1 // SPDX-License-Identifier: MIT
2 /*
3 * Copyright © 2014-2019 Intel Corporation
4 */
5
6 #include <linux/debugfs.h>
7
8 #include "gt/intel_gt.h"
9 #include "i915_drv.h"
10 #include "i915_memcpy.h"
11 #include "intel_guc_log.h"
12
13 static void guc_log_capture_logs(struct intel_guc_log *log);
14
15 /**
16 * DOC: GuC firmware log
17 *
18 * Firmware log is enabled by setting i915.guc_log_level to the positive level.
19 * Log data is printed out via reading debugfs i915_guc_log_dump. Reading from
20 * i915_guc_load_status will print out firmware loading status and scratch
21 * registers value.
22 */
23
guc_action_flush_log_complete(struct intel_guc * guc)24 static int guc_action_flush_log_complete(struct intel_guc *guc)
25 {
26 u32 action[] = {
27 INTEL_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE
28 };
29
30 return intel_guc_send(guc, action, ARRAY_SIZE(action));
31 }
32
guc_action_flush_log(struct intel_guc * guc)33 static int guc_action_flush_log(struct intel_guc *guc)
34 {
35 u32 action[] = {
36 INTEL_GUC_ACTION_FORCE_LOG_BUFFER_FLUSH,
37 0
38 };
39
40 return intel_guc_send(guc, action, ARRAY_SIZE(action));
41 }
42
guc_action_control_log(struct intel_guc * guc,bool enable,bool default_logging,u32 verbosity)43 static int guc_action_control_log(struct intel_guc *guc, bool enable,
44 bool default_logging, u32 verbosity)
45 {
46 u32 action[] = {
47 INTEL_GUC_ACTION_UK_LOG_ENABLE_LOGGING,
48 (enable ? GUC_LOG_CONTROL_LOGGING_ENABLED : 0) |
49 (verbosity << GUC_LOG_CONTROL_VERBOSITY_SHIFT) |
50 (default_logging ? GUC_LOG_CONTROL_DEFAULT_LOGGING : 0)
51 };
52
53 GEM_BUG_ON(verbosity > GUC_LOG_VERBOSITY_MAX);
54
55 return intel_guc_send(guc, action, ARRAY_SIZE(action));
56 }
57
log_to_guc(struct intel_guc_log * log)58 static inline struct intel_guc *log_to_guc(struct intel_guc_log *log)
59 {
60 return container_of(log, struct intel_guc, log);
61 }
62
guc_log_enable_flush_events(struct intel_guc_log * log)63 static void guc_log_enable_flush_events(struct intel_guc_log *log)
64 {
65 intel_guc_enable_msg(log_to_guc(log),
66 INTEL_GUC_RECV_MSG_FLUSH_LOG_BUFFER |
67 INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED);
68 }
69
guc_log_disable_flush_events(struct intel_guc_log * log)70 static void guc_log_disable_flush_events(struct intel_guc_log *log)
71 {
72 intel_guc_disable_msg(log_to_guc(log),
73 INTEL_GUC_RECV_MSG_FLUSH_LOG_BUFFER |
74 INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED);
75 }
76
77 /*
78 * Sub buffer switch callback. Called whenever relay has to switch to a new
79 * sub buffer, relay stays on the same sub buffer if 0 is returned.
80 */
subbuf_start_callback(struct rchan_buf * buf,void * subbuf,void * prev_subbuf,size_t prev_padding)81 static int subbuf_start_callback(struct rchan_buf *buf,
82 void *subbuf,
83 void *prev_subbuf,
84 size_t prev_padding)
85 {
86 /*
87 * Use no-overwrite mode by default, where relay will stop accepting
88 * new data if there are no empty sub buffers left.
89 * There is no strict synchronization enforced by relay between Consumer
90 * and Producer. In overwrite mode, there is a possibility of getting
91 * inconsistent/garbled data, the producer could be writing on to the
92 * same sub buffer from which Consumer is reading. This can't be avoided
93 * unless Consumer is fast enough and can always run in tandem with
94 * Producer.
95 */
96 if (relay_buf_full(buf))
97 return 0;
98
99 return 1;
100 }
101
102 /*
103 * file_create() callback. Creates relay file in debugfs.
