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
12#include "i915_drv.h"
13#include "i915_irq.h"
14#include "i915_reg.h"
15#include "i915_wait_util.h"
16#include "intel_guc.h"
17#include "intel_guc_ads.h"
18#include "intel_guc_capture.h"
19#include "intel_guc_print.h"
20#include "intel_guc_slpc.h"
21#include "intel_guc_submission.h"
22
23/**
24 * DOC: GuC
25 *
26 * The GuC is a microcontroller inside the GT HW, introduced in gen9. The GuC is
27 * designed to offload some of the functionality usually performed by the host
28 * driver; currently the main operations it can take care of are:
29 *
30 * - Authentication of the HuC, which is required to fully enable HuC usage.
31 * - Low latency graphics context scheduling (a.k.a. GuC submission).
32 * - GT Power management.
33 *
34 * The enable_guc module parameter can be used to select which of those
35 * operations to enable within GuC. Note that not all the operations are
36 * supported on all gen9+ platforms.
37 *
38 * Enabling the GuC is not mandatory and therefore the firmware is only loaded
39 * if at least one of the operations is selected. However, not loading the GuC
40 * might result in the loss of some features that do require the GuC (currently
41 * just the HuC, but more are expected to land in the future).
42 */
43
44void intel_guc_notify(struct intel_guc *guc)
45{
46 struct intel_gt *gt = guc_to_gt(guc);
47
48 /*
49 * On Gen11+, the value written to the register is passes as a payload
50 * to the FW. However, the FW currently treats all values the same way
51 * (H2G interrupt), so we can just write the value that the HW expects
52 * on older gens.
53 */
54 intel_uncore_write(uncore: gt->uncore, reg: guc->notify_reg, GUC_SEND_TRIGGER);
55}
56
57static inline i915_reg_t guc_send_reg(struct intel_guc *guc, u32 i)
58{
59 GEM_BUG_ON(!guc->send_regs.base);
60 GEM_BUG_ON(!guc->send_regs.count);
61 GEM_BUG_ON(i >= guc->send_regs.count);
62
63 return _MMIO(guc->send_regs.base + 4 * i);
64}
65
66void intel_guc_init_send_regs(struct intel_guc *guc)
67{
68 struct intel_gt *gt = guc_to_gt(guc);
69 enum forcewake_domains fw_domains = 0;
70 unsigned int i;
71
72 GEM_BUG_ON(!guc->send_regs.base);
73 GEM_BUG_ON(!guc->send_regs.count);
74
75 for (i = 0; i < guc->send_regs.count; i++) {
76 fw_domains |= intel_uncore_forcewake_for_reg(uncore: gt->uncore,
77 reg: guc_send_reg(guc, i),
78 FW_REG_READ | FW_REG_WRITE);
79 }
80 guc->send_regs.fw_domains = fw_domains;
81}
82
83static void gen9_reset_guc_interrupts(struct intel_guc *guc)
84{
85 struct intel_gt *gt = guc_to_gt(guc);
86
87 assert_rpm_wakelock_held(rpm: &gt->i915->runtime_pm);
88
89 spin_lock_irq(lock: gt->irq_lock);
90 gen6_gt_pm_reset_iir(gt, reset_mask: gt->pm_guc_events);
91 spin_unlock_irq(lock: gt->irq_lock);
92}
93
94static void gen9_enable_guc_interrupts(struct intel_guc *guc)
95{
96 struct intel_gt *gt = guc_to_gt(guc);
97
98 assert_rpm_wakelock_held(rpm: &gt->i915->runtime_pm);
99
100 spin_lock_irq(lock: gt->irq_lock);
101 guc_WARN_ON_ONCE(guc, intel_uncore_read(gt->uncore, GEN8_GT_IIR(2)) &
102 gt->pm_guc_events);
103 gen6_gt_pm_enable_irq(gt, enable_mask: gt->pm_guc_events);
104 spin_unlock_irq(lock: gt->irq_lock);
105
106 guc->interrupts.enabled = true;
107}
108
109static void gen9_disable_guc_interrupts(struct intel_guc *guc)
110{
111 struct intel_gt *gt = guc_to_gt(guc);
112
113 assert_rpm_wakelock_held(rpm: &gt->i915->runtime_pm);
114 guc->interrupts.enabled = false;
115
116 spin_lock_irq(lock: gt->irq_lock);
117
118 gen6_gt_pm_disable_irq(gt, disable_mask: gt->pm_guc_events);
119
120 spin_unlock_irq(lock: gt->irq_lock);
121 intel_synchronize_irq(i915: gt->i915);
122
123 gen9_reset_guc_interrupts(guc);
124}
125
126static bool __gen11_reset_guc_interrupts(struct intel_gt *gt)
127{
