xref: /netbsd-src/sys/external/bsd/drm2/dist/drm/i915/intel_pm.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*
2  * Copyright © 2012 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Eugeni Dodonov <eugeni.dodonov@intel.com>
25  *
26  */
27 
28 #include <linux/cpufreq.h>
29 #include "i915_drv.h"
30 #include "intel_drv.h"
31 #ifndef __NetBSD__
32 #include "../../../platform/x86/intel_ips.h"
33 #endif
34 #include <linux/module.h>
35 #include <linux/kgdb.h>
36 #include <linux/log2.h>
37 #include <linux/math64.h>
38 #include <linux/time.h>
39 #include <asm/param.h>
40 
41 #define FORCEWAKE_ACK_TIMEOUT_MS 2
42 
43 /* FBC, or Frame Buffer Compression, is a technique employed to compress the
44  * framebuffer contents in-memory, aiming at reducing the required bandwidth
45  * during in-memory transfers and, therefore, reduce the power packet.
46  *
47  * The benefits of FBC are mostly visible with solid backgrounds and
48  * variation-less patterns.
49  *
50  * FBC-related functionality can be enabled by the means of the
51  * i915.i915_enable_fbc parameter
52  */
53 
54 static bool intel_crtc_active(struct drm_crtc *crtc)
55 {
56 	/* Be paranoid as we can arrive here with only partial
57 	 * state retrieved from the hardware during setup.
58 	 */
59 	return to_intel_crtc(crtc)->active && crtc->fb && crtc->mode.clock;
60 }
61 
62 static void i8xx_disable_fbc(struct drm_device *dev)
63 {
64 	struct drm_i915_private *dev_priv = dev->dev_private;
65 	u32 fbc_ctl;
66 
67 	/* Disable compression */
68 	fbc_ctl = I915_READ(FBC_CONTROL);
69 	if ((fbc_ctl & FBC_CTL_EN) == 0)
70 		return;
71 
72 	fbc_ctl &= ~FBC_CTL_EN;
73 	I915_WRITE(FBC_CONTROL, fbc_ctl);
74 
75 	/* Wait for compressing bit to clear */
76 	if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) {
77 		DRM_DEBUG_KMS("FBC idle timed out\n");
78 		return;
79 	}
80 
81 	DRM_DEBUG_KMS("disabled FBC\n");
82 }
83 
84 static void i8xx_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
85 {
86 	struct drm_device *dev = crtc->dev;
87 	struct drm_i915_private *dev_priv = dev->dev_private;
88 	struct drm_framebuffer *fb = crtc->fb;
89 	struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
90 	struct drm_i915_gem_object *obj = intel_fb->obj;
91 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
92 	int cfb_pitch;
93 	int plane, i;
94 	u32 fbc_ctl, fbc_ctl2;
95 
96 	cfb_pitch = dev_priv->cfb_size / FBC_LL_SIZE;
97 	if (fb->pitches[0] < cfb_pitch)
98 		cfb_pitch = fb->pitches[0];
99 
100 	/* FBC_CTL wants 64B units */
101 	cfb_pitch = (cfb_pitch / 64) - 1;
102 	plane = intel_crtc->plane == 0 ? FBC_CTL_PLANEA : FBC_CTL_PLANEB;
103 
104 	/* Clear old tags */
105 	for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++)
106 		I915_WRITE(FBC_TAG + (i * 4), 0);
107 
108 	/* Set it up... */
109 	fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE;
110 	fbc_ctl2 |= plane;
111 	I915_WRITE(FBC_CONTROL2, fbc_ctl2);
112 	I915_WRITE(FBC_FENCE_OFF, crtc->y);
113 
114 	/* enable it... */
115 	fbc_ctl = FBC_CTL_EN | FBC_CTL_PERIODIC;
116 	if (IS_I945GM(dev))
117 		fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */
118 	fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT;
119 	fbc_ctl |= (interval & 0x2fff) << FBC_CTL_INTERVAL_SHIFT;
120 	fbc_ctl |= obj->fence_reg;
121 	I915_WRITE(FBC_CONTROL, fbc_ctl);
122 
123 	DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %d, ",
124 		      cfb_pitch, crtc->y, intel_crtc->plane);
125 }
126 
127 static bool i8xx_fbc_enabled(struct drm_device *dev)
128 {
129 	struct drm_i915_private *dev_priv = dev->dev_private;
130 
131 	return I915_READ(FBC_CONTROL) & FBC_CTL_EN;
132 }
133 
134 static void g4x_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
135 {
136 	struct drm_device *dev = crtc->dev;
137 	struct drm_i915_private *dev_priv = dev->dev_private;
138 	struct drm_framebuffer *fb = crtc->fb;
139 	struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
140 	struct drm_i915_gem_object *obj = intel_fb->obj;
141 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
142 	int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
143 	unsigned long stall_watermark = 200;
144 	u32 dpfc_ctl;
145 
146 	dpfc_ctl = plane | DPFC_SR_EN | DPFC_CTL_LIMIT_1X;
147 	dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg;
148 	I915_WRITE(DPFC_CHICKEN, DPFC_HT_MODIFY);
149 
150 	I915_WRITE(DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
151 		   (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
152 		   (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
153 	I915_WRITE(DPFC_FENCE_YOFF, crtc->y);
154 
155 	/* enable it... */
156 	I915_WRITE(DPFC_CONTROL, I915_READ(DPFC_CONTROL) | DPFC_CTL_EN);
157 
158 	DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc->plane);
159 }
160 
161 static void g4x_disable_fbc(struct drm_device *dev)
162 {
163 	struct drm_i915_private *dev_priv = dev->dev_private;
164 	u32 dpfc_ctl;
165 
166 	/* Disable compression */
167 	dpfc_ctl = I915_READ(DPFC_CONTROL);
168 	if (dpfc_ctl & DPFC_CTL_EN) {
169 		dpfc_ctl &= ~DPFC_CTL_EN;
170 		I915_WRITE(DPFC_CONTROL, dpfc_ctl);
171 
172 		DRM_DEBUG_KMS("disabled FBC\n");
173 	}
174 }
175 
176 static bool g4x_fbc_enabled(struct drm_device *dev)
177 {
178 	struct drm_i915_private *dev_priv = dev->dev_private;
179 
180 	return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN;
181 }
182 
183 static void sandybridge_blit_fbc_update(struct drm_device *dev)
184 {
185 	struct drm_i915_private *dev_priv = dev->dev_private;
186 	u32 blt_ecoskpd;
187 
188 	/* Make sure blitter notifies FBC of writes */
189 	gen6_gt_force_wake_get(dev_priv);
190 	blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD);
191 	blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY <<
192 		GEN6_BLITTER_LOCK_SHIFT;
193 	I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
194 	blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY;
195 	I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
196 	blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY <<
197 			 GEN6_BLITTER_LOCK_SHIFT);
198 	I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
199 	POSTING_READ(GEN6_BLITTER_ECOSKPD);
200 	gen6_gt_force_wake_put(dev_priv);
201 }
202 
203 static void ironlake_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
204 {
205 	struct drm_device *dev = crtc->dev;
206 	struct drm_i915_private *dev_priv = dev->dev_private;
207 	struct drm_framebuffer *fb = crtc->fb;
208 	struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
209 	struct drm_i915_gem_object *obj = intel_fb->obj;
210 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
211 	int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
212 	unsigned long stall_watermark = 200;
213 	u32 dpfc_ctl;
214 
215 	dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
216 	dpfc_ctl &= DPFC_RESERVED;
217 	dpfc_ctl |= (plane | DPFC_CTL_LIMIT_1X);
218 	/* Set persistent mode for front-buffer rendering, ala X. */
219 	dpfc_ctl |= DPFC_CTL_PERSISTENT_MODE;
220 	dpfc_ctl |= (DPFC_CTL_FENCE_EN | obj->fence_reg);
221 	I915_WRITE(ILK_DPFC_CHICKEN, DPFC_HT_MODIFY);
222 
223 	I915_WRITE(ILK_DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
224 		   (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
225 		   (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
226 	I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
227 	I915_WRITE(ILK_FBC_RT_BASE, obj->gtt_offset | ILK_FBC_RT_VALID);
228 	/* enable it... */
229 	I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
230 
231 	if (IS_GEN6(dev)) {
232 		I915_WRITE(SNB_DPFC_CTL_SA,
233 			   SNB_CPU_FENCE_ENABLE | obj->fence_reg);
234 		I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
235 		sandybridge_blit_fbc_update(dev);
236 	}
237 
238 	DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc->plane);
239 }
240 
241 static void ironlake_disable_fbc(struct drm_device *dev)
242 {
243 	struct drm_i915_private *dev_priv = dev->dev_private;
244 	u32 dpfc_ctl;
245 
246 	/* Disable compression */
247 	dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
248 	if (dpfc_ctl & DPFC_CTL_EN) {
249 		dpfc_ctl &= ~DPFC_CTL_EN;
250 		I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl);
251 
252 		DRM_DEBUG_KMS("disabled FBC\n");
253 	}
254 }
255 
256 static bool ironlake_fbc_enabled(struct drm_device *dev)
257 {
258 	struct drm_i915_private *dev_priv = dev->dev_private;
259 
260 	return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
261 }
262 
263 bool intel_fbc_enabled(struct drm_device *dev)
264 {
265 	struct drm_i915_private *dev_priv = dev->dev_private;
266 
267 	if (!dev_priv->display.fbc_enabled)
268 		return false;
269 
270 	return dev_priv->display.fbc_enabled(dev);
271 }
272 
273 static void intel_fbc_work_fn(struct work_struct *__work)
274 {
275 	struct intel_fbc_work *work =
276 		container_of(to_delayed_work(__work),
277 			     struct intel_fbc_work, work);
278 	struct drm_device *dev = work->crtc->dev;
279 	struct drm_i915_private *dev_priv = dev->dev_private;
280 
281 	mutex_lock(&dev->struct_mutex);
282 	if (work == dev_priv->fbc_work) {
283 		/* Double check that we haven't switched fb without cancelling
284 		 * the prior work.
285 		 */
286 		if (work->crtc->fb == work->fb) {
287 			dev_priv->display.enable_fbc(work->crtc,
288 						     work->interval);
289 
290 			dev_priv->cfb_plane = to_intel_crtc(work->crtc)->plane;
291 			dev_priv->cfb_fb = work->crtc->fb->base.id;
292 			dev_priv->cfb_y = work->crtc->y;
293 		}
294 
295 		dev_priv->fbc_work = NULL;
296 	}
297 	mutex_unlock(&dev->struct_mutex);
298 
299 	kfree(work);
300 }
301 
302 static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
303 {
304 	if (dev_priv->fbc_work == NULL)
305 		return;
306 
307 	DRM_DEBUG_KMS("cancelling pending FBC enable\n");
308 
309 	/* Synchronisation is provided by struct_mutex and checking of
310 	 * dev_priv->fbc_work, so we can perform the cancellation
311 	 * entirely asynchronously.
312 	 */
313 	if (cancel_delayed_work(&dev_priv->fbc_work->work))
314 		/* tasklet was killed before being run, clean up */
315 		kfree(dev_priv->fbc_work);
316 
317 	/* Mark the work as no longer wanted so that if it does
318 	 * wake-up (because the work was already running and waiting
319 	 * for our mutex), it will discover that is no longer
320 	 * necessary to run.
321 	 */
322 	dev_priv->fbc_work = NULL;
323 }
324 
325 void intel_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
326 {
327 	struct intel_fbc_work *work;
328 	struct drm_device *dev = crtc->dev;
329 	struct drm_i915_private *dev_priv = dev->dev_private;
330 
331 	if (!dev_priv->display.enable_fbc)
332 		return;
333 
334 	intel_cancel_fbc_work(dev_priv);
335 
336 	work = kzalloc(sizeof *work, GFP_KERNEL);
337 	if (work == NULL) {
338 		dev_priv->display.enable_fbc(crtc, interval);
339 		return;
340 	}
341 
342 	work->crtc = crtc;
343 	work->fb = crtc->fb;
344 	work->interval = interval;
345 	INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
346 
347 	dev_priv->fbc_work = work;
348 
349 	DRM_DEBUG_KMS("scheduling delayed FBC enable\n");
350 
351 	/* Delay the actual enabling to let pageflipping cease and the
352 	 * display to settle before starting the compression. Note that
353 	 * this delay also serves a second purpose: it allows for a
354 	 * vblank to pass after disabling the FBC before we attempt
355 	 * to modify the control registers.
356 	 *
357 	 * A more complicated solution would involve tracking vblanks
358 	 * following the termination of the page-flipping sequence
359 	 * and indeed performing the enable as a co-routine and not
360 	 * waiting synchronously upon the vblank.
361 	 */
362 	schedule_delayed_work(&work->work, msecs_to_jiffies(50));
363 }
364 
365 void intel_disable_fbc(struct drm_device *dev)
366 {
367 	struct drm_i915_private *dev_priv = dev->dev_private;
368 
369 	intel_cancel_fbc_work(dev_priv);
370 
371 	if (!dev_priv->display.disable_fbc)
372 		return;
373 
374 	dev_priv->display.disable_fbc(dev);
375 	dev_priv->cfb_plane = -1;
376 }
377 
378 /**
379  * intel_update_fbc - enable/disable FBC as needed
380  * @dev: the drm_device
381  *
382  * Set up the framebuffer compression hardware at mode set time.  We
383  * enable it if possible:
384  *   - plane A only (on pre-965)
385  *   - no pixel mulitply/line duplication
386  *   - no alpha buffer discard
387  *   - no dual wide
388  *   - framebuffer <= 2048 in width, 1536 in height
389  *
390  * We can't assume that any compression will take place (worst case),
391  * so the compressed buffer has to be the same size as the uncompressed
392  * one.  It also must reside (along with the line length buffer) in
393  * stolen memory.
394  *
395  * We need to enable/disable FBC on a global basis.
396  */
397 void intel_update_fbc(struct drm_device *dev)
398 {
399 	struct drm_i915_private *dev_priv = dev->dev_private;
400 	struct drm_crtc *crtc = NULL, *tmp_crtc;
401 	struct intel_crtc *intel_crtc;
402 	struct drm_framebuffer *fb;
403 	struct intel_framebuffer *intel_fb;
404 	struct drm_i915_gem_object *obj;
405 	int enable_fbc;
406 
407 	if (!i915_powersave)
408 		return;
409 
410 	if (!I915_HAS_FBC(dev))
411 		return;
412 
413 	/*
414 	 * If FBC is already on, we just have to verify that we can
415 	 * keep it that way...
416 	 * Need to disable if:
417 	 *   - more than one pipe is active
418 	 *   - changing FBC params (stride, fence, mode)
419 	 *   - new fb is too large to fit in compressed buffer
420 	 *   - going to an unsupported config (interlace, pixel multiply, etc.)
421 	 */
422 	list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) {
423 		if (intel_crtc_active(tmp_crtc) &&
424 		    !to_intel_crtc(tmp_crtc)->primary_disabled) {
425 			if (crtc) {
426 				DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
427 				dev_priv->no_fbc_reason = FBC_MULTIPLE_PIPES;
428 				goto out_disable;
429 			}
430 			crtc = tmp_crtc;
431 		}
432 	}
433 
434 	if (!crtc || crtc->fb == NULL) {
435 		DRM_DEBUG_KMS("no output, disabling\n");
436 		dev_priv->no_fbc_reason = FBC_NO_OUTPUT;
437 		goto out_disable;
438 	}
439 
440 	intel_crtc = to_intel_crtc(crtc);
441 	fb = crtc->fb;
442 	intel_fb = to_intel_framebuffer(fb);
443 	obj = intel_fb->obj;
444 
445 	enable_fbc = i915_enable_fbc;
446 	if (enable_fbc < 0) {
447 		DRM_DEBUG_KMS("fbc set to per-chip default\n");
448 		enable_fbc = 1;
449 		if (INTEL_INFO(dev)->gen <= 6)
450 			enable_fbc = 0;
451 	}
452 	if (!enable_fbc) {
453 		DRM_DEBUG_KMS("fbc disabled per module param\n");
454 		dev_priv->no_fbc_reason = FBC_MODULE_PARAM;
455 		goto out_disable;
456 	}
457 	if (intel_fb->obj->base.size > dev_priv->cfb_size) {
458 		DRM_DEBUG_KMS("framebuffer too large, disabling "
459 			      "compression\n");
460 		dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
461 		goto out_disable;
462 	}
463 	if ((crtc->mode.flags & DRM_MODE_FLAG_INTERLACE) ||
464 	    (crtc->mode.flags & DRM_MODE_FLAG_DBLSCAN)) {
465 		DRM_DEBUG_KMS("mode incompatible with compression, "
466 			      "disabling\n");
467 		dev_priv->no_fbc_reason = FBC_UNSUPPORTED_MODE;
468 		goto out_disable;
469 	}
470 	if ((crtc->mode.hdisplay > 2048) ||
471 	    (crtc->mode.vdisplay > 1536)) {
472 		DRM_DEBUG_KMS("mode too large for compression, disabling\n");
473 		dev_priv->no_fbc_reason = FBC_MODE_TOO_LARGE;
474 		goto out_disable;
475 	}
476 	if ((IS_I915GM(dev) || IS_I945GM(dev)) && intel_crtc->plane != 0) {
477 		DRM_DEBUG_KMS("plane not 0, disabling compression\n");
478 		dev_priv->no_fbc_reason = FBC_BAD_PLANE;
479 		goto out_disable;
480 	}
481 
482 	/* The use of a CPU fence is mandatory in order to detect writes
483 	 * by the CPU to the scanout and trigger updates to the FBC.
484 	 */
485 	if (obj->tiling_mode != I915_TILING_X ||
486 	    obj->fence_reg == I915_FENCE_REG_NONE) {
487 		DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
488 		dev_priv->no_fbc_reason = FBC_NOT_TILED;
489 		goto out_disable;
490 	}
491 
492 	/* If the kernel debugger is active, always disable compression */
493 	if (in_dbg_master())
494 		goto out_disable;
495 
496 	/* If the scanout has not changed, don't modify the FBC settings.
497 	 * Note that we make the fundamental assumption that the fb->obj
498 	 * cannot be unpinned (and have its GTT offset and fence revoked)
499 	 * without first being decoupled from the scanout and FBC disabled.
500 	 */
501 	if (dev_priv->cfb_plane == intel_crtc->plane &&
502 	    dev_priv->cfb_fb == fb->base.id &&
503 	    dev_priv->cfb_y == crtc->y)
504 		return;
505 
506 	if (intel_fbc_enabled(dev)) {
507 		/* We update FBC along two paths, after changing fb/crtc
508 		 * configuration (modeswitching) and after page-flipping
509 		 * finishes. For the latter, we know that not only did
510 		 * we disable the FBC at the start of the page-flip
511 		 * sequence, but also more than one vblank has passed.
512 		 *
513 		 * For the former case of modeswitching, it is possible
514 		 * to switch between two FBC valid configurations
515 		 * instantaneously so we do need to disable the FBC
516 		 * before we can modify its control registers. We also
517 		 * have to wait for the next vblank for that to take
518 		 * effect. However, since we delay enabling FBC we can
519 		 * assume that a vblank has passed since disabling and
520 		 * that we can safely alter the registers in the deferred
521 		 * callback.
522 		 *
523 		 * In the scenario that we go from a valid to invalid
524 		 * and then back to valid FBC configuration we have
525 		 * no strict enforcement that a vblank occurred since
526 		 * disabling the FBC. However, along all current pipe
527 		 * disabling paths we do need to wait for a vblank at
528 		 * some point. And we wait before enabling FBC anyway.
