xref: /openbmc/linux/drivers/gpu/drm/i915/gt/uc/intel_guc.c (revision baf2c002)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2014-2019 Intel Corporation
4  */
5 
6 #include "gem/i915_gem_lmem.h"
7 #include "gt/intel_gt.h"
8 #include "gt/intel_gt_irq.h"
9 #include "gt/intel_gt_pm_irq.h"
10 #include "gt/intel_gt_regs.h"
11 #include "intel_guc.h"
12 #include "intel_guc_ads.h"
13 #include "intel_guc_capture.h"
14 #include "intel_guc_slpc.h"
15 #include "intel_guc_submission.h"
16 #include "i915_drv.h"
17 #include "i915_irq.h"
18 
19 /**
20  * DOC: GuC
21  *
22  * The GuC is a microcontroller inside the GT HW, introduced in gen9. The GuC is
23  * designed to offload some of the functionality usually performed by the host
24  * driver; currently the main operations it can take care of are:
25  *
26  * - Authentication of the HuC, which is required to fully enable HuC usage.
27  * - Low latency graphics context scheduling (a.k.a. GuC submission).
28  * - GT Power management.
29  *
30  * The enable_guc module parameter can be used to select which of those
31  * operations to enable within GuC. Note that not all the operations are
32  * supported on all gen9+ platforms.
33  *
34  * Enabling the GuC is not mandatory and therefore the firmware is only loaded
35  * if at least one of the operations is selected. However, not loading the GuC
36  * might result in the loss of some features that do require the GuC (currently
37  * just the HuC, but more are expected to land in the future).
38  */
39 
40 void intel_guc_notify(struct intel_guc *guc)
41 {
42 	struct intel_gt *gt = guc_to_gt(guc);
43 
44 	/*
45 	 * On Gen11+, the value written to the register is passes as a payload
46 	 * to the FW. However, the FW currently treats all values the same way
47 	 * (H2G interrupt), so we can just write the value that the HW expects
48 	 * on older gens.
49 	 */
50 	intel_uncore_write(gt->uncore, guc->notify_reg, GUC_SEND_TRIGGER);
51 }
52 
53 static inline i915_reg_t guc_send_reg(struct intel_guc *guc, u32 i)
54 {
55 	GEM_BUG_ON(!guc->send_regs.base);
56 	GEM_BUG_ON(!guc->send_regs.count);
57 	GEM_BUG_ON(i >= guc->send_regs.count);
58 
59 	return _MMIO(guc->send_regs.base + 4 * i);
60 }
61 
62 void intel_guc_init_send_regs(struct intel_guc *guc)
63 {
64 	struct intel_gt *gt = guc_to_gt(guc);
65 	enum forcewake_domains fw_domains = 0;
66 	unsigned int i;
67 
68 	GEM_BUG_ON(!guc->send_regs.base);
69 	GEM_BUG_ON(!guc->send_regs.count);
70 
71 	for (i = 0; i < guc->send_regs.count; i++) {
72 		fw_domains |= intel_uncore_forcewake_for_reg(gt->uncore,
73 					guc_send_reg(guc, i),
74 					FW_REG_READ | FW_REG_WRITE);
75 	}
76 	guc->send_regs.fw_domains = fw_domains;
77 }
78 
79 static void gen9_reset_guc_interrupts(struct intel_guc *guc)
80 {
81 	struct intel_gt *gt = guc_to_gt(guc);
82 
83 	assert_rpm_wakelock_held(&gt->i915->runtime_pm);
84 
85 	spin_lock_irq(&gt->irq_lock);
86 	gen6_gt_pm_reset_iir(gt, gt->pm_guc_events);
87 	spin_unlock_irq(&gt->irq_lock);
88 }
89 
90 static void gen9_enable_guc_interrupts(struct intel_guc *guc)
91 {
92 	struct intel_gt *gt = guc_to_gt(guc);
93 
94 	assert_rpm_wakelock_held(&gt->i915->runtime_pm);
95 
96 	spin_lock_irq(&gt->irq_lock);
97 	WARN_ON_ONCE(intel_uncore_read(gt->uncore, GEN8_GT_IIR(2)) &