128 u32 irq = gt->type == GT_MEDIA ? MTL_MGUC : GEN11_GUC;
129
130 lockdep_assert_held(gt->irq_lock);
131 return gen11_gt_reset_one_iir(gt, bank: 0, bit: irq);
132}
133
134static void gen11_reset_guc_interrupts(struct intel_guc *guc)
135{
136 struct intel_gt *gt = guc_to_gt(guc);
137
138 spin_lock_irq(lock: gt->irq_lock);
139 __gen11_reset_guc_interrupts(gt);
140 spin_unlock_irq(lock: gt->irq_lock);
141}
142
143static void gen11_enable_guc_interrupts(struct intel_guc *guc)
144{
145 struct intel_gt *gt = guc_to_gt(guc);
146
147 spin_lock_irq(lock: gt->irq_lock);
148 __gen11_reset_guc_interrupts(gt);
149 spin_unlock_irq(lock: gt->irq_lock);
150
151 guc->interrupts.enabled = true;
152}
153
154static void gen11_disable_guc_interrupts(struct intel_guc *guc)
155{
156 struct intel_gt *gt = guc_to_gt(guc);
157
158 guc->interrupts.enabled = false;
159 intel_synchronize_irq(i915: gt->i915);
160
161 gen11_reset_guc_interrupts(guc);
162}
163
164static void guc_dead_worker_func(struct work_struct *w)
165{
166 struct intel_guc *guc = container_of(w, struct intel_guc, dead_guc_worker);
167 struct intel_gt *gt = guc_to_gt(guc);
168 unsigned long last = guc->last_dead_guc_jiffies;
169 unsigned long delta = jiffies_to_msecs(j: jiffies - last);
170
171 if (delta < 500) {
172 intel_gt_set_wedged(gt);
173 } else {
174 intel_gt_handle_error(gt, ALL_ENGINES, I915_ERROR_CAPTURE, fmt: "dead GuC");
175 guc->last_dead_guc_jiffies = jiffies;
176 }
177}
178
179void intel_guc_init_early(struct intel_guc *guc)
180{
181 struct intel_gt *gt = guc_to_gt(guc);
182 struct drm_i915_private *i915 = gt->i915;
183
184 intel_uc_fw_init_early(uc_fw: &guc->fw, type: INTEL_UC_FW_TYPE_GUC, needs_ggtt_mapping: true);
185 intel_guc_ct_init_early(ct: &guc->ct);
186 intel_guc_log_init_early(log: &guc->log);
187 intel_guc_submission_init_early(guc);
188 intel_guc_slpc_init_early(slpc: &guc->slpc);
189 intel_guc_rc_init_early(guc);
190
191 INIT_WORK(&guc->dead_guc_worker, guc_dead_worker_func);
192
193 mutex_init(&guc->send_mutex);
194 spin_lock_init(&guc->irq_lock);
195 if (GRAPHICS_VER(i915) >= 11) {
196 guc->interrupts.reset = gen11_reset_guc_interrupts;
197 guc->interrupts.enable = gen11_enable_guc_interrupts;
198 guc->interrupts.disable = gen11_disable_guc_interrupts;
199 if (gt->type == GT_MEDIA) {
200 guc->notify_reg = MEDIA_GUC_HOST_INTERRUPT;
201 guc->send_regs.base = i915_mmio_reg_offset(MEDIA_SOFT_SCRATCH(0));
202 } else {
203 guc->notify_reg = GEN11_GUC_HOST_INTERRUPT;
204 guc->send_regs.base = i915_mmio_reg_offset(GEN11_SOFT_SCRATCH(0));
205 }
206
207 guc->send_regs.count = GEN11_SOFT_SCRATCH_COUNT;
208
209 } else {
210 guc->notify_reg = GUC_SEND_INTERRUPT;
211 guc->interrupts.reset = gen9_reset_guc_interrupts;
212 guc->interrupts.enable = gen9_enable_guc_interrupts;
213 guc->interrupts.disable = gen9_disable_guc_interrupts;
214 guc->send_regs.base = i915_mmio_reg_offset(SOFT_SCRATCH(0));
215 guc->send_regs.count = GUC_MAX_MMIO_MSG_LEN;
216 BUILD_BUG_ON(GUC_MAX_MMIO_MSG_LEN > SOFT_SCRATCH_COUNT);
217 }
218
219 intel_guc_enable_msg(guc, mask: INTEL_GUC_RECV_MSG_EXCEPTION |
220 INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED);
221}
222
223void intel_guc_init_late(struct intel_guc *guc)
224{
225 intel_guc_ads_init_late(guc);
226}
227
228static u32 guc_ctl_debug_flags(struct intel_guc *guc)
229{
230 u32 level = intel_guc_log_get_level(log: &guc->log);
231 u32 flags = 0;
232
233 if (!GUC_LOG_LEVEL_IS_VERBOSE(level))
234 flags |= GUC_LOG_DISABLED;
235 else
236 flags |= GUC_LOG_LEVEL_TO_VERBOSITY(level) <<
237 GUC_LOG_VERBOSITY_SHIFT;
238
239 return flags;
240}
241
242static u32 guc_ctl_feature_flags(struct intel_guc *guc)
243{
244 struct intel_gt *gt = guc_to_gt(guc);
245 u32 flags = 0;
246
247 /*
248 * Enable PXP GuC autoteardown flow.