529 		 */
530 		DRM_DEBUG_KMS("disabling active FBC for update\n");
531 		intel_disable_fbc(dev);
532 	}
533 
534 	intel_enable_fbc(crtc, 500);
535 	return;
536 
537 out_disable:
538 	/* Multiple disables should be harmless */
539 	if (intel_fbc_enabled(dev)) {
540 		DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
541 		intel_disable_fbc(dev);
542 	}
543 }
544 
545 static void i915_pineview_get_mem_freq(struct drm_device *dev)
546 {
547 	drm_i915_private_t *dev_priv = dev->dev_private;
548 	u32 tmp;
549 
550 	tmp = I915_READ(CLKCFG);
551 
552 	switch (tmp & CLKCFG_FSB_MASK) {
553 	case CLKCFG_FSB_533:
554 		dev_priv->fsb_freq = 533; /* 133*4 */
555 		break;
556 	case CLKCFG_FSB_800:
557 		dev_priv->fsb_freq = 800; /* 200*4 */
558 		break;
559 	case CLKCFG_FSB_667:
560 		dev_priv->fsb_freq =  667; /* 167*4 */
561 		break;
562 	case CLKCFG_FSB_400:
563 		dev_priv->fsb_freq = 400; /* 100*4 */
564 		break;
565 	}
566 
567 	switch (tmp & CLKCFG_MEM_MASK) {
568 	case CLKCFG_MEM_533:
569 		dev_priv->mem_freq = 533;
570 		break;
571 	case CLKCFG_MEM_667:
572 		dev_priv->mem_freq = 667;
573 		break;
574 	case CLKCFG_MEM_800:
575 		dev_priv->mem_freq = 800;
576 		break;
577 	}
578 
579 	/* detect pineview DDR3 setting */
580 	tmp = I915_READ(CSHRDDR3CTL);
581 	dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
582 }
583 
584 static void i915_ironlake_get_mem_freq(struct drm_device *dev)
585 {
586 	drm_i915_private_t *dev_priv = dev->dev_private;
587 	u16 ddrpll, csipll;
588 
589 	ddrpll = I915_READ16(DDRMPLL1);
590 	csipll = I915_READ16(CSIPLL0);
591 
592 	switch (ddrpll & 0xff) {
593 	case 0xc:
594 		dev_priv->mem_freq = 800;
595 		break;
596 	case 0x10:
597 		dev_priv->mem_freq = 1066;
598 		break;
599 	case 0x14:
600 		dev_priv->mem_freq = 1333;
601 		break;
602 	case 0x18:
603 		dev_priv->mem_freq = 1600;
604 		break;
605 	default:
606 		DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
607 				 ddrpll & 0xff);
608 		dev_priv->mem_freq = 0;
609 		break;
610 	}
611 
612 	dev_priv->ips.r_t = dev_priv->mem_freq;
613 
614 	switch (csipll & 0x3ff) {
615 	case 0x00c:
616 		dev_priv->fsb_freq = 3200;
617 		break;
618 	case 0x00e:
619 		dev_priv->fsb_freq = 3733;
620 		break;
621 	case 0x010:
622 		dev_priv->fsb_freq = 4266;
623 		break;
624 	case 0x012:
625 		dev_priv->fsb_freq = 4800;
626 		break;
627 	case 0x014:
628 		dev_priv->fsb_freq = 5333;
629 		break;
630 	case 0x016:
631 		dev_priv->fsb_freq = 5866;
632 		break;
633 	case 0x018:
634 		dev_priv->fsb_freq = 6400;
635 		break;
636 	default:
637 		DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
638 				 csipll & 0x3ff);
639 		dev_priv->fsb_freq = 0;
640 		break;
641 	}
642 
643 	if (dev_priv->fsb_freq == 3200) {
644 		dev_priv->ips.c_m = 0;
645 	} else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
646 		dev_priv->ips.c_m = 1;
647 	} else {
648 		dev_priv->ips.c_m = 2;
649 	}
650 }
651 
652 static const struct cxsr_latency cxsr_latency_table[] = {
653 	{1, 0, 800, 400, 3382, 33382, 3983, 33983},    /* DDR2-400 SC */
654 	{1, 0, 800, 667, 3354, 33354, 3807, 33807},    /* DDR2-667 SC */
655 	{1, 0, 800, 800, 3347, 33347, 3763, 33763},    /* DDR2-800 SC */
656 	{1, 1, 800, 667, 6420, 36420, 6873, 36873},    /* DDR3-667 SC */
657 	{1, 1, 800, 800, 5902, 35902, 6318, 36318},    /* DDR3-800 SC */
658 
659 	{1, 0, 667, 400, 3400, 33400, 4021, 34021},    /* DDR2-400 SC */
660 	{1, 0, 667, 667, 3372, 33372, 3845, 33845},    /* DDR2-667 SC */
661 	{1, 0, 667, 800, 3386, 33386, 3822, 33822},    /* DDR2-800 SC */
662 	{1, 1, 667, 667, 6438, 36438, 6911, 36911},    /* DDR3-667 SC */
663 	{1, 1, 667, 800, 5941, 35941, 6377, 36377},    /* DDR3-800 SC */
664 
665 	{1, 0, 400, 400, 3472, 33472, 4173, 34173},    /* DDR2-400 SC */
666 	{1, 0, 400, 667, 3443, 33443, 3996, 33996},    /* DDR2-667 SC */
667 	{1, 0, 400, 800, 3430, 33430, 3946, 33946},    /* DDR2-800 SC */
668 	{1, 1, 400, 667, 6509, 36509, 7062, 37062},    /* DDR3-667 SC */
669 	{1, 1, 400, 800, 5985, 35985, 6501, 36501},    /* DDR3-800 SC */
670 
671 	{0, 0, 800, 400, 3438, 33438, 4065, 34065},    /* DDR2-400 SC */
672 	{0, 0, 800, 667, 3410, 33410, 3889, 33889},    /* DDR2-667 SC */
673 	{0, 0, 800, 800, 3403, 33403, 3845, 33845},    /* DDR2-800 SC */
674 	{0, 1, 800, 667, 6476, 36476, 6955, 36955},    /* DDR3-667 SC */
675 	{0, 1, 800, 800, 5958, 35958, 6400, 36400},    /* DDR3-800 SC */
676 
677 	{0, 0, 667, 400, 3456, 33456, 4103, 34106},    /* DDR2-400 SC */
678 	{0, 0, 667, 667, 3428, 33428, 3927, 33927},    /* DDR2-667 SC */
679 	{0, 0, 667, 800, 3443, 33443, 3905, 33905},    /* DDR2-800 SC */
680 	{0, 1, 667, 667, 6494, 36494, 6993, 36993},    /* DDR3-667 SC */
681 	{0, 1, 667, 800, 5998, 35998, 6460, 36460},    /* DDR3-800 SC */
682 
683 	{0, 0, 400, 400, 3528, 33528, 4255, 34255},    /* DDR2-400 SC */
684 	{0, 0, 400, 667, 3500, 33500, 4079, 34079},    /* DDR2-667 SC */
685 	{0, 0, 400, 800, 3487, 33487, 4029, 34029},    /* DDR2-800 SC */
686 	{0, 1, 400, 667, 6566, 36566, 7145, 37145},    /* DDR3-667 SC */
687 	{0, 1, 400, 800, 6042, 36042, 6584, 36584},    /* DDR3-800 SC */
688 };
689 
690 static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
691 							 int is_ddr3,
692 							 int fsb,
693 							 int mem)
694 {
695 	const struct cxsr_latency *latency;
696 	int i;
697 
698 	if (fsb == 0 || mem == 0)
699 		return NULL;
700 
701 	for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
702 		latency = &cxsr_latency_table[i];
703 		if (is_desktop == latency->is_desktop &&
704 		    is_ddr3 == latency->is_ddr3 &&
705 		    fsb == latency->fsb_freq && mem == latency->mem_freq)
706 			return latency;
707 	}
708 
709 	DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
710 
711 	return NULL;
712 }
713 
714 static void pineview_disable_cxsr(struct drm_device *dev)
715 {
716 	struct drm_i915_private *dev_priv = dev->dev_private;
717 
718 	/* deactivate cxsr */
719 	I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
720 }
721 
722 /*
723  * Latency for FIFO fetches is dependent on several factors:
724  *   - memory configuration (speed, channels)
725  *   - chipset
726  *   - current MCH state
727  * It can be fairly high in some situations, so here we assume a fairly
728  * pessimal value.  It's a tradeoff between extra memory fetches (if we
729  * set this value too high, the FIFO will fetch frequently to stay full)
730  * and power consumption (set it too low to save power and we might see
731  * FIFO underruns and display "flicker").
732  *
733  * A value of 5us seems to be a good balance; safe for very low end
734  * platforms but not overly aggressive on lower latency configs.
735  */
736 static const int latency_ns = 5000;
737 
738 static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
739 {
740 	struct drm_i915_private *dev_priv = dev->dev_private;
741 	uint32_t dsparb = I915_READ(DSPARB);
742 	int size;
743 
744 	size = dsparb & 0x7f;
745 	if (plane)
746 		size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
747 
748 	DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
749 		      plane ? "B" : "A", size);
750 
751 	return size;
752 }
753 
754 static int i85x_get_fifo_size(struct drm_device *dev, int plane)
755 {
756 	struct drm_i915_private *dev_priv = dev->dev_private;
757 	uint32_t dsparb = I915_READ(DSPARB);
758 	int size;
759 
760 	size = dsparb & 0x1ff;
761 	if (plane)
762 		size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
763 	size >>= 1; /* Convert to cachelines */
764 
765 	DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
766 		      plane ? "B" : "A", size);
767 
768 	return size;
769 }
770 
771 static int i845_get_fifo_size(struct drm_device *dev, int plane)
772 {
773 	struct drm_i915_private *dev_priv = dev->dev_private;
774 	uint32_t dsparb = I915_READ(DSPARB);
775 	int size;
776 
777 	size = dsparb & 0x7f;
778 	size >>= 2; /* Convert to cachelines */
779 
780 	DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
781 		      plane ? "B" : "A",
782 		      size);
783 
784 	return size;
785 }
786 
787 static int i830_get_fifo_size(struct drm_device *dev, int plane)
788 {
789 	struct drm_i915_private *dev_priv = dev->dev_private;
790 	uint32_t dsparb = I915_READ(DSPARB);
791 	int size;
792 
793 	size = dsparb & 0x7f;
794 	size >>= 1; /* Convert to cachelines */
795 
796 	DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
797 		      plane ? "B" : "A", size);
798 
799 	return size;
800 }
801 
802 /* Pineview has different values for various configs */
803 static const struct intel_watermark_params pineview_display_wm = {
804 	PINEVIEW_DISPLAY_FIFO,
805 	PINEVIEW_MAX_WM,
806 	PINEVIEW_DFT_WM,
807 	PINEVIEW_GUARD_WM,
808 	PINEVIEW_FIFO_LINE_SIZE
809 };
810 static const struct intel_watermark_params pineview_display_hplloff_wm = {
811 	PINEVIEW_DISPLAY_FIFO,
812 	PINEVIEW_MAX_WM,
813 	PINEVIEW_DFT_HPLLOFF_WM,
814 	PINEVIEW_GUARD_WM,
815 	PINEVIEW_FIFO_LINE_SIZE
816 };
817 static const struct intel_watermark_params pineview_cursor_wm = {
818 	PINEVIEW_CURSOR_FIFO,
819 	PINEVIEW_CURSOR_MAX_WM,
820 	PINEVIEW_CURSOR_DFT_WM,
821 	PINEVIEW_CURSOR_GUARD_WM,
822 	PINEVIEW_FIFO_LINE_SIZE,
823 };
824 static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
825 	PINEVIEW_CURSOR_FIFO,
826 	PINEVIEW_CURSOR_MAX_WM,
827 	PINEVIEW_CURSOR_DFT_WM,
828 	PINEVIEW_CURSOR_GUARD_WM,
829 	PINEVIEW_FIFO_LINE_SIZE
830 };
831 static const struct intel_watermark_params g4x_wm_info = {
832 	G4X_FIFO_SIZE,
833 	G4X_MAX_WM,
834 	G4X_MAX_WM,
835 	2,
836 	G4X_FIFO_LINE_SIZE,
837 };
838 static const struct intel_watermark_params g4x_cursor_wm_info = {
839 	I965_CURSOR_FIFO,
840 	I965_CURSOR_MAX_WM,
841 	I965_CURSOR_DFT_WM,
842 	2,
843 	G4X_FIFO_LINE_SIZE,
844 };
845 static const struct intel_watermark_params valleyview_wm_info = {
846 	VALLEYVIEW_FIFO_SIZE,
847 	VALLEYVIEW_MAX_WM,
848 	VALLEYVIEW_MAX_WM,
849 	2,
850 	G4X_FIFO_LINE_SIZE,
851 };
852 static const struct intel_watermark_params valleyview_cursor_wm_info = {
853 	I965_CURSOR_FIFO,
854 	VALLEYVIEW_CURSOR_MAX_WM,
855 	I965_CURSOR_DFT_WM,
856 	2,
857 	G4X_FIFO_LINE_SIZE,
858 };
859 static const struct intel_watermark_params i965_cursor_wm_info = {
860 	I965_CURSOR_FIFO,
861 	I965_CURSOR_MAX_WM,
862 	I965_CURSOR_DFT_WM,
863 	2,
864 	I915_FIFO_LINE_SIZE,
865 };
866 static const struct intel_watermark_params i945_wm_info = {
867 	I945_FIFO_SIZE,
868 	I915_MAX_WM,
869 	1,
870 	2,
871 	I915_FIFO_LINE_SIZE
872 };
873 static const struct intel_watermark_params i915_wm_info = {
874 	I915_FIFO_SIZE,
875 	I915_MAX_WM,
876 	1,
877 	2,
878 	I915_FIFO_LINE_SIZE
879 };
880 static const struct intel_watermark_params i855_wm_info = {
881 	I855GM_FIFO_SIZE,
882 	I915_MAX_WM,
883 	1,
884 	2,
885 	I830_FIFO_LINE_SIZE
886 };
887 static const struct intel_watermark_params i830_wm_info = {
888 	I830_FIFO_SIZE,
889 	I915_MAX_WM,
890 	1,
891 	2,
892 	I830_FIFO_LINE_SIZE
893 };
894 
895 static const struct intel_watermark_params ironlake_display_wm_info = {
896 	ILK_DISPLAY_FIFO,
897 	ILK_DISPLAY_MAXWM,
898 	ILK_DISPLAY_DFTWM,
899 	2,
900 	ILK_FIFO_LINE_SIZE
901 };
902 static const struct intel_watermark_params ironlake_cursor_wm_info = {
903 	ILK_CURSOR_FIFO,
904 	ILK_CURSOR_MAXWM,
905 	ILK_CURSOR_DFTWM,
906 	2,
907 	ILK_FIFO_LINE_SIZE
908 };
909 static const struct intel_watermark_params ironlake_display_srwm_info = {
910 	ILK_DISPLAY_SR_FIFO,
911 	ILK_DISPLAY_MAX_SRWM,
912 	ILK_DISPLAY_DFT_SRWM,
913 	2,
914 	ILK_FIFO_LINE_SIZE
915 };
916 static const struct intel_watermark_params ironlake_cursor_srwm_info = {
917 	ILK_CURSOR_SR_FIFO,
918 	ILK_CURSOR_MAX_SRWM,
919 	ILK_CURSOR_DFT_SRWM,
920 	2,
921 	ILK_FIFO_LINE_SIZE
922 };
923 
924 static const struct intel_watermark_params sandybridge_display_wm_info = {
925 	SNB_DISPLAY_FIFO,
926 	SNB_DISPLAY_MAXWM,
927 	SNB_DISPLAY_DFTWM,
928 	2,
929 	SNB_FIFO_LINE_SIZE
930 };
931 static const struct intel_watermark_params sandybridge_cursor_wm_info = {
932 	SNB_CURSOR_FIFO,
933 	SNB_CURSOR_MAXWM,
934 	SNB_CURSOR_DFTWM,
935 	2,
936 	SNB_FIFO_LINE_SIZE
937 };
938 static const struct intel_watermark_params sandybridge_display_srwm_info = {
939 	SNB_DISPLAY_SR_FIFO,
940 	SNB_DISPLAY_MAX_SRWM,
941 	SNB_DISPLAY_DFT_SRWM,
942 	2,
943 	SNB_FIFO_LINE_SIZE
944 };
945 static const struct intel_watermark_params sandybridge_cursor_srwm_info = {
946 	SNB_CURSOR_SR_FIFO,
947 	SNB_CURSOR_MAX_SRWM,
948 	SNB_CURSOR_DFT_SRWM,
949 	2,
950 	SNB_FIFO_LINE_SIZE
951 };
952 
953 
954 /**
955  * intel_calculate_wm - calculate watermark level
956  * @clock_in_khz: pixel clock
957  * @wm: chip FIFO params
958  * @pixel_size: display pixel size
959  * @latency_ns: memory latency for the platform
960  *
961  * Calculate the watermark level (the level at which the display plane will
962  * start fetching from memory again).  Each chip has a different display
963  * FIFO size and allocation, so the caller needs to figure that out and pass
964  * in the correct intel_watermark_params structure.
965  *
966  * As the pixel clock runs, the FIFO will be drained at a rate that depends
967  * on the pixel size.  When it reaches the watermark level, it'll start
968  * fetching FIFO line sized based chunks from memory until the FIFO fills
969  * past the watermark point.  If the FIFO drains completely, a FIFO underrun
970  * will occur, and a display engine hang could result.
971  */
972 static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
973 					const struct intel_watermark_params *wm,
974 					int fifo_size,
975 					int pixel_size,
976 					unsigned long latency_ns)
977 {
978 	long entries_required, wm_size;
979 
980 	/*
981 	 * Note: we need to make sure we don't overflow for various clock &
982 	 * latency values.
983 	 * clocks go from a few thousand to several hundred thousand.
984 	 * latency is usually a few thousand
985 	 */
986 	entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
987 		1000;
988 	entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
989 
990 	DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
991 
992 	wm_size = fifo_size - (entries_required + wm->guard_size);
993 
994 	DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
995 
996 	/* Don't promote wm_size to unsigned... */
997 	if (wm_size > (long)wm->max_wm)
998 		wm_size = wm->max_wm;
999 	if (wm_size <= 0)
1000 		wm_size = wm->default_wm;
1001 	return wm_size;
1002 }
1003 
1004 static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1005 {
1006 	struct drm_crtc *crtc, *enabled = NULL;
1007 
1008 	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
1009 		if (intel_crtc_active(crtc)) {
1010 			if (enabled)
1011 				return NULL;
1012 			enabled = crtc;
1013 		}
1014 	}
1015 
1016 	return enabled;
1017 }
1018 
1019 static void pineview_update_wm(struct drm_device *dev)
1020 {
1021 	struct drm_i915_private *dev_priv = dev->dev_private;
1022 	struct drm_crtc *crtc;
1023 	const struct cxsr_latency *latency;
1024 	u32 reg;
1025 	unsigned long wm;
1026 
1027 	latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1028 					 dev_priv->fsb_freq, dev_priv->mem_freq);
1029 	if (!latency) {
1030 		DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1031 		pineview_disable_cxsr(dev);
1032 		return;
1033 	}
1034 
1035 	crtc = single_enabled_crtc(dev);
1036 	if (crtc) {
1037 		int clock = crtc->mode.clock;
1038 		int pixel_size = crtc->fb->bits_per_pixel / 8;
1039 
1040 		/* Display SR */
1041 		wm = intel_calculate_wm(clock, &pineview_display_wm,
1042 					pineview_display_wm.fifo_size,
1043 					pixel_size, latency->display_sr);
1044 		reg = I915_READ(DSPFW1);
1045 		reg &= ~DSPFW_SR_MASK;
1046 		reg |= wm << DSPFW_SR_SHIFT;
1047 		I915_WRITE(DSPFW1, reg);
1048 		DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1049 
1050 		/* cursor SR */
1051 		wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1052 					pineview_display_wm.fifo_size,
1053 					pixel_size, latency->cursor_sr);
1054 		reg = I915_READ(DSPFW3);
1055 		reg &= ~DSPFW_CURSOR_SR_MASK;
1056 		reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1057 		I915_WRITE(DSPFW3, reg);
1058 
1059 		/* Display HPLL off SR */
1060 		wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1061 					pineview_display_hplloff_wm.fifo_size,
1062 					pixel_size, latency->display_hpll_disable);
1063 		reg = I915_READ(DSPFW3);
1064 		reg &= ~DSPFW_HPLL_SR_MASK;
1065 		reg |= wm & DSPFW_HPLL_SR_MASK;
1066 		I915_WRITE(DSPFW3, reg);
1067 
1068 		/* cursor HPLL off SR */
1069 		wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1070 					pineview_display_hplloff_wm.fifo_size,
1071 					pixel_size, latency->cursor_hpll_disable);
1072 		reg = I915_READ(DSPFW3);
1073 		reg &= ~DSPFW_HPLL_CURSOR_MASK;
1074 		reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1075 		I915_WRITE(DSPFW3, reg);
1076 		DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1077 
1078 		/* activate cxsr */
1079 		I915_WRITE(DSPFW3,
1080 			   I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1081 		DRM_DEBUG_KMS("Self-refresh is enabled\n");
1082 	} else {
1083 		pineview_disable_cxsr(dev);
1084 		DRM_DEBUG_KMS("Self-refresh is disabled\n");
1085 	}
1086 }
1087 
1088 static bool g4x_compute_wm0(struct drm_device *dev,
1089 			    int plane,
1090 			    const struct intel_watermark_params *display,
1091 			    int display_latency_ns,
1092 			    const struct intel_watermark_params *cursor,
1093 			    int cursor_latency_ns,
1094 			    int *plane_wm,
1095 			    int *cursor_wm)
1096 {
1097 	struct drm_crtc *crtc;
1098 	int htotal, hdisplay, clock, pixel_size;
1099 	int line_time_us, line_count;
1100 	int entries, tlb_miss;
1101 
1102 	crtc = intel_get_crtc_for_plane(dev, plane);
1103 	if (!intel_crtc_active(crtc)) {
1104 		*cursor_wm = cursor->guard_size;
1105 		*plane_wm = display->guard_size;
1106 		return false;
1107 	}
1108 
1109 	htotal = crtc->mode.htotal;
1110 	hdisplay = crtc->mode.hdisplay;
1111 	clock = crtc->mode.clock;
1112 	pixel_size = crtc->fb->bits_per_pixel / 8;
1113 
1114 	/* Use the small buffer method to calculate plane watermark */
1115 	entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1116 	tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1117 	if (tlb_miss > 0)
1118 		entries += tlb_miss;
1119 	entries = DIV_ROUND_UP(entries, display->cacheline_size);
1120 	*plane_wm = entries + display->guard_size;
1121 	if (*plane_wm > (int)display->max_wm)
1122 		*plane_wm = display->max_wm;
1123 
1124 	/* Use the large buffer method to calculate cursor watermark */
1125 	line_time_us = ((htotal * 1000) / clock);
1126 	line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
1127 	entries = line_count * 64 * pixel_size;
1128 	tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1129 	if (tlb_miss > 0)
1130 		entries += tlb_miss;
1131 	entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1132 	*cursor_wm = entries + cursor->guard_size;
1133 	if (*cursor_wm > (int)cursor->max_wm)
1134 		*cursor_wm = (int)cursor->max_wm;
1135 
1136 	return true;
1137 }
1138 
1139 /*
1140  * Check the wm result.