98 		     gt->pm_guc_events);
99 	gen6_gt_pm_enable_irq(gt, gt->pm_guc_events);
100 	spin_unlock_irq(&gt->irq_lock);
101 }
102 
103 static void gen9_disable_guc_interrupts(struct intel_guc *guc)
104 {
105 	struct intel_gt *gt = guc_to_gt(guc);
106 
107 	assert_rpm_wakelock_held(&gt->i915->runtime_pm);
108 
109 	spin_lock_irq(&gt->irq_lock);
110 
111 	gen6_gt_pm_disable_irq(gt, gt->pm_guc_events);
112 
113 	spin_unlock_irq(&gt->irq_lock);
114 	intel_synchronize_irq(gt->i915);
115 
116 	gen9_reset_guc_interrupts(guc);
117 }
118 
119 static void gen11_reset_guc_interrupts(struct intel_guc *guc)
120 {
121 	struct intel_gt *gt = guc_to_gt(guc);
122 
123 	spin_lock_irq(&gt->irq_lock);
124 	gen11_gt_reset_one_iir(gt, 0, GEN11_GUC);
125 	spin_unlock_irq(&gt->irq_lock);
126 }
127 
128 static void gen11_enable_guc_interrupts(struct intel_guc *guc)
129 {
130 	struct intel_gt *gt = guc_to_gt(guc);
131 	u32 events = REG_FIELD_PREP(ENGINE1_MASK, GUC_INTR_GUC2HOST);
132 
133 	spin_lock_irq(&gt->irq_lock);
134 	WARN_ON_ONCE(gen11_gt_reset_one_iir(gt, 0, GEN11_GUC));
135 	intel_uncore_write(gt->uncore,
136 			   GEN11_GUC_SG_INTR_ENABLE, events);
137 	intel_uncore_write(gt->uncore,
138 			   GEN11_GUC_SG_INTR_MASK, ~events);
139 	spin_unlock_irq(&gt->irq_lock);
140 }
141 
142 static void gen11_disable_guc_interrupts(struct intel_guc *guc)
143 {
144 	struct intel_gt *gt = guc_to_gt(guc);
145 
146 	spin_lock_irq(&gt->irq_lock);
147 
148 	intel_uncore_write(gt->uncore, GEN11_GUC_SG_INTR_MASK, ~0);
149 	intel_uncore_write(gt->uncore, GEN11_GUC_SG_INTR_ENABLE, 0);
150 
151 	spin_unlock_irq(&gt->irq_lock);
152 	intel_synchronize_irq(gt->i915);
153 
154 	gen11_reset_guc_interrupts(guc);
155 }
156 
157 void intel_guc_init_early(struct intel_guc *guc)
158 {
159 	struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
160 
161 	intel_uc_fw_init_early(&guc->fw, INTEL_UC_FW_TYPE_GUC);
162 	intel_guc_ct_init_early(&guc->ct);
163 	intel_guc_log_init_early(&guc->log);
164 	intel_guc_submission_init_early(guc);
165 	intel_guc_slpc_init_early(&guc->slpc);
166 	intel_guc_rc_init_early(guc);
167 
168 	mutex_init(&guc->send_mutex);
169 	spin_lock_init(&guc->irq_lock);
170 	if (GRAPHICS_VER(i915) >= 11) {
171 		guc->notify_reg = GEN11_GUC_HOST_INTERRUPT;
172 		guc->interrupts.reset = gen11_reset_guc_interrupts;
173 		guc->interrupts.enable = gen11_enable_guc_interrupts;
174 		guc->interrupts.disable = gen11_disable_guc_interrupts;
175 		guc->send_regs.base =
176 			i915_mmio_reg_offset(GEN11_SOFT_SCRATCH(0));
177 		guc->send_regs.count = GEN11_SOFT_SCRATCH_COUNT;
178 
179 	} else {
180 		guc->notify_reg = GUC_SEND_INTERRUPT;
181 		guc->interrupts.reset = gen9_reset_guc_interrupts;
182 		guc->interrupts.enable = gen9_enable_guc_interrupts;
183 		guc->interrupts.disable = gen9_disable_guc_interrupts;
184 		guc->send_regs.base = i915_mmio_reg_offset(SOFT_SCRATCH(0));
185 		guc->send_regs.count = GUC_MAX_MMIO_MSG_LEN;
186 		BUILD_BUG_ON(GUC_MAX_MMIO_MSG_LEN > SOFT_SCRATCH_COUNT);
187 	}
188 
189 	intel_guc_enable_msg(guc, INTEL_GUC_RECV_MSG_EXCEPTION |
190 				  INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED);
191 }
192 
193 void intel_guc_init_late(struct intel_guc *guc)
194 {
195 	intel_guc_ads_init_late(guc);
196 }
197 
198 static u32 guc_ctl_debug_flags(struct intel_guc *guc)
