xref: /linux/drivers/gpu/drm/i915/intel_uncore.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Copyright © 2013 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 
24 #include "i915_drv.h"
25 #include "intel_drv.h"
26 #include "i915_vgpu.h"
27 
28 #include <linux/pm_runtime.h>
29 
30 #define FORCEWAKE_ACK_TIMEOUT_MS 2
31 
32 #define __raw_i915_read8(dev_priv__, reg__) readb((dev_priv__)->regs + (reg__))
33 #define __raw_i915_write8(dev_priv__, reg__, val__) writeb(val__, (dev_priv__)->regs + (reg__))
34 
35 #define __raw_i915_read16(dev_priv__, reg__) readw((dev_priv__)->regs + (reg__))
36 #define __raw_i915_write16(dev_priv__, reg__, val__) writew(val__, (dev_priv__)->regs + (reg__))
37 
38 #define __raw_i915_read32(dev_priv__, reg__) readl((dev_priv__)->regs + (reg__))
39 #define __raw_i915_write32(dev_priv__, reg__, val__) writel(val__, (dev_priv__)->regs + (reg__))
40 
41 #define __raw_i915_read64(dev_priv__, reg__) readq((dev_priv__)->regs + (reg__))
42 #define __raw_i915_write64(dev_priv__, reg__, val__) writeq(val__, (dev_priv__)->regs + (reg__))
43 
44 #define __raw_posting_read(dev_priv__, reg__) (void)__raw_i915_read32(dev_priv__, reg__)
45 
46 static const char * const forcewake_domain_names[] = {
47 	"render",
48 	"blitter",
49 	"media",
50 };
51 
52 const char *
53 intel_uncore_forcewake_domain_to_str(const enum forcewake_domain_id id)
54 {
55 	BUILD_BUG_ON((sizeof(forcewake_domain_names)/sizeof(const char *)) !=
56 		     FW_DOMAIN_ID_COUNT);
57 
58 	if (id >= 0 && id < FW_DOMAIN_ID_COUNT)
59 		return forcewake_domain_names[id];
60 
61 	WARN_ON(id);
62 
63 	return "unknown";
64 }
65 
66 static void
67 assert_device_not_suspended(struct drm_i915_private *dev_priv)
68 {
69 	WARN_ONCE(HAS_RUNTIME_PM(dev_priv->dev) && dev_priv->pm.suspended,
70 		  "Device suspended\n");
71 }
72 
73 static inline void
74 fw_domain_reset(const struct intel_uncore_forcewake_domain *d)
75 {
76 	WARN_ON(d->reg_set == 0);
77 	__raw_i915_write32(d->i915, d->reg_set, d->val_reset);
78 }
79 
80 static inline void
81 fw_domain_arm_timer(struct intel_uncore_forcewake_domain *d)
82 {
83 	mod_timer_pinned(&d->timer, jiffies + 1);
84 }
85 
86 static inline void
87 fw_domain_wait_ack_clear(const struct intel_uncore_forcewake_domain *d)
88 {
89 	if (wait_for_atomic((__raw_i915_read32(d->i915, d->reg_ack) &
90 			     FORCEWAKE_KERNEL) == 0,
91 			    FORCEWAKE_ACK_TIMEOUT_MS))
92 		DRM_ERROR("%s: timed out waiting for forcewake ack to clear.\n",
93 			  intel_uncore_forcewake_domain_to_str(d->id));
94 }
95 
96 static inline void
97 fw_domain_get(const struct intel_uncore_forcewake_domain *d)
98 {
99 	__raw_i915_write32(d->i915, d->reg_set, d->val_set);
100 }
101 
102 static inline void
103 fw_domain_wait_ack(const struct intel_uncore_forcewake_domain *d)
104 {
105 	if (wait_for_atomic((__raw_i915_read32(d->i915, d->reg_ack) &
106 			     FORCEWAKE_KERNEL),
107 			    FORCEWAKE_ACK_TIMEOUT_MS))
108 		DRM_ERROR("%s: timed out waiting for forcewake ack request.\n",
109 			  intel_uncore_forcewake_domain_to_str(d->id));
110 }
111 
112 static inline void
113 fw_domain_put(const struct intel_uncore_forcewake_domain *d)
114 {
115 	__raw_i915_write32(d->i915, d->reg_set, d->val_clear);
116 }
117 
118 static inline void
119 fw_domain_posting_read(const struct intel_uncore_forcewake_domain *d)
120 {
121 	/* something from same cacheline, but not from the set register */
122 	if (d->reg_post)
123 		__raw_posting_read(d->i915, d->reg_post);
124 }
125 
126 static void
127 fw_domains_get(struct drm_i915_private *dev_priv, enum forcewake_domains fw_domains)
128 {
129 	struct intel_uncore_forcewake_domain *d;
130 	enum forcewake_domain_id id;
131 
132 	for_each_fw_domain_mask(d, fw_domains, dev_priv, id) {
133 		fw_domain_wait_ack_clear(d);
134 		fw_domain_get(d);
135 		fw_domain_wait_ack(d);
136 	}
137 }
138 
139 static void
140 fw_domains_put(struct drm_i915_private *dev_priv, enum forcewake_domains fw_domains)
141 {
142 	struct intel_uncore_forcewake_domain *d;
143 	enum forcewake_domain_id id;
144 
145 	for_each_fw_domain_mask(d, fw_domains, dev_priv, id) {
146 		fw_domain_put(d);
147 		fw_domain_posting_read(d);
148 	}
149 }
150 
151 static void
152 fw_domains_posting_read(struct drm_i915_private *dev_priv)
153 {
154 	struct intel_uncore_forcewake_domain *d;
155 	enum forcewake_domain_id id;
156 
157 	/* No need to do for all, just do for first found */
158 	for_each_fw_domain(d, dev_priv, id) {
159 		fw_domain_posting_read(d);
160 		break;
161 	}
162 }
163 
164 static void
165 fw_domains_reset(struct drm_i915_private *dev_priv, enum forcewake_domains fw_domains)
166 {
167 	struct intel_uncore_forcewake_domain *d;
168 	enum forcewake_domain_id id;
169 
170 	if (dev_priv->uncore.fw_domains == 0)
171 		return;
172 
173 	for_each_fw_domain_mask(d, fw_domains, dev_priv, id)
174 		fw_domain_reset(d);
175 
176 	fw_domains_posting_read(dev_priv);
177 }
178 
179 static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv)
180 {
181 	/* w/a for a sporadic read returning 0 by waiting for the GT
182 	 * thread to wake up.
183 	 */
184 	if (wait_for_atomic_us((__raw_i915_read32(dev_priv, GEN6_GT_THREAD_STATUS_REG) &
185 				GEN6_GT_THREAD_STATUS_CORE_MASK) == 0, 500))
186 		DRM_ERROR("GT thread status wait timed out\n");
187 }
188 
189 static void fw_domains_get_with_thread_status(struct drm_i915_private *dev_priv,
190 					      enum forcewake_domains fw_domains)
191 {
192 	fw_domains_get(dev_priv, fw_domains);
193 
194 	/* WaRsForcewakeWaitTC0:snb,ivb,hsw,bdw,vlv */
195 	__gen6_gt_wait_for_thread_c0(dev_priv);
196 }
197 
198 static void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv)
199 {
200 	u32 gtfifodbg;
201 
202 	gtfifodbg = __raw_i915_read32(dev_priv, GTFIFODBG);
203 	if (WARN(gtfifodbg, "GT wake FIFO error 0x%x\n", gtfifodbg))
204 		__raw_i915_write32(dev_priv, GTFIFODBG, gtfifodbg);
205 }
206 
207 static void fw_domains_put_with_fifo(struct drm_i915_private *dev_priv,
208 				     enum forcewake_domains fw_domains)
209 {
210 	fw_domains_put(dev_priv, fw_domains);
211 	gen6_gt_check_fifodbg(dev_priv);
212 }
213 
214 static inline u32 fifo_free_entries(struct drm_i915_private *dev_priv)
215 {
216 	u32 count = __raw_i915_read32(dev_priv, GTFIFOCTL);
217 
218 	return count & GT_FIFO_FREE_ENTRIES_MASK;
219 }
220 
221 static int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv)
222 {
223 	int ret = 0;
224 
225 	/* On VLV, FIFO will be shared by both SW and HW.