249 * NB: MTL does things differently.
250 */
251 if (HAS_PXP(gt->i915) && !IS_METEORLAKE(gt->i915))
252 flags |= GUC_CTL_ENABLE_GUC_PXP_CTL;
253
254 if (!intel_guc_submission_is_used(guc))
255 flags |= GUC_CTL_DISABLE_SCHEDULER;
256
257 if (intel_guc_slpc_is_used(guc))
258 flags |= GUC_CTL_ENABLE_SLPC;
259
260 return flags;
261}
262
263static u32 guc_ctl_log_params_flags(struct intel_guc *guc)
264{
265 struct intel_guc_log *log = &guc->log;
266 u32 offset, flags;
267
268 GEM_BUG_ON(!log->sizes_initialised);
269
270 offset = intel_guc_ggtt_offset(guc, vma: log->vma) >> PAGE_SHIFT;
271
272 flags = GUC_LOG_VALID |
273 GUC_LOG_NOTIFY_ON_HALF_FULL |
274 log->sizes[GUC_LOG_SECTIONS_DEBUG].flag |
275 log->sizes[GUC_LOG_SECTIONS_CAPTURE].flag |
276 (log->sizes[GUC_LOG_SECTIONS_CRASH].count << GUC_LOG_CRASH_SHIFT) |
277 (log->sizes[GUC_LOG_SECTIONS_DEBUG].count << GUC_LOG_DEBUG_SHIFT) |
278 (log->sizes[GUC_LOG_SECTIONS_CAPTURE].count << GUC_LOG_CAPTURE_SHIFT) |
279 (offset << GUC_LOG_BUF_ADDR_SHIFT);
280
281 return flags;
282}
283
284static u32 guc_ctl_ads_flags(struct intel_guc *guc)
285{
286 u32 ads = intel_guc_ggtt_offset(guc, vma: guc->ads_vma) >> PAGE_SHIFT;
287 u32 flags = ads << GUC_ADS_ADDR_SHIFT;
288
289 return flags;
290}
291
292static u32 guc_ctl_wa_flags(struct intel_guc *guc)
293{
294 struct intel_gt *gt = guc_to_gt(guc);
295 u32 flags = 0;
296
297 /* Wa_22012773006:gen11,gen12 < XeHP */
298 if (GRAPHICS_VER(gt->i915) >= 11 &&
299 GRAPHICS_VER_FULL(gt->i915) < IP_VER(12, 55))
300 flags |= GUC_WA_POLLCS;
301
302 /* Wa_14014475959 */
303 if (IS_GFX_GT_IP_STEP(gt, IP_VER(12, 70), STEP_A0, STEP_B0) ||
304 IS_DG2(gt->i915))
305 flags |= GUC_WA_HOLD_CCS_SWITCHOUT;
306
307 /* Wa_16019325821 */
308 /* Wa_14019159160 */
309 if (IS_GFX_GT_IP_RANGE(gt, IP_VER(12, 70), IP_VER(12, 74)))
310 flags |= GUC_WA_RCS_CCS_SWITCHOUT;
311
312 /*
313 * Wa_14012197797
314 * Wa_22011391025
315 *
316 * The same WA bit is used for both and 22011391025 is applicable to
317 * all DG2.
318 *
319 * Platforms post DG2 prevent this issue in hardware by stalling
320 * submissions. With this flag GuC will schedule as to avoid such
321 * stalls.
322 */
323 if (IS_DG2(gt->i915) ||
324 (CCS_MASK(gt) && GRAPHICS_VER_FULL(gt->i915) >= IP_VER(12, 70)))
325 flags |= GUC_WA_DUAL_QUEUE;
326
327 /* Wa_22011802037: graphics version 11/12 */
328 if (intel_engine_reset_needs_wa_22011802037(gt))
329 flags |= GUC_WA_PRE_PARSER;
330
331 /*
332 * Wa_22012727170
333 * Wa_22012727685
334 */
335 if (IS_DG2_G11(gt->i915))
336 flags |= GUC_WA_CONTEXT_ISOLATION;
337
338 /*
339 * Wa_14018913170: Applicable to all platforms supported by i915 so
340 * don't bother testing for all X/Y/Z platforms explicitly.
341 */
342 if (GUC_FIRMWARE_VER(guc) >= MAKE_GUC_VER(70, 7, 0))
343 flags |= GUC_WA_ENABLE_TSC_CHECK_ON_RC6;
344
345 return flags;
346}
347
348static u32 guc_ctl_devid(struct intel_guc *guc)
349{
350 struct drm_i915_private *i915 = guc_to_i915(guc);
351
352 return (INTEL_DEVID(i915) << 16) | INTEL_REVID(i915);
353}
354
355/*
356 * Initialise the GuC parameter block before starting the firmware
357 * transfer. These parameters are read by the firmware on startup
358 * and cannot be changed thereafter.