199 {
200 	u32 level = intel_guc_log_get_level(&guc->log);
201 	u32 flags = 0;
202 
203 	if (!GUC_LOG_LEVEL_IS_VERBOSE(level))
204 		flags |= GUC_LOG_DISABLED;
205 	else
206 		flags |= GUC_LOG_LEVEL_TO_VERBOSITY(level) <<
207 			 GUC_LOG_VERBOSITY_SHIFT;
208 
209 	return flags;
210 }
211 
212 static u32 guc_ctl_feature_flags(struct intel_guc *guc)
213 {
214 	u32 flags = 0;
215 
216 	if (!intel_guc_submission_is_used(guc))
217 		flags |= GUC_CTL_DISABLE_SCHEDULER;
218 
219 	if (intel_guc_slpc_is_used(guc))
220 		flags |= GUC_CTL_ENABLE_SLPC;
221 
222 	return flags;
223 }
224 
225 static u32 guc_ctl_log_params_flags(struct intel_guc *guc)
226 {
227 	u32 offset = intel_guc_ggtt_offset(guc, guc->log.vma) >> PAGE_SHIFT;
228 	u32 flags;
229 
230 	#if (((CRASH_BUFFER_SIZE) % SZ_1M) == 0)
231 	#define LOG_UNIT SZ_1M
232 	#define LOG_FLAG GUC_LOG_LOG_ALLOC_UNITS
233 	#else
234 	#define LOG_UNIT SZ_4K
235 	#define LOG_FLAG 0
236 	#endif
237 
238 	#if (((CAPTURE_BUFFER_SIZE) % SZ_1M) == 0)
239 	#define CAPTURE_UNIT SZ_1M
240 	#define CAPTURE_FLAG GUC_LOG_CAPTURE_ALLOC_UNITS
241 	#else
242 	#define CAPTURE_UNIT SZ_4K
243 	#define CAPTURE_FLAG 0
244 	#endif
245 
246 	BUILD_BUG_ON(!CRASH_BUFFER_SIZE);
247 	BUILD_BUG_ON(!IS_ALIGNED(CRASH_BUFFER_SIZE, LOG_UNIT));
248 	BUILD_BUG_ON(!DEBUG_BUFFER_SIZE);
249 	BUILD_BUG_ON(!IS_ALIGNED(DEBUG_BUFFER_SIZE, LOG_UNIT));
250 	BUILD_BUG_ON(!CAPTURE_BUFFER_SIZE);
251 	BUILD_BUG_ON(!IS_ALIGNED(CAPTURE_BUFFER_SIZE, CAPTURE_UNIT));
252 
253 	BUILD_BUG_ON((CRASH_BUFFER_SIZE / LOG_UNIT - 1) >
254 			(GUC_LOG_CRASH_MASK >> GUC_LOG_CRASH_SHIFT));
255 	BUILD_BUG_ON((DEBUG_BUFFER_SIZE / LOG_UNIT - 1) >
256 			(GUC_LOG_DEBUG_MASK >> GUC_LOG_DEBUG_SHIFT));
257 	BUILD_BUG_ON((CAPTURE_BUFFER_SIZE / CAPTURE_UNIT - 1) >
258 			(GUC_LOG_CAPTURE_MASK >> GUC_LOG_CAPTURE_SHIFT));
259 
260 	flags = GUC_LOG_VALID |
261 		GUC_LOG_NOTIFY_ON_HALF_FULL |
262 		CAPTURE_FLAG |
263 		LOG_FLAG |
264 		((CRASH_BUFFER_SIZE / LOG_UNIT - 1) << GUC_LOG_CRASH_SHIFT) |
265 		((DEBUG_BUFFER_SIZE / LOG_UNIT - 1) << GUC_LOG_DEBUG_SHIFT) |
266 		((CAPTURE_BUFFER_SIZE / CAPTURE_UNIT - 1) << GUC_LOG_CAPTURE_SHIFT) |
267 		(offset << GUC_LOG_BUF_ADDR_SHIFT);
268 
269 	#undef LOG_UNIT
270 	#undef LOG_FLAG
271 	#undef CAPTURE_UNIT
272 	#undef CAPTURE_FLAG
273 
274 	return flags;
275 }
276 
277 static u32 guc_ctl_ads_flags(struct intel_guc *guc)
278 {
279 	u32 ads = intel_guc_ggtt_offset(guc, guc->ads_vma) >> PAGE_SHIFT;
280 	u32 flags = ads << GUC_ADS_ADDR_SHIFT;
281 
282 	return flags;
283 }
284 
285 static u32 guc_ctl_wa_flags(struct intel_guc *guc)
286 {
287 	struct intel_gt *gt = guc_to_gt(guc);
288 	u32 flags = 0;
289 
290 	/* Wa_22012773006:gen11,gen12 < XeHP */
291 	if (GRAPHICS_VER(gt->i915) >= 11 &&
292 	    GRAPHICS_VER_FULL(gt->i915) < IP_VER(12, 50))
293 		flags |= GUC_WA_POLLCS;
294 
295 	/* Wa_16011759253:dg2_g10:a0 */
296 	if (IS_DG2_GRAPHICS_STEP(gt->i915, G10, STEP_A0, STEP_B0))
297 		flags |= GUC_WA_GAM_CREDITS;
298 
299 	/* Wa_14014475959:dg2 */
300 	if (IS_DG2(gt->i915))
301 		flags |= GUC_WA_HOLD_CCS_SWITCHOUT;
302 
303 	/*
304 	 * Wa_14012197797:dg2_g10:a0,dg2_g11:a0
305 	 * Wa_22011391025:dg2_g10,dg2_g11,dg2_g12
306 	 *
307 	 * The same WA bit is used for both and 22011391025 is applicable to
308 	 * all DG2.