1141  *
1142  * If any calculated watermark values is larger than the maximum value that
1143  * can be programmed into the associated watermark register, that watermark
1144  * must be disabled.
1145  */
1146 static bool g4x_check_srwm(struct drm_device *dev,
1147 			   int display_wm, int cursor_wm,
1148 			   const struct intel_watermark_params *display,
1149 			   const struct intel_watermark_params *cursor)
1150 {
1151 	DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1152 		      display_wm, cursor_wm);
1153 
1154 	if (display_wm > display->max_wm) {
1155 		DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1156 			      display_wm, display->max_wm);
1157 		return false;
1158 	}
1159 
1160 	if (cursor_wm > cursor->max_wm) {
1161 		DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1162 			      cursor_wm, cursor->max_wm);
1163 		return false;
1164 	}
1165 
1166 	if (!(display_wm || cursor_wm)) {
1167 		DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1168 		return false;
1169 	}
1170 
1171 	return true;
1172 }
1173 
1174 static bool g4x_compute_srwm(struct drm_device *dev,
1175 			     int plane,
1176 			     int latency_ns,
1177 			     const struct intel_watermark_params *display,
1178 			     const struct intel_watermark_params *cursor,
1179 			     int *display_wm, int *cursor_wm)
1180 {
1181 	struct drm_crtc *crtc;
1182 	int hdisplay, htotal, pixel_size, clock;
1183 	unsigned long line_time_us;
1184 	int line_count, line_size;
1185 	int small, large;
1186 	int entries;
1187 
1188 	if (!latency_ns) {
1189 		*display_wm = *cursor_wm = 0;
1190 		return false;
1191 	}
1192 
1193 	crtc = intel_get_crtc_for_plane(dev, plane);
1194 	hdisplay = crtc->mode.hdisplay;
1195 	htotal = crtc->mode.htotal;
1196 	clock = crtc->mode.clock;
1197 	pixel_size = crtc->fb->bits_per_pixel / 8;
1198 
1199 	line_time_us = (htotal * 1000) / clock;
1200 	line_count = (latency_ns / line_time_us + 1000) / 1000;
1201 	line_size = hdisplay * pixel_size;
1202 
1203 	/* Use the minimum of the small and large buffer method for primary */
1204 	small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1205 	large = line_count * line_size;
1206 
1207 	entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1208 	*display_wm = entries + display->guard_size;
1209 
1210 	/* calculate the self-refresh watermark for display cursor */
1211 	entries = line_count * pixel_size * 64;
1212 	entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1213 	*cursor_wm = entries + cursor->guard_size;
1214 
1215 	return g4x_check_srwm(dev,
1216 			      *display_wm, *cursor_wm,
1217 			      display, cursor);
1218 }
1219 
1220 static bool vlv_compute_drain_latency(struct drm_device *dev,
1221 				     int plane,
1222 				     int *plane_prec_mult,
1223 				     int *plane_dl,
1224 				     int *cursor_prec_mult,
1225 				     int *cursor_dl)
1226 {
1227 	struct drm_crtc *crtc;
1228 	int clock, pixel_size;
1229 	int entries;
1230 
1231 	crtc = intel_get_crtc_for_plane(dev, plane);
1232 	if (!intel_crtc_active(crtc))
1233 		return false;
1234 
1235 	clock = crtc->mode.clock;	/* VESA DOT Clock */
1236 	pixel_size = crtc->fb->bits_per_pixel / 8;	/* BPP */
1237 
1238 	entries = (clock / 1000) * pixel_size;
1239 	*plane_prec_mult = (entries > 256) ?
1240 		DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1241 	*plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1242 						     pixel_size);
1243 
1244 	entries = (clock / 1000) * 4;	/* BPP is always 4 for cursor */
1245 	*cursor_prec_mult = (entries > 256) ?
1246 		DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1247 	*cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1248 
1249 	return true;
1250 }
1251 
1252 /*
1253  * Update drain latency registers of memory arbiter
1254  *
1255  * Valleyview SoC has a new memory arbiter and needs drain latency registers
1256  * to be programmed. Each plane has a drain latency multiplier and a drain
1257  * latency value.
1258  */
1259 
1260 static void vlv_update_drain_latency(struct drm_device *dev)
1261 {
1262 	struct drm_i915_private *dev_priv = dev->dev_private;
1263 	int planea_prec, planea_dl, planeb_prec, planeb_dl;
1264 	int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1265 	int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1266 							either 16 or 32 */
1267 
1268 	/* For plane A, Cursor A */
1269 	if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1270 				      &cursor_prec_mult, &cursora_dl)) {
1271 		cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1272 			DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1273 		planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1274 			DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1275 
1276 		I915_WRITE(VLV_DDL1, cursora_prec |
1277 				(cursora_dl << DDL_CURSORA_SHIFT) |
1278 				planea_prec | planea_dl);
1279 	}
1280 
1281 	/* For plane B, Cursor B */
1282 	if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1283 				      &cursor_prec_mult, &cursorb_dl)) {
1284 		cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1285 			DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1286 		planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1287 			DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1288 
1289 		I915_WRITE(VLV_DDL2, cursorb_prec |
1290 				(cursorb_dl << DDL_CURSORB_SHIFT) |
1291 				planeb_prec | planeb_dl);
1292 	}
1293 }
1294 
1295 #define single_plane_enabled(mask) is_power_of_2(mask)
1296 
1297 static void valleyview_update_wm(struct drm_device *dev)
1298 {
1299 	static const int sr_latency_ns = 12000;
1300 	struct drm_i915_private *dev_priv = dev->dev_private;
1301 	int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1302 	int plane_sr, cursor_sr;
1303 	int ignore_plane_sr, ignore_cursor_sr;
1304 	unsigned int enabled = 0;
1305 
1306 	vlv_update_drain_latency(dev);
1307 
1308 	if (g4x_compute_wm0(dev, 0,
1309 			    &valleyview_wm_info, latency_ns,
1310 			    &valleyview_cursor_wm_info, latency_ns,
1311 			    &planea_wm, &cursora_wm))
1312 		enabled |= 1;
1313 
1314 	if (g4x_compute_wm0(dev, 1,
1315 			    &valleyview_wm_info, latency_ns,
1316 			    &valleyview_cursor_wm_info, latency_ns,
1317 			    &planeb_wm, &cursorb_wm))
1318 		enabled |= 2;
1319 
1320 	if (single_plane_enabled(enabled) &&
1321 	    g4x_compute_srwm(dev, ffs(enabled) - 1,
1322 			     sr_latency_ns,
1323 			     &valleyview_wm_info,
1324 			     &valleyview_cursor_wm_info,
1325 			     &plane_sr, &ignore_cursor_sr) &&
1326 	    g4x_compute_srwm(dev, ffs(enabled) - 1,
1327 			     2*sr_latency_ns,
1328 			     &valleyview_wm_info,
1329 			     &valleyview_cursor_wm_info,
1330 			     &ignore_plane_sr, &cursor_sr)) {
1331 		I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
1332 	} else {
1333 		I915_WRITE(FW_BLC_SELF_VLV,
1334 			   I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
1335 		plane_sr = cursor_sr = 0;
1336 	}
1337 
1338 	DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1339 		      planea_wm, cursora_wm,
1340 		      planeb_wm, cursorb_wm,
1341 		      plane_sr, cursor_sr);
1342 
1343 	I915_WRITE(DSPFW1,
1344 		   (plane_sr << DSPFW_SR_SHIFT) |
1345 		   (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1346 		   (planeb_wm << DSPFW_PLANEB_SHIFT) |
1347 		   planea_wm);
1348 	I915_WRITE(DSPFW2,
1349 		   (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1350 		   (cursora_wm << DSPFW_CURSORA_SHIFT));
1351 	I915_WRITE(DSPFW3,
1352 		   (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1353 		   (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1354 }
1355 
1356 static void g4x_update_wm(struct drm_device *dev)
1357 {
1358 	static const int sr_latency_ns = 12000;
1359 	struct drm_i915_private *dev_priv = dev->dev_private;
1360 	int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1361 	int plane_sr, cursor_sr;
1362 	unsigned int enabled = 0;
1363 
1364 	if (g4x_compute_wm0(dev, 0,
1365 			    &g4x_wm_info, latency_ns,
1366 			    &g4x_cursor_wm_info, latency_ns,
1367 			    &planea_wm, &cursora_wm))
1368 		enabled |= 1;
1369 
1370 	if (g4x_compute_wm0(dev, 1,
1371 			    &g4x_wm_info, latency_ns,
1372 			    &g4x_cursor_wm_info, latency_ns,
1373 			    &planeb_wm, &cursorb_wm))
1374 		enabled |= 2;
1375 
1376 	if (single_plane_enabled(enabled) &&
1377 	    g4x_compute_srwm(dev, ffs(enabled) - 1,
1378 			     sr_latency_ns,
1379 			     &g4x_wm_info,
1380 			     &g4x_cursor_wm_info,
1381 			     &plane_sr, &cursor_sr)) {
1382 		I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1383 	} else {
1384 		I915_WRITE(FW_BLC_SELF,
1385 			   I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
1386 		plane_sr = cursor_sr = 0;
1387 	}
1388 
1389 	DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1390 		      planea_wm, cursora_wm,
1391 		      planeb_wm, cursorb_wm,
1392 		      plane_sr, cursor_sr);
1393 
1394 	I915_WRITE(DSPFW1,
1395 		   (plane_sr << DSPFW_SR_SHIFT) |
1396 		   (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1397 		   (planeb_wm << DSPFW_PLANEB_SHIFT) |
1398 		   planea_wm);
1399 	I915_WRITE(DSPFW2,
1400 		   (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1401 		   (cursora_wm << DSPFW_CURSORA_SHIFT));
1402 	/* HPLL off in SR has some issues on G4x... disable it */
1403 	I915_WRITE(DSPFW3,
1404 		   (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
1405 		   (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1406 }
1407 
1408 static void i965_update_wm(struct drm_device *dev)
1409 {
1410 	struct drm_i915_private *dev_priv = dev->dev_private;
1411 	struct drm_crtc *crtc;
1412 	int srwm = 1;
1413 	int cursor_sr = 16;
1414 
1415 	/* Calc sr entries for one plane configs */
1416 	crtc = single_enabled_crtc(dev);
1417 	if (crtc) {
1418 		/* self-refresh has much higher latency */
1419 		static const int sr_latency_ns = 12000;
1420 		int clock = crtc->mode.clock;
1421 		int htotal = crtc->mode.htotal;
1422 		int hdisplay = crtc->mode.hdisplay;
1423 		int pixel_size = crtc->fb->bits_per_pixel / 8;
1424 		unsigned long line_time_us;
1425 		int entries;
1426 
1427 		line_time_us = ((htotal * 1000) / clock);
1428 
1429 		/* Use ns/us then divide to preserve precision */
1430 		entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1431 			pixel_size * hdisplay;
1432 		entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1433 		srwm = I965_FIFO_SIZE - entries;
1434 		if (srwm < 0)
1435 			srwm = 1;
1436 		srwm &= 0x1ff;
1437 		DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1438 			      entries, srwm);
1439 
1440 		entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1441 			pixel_size * 64;
1442 		entries = DIV_ROUND_UP(entries,
1443 					  i965_cursor_wm_info.cacheline_size);
1444 		cursor_sr = i965_cursor_wm_info.fifo_size -
1445 			(entries + i965_cursor_wm_info.guard_size);
1446 
1447 		if (cursor_sr > i965_cursor_wm_info.max_wm)
1448 			cursor_sr = i965_cursor_wm_info.max_wm;
1449 
1450 		DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1451 			      "cursor %d\n", srwm, cursor_sr);
1452 
1453 		if (IS_CRESTLINE(dev))
1454 			I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1455 	} else {
1456 		/* Turn off self refresh if both pipes are enabled */
1457 		if (IS_CRESTLINE(dev))
1458 			I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1459 				   & ~FW_BLC_SELF_EN);
1460 	}
1461 
1462 	DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1463 		      srwm);
1464 
1465 	/* 965 has limitations... */
1466 	I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1467 		   (8 << 16) | (8 << 8) | (8 << 0));
1468 	I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1469 	/* update cursor SR watermark */
1470 	I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1471 }
1472 
1473 static void i9xx_update_wm(struct drm_device *dev)
1474 {
1475 	struct drm_i915_private *dev_priv = dev->dev_private;
1476 	const struct intel_watermark_params *wm_info;
1477 	uint32_t fwater_lo;
1478 	uint32_t fwater_hi;
1479 	int cwm, srwm = 1;
1480 	int fifo_size;
1481 	int planea_wm, planeb_wm;
1482 	struct drm_crtc *crtc, *enabled = NULL;
1483 
1484 	if (IS_I945GM(dev))
1485 		wm_info = &i945_wm_info;
1486 	else if (!IS_GEN2(dev))
1487 		wm_info = &i915_wm_info;
1488 	else
1489 		wm_info = &i855_wm_info;
1490 
1491 	fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1492 	crtc = intel_get_crtc_for_plane(dev, 0);
1493 	if (intel_crtc_active(crtc)) {
1494 		int cpp = crtc->fb->bits_per_pixel / 8;
1495 		if (IS_GEN2(dev))
1496 			cpp = 4;
1497 
1498 		planea_wm = intel_calculate_wm(crtc->mode.clock,
1499 					       wm_info, fifo_size, cpp,
1500 					       latency_ns);
1501 		enabled = crtc;
1502 	} else
1503 		planea_wm = fifo_size - wm_info->guard_size;
1504 
1505 	fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1506 	crtc = intel_get_crtc_for_plane(dev, 1);
1507 	if (intel_crtc_active(crtc)) {
1508 		int cpp = crtc->fb->bits_per_pixel / 8;
1509 		if (IS_GEN2(dev))
1510 			cpp = 4;
1511 
1512 		planeb_wm = intel_calculate_wm(crtc->mode.clock,
1513 					       wm_info, fifo_size, cpp,
1514 					       latency_ns);
1515 		if (enabled == NULL)
1516 			enabled = crtc;
1517 		else
1518 			enabled = NULL;
1519 	} else
1520 		planeb_wm = fifo_size - wm_info->guard_size;
1521 
1522 	DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1523 
1524 	/*
1525 	 * Overlay gets an aggressive default since video jitter is bad.
1526 	 */
1527 	cwm = 2;
1528 
1529 	/* Play safe and disable self-refresh before adjusting watermarks. */
1530 	if (IS_I945G(dev) || IS_I945GM(dev))
1531 		I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1532 	else if (IS_I915GM(dev))
1533 		I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN);
1534 
1535 	/* Calc sr entries for one plane configs */
1536 	if (HAS_FW_BLC(dev) && enabled) {
1537 		/* self-refresh has much higher latency */
1538 		static const int sr_latency_ns = 6000;
1539 		int clock = enabled->mode.clock;
1540 		int htotal = enabled->mode.htotal;
1541 		int hdisplay = enabled->mode.hdisplay;
1542 		int pixel_size = enabled->fb->bits_per_pixel / 8;
1543 		unsigned long line_time_us;
1544 		int entries;
1545 
1546 		line_time_us = (htotal * 1000) / clock;
1547 
1548 		/* Use ns/us then divide to preserve precision */
1549 		entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1550 			pixel_size * hdisplay;
1551 		entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1552 		DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1553 		srwm = wm_info->fifo_size - entries;
1554 		if (srwm < 0)
1555 			srwm = 1;
1556 
1557 		if (IS_I945G(dev) || IS_I945GM(dev))
1558 			I915_WRITE(FW_BLC_SELF,
1559 				   FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1560 		else if (IS_I915GM(dev))
1561 			I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1562 	}
1563 
1564 	DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1565 		      planea_wm, planeb_wm, cwm, srwm);
1566 
1567 	fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1568 	fwater_hi = (cwm & 0x1f);
1569 
1570 	/* Set request length to 8 cachelines per fetch */
1571 	fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1572 	fwater_hi = fwater_hi | (1 << 8);
1573 
1574 	I915_WRITE(FW_BLC, fwater_lo);
1575 	I915_WRITE(FW_BLC2, fwater_hi);
1576 
1577 	if (HAS_FW_BLC(dev)) {
1578 		if (enabled) {
1579 			if (IS_I945G(dev) || IS_I945GM(dev))
1580 				I915_WRITE(FW_BLC_SELF,
1581 					   FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1582 			else if (IS_I915GM(dev))
1583 				I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN);
1584 			DRM_DEBUG_KMS("memory self refresh enabled\n");
1585 		} else
1586 			DRM_DEBUG_KMS("memory self refresh disabled\n");
1587 	}
1588 }
1589 
1590 static void i830_update_wm(struct drm_device *dev)
1591 {
1592 	struct drm_i915_private *dev_priv = dev->dev_private;
1593 	struct drm_crtc *crtc;
1594 	uint32_t fwater_lo;
1595 	int planea_wm;
1596 
1597 	crtc = single_enabled_crtc(dev);
1598 	if (crtc == NULL)
1599 		return;
1600 
1601 	planea_wm = intel_calculate_wm(crtc->mode.clock, &i830_wm_info,
1602 				       dev_priv->display.get_fifo_size(dev, 0),
1603 				       4, latency_ns);
1604 	fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1605 	fwater_lo |= (3<<8) | planea_wm;
1606 
1607 	DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1608 
1609 	I915_WRITE(FW_BLC, fwater_lo);
1610 }
1611 
1612 #define ILK_LP0_PLANE_LATENCY		700
1613 #define ILK_LP0_CURSOR_LATENCY		1300
1614 
1615 /*
1616  * Check the wm result.
1617  *
1618  * If any calculated watermark values is larger than the maximum value that
1619  * can be programmed into the associated watermark register, that watermark
1620  * must be disabled.
1621  */
1622 static bool ironlake_check_srwm(struct drm_device *dev, int level,
1623 				int fbc_wm, int display_wm, int cursor_wm,
1624 				const struct intel_watermark_params *display,
1625 				const struct intel_watermark_params *cursor)
1626 {
1627 	struct drm_i915_private *dev_priv = dev->dev_private;
1628 
1629 	DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d,"
1630 		      " cursor %d\n", level, display_wm, fbc_wm, cursor_wm);
1631 
1632 	if (fbc_wm > SNB_FBC_MAX_SRWM) {
1633 		DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n",
1634 			      fbc_wm, SNB_FBC_MAX_SRWM, level);
1635 
1636 		/* fbc has it's own way to disable FBC WM */
1637 		I915_WRITE(DISP_ARB_CTL,
1638 			   I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS);
1639 		return false;
1640 	}
1641 
1642 	if (display_wm > display->max_wm) {
1643 		DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n",
1644 			      display_wm, SNB_DISPLAY_MAX_SRWM, level);
1645 		return false;
1646 	}
1647 
1648 	if (cursor_wm > cursor->max_wm) {
1649 		DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n",
1650 			      cursor_wm, SNB_CURSOR_MAX_SRWM, level);
1651 		return false;
1652 	}
1653 
1654 	if (!(fbc_wm || display_wm || cursor_wm)) {
1655 		DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level);
1656 		return false;
1657 	}
1658 
1659 	return true;
1660 }
1661 
1662 /*
1663  * Compute watermark values of WM[1-3],
1664  */
1665 static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane,
1666 				  int latency_ns,
1667 				  const struct intel_watermark_params *display,
1668 				  const struct intel_watermark_params *cursor,
1669 				  int *fbc_wm, int *display_wm, int *cursor_wm)
1670 {
1671 	struct drm_crtc *crtc;
1672 	unsigned long line_time_us;
1673 	int hdisplay, htotal, pixel_size, clock;
1674 	int line_count, line_size;
1675 	int small, large;
1676 	int entries;
1677 
1678 	if (!latency_ns) {
1679 		*fbc_wm = *display_wm = *cursor_wm = 0;
1680 		return false;
1681 	}
1682 
1683 	crtc = intel_get_crtc_for_plane(dev, plane);
1684 	hdisplay = crtc->mode.hdisplay;
1685 	htotal = crtc->mode.htotal;
1686 	clock = crtc->mode.clock;
1687 	pixel_size = crtc->fb->bits_per_pixel / 8;
1688 
1689 	line_time_us = (htotal * 1000) / clock;
1690 	line_count = (latency_ns / line_time_us + 1000) / 1000;
1691 	line_size = hdisplay * pixel_size;
1692 
1693 	/* Use the minimum of the small and large buffer method for primary */
1694 	small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1695 	large = line_count * line_size;
1696 
1697 	entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1698 	*display_wm = entries + display->guard_size;
1699 
1700 	/*
1701 	 * Spec says:
1702 	 * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2
1703 	 */
1704 	*fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2;
1705 
1706 	/* calculate the self-refresh watermark for display cursor */
1707 	entries = line_count * pixel_size * 64;
1708 	entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1709 	*cursor_wm = entries + cursor->guard_size;
1710 
1711 	return ironlake_check_srwm(dev, level,
1712 				   *fbc_wm, *display_wm, *cursor_wm,
1713 				   display, cursor);
1714 }
1715 
1716 static void ironlake_update_wm(struct drm_device *dev)
1717 {
1718 	struct drm_i915_private *dev_priv = dev->dev_private;
1719 	int fbc_wm, plane_wm, cursor_wm;
1720 	unsigned int enabled;
1721 
1722 	enabled = 0;
1723 	if (g4x_compute_wm0(dev, 0,
1724 			    &ironlake_display_wm_info,
1725 			    ILK_LP0_PLANE_LATENCY,
1726 			    &ironlake_cursor_wm_info,
1727 			    ILK_LP0_CURSOR_LATENCY,
1728 			    &plane_wm, &cursor_wm)) {
1729 		I915_WRITE(WM0_PIPEA_ILK,
1730 			   (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1731 		DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1732 			      " plane %d, " "cursor: %d\n",
1733 			      plane_wm, cursor_wm);
1734 		enabled |= 1;
1735 	}
1736 
1737 	if (g4x_compute_wm0(dev, 1,
1738 			    &ironlake_display_wm_info,
1739 			    ILK_LP0_PLANE_LATENCY,
1740 			    &ironlake_cursor_wm_info,
1741 			    ILK_LP0_CURSOR_LATENCY,
1742 			    &plane_wm, &cursor_wm)) {
1743 		I915_WRITE(WM0_PIPEB_ILK,
1744 			   (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1745 		DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1746 			      " plane %d, cursor: %d\n",
1747 			      plane_wm, cursor_wm);
1748 		enabled |= 2;
1749 	}
1750 
1751 	/*
1752 	 * Calculate and update the self-refresh watermark only when one
1753 	 * display plane is used.
