3674 struct drm_device *dev = dev_priv->dev;
3676 mutex_lock(&dev_priv->rps.hw_lock);
3678 if (IS_VALLEYVIEW(dev)) {
3679 valleyview_enable_rps(dev);
3681 gen6_enable_rps(dev);
3682 gen6_update_ring_freq(dev);
3684 mutex_unlock(&dev_priv->rps.hw_lock);
3687 void intel_enable_gt_powersave(struct drm_device *dev)
3689 struct drm_i915_private *dev_priv = dev->dev_private;
3691 if (IS_IRONLAKE_M(dev)) {
3692 ironlake_enable_drps(dev);
3693 ironlake_enable_rc6(dev);
3694 intel_init_emon(dev);
3695 } else if (IS_GEN6(dev) || IS_GEN7(dev)) {
3697 * PCU communication is slow and this doesn't need to be
3698 * done at any specific time, so do this out of our fast path
3699 * to make resume and init faster.
3701 schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
3702 round_jiffies_up_relative(HZ));
3706 static void ibx_init_clock_gating(struct drm_device *dev)
3708 struct drm_i915_private *dev_priv = dev->dev_private;
3711 * On Ibex Peak and Cougar Point, we need to disable clock
3712 * gating for the panel power sequencer or it will fail to
3713 * start up when no ports are active.
3715 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
3718 static void ironlake_init_clock_gating(struct drm_device *dev)
3720 struct drm_i915_private *dev_priv = dev->dev_private;
3721 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
3723 /* Required for FBC */
3724 dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
3725 ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
3726 ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
3728 I915_WRITE(PCH_3DCGDIS0,
3729 MARIUNIT_CLOCK_GATE_DISABLE |
3730 SVSMUNIT_CLOCK_GATE_DISABLE);
3731 I915_WRITE(PCH_3DCGDIS1,
3732 VFMUNIT_CLOCK_GATE_DISABLE);
3735 * According to the spec the following bits should be set in
3736 * order to enable memory self-refresh
3737 * The bit 22/21 of 0x42004
3738 * The bit 5 of 0x42020
3739 * The bit 15 of 0x45000
3741 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3742 (I915_READ(ILK_DISPLAY_CHICKEN2) |
3743 ILK_DPARB_GATE | ILK_VSDPFD_FULL));
3744 dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
3745 I915_WRITE(DISP_ARB_CTL,
3746 (I915_READ(DISP_ARB_CTL) |
3748 I915_WRITE(WM3_LP_ILK, 0);
3749 I915_WRITE(WM2_LP_ILK, 0);
3750 I915_WRITE(WM1_LP_ILK, 0);
3753 * Based on the document from hardware guys the following bits
3754 * should be set unconditionally in order to enable FBC.
3755 * The bit 22 of 0x42000
3756 * The bit 22 of 0x42004
3757 * The bit 7,8,9 of 0x42020.
3759 if (IS_IRONLAKE_M(dev)) {
3760 I915_WRITE(ILK_DISPLAY_CHICKEN1,
3761 I915_READ(ILK_DISPLAY_CHICKEN1) |
3763 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3764 I915_READ(ILK_DISPLAY_CHICKEN2) |
3768 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
3770 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3771 I915_READ(ILK_DISPLAY_CHICKEN2) |
3772 ILK_ELPIN_409_SELECT);
3773 I915_WRITE(_3D_CHICKEN2,
3774 _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
3775 _3D_CHICKEN2_WM_READ_PIPELINED);
3777 /* WaDisableRenderCachePipelinedFlush */
3778 I915_WRITE(CACHE_MODE_0,
3779 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3781 ibx_init_clock_gating(dev);
3784 static void cpt_init_clock_gating(struct drm_device *dev)
3786 struct drm_i915_private *dev_priv = dev->dev_private;
3791 * On Ibex Peak and Cougar Point, we need to disable clock
3792 * gating for the panel power sequencer or it will fail to
3793 * start up when no ports are active.
3795 I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
3796 I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
3797 DPLS_EDP_PPS_FIX_DIS);
3798 /* The below fixes the weird display corruption, a few pixels shifted
3799 * downward, on (only) LVDS of some HP laptops with IVY.
