1 // SPDX-License-Identifier: MIT
2 /*
3 * Copyright © 2016-2019 Intel Corporation
4 */
5
6 #include "gt/intel_gt.h"
7 #include "gt/intel_reset.h"
8 #include "intel_guc.h"
9 #include "intel_guc_ads.h"
10 #include "intel_guc_submission.h"
11 #include "intel_uc.h"
12
13 #include "i915_drv.h"
14
15 static const struct intel_uc_ops uc_ops_off;
16 static const struct intel_uc_ops uc_ops_on;
17
18 /* Reset GuC providing us with fresh state for both GuC and HuC.
19 */
__intel_uc_reset_hw(struct intel_uc * uc)20 static int __intel_uc_reset_hw(struct intel_uc *uc)
21 {
22 struct intel_gt *gt = uc_to_gt(uc);
23 int ret;
24 u32 guc_status;
25
26 ret = i915_inject_probe_error(gt->i915, -ENXIO);
27 if (ret)
28 return ret;
29
30 ret = intel_reset_guc(gt);
31 if (ret) {
32 DRM_ERROR("Failed to reset GuC, ret = %d\n", ret);
33 return ret;
34 }
35
36 guc_status = intel_uncore_read(gt->uncore, GUC_STATUS);
37 WARN(!(guc_status & GS_MIA_IN_RESET),
38 "GuC status: 0x%x, MIA core expected to be in reset\n",
39 guc_status);
40
41 return ret;
42 }
43
__confirm_options(struct intel_uc * uc)44 static void __confirm_options(struct intel_uc *uc)
45 {
46 struct drm_i915_private *i915 = uc_to_gt(uc)->i915;
47
48 drm_dbg(&i915->drm,
49 "enable_guc=%d (guc:%s submission:%s huc:%s)\n",
50 i915->params.enable_guc,
51 yesno(intel_uc_wants_guc(uc)),
52 yesno(intel_uc_wants_guc_submission(uc)),
53 yesno(intel_uc_wants_huc(uc)));
54
55 if (i915->params.enable_guc == -1)
56 return;
57
58 if (i915->params.enable_guc == 0) {
59 GEM_BUG_ON(intel_uc_wants_guc(uc));
60 GEM_BUG_ON(intel_uc_wants_guc_submission(uc));
61 GEM_BUG_ON(intel_uc_wants_huc(uc));
62 return;
63 }
64
65 if (!intel_uc_supports_guc(uc))
66 drm_info(&i915->drm,
67 "Incompatible option enable_guc=%d - %s\n",
68 i915->params.enable_guc, "GuC is not supported!");
69
70 if (i915->params.enable_guc & ENABLE_GUC_LOAD_HUC &&
71 !intel_uc_supports_huc(uc))
72 drm_info(&i915->drm,
73 "Incompatible option enable_guc=%d - %s\n",
74 i915->params.enable_guc, "HuC is not supported!");
75
76 if (i915->params.enable_guc & ENABLE_GUC_SUBMISSION &&
77 !intel_uc_supports_guc_submission(uc))
78 drm_info(&i915->drm,
79 "Incompatible option enable_guc=%d - %s\n",
80 i915->params.enable_guc, "GuC submission is N/A");
81
82 if (i915->params.enable_guc & ~(ENABLE_GUC_SUBMISSION |
83 ENABLE_GUC_LOAD_HUC))
84 drm_info(&i915->drm,
85 "Incompatible option enable_guc=%d - %s\n",
86 i915->params.enable_guc, "undocumented flag");
87 }
88
intel_uc_init_early(struct intel_uc * uc)89 void intel_uc_init_early(struct intel_uc *uc)
90 {
91 intel_guc_init_early(&uc->guc);
92 intel_huc_init_early(&uc->huc);
93
94 __confirm_options(uc);
95
96 if (intel_uc_wants_guc(uc))
97 uc->ops = &uc_ops_on;
98 else
99 uc->ops = &uc_ops_off;
100 }
101
intel_uc_driver_late_release(struct intel_uc * uc)102 void intel_uc_driver_late_release(struct intel_uc *uc)
103 {
104 }
105
106 /**
107 * intel_uc_init_mmio - setup uC MMIO access
108 * @uc: the intel_uc structure
109 *
110 * Setup minimal state necessary for MMIO accesses later in the
111 * initialization sequence.