1754 	 */
1755 	I915_WRITE(WM3_LP_ILK, 0);
1756 	I915_WRITE(WM2_LP_ILK, 0);
1757 	I915_WRITE(WM1_LP_ILK, 0);
1758 
1759 	if (!single_plane_enabled(enabled))
1760 		return;
1761 	enabled = ffs(enabled) - 1;
1762 
1763 	/* WM1 */
1764 	if (!ironlake_compute_srwm(dev, 1, enabled,
1765 				   ILK_READ_WM1_LATENCY() * 500,
1766 				   &ironlake_display_srwm_info,
1767 				   &ironlake_cursor_srwm_info,
1768 				   &fbc_wm, &plane_wm, &cursor_wm))
1769 		return;
1770 
1771 	I915_WRITE(WM1_LP_ILK,
1772 		   WM1_LP_SR_EN |
1773 		   (ILK_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1774 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1775 		   (plane_wm << WM1_LP_SR_SHIFT) |
1776 		   cursor_wm);
1777 
1778 	/* WM2 */
1779 	if (!ironlake_compute_srwm(dev, 2, enabled,
1780 				   ILK_READ_WM2_LATENCY() * 500,
1781 				   &ironlake_display_srwm_info,
1782 				   &ironlake_cursor_srwm_info,
1783 				   &fbc_wm, &plane_wm, &cursor_wm))
1784 		return;
1785 
1786 	I915_WRITE(WM2_LP_ILK,
1787 		   WM2_LP_EN |
1788 		   (ILK_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1789 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1790 		   (plane_wm << WM1_LP_SR_SHIFT) |
1791 		   cursor_wm);
1792 
1793 	/*
1794 	 * WM3 is unsupported on ILK, probably because we don't have latency
1795 	 * data for that power state
1796 	 */
1797 }
1798 
1799 static void sandybridge_update_wm(struct drm_device *dev)
1800 {
1801 	struct drm_i915_private *dev_priv = dev->dev_private;
1802 	int latency = SNB_READ_WM0_LATENCY() * 100;	/* In unit 0.1us */
1803 	u32 val;
1804 	int fbc_wm, plane_wm, cursor_wm;
1805 	unsigned int enabled;
1806 
1807 	enabled = 0;
1808 	if (g4x_compute_wm0(dev, 0,
1809 			    &sandybridge_display_wm_info, latency,
1810 			    &sandybridge_cursor_wm_info, latency,
1811 			    &plane_wm, &cursor_wm)) {
1812 		val = I915_READ(WM0_PIPEA_ILK);
1813 		val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1814 		I915_WRITE(WM0_PIPEA_ILK, val |
1815 			   ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1816 		DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1817 			      " plane %d, " "cursor: %d\n",
1818 			      plane_wm, cursor_wm);
1819 		enabled |= 1;
1820 	}
1821 
1822 	if (g4x_compute_wm0(dev, 1,
1823 			    &sandybridge_display_wm_info, latency,
1824 			    &sandybridge_cursor_wm_info, latency,
1825 			    &plane_wm, &cursor_wm)) {
1826 		val = I915_READ(WM0_PIPEB_ILK);
1827 		val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1828 		I915_WRITE(WM0_PIPEB_ILK, val |
1829 			   ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1830 		DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1831 			      " plane %d, cursor: %d\n",
1832 			      plane_wm, cursor_wm);
1833 		enabled |= 2;
1834 	}
1835 
1836 	/*
1837 	 * Calculate and update the self-refresh watermark only when one
1838 	 * display plane is used.
1839 	 *
1840 	 * SNB support 3 levels of watermark.
1841 	 *
1842 	 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1843 	 * and disabled in the descending order
1844 	 *
1845 	 */
1846 	I915_WRITE(WM3_LP_ILK, 0);
1847 	I915_WRITE(WM2_LP_ILK, 0);
1848 	I915_WRITE(WM1_LP_ILK, 0);
1849 
1850 	if (!single_plane_enabled(enabled) ||
1851 	    dev_priv->sprite_scaling_enabled)
1852 		return;
1853 	enabled = ffs(enabled) - 1;
1854 
1855 	/* WM1 */
1856 	if (!ironlake_compute_srwm(dev, 1, enabled,
1857 				   SNB_READ_WM1_LATENCY() * 500,
1858 				   &sandybridge_display_srwm_info,
1859 				   &sandybridge_cursor_srwm_info,
1860 				   &fbc_wm, &plane_wm, &cursor_wm))
1861 		return;
1862 
1863 	I915_WRITE(WM1_LP_ILK,
1864 		   WM1_LP_SR_EN |
1865 		   (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1866 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1867 		   (plane_wm << WM1_LP_SR_SHIFT) |
1868 		   cursor_wm);
1869 
1870 	/* WM2 */
1871 	if (!ironlake_compute_srwm(dev, 2, enabled,
1872 				   SNB_READ_WM2_LATENCY() * 500,
1873 				   &sandybridge_display_srwm_info,
1874 				   &sandybridge_cursor_srwm_info,
1875 				   &fbc_wm, &plane_wm, &cursor_wm))
1876 		return;
1877 
1878 	I915_WRITE(WM2_LP_ILK,
1879 		   WM2_LP_EN |
1880 		   (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1881 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1882 		   (plane_wm << WM1_LP_SR_SHIFT) |
1883 		   cursor_wm);
1884 
1885 	/* WM3 */
1886 	if (!ironlake_compute_srwm(dev, 3, enabled,
1887 				   SNB_READ_WM3_LATENCY() * 500,
1888 				   &sandybridge_display_srwm_info,
1889 				   &sandybridge_cursor_srwm_info,
1890 				   &fbc_wm, &plane_wm, &cursor_wm))
1891 		return;
1892 
1893 	I915_WRITE(WM3_LP_ILK,
1894 		   WM3_LP_EN |
1895 		   (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1896 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1897 		   (plane_wm << WM1_LP_SR_SHIFT) |
1898 		   cursor_wm);
1899 }
1900 
1901 static void ivybridge_update_wm(struct drm_device *dev)
1902 {
1903 	struct drm_i915_private *dev_priv = dev->dev_private;
1904 	int latency = SNB_READ_WM0_LATENCY() * 100;	/* In unit 0.1us */
1905 	u32 val;
1906 	int fbc_wm, plane_wm, cursor_wm;
1907 	int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm;
1908 	unsigned int enabled;
1909 
1910 	enabled = 0;
1911 	if (g4x_compute_wm0(dev, 0,
1912 			    &sandybridge_display_wm_info, latency,
1913 			    &sandybridge_cursor_wm_info, latency,
1914 			    &plane_wm, &cursor_wm)) {
1915 		val = I915_READ(WM0_PIPEA_ILK);
1916 		val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1917 		I915_WRITE(WM0_PIPEA_ILK, val |
1918 			   ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1919 		DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1920 			      " plane %d, " "cursor: %d\n",
1921 			      plane_wm, cursor_wm);
1922 		enabled |= 1;
1923 	}
1924 
1925 	if (g4x_compute_wm0(dev, 1,
1926 			    &sandybridge_display_wm_info, latency,
1927 			    &sandybridge_cursor_wm_info, latency,
1928 			    &plane_wm, &cursor_wm)) {
1929 		val = I915_READ(WM0_PIPEB_ILK);
1930 		val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1931 		I915_WRITE(WM0_PIPEB_ILK, val |
1932 			   ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1933 		DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1934 			      " plane %d, cursor: %d\n",
1935 			      plane_wm, cursor_wm);
1936 		enabled |= 2;
1937 	}
1938 
1939 	if (g4x_compute_wm0(dev, 2,
1940 			    &sandybridge_display_wm_info, latency,
1941 			    &sandybridge_cursor_wm_info, latency,
1942 			    &plane_wm, &cursor_wm)) {
1943 		val = I915_READ(WM0_PIPEC_IVB);
1944 		val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1945 		I915_WRITE(WM0_PIPEC_IVB, val |
1946 			   ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1947 		DRM_DEBUG_KMS("FIFO watermarks For pipe C -"
1948 			      " plane %d, cursor: %d\n",
1949 			      plane_wm, cursor_wm);
1950 		enabled |= 3;
1951 	}
1952 
1953 	/*
1954 	 * Calculate and update the self-refresh watermark only when one
1955 	 * display plane is used.
1956 	 *
1957 	 * SNB support 3 levels of watermark.
1958 	 *
1959 	 * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1960 	 * and disabled in the descending order
1961 	 *
1962 	 */
1963 	I915_WRITE(WM3_LP_ILK, 0);
1964 	I915_WRITE(WM2_LP_ILK, 0);
1965 	I915_WRITE(WM1_LP_ILK, 0);
1966 
1967 	if (!single_plane_enabled(enabled) ||
1968 	    dev_priv->sprite_scaling_enabled)
1969 		return;
1970 	enabled = ffs(enabled) - 1;
1971 
1972 	/* WM1 */
1973 	if (!ironlake_compute_srwm(dev, 1, enabled,
1974 				   SNB_READ_WM1_LATENCY() * 500,
1975 				   &sandybridge_display_srwm_info,
1976 				   &sandybridge_cursor_srwm_info,
1977 				   &fbc_wm, &plane_wm, &cursor_wm))
1978 		return;
1979 
1980 	I915_WRITE(WM1_LP_ILK,
1981 		   WM1_LP_SR_EN |
1982 		   (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1983 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1984 		   (plane_wm << WM1_LP_SR_SHIFT) |
1985 		   cursor_wm);
1986 
1987 	/* WM2 */
1988 	if (!ironlake_compute_srwm(dev, 2, enabled,
1989 				   SNB_READ_WM2_LATENCY() * 500,
1990 				   &sandybridge_display_srwm_info,
1991 				   &sandybridge_cursor_srwm_info,
1992 				   &fbc_wm, &plane_wm, &cursor_wm))
1993 		return;
1994 
1995 	I915_WRITE(WM2_LP_ILK,
1996 		   WM2_LP_EN |
1997 		   (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) |
1998 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
1999 		   (plane_wm << WM1_LP_SR_SHIFT) |
2000 		   cursor_wm);
2001 
2002 	/* WM3, note we have to correct the cursor latency */
2003 	if (!ironlake_compute_srwm(dev, 3, enabled,
2004 				   SNB_READ_WM3_LATENCY() * 500,
2005 				   &sandybridge_display_srwm_info,
2006 				   &sandybridge_cursor_srwm_info,
2007 				   &fbc_wm, &plane_wm, &ignore_cursor_wm) ||
2008 	    !ironlake_compute_srwm(dev, 3, enabled,
2009 				   2 * SNB_READ_WM3_LATENCY() * 500,
2010 				   &sandybridge_display_srwm_info,
2011 				   &sandybridge_cursor_srwm_info,
2012 				   &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm))
2013 		return;
2014 
2015 	I915_WRITE(WM3_LP_ILK,
2016 		   WM3_LP_EN |
2017 		   (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) |
2018 		   (fbc_wm << WM1_LP_FBC_SHIFT) |
2019 		   (plane_wm << WM1_LP_SR_SHIFT) |
2020 		   cursor_wm);
2021 }
2022 
2023 static void
2024 haswell_update_linetime_wm(struct drm_device *dev, int pipe,
2025 				 struct drm_display_mode *mode)
2026 {
2027 	struct drm_i915_private *dev_priv = dev->dev_private;
2028 	u32 temp;
2029 
2030 	temp = I915_READ(PIPE_WM_LINETIME(pipe));
2031 	temp &= ~PIPE_WM_LINETIME_MASK;
2032 
2033 	/* The WM are computed with base on how long it takes to fill a single
2034 	 * row at the given clock rate, multiplied by 8.
2035 	 * */
2036 	temp |= PIPE_WM_LINETIME_TIME(
2037 		((mode->crtc_hdisplay * 1000) / mode->clock) * 8);
2038 
2039 	/* IPS watermarks are only used by pipe A, and are ignored by
2040 	 * pipes B and C.  They are calculated similarly to the common
2041 	 * linetime values, except that we are using CD clock frequency
2042 	 * in MHz instead of pixel rate for the division.
2043 	 *
2044 	 * This is a placeholder for the IPS watermark calculation code.
2045 	 */
2046 
2047 	I915_WRITE(PIPE_WM_LINETIME(pipe), temp);
2048 }
2049 
2050 static bool
2051 sandybridge_compute_sprite_wm(struct drm_device *dev, int plane,
2052 			      uint32_t sprite_width, int pixel_size,
2053 			      const struct intel_watermark_params *display,
2054 			      int display_latency_ns, int *sprite_wm)
2055 {
2056 	struct drm_crtc *crtc;
2057 	int clock;
2058 	int entries, tlb_miss;
2059 
2060 	crtc = intel_get_crtc_for_plane(dev, plane);
2061 	if (!intel_crtc_active(crtc)) {
2062 		*sprite_wm = display->guard_size;
2063 		return false;
2064 	}
2065 
2066 	clock = crtc->mode.clock;
2067 
2068 	/* Use the small buffer method to calculate the sprite watermark */
2069 	entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
2070 	tlb_miss = display->fifo_size*display->cacheline_size -
2071 		sprite_width * 8;
2072 	if (tlb_miss > 0)
2073 		entries += tlb_miss;
2074 	entries = DIV_ROUND_UP(entries, display->cacheline_size);
2075 	*sprite_wm = entries + display->guard_size;
2076 	if (*sprite_wm > (int)display->max_wm)
2077 		*sprite_wm = display->max_wm;
2078 
2079 	return true;
2080 }
2081 
2082 static bool
2083 sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane,
2084 				uint32_t sprite_width, int pixel_size,
2085 				const struct intel_watermark_params *display,
2086 				int latency_ns, int *sprite_wm)
2087 {
2088 	struct drm_crtc *crtc;
2089 	unsigned long line_time_us;
2090 	int clock;
2091 	int line_count, line_size;
2092 	int small, large;
2093 	int entries;
2094 
2095 	if (!latency_ns) {
2096 		*sprite_wm = 0;
2097 		return false;
2098 	}
2099 
2100 	crtc = intel_get_crtc_for_plane(dev, plane);
2101 	clock = crtc->mode.clock;
2102 	if (!clock) {
2103 		*sprite_wm = 0;
2104 		return false;
2105 	}
2106 
2107 	line_time_us = (sprite_width * 1000) / clock;
2108 	if (!line_time_us) {
2109 		*sprite_wm = 0;
2110 		return false;
2111 	}
2112 
2113 	line_count = (latency_ns / line_time_us + 1000) / 1000;
2114 	line_size = sprite_width * pixel_size;
2115 
2116 	/* Use the minimum of the small and large buffer method for primary */
2117 	small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
2118 	large = line_count * line_size;
2119 
2120 	entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
2121 	*sprite_wm = entries + display->guard_size;
2122 
2123 	return *sprite_wm > 0x3ff ? false : true;
2124 }
2125 
2126 static void sandybridge_update_sprite_wm(struct drm_device *dev, int pipe,
2127 					 uint32_t sprite_width, int pixel_size)
2128 {
2129 	struct drm_i915_private *dev_priv = dev->dev_private;
2130 	int latency = SNB_READ_WM0_LATENCY() * 100;	/* In unit 0.1us */
2131 	u32 val;
2132 	int sprite_wm, reg;
2133 	int ret;
2134 
2135 	switch (pipe) {
2136 	case 0:
2137 		reg = WM0_PIPEA_ILK;
2138 		break;
2139 	case 1:
2140 		reg = WM0_PIPEB_ILK;
2141 		break;
2142 	case 2:
2143 		reg = WM0_PIPEC_IVB;
2144 		break;
2145 	default:
2146 		return; /* bad pipe */
2147 	}
2148 
2149 	ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size,
2150 					    &sandybridge_display_wm_info,
2151 					    latency, &sprite_wm);
2152 	if (!ret) {
2153 		DRM_DEBUG_KMS("failed to compute sprite wm for pipe %d\n",
2154 			      pipe);
2155 		return;
2156 	}
2157 
2158 	val = I915_READ(reg);
2159 	val &= ~WM0_PIPE_SPRITE_MASK;
2160 	I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT));
2161 	DRM_DEBUG_KMS("sprite watermarks For pipe %d - %d\n", pipe, sprite_wm);
2162 
2163 
2164 	ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2165 					      pixel_size,
2166 					      &sandybridge_display_srwm_info,
2167 					      SNB_READ_WM1_LATENCY() * 500,
2168 					      &sprite_wm);
2169 	if (!ret) {
2170 		DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %d\n",
2171 			      pipe);
2172 		return;
2173 	}
2174 	I915_WRITE(WM1S_LP_ILK, sprite_wm);
2175 
2176 	/* Only IVB has two more LP watermarks for sprite */
2177 	if (!IS_IVYBRIDGE(dev))
2178 		return;
2179 
2180 	ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2181 					      pixel_size,
2182 					      &sandybridge_display_srwm_info,
2183 					      SNB_READ_WM2_LATENCY() * 500,
2184 					      &sprite_wm);
2185 	if (!ret) {
2186 		DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %d\n",
2187 			      pipe);
2188 		return;
2189 	}
2190 	I915_WRITE(WM2S_LP_IVB, sprite_wm);
2191 
2192 	ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
2193 					      pixel_size,
2194 					      &sandybridge_display_srwm_info,
2195 					      SNB_READ_WM3_LATENCY() * 500,
2196 					      &sprite_wm);
2197 	if (!ret) {
2198 		DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %d\n",
2199 			      pipe);
2200 		return;
2201 	}
2202 	I915_WRITE(WM3S_LP_IVB, sprite_wm);
2203 }
2204 
2205 /**
2206  * intel_update_watermarks - update FIFO watermark values based on current modes
2207  *
2208  * Calculate watermark values for the various WM regs based on current mode
2209  * and plane configuration.
2210  *
2211  * There are several cases to deal with here:
2212  *   - normal (i.e. non-self-refresh)
2213  *   - self-refresh (SR) mode
2214  *   - lines are large relative to FIFO size (buffer can hold up to 2)
2215  *   - lines are small relative to FIFO size (buffer can hold more than 2
2216  *     lines), so need to account for TLB latency
2217  *
2218  *   The normal calculation is:
2219  *     watermark = dotclock * bytes per pixel * latency
2220  *   where latency is platform & configuration dependent (we assume pessimal
2221  *   values here).
2222  *
2223  *   The SR calculation is:
2224  *     watermark = (trunc(latency/line time)+1) * surface width *
2225  *       bytes per pixel
2226  *   where
2227  *     line time = htotal / dotclock
2228  *     surface width = hdisplay for normal plane and 64 for cursor
2229  *   and latency is assumed to be high, as above.
2230  *
2231  * The final value programmed to the register should always be rounded up,
2232  * and include an extra 2 entries to account for clock crossings.
2233  *
2234  * We don't use the sprite, so we can ignore that.  And on Crestline we have
2235  * to set the non-SR watermarks to 8.