3801 for_each_pipe(pipe) {
3802 val = I915_READ(TRANS_CHICKEN2(pipe));
3803 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
3804 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
3805 if (dev_priv->fdi_rx_polarity_inverted)
3806 val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
3807 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
3808 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
3809 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
3810 I915_WRITE(TRANS_CHICKEN2(pipe), val);
3812 /* WADP0ClockGatingDisable */
3813 for_each_pipe(pipe) {
3814 I915_WRITE(TRANS_CHICKEN1(pipe),
3815 TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
3819 static void gen6_check_mch_setup(struct drm_device *dev)
3821 struct drm_i915_private *dev_priv = dev->dev_private;
3824 tmp = I915_READ(MCH_SSKPD);
3825 if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
3826 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
3827 DRM_INFO("This can cause pipe underruns and display issues.\n");
3828 DRM_INFO("Please upgrade your BIOS to fix this.\n");
3832 static void gen6_init_clock_gating(struct drm_device *dev)
3834 struct drm_i915_private *dev_priv = dev->dev_private;
3836 uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
3838 I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
3840 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3841 I915_READ(ILK_DISPLAY_CHICKEN2) |
3842 ILK_ELPIN_409_SELECT);
3844 /* WaDisableHiZPlanesWhenMSAAEnabled */
3845 I915_WRITE(_3D_CHICKEN,
3846 _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
3848 /* WaSetupGtModeTdRowDispatch */
3849 if (IS_SNB_GT1(dev))
3850 I915_WRITE(GEN6_GT_MODE,
3851 _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
3853 I915_WRITE(WM3_LP_ILK, 0);
3854 I915_WRITE(WM2_LP_ILK, 0);
3855 I915_WRITE(WM1_LP_ILK, 0);
3857 I915_WRITE(CACHE_MODE_0,
3858 _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
3860 I915_WRITE(GEN6_UCGCTL1,
3861 I915_READ(GEN6_UCGCTL1) |
3862 GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
3863 GEN6_CSUNIT_CLOCK_GATE_DISABLE);
3865 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3866 * gating disable must be set. Failure to set it results in
3867 * flickering pixels due to Z write ordering failures after
3868 * some amount of runtime in the Mesa "fire" demo, and Unigine
3869 * Sanctuary and Tropics, and apparently anything else with
3870 * alpha test or pixel discard.
3872 * According to the spec, bit 11 (RCCUNIT) must also be set,
3873 * but we didn't debug actual testcases to find it out.
3875 * Also apply WaDisableVDSUnitClockGating and
3876 * WaDisableRCPBUnitClockGating.
3878 I915_WRITE(GEN6_UCGCTL2,
3879 GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
3880 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
3881 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
3883 /* Bspec says we need to always set all mask bits. */
3884 I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) |
3885 _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL);
3888 * According to the spec the following bits should be
3889 * set in order to enable memory self-refresh and fbc:
3890 * The bit21 and bit22 of 0x42000
3891 * The bit21 and bit22 of 0x42004
3892 * The bit5 and bit7 of 0x42020
3893 * The bit14 of 0x70180
3894 * The bit14 of 0x71180
3896 I915_WRITE(ILK_DISPLAY_CHICKEN1,
3897 I915_READ(ILK_DISPLAY_CHICKEN1) |
3898 ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
3899 I915_WRITE(ILK_DISPLAY_CHICKEN2,
3900 I915_READ(ILK_DISPLAY_CHICKEN2) |
3901 ILK_DPARB_GATE | ILK_VSDPFD_FULL);
3902 I915_WRITE(ILK_DSPCLK_GATE_D,
3903 I915_READ(ILK_DSPCLK_GATE_D) |
3904 ILK_DPARBUNIT_CLOCK_GATE_ENABLE |
3905 ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
3907 /* WaMbcDriverBootEnable */
3908 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
3909 GEN6_MBCTL_ENABLE_BOOT_FETCH);
3911 for_each_pipe(pipe) {
3912 I915_WRITE(DSPCNTR(pipe),
3913 I915_READ(DSPCNTR(pipe)) |
3914 DISPPLANE_TRICKLE_FEED_DISABLE);
3915 intel_flush_display_plane(dev_priv, pipe);
3918 /* The default value should be 0x200 according to docs, but the two
3919 * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */
3920 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff));
3921 I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI));
3923 cpt_init_clock_gating(dev);
3925 gen6_check_mch_setup(dev);
3928 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
3930 uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
3932 reg &= ~GEN7_FF_SCHED_MASK;
3933 reg |= GEN7_FF_TS_SCHED_HW;
3934 reg |= GEN7_FF_VS_SCHED_HW;
3935 reg |= GEN7_FF_DS_SCHED_HW;
3937 /* WaVSRefCountFullforceMissDisable */
3938 if (IS_HASWELL(dev_priv->dev))
3939 reg &= ~GEN7_FF_VS_REF_CNT_FFME;
3941 I915_WRITE(GEN7_FF_THREAD_MODE, reg);
3944 static void lpt_init_clock_gating(struct drm_device *dev)
3946 struct drm_i915_private *dev_priv = dev->dev_private;
3949 * TODO: this bit should only be enabled when really needed, then
3950 * disabled when not needed anymore in order to save power.