112 */
intel_uc_init_mmio(struct intel_uc * uc)113 void intel_uc_init_mmio(struct intel_uc *uc)
114 {
115 intel_guc_init_send_regs(&uc->guc);
116 }
117
__uc_capture_load_err_log(struct intel_uc * uc)118 static void __uc_capture_load_err_log(struct intel_uc *uc)
119 {
120 struct intel_guc *guc = &uc->guc;
121
122 if (guc->log.vma && !uc->load_err_log)
123 uc->load_err_log = i915_gem_object_get(guc->log.vma->obj);
124 }
125
__uc_free_load_err_log(struct intel_uc * uc)126 static void __uc_free_load_err_log(struct intel_uc *uc)
127 {
128 struct drm_i915_gem_object *log = fetch_and_zero(&uc->load_err_log);
129
130 if (log)
131 i915_gem_object_put(log);
132 }
133
intel_uc_driver_remove(struct intel_uc * uc)134 void intel_uc_driver_remove(struct intel_uc *uc)
135 {
136 intel_uc_fini_hw(uc);
137 intel_uc_fini(uc);
138 __uc_free_load_err_log(uc);
139 }
140
guc_communication_enabled(struct intel_guc * guc)141 static inline bool guc_communication_enabled(struct intel_guc *guc)
142 {
143 return intel_guc_ct_enabled(&guc->ct);
144 }
145
146 /*
147 * Events triggered while CT buffers are disabled are logged in the SCRATCH_15
148 * register using the same bits used in the CT message payload. Since our
149 * communication channel with guc is turned off at this point, we can save the
150 * message and handle it after we turn it back on.
151 */
guc_clear_mmio_msg(struct intel_guc * guc)152 static void guc_clear_mmio_msg(struct intel_guc *guc)
153 {
154 intel_uncore_write(guc_to_gt(guc)->uncore, SOFT_SCRATCH(15), 0);
155 }
156
guc_get_mmio_msg(struct intel_guc * guc)157 static void guc_get_mmio_msg(struct intel_guc *guc)
158 {
159 u32 val;
160
161 spin_lock_irq(&guc->irq_lock);
162
163 val = intel_uncore_read(guc_to_gt(guc)->uncore, SOFT_SCRATCH(15));
164 guc->mmio_msg |= val & guc->msg_enabled_mask;
165
166 /*
167 * clear all events, including the ones we're not currently servicing,
168 * to make sure we don't try to process a stale message if we enable
169 * handling of more events later.
170 */
171 guc_clear_mmio_msg(guc);
172
173 spin_unlock_irq(&guc->irq_lock);
174 }
175
guc_handle_mmio_msg(struct intel_guc * guc)176 static void guc_handle_mmio_msg(struct intel_guc *guc)
177 {
178 struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
179
180 /* we need communication to be enabled to reply to GuC */
181 GEM_BUG_ON(!guc_communication_enabled(guc));
182
183 if (!guc->mmio_msg)
184 return;
185
186 spin_lock_irq(&i915->irq_lock);
187 intel_guc_to_host_process_recv_msg(guc, &guc->mmio_msg, 1);
188 spin_unlock_irq(&i915->irq_lock);
189
190 guc->mmio_msg = 0;
191 }
192
guc_reset_interrupts(struct intel_guc * guc)193 static void guc_reset_interrupts(struct intel_guc *guc)
194 {
195 guc->interrupts.reset(guc);
196 }
197
guc_enable_interrupts(struct intel_guc * guc)198 static void guc_enable_interrupts(struct intel_guc *guc)
199 {
200 guc->interrupts.enable(guc);
201 }
202
guc_disable_interrupts(struct intel_guc * guc)203 static void guc_disable_interrupts(struct intel_guc *guc)
204 {
205 guc->interrupts.disable(guc);
206 }
207