226 	 * So, we need to read the FREE_ENTRIES everytime */
227 	if (IS_VALLEYVIEW(dev_priv->dev))
228 		dev_priv->uncore.fifo_count = fifo_free_entries(dev_priv);
229 
230 	if (dev_priv->uncore.fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) {
231 		int loop = 500;
232 		u32 fifo = fifo_free_entries(dev_priv);
233 
234 		while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) {
235 			udelay(10);
236 			fifo = fifo_free_entries(dev_priv);
237 		}
238 		if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES))
239 			++ret;
240 		dev_priv->uncore.fifo_count = fifo;
241 	}
242 	dev_priv->uncore.fifo_count--;
243 
244 	return ret;
245 }
246 
247 static void intel_uncore_fw_release_timer(unsigned long arg)
248 {
249 	struct intel_uncore_forcewake_domain *domain = (void *)arg;
250 	unsigned long irqflags;
251 
252 	assert_device_not_suspended(domain->i915);
253 
254 	spin_lock_irqsave(&domain->i915->uncore.lock, irqflags);
255 	if (WARN_ON(domain->wake_count == 0))
256 		domain->wake_count++;
257 
258 	if (--domain->wake_count == 0)
259 		domain->i915->uncore.funcs.force_wake_put(domain->i915,
260 							  1 << domain->id);
261 
262 	spin_unlock_irqrestore(&domain->i915->uncore.lock, irqflags);
263 }
264 
265 void intel_uncore_forcewake_reset(struct drm_device *dev, bool restore)
266 {
267 	struct drm_i915_private *dev_priv = dev->dev_private;
268 	unsigned long irqflags;
269 	struct intel_uncore_forcewake_domain *domain;
270 	int retry_count = 100;
271 	enum forcewake_domain_id id;
272 	enum forcewake_domains fw = 0, active_domains;
273 
274 	/* Hold uncore.lock across reset to prevent any register access
275 	 * with forcewake not set correctly. Wait until all pending
276 	 * timers are run before holding.
277 	 */
278 	while (1) {
279 		active_domains = 0;
280 
281 		for_each_fw_domain(domain, dev_priv, id) {
282 			if (del_timer_sync(&domain->timer) == 0)
283 				continue;
284 
285 			intel_uncore_fw_release_timer((unsigned long)domain);
286 		}
287 
288 		spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
289 
290 		for_each_fw_domain(domain, dev_priv, id) {
291 			if (timer_pending(&domain->timer))
292 				active_domains |= (1 << id);
293 		}
294 
295 		if (active_domains == 0)
296 			break;
297 
298 		if (--retry_count == 0) {
299 			DRM_ERROR("Timed out waiting for forcewake timers to finish\n");
300 			break;
301 		}
302 
303 		spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
304 		cond_resched();
305 	}
306 
307 	WARN_ON(active_domains);
308 
309 	for_each_fw_domain(domain, dev_priv, id)
310 		if (domain->wake_count)
311 			fw |= 1 << id;
312 
313 	if (fw)
314 		dev_priv->uncore.funcs.force_wake_put(dev_priv, fw);
315 
316 	fw_domains_reset(dev_priv, FORCEWAKE_ALL);
317 
318 	if (restore) { /* If reset with a user forcewake, try to restore */
319 		if (fw)
320 			dev_priv->uncore.funcs.force_wake_get(dev_priv, fw);
321 
322 		if (IS_GEN6(dev) || IS_GEN7(dev))
323 			dev_priv->uncore.fifo_count =
324 				fifo_free_entries(dev_priv);
325 	}
326 
327 	if (!restore)
328 		assert_forcewakes_inactive(dev_priv);
329 
330 	spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
331 }
332 
333 static void intel_uncore_ellc_detect(struct drm_device *dev)
334 {
335 	struct drm_i915_private *dev_priv = dev->dev_private;
336 
337 	if ((IS_HASWELL(dev) || IS_BROADWELL(dev) ||
338 	     INTEL_INFO(dev)->gen >= 9) &&
339 	    (__raw_i915_read32(dev_priv, HSW_EDRAM_PRESENT) & EDRAM_ENABLED)) {
340 		/* The docs do not explain exactly how the calculation can be
341 		 * made. It is somewhat guessable, but for now, it's always
342 		 * 128MB.
343 		 * NB: We can't write IDICR yet because we do not have gt funcs
344 		 * set up */
345 		dev_priv->ellc_size = 128;
346 		DRM_INFO("Found %zuMB of eLLC\n", dev_priv->ellc_size);
347 	}
348 }
349 
350 static void __intel_uncore_early_sanitize(struct drm_device *dev,
351 					  bool restore_forcewake)
352 {
353 	struct drm_i915_private *dev_priv = dev->dev_private;
354 
355 	if (HAS_FPGA_DBG_UNCLAIMED(dev))
356 		__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
357 
358 	/* clear out old GT FIFO errors */
359 	if (IS_GEN6(dev) || IS_GEN7(dev))
360 		__raw_i915_write32(dev_priv, GTFIFODBG,
361 				   __raw_i915_read32(dev_priv, GTFIFODBG));
362 
363 	/* WaDisableShadowRegForCpd:chv */
364 	if (IS_CHERRYVIEW(dev)) {
365 		__raw_i915_write32(dev_priv, GTFIFOCTL,
366 				   __raw_i915_read32(dev_priv, GTFIFOCTL) |
367 				   GT_FIFO_CTL_BLOCK_ALL_POLICY_STALL |
368 				   GT_FIFO_CTL_RC6_POLICY_STALL);
369 	}
370 
371 	intel_uncore_forcewake_reset(dev, restore_forcewake);
372 }
373 
374 void intel_uncore_early_sanitize(struct drm_device *dev, bool restore_forcewake)
375 {
376 	__intel_uncore_early_sanitize(dev, restore_forcewake);
377 	i915_check_and_clear_faults(dev);
378 }
379 
380 void intel_uncore_sanitize(struct drm_device *dev)
381 {
382 	/* BIOS often leaves RC6 enabled, but disable it for hw init */
383 	intel_disable_gt_powersave(dev);
384 }
385 
386 static void __intel_uncore_forcewake_get(struct drm_i915_private *dev_priv,
387 					 enum forcewake_domains fw_domains)
388 {
389 	struct intel_uncore_forcewake_domain *domain;
390 	enum forcewake_domain_id id;
391 
392 	if (!dev_priv->uncore.funcs.force_wake_get)
393 		return;
394 
395 	fw_domains &= dev_priv->uncore.fw_domains;
396 
397 	for_each_fw_domain_mask(domain, fw_domains, dev_priv, id) {
398 		if (domain->wake_count++)
399 			fw_domains &= ~(1 << id);
400 	}
401 
402 	if (fw_domains)
403 		dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
404 }
405 
406 /**
407  * intel_uncore_forcewake_get - grab forcewake domain references
408  * @dev_priv: i915 device instance
409  * @fw_domains: forcewake domains to get reference on
410  *
411  * This function can be used get GT's forcewake domain references.
412  * Normal register access will handle the forcewake domains automatically.
413  * However if some sequence requires the GT to not power down a particular
414  * forcewake domains this function should be called at the beginning of the
415  * sequence. And subsequently the reference should be dropped by symmetric
416  * call to intel_unforce_forcewake_put(). Usually caller wants all the domains
417  * to be kept awake so the @fw_domains would be then FORCEWAKE_ALL.