3952 if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
3953 I915_WRITE(SOUTH_DSPCLK_GATE_D,
3954 I915_READ(SOUTH_DSPCLK_GATE_D) |
3955 PCH_LP_PARTITION_LEVEL_DISABLE);
3958 static void haswell_init_clock_gating(struct drm_device *dev)
3960 struct drm_i915_private *dev_priv = dev->dev_private;
3963 I915_WRITE(WM3_LP_ILK, 0);
3964 I915_WRITE(WM2_LP_ILK, 0);
3965 I915_WRITE(WM1_LP_ILK, 0);
3967 /* According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3968 * This implements the WaDisableRCZUnitClockGating workaround.
3970 I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
3972 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3973 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
3974 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
3976 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3977 I915_WRITE(GEN7_L3CNTLREG1,
3978 GEN7_WA_FOR_GEN7_L3_CONTROL);
3979 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
3980 GEN7_WA_L3_CHICKEN_MODE);
3982 /* This is required by WaCatErrorRejectionIssue */
3983 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
3984 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
3985 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
3987 for_each_pipe(pipe) {
3988 I915_WRITE(DSPCNTR(pipe),
3989 I915_READ(DSPCNTR(pipe)) |
3990 DISPPLANE_TRICKLE_FEED_DISABLE);
3991 intel_flush_display_plane(dev_priv, pipe);
3994 gen7_setup_fixed_func_scheduler(dev_priv);
3996 /* WaDisable4x2SubspanOptimization */
3997 I915_WRITE(CACHE_MODE_1,
3998 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
4000 /* WaMbcDriverBootEnable */
4001 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
4002 GEN6_MBCTL_ENABLE_BOOT_FETCH);
4004 /* WaSwitchSolVfFArbitrationPriority */
4005 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
4007 /* XXX: This is a workaround for early silicon revisions and should be
4012 WM_DBG_DISALLOW_MULTIPLE_LP |
4013 WM_DBG_DISALLOW_SPRITE |
4014 WM_DBG_DISALLOW_MAXFIFO);
4016 lpt_init_clock_gating(dev);
4019 static void ivybridge_init_clock_gating(struct drm_device *dev)
4021 struct drm_i915_private *dev_priv = dev->dev_private;
4025 I915_WRITE(WM3_LP_ILK, 0);
4026 I915_WRITE(WM2_LP_ILK, 0);
4027 I915_WRITE(WM1_LP_ILK, 0);
4029 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
4031 /* WaDisableEarlyCull */
4032 I915_WRITE(_3D_CHICKEN3,
4033 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4035 /* WaDisableBackToBackFlipFix */
4036 I915_WRITE(IVB_CHICKEN3,
4037 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4038 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4040 /* WaDisablePSDDualDispatchEnable */
4041 if (IS_IVB_GT1(dev))
4042 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4043 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
4045 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2,
4046 _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
4048 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
4049 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4050 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4052 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
4053 I915_WRITE(GEN7_L3CNTLREG1,
4054 GEN7_WA_FOR_GEN7_L3_CONTROL);
4055 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
4056 GEN7_WA_L3_CHICKEN_MODE);
4057 if (IS_IVB_GT1(dev))
4058 I915_WRITE(GEN7_ROW_CHICKEN2,
4059 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
4061 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
4062 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
4065 /* WaForceL3Serialization */
4066 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4067 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4069 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4070 * gating disable must be set. Failure to set it results in
4071 * flickering pixels due to Z write ordering failures after
4072 * some amount of runtime in the Mesa "fire" demo, and Unigine
4073 * Sanctuary and Tropics, and apparently anything else with
4074 * alpha test or pixel discard.
4076 * According to the spec, bit 11 (RCCUNIT) must also be set,
4077 * but we didn't debug actual testcases to find it out.
4079 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
4080 * This implements the WaDisableRCZUnitClockGating workaround.