359 */
360static void guc_init_params(struct intel_guc *guc)
361{
362 u32 *params = guc->params;
363 int i;
364
365 BUILD_BUG_ON(sizeof(guc->params) != GUC_CTL_MAX_DWORDS * sizeof(u32));
366
367 params[GUC_CTL_LOG_PARAMS] = guc_ctl_log_params_flags(guc);
368 params[GUC_CTL_FEATURE] = guc_ctl_feature_flags(guc);
369 params[GUC_CTL_DEBUG] = guc_ctl_debug_flags(guc);
370 params[GUC_CTL_ADS] = guc_ctl_ads_flags(guc);
371 params[GUC_CTL_WA] = guc_ctl_wa_flags(guc);
372 params[GUC_CTL_DEVID] = guc_ctl_devid(guc);
373
374 for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
375 guc_dbg(guc, "param[%2d] = %#x\n", i, params[i]);
376}
377
378/*
379 * Initialise the GuC parameter block before starting the firmware
380 * transfer. These parameters are read by the firmware on startup
381 * and cannot be changed thereafter.
382 */
383void intel_guc_write_params(struct intel_guc *guc)
384{
385 struct intel_uncore *uncore = guc_to_gt(guc)->uncore;
386 int i;
387
388 /*
389 * All SOFT_SCRATCH registers are in FORCEWAKE_GT domain and
390 * they are power context saved so it's ok to release forcewake
391 * when we are done here and take it again at xfer time.
392 */
393 intel_uncore_forcewake_get(uncore, domains: FORCEWAKE_GT);
394
395 intel_uncore_write(uncore, SOFT_SCRATCH(0), val: 0);
396
397 for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
398 intel_uncore_write(uncore, SOFT_SCRATCH(1 + i), val: guc->params[i]);
399
400 intel_uncore_forcewake_put(uncore, domains: FORCEWAKE_GT);
401}
402
403void intel_guc_dump_time_info(struct intel_guc *guc, struct drm_printer *p)
404{
405 struct intel_gt *gt = guc_to_gt(guc);
406 intel_wakeref_t wakeref;
407 u32 stamp = 0;
408 u64 ktime;
409
410 with_intel_runtime_pm(&gt->i915->runtime_pm, wakeref)
411 stamp = intel_uncore_read(uncore: gt->uncore, GUCPMTIMESTAMP);
412 ktime = ktime_get_boottime_ns();
413
414 drm_printf(p, f: "Kernel timestamp: 0x%08llX [%llu]\n", ktime, ktime);
415 drm_printf(p, f: "GuC timestamp: 0x%08X [%u]\n", stamp, stamp);
416 drm_printf(p, f: "CS timestamp frequency: %u Hz, %u ns\n",
417 gt->clock_frequency, gt->clock_period_ns);
418}
419
420int intel_guc_init(struct intel_guc *guc)
421{
422 int ret;
423
424 ret = intel_uc_fw_init(uc_fw: &guc->fw);
425 if (ret)
426 goto out;
427
428 ret = intel_guc_log_create(log: &guc->log);
429 if (ret)
430 goto err_fw;
431
432 ret = intel_guc_capture_init(guc);
433 if (ret)
434 goto err_log;
435
436 ret = intel_guc_ads_create(guc);
437 if (ret)
438 goto err_capture;
439
440 GEM_BUG_ON(!guc->ads_vma);
441
442 ret = intel_guc_ct_init(ct: &guc->ct);
443 if (ret)
444 goto err_ads;
445
446 if (intel_guc_submission_is_used(guc)) {
447 /*
448 * This is stuff we need to have available at fw load time
449 * if we are planning to enable submission later
450 */
451 ret = intel_guc_submission_init(guc);
452 if (ret)
453 goto err_ct;
454 }
455
456 if (intel_guc_slpc_is_used(guc)) {
457 ret = intel_guc_slpc_init(slpc: &guc->slpc);
458 if (ret)
459 goto err_submission;
460 }
461
462 /* now that everything is perma-pinned, initialize the parameters */
463 guc_init_params(guc);
464
465 intel_uc_fw_change_status(uc_fw: &guc->fw, status: INTEL_UC_FIRMWARE_LOADABLE);
466
467 return 0;
468
469err_submission:
470 intel_guc_submission_fini(guc);
471err_ct:
472 intel_guc_ct_fini(ct: &guc->ct);
473err_ads:
474 intel_guc_ads_destroy(guc);
475err_capture:
476 intel_guc_capture_destroy(guc);
477err_log:
478 intel_guc_log_destroy(log: &guc->log);
479err_fw:
480 intel_uc_fw_fini(uc_fw: &guc->fw);
481out:
482 intel_uc_fw_change_status(uc_fw: &guc->fw, status: INTEL_UC_FIRMWARE_INIT_FAIL);
483 guc_probe_error(guc, "failed with %pe\n", ERR_PTR(ret));
484 return ret;
485}
486
487void intel_guc_fini(struct intel_guc *guc)
488{
489 if (!intel_uc_fw_is_loadable(uc_fw: &guc->fw))
490 return;
491
492 flush_work(work: &guc->dead_guc_worker);
493
494 if (intel_guc_slpc_is_used(guc))
495 intel_guc_slpc_fini(slpc: &guc->slpc);
496
497 if (intel_guc_submission_is_used(guc))
498 intel_guc_submission_fini(guc);
499
500 intel_guc_ct_fini(ct: &guc->ct);
501
502 intel_guc_ads_destroy(guc);
503 intel_guc_capture_destroy(guc);
504 intel_guc_log_destroy(log: &guc->log);
505 intel_uc_fw_fini(uc_fw: &guc->fw);
506}
507
508/*
509 * This function implements the MMIO based host to GuC interface.