104 */
create_buf_file_callback(const char * filename,struct dentry * parent,umode_t mode,struct rchan_buf * buf,int * is_global)105 static struct dentry *create_buf_file_callback(const char *filename,
106 struct dentry *parent,
107 umode_t mode,
108 struct rchan_buf *buf,
109 int *is_global)
110 {
111 struct dentry *buf_file;
112
113 /*
114 * This to enable the use of a single buffer for the relay channel and
115 * correspondingly have a single file exposed to User, through which
116 * it can collect the logs in order without any post-processing.
117 * Need to set 'is_global' even if parent is NULL for early logging.
118 */
119 *is_global = 1;
120
121 if (!parent)
122 return NULL;
123
124 buf_file = debugfs_create_file(filename, mode,
125 parent, buf, &relay_file_operations);
126 if (IS_ERR(buf_file))
127 return NULL;
128
129 return buf_file;
130 }
131
132 /*
133 * file_remove() default callback. Removes relay file in debugfs.
134 */
remove_buf_file_callback(struct dentry * dentry)135 static int remove_buf_file_callback(struct dentry *dentry)
136 {
137 debugfs_remove(dentry);
138 return 0;
139 }
140
141 /* relay channel callbacks */
142 static struct rchan_callbacks relay_callbacks = {
143 .subbuf_start = subbuf_start_callback,
144 .create_buf_file = create_buf_file_callback,
145 .remove_buf_file = remove_buf_file_callback,
146 };
147
guc_move_to_next_buf(struct intel_guc_log * log)148 static void guc_move_to_next_buf(struct intel_guc_log *log)
149 {
150 /*
151 * Make sure the updates made in the sub buffer are visible when
152 * Consumer sees the following update to offset inside the sub buffer.
153 */
154 smp_wmb();
155
156 /* All data has been written, so now move the offset of sub buffer. */
157 relay_reserve(log->relay.channel, log->vma->obj->base.size);
158
159 /* Switch to the next sub buffer */
160 relay_flush(log->relay.channel);
161 }
162
guc_get_write_buffer(struct intel_guc_log * log)163 static void *guc_get_write_buffer(struct intel_guc_log *log)
164 {
165 /*
166 * Just get the base address of a new sub buffer and copy data into it
167 * ourselves. NULL will be returned in no-overwrite mode, if all sub
168 * buffers are full. Could have used the relay_write() to indirectly
169 * copy the data, but that would have been bit convoluted, as we need to
170 * write to only certain locations inside a sub buffer which cannot be
171 * done without using relay_reserve() along with relay_write(). So its
172 * better to use relay_reserve() alone.
173 */
174 return relay_reserve(log->relay.channel, 0);
175 }
176
guc_check_log_buf_overflow(struct intel_guc_log * log,enum guc_log_buffer_type type,unsigned int full_cnt)177 static bool guc_check_log_buf_overflow(struct intel_guc_log *log,
178 enum guc_log_buffer_type type,
179 unsigned int full_cnt)
180 {
181 unsigned int prev_full_cnt = log->stats[type].sampled_overflow;
182 bool overflow = false;
183
184 if (full_cnt != prev_full_cnt) {
185 overflow = true;
186
187 log->stats[type].overflow = full_cnt;
188 log->stats[type].sampled_overflow += full_cnt - prev_full_cnt;
189
190 if (full_cnt < prev_full_cnt) {
191 /* buffer_full_cnt is a 4 bit counter */
192 log->stats[type].sampled_overflow += 16;
193 }
194
195 dev_notice_ratelimited(guc_to_gt(log_to_guc(log))->i915->drm.dev,
196 "GuC log buffer overflow\n");
197 }
198
199 return overflow;
200 }
201
guc_get_log_buffer_size(enum guc_log_buffer_type type)202 static unsigned int guc_get_log_buffer_size(enum guc_log_buffer_type type)
203 {
204 switch (type) {
205 case GUC_ISR_LOG_BUFFER:
206 return ISR_BUFFER_SIZE;
207 case GUC_DPC_LOG_BUFFER:
208 return DPC_BUFFER_SIZE;
209 case GUC_CRASH_DUMP_LOG_BUFFER:
210 return CRASH_BUFFER_SIZE;
211 default:
212 MISSING_CASE(type);
213 }
214
215 return 0;
216 }
217
guc_read_update_log_buffer(struct intel_guc_log * log)218 static void guc_read_update_log_buffer(struct intel_guc_log *log)
219 {
220 unsigned int buffer_size, read_offset, write_offset, bytes_to_copy, full_cnt;
221 struct guc_log_buffer_state *log_buf_state, *log_buf_snapshot_state;
222 struct guc_log_buffer_state log_buf_state_local;
223 enum guc_log_buffer_type type;
224 void *src_data, *dst_data;
225 bool new_overflow;
226
227 mutex_lock(&log->relay.lock);
228
229 if (WARN_ON(!intel_guc_log_relay_enabled(log)))
230 goto out_unlock;
231
232 /* Get the pointer to shared GuC log buffer */
233 log_buf_state = src_data = log->relay.buf_addr;
234
235 /* Get the pointer to local buffer to store the logs */
236 log_buf_snapshot_state = dst_data = guc_get_write_buffer(log);
237
238 if (unlikely(!log_buf_snapshot_state)) {
239 /*
240 * Used rate limited to avoid deluge of messages, logs might be
241 * getting consumed by User at a slow rate.