4082 I915_WRITE(GEN6_UCGCTL2,
4083 GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
4084 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4086 /* This is required by WaCatErrorRejectionIssue */
4087 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4088 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4089 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4091 for_each_pipe(pipe) {
4092 I915_WRITE(DSPCNTR(pipe),
4093 I915_READ(DSPCNTR(pipe)) |
4094 DISPPLANE_TRICKLE_FEED_DISABLE);
4095 intel_flush_display_plane(dev_priv, pipe);
4098 /* WaMbcDriverBootEnable */
4099 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
4100 GEN6_MBCTL_ENABLE_BOOT_FETCH);
4102 gen7_setup_fixed_func_scheduler(dev_priv);
4104 /* WaDisable4x2SubspanOptimization */
4105 I915_WRITE(CACHE_MODE_1,
4106 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
4108 snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
4109 snpcr &= ~GEN6_MBC_SNPCR_MASK;
4110 snpcr |= GEN6_MBC_SNPCR_MED;
4111 I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
4113 if (!HAS_PCH_NOP(dev))
4114 cpt_init_clock_gating(dev);
4116 gen6_check_mch_setup(dev);
4119 static void valleyview_init_clock_gating(struct drm_device *dev)
4121 struct drm_i915_private *dev_priv = dev->dev_private;
4124 I915_WRITE(WM3_LP_ILK, 0);
4125 I915_WRITE(WM2_LP_ILK, 0);
4126 I915_WRITE(WM1_LP_ILK, 0);
4128 I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
4130 /* WaDisableEarlyCull */
4131 I915_WRITE(_3D_CHICKEN3,
4132 _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
4134 /* WaDisableBackToBackFlipFix */
4135 I915_WRITE(IVB_CHICKEN3,
4136 CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
4137 CHICKEN3_DGMG_DONE_FIX_DISABLE);
4139 /* WaDisablePSDDualDispatchEnable */
4140 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
4141 _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
4142 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
4144 /* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
4145 I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
4146 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
4148 /* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
4149 I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS);
4150 I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE);
4152 /* WaForceL3Serialization */
4153 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4154 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4156 /* WaDisableDopClockGating */
4157 I915_WRITE(GEN7_ROW_CHICKEN2,
4158 _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
4160 /* WaForceL3Serialization */
4161 I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
4162 ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
4164 /* This is required by WaCatErrorRejectionIssue */
4165 I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
4166 I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
4167 GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
4169 /* WaMbcDriverBootEnable */
4170 I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
4171 GEN6_MBCTL_ENABLE_BOOT_FETCH);
4174 /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4175 * gating disable must be set. Failure to set it results in
4176 * flickering pixels due to Z write ordering failures after
4177 * some amount of runtime in the Mesa "fire" demo, and Unigine
4178 * Sanctuary and Tropics, and apparently anything else with
4179 * alpha test or pixel discard.
4181 * According to the spec, bit 11 (RCCUNIT) must also be set,
4182 * but we didn't debug actual testcases to find it out.
4184 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
4185 * This implements the WaDisableRCZUnitClockGating workaround.
4187 * Also apply WaDisableVDSUnitClockGating and
4188 * WaDisableRCPBUnitClockGating.
4190 I915_WRITE(GEN6_UCGCTL2,
4191 GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
4192 GEN7_TDLUNIT_CLOCK_GATE_DISABLE |
4193 GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
4194 GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
4195 GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4197 I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
4199 for_each_pipe(pipe) {
4200 I915_WRITE(DSPCNTR(pipe),
4201 I915_READ(DSPCNTR(pipe)) |
4202 DISPPLANE_TRICKLE_FEED_DISABLE);
4203 intel_flush_display_plane(dev_priv, pipe);
4206 I915_WRITE(CACHE_MODE_1,
4207 _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