510 */
511int intel_guc_send_mmio(struct intel_guc *guc, const u32 *request, u32 len,
512 u32 *response_buf, u32 response_buf_size)
513{
514 struct intel_uncore *uncore = guc_to_gt(guc)->uncore;
515 u32 header;
516 int i;
517 int ret;
518
519 GEM_BUG_ON(!len);
520 GEM_BUG_ON(len > guc->send_regs.count);
521
522 GEM_BUG_ON(FIELD_GET(GUC_HXG_MSG_0_ORIGIN, request[0]) != GUC_HXG_ORIGIN_HOST);
523 GEM_BUG_ON(FIELD_GET(GUC_HXG_MSG_0_TYPE, request[0]) != GUC_HXG_TYPE_REQUEST);
524
525 mutex_lock(lock: &guc->send_mutex);
526 intel_uncore_forcewake_get(uncore, domains: guc->send_regs.fw_domains);
527
528retry:
529 for (i = 0; i < len; i++)
530 intel_uncore_write(uncore, reg: guc_send_reg(guc, i), val: request[i]);
531
532 intel_uncore_posting_read(uncore, guc_send_reg(guc, i - 1));
533
534 intel_guc_notify(guc);
535
536 /*
537 * No GuC command should ever take longer than 10ms.
538 * Fast commands should still complete in 10us.
539 */
540 ret = __intel_wait_for_register_fw(uncore,
541 reg: guc_send_reg(guc, i: 0),
542 GUC_HXG_MSG_0_ORIGIN,
543 FIELD_PREP(GUC_HXG_MSG_0_ORIGIN,
544 GUC_HXG_ORIGIN_GUC),
545 fast_timeout_us: 10, slow_timeout_ms: 10, out_value: &header);
546 if (unlikely(ret)) {
547timeout:
548 guc_err(guc, "mmio request %#x: no reply %x\n",
549 request[0], header);
550 goto out;
551 }
552
553 if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) == GUC_HXG_TYPE_NO_RESPONSE_BUSY) {
554#define done ({ header = intel_uncore_read(uncore, guc_send_reg(guc, 0)); \
555 FIELD_GET(GUC_HXG_MSG_0_ORIGIN, header) != GUC_HXG_ORIGIN_GUC || \
556 FIELD_GET(GUC_HXG_MSG_0_TYPE, header) != GUC_HXG_TYPE_NO_RESPONSE_BUSY; })
557
558 ret = wait_for(done, 1000);
559 if (unlikely(ret))
560 goto timeout;
561 if (unlikely(FIELD_GET(GUC_HXG_MSG_0_ORIGIN, header) !=
562 GUC_HXG_ORIGIN_GUC))
563 goto proto;
564#undef done
565 }
566
567 if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) == GUC_HXG_TYPE_NO_RESPONSE_RETRY) {
568 u32 reason = FIELD_GET(GUC_HXG_RETRY_MSG_0_REASON, header);
569
570 guc_dbg(guc, "mmio request %#x: retrying, reason %u\n",
571 request[0], reason);
572 goto retry;
573 }
574
575 if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) == GUC_HXG_TYPE_RESPONSE_FAILURE) {
576 u32 hint = FIELD_GET(GUC_HXG_FAILURE_MSG_0_HINT, header);
577 u32 error = FIELD_GET(GUC_HXG_FAILURE_MSG_0_ERROR, header);
578
579 guc_err(guc, "mmio request %#x: failure %x/%u\n",
580 request[0], error, hint);
581 ret = -ENXIO;
582 goto out;
583 }
584
585 if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) != GUC_HXG_TYPE_RESPONSE_SUCCESS) {
586proto:
587 guc_err(guc, "mmio request %#x: unexpected reply %#x\n",
588 request[0], header);
589 ret = -EPROTO;
590 goto out;
591 }
592
593 if (response_buf) {
594 int count = min(response_buf_size, guc->send_regs.count);
595
596 GEM_BUG_ON(!count);
597
598 response_buf[0] = header;
599
600 for (i = 1; i < count; i++)
601 response_buf[i] = intel_uncore_read(uncore,
602 reg: guc_send_reg(guc, i));
603
604 /* Use number of copied dwords as our return value */
605 ret = count;
606 } else {
607 /* Use data from the GuC response as our return value */
608 ret = FIELD_GET(GUC_HXG_RESPONSE_MSG_0_DATA0, header);
609 }
610
611out:
612 intel_uncore_forcewake_put(uncore, domains: guc->send_regs.fw_domains);
613 mutex_unlock(lock: &guc->send_mutex);
614
615 return ret;
616}
617
618int intel_guc_crash_process_msg(struct intel_guc *guc, u32 action)