guc_enable_communication(struct intel_guc * guc)208 static int guc_enable_communication(struct intel_guc *guc)
209 {
210 struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
211 int ret;
212
213 GEM_BUG_ON(guc_communication_enabled(guc));
214
215 ret = i915_inject_probe_error(i915, -ENXIO);
216 if (ret)
217 return ret;
218
219 ret = intel_guc_ct_enable(&guc->ct);
220 if (ret)
221 return ret;
222
223 /* check for mmio messages received before/during the CT enable */
224 guc_get_mmio_msg(guc);
225 guc_handle_mmio_msg(guc);
226
227 guc_enable_interrupts(guc);
228
229 /* check for CT messages received before we enabled interrupts */
230 spin_lock_irq(&i915->irq_lock);
231 intel_guc_ct_event_handler(&guc->ct);
232 spin_unlock_irq(&i915->irq_lock);
233
234 DRM_INFO("GuC communication enabled\n");
235
236 return 0;
237 }
238
guc_disable_communication(struct intel_guc * guc)239 static void guc_disable_communication(struct intel_guc *guc)
240 {
241 /*
242 * Events generated during or after CT disable are logged by guc in
243 * via mmio. Make sure the register is clear before disabling CT since
244 * all events we cared about have already been processed via CT.
245 */
246 guc_clear_mmio_msg(guc);
247
248 guc_disable_interrupts(guc);
249
250 intel_guc_ct_disable(&guc->ct);
251
252 /*
253 * Check for messages received during/after the CT disable. We do not
254 * expect any messages to have arrived via CT between the interrupt
255 * disable and the CT disable because GuC should've been idle until we
256 * triggered the CT disable protocol.
257 */
258 guc_get_mmio_msg(guc);
259
260 DRM_INFO("GuC communication disabled\n");
261 }
262
__uc_fetch_firmwares(struct intel_uc * uc)263 static void __uc_fetch_firmwares(struct intel_uc *uc)
264 {
265 int err;
266
267 GEM_BUG_ON(!intel_uc_wants_guc(uc));
268
269 err = intel_uc_fw_fetch(&uc->guc.fw);
270 if (err) {
271 /* Make sure we transition out of transient "SELECTED" state */
272 if (intel_uc_wants_huc(uc)) {
273 drm_dbg(&uc_to_gt(uc)->i915->drm,
274 "Failed to fetch GuC: %d disabling HuC\n", err);
275 intel_uc_fw_change_status(&uc->huc.fw,
276 INTEL_UC_FIRMWARE_ERROR);
277 }
278
279 return;
280 }
281
282 if (intel_uc_wants_huc(uc))
283 intel_uc_fw_fetch(&uc->huc.fw);
284 }
285
__uc_cleanup_firmwares(struct intel_uc * uc)286 static void __uc_cleanup_firmwares(struct intel_uc *uc)
287 {
288 intel_uc_fw_cleanup_fetch(&uc->huc.fw);
289 intel_uc_fw_cleanup_fetch(&uc->guc.fw);
290 }
291
__uc_init(struct intel_uc * uc)292 static int __uc_init(struct intel_uc *uc)
293 {
294 struct intel_guc *guc = &uc->guc;
295 struct intel_huc *huc = &uc->huc;
296 int ret;
297
298 GEM_BUG_ON(!intel_uc_wants_guc(uc));
299
300 if (!intel_uc_uses_guc(uc))
301 return 0;
302
303 if (i915_inject_probe_failure(uc_to_gt(uc)->i915))
304 return -ENOMEM;
305
306 /* XXX: GuC submission is unavailable for now */
307 GEM_BUG_ON(intel_uc_uses_guc_submission(uc));
308
309 ret = intel_guc_init(guc);
310 if (ret)
311 return ret;
312
313 if (intel_uc_uses_huc(uc)) {
314 ret = intel_huc_init(huc);
315 if (ret)
316 goto out_guc;