4210 * WaDisableVLVClockGating_VBIIssue
4211 * Disable clock gating on th GCFG unit to prevent a delay
4212 * in the reporting of vblank events.
4214 I915_WRITE(VLV_GUNIT_CLOCK_GATE, 0xffffffff);
4216 /* Conservative clock gating settings for now */
4217 I915_WRITE(0x9400, 0xffffffff);
4218 I915_WRITE(0x9404, 0xffffffff);
4219 I915_WRITE(0x9408, 0xffffffff);
4220 I915_WRITE(0x940c, 0xffffffff);
4221 I915_WRITE(0x9410, 0xffffffff);
4222 I915_WRITE(0x9414, 0xffffffff);
4223 I915_WRITE(0x9418, 0xffffffff);
4226 static void g4x_init_clock_gating(struct drm_device *dev)
4228 struct drm_i915_private *dev_priv = dev->dev_private;
4229 uint32_t dspclk_gate;
4231 I915_WRITE(RENCLK_GATE_D1, 0);
4232 I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
4233 GS_UNIT_CLOCK_GATE_DISABLE |
4234 CL_UNIT_CLOCK_GATE_DISABLE);
4235 I915_WRITE(RAMCLK_GATE_D, 0);
4236 dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
4237 OVRUNIT_CLOCK_GATE_DISABLE |
4238 OVCUNIT_CLOCK_GATE_DISABLE;
4240 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
4241 I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4243 /* WaDisableRenderCachePipelinedFlush */
4244 I915_WRITE(CACHE_MODE_0,
4245 _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
4248 static void crestline_init_clock_gating(struct drm_device *dev)
4250 struct drm_i915_private *dev_priv = dev->dev_private;
4252 I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
4253 I915_WRITE(RENCLK_GATE_D2, 0);
4254 I915_WRITE(DSPCLK_GATE_D, 0);
4255 I915_WRITE(RAMCLK_GATE_D, 0);
4256 I915_WRITE16(DEUC, 0);
4259 static void broadwater_init_clock_gating(struct drm_device *dev)
4261 struct drm_i915_private *dev_priv = dev->dev_private;
4263 I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
4264 I965_RCC_CLOCK_GATE_DISABLE |
4265 I965_RCPB_CLOCK_GATE_DISABLE |
4266 I965_ISC_CLOCK_GATE_DISABLE |
4267 I965_FBC_CLOCK_GATE_DISABLE);
4268 I915_WRITE(RENCLK_GATE_D2, 0);
4271 static void gen3_init_clock_gating(struct drm_device *dev)
4273 struct drm_i915_private *dev_priv = dev->dev_private;
4274 u32 dstate = I915_READ(D_STATE);
4276 dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
4277 DSTATE_DOT_CLOCK_GATING;
4278 I915_WRITE(D_STATE, dstate);
4280 if (IS_PINEVIEW(dev))
4281 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
4283 /* IIR "flip pending" means done if this bit is set */
4284 I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
4287 static void i85x_init_clock_gating(struct drm_device *dev)
4289 struct drm_i915_private *dev_priv = dev->dev_private;
4291 I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
4294 static void i830_init_clock_gating(struct drm_device *dev)
4296 struct drm_i915_private *dev_priv = dev->dev_private;
4298 I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
4301 void intel_init_clock_gating(struct drm_device *dev)
4303 struct drm_i915_private *dev_priv = dev->dev_private;
4305 dev_priv->display.init_clock_gating(dev);
4309 * We should only use the power well if we explicitly asked the hardware to
4310 * enable it, so check if it's enabled and also check if we've requested it to
4313 bool intel_using_power_well(struct drm_device *dev)
4315 struct drm_i915_private *dev_priv = dev->dev_private;
4317 if (IS_HASWELL(dev))
4318 return I915_READ(HSW_PWR_WELL_DRIVER) ==
4319 (HSW_PWR_WELL_ENABLE | HSW_PWR_WELL_STATE);
4324 void intel_set_power_well(struct drm_device *dev, bool enable)
4326 struct drm_i915_private *dev_priv = dev->dev_private;
4327 bool is_enabled, enable_requested;
4330 if (!HAS_POWER_WELL(dev))
4333 if (!i915_disable_power_well && !enable)
4336 tmp = I915_READ(HSW_PWR_WELL_DRIVER);
4337 is_enabled = tmp & HSW_PWR_WELL_STATE;
4338 enable_requested = tmp & HSW_PWR_WELL_ENABLE;
4341 if (!enable_requested)
4342 I915_WRITE(HSW_PWR_WELL_DRIVER, HSW_PWR_WELL_ENABLE);
4345 DRM_DEBUG_KMS("Enabling power well\n");
4346 if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
4347 HSW_PWR_WELL_STATE), 20))
4348 DRM_ERROR("Timeout enabling power well\n");
4351 if (enable_requested) {
4352 I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
4353 DRM_DEBUG_KMS("Requesting to disable the power well\n");
4359 * Starting with Haswell, we have a "Power Down Well" that can be turned off
4360 * when not needed anymore. We have 4 registers that can request the power well
4361 * to be enabled, and it will only be disabled if none of the registers is
4362 * requesting it to be enabled.