619{
620 if (action == INTEL_GUC_ACTION_NOTIFY_CRASH_DUMP_POSTED)
621 guc_err(guc, "Crash dump notification\n");
622 else if (action == INTEL_GUC_ACTION_NOTIFY_EXCEPTION)
623 guc_err(guc, "Exception notification\n");
624 else
625 guc_err(guc, "Unknown crash notification: 0x%04X\n", action);
626
627 queue_work(wq: system_unbound_wq, work: &guc->dead_guc_worker);
628
629 return 0;
630}
631
632int intel_guc_to_host_process_recv_msg(struct intel_guc *guc,
633 const u32 *payload, u32 len)
634{
635 u32 msg;
636
637 if (unlikely(!len))
638 return -EPROTO;
639
640 /* Make sure to handle only enabled messages */
641 msg = payload[0] & guc->msg_enabled_mask;
642
643 if (msg & INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED)
644 guc_err(guc, "Received early crash dump notification!\n");
645 if (msg & INTEL_GUC_RECV_MSG_EXCEPTION)
646 guc_err(guc, "Received early exception notification!\n");
647
648 if (msg & (INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED | INTEL_GUC_RECV_MSG_EXCEPTION))
649 queue_work(wq: system_unbound_wq, work: &guc->dead_guc_worker);
650
651 return 0;
652}
653
654/**
655 * intel_guc_auth_huc() - Send action to GuC to authenticate HuC ucode
656 * @guc: intel_guc structure
657 * @rsa_offset: rsa offset w.r.t ggtt base of huc vma
658 *
659 * Triggers a HuC firmware authentication request to the GuC via intel_guc_send
660 * INTEL_GUC_ACTION_AUTHENTICATE_HUC interface. This function is invoked by
661 * intel_huc_auth().
662 *
663 * Return: non-zero code on error
664 */
665int intel_guc_auth_huc(struct intel_guc *guc, u32 rsa_offset)
666{
667 u32 action[] = {
668 INTEL_GUC_ACTION_AUTHENTICATE_HUC,
669 rsa_offset
670 };
671
672 return intel_guc_send(guc, action, ARRAY_SIZE(action));
673}
674
675/**
676 * intel_guc_suspend() - notify GuC entering suspend state
677 * @guc: the guc
678 */
679int intel_guc_suspend(struct intel_guc *guc)
680{
681 int ret;
682 u32 action[] = {
683 INTEL_GUC_ACTION_CLIENT_SOFT_RESET,
684 };
685
686 if (!intel_guc_is_ready(guc))
687 return 0;
688
689 if (intel_guc_submission_is_used(guc)) {
690 flush_work(work: &guc->dead_guc_worker);
691
692 /*
693 * This H2G MMIO command tears down the GuC in two steps. First it will
694 * generate a G2H CTB for every active context indicating a reset. In
695 * practice the i915 shouldn't ever get a G2H as suspend should only be
696 * called when the GPU is idle. Next, it tears down the CTBs and this
697 * H2G MMIO command completes.
698 *
699 * Don't abort on a failure code from the GuC. Keep going and do the
700 * clean up in sanitize() and re-initialisation on resume and hopefully
701 * the error here won't be problematic.
702 */
703 ret = intel_guc_send_mmio(guc, request: action, ARRAY_SIZE(action), NULL, response_buf_size: 0);
704 if (ret)
705 guc_err(guc, "suspend: RESET_CLIENT action failed with %pe\n",
706 ERR_PTR(ret));
707 }
708
709 /* Signal that the GuC isn't running. */
710 intel_guc_sanitize(guc);
711
712 return 0;
713}
714
715/**
716 * intel_guc_resume() - notify GuC resuming from suspend state
717 * @guc: the guc
718 */
719int intel_guc_resume(struct intel_guc *guc)
720{
721 /*
722 * NB: This function can still be called even if GuC submission is
723 * disabled, e.g. if GuC is enabled for HuC authentication only. Thus,
724 * if any code is later added here, it must be support doing nothing
725 * if submission is disabled (as per intel_guc_suspend).