309 	 */
310 	if (IS_DG2(gt->i915))
311 		flags |= GUC_WA_DUAL_QUEUE;
312 
313 	/* Wa_22011802037: graphics version 11/12 */
314 	if (IS_GRAPHICS_VER(gt->i915, 11, 12))
315 		flags |= GUC_WA_PRE_PARSER;
316 
317 	/* Wa_16011777198:dg2 */
318 	if (IS_DG2_GRAPHICS_STEP(gt->i915, G10, STEP_A0, STEP_C0) ||
319 	    IS_DG2_GRAPHICS_STEP(gt->i915, G11, STEP_A0, STEP_B0))
320 		flags |= GUC_WA_RCS_RESET_BEFORE_RC6;
321 
322 	/*
323 	 * Wa_22012727170:dg2_g10[a0-c0), dg2_g11[a0..)
324 	 * Wa_22012727685:dg2_g11[a0..)
325 	 */
326 	if (IS_DG2_GRAPHICS_STEP(gt->i915, G10, STEP_A0, STEP_C0) ||
327 	    IS_DG2_GRAPHICS_STEP(gt->i915, G11, STEP_A0, STEP_FOREVER))
328 		flags |= GUC_WA_CONTEXT_ISOLATION;
329 
330 	/* Wa_16015675438 */
331 	if (!RCS_MASK(gt))
332 		flags |= GUC_WA_RCS_REGS_IN_CCS_REGS_LIST;
333 
334 	return flags;
335 }
336 
337 static u32 guc_ctl_devid(struct intel_guc *guc)
338 {
339 	struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
340 
341 	return (INTEL_DEVID(i915) << 16) | INTEL_REVID(i915);
342 }
343 
344 /*
345  * Initialise the GuC parameter block before starting the firmware
346  * transfer. These parameters are read by the firmware on startup
347  * and cannot be changed thereafter.
348  */
349 static void guc_init_params(struct intel_guc *guc)
350 {
351 	u32 *params = guc->params;
352 	int i;
353 
354 	BUILD_BUG_ON(sizeof(guc->params) != GUC_CTL_MAX_DWORDS * sizeof(u32));
355 
356 	params[GUC_CTL_LOG_PARAMS] = guc_ctl_log_params_flags(guc);
357 	params[GUC_CTL_FEATURE] = guc_ctl_feature_flags(guc);
358 	params[GUC_CTL_DEBUG] = guc_ctl_debug_flags(guc);
359 	params[GUC_CTL_ADS] = guc_ctl_ads_flags(guc);
360 	params[GUC_CTL_WA] = guc_ctl_wa_flags(guc);
361 	params[GUC_CTL_DEVID] = guc_ctl_devid(guc);
362 
363 	for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
364 		DRM_DEBUG_DRIVER("param[%2d] = %#x\n", i, params[i]);
365 }
366 
367 /*
368  * Initialise the GuC parameter block before starting the firmware
369  * transfer. These parameters are read by the firmware on startup
370  * and cannot be changed thereafter.
371  */
372 void intel_guc_write_params(struct intel_guc *guc)
373 {
374 	struct intel_uncore *uncore = guc_to_gt(guc)->uncore;
375 	int i;
376 
377 	/*
378 	 * All SOFT_SCRATCH registers are in FORCEWAKE_GT domain and
379 	 * they are power context saved so it's ok to release forcewake
380 	 * when we are done here and take it again at xfer time.
381 	 */
382 	intel_uncore_forcewake_get(uncore, FORCEWAKE_GT);
383 
384 	intel_uncore_write(uncore, SOFT_SCRATCH(0), 0);
385 
386 	for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
387 		intel_uncore_write(uncore, SOFT_SCRATCH(1 + i), guc->params[i]);
388 
389 	intel_uncore_forcewake_put(uncore, FORCEWAKE_GT);
390 }
391 
392 int intel_guc_init(struct intel_guc *guc)
393 {
394 	struct intel_gt *gt = guc_to_gt(guc);
395 	int ret;
396 
397 	ret = intel_uc_fw_init(&guc->fw);
398 	if (ret)
399 		goto out;
400 
401 	ret = intel_guc_log_create(&guc->log);
402 	if (ret)
403 		goto err_fw;
404 
405 	ret = intel_guc_capture_init(guc);
406 	if (ret)
407 		goto err_log;
408 
409 	ret = intel_guc_ads_create(guc);
410 	if (ret)
411 		goto err_capture;
412 
413 	GEM_BUG_ON(!guc->ads_vma);
414 
415 	ret = intel_guc_ct_init(&guc->ct);
416 	if (ret)
417 		goto err_ads;
418 
419 	if (intel_guc_submission_is_used(guc)) {
420 		/*
421 		 * This is stuff we need to have available at fw load time
422 		 * if we are planning to enable submission later
423 		 */
424 		ret = intel_guc_submission_init(guc);
425 		if (ret)
426 			goto err_ct;
427 	}
428 
429 	if (intel_guc_slpc_is_used(guc)) {
430 		ret = intel_guc_slpc_init(&guc->slpc);
431 		if (ret)
432 			goto err_submission;
433 	}
434 
435 	/* now that everything is perma-pinned, initialize the parameters */
436 	guc_init_params(guc);
437 