418  */
419 void intel_uncore_forcewake_get(struct drm_i915_private *dev_priv,
420 				enum forcewake_domains fw_domains)
421 {
422 	unsigned long irqflags;
423 
424 	if (!dev_priv->uncore.funcs.force_wake_get)
425 		return;
426 
427 	WARN_ON(dev_priv->pm.suspended);
428 
429 	spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
430 	__intel_uncore_forcewake_get(dev_priv, fw_domains);
431 	spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
432 }
433 
434 /**
435  * intel_uncore_forcewake_get__locked - grab forcewake domain references
436  * @dev_priv: i915 device instance
437  * @fw_domains: forcewake domains to get reference on
438  *
439  * See intel_uncore_forcewake_get(). This variant places the onus
440  * on the caller to explicitly handle the dev_priv->uncore.lock spinlock.
441  */
442 void intel_uncore_forcewake_get__locked(struct drm_i915_private *dev_priv,
443 					enum forcewake_domains fw_domains)
444 {
445 	assert_spin_locked(&dev_priv->uncore.lock);
446 
447 	if (!dev_priv->uncore.funcs.force_wake_get)
448 		return;
449 
450 	__intel_uncore_forcewake_get(dev_priv, fw_domains);
451 }
452 
453 static void __intel_uncore_forcewake_put(struct drm_i915_private *dev_priv,
454 					 enum forcewake_domains fw_domains)
455 {
456 	struct intel_uncore_forcewake_domain *domain;
457 	enum forcewake_domain_id id;
458 
459 	if (!dev_priv->uncore.funcs.force_wake_put)
460 		return;
461 
462 	fw_domains &= dev_priv->uncore.fw_domains;
463 
464 	for_each_fw_domain_mask(domain, fw_domains, dev_priv, id) {
465 		if (WARN_ON(domain->wake_count == 0))
466 			continue;
467 
468 		if (--domain->wake_count)
469 			continue;
470 
471 		domain->wake_count++;
472 		fw_domain_arm_timer(domain);
473 	}
474 }
475 
476 /**
477  * intel_uncore_forcewake_put - release a forcewake domain reference
478  * @dev_priv: i915 device instance
479  * @fw_domains: forcewake domains to put references
480  *
481  * This function drops the device-level forcewakes for specified
482  * domains obtained by intel_uncore_forcewake_get().
483  */
484 void intel_uncore_forcewake_put(struct drm_i915_private *dev_priv,
485 				enum forcewake_domains fw_domains)
486 {
487 	unsigned long irqflags;
488 
489 	if (!dev_priv->uncore.funcs.force_wake_put)
490 		return;
491 
492 	spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
493 	__intel_uncore_forcewake_put(dev_priv, fw_domains);
494 	spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
495 }
496 
497 /**
498  * intel_uncore_forcewake_put__locked - grab forcewake domain references
499  * @dev_priv: i915 device instance
500  * @fw_domains: forcewake domains to get reference on
501  *
502  * See intel_uncore_forcewake_put(). This variant places the onus
503  * on the caller to explicitly handle the dev_priv->uncore.lock spinlock.
504  */
505 void intel_uncore_forcewake_put__locked(struct drm_i915_private *dev_priv,
506 					enum forcewake_domains fw_domains)
507 {
508 	assert_spin_locked(&dev_priv->uncore.lock);
509 
510 	if (!dev_priv->uncore.funcs.force_wake_put)
511 		return;
512 
513 	__intel_uncore_forcewake_put(dev_priv, fw_domains);
514 }
515 
516 void assert_forcewakes_inactive(struct drm_i915_private *dev_priv)
517 {
518 	struct intel_uncore_forcewake_domain *domain;
519 	enum forcewake_domain_id id;
520 
521 	if (!dev_priv->uncore.funcs.force_wake_get)
522 		return;
523 
524 	for_each_fw_domain(domain, dev_priv, id)
525 		WARN_ON(domain->wake_count);
526 }
527 
528 /* We give fast paths for the really cool registers */
529 #define NEEDS_FORCE_WAKE(dev_priv, reg) \
530 	 ((reg) < 0x40000 && (reg) != FORCEWAKE)
531 
532 #define REG_RANGE(reg, start, end) ((reg) >= (start) && (reg) < (end))
533 
534 #define FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg) \
535 	(REG_RANGE((reg), 0x2000, 0x4000) || \
536 	 REG_RANGE((reg), 0x5000, 0x8000) || \
537 	 REG_RANGE((reg), 0xB000, 0x12000) || \
538 	 REG_RANGE((reg), 0x2E000, 0x30000))
539 
540 #define FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg) \
541 	(REG_RANGE((reg), 0x12000, 0x14000) || \
542 	 REG_RANGE((reg), 0x22000, 0x24000) || \
543 	 REG_RANGE((reg), 0x30000, 0x40000))
544 
545 #define FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg) \
546 	(REG_RANGE((reg), 0x2000, 0x4000) || \
547 	 REG_RANGE((reg), 0x5200, 0x8000) || \
548 	 REG_RANGE((reg), 0x8300, 0x8500) || \
549 	 REG_RANGE((reg), 0xB000, 0xB480) || \
550 	 REG_RANGE((reg), 0xE000, 0xE800))
551 
552 #define FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg) \
553 	(REG_RANGE((reg), 0x8800, 0x8900) || \
554 	 REG_RANGE((reg), 0xD000, 0xD800) || \
555 	 REG_RANGE((reg), 0x12000, 0x14000) || \
556 	 REG_RANGE((reg), 0x1A000, 0x1C000) || \
557 	 REG_RANGE((reg), 0x1E800, 0x1EA00) || \
558 	 REG_RANGE((reg), 0x30000, 0x38000))
559 
560 #define FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg) \
561 	(REG_RANGE((reg), 0x4000, 0x5000) || \
562 	 REG_RANGE((reg), 0x8000, 0x8300) || \
563 	 REG_RANGE((reg), 0x8500, 0x8600) || \
564 	 REG_RANGE((reg), 0x9000, 0xB000) || \
565 	 REG_RANGE((reg), 0xF000, 0x10000))
566 
567 #define FORCEWAKE_GEN9_UNCORE_RANGE_OFFSET(reg) \
568 	REG_RANGE((reg), 0xB00,  0x2000)
569 
570 #define FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg) \
571 	(REG_RANGE((reg), 0x2000, 0x2700) || \
572 	 REG_RANGE((reg), 0x3000, 0x4000) || \
573 	 REG_RANGE((reg), 0x5200, 0x8000) || \
574 	 REG_RANGE((reg), 0x8140, 0x8160) || \
575 	 REG_RANGE((reg), 0x8300, 0x8500) || \
576 	 REG_RANGE((reg), 0x8C00, 0x8D00) || \
577 	 REG_RANGE((reg), 0xB000, 0xB480) || \
578 	 REG_RANGE((reg), 0xE000, 0xE900) || \
579 	 REG_RANGE((reg), 0x24400, 0x24800))
580 
581 #define FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg) \
582 	(REG_RANGE((reg), 0x8130, 0x8140) || \
583 	 REG_RANGE((reg), 0x8800, 0x8A00) || \
584 	 REG_RANGE((reg), 0xD000, 0xD800) || \
585 	 REG_RANGE((reg), 0x12000, 0x14000) || \
586 	 REG_RANGE((reg), 0x1A000, 0x1EA00) || \
587 	 REG_RANGE((reg), 0x30000, 0x40000))
588 
589 #define FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg) \
590 	REG_RANGE((reg), 0x9400, 0x9800)