4364 void intel_init_power_well(struct drm_device *dev)
4366 struct drm_i915_private *dev_priv = dev->dev_private;
4368 if (!HAS_POWER_WELL(dev))
4371 /* For now, we need the power well to be always enabled. */
4372 intel_set_power_well(dev, true);
4374 /* We're taking over the BIOS, so clear any requests made by it since
4375 * the driver is in charge now. */
4376 if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE)
4377 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
4380 /* Set up chip specific power management-related functions */
4381 void intel_init_pm(struct drm_device *dev)
4383 struct drm_i915_private *dev_priv = dev->dev_private;
4385 if (I915_HAS_FBC(dev)) {
4386 if (HAS_PCH_SPLIT(dev)) {
4387 dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
4388 dev_priv->display.enable_fbc = ironlake_enable_fbc;
4389 dev_priv->display.disable_fbc = ironlake_disable_fbc;
4390 } else if (IS_GM45(dev)) {
4391 dev_priv->display.fbc_enabled = g4x_fbc_enabled;
4392 dev_priv->display.enable_fbc = g4x_enable_fbc;
4393 dev_priv->display.disable_fbc = g4x_disable_fbc;
4394 } else if (IS_CRESTLINE(dev)) {
4395 dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
4396 dev_priv->display.enable_fbc = i8xx_enable_fbc;
4397 dev_priv->display.disable_fbc = i8xx_disable_fbc;
4399 /* 855GM needs testing */
4403 if (IS_PINEVIEW(dev))
4404 i915_pineview_get_mem_freq(dev);
4405 else if (IS_GEN5(dev))
4406 i915_ironlake_get_mem_freq(dev);
4408 /* For FIFO watermark updates */
4409 if (HAS_PCH_SPLIT(dev)) {
4411 if (I915_READ(MLTR_ILK) & ILK_SRLT_MASK)
4412 dev_priv->display.update_wm = ironlake_update_wm;
4414 DRM_DEBUG_KMS("Failed to get proper latency. "
4416 dev_priv->display.update_wm = NULL;
4418 dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
4419 } else if (IS_GEN6(dev)) {
4420 if (SNB_READ_WM0_LATENCY()) {
4421 dev_priv->display.update_wm = sandybridge_update_wm;
4422 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4424 DRM_DEBUG_KMS("Failed to read display plane latency. "
4426 dev_priv->display.update_wm = NULL;
4428 dev_priv->display.init_clock_gating = gen6_init_clock_gating;
4429 } else if (IS_IVYBRIDGE(dev)) {
4430 if (SNB_READ_WM0_LATENCY()) {
4431 dev_priv->display.update_wm = ivybridge_update_wm;
4432 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4434 DRM_DEBUG_KMS("Failed to read display plane latency. "
4436 dev_priv->display.update_wm = NULL;
4438 dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
4439 } else if (IS_HASWELL(dev)) {
4440 if (SNB_READ_WM0_LATENCY()) {
4441 dev_priv->display.update_wm = sandybridge_update_wm;
4442 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4443 dev_priv->display.update_linetime_wm = haswell_update_linetime_wm;
4445 DRM_DEBUG_KMS("Failed to read display plane latency. "
4447 dev_priv->display.update_wm = NULL;
4449 dev_priv->display.init_clock_gating = haswell_init_clock_gating;
4451 dev_priv->display.update_wm = NULL;
4452 } else if (IS_VALLEYVIEW(dev)) {
4453 dev_priv->display.update_wm = valleyview_update_wm;
4454 dev_priv->display.init_clock_gating =
4455 valleyview_init_clock_gating;
4456 } else if (IS_PINEVIEW(dev)) {
4457 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
4460 dev_priv->mem_freq)) {
4461 DRM_INFO("failed to find known CxSR latency "
4462 "(found ddr%s fsb freq %d, mem freq %d), "
4464 (dev_priv->is_ddr3 == 1) ? "3" : "2",
4465 dev_priv->fsb_freq, dev_priv->mem_freq);
4466 /* Disable CxSR and never update its watermark again */
4467 pineview_disable_cxsr(dev);
4468 dev_priv->display.update_wm = NULL;
4470 dev_priv->display.update_wm = pineview_update_wm;
4471 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
4472 } else if (IS_G4X(dev)) {
4473 dev_priv->display.update_wm = g4x_update_wm;
4474 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
4475 } else if (IS_GEN4(dev)) {
4476 dev_priv->display.update_wm = i965_update_wm;
4477 if (IS_CRESTLINE(dev))
4478 dev_priv->display.init_clock_gating = crestline_init_clock_gating;
4479 else if (IS_BROADWATER(dev))
4480 dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
4481 } else if (IS_GEN3(dev)) {
4482 dev_priv->display.update_wm = i9xx_update_wm;
4483 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
4484 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
4485 } else if (IS_I865G(dev)) {
4486 dev_priv->display.update_wm = i830_update_wm;
4487 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
4488 dev_priv->display.get_fifo_size = i830_get_fifo_size;
4489 } else if (IS_I85X(dev)) {
4490 dev_priv->display.update_wm = i9xx_update_wm;
4491 dev_priv->display.get_fifo_size = i85x_get_fifo_size;
4492 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
4494 dev_priv->display.update_wm = i830_update_wm;
4495 dev_priv->display.init_clock_gating = i830_init_clock_gating;
4497 dev_priv->display.get_fifo_size = i845_get_fifo_size;
4499 dev_priv->display.get_fifo_size = i830_get_fifo_size;
4503 static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv)
4505 u32 gt_thread_status_mask;
4507 if (IS_HASWELL(dev_priv->dev))
4508 gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK_HSW;
4510 gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK;
4512 /* w/a for a sporadic read returning 0 by waiting for the GT
4513 * thread to wake up.