242 */
243 DRM_ERROR_RATELIMITED("no sub-buffer to capture logs\n");
244 log->relay.full_count++;
245
246 goto out_unlock;
247 }
248
249 /* Actual logs are present from the 2nd page */
250 src_data += PAGE_SIZE;
251 dst_data += PAGE_SIZE;
252
253 for (type = GUC_ISR_LOG_BUFFER; type < GUC_MAX_LOG_BUFFER; type++) {
254 /*
255 * Make a copy of the state structure, inside GuC log buffer
256 * (which is uncached mapped), on the stack to avoid reading
257 * from it multiple times.
258 */
259 memcpy(&log_buf_state_local, log_buf_state,
260 sizeof(struct guc_log_buffer_state));
261 buffer_size = guc_get_log_buffer_size(type);
262 read_offset = log_buf_state_local.read_ptr;
263 write_offset = log_buf_state_local.sampled_write_ptr;
264 full_cnt = log_buf_state_local.buffer_full_cnt;
265
266 /* Bookkeeping stuff */
267 log->stats[type].flush += log_buf_state_local.flush_to_file;
268 new_overflow = guc_check_log_buf_overflow(log, type, full_cnt);
269
270 /* Update the state of shared log buffer */
271 log_buf_state->read_ptr = write_offset;
272 log_buf_state->flush_to_file = 0;
273 log_buf_state++;
274
275 /* First copy the state structure in snapshot buffer */
276 memcpy(log_buf_snapshot_state, &log_buf_state_local,
277 sizeof(struct guc_log_buffer_state));
278
279 /*
280 * The write pointer could have been updated by GuC firmware,
281 * after sending the flush interrupt to Host, for consistency
282 * set write pointer value to same value of sampled_write_ptr
283 * in the snapshot buffer.
284 */
285 log_buf_snapshot_state->write_ptr = write_offset;
286 log_buf_snapshot_state++;
287
288 /* Now copy the actual logs. */
289 if (unlikely(new_overflow)) {
290 /* copy the whole buffer in case of overflow */
291 read_offset = 0;
292 write_offset = buffer_size;
293 } else if (unlikely((read_offset > buffer_size) ||
294 (write_offset > buffer_size))) {
295 DRM_ERROR("invalid log buffer state\n");
296 /* copy whole buffer as offsets are unreliable */
297 read_offset = 0;
298 write_offset = buffer_size;
299 }
300
301 /* Just copy the newly written data */
302 if (read_offset > write_offset) {
303 i915_memcpy_from_wc(dst_data, src_data, write_offset);
304 bytes_to_copy = buffer_size - read_offset;
305 } else {
306 bytes_to_copy = write_offset - read_offset;
307 }
308 i915_memcpy_from_wc(dst_data + read_offset,
309 src_data + read_offset, bytes_to_copy);
310
311 src_data += buffer_size;
312 dst_data += buffer_size;
313 }
314
315 guc_move_to_next_buf(log);
316
317 out_unlock:
318 mutex_unlock(&log->relay.lock);
319 }
320
capture_logs_work(struct work_struct * work)321 static void capture_logs_work(struct work_struct *work)
322 {
323 struct intel_guc_log *log =
324 container_of(work, struct intel_guc_log, relay.flush_work);
325
326 guc_log_capture_logs(log);
327 }
328
guc_log_map(struct intel_guc_log * log)329 static int guc_log_map(struct intel_guc_log *log)
330 {
331 void *vaddr;
332
333 lockdep_assert_held(&log->relay.lock);
334
335 if (!log->vma)
336 return -ENODEV;
337
338 /*
339 * Create a WC (Uncached for read) vmalloc mapping of log
340 * buffer pages, so that we can directly get the data
341 * (up-to-date) from memory.