591 
592 #define FORCEWAKE_GEN9_BLITTER_RANGE_OFFSET(reg) \
593 	((reg) < 0x40000 &&\
594 	 !FORCEWAKE_GEN9_UNCORE_RANGE_OFFSET(reg) && \
595 	 !FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg) && \
596 	 !FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg) && \
597 	 !FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg))
598 
599 static void
600 ilk_dummy_write(struct drm_i915_private *dev_priv)
601 {
602 	/* WaIssueDummyWriteToWakeupFromRC6:ilk Issue a dummy write to wake up
603 	 * the chip from rc6 before touching it for real. MI_MODE is masked,
604 	 * hence harmless to write 0 into. */
605 	__raw_i915_write32(dev_priv, MI_MODE, 0);
606 }
607 
608 static void
609 hsw_unclaimed_reg_debug(struct drm_i915_private *dev_priv, u32 reg, bool read,
610 			bool before)
611 {
612 	const char *op = read ? "reading" : "writing to";
613 	const char *when = before ? "before" : "after";
614 
615 	if (!i915.mmio_debug)
616 		return;
617 
618 	if (__raw_i915_read32(dev_priv, FPGA_DBG) & FPGA_DBG_RM_NOCLAIM) {
619 		WARN(1, "Unclaimed register detected %s %s register 0x%x\n",
620 		     when, op, reg);
621 		__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
622 		i915.mmio_debug--; /* Only report the first N failures */
623 	}
624 }
625 
626 static void
627 hsw_unclaimed_reg_detect(struct drm_i915_private *dev_priv)
628 {
629 	static bool mmio_debug_once = true;
630 
631 	if (i915.mmio_debug || !mmio_debug_once)
632 		return;
633 
634 	if (__raw_i915_read32(dev_priv, FPGA_DBG) & FPGA_DBG_RM_NOCLAIM) {
635 		DRM_DEBUG("Unclaimed register detected, "
636 			  "enabling oneshot unclaimed register reporting. "
637 			  "Please use i915.mmio_debug=N for more information.\n");
638 		__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
639 		i915.mmio_debug = mmio_debug_once--;
640 	}
641 }
642 
643 #define GEN2_READ_HEADER(x) \
644 	u##x val = 0; \
645 	assert_device_not_suspended(dev_priv);
646 
647 #define GEN2_READ_FOOTER \
648 	trace_i915_reg_rw(false, reg, val, sizeof(val), trace); \
649 	return val
650 
651 #define __gen2_read(x) \
652 static u##x \
653 gen2_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
654 	GEN2_READ_HEADER(x); \
655 	val = __raw_i915_read##x(dev_priv, reg); \
656 	GEN2_READ_FOOTER; \
657 }
658 
659 #define __gen5_read(x) \
660 static u##x \
661 gen5_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
662 	GEN2_READ_HEADER(x); \
663 	ilk_dummy_write(dev_priv); \
664 	val = __raw_i915_read##x(dev_priv, reg); \
665 	GEN2_READ_FOOTER; \
666 }
667 
668 __gen5_read(8)
669 __gen5_read(16)
670 __gen5_read(32)
671 __gen5_read(64)
672 __gen2_read(8)
673 __gen2_read(16)
674 __gen2_read(32)
675 __gen2_read(64)
676 
677 #undef __gen5_read
678 #undef __gen2_read
679 
680 #undef GEN2_READ_FOOTER
681 #undef GEN2_READ_HEADER
682 
683 #define GEN6_READ_HEADER(x) \
684 	unsigned long irqflags; \
685 	u##x val = 0; \
686 	assert_device_not_suspended(dev_priv); \
687 	spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)
688 
689 #define GEN6_READ_FOOTER \
690 	spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags); \
691 	trace_i915_reg_rw(false, reg, val, sizeof(val), trace); \
692 	return val
693 
694 static inline void __force_wake_get(struct drm_i915_private *dev_priv,
695 				    enum forcewake_domains fw_domains)
696 {
697 	struct intel_uncore_forcewake_domain *domain;
698 	enum forcewake_domain_id id;
699 
700 	if (WARN_ON(!fw_domains))
701 		return;
702 
703 	/* Ideally GCC would be constant-fold and eliminate this loop */
704 	for_each_fw_domain_mask(domain, fw_domains, dev_priv, id) {
705 		if (domain->wake_count) {
706 			fw_domains &= ~(1 << id);
707 			continue;
708 		}
709 
710 		domain->wake_count++;
711 		fw_domain_arm_timer(domain);
712 	}
713 
714 	if (fw_domains)
715 		dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
716 }
717 
718 #define __vgpu_read(x) \
719 static u##x \
720 vgpu_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
721 	GEN6_READ_HEADER(x); \
722 	val = __raw_i915_read##x(dev_priv, reg); \
723 	GEN6_READ_FOOTER; \
724 }
725 
726 #define __gen6_read(x) \
727 static u##x \
728 gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
729 	GEN6_READ_HEADER(x); \
730 	hsw_unclaimed_reg_debug(dev_priv, reg, true, true); \
731 	if (NEEDS_FORCE_WAKE((dev_priv), (reg))) \
732 		__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
733 	val = __raw_i915_read##x(dev_priv, reg); \
734 	hsw_unclaimed_reg_debug(dev_priv, reg, true, false); \
735 	GEN6_READ_FOOTER; \
736 }
737 
738 #define __vlv_read(x) \
739 static u##x \
740 vlv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
741 	GEN6_READ_HEADER(x); \
742 	if (FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg)) \
743 		__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
744 	else if (FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg)) \
745 		__force_wake_get(dev_priv, FORCEWAKE_MEDIA); \
746 	val = __raw_i915_read##x(dev_priv, reg); \
747 	GEN6_READ_FOOTER; \
748 }
749 
750 #define __chv_read(x) \
751 static u##x \
752 chv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
753 	GEN6_READ_HEADER(x); \
754 	if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \
755 		__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
756 	else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \
757 		__force_wake_get(dev_priv, FORCEWAKE_MEDIA); \
758 	else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \
759 		__force_wake_get(dev_priv, \
760 				 FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \
761 	val = __raw_i915_read##x(dev_priv, reg); \
762 	GEN6_READ_FOOTER; \
763 }
764 
765 #define SKL_NEEDS_FORCE_WAKE(dev_priv, reg)	\
766 	 ((reg) < 0x40000 && !FORCEWAKE_GEN9_UNCORE_RANGE_OFFSET(reg))
767 
768 #define __gen9_read(x) \
769 static u##x \
770 gen9_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
771 	enum forcewake_domains fw_engine; \
772 	GEN6_READ_HEADER(x); \
773 	if (!SKL_NEEDS_FORCE_WAKE((dev_priv), (reg)))	\
774 		fw_engine = 0; \
775 	else if (FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg))	\