2236  */
2237 void intel_update_watermarks(struct drm_device *dev)
2238 {
2239 	struct drm_i915_private *dev_priv = dev->dev_private;
2240 
2241 	if (dev_priv->display.update_wm)
2242 		dev_priv->display.update_wm(dev);
2243 }
2244 
2245 void intel_update_linetime_watermarks(struct drm_device *dev,
2246 		int pipe, struct drm_display_mode *mode)
2247 {
2248 	struct drm_i915_private *dev_priv = dev->dev_private;
2249 
2250 	if (dev_priv->display.update_linetime_wm)
2251 		dev_priv->display.update_linetime_wm(dev, pipe, mode);
2252 }
2253 
2254 void intel_update_sprite_watermarks(struct drm_device *dev, int pipe,
2255 				    uint32_t sprite_width, int pixel_size)
2256 {
2257 	struct drm_i915_private *dev_priv = dev->dev_private;
2258 
2259 	if (dev_priv->display.update_sprite_wm)
2260 		dev_priv->display.update_sprite_wm(dev, pipe, sprite_width,
2261 						   pixel_size);
2262 }
2263 
2264 static struct drm_i915_gem_object *
2265 intel_alloc_context_page(struct drm_device *dev)
2266 {
2267 	struct drm_i915_gem_object *ctx;
2268 	int ret;
2269 
2270 	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2271 
2272 	ctx = i915_gem_alloc_object(dev, 4096);
2273 	if (!ctx) {
2274 		DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
2275 		return NULL;
2276 	}
2277 
2278 	ret = i915_gem_object_pin(ctx, 4096, true, false);
2279 	if (ret) {
2280 		DRM_ERROR("failed to pin power context: %d\n", ret);
2281 		goto err_unref;
2282 	}
2283 
2284 	ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
2285 	if (ret) {
2286 		DRM_ERROR("failed to set-domain on power context: %d\n", ret);
2287 		goto err_unpin;
2288 	}
2289 
2290 	return ctx;
2291 
2292 err_unpin:
2293 	i915_gem_object_unpin(ctx);
2294 err_unref:
2295 	drm_gem_object_unreference(&ctx->base);
2296 	mutex_unlock(&dev->struct_mutex);
2297 	return NULL;
2298 }
2299 
2300 /**
2301  * Lock protecting IPS related data structures
2302  */
2303 #ifdef __NetBSD__
2304 spinlock_t mchdev_lock;
2305 #else
2306 DEFINE_SPINLOCK(mchdev_lock);
2307 #endif
2308 
2309 /* Global for IPS driver to get at the current i915 device. Protected by
2310  * mchdev_lock. */
2311 static struct drm_i915_private *i915_mch_dev;
2312 
2313 bool ironlake_set_drps(struct drm_device *dev, u8 val)
2314 {
2315 	struct drm_i915_private *dev_priv = dev->dev_private;
2316 	u16 rgvswctl;
2317 
2318 	assert_spin_locked(&mchdev_lock);
2319 
2320 	rgvswctl = I915_READ16(MEMSWCTL);
2321 	if (rgvswctl & MEMCTL_CMD_STS) {
2322 		DRM_DEBUG("gpu busy, RCS change rejected\n");
2323 		return false; /* still busy with another command */
2324 	}
2325 
2326 	rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
2327 		(val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
2328 	I915_WRITE16(MEMSWCTL, rgvswctl);
2329 	POSTING_READ16(MEMSWCTL);
2330 
2331 	rgvswctl |= MEMCTL_CMD_STS;
2332 	I915_WRITE16(MEMSWCTL, rgvswctl);
2333 
2334 	return true;
2335 }
2336 
2337 static void ironlake_enable_drps(struct drm_device *dev)
2338 {
2339 	struct drm_i915_private *dev_priv = dev->dev_private;
2340 	u32 rgvmodectl = I915_READ(MEMMODECTL);
2341 	u8 fmax, fmin, fstart, vstart;
2342 
2343 	spin_lock_irq(&mchdev_lock);
2344 
2345 	/* Enable temp reporting */
2346 	I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
2347 	I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
2348 
2349 	/* 100ms RC evaluation intervals */
2350 	I915_WRITE(RCUPEI, 100000);
2351 	I915_WRITE(RCDNEI, 100000);
2352 
2353 	/* Set max/min thresholds to 90ms and 80ms respectively */
2354 	I915_WRITE(RCBMAXAVG, 90000);
2355 	I915_WRITE(RCBMINAVG, 80000);
2356 
2357 	I915_WRITE(MEMIHYST, 1);
2358 
2359 	/* Set up min, max, and cur for interrupt handling */
2360 	fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
2361 	fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
2362 	fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
2363 		MEMMODE_FSTART_SHIFT;
2364 
2365 	vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
2366 		PXVFREQ_PX_SHIFT;
2367 
2368 	dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
2369 	dev_priv->ips.fstart = fstart;
2370 
2371 	dev_priv->ips.max_delay = fstart;
2372 	dev_priv->ips.min_delay = fmin;
2373 	dev_priv->ips.cur_delay = fstart;
2374 
2375 	DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
2376 			 fmax, fmin, fstart);
2377 
2378 	I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
2379 
2380 	/*
2381 	 * Interrupts will be enabled in ironlake_irq_postinstall
2382 	 */
2383 
2384 	I915_WRITE(VIDSTART, vstart);
2385 	POSTING_READ(VIDSTART);
2386 
2387 	rgvmodectl |= MEMMODE_SWMODE_EN;
2388 	I915_WRITE(MEMMODECTL, rgvmodectl);
2389 
2390 	if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
2391 		DRM_ERROR("stuck trying to change perf mode\n");
2392 	mdelay(1);
2393 
2394 	ironlake_set_drps(dev, fstart);
2395 
2396 	dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
2397 		I915_READ(0x112e0);
2398 	dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
2399 	dev_priv->ips.last_count2 = I915_READ(0x112f4);
2400 	getrawmonotonic(&dev_priv->ips.last_time2);
2401 
2402 	spin_unlock_irq(&mchdev_lock);
2403 }
2404 
2405 static void ironlake_disable_drps(struct drm_device *dev)
2406 {
2407 	struct drm_i915_private *dev_priv = dev->dev_private;
2408 	u16 rgvswctl;
2409 
2410 	spin_lock_irq(&mchdev_lock);
2411 
2412 	rgvswctl = I915_READ16(MEMSWCTL);
2413 
2414 	/* Ack interrupts, disable EFC interrupt */
2415 	I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
2416 	I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
2417 	I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
2418 	I915_WRITE(DEIIR, DE_PCU_EVENT);
2419 	I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
2420 
2421 	/* Go back to the starting frequency */
2422 	ironlake_set_drps(dev, dev_priv->ips.fstart);
2423 	mdelay(1);
2424 	rgvswctl |= MEMCTL_CMD_STS;
2425 	I915_WRITE(MEMSWCTL, rgvswctl);
2426 	mdelay(1);
2427 
2428 	spin_unlock_irq(&mchdev_lock);
2429 }
2430 
2431 /* There's a funny hw issue where the hw returns all 0 when reading from
2432  * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
2433  * ourselves, instead of doing a rmw cycle (which might result in us clearing
2434  * all limits and the gpu stuck at whatever frequency it is at atm).
2435  */
2436 static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 *val)
2437 {
2438 	u32 limits;
2439 
2440 	limits = 0;
2441 
2442 	if (*val >= dev_priv->rps.max_delay)
2443 		*val = dev_priv->rps.max_delay;
2444 	limits |= dev_priv->rps.max_delay << 24;
2445 
2446 	/* Only set the down limit when we've reached the lowest level to avoid
2447 	 * getting more interrupts, otherwise leave this clear. This prevents a
2448 	 * race in the hw when coming out of rc6: There's a tiny window where
2449 	 * the hw runs at the minimal clock before selecting the desired
2450 	 * frequency, if the down threshold expires in that window we will not
2451 	 * receive a down interrupt. */
2452 	if (*val <= dev_priv->rps.min_delay) {
2453 		*val = dev_priv->rps.min_delay;
2454 		limits |= dev_priv->rps.min_delay << 16;
2455 	}
2456 
2457 	return limits;
2458 }
2459 
2460 void gen6_set_rps(struct drm_device *dev, u8 val)
2461 {
2462 	struct drm_i915_private *dev_priv = dev->dev_private;
2463 	u32 limits = gen6_rps_limits(dev_priv, &val);
2464 
2465 	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2466 	WARN_ON(val > dev_priv->rps.max_delay);
2467 	WARN_ON(val < dev_priv->rps.min_delay);
2468 
2469 	if (val == dev_priv->rps.cur_delay)
2470 		return;
2471 
2472 	I915_WRITE(GEN6_RPNSWREQ,
2473 		   GEN6_FREQUENCY(val) |
2474 		   GEN6_OFFSET(0) |
2475 		   GEN6_AGGRESSIVE_TURBO);
2476 
2477 	/* Make sure we continue to get interrupts
2478 	 * until we hit the minimum or maximum frequencies.
2479 	 */
2480 	I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
2481 
2482 	POSTING_READ(GEN6_RPNSWREQ);
2483 
2484 	dev_priv->rps.cur_delay = val;
2485 
2486 	trace_intel_gpu_freq_change(val * 50);
2487 }
2488 
2489 static void gen6_disable_rps(struct drm_device *dev)
2490 {
2491 	struct drm_i915_private *dev_priv = dev->dev_private;
2492 
2493 	I915_WRITE(GEN6_RC_CONTROL, 0);
2494 	I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
2495 	I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
2496 	I915_WRITE(GEN6_PMIER, 0);
2497 	/* Complete PM interrupt masking here doesn't race with the rps work
2498 	 * item again unmasking PM interrupts because that is using a different
2499 	 * register (PMIMR) to mask PM interrupts. The only risk is in leaving
2500 	 * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
2501 
2502 	spin_lock_irq(&dev_priv->rps.lock);
2503 	dev_priv->rps.pm_iir = 0;
2504 	spin_unlock_irq(&dev_priv->rps.lock);
2505 
2506 	I915_WRITE(GEN6_PMIIR, I915_READ(GEN6_PMIIR));
2507 }
2508 
2509 int intel_enable_rc6(const struct drm_device *dev)
2510 {
2511 	/* Respect the kernel parameter if it is set */
2512 	if (i915_enable_rc6 >= 0)
2513 		return i915_enable_rc6;
2514 
2515 	/* Disable RC6 on Ironlake */
2516 	if (INTEL_INFO(dev)->gen == 5)
2517 		return 0;
2518 
2519 	if (IS_HASWELL(dev)) {
2520 		DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
2521 		return INTEL_RC6_ENABLE;
2522 	}
2523 
2524 	/* snb/ivb have more than one rc6 state. */
2525 	if (INTEL_INFO(dev)->gen == 6) {
2526 		DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
2527 		return INTEL_RC6_ENABLE;
2528 	}
2529 
2530 	DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n");
2531 	return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
2532 }
2533 
2534 static void gen6_enable_rps(struct drm_device *dev)
2535 {
2536 	struct drm_i915_private *dev_priv = dev->dev_private;
2537 	struct intel_ring_buffer *ring;
2538 	u32 rp_state_cap;
2539 	u32 gt_perf_status;
2540 	u32 rc6vids, pcu_mbox, rc6_mask = 0;
2541 	u32 gtfifodbg;
2542 	int rc6_mode;
2543 	int i, ret;
2544 
2545 	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2546 
2547 	/* Here begins a magic sequence of register writes to enable
2548 	 * auto-downclocking.
2549 	 *
2550 	 * Perhaps there might be some value in exposing these to
2551 	 * userspace...
2552 	 */
2553 	I915_WRITE(GEN6_RC_STATE, 0);
2554 
2555 	/* Clear the DBG now so we don't confuse earlier errors */
2556 	if ((gtfifodbg = I915_READ(GTFIFODBG))) {
2557 		DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
2558 		I915_WRITE(GTFIFODBG, gtfifodbg);
2559 	}
2560 
2561 	gen6_gt_force_wake_get(dev_priv);
2562 
2563 	rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
2564 	gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
2565 
2566 	/* In units of 100MHz */
2567 	dev_priv->rps.max_delay = rp_state_cap & 0xff;
2568 	dev_priv->rps.min_delay = (rp_state_cap & 0xff0000) >> 16;
2569 	dev_priv->rps.cur_delay = 0;
2570 
2571 	/* disable the counters and set deterministic thresholds */
2572 	I915_WRITE(GEN6_RC_CONTROL, 0);
2573 
2574 	I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
2575 	I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
2576 	I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
2577 	I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
2578 	I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
2579 
2580 	for_each_ring(ring, dev_priv, i)
2581 		I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
2582 
2583 	I915_WRITE(GEN6_RC_SLEEP, 0);
2584 	I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
2585 	I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
2586 	I915_WRITE(GEN6_RC6p_THRESHOLD, 100000);
2587 	I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
2588 
2589 	/* Check if we are enabling RC6 */
2590 	rc6_mode = intel_enable_rc6(dev_priv->dev);
2591 	if (rc6_mode & INTEL_RC6_ENABLE)
2592 		rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
2593 
2594 	/* We don't use those on Haswell */
2595 	if (!IS_HASWELL(dev)) {
2596 		if (rc6_mode & INTEL_RC6p_ENABLE)
2597 			rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
2598 
2599 		if (rc6_mode & INTEL_RC6pp_ENABLE)
2600 			rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
2601 	}
2602 
2603 	DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
2604 			(rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
2605 			(rc6_mask & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
2606 			(rc6_mask & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
2607 
2608 	I915_WRITE(GEN6_RC_CONTROL,
2609 		   rc6_mask |
2610 		   GEN6_RC_CTL_EI_MODE(1) |
2611 		   GEN6_RC_CTL_HW_ENABLE);
2612 
2613 	I915_WRITE(GEN6_RPNSWREQ,
2614 		   GEN6_FREQUENCY(10) |
2615 		   GEN6_OFFSET(0) |
2616 		   GEN6_AGGRESSIVE_TURBO);
2617 	I915_WRITE(GEN6_RC_VIDEO_FREQ,
2618 		   GEN6_FREQUENCY(12));
2619 
2620 	I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
2621 	I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
2622 		   dev_priv->rps.max_delay << 24 |
2623 		   dev_priv->rps.min_delay << 16);
2624 
2625 	I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
2626 	I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
2627 	I915_WRITE(GEN6_RP_UP_EI, 66000);
2628 	I915_WRITE(GEN6_RP_DOWN_EI, 350000);
2629 
2630 	I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
2631 	I915_WRITE(GEN6_RP_CONTROL,
2632 		   GEN6_RP_MEDIA_TURBO |
2633 		   GEN6_RP_MEDIA_HW_NORMAL_MODE |
2634 		   GEN6_RP_MEDIA_IS_GFX |
2635 		   GEN6_RP_ENABLE |
2636 		   GEN6_RP_UP_BUSY_AVG |
2637 		   (IS_HASWELL(dev) ? GEN7_RP_DOWN_IDLE_AVG : GEN6_RP_DOWN_IDLE_CONT));
2638 
2639 	ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
2640 	if (!ret) {
2641 		pcu_mbox = 0;
2642 		ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
2643 		if (ret && pcu_mbox & (1<<31)) { /* OC supported */
2644 			dev_priv->rps.max_delay = pcu_mbox & 0xff;
2645 			DRM_DEBUG_DRIVER("overclocking supported, adjusting frequency max to %dMHz\n", pcu_mbox * 50);
2646 		}
2647 	} else {
2648 		DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
2649 	}
2650 
2651 	gen6_set_rps(dev_priv->dev, (gt_perf_status & 0xff00) >> 8);
2652 
2653 	/* requires MSI enabled */
2654 	I915_WRITE(GEN6_PMIER, GEN6_PM_DEFERRED_EVENTS);
2655 	spin_lock_irq(&dev_priv->rps.lock);
2656 	WARN_ON(dev_priv->rps.pm_iir != 0);
2657 	I915_WRITE(GEN6_PMIMR, 0);
2658 	spin_unlock_irq(&dev_priv->rps.lock);
2659 	/* enable all PM interrupts */
2660 	I915_WRITE(GEN6_PMINTRMSK, 0);
2661 
2662 	rc6vids = 0;
2663 	ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
2664 	if (IS_GEN6(dev) && ret) {
2665 		DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
2666 	} else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
2667 		DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
2668 			  GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
2669 		rc6vids &= 0xffff00;
2670 		rc6vids |= GEN6_ENCODE_RC6_VID(450);
2671 		ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
2672 		if (ret)
2673 			DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
2674 	}
2675 
2676 	gen6_gt_force_wake_put(dev_priv);
2677 }
2678 
2679 static void gen6_update_ring_freq(struct drm_device *dev)
2680 {
2681 	struct drm_i915_private *dev_priv = dev->dev_private;
2682 	int min_freq = 15;
2683 	int gpu_freq;
2684 	unsigned int ia_freq, max_ia_freq;
2685 	int scaling_factor = 180;
2686 
2687 	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
2688 
2689 #ifdef __NetBSD__		/* XXX cpufreq */
2690 	{
2691 		extern uint64_t tsc_freq; /* x86 TSC frequency in Hz */
2692 		max_ia_freq = (tsc_freq / 1000);
2693 	}
2694 #else
2695 	max_ia_freq = cpufreq_quick_get_max(0);
2696 	/*
2697 	 * Default to measured freq if none found, PCU will ensure we don't go
2698 	 * over
2699 	 */
2700 	if (!max_ia_freq)
2701 		max_ia_freq = tsc_khz;
2702 #endif
2703 
2704 	/* Convert from kHz to MHz */
2705 	max_ia_freq /= 1000;
2706 
2707 	/*
2708 	 * For each potential GPU frequency, load a ring frequency we'd like
2709 	 * to use for memory access.  We do this by specifying the IA frequency
2710 	 * the PCU should use as a reference to determine the ring frequency.
2711 	 */
2712 	for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay;
2713 	     gpu_freq--) {
2714 		int diff = dev_priv->rps.max_delay - gpu_freq;
2715 
2716 		/*
2717 		 * For GPU frequencies less than 750MHz, just use the lowest
2718 		 * ring freq.
2719 		 */
2720 		if (gpu_freq < min_freq)
2721 			ia_freq = 800;
2722 		else
2723 			ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
2724 		ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
2725 		ia_freq <<= GEN6_PCODE_FREQ_IA_RATIO_SHIFT;
2726 
2727 		sandybridge_pcode_write(dev_priv,
2728 					GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
2729 					ia_freq | gpu_freq);
2730 	}
2731 }
2732 
2733 void ironlake_teardown_rc6(struct drm_device *dev)
2734 {
2735 	struct drm_i915_private *dev_priv = dev->dev_private;
2736 
2737 	if (dev_priv->ips.renderctx) {
2738 		i915_gem_object_unpin(dev_priv->ips.renderctx);
2739 		drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
2740 		dev_priv->ips.renderctx = NULL;
2741 	}
2742 
2743 	if (dev_priv->ips.pwrctx) {
2744 		i915_gem_object_unpin(dev_priv->ips.pwrctx);
2745 		drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
2746 		dev_priv->ips.pwrctx = NULL;
2747 	}
2748 }
2749 
2750 static void ironlake_disable_rc6(struct drm_device *dev)
2751 {
2752 	struct drm_i915_private *dev_priv = dev->dev_private;
2753 
2754 	if (I915_READ(PWRCTXA)) {
2755 		/* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
2756 		I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
2757 		wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
2758 			 50);
2759 
2760 		I915_WRITE(PWRCTXA, 0);
2761 		POSTING_READ(PWRCTXA);
2762 
2763 		I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
2764 		POSTING_READ(RSTDBYCTL);
2765 	}
2766 }
2767 
2768 static int ironlake_setup_rc6(struct drm_device *dev)
2769 {
2770 	struct drm_i915_private *dev_priv = dev->dev_private;
2771 
2772 	if (dev_priv->ips.renderctx == NULL)
2773 		dev_priv->ips.renderctx = intel_alloc_context_page(dev);
2774 	if (!dev_priv->ips.renderctx)
2775 		return -ENOMEM;
2776 
2777 	if (dev_priv->ips.pwrctx == NULL)
2778 		dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
2779 	if (!dev_priv->ips.pwrctx) {
2780 		ironlake_teardown_rc6(dev);
2781 		return -ENOMEM;
2782 	}
2783 
2784 	return 0;
2785 }
2786 
2787 static void ironlake_enable_rc6(struct drm_device *dev)
2788 {
2789 	struct drm_i915_private *dev_priv = dev->dev_private;
2790 	struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
2791 	bool was_interruptible;
2792 	int ret;
2793 
2794 	/* rc6 disabled by default due to repeated reports of hanging during
2795 	 * boot and resume.
2796 	 */
2797 	if (!intel_enable_rc6(dev))
2798 		return;
2799 
2800 	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
2801 
2802 	ret = ironlake_setup_rc6(dev);
2803 	if (ret)
2804 		return;
2805 
2806 	was_interruptible = dev_priv->mm.interruptible;
2807 	dev_priv->mm.interruptible = false;
2808 
2809 	/*
2810 	 * GPU can automatically power down the render unit if given a page
2811 	 * to save state.