438 	/* We need to notify the guc whenever we change the GGTT */
439 	i915_ggtt_enable_guc(gt->ggtt);
440 
441 	intel_uc_fw_change_status(&guc->fw, INTEL_UC_FIRMWARE_LOADABLE);
442 
443 	return 0;
444 
445 err_submission:
446 	intel_guc_submission_fini(guc);
447 err_ct:
448 	intel_guc_ct_fini(&guc->ct);
449 err_ads:
450 	intel_guc_ads_destroy(guc);
451 err_capture:
452 	intel_guc_capture_destroy(guc);
453 err_log:
454 	intel_guc_log_destroy(&guc->log);
455 err_fw:
456 	intel_uc_fw_fini(&guc->fw);
457 out:
458 	i915_probe_error(gt->i915, "failed with %d\n", ret);
459 	return ret;
460 }
461 
462 void intel_guc_fini(struct intel_guc *guc)
463 {
464 	struct intel_gt *gt = guc_to_gt(guc);
465 
466 	if (!intel_uc_fw_is_loadable(&guc->fw))
467 		return;
468 
469 	i915_ggtt_disable_guc(gt->ggtt);
470 
471 	if (intel_guc_slpc_is_used(guc))
472 		intel_guc_slpc_fini(&guc->slpc);
473 
474 	if (intel_guc_submission_is_used(guc))
475 		intel_guc_submission_fini(guc);
476 
477 	intel_guc_ct_fini(&guc->ct);
478 
479 	intel_guc_ads_destroy(guc);
480 	intel_guc_capture_destroy(guc);
481 	intel_guc_log_destroy(&guc->log);
482 	intel_uc_fw_fini(&guc->fw);
483 }
484 
485 /*
486  * This function implements the MMIO based host to GuC interface.
487  */
488 int intel_guc_send_mmio(struct intel_guc *guc, const u32 *request, u32 len,
489 			u32 *response_buf, u32 response_buf_size)
490 {
491 	struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
492 	struct intel_uncore *uncore = guc_to_gt(guc)->uncore;
493 	u32 header;
494 	int i;
495 	int ret;
496 
497 	GEM_BUG_ON(!len);
498 	GEM_BUG_ON(len > guc->send_regs.count);
499 
500 	GEM_BUG_ON(FIELD_GET(GUC_HXG_MSG_0_ORIGIN, request[0]) != GUC_HXG_ORIGIN_HOST);
501 	GEM_BUG_ON(FIELD_GET(GUC_HXG_MSG_0_TYPE, request[0]) != GUC_HXG_TYPE_REQUEST);
502 
503 	mutex_lock(&guc->send_mutex);
504 	intel_uncore_forcewake_get(uncore, guc->send_regs.fw_domains);
505 
506 retry:
507 	for (i = 0; i < len; i++)
508 		intel_uncore_write(uncore, guc_send_reg(guc, i), request[i]);
509 
510 	intel_uncore_posting_read(uncore, guc_send_reg(guc, i - 1));
511 
512 	intel_guc_notify(guc);
513 
514 	/*
515 	 * No GuC command should ever take longer than 10ms.
516 	 * Fast commands should still complete in 10us.
517 	 */
518 	ret = __intel_wait_for_register_fw(uncore,
519 					   guc_send_reg(guc, 0),
520 					   GUC_HXG_MSG_0_ORIGIN,
521 					   FIELD_PREP(GUC_HXG_MSG_0_ORIGIN,
522 						      GUC_HXG_ORIGIN_GUC),
523 					   10, 10, &header);
524 	if (unlikely(ret)) {
525 timeout:
526 		drm_err(&i915->drm, "mmio request %#x: no reply %x\n",
527 			request[0], header);
528 		goto out;
529 	}
530 
531 	if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) == GUC_HXG_TYPE_NO_RESPONSE_BUSY) {
532 #define done ({ header = intel_uncore_read(uncore, guc_send_reg(guc, 0)); \
533 		FIELD_GET(GUC_HXG_MSG_0_ORIGIN, header) != GUC_HXG_ORIGIN_GUC || \
534 		FIELD_GET(GUC_HXG_MSG_0_TYPE, header) != GUC_HXG_TYPE_NO_RESPONSE_BUSY; })
535 
536 		ret = wait_for(done, 1000);
537 		if (unlikely(ret))
538 			goto timeout;
539 		if (unlikely(FIELD_GET(GUC_HXG_MSG_0_ORIGIN, header) !=
540 				       GUC_HXG_ORIGIN_GUC))
541 			goto proto;
542 #undef done
543 	}
544 
545 	if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) == GUC_HXG_TYPE_NO_RESPONSE_RETRY) {
546 		u32 reason = FIELD_GET(GUC_HXG_RETRY_MSG_0_REASON, header);
547 
548 		drm_dbg(&i915->drm, "mmio request %#x: retrying, reason %u\n",
549 			request[0], reason);
550 		goto retry;
551 	}
552 
553 	if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) == GUC_HXG_TYPE_RESPONSE_FAILURE) {
554 		u32 hint = FIELD_GET(GUC_HXG_FAILURE_MSG_0_HINT, header);
555 		u32 error = FIELD_GET(GUC_HXG_FAILURE_MSG_0_ERROR, header);
556 
557 		drm_err(&i915->drm, "mmio request %#x: failure %x/%u\n",
558 			request[0], error, hint);
559 		ret = -ENXIO;
560 		goto out;
561 	}
562 