776 		fw_engine = FORCEWAKE_RENDER; \
777 	else if (FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg)) \
778 		fw_engine = FORCEWAKE_MEDIA; \
779 	else if (FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg)) \
780 		fw_engine = FORCEWAKE_RENDER | FORCEWAKE_MEDIA; \
781 	else \
782 		fw_engine = FORCEWAKE_BLITTER; \
783 	if (fw_engine) \
784 		__force_wake_get(dev_priv, fw_engine); \
785 	val = __raw_i915_read##x(dev_priv, reg); \
786 	GEN6_READ_FOOTER; \
787 }
788 
789 __vgpu_read(8)
790 __vgpu_read(16)
791 __vgpu_read(32)
792 __vgpu_read(64)
793 __gen9_read(8)
794 __gen9_read(16)
795 __gen9_read(32)
796 __gen9_read(64)
797 __chv_read(8)
798 __chv_read(16)
799 __chv_read(32)
800 __chv_read(64)
801 __vlv_read(8)
802 __vlv_read(16)
803 __vlv_read(32)
804 __vlv_read(64)
805 __gen6_read(8)
806 __gen6_read(16)
807 __gen6_read(32)
808 __gen6_read(64)
809 
810 #undef __gen9_read
811 #undef __chv_read
812 #undef __vlv_read
813 #undef __gen6_read
814 #undef __vgpu_read
815 #undef GEN6_READ_FOOTER
816 #undef GEN6_READ_HEADER
817 
818 #define GEN2_WRITE_HEADER \
819 	trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \
820 	assert_device_not_suspended(dev_priv); \
821 
822 #define GEN2_WRITE_FOOTER
823 
824 #define __gen2_write(x) \
825 static void \
826 gen2_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
827 	GEN2_WRITE_HEADER; \
828 	__raw_i915_write##x(dev_priv, reg, val); \
829 	GEN2_WRITE_FOOTER; \
830 }
831 
832 #define __gen5_write(x) \
833 static void \
834 gen5_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
835 	GEN2_WRITE_HEADER; \
836 	ilk_dummy_write(dev_priv); \
837 	__raw_i915_write##x(dev_priv, reg, val); \
838 	GEN2_WRITE_FOOTER; \
839 }
840 
841 __gen5_write(8)
842 __gen5_write(16)
843 __gen5_write(32)
844 __gen5_write(64)
845 __gen2_write(8)
846 __gen2_write(16)
847 __gen2_write(32)
848 __gen2_write(64)
849 
850 #undef __gen5_write
851 #undef __gen2_write
852 
853 #undef GEN2_WRITE_FOOTER
854 #undef GEN2_WRITE_HEADER
855 
856 #define GEN6_WRITE_HEADER \
857 	unsigned long irqflags; \
858 	trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \
859 	assert_device_not_suspended(dev_priv); \
860 	spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)
861 
862 #define GEN6_WRITE_FOOTER \
863 	spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags)
864 
865 #define __gen6_write(x) \
866 static void \
867 gen6_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
868 	u32 __fifo_ret = 0; \
869 	GEN6_WRITE_HEADER; \
870 	if (NEEDS_FORCE_WAKE((dev_priv), (reg))) { \
871 		__fifo_ret = __gen6_gt_wait_for_fifo(dev_priv); \
872 	} \
873 	__raw_i915_write##x(dev_priv, reg, val); \
874 	if (unlikely(__fifo_ret)) { \
875 		gen6_gt_check_fifodbg(dev_priv); \
876 	} \
877 	GEN6_WRITE_FOOTER; \
878 }
879 
880 #define __hsw_write(x) \
881 static void \
882 hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
883 	u32 __fifo_ret = 0; \
884 	GEN6_WRITE_HEADER; \
885 	if (NEEDS_FORCE_WAKE((dev_priv), (reg))) { \
886 		__fifo_ret = __gen6_gt_wait_for_fifo(dev_priv); \
887 	} \
888 	hsw_unclaimed_reg_debug(dev_priv, reg, false, true); \
889 	__raw_i915_write##x(dev_priv, reg, val); \
890 	if (unlikely(__fifo_ret)) { \
891 		gen6_gt_check_fifodbg(dev_priv); \
892 	} \
893 	hsw_unclaimed_reg_debug(dev_priv, reg, false, false); \
894 	hsw_unclaimed_reg_detect(dev_priv); \
895 	GEN6_WRITE_FOOTER; \
896 }
897 
898 #define __vgpu_write(x) \
899 static void vgpu_write##x(struct drm_i915_private *dev_priv, \
900 			  off_t reg, u##x val, bool trace) { \
901 	GEN6_WRITE_HEADER; \
902 	__raw_i915_write##x(dev_priv, reg, val); \
903 	GEN6_WRITE_FOOTER; \
904 }
905 
906 static const u32 gen8_shadowed_regs[] = {
907 	FORCEWAKE_MT,
908 	GEN6_RPNSWREQ,
909 	GEN6_RC_VIDEO_FREQ,
910 	RING_TAIL(RENDER_RING_BASE),
911 	RING_TAIL(GEN6_BSD_RING_BASE),
912 	RING_TAIL(VEBOX_RING_BASE),
913 	RING_TAIL(BLT_RING_BASE),
914 	/* TODO: Other registers are not yet used */
915 };
916 
917 static bool is_gen8_shadowed(struct drm_i915_private *dev_priv, u32 reg)
918 {
919 	int i;
920 	for (i = 0; i < ARRAY_SIZE(gen8_shadowed_regs); i++)
921 		if (reg == gen8_shadowed_regs[i])
922 			return true;
923 
924 	return false;
925 }
926 
927 #define __gen8_write(x) \
928 static void \
929 gen8_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
930 	GEN6_WRITE_HEADER; \
931 	hsw_unclaimed_reg_debug(dev_priv, reg, false, true); \
932 	if (reg < 0x40000 && !is_gen8_shadowed(dev_priv, reg)) \
933 		__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
934 	__raw_i915_write##x(dev_priv, reg, val); \
935 	hsw_unclaimed_reg_debug(dev_priv, reg, false, false); \
936 	hsw_unclaimed_reg_detect(dev_priv); \
937 	GEN6_WRITE_FOOTER; \
938 }
939 
940 #define __chv_write(x) \
941 static void \
942 chv_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
943 	bool shadowed = is_gen8_shadowed(dev_priv, reg); \
944 	GEN6_WRITE_HEADER; \
945 	if (!shadowed) { \
946 		if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \
947 			__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
948 		else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \
949 			__force_wake_get(dev_priv, FORCEWAKE_MEDIA); \
950 		else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \
951 			__force_wake_get(dev_priv, FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \
952 	} \
953 	__raw_i915_write##x(dev_priv, reg, val); \
954 	GEN6_WRITE_FOOTER; \
955 }
956 
957 static const u32 gen9_shadowed_regs[] = {
958 	RING_TAIL(RENDER_RING_BASE),
959 	RING_TAIL(GEN6_BSD_RING_BASE),
960 	RING_TAIL(VEBOX_RING_BASE),
961 	RING_TAIL(BLT_RING_BASE),
962 	FORCEWAKE_BLITTER_GEN9,
963 	FORCEWAKE_RENDER_GEN9,
964 	FORCEWAKE_MEDIA_GEN9,
965 	GEN6_RPNSWREQ,
966 	GEN6_RC_VIDEO_FREQ,
967 	/* TODO: Other registers are not yet used */
968 };
969 
970 static bool is_gen9_shadowed(struct drm_i915_private *dev_priv, u32 reg)
971 {
972 	int i;
973 	for (i = 0; i < ARRAY_SIZE(gen9_shadowed_regs); i++)
974 		if (reg == gen9_shadowed_regs[i])
975 			return true;
976 
977 	return false;
978 }
979 
980 #define __gen9_write(x) \
981 static void \