317 }
318
319 return 0;
320
321 out_guc:
322 intel_guc_fini(guc);
323 return ret;
324 }
325
__uc_fini(struct intel_uc * uc)326 static void __uc_fini(struct intel_uc *uc)
327 {
328 intel_huc_fini(&uc->huc);
329 intel_guc_fini(&uc->guc);
330 }
331
__uc_sanitize(struct intel_uc * uc)332 static int __uc_sanitize(struct intel_uc *uc)
333 {
334 struct intel_guc *guc = &uc->guc;
335 struct intel_huc *huc = &uc->huc;
336
337 GEM_BUG_ON(!intel_uc_supports_guc(uc));
338
339 intel_huc_sanitize(huc);
340 intel_guc_sanitize(guc);
341
342 return __intel_uc_reset_hw(uc);
343 }
344
345 /* Initialize and verify the uC regs related to uC positioning in WOPCM */
uc_init_wopcm(struct intel_uc * uc)346 static int uc_init_wopcm(struct intel_uc *uc)
347 {
348 struct intel_gt *gt = uc_to_gt(uc);
349 struct intel_uncore *uncore = gt->uncore;
350 u32 base = intel_wopcm_guc_base(>->i915->wopcm);
351 u32 size = intel_wopcm_guc_size(>->i915->wopcm);
352 u32 huc_agent = intel_uc_uses_huc(uc) ? HUC_LOADING_AGENT_GUC : 0;
353 u32 mask;
354 int err;
355
356 if (unlikely(!base || !size)) {
357 i915_probe_error(gt->i915, "Unsuccessful WOPCM partitioning\n");
358 return -E2BIG;
359 }
360
361 GEM_BUG_ON(!intel_uc_supports_guc(uc));
362 GEM_BUG_ON(!(base & GUC_WOPCM_OFFSET_MASK));
363 GEM_BUG_ON(base & ~GUC_WOPCM_OFFSET_MASK);
364 GEM_BUG_ON(!(size & GUC_WOPCM_SIZE_MASK));
365 GEM_BUG_ON(size & ~GUC_WOPCM_SIZE_MASK);
366
367 err = i915_inject_probe_error(gt->i915, -ENXIO);
368 if (err)
369 return err;
370
371 mask = GUC_WOPCM_SIZE_MASK | GUC_WOPCM_SIZE_LOCKED;
372 err = intel_uncore_write_and_verify(uncore, GUC_WOPCM_SIZE, size, mask,
373 size | GUC_WOPCM_SIZE_LOCKED);
374 if (err)
375 goto err_out;
376
377 mask = GUC_WOPCM_OFFSET_MASK | GUC_WOPCM_OFFSET_VALID | huc_agent;
378 err = intel_uncore_write_and_verify(uncore, DMA_GUC_WOPCM_OFFSET,
379 base | huc_agent, mask,
380 base | huc_agent |
381 GUC_WOPCM_OFFSET_VALID);
382 if (err)
383 goto err_out;
384
385 return 0;
386
387 err_out:
388 i915_probe_error(gt->i915, "Failed to init uC WOPCM registers!\n");
389 i915_probe_error(gt->i915, "%s(%#x)=%#x\n", "DMA_GUC_WOPCM_OFFSET",
390 i915_mmio_reg_offset(DMA_GUC_WOPCM_OFFSET),
391 intel_uncore_read(uncore, DMA_GUC_WOPCM_OFFSET));
392 i915_probe_error(gt->i915, "%s(%#x)=%#x\n", "GUC_WOPCM_SIZE",
393 i915_mmio_reg_offset(GUC_WOPCM_SIZE),
394 intel_uncore_read(uncore, GUC_WOPCM_SIZE));
395
396 return err;
397 }
398
uc_is_wopcm_locked(struct intel_uc * uc)399 static bool uc_is_wopcm_locked(struct intel_uc *uc)
400 {
401 struct intel_gt *gt = uc_to_gt(uc);
402 struct intel_uncore *uncore = gt->uncore;
403
404 return (intel_uncore_read(uncore, GUC_WOPCM_SIZE) & GUC_WOPCM_SIZE_LOCKED) ||
405 (intel_uncore_read(uncore, DMA_GUC_WOPCM_OFFSET) & GUC_WOPCM_OFFSET_VALID);
406 }
407
__uc_check_hw(struct intel_uc * uc)408 static int __uc_check_hw(struct intel_uc *uc)
409 {
410 if (!intel_uc_supports_guc(uc))
411 return 0;
412
413 /*
414 * We can silently continue without GuC only if it was never enabled
415 * before on this system after reboot, otherwise we risk GPU hangs.