563 	if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) != GUC_HXG_TYPE_RESPONSE_SUCCESS) {
564 proto:
565 		drm_err(&i915->drm, "mmio request %#x: unexpected reply %#x\n",
566 			request[0], header);
567 		ret = -EPROTO;
568 		goto out;
569 	}
570 
571 	if (response_buf) {
572 		int count = min(response_buf_size, guc->send_regs.count);
573 
574 		GEM_BUG_ON(!count);
575 
576 		response_buf[0] = header;
577 
578 		for (i = 1; i < count; i++)
579 			response_buf[i] = intel_uncore_read(uncore,
580 							    guc_send_reg(guc, i));
581 
582 		/* Use number of copied dwords as our return value */
583 		ret = count;
584 	} else {
585 		/* Use data from the GuC response as our return value */
586 		ret = FIELD_GET(GUC_HXG_RESPONSE_MSG_0_DATA0, header);
587 	}
588 
589 out:
590 	intel_uncore_forcewake_put(uncore, guc->send_regs.fw_domains);
591 	mutex_unlock(&guc->send_mutex);
592 
593 	return ret;
594 }
595 
596 int intel_guc_to_host_process_recv_msg(struct intel_guc *guc,
597 				       const u32 *payload, u32 len)
598 {
599 	u32 msg;
600 
601 	if (unlikely(!len))
602 		return -EPROTO;
603 
604 	/* Make sure to handle only enabled messages */
605 	msg = payload[0] & guc->msg_enabled_mask;
606 
607 	if (msg & INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED)
608 		drm_err(&guc_to_gt(guc)->i915->drm, "Received early GuC crash dump notification!\n");
609 	if (msg & INTEL_GUC_RECV_MSG_EXCEPTION)
610 		drm_err(&guc_to_gt(guc)->i915->drm, "Received early GuC exception notification!\n");
611 
612 	return 0;
613 }
614 
615 /**
616  * intel_guc_auth_huc() - Send action to GuC to authenticate HuC ucode
617  * @guc: intel_guc structure
618  * @rsa_offset: rsa offset w.r.t ggtt base of huc vma
619  *
620  * Triggers a HuC firmware authentication request to the GuC via intel_guc_send
621  * INTEL_GUC_ACTION_AUTHENTICATE_HUC interface. This function is invoked by
622  * intel_huc_auth().
623  *
624  * Return:	non-zero code on error
625  */
626 int intel_guc_auth_huc(struct intel_guc *guc, u32 rsa_offset)
627 {
628 	u32 action[] = {
629 		INTEL_GUC_ACTION_AUTHENTICATE_HUC,
630 		rsa_offset
631 	};
632 
633 	return intel_guc_send(guc, action, ARRAY_SIZE(action));
634 }
635 
636 /**
637  * intel_guc_suspend() - notify GuC entering suspend state
638  * @guc:	the guc
639  */
640 int intel_guc_suspend(struct intel_guc *guc)
641 {
642 	int ret;
643 	u32 action[] = {
644 		INTEL_GUC_ACTION_CLIENT_SOFT_RESET,
645 	};
646 
647 	if (!intel_guc_is_ready(guc))
648 		return 0;
649 
650 	if (intel_guc_submission_is_used(guc)) {
651 		/*
652 		 * This H2G MMIO command tears down the GuC in two steps. First it will
653 		 * generate a G2H CTB for every active context indicating a reset. In
654 		 * practice the i915 shouldn't ever get a G2H as suspend should only be
655 		 * called when the GPU is idle. Next, it tears down the CTBs and this
656 		 * H2G MMIO command completes.
657 		 *
658 		 * Don't abort on a failure code from the GuC. Keep going and do the
659 		 * clean up in santize() and re-initialisation on resume and hopefully
660 		 * the error here won't be problematic.
661 		 */
662 		ret = intel_guc_send_mmio(guc, action, ARRAY_SIZE(action), NULL, 0);
663 		if (ret)
664 			DRM_ERROR("GuC suspend: RESET_CLIENT action failed with error %d!\n", ret);
665 	}
666 
667 	/* Signal that the GuC isn't running. */
668 	intel_guc_sanitize(guc);
669 
670 	return 0;
671 }
672 
673 /**
674  * intel_guc_resume() - notify GuC resuming from suspend state
675  * @guc:	the guc
676  */
677 int intel_guc_resume(struct intel_guc *guc)
678 {
679 	/*
680 	 * NB: This function can still be called even if GuC submission is
681 	 * disabled, e.g. if GuC is enabled for HuC authentication only. Thus,
682 	 * if any code is later added here, it must be support doing nothing
683 	 * if submission is disabled (as per intel_guc_suspend).