982 gen9_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, \
983 		bool trace) { \
984 	enum forcewake_domains fw_engine; \
985 	GEN6_WRITE_HEADER; \
986 	if (!SKL_NEEDS_FORCE_WAKE((dev_priv), (reg)) ||	\
987 	    is_gen9_shadowed(dev_priv, reg)) \
988 		fw_engine = 0; \
989 	else if (FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg)) \
990 		fw_engine = FORCEWAKE_RENDER; \
991 	else if (FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg)) \
992 		fw_engine = FORCEWAKE_MEDIA; \
993 	else if (FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg)) \
994 		fw_engine = FORCEWAKE_RENDER | FORCEWAKE_MEDIA; \
995 	else \
996 		fw_engine = FORCEWAKE_BLITTER; \
997 	if (fw_engine) \
998 		__force_wake_get(dev_priv, fw_engine); \
999 	__raw_i915_write##x(dev_priv, reg, val); \
1000 	GEN6_WRITE_FOOTER; \
1001 }
1002 
1003 __gen9_write(8)
1004 __gen9_write(16)
1005 __gen9_write(32)
1006 __gen9_write(64)
1007 __chv_write(8)
1008 __chv_write(16)
1009 __chv_write(32)
1010 __chv_write(64)
1011 __gen8_write(8)
1012 __gen8_write(16)
1013 __gen8_write(32)
1014 __gen8_write(64)
1015 __hsw_write(8)
1016 __hsw_write(16)
1017 __hsw_write(32)
1018 __hsw_write(64)
1019 __gen6_write(8)
1020 __gen6_write(16)
1021 __gen6_write(32)
1022 __gen6_write(64)
1023 __vgpu_write(8)
1024 __vgpu_write(16)
1025 __vgpu_write(32)
1026 __vgpu_write(64)
1027 
1028 #undef __gen9_write
1029 #undef __chv_write
1030 #undef __gen8_write
1031 #undef __hsw_write
1032 #undef __gen6_write
1033 #undef __vgpu_write
1034 #undef GEN6_WRITE_FOOTER
1035 #undef GEN6_WRITE_HEADER
1036 
1037 #define ASSIGN_WRITE_MMIO_VFUNCS(x) \
1038 do { \
1039 	dev_priv->uncore.funcs.mmio_writeb = x##_write8; \
1040 	dev_priv->uncore.funcs.mmio_writew = x##_write16; \
1041 	dev_priv->uncore.funcs.mmio_writel = x##_write32; \
1042 	dev_priv->uncore.funcs.mmio_writeq = x##_write64; \
1043 } while (0)
1044 
1045 #define ASSIGN_READ_MMIO_VFUNCS(x) \
1046 do { \
1047 	dev_priv->uncore.funcs.mmio_readb = x##_read8; \
1048 	dev_priv->uncore.funcs.mmio_readw = x##_read16; \
1049 	dev_priv->uncore.funcs.mmio_readl = x##_read32; \
1050 	dev_priv->uncore.funcs.mmio_readq = x##_read64; \
1051 } while (0)
1052 
1053 
1054 static void fw_domain_init(struct drm_i915_private *dev_priv,
1055 			   enum forcewake_domain_id domain_id,
1056 			   u32 reg_set, u32 reg_ack)
1057 {
1058 	struct intel_uncore_forcewake_domain *d;
1059 
1060 	if (WARN_ON(domain_id >= FW_DOMAIN_ID_COUNT))
1061 		return;
1062 
1063 	d = &dev_priv->uncore.fw_domain[domain_id];
1064 
1065 	WARN_ON(d->wake_count);
1066 
1067 	d->wake_count = 0;
1068 	d->reg_set = reg_set;
1069 	d->reg_ack = reg_ack;
1070 
1071 	if (IS_GEN6(dev_priv)) {
1072 		d->val_reset = 0;
1073 		d->val_set = FORCEWAKE_KERNEL;
1074 		d->val_clear = 0;
1075 	} else {
1076 		/* WaRsClearFWBitsAtReset:bdw,skl */
1077 		d->val_reset = _MASKED_BIT_DISABLE(0xffff);
1078 		d->val_set = _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL);
1079 		d->val_clear = _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL);
1080 	}
1081 
1082 	if (IS_VALLEYVIEW(dev_priv))
1083 		d->reg_post = FORCEWAKE_ACK_VLV;
1084 	else if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv) || IS_GEN8(dev_priv))
1085 		d->reg_post = ECOBUS;
1086 	else
1087 		d->reg_post = 0;
1088 
1089 	d->i915 = dev_priv;
1090 	d->id = domain_id;
1091 
1092 	setup_timer(&d->timer, intel_uncore_fw_release_timer, (unsigned long)d);
1093 
1094 	dev_priv->uncore.fw_domains |= (1 << domain_id);
1095 
1096 	fw_domain_reset(d);
1097 }
1098 
1099 static void intel_uncore_fw_domains_init(struct drm_device *dev)
1100 {
1101 	struct drm_i915_private *dev_priv = dev->dev_private;
1102 
1103 	if (INTEL_INFO(dev_priv->dev)->gen <= 5)
1104 		return;
1105 
1106 	if (IS_GEN9(dev)) {
1107 		dev_priv->uncore.funcs.force_wake_get = fw_domains_get;
1108 		dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
1109 		fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
1110 			       FORCEWAKE_RENDER_GEN9,
1111 			       FORCEWAKE_ACK_RENDER_GEN9);
1112 		fw_domain_init(dev_priv, FW_DOMAIN_ID_BLITTER,
1113 			       FORCEWAKE_BLITTER_GEN9,
1114 			       FORCEWAKE_ACK_BLITTER_GEN9);
1115 		fw_domain_init(dev_priv, FW_DOMAIN_ID_MEDIA,
1116 			       FORCEWAKE_MEDIA_GEN9, FORCEWAKE_ACK_MEDIA_GEN9);
1117 	} else if (IS_VALLEYVIEW(dev)) {
1118 		dev_priv->uncore.funcs.force_wake_get = fw_domains_get;
1119 		if (!IS_CHERRYVIEW(dev))
1120 			dev_priv->uncore.funcs.force_wake_put =
1121 				fw_domains_put_with_fifo;
1122 		else
1123 			dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
1124 		fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
1125 			       FORCEWAKE_VLV, FORCEWAKE_ACK_VLV);
1126 		fw_domain_init(dev_priv, FW_DOMAIN_ID_MEDIA,
1127 			       FORCEWAKE_MEDIA_VLV, FORCEWAKE_ACK_MEDIA_VLV);
1128 	} else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
1129 		dev_priv->uncore.funcs.force_wake_get =
1130 			fw_domains_get_with_thread_status;
1131 		dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
1132 		fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
1133 			       FORCEWAKE_MT, FORCEWAKE_ACK_HSW);
1134 	} else if (IS_IVYBRIDGE(dev)) {
1135 		u32 ecobus;
1136 
1137 		/* IVB configs may use multi-threaded forcewake */
1138 
1139 		/* A small trick here - if the bios hasn't configured
1140 		 * MT forcewake, and if the device is in RC6, then
1141 		 * force_wake_mt_get will not wake the device and the
1142 		 * ECOBUS read will return zero. Which will be
1143 		 * (correctly) interpreted by the test below as MT
1144 		 * forcewake being disabled.
1145 		 */
1146 		dev_priv->uncore.funcs.force_wake_get =
1147 			fw_domains_get_with_thread_status;
1148 		dev_priv->uncore.funcs.force_wake_put =
1149 			fw_domains_put_with_fifo;
1150 
1151 		/* We need to init first for ECOBUS access and then
1152 		 * determine later if we want to reinit, in case of MT access is
1153 		 * not working. In this stage we don't know which flavour this
1154 		 * ivb is, so it is better to reset also the gen6 fw registers
1155 		 * before the ecobus check.