4515 if (wait_for_atomic_us((I915_READ_NOTRACE(GEN6_GT_THREAD_STATUS_REG) & gt_thread_status_mask) == 0, 500))
4516 DRM_ERROR("GT thread status wait timed out\n");
4519 static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv)
4521 I915_WRITE_NOTRACE(FORCEWAKE, 0);
4522 POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */
4525 static void __gen6_gt_force_wake_get(struct drm_i915_private *dev_priv)
4527 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1) == 0,
4528 FORCEWAKE_ACK_TIMEOUT_MS))
4529 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4531 I915_WRITE_NOTRACE(FORCEWAKE, 1);
4532 POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */
4534 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1),
4535 FORCEWAKE_ACK_TIMEOUT_MS))
4536 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4538 __gen6_gt_wait_for_thread_c0(dev_priv);
4541 static void __gen6_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv)
4543 I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(0xffff));
4544 /* something from same cacheline, but !FORCEWAKE_MT */
4545 POSTING_READ(ECOBUS);
4548 static void __gen6_gt_force_wake_mt_get(struct drm_i915_private *dev_priv)
4552 if (IS_HASWELL(dev_priv->dev))
4553 forcewake_ack = FORCEWAKE_ACK_HSW;
4555 forcewake_ack = FORCEWAKE_MT_ACK;
4557 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL) == 0,
4558 FORCEWAKE_ACK_TIMEOUT_MS))
4559 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4561 I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4562 /* something from same cacheline, but !FORCEWAKE_MT */
4563 POSTING_READ(ECOBUS);
4565 if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL),
4566 FORCEWAKE_ACK_TIMEOUT_MS))
4567 DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4569 __gen6_gt_wait_for_thread_c0(dev_priv);
4573 * Generally this is called implicitly by the register read function. However,
4574 * if some sequence requires the GT to not power down then this function should
4575 * be called at the beginning of the sequence followed by a call to
4576 * gen6_gt_force_wake_put() at the end of the sequence.
4578 void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv)
4580 unsigned long irqflags;
4582 spin_lock_irqsave(&dev_priv->gt_lock, irqflags);
4583 if (dev_priv->forcewake_count++ == 0)
4584 dev_priv->gt.force_wake_get(dev_priv);
4585 spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags);
4588 void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv)
4591 gtfifodbg = I915_READ_NOTRACE(GTFIFODBG);
4592 if (WARN(gtfifodbg & GT_FIFO_CPU_ERROR_MASK,
4593 "MMIO read or write has been dropped %x\n", gtfifodbg))
4594 I915_WRITE_NOTRACE(GTFIFODBG, GT_FIFO_CPU_ERROR_MASK);
4597 static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv)
4599 I915_WRITE_NOTRACE(FORCEWAKE, 0);
4600 /* something from same cacheline, but !FORCEWAKE */
4601 POSTING_READ(ECOBUS);
4602 gen6_gt_check_fifodbg(dev_priv);
4605 static void __gen6_gt_force_wake_mt_put(struct drm_i915_private *dev_priv)
4607 I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4608 /* something from same cacheline, but !FORCEWAKE_MT */
4609 POSTING_READ(ECOBUS);
4610 gen6_gt_check_fifodbg(dev_priv);
4614 * see gen6_gt_force_wake_get()
4616 void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv)
4618 unsigned long irqflags;
4620 spin_lock_irqsave(&dev_priv->gt_lock, irqflags);
4621 if (--dev_priv->forcewake_count == 0)
4622 dev_priv->gt.force_wake_put(dev_priv);
4623 spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags);
4626 int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv)
4630 if (dev_priv->gt_fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) {
4632 u32 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES);
4633 while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) {
4635 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES);
4637 if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES))
4639 dev_priv->gt_fifo_count = fifo;
4641 dev_priv->gt_fifo_count--;
4646 static void vlv_force_wake_reset(struct drm_i915_private *dev_priv)
4648 I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(0xffff));
4649 /* something from same cacheline, but !FORCEWAKE_VLV */
4650 POSTING_READ(FORCEWAKE_ACK_VLV);
4653 static void vlv_force_wake_get(struct drm_i915_private *dev_priv)
4655 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL) == 0,
4656 FORCEWAKE_ACK_TIMEOUT_MS))
4657 DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4659 I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4660 I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV,