684 	 */
685 	return 0;
686 }
687 
688 /**
689  * DOC: GuC Memory Management
690  *
691  * GuC can't allocate any memory for its own usage, so all the allocations must
692  * be handled by the host driver. GuC accesses the memory via the GGTT, with the
693  * exception of the top and bottom parts of the 4GB address space, which are
694  * instead re-mapped by the GuC HW to memory location of the FW itself (WOPCM)
695  * or other parts of the HW. The driver must take care not to place objects that
696  * the GuC is going to access in these reserved ranges. The layout of the GuC
697  * address space is shown below:
698  *
699  * ::
700  *
701  *     +===========> +====================+ <== FFFF_FFFF
702  *     ^             |      Reserved      |
703  *     |             +====================+ <== GUC_GGTT_TOP
704  *     |             |                    |
705  *     |             |        DRAM        |
706  *    GuC            |                    |
707  *  Address    +===> +====================+ <== GuC ggtt_pin_bias
708  *   Space     ^     |                    |
709  *     |       |     |                    |
710  *     |      GuC    |        GuC         |
711  *     |     WOPCM   |       WOPCM        |
712  *     |      Size   |                    |
713  *     |       |     |                    |
714  *     v       v     |                    |
715  *     +=======+===> +====================+ <== 0000_0000
716  *
717  * The lower part of GuC Address Space [0, ggtt_pin_bias) is mapped to GuC WOPCM
718  * while upper part of GuC Address Space [ggtt_pin_bias, GUC_GGTT_TOP) is mapped
719  * to DRAM. The value of the GuC ggtt_pin_bias is the GuC WOPCM size.
720  */
721 
722 /**
723  * intel_guc_allocate_vma() - Allocate a GGTT VMA for GuC usage
724  * @guc:	the guc
725  * @size:	size of area to allocate (both virtual space and memory)
726  *
727  * This is a wrapper to create an object for use with the GuC. In order to
728  * use it inside the GuC, an object needs to be pinned lifetime, so we allocate
729  * both some backing storage and a range inside the Global GTT. We must pin
730  * it in the GGTT somewhere other than than [0, GUC ggtt_pin_bias) because that
731  * range is reserved inside GuC.
732  *
733  * Return:	A i915_vma if successful, otherwise an ERR_PTR.
734  */
735 struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size)
736 {
737 	struct intel_gt *gt = guc_to_gt(guc);
738 	struct drm_i915_gem_object *obj;
739 	struct i915_vma *vma;
740 	u64 flags;
741 	int ret;
742 
743 	if (HAS_LMEM(gt->i915))
744 		obj = i915_gem_object_create_lmem(gt->i915, size,
745 						  I915_BO_ALLOC_CPU_CLEAR |
746 						  I915_BO_ALLOC_CONTIGUOUS |
747 						  I915_BO_ALLOC_PM_EARLY);
748 	else
749 		obj = i915_gem_object_create_shmem(gt->i915, size);
750 
751 	if (IS_ERR(obj))
752 		return ERR_CAST(obj);
753 
754 	vma = i915_vma_instance(obj, &gt->ggtt->vm, NULL);
755 	if (IS_ERR(vma))
756 		goto err;
757 
758 	flags = PIN_OFFSET_BIAS | i915_ggtt_pin_bias(vma);
759 	ret = i915_ggtt_pin(vma, NULL, 0, flags);
760 	if (ret) {
761 		vma = ERR_PTR(ret);
762 		goto err;
763 	}
764 
765 	return i915_vma_make_unshrinkable(vma);
766 
767 err:
768 	i915_gem_object_put(obj);
769 	return vma;
770 }
771 
772 /**
773  * intel_guc_allocate_and_map_vma() - Allocate and map VMA for GuC usage
774  * @guc:	the guc
775  * @size:	size of area to allocate (both virtual space and memory)
776  * @out_vma:	return variable for the allocated vma pointer
777  * @out_vaddr:	return variable for the obj mapping
778  *
779  * This wrapper calls intel_guc_allocate_vma() and then maps the allocated
780  * object with I915_MAP_WB.
781  *
782  * Return:	0 if successful, a negative errno code otherwise.