1156 		 */
1157 
1158 		__raw_i915_write32(dev_priv, FORCEWAKE, 0);
1159 		__raw_posting_read(dev_priv, ECOBUS);
1160 
1161 		fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
1162 			       FORCEWAKE_MT, FORCEWAKE_MT_ACK);
1163 
1164 		mutex_lock(&dev->struct_mutex);
1165 		fw_domains_get_with_thread_status(dev_priv, FORCEWAKE_ALL);
1166 		ecobus = __raw_i915_read32(dev_priv, ECOBUS);
1167 		fw_domains_put_with_fifo(dev_priv, FORCEWAKE_ALL);
1168 		mutex_unlock(&dev->struct_mutex);
1169 
1170 		if (!(ecobus & FORCEWAKE_MT_ENABLE)) {
1171 			DRM_INFO("No MT forcewake available on Ivybridge, this can result in issues\n");
1172 			DRM_INFO("when using vblank-synced partial screen updates.\n");
1173 			fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
1174 				       FORCEWAKE, FORCEWAKE_ACK);
1175 		}
1176 	} else if (IS_GEN6(dev)) {
1177 		dev_priv->uncore.funcs.force_wake_get =
1178 			fw_domains_get_with_thread_status;
1179 		dev_priv->uncore.funcs.force_wake_put =
1180 			fw_domains_put_with_fifo;
1181 		fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
1182 			       FORCEWAKE, FORCEWAKE_ACK);
1183 	}
1184 
1185 	/* All future platforms are expected to require complex power gating */
1186 	WARN_ON(dev_priv->uncore.fw_domains == 0);
1187 }
1188 
1189 void intel_uncore_init(struct drm_device *dev)
1190 {
1191 	struct drm_i915_private *dev_priv = dev->dev_private;
1192 
1193 	i915_check_vgpu(dev);
1194 
1195 	intel_uncore_ellc_detect(dev);
1196 	intel_uncore_fw_domains_init(dev);
1197 	__intel_uncore_early_sanitize(dev, false);
1198 
1199 	switch (INTEL_INFO(dev)->gen) {
1200 	default:
1201 		MISSING_CASE(INTEL_INFO(dev)->gen);
1202 		return;
1203 	case 9:
1204 		ASSIGN_WRITE_MMIO_VFUNCS(gen9);
1205 		ASSIGN_READ_MMIO_VFUNCS(gen9);
1206 		break;
1207 	case 8:
1208 		if (IS_CHERRYVIEW(dev)) {
1209 			ASSIGN_WRITE_MMIO_VFUNCS(chv);
1210 			ASSIGN_READ_MMIO_VFUNCS(chv);
1211 
1212 		} else {
1213 			ASSIGN_WRITE_MMIO_VFUNCS(gen8);
1214 			ASSIGN_READ_MMIO_VFUNCS(gen6);
1215 		}
1216 		break;
1217 	case 7:
1218 	case 6:
1219 		if (IS_HASWELL(dev)) {
1220 			ASSIGN_WRITE_MMIO_VFUNCS(hsw);
1221 		} else {
1222 			ASSIGN_WRITE_MMIO_VFUNCS(gen6);
1223 		}
1224 
1225 		if (IS_VALLEYVIEW(dev)) {
1226 			ASSIGN_READ_MMIO_VFUNCS(vlv);
1227 		} else {
1228 			ASSIGN_READ_MMIO_VFUNCS(gen6);
1229 		}
1230 		break;
1231 	case 5:
1232 		ASSIGN_WRITE_MMIO_VFUNCS(gen5);
1233 		ASSIGN_READ_MMIO_VFUNCS(gen5);
1234 		break;
1235 	case 4:
1236 	case 3:
1237 	case 2:
1238 		ASSIGN_WRITE_MMIO_VFUNCS(gen2);
1239 		ASSIGN_READ_MMIO_VFUNCS(gen2);
1240 		break;
1241 	}
1242 
1243 	if (intel_vgpu_active(dev)) {
1244 		ASSIGN_WRITE_MMIO_VFUNCS(vgpu);
1245 		ASSIGN_READ_MMIO_VFUNCS(vgpu);
1246 	}
1247 
1248 	i915_check_and_clear_faults(dev);
1249 }
1250 #undef ASSIGN_WRITE_MMIO_VFUNCS
1251 #undef ASSIGN_READ_MMIO_VFUNCS
1252 
1253 void intel_uncore_fini(struct drm_device *dev)
1254 {
1255 	/* Paranoia: make sure we have disabled everything before we exit. */
1256 	intel_uncore_sanitize(dev);
1257 	intel_uncore_forcewake_reset(dev, false);
1258 }
1259 
1260 #define GEN_RANGE(l, h) GENMASK(h, l)
1261 
1262 static const struct register_whitelist {
1263 	uint64_t offset;
1264 	uint32_t size;
1265 	/* supported gens, 0x10 for 4, 0x30 for 4 and 5, etc. */
1266 	uint32_t gen_bitmask;
1267 } whitelist[] = {
1268 	{ RING_TIMESTAMP(RENDER_RING_BASE), 8, GEN_RANGE(4, 9) },
1269 };
1270 
1271 int i915_reg_read_ioctl(struct drm_device *dev,
1272 			void *data, struct drm_file *file)
1273 {
1274 	struct drm_i915_private *dev_priv = dev->dev_private;
1275 	struct drm_i915_reg_read *reg = data;
1276 	struct register_whitelist const *entry = whitelist;
1277 	unsigned size;
1278 	u64 offset;
1279 	int i, ret = 0;
1280 
1281 	for (i = 0; i < ARRAY_SIZE(whitelist); i++, entry++) {
1282 		if (entry->offset == (reg->offset & -entry->size) &&
1283 		    (1 << INTEL_INFO(dev)->gen & entry->gen_bitmask))
1284 			break;
1285 	}
1286 
1287 	if (i == ARRAY_SIZE(whitelist))
1288 		return -EINVAL;
1289 
1290 	/* We use the low bits to encode extra flags as the register should
1291 	 * be naturally aligned (and those that are not so aligned merely
1292 	 * limit the available flags for that register).
1293 	 */
1294 	offset = entry->offset;
1295 	size = entry->size;
1296 	size |= reg->offset ^ offset;
1297 
1298 	intel_runtime_pm_get(dev_priv);
1299 
1300 	switch (size) {
1301 	case 8 | 1:
1302 		reg->val = I915_READ64_2x32(offset, offset+4);
1303 		break;
1304 	case 8:
1305 		reg->val = I915_READ64(offset);
1306 		break;
1307 	case 4:
1308 		reg->val = I915_READ(offset);
1309 		break;
1310 	case 2:
1311 		reg->val = I915_READ16(offset);
1312 		break;
1313 	case 1:
1314 		reg->val = I915_READ8(offset);
1315 		break;
1316 	default:
1317 		ret = -EINVAL;
1318 		goto out;
1319 	}
1320 
1321 out:
1322 	intel_runtime_pm_put(dev_priv);
1323 	return ret;
1324 }
1325 
1326 int i915_get_reset_stats_ioctl(struct drm_device *dev,
1327 			       void *data, struct drm_file *file)
1328 {
1329 	struct drm_i915_private *dev_priv = dev->dev_private;
1330 	struct drm_i915_reset_stats *args = data;
1331 	struct i915_ctx_hang_stats *hs;
1332 	struct intel_context *ctx;
1333 	int ret;
1334 
1335 	if (args->flags || args->pad)
1336 		return -EINVAL;
1337 
1338 	if (args->ctx_id == DEFAULT_CONTEXT_HANDLE && !capable(CAP_SYS_ADMIN))
1339 		return -EPERM;
1340 
1341 	ret = mutex_lock_interruptible(&dev->struct_mutex);
1342 	if (ret)
1343 		return ret;
1344 
1345 	ctx = i915_gem_context_get(file->driver_priv, args->ctx_id);
1346 	if (IS_ERR(ctx)) {
1347 		mutex_unlock(&dev->struct_mutex);
1348 		return PTR_ERR(ctx);
1349 	}
1350 	hs = &ctx->hang_stats;
1351 
1352 	if (capable(CAP_SYS_ADMIN))
1353 		args->reset_count = i915_reset_count(&dev_priv->gpu_error);
1354 	else
1355 		args->reset_count = 0;
1356 
1357 	args->batch_active = hs->batch_active;