2812 	 */
2813 	ret = intel_ring_begin(ring, 6);
2814 	if (ret) {
2815 		ironlake_teardown_rc6(dev);
2816 		dev_priv->mm.interruptible = was_interruptible;
2817 		return;
2818 	}
2819 
2820 	intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
2821 	intel_ring_emit(ring, MI_SET_CONTEXT);
2822 	intel_ring_emit(ring, dev_priv->ips.renderctx->gtt_offset |
2823 			MI_MM_SPACE_GTT |
2824 			MI_SAVE_EXT_STATE_EN |
2825 			MI_RESTORE_EXT_STATE_EN |
2826 			MI_RESTORE_INHIBIT);
2827 	intel_ring_emit(ring, MI_SUSPEND_FLUSH);
2828 	intel_ring_emit(ring, MI_NOOP);
2829 	intel_ring_emit(ring, MI_FLUSH);
2830 	intel_ring_advance(ring);
2831 
2832 	/*
2833 	 * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
2834 	 * does an implicit flush, combined with MI_FLUSH above, it should be
2835 	 * safe to assume that renderctx is valid
2836 	 */
2837 	ret = intel_ring_idle(ring);
2838 	dev_priv->mm.interruptible = was_interruptible;
2839 	if (ret) {
2840 		DRM_ERROR("failed to enable ironlake power power savings\n");
2841 		ironlake_teardown_rc6(dev);
2842 		return;
2843 	}
2844 
2845 	I915_WRITE(PWRCTXA, dev_priv->ips.pwrctx->gtt_offset | PWRCTX_EN);
2846 	I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
2847 }
2848 
2849 static unsigned long intel_pxfreq(u32 vidfreq)
2850 {
2851 	unsigned long freq;
2852 	int div = (vidfreq & 0x3f0000) >> 16;
2853 	int post = (vidfreq & 0x3000) >> 12;
2854 	int pre = (vidfreq & 0x7);
2855 
2856 	if (!pre)
2857 		return 0;
2858 
2859 	freq = ((div * 133333) / ((1<<post) * pre));
2860 
2861 	return freq;
2862 }
2863 
2864 static const struct cparams {
2865 	u16 i;
2866 	u16 t;
2867 	u16 m;
2868 	u16 c;
2869 } cparams[] = {
2870 	{ 1, 1333, 301, 28664 },
2871 	{ 1, 1066, 294, 24460 },
2872 	{ 1, 800, 294, 25192 },
2873 	{ 0, 1333, 276, 27605 },
2874 	{ 0, 1066, 276, 27605 },
2875 	{ 0, 800, 231, 23784 },
2876 };
2877 
2878 static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
2879 {
2880 	u64 total_count, diff, ret;
2881 	u32 count1, count2, count3, m = 0, c = 0;
2882 	unsigned long now = jiffies_to_msecs(jiffies), diff1;
2883 	int i;
2884 
2885 	assert_spin_locked(&mchdev_lock);
2886 
2887 	diff1 = now - dev_priv->ips.last_time1;
2888 
2889 	/* Prevent division-by-zero if we are asking too fast.
2890 	 * Also, we don't get interesting results if we are polling
2891 	 * faster than once in 10ms, so just return the saved value
2892 	 * in such cases.
2893 	 */
2894 	if (diff1 <= 10)
2895 		return dev_priv->ips.chipset_power;
2896 
2897 	count1 = I915_READ(DMIEC);
2898 	count2 = I915_READ(DDREC);
2899 	count3 = I915_READ(CSIEC);
2900 
2901 	total_count = count1 + count2 + count3;
2902 
2903 	/* FIXME: handle per-counter overflow */
2904 	if (total_count < dev_priv->ips.last_count1) {
2905 		diff = ~0UL - dev_priv->ips.last_count1;
2906 		diff += total_count;
2907 	} else {
2908 		diff = total_count - dev_priv->ips.last_count1;
2909 	}
2910 
2911 	for (i = 0; i < ARRAY_SIZE(cparams); i++) {
2912 		if (cparams[i].i == dev_priv->ips.c_m &&
2913 		    cparams[i].t == dev_priv->ips.r_t) {
2914 			m = cparams[i].m;
2915 			c = cparams[i].c;
2916 			break;
2917 		}
2918 	}
2919 
2920 	diff = div_u64(diff, diff1);
2921 	ret = ((m * diff) + c);
2922 	ret = div_u64(ret, 10);
2923 
2924 	dev_priv->ips.last_count1 = total_count;
2925 	dev_priv->ips.last_time1 = now;
2926 
2927 	dev_priv->ips.chipset_power = ret;
2928 
2929 	return ret;
2930 }
2931 
2932 unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
2933 {
2934 	unsigned long val;
2935 
2936 	if (dev_priv->info->gen != 5)
2937 		return 0;
2938 
2939 	spin_lock_irq(&mchdev_lock);
2940 
2941 	val = __i915_chipset_val(dev_priv);
2942 
2943 	spin_unlock_irq(&mchdev_lock);
2944 
2945 	return val;
2946 }
2947 
2948 unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
2949 {
2950 	unsigned long m, x, b;
2951 	u32 tsfs;
2952 
2953 	tsfs = I915_READ(TSFS);
2954 
2955 	m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
2956 	x = I915_READ8(TR1);
2957 
2958 	b = tsfs & TSFS_INTR_MASK;
2959 
2960 	return ((m * x) / 127) - b;
2961 }
2962 
2963 static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
2964 {
2965 	static const struct v_table {
2966 		u16 vd; /* in .1 mil */
2967 		u16 vm; /* in .1 mil */
2968 	} v_table[] = {
2969 		{ 0, 0, },
2970 		{ 375, 0, },
2971 		{ 500, 0, },
2972 		{ 625, 0, },
2973 		{ 750, 0, },
2974 		{ 875, 0, },
2975 		{ 1000, 0, },
2976 		{ 1125, 0, },
2977 		{ 4125, 3000, },
2978 		{ 4125, 3000, },
2979 		{ 4125, 3000, },
2980 		{ 4125, 3000, },
2981 		{ 4125, 3000, },
2982 		{ 4125, 3000, },
2983 		{ 4125, 3000, },
2984 		{ 4125, 3000, },
2985 		{ 4125, 3000, },
2986 		{ 4125, 3000, },
2987 		{ 4125, 3000, },
2988 		{ 4125, 3000, },
2989 		{ 4125, 3000, },
2990 		{ 4125, 3000, },
2991 		{ 4125, 3000, },
2992 		{ 4125, 3000, },
2993 		{ 4125, 3000, },
2994 		{ 4125, 3000, },
2995 		{ 4125, 3000, },
2996 		{ 4125, 3000, },
2997 		{ 4125, 3000, },
2998 		{ 4125, 3000, },
2999 		{ 4125, 3000, },
3000 		{ 4125, 3000, },
3001 		{ 4250, 3125, },
3002 		{ 4375, 3250, },
3003 		{ 4500, 3375, },
3004 		{ 4625, 3500, },
3005 		{ 4750, 3625, },
3006 		{ 4875, 3750, },
3007 		{ 5000, 3875, },
3008 		{ 5125, 4000, },
3009 		{ 5250, 4125, },
3010 		{ 5375, 4250, },
3011 		{ 5500, 4375, },
3012 		{ 5625, 4500, },
3013 		{ 5750, 4625, },
3014 		{ 5875, 4750, },
3015 		{ 6000, 4875, },
3016 		{ 6125, 5000, },
3017 		{ 6250, 5125, },
3018 		{ 6375, 5250, },
3019 		{ 6500, 5375, },
3020 		{ 6625, 5500, },
3021 		{ 6750, 5625, },
3022 		{ 6875, 5750, },
3023 		{ 7000, 5875, },
3024 		{ 7125, 6000, },
3025 		{ 7250, 6125, },
3026 		{ 7375, 6250, },
3027 		{ 7500, 6375, },
3028 		{ 7625, 6500, },
3029 		{ 7750, 6625, },
3030 		{ 7875, 6750, },
3031 		{ 8000, 6875, },
3032 		{ 8125, 7000, },
3033 		{ 8250, 7125, },
3034 		{ 8375, 7250, },
3035 		{ 8500, 7375, },
3036 		{ 8625, 7500, },
3037 		{ 8750, 7625, },
3038 		{ 8875, 7750, },
3039 		{ 9000, 7875, },
3040 		{ 9125, 8000, },
3041 		{ 9250, 8125, },
3042 		{ 9375, 8250, },
3043 		{ 9500, 8375, },
3044 		{ 9625, 8500, },
3045 		{ 9750, 8625, },
3046 		{ 9875, 8750, },
3047 		{ 10000, 8875, },
3048 		{ 10125, 9000, },
3049 		{ 10250, 9125, },
3050 		{ 10375, 9250, },
3051 		{ 10500, 9375, },
3052 		{ 10625, 9500, },
3053 		{ 10750, 9625, },
3054 		{ 10875, 9750, },
3055 		{ 11000, 9875, },
3056 		{ 11125, 10000, },
3057 		{ 11250, 10125, },
3058 		{ 11375, 10250, },
3059 		{ 11500, 10375, },
3060 		{ 11625, 10500, },
3061 		{ 11750, 10625, },
3062 		{ 11875, 10750, },
3063 		{ 12000, 10875, },
3064 		{ 12125, 11000, },
3065 		{ 12250, 11125, },
3066 		{ 12375, 11250, },
3067 		{ 12500, 11375, },
3068 		{ 12625, 11500, },
3069 		{ 12750, 11625, },
3070 		{ 12875, 11750, },
3071 		{ 13000, 11875, },
3072 		{ 13125, 12000, },
3073 		{ 13250, 12125, },
3074 		{ 13375, 12250, },
3075 		{ 13500, 12375, },
3076 		{ 13625, 12500, },
3077 		{ 13750, 12625, },
3078 		{ 13875, 12750, },
3079 		{ 14000, 12875, },
3080 		{ 14125, 13000, },
3081 		{ 14250, 13125, },
3082 		{ 14375, 13250, },
3083 		{ 14500, 13375, },
3084 		{ 14625, 13500, },
3085 		{ 14750, 13625, },
3086 		{ 14875, 13750, },
3087 		{ 15000, 13875, },
3088 		{ 15125, 14000, },
3089 		{ 15250, 14125, },
3090 		{ 15375, 14250, },
3091 		{ 15500, 14375, },
3092 		{ 15625, 14500, },
3093 		{ 15750, 14625, },
3094 		{ 15875, 14750, },
3095 		{ 16000, 14875, },
3096 		{ 16125, 15000, },
3097 	};
3098 	if (dev_priv->info->is_mobile)
3099 		return v_table[pxvid].vm;
3100 	else
3101 		return v_table[pxvid].vd;
3102 }
3103 
3104 static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
3105 {
3106 	struct timespec now, diff1;
3107 	u64 diff;
3108 	unsigned long diffms;
3109 	u32 count;
3110 
3111 	assert_spin_locked(&mchdev_lock);
3112 
3113 	getrawmonotonic(&now);
3114 	diff1 = timespec_sub(now, dev_priv->ips.last_time2);
3115 
3116 	/* Don't divide by 0 */
3117 	diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
3118 	if (!diffms)
3119 		return;
3120 
3121 	count = I915_READ(GFXEC);
3122 
3123 	if (count < dev_priv->ips.last_count2) {
3124 		diff = ~0UL - dev_priv->ips.last_count2;
3125 		diff += count;
3126 	} else {
3127 		diff = count - dev_priv->ips.last_count2;
3128 	}
3129 
3130 	dev_priv->ips.last_count2 = count;
3131 	dev_priv->ips.last_time2 = now;
3132 
3133 	/* More magic constants... */
3134 	diff = diff * 1181;
3135 	diff = div_u64(diff, diffms * 10);
3136 	dev_priv->ips.gfx_power = diff;
3137 }
3138 
3139 void i915_update_gfx_val(struct drm_i915_private *dev_priv)
3140 {
3141 	if (dev_priv->info->gen != 5)
3142 		return;
3143 
3144 	spin_lock_irq(&mchdev_lock);
3145 
3146 	__i915_update_gfx_val(dev_priv);
3147 
3148 	spin_unlock_irq(&mchdev_lock);
3149 }
3150 
3151 static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
3152 {
3153 	unsigned long t, corr, state1, corr2, state2;
3154 	u32 pxvid, ext_v;
3155 
3156 	assert_spin_locked(&mchdev_lock);
3157 
3158 	pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4));
3159 	pxvid = (pxvid >> 24) & 0x7f;
3160 	ext_v = pvid_to_extvid(dev_priv, pxvid);
3161 
3162 	state1 = ext_v;
3163 
3164 	t = i915_mch_val(dev_priv);
3165 
3166 	/* Revel in the empirically derived constants */
3167 
3168 	/* Correction factor in 1/100000 units */
3169 	if (t > 80)
3170 		corr = ((t * 2349) + 135940);
3171 	else if (t >= 50)
3172 		corr = ((t * 964) + 29317);
3173 	else /* < 50 */
3174 		corr = ((t * 301) + 1004);
3175 
3176 	corr = corr * ((150142 * state1) / 10000 - 78642);
3177 	corr /= 100000;
3178 	corr2 = (corr * dev_priv->ips.corr);
3179 
3180 	state2 = (corr2 * state1) / 10000;
3181 	state2 /= 100; /* convert to mW */
3182 
3183 	__i915_update_gfx_val(dev_priv);
3184 
3185 	return dev_priv->ips.gfx_power + state2;
3186 }
3187 
3188 unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
3189 {
3190 	unsigned long val;
3191 
3192 	if (dev_priv->info->gen != 5)
3193 		return 0;
3194 
3195 	spin_lock_irq(&mchdev_lock);
3196 
3197 	val = __i915_gfx_val(dev_priv);
3198 
3199 	spin_unlock_irq(&mchdev_lock);
3200 
3201 	return val;
3202 }
3203 
3204 #ifndef __NetBSD__		/* XXX IPS */
3205 /**
3206  * i915_read_mch_val - return value for IPS use
3207  *
3208  * Calculate and return a value for the IPS driver to use when deciding whether
3209  * we have thermal and power headroom to increase CPU or GPU power budget.
3210  */
3211 unsigned long i915_read_mch_val(void)
3212 {
3213 	struct drm_i915_private *dev_priv;
3214 	unsigned long chipset_val, graphics_val, ret = 0;
3215 
3216 	spin_lock_irq(&mchdev_lock);
3217 	if (!i915_mch_dev)
3218 		goto out_unlock;
3219 	dev_priv = i915_mch_dev;
3220 
3221 	chipset_val = __i915_chipset_val(dev_priv);
3222 	graphics_val = __i915_gfx_val(dev_priv);
3223 
3224 	ret = chipset_val + graphics_val;
3225 
3226 out_unlock:
3227 	spin_unlock_irq(&mchdev_lock);
3228 
3229 	return ret;
3230 }
3231 EXPORT_SYMBOL_GPL(i915_read_mch_val);
3232 
3233 /**
3234  * i915_gpu_raise - raise GPU frequency limit
3235  *
3236  * Raise the limit; IPS indicates we have thermal headroom.
3237  */
3238 bool i915_gpu_raise(void)
3239 {
3240 	struct drm_i915_private *dev_priv;
3241 	bool ret = true;
3242 
3243 	spin_lock_irq(&mchdev_lock);
3244 	if (!i915_mch_dev) {
3245 		ret = false;
3246 		goto out_unlock;
3247 	}
3248 	dev_priv = i915_mch_dev;
3249 
3250 	if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
3251 		dev_priv->ips.max_delay--;
3252 
3253 out_unlock:
3254 	spin_unlock_irq(&mchdev_lock);
3255 
3256 	return ret;
3257 }
3258 EXPORT_SYMBOL_GPL(i915_gpu_raise);
3259 
3260 /**
3261  * i915_gpu_lower - lower GPU frequency limit
3262  *
3263  * IPS indicates we're close to a thermal limit, so throttle back the GPU
3264  * frequency maximum.
3265  */
3266 bool i915_gpu_lower(void)
3267 {
3268 	struct drm_i915_private *dev_priv;
3269 	bool ret = true;
3270 
3271 	spin_lock_irq(&mchdev_lock);
3272 	if (!i915_mch_dev) {
3273 		ret = false;
3274 		goto out_unlock;
3275 	}
3276 	dev_priv = i915_mch_dev;
3277 
3278 	if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
3279 		dev_priv->ips.max_delay++;
3280 
3281 out_unlock:
3282 	spin_unlock_irq(&mchdev_lock);
3283 
3284 	return ret;
3285 }
3286 EXPORT_SYMBOL_GPL(i915_gpu_lower);
3287 
3288 /**
3289  * i915_gpu_busy - indicate GPU business to IPS
3290  *
3291  * Tell the IPS driver whether or not the GPU is busy.
3292  */
3293 bool i915_gpu_busy(void)
3294 {
3295 	struct drm_i915_private *dev_priv;
3296 	struct intel_ring_buffer *ring;
3297 	bool ret = false;
3298 	int i;
3299 
3300 	spin_lock_irq(&mchdev_lock);
3301 	if (!i915_mch_dev)
3302 		goto out_unlock;
3303 	dev_priv = i915_mch_dev;
3304 
3305 	for_each_ring(ring, dev_priv, i)
3306 		ret |= !list_empty(&ring->request_list);
3307 
3308 out_unlock:
3309 	spin_unlock_irq(&mchdev_lock);
3310 
3311 	return ret;
3312 }
3313 EXPORT_SYMBOL_GPL(i915_gpu_busy);
3314 
3315 /**
3316  * i915_gpu_turbo_disable - disable graphics turbo
3317  *
3318  * Disable graphics turbo by resetting the max frequency and setting the
3319  * current frequency to the default.
3320  */
3321 bool i915_gpu_turbo_disable(void)
3322 {
3323 	struct drm_i915_private *dev_priv;
3324 	bool ret = true;
3325 
3326 	spin_lock_irq(&mchdev_lock);
3327 	if (!i915_mch_dev) {
3328 		ret = false;
3329 		goto out_unlock;
3330 	}
3331 	dev_priv = i915_mch_dev;
3332 
3333 	dev_priv->ips.max_delay = dev_priv->ips.fstart;
3334 
3335 	if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
3336 		ret = false;
3337 
3338 out_unlock:
3339 	spin_unlock_irq(&mchdev_lock);
3340 
3341 	return ret;
3342 }
3343 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
3344 #endif
3345 
3346 /**
3347  * Tells the intel_ips driver that the i915 driver is now loaded, if
3348  * IPS got loaded first.
3349  *
3350  * This awkward dance is so that neither module has to depend on the
3351  * other in order for IPS to do the appropriate communication of
3352  * GPU turbo limits to i915.
3353  */
3354 static void
3355 ips_ping_for_i915_load(void)
3356 {
3357 #ifndef __NetBSD__		/* XXX whattakludge for Linux module mania */
3358 	void (*link)(void);
3359 
3360 	link = symbol_get(ips_link_to_i915_driver);
3361 	if (link) {
3362 		link();
3363 		symbol_put(ips_link_to_i915_driver);
3364 	}
3365 #endif
3366 }
3367 
3368 void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
3369 {
3370 #ifdef __NetBSD__		/* XXX */
3371 	/*
3372 	 * This seems as good a place as any to initialize mchdev_lock.
3373 	 * Taking the lock in the rest of this routine is silly, but...