783  */
784 int intel_guc_allocate_and_map_vma(struct intel_guc *guc, u32 size,
785 				   struct i915_vma **out_vma, void **out_vaddr)
786 {
787 	struct i915_vma *vma;
788 	void *vaddr;
789 
790 	vma = intel_guc_allocate_vma(guc, size);
791 	if (IS_ERR(vma))
792 		return PTR_ERR(vma);
793 
794 	vaddr = i915_gem_object_pin_map_unlocked(vma->obj,
795 						 i915_coherent_map_type(guc_to_gt(guc)->i915,
796 									vma->obj, true));
797 	if (IS_ERR(vaddr)) {
798 		i915_vma_unpin_and_release(&vma, 0);
799 		return PTR_ERR(vaddr);
800 	}
801 
802 	*out_vma = vma;
803 	*out_vaddr = vaddr;
804 
805 	return 0;
806 }
807 
808 static int __guc_action_self_cfg(struct intel_guc *guc, u16 key, u16 len, u64 value)
809 {
810 	u32 request[HOST2GUC_SELF_CFG_REQUEST_MSG_LEN] = {
811 		FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
812 		FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
813 		FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION, GUC_ACTION_HOST2GUC_SELF_CFG),
814 		FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_KEY, key) |
815 		FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_LEN, len),
816 		FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_2_VALUE32, lower_32_bits(value)),
817 		FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_3_VALUE64, upper_32_bits(value)),
818 	};
819 	int ret;
820 
821 	GEM_BUG_ON(len > 2);
822 	GEM_BUG_ON(len == 1 && upper_32_bits(value));
823 
824 	/* Self config must go over MMIO */
825 	ret = intel_guc_send_mmio(guc, request, ARRAY_SIZE(request), NULL, 0);
826 
827 	if (unlikely(ret < 0))
828 		return ret;
829 	if (unlikely(ret > 1))
830 		return -EPROTO;
831 	if (unlikely(!ret))
832 		return -ENOKEY;
833 
834 	return 0;
835 }
836 
837 static int __guc_self_cfg(struct intel_guc *guc, u16 key, u16 len, u64 value)
838 {
839 	struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
840 	int err = __guc_action_self_cfg(guc, key, len, value);
841 
842 	if (unlikely(err))
843 		i915_probe_error(i915, "Unsuccessful self-config (%pe) key %#hx value %#llx\n",
844 				 ERR_PTR(err), key, value);
845 	return err;
846 }
847 
848 int intel_guc_self_cfg32(struct intel_guc *guc, u16 key, u32 value)
849 {
850 	return __guc_self_cfg(guc, key, 1, value);
851 }
852 
853 int intel_guc_self_cfg64(struct intel_guc *guc, u16 key, u64 value)
854 {
855 	return __guc_self_cfg(guc, key, 2, value);
856 }
857 
858 /**
859  * intel_guc_load_status - dump information about GuC load status
860  * @guc: the GuC
861  * @p: the &drm_printer
862  *
863  * Pretty printer for GuC load status.
864  */
865 void intel_guc_load_status(struct intel_guc *guc, struct drm_printer *p)
866 {
867 	struct intel_gt *gt = guc_to_gt(guc);
868 	struct intel_uncore *uncore = gt->uncore;
869 	intel_wakeref_t wakeref;
870 
871 	if (!intel_guc_is_supported(guc)) {
872 		drm_printf(p, "GuC not supported\n");
873 		return;
874 	}
875 
876 	if (!intel_guc_is_wanted(guc)) {
877 		drm_printf(p, "GuC disabled\n");
878 		return;
879 	}
880 
881 	intel_uc_fw_dump(&guc->fw, p);
882 
883 	with_intel_runtime_pm(uncore->rpm, wakeref) {
884 		u32 status = intel_uncore_read(uncore, GUC_STATUS);
885 		u32 i;
886 
887 		drm_printf(p, "\nGuC status 0x%08x:\n", status);
888 		drm_printf(p, "\tBootrom status = 0x%x\n",
889 			   (status & GS_BOOTROM_MASK) >> GS_BOOTROM_SHIFT);
890 		drm_printf(p, "\tuKernel status = 0x%x\n",
891 			   (status & GS_UKERNEL_MASK) >> GS_UKERNEL_SHIFT);
892 		drm_printf(p, "\tMIA Core status = 0x%x\n",
893 			   (status & GS_MIA_MASK) >> GS_MIA_SHIFT);
894 		drm_puts(p, "\nScratch registers:\n");
895 		for (i = 0; i < 16; i++) {
896 			drm_printf(p, "\t%2d: \t0x%x\n",
897 				   i, intel_uncore_read(uncore, SOFT_SCRATCH(i)));
898 		}
899 	}
900 }
901 
902 void intel_guc_write_barrier(struct intel_guc *guc)
903 {
904 	struct intel_gt *gt = guc_to_gt(guc);
905 
906 	if (i915_gem_object_is_lmem(guc->ct.vma->obj)) {
907 		/*
908 		 * Ensure intel_uncore_write_fw can be used rather than
909 		 * intel_uncore_write.
910 		 */
911 		GEM_BUG_ON(guc->send_regs.fw_domains);
912 
913 		/*
914 		 * This register is used by the i915 and GuC for MMIO based
915 		 * communication. Once we are in this code CTBs are the only
916 		 * method the i915 uses to communicate with the GuC so it is
917 		 * safe to write to this register (a value of 0 is NOP for MMIO
918 		 * communication). If we ever start mixing CTBs and MMIOs a new
919 		 * register will have to be chosen. This function is also used
920 		 * to enforce ordering of a work queue item write and an update
921 		 * to the process descriptor. When a work queue is being used,
922 		 * CTBs are also the only mechanism of communication.
923 		 */
924 		intel_uncore_write_fw(gt->uncore, GEN11_SOFT_SCRATCH(0), 0);
925 	} else {
926 		/* wmb() sufficient for a barrier if in smem */
927 		wmb();
928 	}
929 }
930