1358 	args->batch_pending = hs->batch_pending;
1359 
1360 	mutex_unlock(&dev->struct_mutex);
1361 
1362 	return 0;
1363 }
1364 
1365 static int i915_reset_complete(struct drm_device *dev)
1366 {
1367 	u8 gdrst;
1368 	pci_read_config_byte(dev->pdev, I915_GDRST, &gdrst);
1369 	return (gdrst & GRDOM_RESET_STATUS) == 0;
1370 }
1371 
1372 static int i915_do_reset(struct drm_device *dev)
1373 {
1374 	/* assert reset for at least 20 usec */
1375 	pci_write_config_byte(dev->pdev, I915_GDRST, GRDOM_RESET_ENABLE);
1376 	udelay(20);
1377 	pci_write_config_byte(dev->pdev, I915_GDRST, 0);
1378 
1379 	return wait_for(i915_reset_complete(dev), 500);
1380 }
1381 
1382 static int g4x_reset_complete(struct drm_device *dev)
1383 {
1384 	u8 gdrst;
1385 	pci_read_config_byte(dev->pdev, I915_GDRST, &gdrst);
1386 	return (gdrst & GRDOM_RESET_ENABLE) == 0;
1387 }
1388 
1389 static int g33_do_reset(struct drm_device *dev)
1390 {
1391 	pci_write_config_byte(dev->pdev, I915_GDRST, GRDOM_RESET_ENABLE);
1392 	return wait_for(g4x_reset_complete(dev), 500);
1393 }
1394 
1395 static int g4x_do_reset(struct drm_device *dev)
1396 {
1397 	struct drm_i915_private *dev_priv = dev->dev_private;
1398 	int ret;
1399 
1400 	pci_write_config_byte(dev->pdev, I915_GDRST,
1401 			      GRDOM_RENDER | GRDOM_RESET_ENABLE);
1402 	ret =  wait_for(g4x_reset_complete(dev), 500);
1403 	if (ret)
1404 		return ret;
1405 
1406 	/* WaVcpClkGateDisableForMediaReset:ctg,elk */
1407 	I915_WRITE(VDECCLK_GATE_D, I915_READ(VDECCLK_GATE_D) | VCP_UNIT_CLOCK_GATE_DISABLE);
1408 	POSTING_READ(VDECCLK_GATE_D);
1409 
1410 	pci_write_config_byte(dev->pdev, I915_GDRST,
1411 			      GRDOM_MEDIA | GRDOM_RESET_ENABLE);
1412 	ret =  wait_for(g4x_reset_complete(dev), 500);
1413 	if (ret)
1414 		return ret;
1415 
1416 	/* WaVcpClkGateDisableForMediaReset:ctg,elk */
1417 	I915_WRITE(VDECCLK_GATE_D, I915_READ(VDECCLK_GATE_D) & ~VCP_UNIT_CLOCK_GATE_DISABLE);
1418 	POSTING_READ(VDECCLK_GATE_D);
1419 
1420 	pci_write_config_byte(dev->pdev, I915_GDRST, 0);
1421 
1422 	return 0;
1423 }
1424 
1425 static int ironlake_do_reset(struct drm_device *dev)
1426 {
1427 	struct drm_i915_private *dev_priv = dev->dev_private;
1428 	int ret;
1429 
1430 	I915_WRITE(MCHBAR_MIRROR_BASE + ILK_GDSR,
1431 		   ILK_GRDOM_RENDER | ILK_GRDOM_RESET_ENABLE);
1432 	ret = wait_for((I915_READ(MCHBAR_MIRROR_BASE + ILK_GDSR) &
1433 			ILK_GRDOM_RESET_ENABLE) == 0, 500);
1434 	if (ret)
1435 		return ret;
1436 
1437 	I915_WRITE(MCHBAR_MIRROR_BASE + ILK_GDSR,
1438 		   ILK_GRDOM_MEDIA | ILK_GRDOM_RESET_ENABLE);
1439 	ret = wait_for((I915_READ(MCHBAR_MIRROR_BASE + ILK_GDSR) &
1440 			ILK_GRDOM_RESET_ENABLE) == 0, 500);
1441 	if (ret)
1442 		return ret;
1443 
1444 	I915_WRITE(MCHBAR_MIRROR_BASE + ILK_GDSR, 0);
1445 
1446 	return 0;
1447 }
1448 
1449 static int gen6_do_reset(struct drm_device *dev)
1450 {
1451 	struct drm_i915_private *dev_priv = dev->dev_private;
1452 	int	ret;
1453 
1454 	/* Reset the chip */
1455 
1456 	/* GEN6_GDRST is not in the gt power well, no need to check
1457 	 * for fifo space for the write or forcewake the chip for
1458 	 * the read
1459 	 */
1460 	__raw_i915_write32(dev_priv, GEN6_GDRST, GEN6_GRDOM_FULL);
1461 
1462 	/* Spin waiting for the device to ack the reset request */
1463 	ret = wait_for((__raw_i915_read32(dev_priv, GEN6_GDRST) & GEN6_GRDOM_FULL) == 0, 500);
1464 
1465 	intel_uncore_forcewake_reset(dev, true);
1466 
1467 	return ret;
1468 }
1469 
1470 static int wait_for_register(struct drm_i915_private *dev_priv,
1471 			     const u32 reg,
1472 			     const u32 mask,
1473 			     const u32 value,
1474 			     const unsigned long timeout_ms)
1475 {
1476 	return wait_for((I915_READ(reg) & mask) == value, timeout_ms);
1477 }
1478 
1479 static int gen8_do_reset(struct drm_device *dev)
1480 {
1481 	struct drm_i915_private *dev_priv = dev->dev_private;
1482 	struct intel_engine_cs *engine;
1483 	int i;
1484 
1485 	for_each_ring(engine, dev_priv, i) {
1486 		I915_WRITE(RING_RESET_CTL(engine->mmio_base),
1487 			   _MASKED_BIT_ENABLE(RESET_CTL_REQUEST_RESET));
1488 
1489 		if (wait_for_register(dev_priv,
1490 				      RING_RESET_CTL(engine->mmio_base),
1491 				      RESET_CTL_READY_TO_RESET,
1492 				      RESET_CTL_READY_TO_RESET,
1493 				      700)) {
1494 			DRM_ERROR("%s: reset request timeout\n", engine->name);
1495 			goto not_ready;
1496 		}
1497 	}
1498 
1499 	return gen6_do_reset(dev);
1500 
1501 not_ready:
1502 	for_each_ring(engine, dev_priv, i)
1503 		I915_WRITE(RING_RESET_CTL(engine->mmio_base),
1504 			   _MASKED_BIT_DISABLE(RESET_CTL_REQUEST_RESET));
1505 
1506 	return -EIO;
1507 }
1508 
1509 static int (*intel_get_gpu_reset(struct drm_device *dev))(struct drm_device *)
1510 {
1511 	if (!i915.reset)
1512 		return NULL;
1513 
1514 	if (INTEL_INFO(dev)->gen >= 8)
1515 		return gen8_do_reset;
1516 	else if (INTEL_INFO(dev)->gen >= 6)
1517 		return gen6_do_reset;
1518 	else if (IS_GEN5(dev))
1519 		return ironlake_do_reset;
1520 	else if (IS_G4X(dev))
1521 		return g4x_do_reset;
1522 	else if (IS_G33(dev))
1523 		return g33_do_reset;
1524 	else if (INTEL_INFO(dev)->gen >= 3)
1525 		return i915_do_reset;
1526 	else
1527 		return NULL;
1528 }
1529 
1530 int intel_gpu_reset(struct drm_device *dev)
1531 {
1532 	int (*reset)(struct drm_device *);
1533 
1534 	reset = intel_get_gpu_reset(dev);
1535 	if (reset == NULL)
1536 		return -ENODEV;
1537 
1538 	return reset(dev);
1539 }
1540 
1541 bool intel_has_gpu_reset(struct drm_device *dev)
1542 {
1543 	return intel_get_gpu_reset(dev) != NULL;
1544 }
1545 
1546 void intel_uncore_check_errors(struct drm_device *dev)
1547 {
1548 	struct drm_i915_private *dev_priv = dev->dev_private;
1549 
1550 	if (HAS_FPGA_DBG_UNCLAIMED(dev) &&
1551 	    (__raw_i915_read32(dev_priv, FPGA_DBG) & FPGA_DBG_RM_NOCLAIM)) {
1552 		DRM_ERROR("Unclaimed register before interrupt\n");
1553 		__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
1554 	}
1555 }
1556