3374 	 */
3375 	spin_lock_init(&mchdev_lock);
3376 #endif
3377 
3378 	/* We only register the i915 ips part with intel-ips once everything is
3379 	 * set up, to avoid intel-ips sneaking in and reading bogus values. */
3380 	spin_lock_irq(&mchdev_lock);
3381 	i915_mch_dev = dev_priv;
3382 	spin_unlock_irq(&mchdev_lock);
3383 
3384 	ips_ping_for_i915_load();
3385 }
3386 
3387 void intel_gpu_ips_teardown(void)
3388 {
3389 #ifdef __NetBSD__
3390 	if (i915_mch_dev == NULL)
3391 		return;
3392 #endif
3393 	spin_lock_irq(&mchdev_lock);
3394 	i915_mch_dev = NULL;
3395 	spin_unlock_irq(&mchdev_lock);
3396 #ifdef __NetBSD__
3397 	spin_lock_destroy(&mchdev_lock);
3398 #endif
3399 }
3400 static void intel_init_emon(struct drm_device *dev)
3401 {
3402 	struct drm_i915_private *dev_priv = dev->dev_private;
3403 	u32 lcfuse;
3404 	u8 pxw[16];
3405 	int i;
3406 
3407 	/* Disable to program */
3408 	I915_WRITE(ECR, 0);
3409 	POSTING_READ(ECR);
3410 
3411 	/* Program energy weights for various events */
3412 	I915_WRITE(SDEW, 0x15040d00);
3413 	I915_WRITE(CSIEW0, 0x007f0000);
3414 	I915_WRITE(CSIEW1, 0x1e220004);
3415 	I915_WRITE(CSIEW2, 0x04000004);
3416 
3417 	for (i = 0; i < 5; i++)
3418 		I915_WRITE(PEW + (i * 4), 0);
3419 	for (i = 0; i < 3; i++)
3420 		I915_WRITE(DEW + (i * 4), 0);
3421 
3422 	/* Program P-state weights to account for frequency power adjustment */
3423 	for (i = 0; i < 16; i++) {
3424 		u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
3425 		unsigned long freq = intel_pxfreq(pxvidfreq);
3426 		unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
3427 			PXVFREQ_PX_SHIFT;
3428 		unsigned long val;
3429 
3430 		val = vid * vid;
3431 		val *= (freq / 1000);
3432 		val *= 255;
3433 		val /= (127*127*900);
3434 		if (val > 0xff)
3435 			DRM_ERROR("bad pxval: %ld\n", val);
3436 		pxw[i] = val;
3437 	}
3438 	/* Render standby states get 0 weight */
3439 	pxw[14] = 0;
3440 	pxw[15] = 0;
3441 
3442 	for (i = 0; i < 4; i++) {
3443 		u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
3444 			(pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
3445 		I915_WRITE(PXW + (i * 4), val);
3446 	}
3447 
3448 	/* Adjust magic regs to magic values (more experimental results) */
3449 	I915_WRITE(OGW0, 0);
3450 	I915_WRITE(OGW1, 0);
3451 	I915_WRITE(EG0, 0x00007f00);
3452 	I915_WRITE(EG1, 0x0000000e);
3453 	I915_WRITE(EG2, 0x000e0000);
3454 	I915_WRITE(EG3, 0x68000300);
3455 	I915_WRITE(EG4, 0x42000000);
3456 	I915_WRITE(EG5, 0x00140031);
3457 	I915_WRITE(EG6, 0);
3458 	I915_WRITE(EG7, 0);
3459 
3460 	for (i = 0; i < 8; i++)
3461 		I915_WRITE(PXWL + (i * 4), 0);
3462 
3463 	/* Enable PMON + select events */
3464 	I915_WRITE(ECR, 0x80000019);
3465 
3466 	lcfuse = I915_READ(LCFUSE02);
3467 
3468 	dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
3469 }
3470 
3471 void intel_disable_gt_powersave(struct drm_device *dev)
3472 {
3473 	struct drm_i915_private *dev_priv = dev->dev_private;
3474 
3475 	if (IS_IRONLAKE_M(dev)) {
3476 		ironlake_disable_drps(dev);
3477 		ironlake_disable_rc6(dev);
3478 	} else if (INTEL_INFO(dev)->gen >= 6 && !IS_VALLEYVIEW(dev)) {
3479 		cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
3480 		mutex_lock(&dev_priv->rps.hw_lock);
3481 		gen6_disable_rps(dev);
3482 		mutex_unlock(&dev_priv->rps.hw_lock);
3483 	}
3484 }
3485 
3486 static void intel_gen6_powersave_work(struct work_struct *work)
3487 {
3488 	struct drm_i915_private *dev_priv =
3489 		container_of(work, struct drm_i915_private,
3490 			     rps.delayed_resume_work.work);
3491 	struct drm_device *dev = dev_priv->dev;
3492 
3493 	mutex_lock(&dev_priv->rps.hw_lock);
3494 	gen6_enable_rps(dev);
3495 	gen6_update_ring_freq(dev);
3496 	mutex_unlock(&dev_priv->rps.hw_lock);
3497 }
3498 
3499 void intel_enable_gt_powersave(struct drm_device *dev)
3500 {
3501 	struct drm_i915_private *dev_priv = dev->dev_private;
3502 
3503 	if (IS_IRONLAKE_M(dev)) {
3504 		ironlake_enable_drps(dev);
3505 		ironlake_enable_rc6(dev);
3506 		intel_init_emon(dev);
3507 	} else if ((IS_GEN6(dev) || IS_GEN7(dev)) && !IS_VALLEYVIEW(dev)) {
3508 		/*
3509 		 * PCU communication is slow and this doesn't need to be
3510 		 * done at any specific time, so do this out of our fast path
3511 		 * to make resume and init faster.
3512 		 */
3513 		schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
3514 				      round_jiffies_up_relative(HZ));
3515 	}
3516 }
3517 
3518 static void ibx_init_clock_gating(struct drm_device *dev)
3519 {
3520 	struct drm_i915_private *dev_priv = dev->dev_private;
3521 
3522 	/*
3523 	 * On Ibex Peak and Cougar Point, we need to disable clock
3524 	 * gating for the panel power sequencer or it will fail to
3525 	 * start up when no ports are active.
3526 	 */
3527 	I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
3528 }
3529 
3530 static void ironlake_init_clock_gating(struct drm_device *dev)
3531 {
3532 	struct drm_i915_private *dev_priv = dev->dev_private;
3533 	uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
3534 
3535 	/* Required for FBC */
3536 	dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
3537 		   ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
3538 		   ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
3539 
3540 	I915_WRITE(PCH_3DCGDIS0,
3541 		   MARIUNIT_CLOCK_GATE_DISABLE |
3542 		   SVSMUNIT_CLOCK_GATE_DISABLE);
3543 	I915_WRITE(PCH_3DCGDIS1,
3544 		   VFMUNIT_CLOCK_GATE_DISABLE);
3545 
3546 	/*
3547 	 * According to the spec the following bits should be set in
3548 	 * order to enable memory self-refresh
3549 	 * The bit 22/21 of 0x42004
3550 	 * The bit 5 of 0x42020
3551 	 * The bit 15 of 0x45000
3552 	 */
3553 	I915_WRITE(ILK_DISPLAY_CHICKEN2,
3554 		   (I915_READ(ILK_DISPLAY_CHICKEN2) |
3555 		    ILK_DPARB_GATE | ILK_VSDPFD_FULL));
3556 	dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
3557 	I915_WRITE(DISP_ARB_CTL,
3558 		   (I915_READ(DISP_ARB_CTL) |
3559 		    DISP_FBC_WM_DIS));
3560 	I915_WRITE(WM3_LP_ILK, 0);
3561 	I915_WRITE(WM2_LP_ILK, 0);
3562 	I915_WRITE(WM1_LP_ILK, 0);
3563 
3564 	/*
3565 	 * Based on the document from hardware guys the following bits
3566 	 * should be set unconditionally in order to enable FBC.
3567 	 * The bit 22 of 0x42000
3568 	 * The bit 22 of 0x42004
3569 	 * The bit 7,8,9 of 0x42020.
3570 	 */
3571 	if (IS_IRONLAKE_M(dev)) {
3572 		I915_WRITE(ILK_DISPLAY_CHICKEN1,
3573 			   I915_READ(ILK_DISPLAY_CHICKEN1) |
3574 			   ILK_FBCQ_DIS);
3575 		I915_WRITE(ILK_DISPLAY_CHICKEN2,
3576 			   I915_READ(ILK_DISPLAY_CHICKEN2) |
3577 			   ILK_DPARB_GATE);
3578 	}
3579 
3580 	I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
3581 
3582 	I915_WRITE(ILK_DISPLAY_CHICKEN2,
3583 		   I915_READ(ILK_DISPLAY_CHICKEN2) |
3584 		   ILK_ELPIN_409_SELECT);
3585 	I915_WRITE(_3D_CHICKEN2,
3586 		   _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
3587 		   _3D_CHICKEN2_WM_READ_PIPELINED);
3588 
3589 	/* WaDisableRenderCachePipelinedFlush */
3590 	I915_WRITE(CACHE_MODE_0,
3591 		   _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
3592 
3593 	ibx_init_clock_gating(dev);
3594 }
3595 
3596 static void cpt_init_clock_gating(struct drm_device *dev)
3597 {
3598 	struct drm_i915_private *dev_priv = dev->dev_private;
3599 	int pipe;
3600 
3601 	/*
3602 	 * On Ibex Peak and Cougar Point, we need to disable clock
3603 	 * gating for the panel power sequencer or it will fail to
3604 	 * start up when no ports are active.
3605 	 */
3606 	I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
3607 	I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
3608 		   DPLS_EDP_PPS_FIX_DIS);
3609 	/* The below fixes the weird display corruption, a few pixels shifted
3610 	 * downward, on (only) LVDS of some HP laptops with IVY.
3611 	 */
3612 	for_each_pipe(pipe)
3613 		I915_WRITE(TRANS_CHICKEN2(pipe), TRANS_CHICKEN2_TIMING_OVERRIDE);
3614 	/* WADP0ClockGatingDisable */
3615 	for_each_pipe(pipe) {
3616 		I915_WRITE(TRANS_CHICKEN1(pipe),
3617 			   TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
3618 	}
3619 }
3620 
3621 static void gen6_init_clock_gating(struct drm_device *dev)
3622 {
3623 	struct drm_i915_private *dev_priv = dev->dev_private;
3624 	int pipe;
3625 	uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
3626 
3627 	I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
3628 
3629 	I915_WRITE(ILK_DISPLAY_CHICKEN2,
3630 		   I915_READ(ILK_DISPLAY_CHICKEN2) |
3631 		   ILK_ELPIN_409_SELECT);
3632 
3633 	/* WaDisableHiZPlanesWhenMSAAEnabled */
3634 	I915_WRITE(_3D_CHICKEN,
3635 		   _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
3636 
3637 	/* WaSetupGtModeTdRowDispatch */
3638 	if (IS_SNB_GT1(dev))
3639 		I915_WRITE(GEN6_GT_MODE,
3640 			   _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
3641 
3642 	I915_WRITE(WM3_LP_ILK, 0);
3643 	I915_WRITE(WM2_LP_ILK, 0);
3644 	I915_WRITE(WM1_LP_ILK, 0);
3645 
3646 	I915_WRITE(CACHE_MODE_0,
3647 		   _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
3648 
3649 	I915_WRITE(GEN6_UCGCTL1,
3650 		   I915_READ(GEN6_UCGCTL1) |
3651 		   GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
3652 		   GEN6_CSUNIT_CLOCK_GATE_DISABLE);
3653 
3654 	/* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3655 	 * gating disable must be set.  Failure to set it results in
3656 	 * flickering pixels due to Z write ordering failures after
3657 	 * some amount of runtime in the Mesa "fire" demo, and Unigine
3658 	 * Sanctuary and Tropics, and apparently anything else with
3659 	 * alpha test or pixel discard.
3660 	 *
3661 	 * According to the spec, bit 11 (RCCUNIT) must also be set,
3662 	 * but we didn't debug actual testcases to find it out.
3663 	 *
3664 	 * Also apply WaDisableVDSUnitClockGating and
3665 	 * WaDisableRCPBUnitClockGating.
3666 	 */
3667 	I915_WRITE(GEN6_UCGCTL2,
3668 		   GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
3669 		   GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
3670 		   GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
3671 
3672 	/* Bspec says we need to always set all mask bits. */
3673 	I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) |
3674 		   _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL);
3675 
3676 	/*
3677 	 * According to the spec the following bits should be
3678 	 * set in order to enable memory self-refresh and fbc:
3679 	 * The bit21 and bit22 of 0x42000
3680 	 * The bit21 and bit22 of 0x42004
3681 	 * The bit5 and bit7 of 0x42020
3682 	 * The bit14 of 0x70180
3683 	 * The bit14 of 0x71180
3684 	 */
3685 	I915_WRITE(ILK_DISPLAY_CHICKEN1,
3686 		   I915_READ(ILK_DISPLAY_CHICKEN1) |
3687 		   ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
3688 	I915_WRITE(ILK_DISPLAY_CHICKEN2,
3689 		   I915_READ(ILK_DISPLAY_CHICKEN2) |
3690 		   ILK_DPARB_GATE | ILK_VSDPFD_FULL);
3691 	I915_WRITE(ILK_DSPCLK_GATE_D,
3692 		   I915_READ(ILK_DSPCLK_GATE_D) |
3693 		   ILK_DPARBUNIT_CLOCK_GATE_ENABLE  |
3694 		   ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
3695 
3696 	/* WaMbcDriverBootEnable */
3697 	I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
3698 		   GEN6_MBCTL_ENABLE_BOOT_FETCH);
3699 
3700 	for_each_pipe(pipe) {
3701 		I915_WRITE(DSPCNTR(pipe),
3702 			   I915_READ(DSPCNTR(pipe)) |
3703 			   DISPPLANE_TRICKLE_FEED_DISABLE);
3704 		intel_flush_display_plane(dev_priv, pipe);
3705 	}
3706 
3707 	/* The default value should be 0x200 according to docs, but the two
3708 	 * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */
3709 	I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff));
3710 	I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI));
3711 
3712 	cpt_init_clock_gating(dev);
3713 }
3714 
3715 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
3716 {
3717 	uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
3718 
3719 	reg &= ~GEN7_FF_SCHED_MASK;
3720 	reg |= GEN7_FF_TS_SCHED_HW;
3721 	reg |= GEN7_FF_VS_SCHED_HW;
3722 	reg |= GEN7_FF_DS_SCHED_HW;
3723 
3724 	I915_WRITE(GEN7_FF_THREAD_MODE, reg);
3725 }
3726 
3727 static void lpt_init_clock_gating(struct drm_device *dev)
3728 {
3729 	struct drm_i915_private *dev_priv = dev->dev_private;
3730 
3731 	/*
3732 	 * TODO: this bit should only be enabled when really needed, then
3733 	 * disabled when not needed anymore in order to save power.
3734 	 */
3735 	if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
3736 		I915_WRITE(SOUTH_DSPCLK_GATE_D,
3737 			   I915_READ(SOUTH_DSPCLK_GATE_D) |
3738 			   PCH_LP_PARTITION_LEVEL_DISABLE);
3739 }
3740 
3741 static void haswell_init_clock_gating(struct drm_device *dev)
3742 {
3743 	struct drm_i915_private *dev_priv = dev->dev_private;
3744 	int pipe;
3745 
3746 	I915_WRITE(WM3_LP_ILK, 0);
3747 	I915_WRITE(WM2_LP_ILK, 0);
3748 	I915_WRITE(WM1_LP_ILK, 0);
3749 
3750 	/* According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3751 	 * This implements the WaDisableRCZUnitClockGating workaround.
3752 	 */
3753 	I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
3754 
3755 	/* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3756 	I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
3757 		   GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
3758 
3759 	/* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3760 	I915_WRITE(GEN7_L3CNTLREG1,
3761 			GEN7_WA_FOR_GEN7_L3_CONTROL);
3762 	I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
3763 			GEN7_WA_L3_CHICKEN_MODE);
3764 
3765 	/* This is required by WaCatErrorRejectionIssue */
3766 	I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
3767 			I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
3768 			GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
3769 
3770 	for_each_pipe(pipe) {
3771 		I915_WRITE(DSPCNTR(pipe),
3772 			   I915_READ(DSPCNTR(pipe)) |
3773 			   DISPPLANE_TRICKLE_FEED_DISABLE);
3774 		intel_flush_display_plane(dev_priv, pipe);
3775 	}
3776 
3777 	gen7_setup_fixed_func_scheduler(dev_priv);
3778 
3779 	/* WaDisable4x2SubspanOptimization */
3780 	I915_WRITE(CACHE_MODE_1,
3781 		   _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
3782 
3783 	/* WaMbcDriverBootEnable */
3784 	I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
3785 		   GEN6_MBCTL_ENABLE_BOOT_FETCH);
3786 
3787 	/* XXX: This is a workaround for early silicon revisions and should be
3788 	 * removed later.
3789 	 */
3790 	I915_WRITE(WM_DBG,
3791 			I915_READ(WM_DBG) |
3792 			WM_DBG_DISALLOW_MULTIPLE_LP |
3793 			WM_DBG_DISALLOW_SPRITE |
3794 			WM_DBG_DISALLOW_MAXFIFO);
3795 
3796 	lpt_init_clock_gating(dev);
3797 }
3798 
3799 static void ivybridge_init_clock_gating(struct drm_device *dev)
3800 {
3801 	struct drm_i915_private *dev_priv = dev->dev_private;
3802 	int pipe;
3803 	uint32_t snpcr;
3804 
3805 	I915_WRITE(WM3_LP_ILK, 0);
3806 	I915_WRITE(WM2_LP_ILK, 0);
3807 	I915_WRITE(WM1_LP_ILK, 0);
3808 
3809 	I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
3810 
3811 	/* WaDisableEarlyCull */
3812 	I915_WRITE(_3D_CHICKEN3,
3813 		   _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
3814 
3815 	/* WaDisableBackToBackFlipFix */
3816 	I915_WRITE(IVB_CHICKEN3,
3817 		   CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
3818 		   CHICKEN3_DGMG_DONE_FIX_DISABLE);
3819 
3820 	/* WaDisablePSDDualDispatchEnable */
3821 	if (IS_IVB_GT1(dev))
3822 		I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
3823 			   _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
3824 	else
3825 		I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2,
3826 			   _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
3827 
3828 	/* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3829 	I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
3830 		   GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
3831 
3832 	/* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3833 	I915_WRITE(GEN7_L3CNTLREG1,
3834 			GEN7_WA_FOR_GEN7_L3_CONTROL);
3835 	I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
3836 		   GEN7_WA_L3_CHICKEN_MODE);
3837 	if (IS_IVB_GT1(dev))
3838 		I915_WRITE(GEN7_ROW_CHICKEN2,
3839 			   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
3840 	else
3841 		I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
3842 			   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
3843 
3844 
3845 	/* WaForceL3Serialization */
3846 	I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
3847 		   ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
3848 
3849 	/* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3850 	 * gating disable must be set.  Failure to set it results in
3851 	 * flickering pixels due to Z write ordering failures after
3852 	 * some amount of runtime in the Mesa "fire" demo, and Unigine
3853 	 * Sanctuary and Tropics, and apparently anything else with
3854 	 * alpha test or pixel discard.
3855 	 *
3856 	 * According to the spec, bit 11 (RCCUNIT) must also be set,
3857 	 * but we didn't debug actual testcases to find it out.
3858 	 *
3859 	 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3860 	 * This implements the WaDisableRCZUnitClockGating workaround.
3861 	 */
3862 	I915_WRITE(GEN6_UCGCTL2,
3863 		   GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
3864 		   GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
3865 
3866 	/* This is required by WaCatErrorRejectionIssue */
3867 	I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
3868 			I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
3869 			GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
3870 
3871 	for_each_pipe(pipe) {
3872 		I915_WRITE(DSPCNTR(pipe),
3873 			   I915_READ(DSPCNTR(pipe)) |
3874 			   DISPPLANE_TRICKLE_FEED_DISABLE);
3875 		intel_flush_display_plane(dev_priv, pipe);
3876 	}
3877 
3878 	/* WaMbcDriverBootEnable */
3879 	I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
3880 		   GEN6_MBCTL_ENABLE_BOOT_FETCH);
3881 
3882 	gen7_setup_fixed_func_scheduler(dev_priv);
3883 
3884 	/* WaDisable4x2SubspanOptimization */
3885 	I915_WRITE(CACHE_MODE_1,
3886 		   _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
3887 
3888 	snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
3889 	snpcr &= ~GEN6_MBC_SNPCR_MASK;
3890 	snpcr |= GEN6_MBC_SNPCR_MED;
3891 	I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
3892 
3893 	cpt_init_clock_gating(dev);
3894 }
3895 
3896 static void valleyview_init_clock_gating(struct drm_device *dev)
3897 {
3898 	struct drm_i915_private *dev_priv = dev->dev_private;
3899 	int pipe;
3900 
3901 	I915_WRITE(WM3_LP_ILK, 0);
3902 	I915_WRITE(WM2_LP_ILK, 0);
3903 	I915_WRITE(WM1_LP_ILK, 0);
3904 
3905 	I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
3906 
3907 	/* WaDisableEarlyCull */
3908 	I915_WRITE(_3D_CHICKEN3,
3909 		   _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
3910 
3911 	/* WaDisableBackToBackFlipFix */
3912 	I915_WRITE(IVB_CHICKEN3,
3913 		   CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
3914 		   CHICKEN3_DGMG_DONE_FIX_DISABLE);
3915 
3916 	I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
3917 		   _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
3918 
3919 	/* Apply the WaDisableRHWOOptimizationForRenderHang workaround. */
3920 	I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
3921 		   GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
3922 
3923 	/* WaApplyL3ControlAndL3ChickenMode requires those two on Ivy Bridge */
3924 	I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS);
3925 	I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE);
3926 
3927 	/* WaForceL3Serialization */
3928 	I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
3929 		   ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
3930 
3931 	/* WaDisableDopClockGating */
3932 	I915_WRITE(GEN7_ROW_CHICKEN2,
3933 		   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
3934 
3935 	/* WaForceL3Serialization */
3936 	I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
3937 		   ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
3938 
3939 	/* This is required by WaCatErrorRejectionIssue */
3940 	I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
3941 		   I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
3942 		   GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
3943 
3944 	/* WaMbcDriverBootEnable */
3945 	I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) |
3946 		   GEN6_MBCTL_ENABLE_BOOT_FETCH);
3947 
3948 
3949 	/* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
3950 	 * gating disable must be set.  Failure to set it results in
3951 	 * flickering pixels due to Z write ordering failures after
3952 	 * some amount of runtime in the Mesa "fire" demo, and Unigine
3953 	 * Sanctuary and Tropics, and apparently anything else with
3954 	 * alpha test or pixel discard.
3955 	 *
3956 	 * According to the spec, bit 11 (RCCUNIT) must also be set,
3957 	 * but we didn't debug actual testcases to find it out.
3958 	 *
3959 	 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
3960 	 * This implements the WaDisableRCZUnitClockGating workaround.
3961 	 *
3962 	 * Also apply WaDisableVDSUnitClockGating and
3963 	 * WaDisableRCPBUnitClockGating.