726 */
727 return 0;
728}
729
730/**
731 * DOC: GuC Memory Management
732 *
733 * GuC can't allocate any memory for its own usage, so all the allocations must
734 * be handled by the host driver. GuC accesses the memory via the GGTT, with the
735 * exception of the top and bottom parts of the 4GB address space, which are
736 * instead re-mapped by the GuC HW to memory location of the FW itself (WOPCM)
737 * or other parts of the HW. The driver must take care not to place objects that
738 * the GuC is going to access in these reserved ranges. The layout of the GuC
739 * address space is shown below:
740 *
741 * ::
742 *
743 * +===========> +====================+ <== FFFF_FFFF
744 * ^ | Reserved |
745 * | +====================+ <== GUC_GGTT_TOP
746 * | | |
747 * | | DRAM |
748 * GuC | |
749 * Address +===> +====================+ <== GuC ggtt_pin_bias
750 * Space ^ | |
751 * | | | |
752 * | GuC | GuC |
753 * | WOPCM | WOPCM |
754 * | Size | |
755 * | | | |
756 * v v | |
757 * +=======+===> +====================+ <== 0000_0000
758 *
759 * The lower part of GuC Address Space [0, ggtt_pin_bias) is mapped to GuC WOPCM
760 * while upper part of GuC Address Space [ggtt_pin_bias, GUC_GGTT_TOP) is mapped
761 * to DRAM. The value of the GuC ggtt_pin_bias is the GuC WOPCM size.
762 */
763
764/**
765 * intel_guc_allocate_vma() - Allocate a GGTT VMA for GuC usage
766 * @guc: the guc
767 * @size: size of area to allocate (both virtual space and memory)
768 *
769 * This is a wrapper to create an object for use with the GuC. In order to
770 * use it inside the GuC, an object needs to be pinned lifetime, so we allocate
771 * both some backing storage and a range inside the Global GTT. We must pin
772 * it in the GGTT somewhere other than than [0, GUC ggtt_pin_bias) because that
773 * range is reserved inside GuC.
774 *
775 * Return: A i915_vma if successful, otherwise an ERR_PTR.
776 */
777struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size)
778{
779 struct intel_gt *gt = guc_to_gt(guc);
780 struct drm_i915_gem_object *obj;
781 struct i915_vma *vma;
782 u64 flags;
783 int ret;
784
785 if (HAS_LMEM(gt->i915))
786 obj = i915_gem_object_create_lmem(i915: gt->i915, size,
787 I915_BO_ALLOC_CPU_CLEAR |
788 I915_BO_ALLOC_CONTIGUOUS |
789 I915_BO_ALLOC_PM_EARLY);
790 else
791 obj = i915_gem_object_create_shmem(i915: gt->i915, size);
792
793 if (IS_ERR(ptr: obj))
794 return ERR_CAST(ptr: obj);
795
796 /*
797 * Wa_22016122933: For Media version 13.0, all Media GT shared
798 * memory needs to be mapped as WC on CPU side and UC (PAT
799 * index 2) on GPU side.
800 */
801 if (intel_gt_needs_wa_22016122933(gt))
802 i915_gem_object_set_cache_coherency(obj, cache_level: I915_CACHE_NONE);
803
804 vma = i915_vma_instance(obj, vm: &gt->ggtt->vm, NULL);
805 if (IS_ERR(ptr: vma))
806 goto err;
807
808 flags = PIN_OFFSET_BIAS | i915_ggtt_pin_bias(vma);
809 ret = i915_ggtt_pin(vma, NULL, align: 0, flags);
810 if (ret) {
811 vma = ERR_PTR(error: ret);
812 goto err;
813 }
814
815 return i915_vma_make_unshrinkable(vma);
816
817err:
818 i915_gem_object_put(obj);
819 return vma;
820}
821
822/**
823 * intel_guc_allocate_and_map_vma() - Allocate and map VMA for GuC usage
824 * @guc: the guc
825 * @size: size of area to allocate (both virtual space and memory)
826 * @out_vma: return variable for the allocated vma pointer
827 * @out_vaddr: return variable for the obj mapping
828 *
829 * This wrapper calls intel_guc_allocate_vma() and then maps the allocated
830 * object with I915_MAP_WB.
831 *
832 * Return: 0 if successful, a negative errno code otherwise.