4661 _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4663 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL),
4664 FORCEWAKE_ACK_TIMEOUT_MS))
4665 DRM_ERROR("Timed out waiting for GT to ack forcewake request.\n");
4667 if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_MEDIA_VLV) &
4669 FORCEWAKE_ACK_TIMEOUT_MS))
4670 DRM_ERROR("Timed out waiting for media to ack forcewake request.\n");
4672 __gen6_gt_wait_for_thread_c0(dev_priv);
4675 static void vlv_force_wake_put(struct drm_i915_private *dev_priv)
4677 I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4678 I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV,
4679 _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4680 /* The below doubles as a POSTING_READ */
4681 gen6_gt_check_fifodbg(dev_priv);
4684 void intel_gt_reset(struct drm_device *dev)
4686 struct drm_i915_private *dev_priv = dev->dev_private;
4688 if (IS_VALLEYVIEW(dev)) {
4689 vlv_force_wake_reset(dev_priv);
4690 } else if (INTEL_INFO(dev)->gen >= 6) {
4691 __gen6_gt_force_wake_reset(dev_priv);
4692 if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
4693 __gen6_gt_force_wake_mt_reset(dev_priv);
4697 void intel_gt_init(struct drm_device *dev)
4699 struct drm_i915_private *dev_priv = dev->dev_private;
4701 spin_lock_init(&dev_priv->gt_lock);
4703 intel_gt_reset(dev);
4705 if (IS_VALLEYVIEW(dev)) {
4706 dev_priv->gt.force_wake_get = vlv_force_wake_get;
4707 dev_priv->gt.force_wake_put = vlv_force_wake_put;
4708 } else if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) {
4709 dev_priv->gt.force_wake_get = __gen6_gt_force_wake_mt_get;
4710 dev_priv->gt.force_wake_put = __gen6_gt_force_wake_mt_put;
4711 } else if (IS_GEN6(dev)) {
4712 dev_priv->gt.force_wake_get = __gen6_gt_force_wake_get;
4713 dev_priv->gt.force_wake_put = __gen6_gt_force_wake_put;
4715 INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
4716 intel_gen6_powersave_work);
4719 int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
4721 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4723 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
4724 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
4728 I915_WRITE(GEN6_PCODE_DATA, *val);
4729 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
4731 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
4733 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
4737 *val = I915_READ(GEN6_PCODE_DATA);
4738 I915_WRITE(GEN6_PCODE_DATA, 0);
4743 int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
4745 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4747 if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
4748 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
4752 I915_WRITE(GEN6_PCODE_DATA, val);
4753 I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
4755 if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
4757 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
4761 I915_WRITE(GEN6_PCODE_DATA, 0);
4766 static int vlv_punit_rw(struct drm_i915_private *dev_priv, u32 port, u8 opcode,
4769 u32 cmd, devfn, be, bar;
4773 devfn = PCI_DEVFN(2, 0);
4775 cmd = (devfn << IOSF_DEVFN_SHIFT) | (opcode << IOSF_OPCODE_SHIFT) |
4776 (port << IOSF_PORT_SHIFT) | (be << IOSF_BYTE_ENABLES_SHIFT) |
4777 (bar << IOSF_BAR_SHIFT);
4779 WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4781 if (I915_READ(VLV_IOSF_DOORBELL_REQ) & IOSF_SB_BUSY) {
4782 DRM_DEBUG_DRIVER("warning: pcode (%s) mailbox access failed\n",
4783 opcode == PUNIT_OPCODE_REG_READ ?
4788 I915_WRITE(VLV_IOSF_ADDR, addr);
4789 if (opcode == PUNIT_OPCODE_REG_WRITE)
4790 I915_WRITE(VLV_IOSF_DATA, *val);
4791 I915_WRITE(VLV_IOSF_DOORBELL_REQ, cmd);
4793 if (wait_for((I915_READ(VLV_IOSF_DOORBELL_REQ) & IOSF_SB_BUSY) == 0,
4795 DRM_ERROR("timeout waiting for pcode %s (%d) to finish\n",
4796 opcode == PUNIT_OPCODE_REG_READ ? "read" : "write",
4801 if (opcode == PUNIT_OPCODE_REG_READ)
4802 *val = I915_READ(VLV_IOSF_DATA);
4803 I915_WRITE(VLV_IOSF_DATA, 0);
4808 int valleyview_punit_read(struct drm_i915_private *dev_priv, u8 addr, u32 *val)
4810 return vlv_punit_rw(dev_priv, IOSF_PORT_PUNIT, PUNIT_OPCODE_REG_READ,
4814 int valleyview_punit_write(struct drm_i915_private *dev_priv, u8 addr, u32 val)
4816 return vlv_punit_rw(dev_priv, IOSF_PORT_PUNIT, PUNIT_OPCODE_REG_WRITE,
4820 int valleyview_nc_read(struct drm_i915_private *dev_priv, u8 addr, u32 *val)
4822 return vlv_punit_rw(dev_priv, IOSF_PORT_NC, PUNIT_OPCODE_REG_READ,
4826 int vlv_gpu_freq(int ddr_freq, int val)
4847 return ((val - 0xbd) * mult) + base;
4850 int vlv_freq_opcode(int ddr_freq, int val)