3964 	 */
3965 	I915_WRITE(GEN6_UCGCTL2,
3966 		   GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
3967 		   GEN7_TDLUNIT_CLOCK_GATE_DISABLE |
3968 		   GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
3969 		   GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
3970 		   GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
3971 
3972 	I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
3973 
3974 	for_each_pipe(pipe) {
3975 		I915_WRITE(DSPCNTR(pipe),
3976 			   I915_READ(DSPCNTR(pipe)) |
3977 			   DISPPLANE_TRICKLE_FEED_DISABLE);
3978 		intel_flush_display_plane(dev_priv, pipe);
3979 	}
3980 
3981 	I915_WRITE(CACHE_MODE_1,
3982 		   _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
3983 
3984 	/*
3985 	 * On ValleyView, the GUnit needs to signal the GT
3986 	 * when flip and other events complete.  So enable
3987 	 * all the GUnit->GT interrupts here
3988 	 */
3989 	I915_WRITE(VLV_DPFLIPSTAT, PIPEB_LINE_COMPARE_INT_EN |
3990 		   PIPEB_HLINE_INT_EN | PIPEB_VBLANK_INT_EN |
3991 		   SPRITED_FLIPDONE_INT_EN | SPRITEC_FLIPDONE_INT_EN |
3992 		   PLANEB_FLIPDONE_INT_EN | PIPEA_LINE_COMPARE_INT_EN |
3993 		   PIPEA_HLINE_INT_EN | PIPEA_VBLANK_INT_EN |
3994 		   SPRITEB_FLIPDONE_INT_EN | SPRITEA_FLIPDONE_INT_EN |
3995 		   PLANEA_FLIPDONE_INT_EN);
3996 
3997 	/*
3998 	 * WaDisableVLVClockGating_VBIIssue
3999 	 * Disable clock gating on th GCFG unit to prevent a delay
4000 	 * in the reporting of vblank events.
4001 	 */
4002 	I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
4003 }
4004 
4005 static void g4x_init_clock_gating(struct drm_device *dev)
4006 {
4007 	struct drm_i915_private *dev_priv = dev->dev_private;
4008 	uint32_t dspclk_gate;
4009 
4010 	I915_WRITE(RENCLK_GATE_D1, 0);
4011 	I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
4012 		   GS_UNIT_CLOCK_GATE_DISABLE |
4013 		   CL_UNIT_CLOCK_GATE_DISABLE);
4014 	I915_WRITE(RAMCLK_GATE_D, 0);
4015 	dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
4016 		OVRUNIT_CLOCK_GATE_DISABLE |
4017 		OVCUNIT_CLOCK_GATE_DISABLE;
4018 	if (IS_GM45(dev))
4019 		dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
4020 	I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
4021 
4022 	/* WaDisableRenderCachePipelinedFlush */
4023 	I915_WRITE(CACHE_MODE_0,
4024 		   _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
4025 }
4026 
4027 static void crestline_init_clock_gating(struct drm_device *dev)
4028 {
4029 	struct drm_i915_private *dev_priv = dev->dev_private;
4030 
4031 	I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
4032 	I915_WRITE(RENCLK_GATE_D2, 0);
4033 	I915_WRITE(DSPCLK_GATE_D, 0);
4034 	I915_WRITE(RAMCLK_GATE_D, 0);
4035 	I915_WRITE16(DEUC, 0);
4036 }
4037 
4038 static void broadwater_init_clock_gating(struct drm_device *dev)
4039 {
4040 	struct drm_i915_private *dev_priv = dev->dev_private;
4041 
4042 	I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
4043 		   I965_RCC_CLOCK_GATE_DISABLE |
4044 		   I965_RCPB_CLOCK_GATE_DISABLE |
4045 		   I965_ISC_CLOCK_GATE_DISABLE |
4046 		   I965_FBC_CLOCK_GATE_DISABLE);
4047 	I915_WRITE(RENCLK_GATE_D2, 0);
4048 }
4049 
4050 static void gen3_init_clock_gating(struct drm_device *dev)
4051 {
4052 	struct drm_i915_private *dev_priv = dev->dev_private;
4053 	u32 dstate = I915_READ(D_STATE);
4054 
4055 	dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
4056 		DSTATE_DOT_CLOCK_GATING;
4057 	I915_WRITE(D_STATE, dstate);
4058 
4059 	if (IS_PINEVIEW(dev))
4060 		I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
4061 
4062 	/* IIR "flip pending" means done if this bit is set */
4063 	I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
4064 }
4065 
4066 static void i85x_init_clock_gating(struct drm_device *dev)
4067 {
4068 	struct drm_i915_private *dev_priv = dev->dev_private;
4069 
4070 	I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
4071 }
4072 
4073 static void i830_init_clock_gating(struct drm_device *dev)
4074 {
4075 	struct drm_i915_private *dev_priv = dev->dev_private;
4076 
4077 	I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
4078 }
4079 
4080 void intel_init_clock_gating(struct drm_device *dev)
4081 {
4082 	struct drm_i915_private *dev_priv = dev->dev_private;
4083 
4084 	dev_priv->display.init_clock_gating(dev);
4085 }
4086 
4087 /* Starting with Haswell, we have different power wells for
4088  * different parts of the GPU. This attempts to enable them all.
4089  */
4090 void intel_init_power_wells(struct drm_device *dev)
4091 {
4092 	struct drm_i915_private *dev_priv = dev->dev_private;
4093 	unsigned long power_wells[] = {
4094 		HSW_PWR_WELL_CTL1,
4095 		HSW_PWR_WELL_CTL2,
4096 		HSW_PWR_WELL_CTL4
4097 	};
4098 	int i;
4099 
4100 	if (!IS_HASWELL(dev))
4101 		return;
4102 
4103 	mutex_lock(&dev->struct_mutex);
4104 
4105 	for (i = 0; i < ARRAY_SIZE(power_wells); i++) {
4106 		int well = I915_READ(power_wells[i]);
4107 
4108 		if ((well & HSW_PWR_WELL_STATE) == 0) {
4109 			I915_WRITE(power_wells[i], well & HSW_PWR_WELL_ENABLE);
4110 			if (wait_for((I915_READ(power_wells[i]) & HSW_PWR_WELL_STATE), 20))
4111 				DRM_ERROR("Error enabling power well %lx\n", power_wells[i]);
4112 		}
4113 	}
4114 
4115 	mutex_unlock(&dev->struct_mutex);
4116 }
4117 
4118 /* Set up chip specific power management-related functions */
4119 void intel_init_pm(struct drm_device *dev)
4120 {
4121 	struct drm_i915_private *dev_priv = dev->dev_private;
4122 
4123 	if (I915_HAS_FBC(dev)) {
4124 		if (HAS_PCH_SPLIT(dev)) {
4125 			dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
4126 			dev_priv->display.enable_fbc = ironlake_enable_fbc;
4127 			dev_priv->display.disable_fbc = ironlake_disable_fbc;
4128 		} else if (IS_GM45(dev)) {
4129 			dev_priv->display.fbc_enabled = g4x_fbc_enabled;
4130 			dev_priv->display.enable_fbc = g4x_enable_fbc;
4131 			dev_priv->display.disable_fbc = g4x_disable_fbc;
4132 		} else if (IS_CRESTLINE(dev)) {
4133 			dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
4134 			dev_priv->display.enable_fbc = i8xx_enable_fbc;
4135 			dev_priv->display.disable_fbc = i8xx_disable_fbc;
4136 		}
4137 		/* 855GM needs testing */
4138 	}
4139 
4140 	/* For cxsr */
4141 	if (IS_PINEVIEW(dev))
4142 		i915_pineview_get_mem_freq(dev);
4143 	else if (IS_GEN5(dev))
4144 		i915_ironlake_get_mem_freq(dev);
4145 
4146 	/* For FIFO watermark updates */
4147 	if (HAS_PCH_SPLIT(dev)) {
4148 		if (IS_GEN5(dev)) {
4149 			if (I915_READ(MLTR_ILK) & ILK_SRLT_MASK)
4150 				dev_priv->display.update_wm = ironlake_update_wm;
4151 			else {
4152 				DRM_DEBUG_KMS("Failed to get proper latency. "
4153 					      "Disable CxSR\n");
4154 				dev_priv->display.update_wm = NULL;
4155 			}
4156 			dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
4157 		} else if (IS_GEN6(dev)) {
4158 			if (SNB_READ_WM0_LATENCY()) {
4159 				dev_priv->display.update_wm = sandybridge_update_wm;
4160 				dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4161 			} else {
4162 				DRM_DEBUG_KMS("Failed to read display plane latency. "
4163 					      "Disable CxSR\n");
4164 				dev_priv->display.update_wm = NULL;
4165 			}
4166 			dev_priv->display.init_clock_gating = gen6_init_clock_gating;
4167 		} else if (IS_IVYBRIDGE(dev)) {
4168 			/* FIXME: detect B0+ stepping and use auto training */
4169 			if (SNB_READ_WM0_LATENCY()) {
4170 				dev_priv->display.update_wm = ivybridge_update_wm;
4171 				dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4172 			} else {
4173 				DRM_DEBUG_KMS("Failed to read display plane latency. "
4174 					      "Disable CxSR\n");
4175 				dev_priv->display.update_wm = NULL;
4176 			}
4177 			dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
4178 		} else if (IS_HASWELL(dev)) {
4179 			if (SNB_READ_WM0_LATENCY()) {
4180 				dev_priv->display.update_wm = sandybridge_update_wm;
4181 				dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
4182 				dev_priv->display.update_linetime_wm = haswell_update_linetime_wm;
4183 			} else {
4184 				DRM_DEBUG_KMS("Failed to read display plane latency. "
4185 					      "Disable CxSR\n");
4186 				dev_priv->display.update_wm = NULL;
4187 			}
4188 			dev_priv->display.init_clock_gating = haswell_init_clock_gating;
4189 		} else
4190 			dev_priv->display.update_wm = NULL;
4191 	} else if (IS_VALLEYVIEW(dev)) {
4192 		dev_priv->display.update_wm = valleyview_update_wm;
4193 		dev_priv->display.init_clock_gating =
4194 			valleyview_init_clock_gating;
4195 	} else if (IS_PINEVIEW(dev)) {
4196 		if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
4197 					    dev_priv->is_ddr3,
4198 					    dev_priv->fsb_freq,
4199 					    dev_priv->mem_freq)) {
4200 			DRM_INFO("failed to find known CxSR latency "
4201 				 "(found ddr%s fsb freq %d, mem freq %d), "
4202 				 "disabling CxSR\n",
4203 				 (dev_priv->is_ddr3 == 1) ? "3" : "2",
4204 				 dev_priv->fsb_freq, dev_priv->mem_freq);
4205 			/* Disable CxSR and never update its watermark again */
4206 			pineview_disable_cxsr(dev);
4207 			dev_priv->display.update_wm = NULL;
4208 		} else
4209 			dev_priv->display.update_wm = pineview_update_wm;
4210 		dev_priv->display.init_clock_gating = gen3_init_clock_gating;
4211 	} else if (IS_G4X(dev)) {
4212 		dev_priv->display.update_wm = g4x_update_wm;
4213 		dev_priv->display.init_clock_gating = g4x_init_clock_gating;
4214 	} else if (IS_GEN4(dev)) {
4215 		dev_priv->display.update_wm = i965_update_wm;
4216 		if (IS_CRESTLINE(dev))
4217 			dev_priv->display.init_clock_gating = crestline_init_clock_gating;
4218 		else if (IS_BROADWATER(dev))
4219 			dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
4220 	} else if (IS_GEN3(dev)) {
4221 		dev_priv->display.update_wm = i9xx_update_wm;
4222 		dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
4223 		dev_priv->display.init_clock_gating = gen3_init_clock_gating;
4224 	} else if (IS_I865G(dev)) {
4225 		dev_priv->display.update_wm = i830_update_wm;
4226 		dev_priv->display.init_clock_gating = i85x_init_clock_gating;
4227 		dev_priv->display.get_fifo_size = i830_get_fifo_size;
4228 	} else if (IS_I85X(dev)) {
4229 		dev_priv->display.update_wm = i9xx_update_wm;
4230 		dev_priv->display.get_fifo_size = i85x_get_fifo_size;
4231 		dev_priv->display.init_clock_gating = i85x_init_clock_gating;
4232 	} else {
4233 		dev_priv->display.update_wm = i830_update_wm;
4234 		dev_priv->display.init_clock_gating = i830_init_clock_gating;
4235 		if (IS_845G(dev))
4236 			dev_priv->display.get_fifo_size = i845_get_fifo_size;
4237 		else
4238 			dev_priv->display.get_fifo_size = i830_get_fifo_size;
4239 	}
4240 }
4241 
4242 static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv)
4243 {
4244 	u32 gt_thread_status_mask;
4245 
4246 	if (IS_HASWELL(dev_priv->dev))
4247 		gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK_HSW;
4248 	else
4249 		gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK;
4250 
4251 	/* w/a for a sporadic read returning 0 by waiting for the GT
4252 	 * thread to wake up.
4253 	 */
4254 	if (wait_for_atomic_us((I915_READ_NOTRACE(GEN6_GT_THREAD_STATUS_REG) & gt_thread_status_mask) == 0, 500))
4255 		DRM_ERROR("GT thread status wait timed out\n");
4256 }
4257 
4258 static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv)
4259 {
4260 	I915_WRITE_NOTRACE(FORCEWAKE, 0);
4261 	POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */
4262 }
4263 
4264 static void __gen6_gt_force_wake_get(struct drm_i915_private *dev_priv)
4265 {
4266 	u32 forcewake_ack;
4267 
4268 	if (IS_HASWELL(dev_priv->dev))
4269 		forcewake_ack = FORCEWAKE_ACK_HSW;
4270 	else
4271 		forcewake_ack = FORCEWAKE_ACK;
4272 
4273 	if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & 1) == 0,
4274 			    FORCEWAKE_ACK_TIMEOUT_MS))
4275 		DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4276 
4277 	I915_WRITE_NOTRACE(FORCEWAKE, FORCEWAKE_KERNEL);
4278 	POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */
4279 
4280 	if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & 1),
4281 			    FORCEWAKE_ACK_TIMEOUT_MS))
4282 		DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4283 
4284 	__gen6_gt_wait_for_thread_c0(dev_priv);
4285 }
4286 
4287 static void __gen6_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv)
4288 {
4289 	I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(0xffff));
4290 	/* something from same cacheline, but !FORCEWAKE_MT */
4291 	POSTING_READ(ECOBUS);
4292 }
4293 
4294 static void __gen6_gt_force_wake_mt_get(struct drm_i915_private *dev_priv)
4295 {
4296 	u32 forcewake_ack;
4297 
4298 	if (IS_HASWELL(dev_priv->dev))
4299 		forcewake_ack = FORCEWAKE_ACK_HSW;
4300 	else
4301 		forcewake_ack = FORCEWAKE_MT_ACK;
4302 
4303 	if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & 1) == 0,
4304 			    FORCEWAKE_ACK_TIMEOUT_MS))
4305 		DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4306 
4307 	I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4308 	/* something from same cacheline, but !FORCEWAKE_MT */
4309 	POSTING_READ(ECOBUS);
4310 
4311 	if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & 1),
4312 			    FORCEWAKE_ACK_TIMEOUT_MS))
4313 		DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4314 
4315 	__gen6_gt_wait_for_thread_c0(dev_priv);
4316 }
4317 
4318 /*
4319  * Generally this is called implicitly by the register read function. However,
4320  * if some sequence requires the GT to not power down then this function should
4321  * be called at the beginning of the sequence followed by a call to
4322  * gen6_gt_force_wake_put() at the end of the sequence.
4323  */
4324 void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv)
4325 {
4326 	unsigned long irqflags;
4327 
4328 	spin_lock_irqsave(&dev_priv->gt_lock, irqflags);
4329 	if (dev_priv->forcewake_count++ == 0)
4330 		dev_priv->gt.force_wake_get(dev_priv);
4331 	spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags);
4332 }
4333 
4334 void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv)
4335 {
4336 	u32 gtfifodbg;
4337 	gtfifodbg = I915_READ_NOTRACE(GTFIFODBG);
4338 	if (WARN(gtfifodbg & GT_FIFO_CPU_ERROR_MASK,
4339 	     "MMIO read or write has been dropped %x\n", gtfifodbg))
4340 		I915_WRITE_NOTRACE(GTFIFODBG, GT_FIFO_CPU_ERROR_MASK);
4341 }
4342 
4343 static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv)
4344 {
4345 	I915_WRITE_NOTRACE(FORCEWAKE, 0);
4346 	/* something from same cacheline, but !FORCEWAKE */
4347 	POSTING_READ(ECOBUS);
4348 	gen6_gt_check_fifodbg(dev_priv);
4349 }
4350 
4351 static void __gen6_gt_force_wake_mt_put(struct drm_i915_private *dev_priv)
4352 {
4353 	I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4354 	/* something from same cacheline, but !FORCEWAKE_MT */
4355 	POSTING_READ(ECOBUS);
4356 	gen6_gt_check_fifodbg(dev_priv);
4357 }
4358 
4359 /*
4360  * see gen6_gt_force_wake_get()
4361  */
4362 void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv)
4363 {
4364 	unsigned long irqflags;
4365 
4366 	spin_lock_irqsave(&dev_priv->gt_lock, irqflags);
4367 	if (--dev_priv->forcewake_count == 0)
4368 		dev_priv->gt.force_wake_put(dev_priv);
4369 	spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags);
4370 }
4371 
4372 int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv)
4373 {
4374 	int ret = 0;
4375 
4376 	if (dev_priv->gt_fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) {
4377 		int loop = 500;
4378 		u32 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES);
4379 		while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) {
4380 			udelay(10);
4381 			fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES);
4382 		}
4383 		if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES))
4384 			++ret;
4385 		dev_priv->gt_fifo_count = fifo;
4386 	}
4387 	dev_priv->gt_fifo_count--;
4388 
4389 	return ret;
4390 }
4391 
4392 static void vlv_force_wake_reset(struct drm_i915_private *dev_priv)
4393 {
4394 	I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(0xffff));
4395 	/* something from same cacheline, but !FORCEWAKE_VLV */
4396 	POSTING_READ(FORCEWAKE_ACK_VLV);
4397 }
4398 
4399 static void vlv_force_wake_get(struct drm_i915_private *dev_priv)
4400 {
4401 	if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & 1) == 0,
4402 			    FORCEWAKE_ACK_TIMEOUT_MS))
4403 		DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
4404 
4405 	I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
4406 
4407 	if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & 1),
4408 			    FORCEWAKE_ACK_TIMEOUT_MS))
4409 		DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
4410 
4411 	__gen6_gt_wait_for_thread_c0(dev_priv);
4412 }
4413 
4414 static void vlv_force_wake_put(struct drm_i915_private *dev_priv)
4415 {
4416 	I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
4417 	/* something from same cacheline, but !FORCEWAKE_VLV */
4418 	POSTING_READ(FORCEWAKE_ACK_VLV);
4419 	gen6_gt_check_fifodbg(dev_priv);
4420 }
4421 
4422 void intel_gt_reset(struct drm_device *dev)
4423 {
4424 	struct drm_i915_private *dev_priv = dev->dev_private;
4425 
4426 	if (IS_VALLEYVIEW(dev)) {
4427 		vlv_force_wake_reset(dev_priv);
4428 	} else if (INTEL_INFO(dev)->gen >= 6) {
4429 		__gen6_gt_force_wake_reset(dev_priv);
4430 		if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
4431 			__gen6_gt_force_wake_mt_reset(dev_priv);
4432 	}
4433 }
4434 
4435 void intel_gt_init(struct drm_device *dev)
4436 {
4437 	struct drm_i915_private *dev_priv = dev->dev_private;
4438 
4439 	spin_lock_init(&dev_priv->gt_lock);
4440 
4441 	intel_gt_reset(dev);
4442 
4443 	if (IS_VALLEYVIEW(dev)) {
4444 		dev_priv->gt.force_wake_get = vlv_force_wake_get;
4445 		dev_priv->gt.force_wake_put = vlv_force_wake_put;
4446 	} else if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) {
4447 		dev_priv->gt.force_wake_get = __gen6_gt_force_wake_mt_get;
4448 		dev_priv->gt.force_wake_put = __gen6_gt_force_wake_mt_put;
4449 	} else if (IS_GEN6(dev)) {
4450 		dev_priv->gt.force_wake_get = __gen6_gt_force_wake_get;
4451 		dev_priv->gt.force_wake_put = __gen6_gt_force_wake_put;
4452 	}
4453 	INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
4454 			  intel_gen6_powersave_work);
4455 }
4456 
4457 #ifdef __NetBSD__		/* XXX gt fini */
4458 void
4459 intel_gt_fini(struct drm_device *dev)
4460 {
4461 	struct drm_i915_private *dev_priv = dev->dev_private;
4462 
4463 	spin_lock_destroy(&dev_priv->gt_lock);
4464 }
4465 #endif
4466 
4467 int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
4468 {
4469 	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4470 
4471 	if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
4472 		DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
4473 		return -EAGAIN;
4474 	}
4475 
4476 	I915_WRITE(GEN6_PCODE_DATA, *val);
4477 	I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
4478 
4479 	if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
4480 		     500)) {
4481 		DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
4482 		return -ETIMEDOUT;
4483 	}
4484 
4485 	*val = I915_READ(GEN6_PCODE_DATA);
4486 	I915_WRITE(GEN6_PCODE_DATA, 0);
4487 
4488 	return 0;
4489 }
4490 
4491 int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
4492 {
4493 	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
4494 
4495 	if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
4496 		DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
4497 		return -EAGAIN;
4498 	}
4499 
4500 	I915_WRITE(GEN6_PCODE_DATA, val);
4501 	I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
4502 
4503 	if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
4504 		     500)) {
4505 		DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
4506 		return -ETIMEDOUT;
4507 	}
4508 
4509 	I915_WRITE(GEN6_PCODE_DATA, 0);
4510 
4511 	return 0;
4512 }
4513