833 */
834int intel_guc_allocate_and_map_vma(struct intel_guc *guc, u32 size,
835 struct i915_vma **out_vma, void **out_vaddr)
836{
837 struct i915_vma *vma;
838 void *vaddr;
839
840 vma = intel_guc_allocate_vma(guc, size);
841 if (IS_ERR(ptr: vma))
842 return PTR_ERR(ptr: vma);
843
844 vaddr = i915_gem_object_pin_map_unlocked(obj: vma->obj,
845 type: intel_gt_coherent_map_type(gt: guc_to_gt(guc),
846 obj: vma->obj, always_coherent: true));
847 if (IS_ERR(ptr: vaddr)) {
848 i915_vma_unpin_and_release(p_vma: &vma, flags: 0);
849 return PTR_ERR(ptr: vaddr);
850 }
851
852 *out_vma = vma;
853 *out_vaddr = vaddr;
854
855 return 0;
856}
857
858static int __guc_action_self_cfg(struct intel_guc *guc, u16 key, u16 len, u64 value)
859{
860 u32 request[HOST2GUC_SELF_CFG_REQUEST_MSG_LEN] = {
861 FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
862 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
863 FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION, GUC_ACTION_HOST2GUC_SELF_CFG),
864 FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_KEY, key) |
865 FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_LEN, len),
866 FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_2_VALUE32, lower_32_bits(value)),
867 FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_3_VALUE64, upper_32_bits(value)),
868 };
869 int ret;
870
871 GEM_BUG_ON(len > 2);
872 GEM_BUG_ON(len == 1 && upper_32_bits(value));
873
874 /* Self config must go over MMIO */
875 ret = intel_guc_send_mmio(guc, request, ARRAY_SIZE(request), NULL, response_buf_size: 0);
876
877 if (unlikely(ret < 0))
878 return ret;
879 if (unlikely(ret > 1))
880 return -EPROTO;
881 if (unlikely(!ret))
882 return -ENOKEY;
883
884 return 0;
885}
886
887static int __guc_self_cfg(struct intel_guc *guc, u16 key, u16 len, u64 value)
888{
889 int err = __guc_action_self_cfg(guc, key, len, value);
890
891 if (unlikely(err))
892 guc_probe_error(guc, "Unsuccessful self-config (%pe) key %#hx value %#llx\n",
893 ERR_PTR(err), key, value);
894 return err;
895}
896
897int intel_guc_self_cfg32(struct intel_guc *guc, u16 key, u32 value)
898{
899 return __guc_self_cfg(guc, key, len: 1, value);
900}
901
902int intel_guc_self_cfg64(struct intel_guc *guc, u16 key, u64 value)
903{
904 return __guc_self_cfg(guc, key, len: 2, value);
905}
906
907/**
908 * intel_guc_load_status - dump information about GuC load status
909 * @guc: the GuC
910 * @p: the &drm_printer
911 *
912 * Pretty printer for GuC load status.
913 */
914void intel_guc_load_status(struct intel_guc *guc, struct drm_printer *p)
915{
916 struct intel_gt *gt = guc_to_gt(guc);
917 struct intel_uncore *uncore = gt->uncore;
918 intel_wakeref_t wakeref;
919
920 if (!intel_guc_is_supported(guc)) {
921 drm_printf(p, f: "GuC not supported\n");
922 return;
923 }
924
925 if (!intel_guc_is_wanted(guc)) {
926 drm_printf(p, f: "GuC disabled\n");
927 return;
928 }
929
930 intel_uc_fw_dump(uc_fw: &guc->fw, p);
931
932 with_intel_runtime_pm(uncore->rpm, wakeref) {
933 u32 status = intel_uncore_read(uncore, GUC_STATUS);
934 u32 i;
935
936 drm_printf(p, f: "GuC status 0x%08x:\n", status);
937 drm_printf(p, f: "\tBootrom status = 0x%x\n",
938 (status & GS_BOOTROM_MASK) >> GS_BOOTROM_SHIFT);
939 drm_printf(p, f: "\tuKernel status = 0x%x\n",
940 (status & GS_UKERNEL_MASK) >> GS_UKERNEL_SHIFT);
941 drm_printf(p, f: "\tMIA Core status = 0x%x\n",
942 (status & GS_MIA_MASK) >> GS_MIA_SHIFT);
943 drm_puts(p, str: "Scratch registers:\n");
944 for (i = 0; i < 16; i++) {
945 drm_printf(p, f: "\t%2d: \t0x%x\n",
946 i, intel_uncore_read(uncore, SOFT_SCRATCH(i)));
947 }
948 }
949}
950
951void intel_guc_write_barrier(struct intel_guc *guc)
952{
953 struct intel_gt *gt = guc_to_gt(guc);
954
955 if (i915_gem_object_is_lmem(obj: guc->ct.vma->obj)) {
956 /*
957 * Ensure intel_uncore_write_fw can be used rather than
958 * intel_uncore_write.
959 */
960 GEM_BUG_ON(guc->send_regs.fw_domains);
961
962 /*
963 * This register is used by the i915 and GuC for MMIO based
964 * communication. Once we are in this code CTBs are the only
965 * method the i915 uses to communicate with the GuC so it is
966 * safe to write to this register (a value of 0 is NOP for MMIO
967 * communication). If we ever start mixing CTBs and MMIOs a new
968 * register will have to be chosen. This function is also used
969 * to enforce ordering of a work queue item write and an update
970 * to the process descriptor. When a work queue is being used,
971 * CTBs are also the only mechanism of communication.
972 */
973 intel_uncore_write_fw(gt->uncore, GEN11_SOFT_SCRATCH(0), 0);
974 } else {
975 /* wmb() sufficient for a barrier if in smem */
976 wmb();
977 }
978}
979