416 * To check if GuC was loaded before we look at WOPCM registers.
417 */
418 if (uc_is_wopcm_locked(uc))
419 return -EIO;
420
421 return 0;
422 }
423
__uc_init_hw(struct intel_uc * uc)424 static int __uc_init_hw(struct intel_uc *uc)
425 {
426 struct drm_i915_private *i915 = uc_to_gt(uc)->i915;
427 struct intel_guc *guc = &uc->guc;
428 struct intel_huc *huc = &uc->huc;
429 int ret, attempts;
430
431 GEM_BUG_ON(!intel_uc_supports_guc(uc));
432 GEM_BUG_ON(!intel_uc_wants_guc(uc));
433
434 if (!intel_uc_fw_is_loadable(&guc->fw)) {
435 ret = __uc_check_hw(uc) ||
436 intel_uc_fw_is_overridden(&guc->fw) ||
437 intel_uc_wants_guc_submission(uc) ?
438 intel_uc_fw_status_to_error(guc->fw.status) : 0;
439 goto err_out;
440 }
441
442 ret = uc_init_wopcm(uc);
443 if (ret)
444 goto err_out;
445
446 guc_reset_interrupts(guc);
447
448 /* WaEnableuKernelHeaderValidFix:skl */
449 /* WaEnableGuCBootHashCheckNotSet:skl,bxt,kbl */
450 if (IS_GEN(i915, 9))
451 attempts = 3;
452 else
453 attempts = 1;
454
455 while (attempts--) {
456 /*
457 * Always reset the GuC just before (re)loading, so
458 * that the state and timing are fairly predictable
459 */
460 ret = __uc_sanitize(uc);
461 if (ret)
462 goto err_out;
463
464 intel_huc_fw_upload(huc);
465 intel_guc_ads_reset(guc);
466 intel_guc_write_params(guc);
467 ret = intel_guc_fw_upload(guc);
468 if (ret == 0)
469 break;
470
471 DRM_DEBUG_DRIVER("GuC fw load failed: %d; will reset and "
472 "retry %d more time(s)\n", ret, attempts);
473 }
474
475 /* Did we succeded or run out of retries? */
476 if (ret)
477 goto err_log_capture;
478
479 ret = guc_enable_communication(guc);
480 if (ret)
481 goto err_log_capture;
482
483 intel_huc_auth(huc);
484
485 ret = intel_guc_sample_forcewake(guc);
486 if (ret)
487 goto err_communication;
488
489 if (intel_uc_uses_guc_submission(uc))
490 intel_guc_submission_enable(guc);
491
492 drm_info(&i915->drm, "%s firmware %s version %u.%u %s:%s\n",
493 intel_uc_fw_type_repr(INTEL_UC_FW_TYPE_GUC), guc->fw.path,
494 guc->fw.major_ver_found, guc->fw.minor_ver_found,
495 "submission",
496 enableddisabled(intel_uc_uses_guc_submission(uc)));
497
498 if (intel_uc_uses_huc(uc)) {
499 drm_info(&i915->drm, "%s firmware %s version %u.%u %s:%s\n",
500 intel_uc_fw_type_repr(INTEL_UC_FW_TYPE_HUC),
501 huc->fw.path,
502 huc->fw.major_ver_found, huc->fw.minor_ver_found,
503 "authenticated",
504 yesno(intel_huc_is_authenticated(huc)));
505 }
506
507 return 0;
508
509 /*
510 * We've failed to load the firmware :(
511 */
512 err_communication:
513 guc_disable_communication(guc);
514 err_log_capture:
515 __uc_capture_load_err_log(uc);
516 err_out:
517 __uc_sanitize(uc);
518
519 if (!ret) {
520 drm_notice(&i915->drm, "GuC is uninitialized\n");
521 /* We want to run without GuC submission */
522 return 0;
523 }
524
525 i915_probe_error(i915, "GuC initialization failed %d\n", ret);
526
527 /* We want to keep KMS alive */
528 return -EIO;
529 }
530
__uc_fini_hw(struct intel_uc * uc)531 static void __uc_fini_hw(struct intel_uc *uc)
532 {
533 struct intel_guc *guc = &uc->guc;
534
535 if (!intel_guc_is_fw_running(guc))
536 return;
537
538 if (intel_uc_uses_guc_submission(uc))
539 intel_guc_submission_disable(guc);
540
541 if (guc_communication_enabled(guc))
542 guc_disable_communication(guc);
543
544 __uc_sanitize(uc);
545 }
546
547 /**
548 * intel_uc_reset_prepare - Prepare for reset
549 * @uc: the intel_uc structure
550 *
551 * Preparing for full gpu reset.
552 */
intel_uc_reset_prepare(struct intel_uc * uc)553 void intel_uc_reset_prepare(struct intel_uc *uc)
554 {
555 struct intel_guc *guc = &uc->guc;
556
557 if (!intel_guc_is_ready(guc))
558 return;
559
560 guc_disable_communication(guc);
561 __uc_sanitize(uc);
562 }
563
intel_uc_runtime_suspend(struct intel_uc * uc)564 void intel_uc_runtime_suspend(struct intel_uc *uc)
565 {
566 struct intel_guc *guc = &uc->guc;
567 int err;
568
569 if (!intel_guc_is_ready(guc))
570 return;
571
572 err = intel_guc_suspend(guc);
573 if (err)
574 DRM_DEBUG_DRIVER("Failed to suspend GuC, err=%d", err);
575
576 guc_disable_communication(guc);
577 }
578
intel_uc_suspend(struct intel_uc * uc)579 void intel_uc_suspend(struct intel_uc *uc)
580 {
581 struct intel_guc *guc = &uc->guc;
582 intel_wakeref_t wakeref;
583
584 if (!intel_guc_is_ready(guc))
585 return;
586
587 with_intel_runtime_pm(uc_to_gt(uc)->uncore->rpm, wakeref)
588 intel_uc_runtime_suspend(uc);
589 }
590
__uc_resume(struct intel_uc * uc,bool enable_communication)591 static int __uc_resume(struct intel_uc *uc, bool enable_communication)
592 {
593 struct intel_guc *guc = &uc->guc;
594 int err;
595
596 if (!intel_guc_is_fw_running(guc))
597 return 0;
598
599 /* Make sure we enable communication if and only if it's disabled */
600 GEM_BUG_ON(enable_communication == guc_communication_enabled(guc));
601
602 if (enable_communication)
603 guc_enable_communication(guc);
604
605 err = intel_guc_resume(guc);
606 if (err) {
607 DRM_DEBUG_DRIVER("Failed to resume GuC, err=%d", err);
608 return err;
609 }
610
611 return 0;
612 }
613
intel_uc_resume(struct intel_uc * uc)614 int intel_uc_resume(struct intel_uc *uc)
615 {
616 /*
617 * When coming out of S3/S4 we sanitize and re-init the HW, so
618 * communication is already re-enabled at this point.
619 */
620 return __uc_resume(uc, false);
621 }
622
intel_uc_runtime_resume(struct intel_uc * uc)623 int intel_uc_runtime_resume(struct intel_uc *uc)
624 {
625 /*
626 * During runtime resume we don't sanitize, so we need to re-init
627 * communication as well.
628 */
629 return __uc_resume(uc, true);
630 }
631
632 static const struct intel_uc_ops uc_ops_off = {
633 .init_hw = __uc_check_hw,
634 };
635
636 static const struct intel_uc_ops uc_ops_on = {
637 .sanitize = __uc_sanitize,
638
639 .init_fw = __uc_fetch_firmwares,
640 .fini_fw = __uc_cleanup_firmwares,
641
642 .init = __uc_init,
643 .fini = __uc_fini,
644
645 .init_hw = __uc_init_hw,
646 .fini_hw = __uc_fini_hw,
647 };
648