342 */
343 vaddr = i915_gem_object_pin_map(log->vma->obj, I915_MAP_WC);
344 if (IS_ERR(vaddr))
345 return PTR_ERR(vaddr);
346
347 log->relay.buf_addr = vaddr;
348
349 return 0;
350 }
351
guc_log_unmap(struct intel_guc_log * log)352 static void guc_log_unmap(struct intel_guc_log *log)
353 {
354 lockdep_assert_held(&log->relay.lock);
355
356 i915_gem_object_unpin_map(log->vma->obj);
357 log->relay.buf_addr = NULL;
358 }
359
intel_guc_log_init_early(struct intel_guc_log * log)360 void intel_guc_log_init_early(struct intel_guc_log *log)
361 {
362 mutex_init(&log->relay.lock);
363 INIT_WORK(&log->relay.flush_work, capture_logs_work);
364 }
365
guc_log_relay_create(struct intel_guc_log * log)366 static int guc_log_relay_create(struct intel_guc_log *log)
367 {
368 struct intel_guc *guc = log_to_guc(log);
369 struct drm_i915_private *dev_priv = guc_to_gt(guc)->i915;
370 struct rchan *guc_log_relay_chan;
371 size_t n_subbufs, subbuf_size;
372 int ret;
373
374 lockdep_assert_held(&log->relay.lock);
375 GEM_BUG_ON(!log->vma);
376
377 /* Keep the size of sub buffers same as shared log buffer */
378 subbuf_size = log->vma->size;
379
380 /*
381 * Store up to 8 snapshots, which is large enough to buffer sufficient
382 * boot time logs and provides enough leeway to User, in terms of
383 * latency, for consuming the logs from relay. Also doesn't take
384 * up too much memory.
385 */
386 n_subbufs = 8;
387
388 guc_log_relay_chan = relay_open("guc_log",
389 dev_priv->drm.primary->debugfs_root,
390 subbuf_size, n_subbufs,
391 &relay_callbacks, dev_priv);
392 if (!guc_log_relay_chan) {
393 DRM_ERROR("Couldn't create relay chan for GuC logging\n");
394
395 ret = -ENOMEM;
396 return ret;
397 }
398
399 GEM_BUG_ON(guc_log_relay_chan->subbuf_size < subbuf_size);
400 log->relay.channel = guc_log_relay_chan;
401
402 return 0;
403 }
404
guc_log_relay_destroy(struct intel_guc_log * log)405 static void guc_log_relay_destroy(struct intel_guc_log *log)
406 {
407 lockdep_assert_held(&log->relay.lock);
408
409 relay_close(log->relay.channel);
410 log->relay.channel = NULL;
411 }
412
guc_log_capture_logs(struct intel_guc_log * log)413 static void guc_log_capture_logs(struct intel_guc_log *log)
414 {
415 struct intel_guc *guc = log_to_guc(log);
416 struct drm_i915_private *dev_priv = guc_to_gt(guc)->i915;
417 intel_wakeref_t wakeref;
418
419 guc_read_update_log_buffer(log);
420
421 /*
422 * Generally device is expected to be active only at this
423 * time, so get/put should be really quick.
424 */
425 with_intel_runtime_pm(&dev_priv->runtime_pm, wakeref)
426 guc_action_flush_log_complete(guc);
427 }
428
__get_default_log_level(struct intel_guc_log * log)429 static u32 __get_default_log_level(struct intel_guc_log *log)
430 {
431 /* A negative value means "use platform/config default" */
432 if (i915_modparams.guc_log_level < 0) {
433 return (IS_ENABLED(CONFIG_DRM_I915_DEBUG) ||
434 IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) ?
435 GUC_LOG_LEVEL_MAX : GUC_LOG_LEVEL_NON_VERBOSE;
436 }
437
438 if (i915_modparams.guc_log_level > GUC_LOG_LEVEL_MAX) {
439 DRM_WARN("Incompatible option detected: %s=%d, %s!\n",
440 "guc_log_level", i915_modparams.guc_log_level,
441 "verbosity too high");
442 return (IS_ENABLED(CONFIG_DRM_I915_DEBUG) ||
443 IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) ?
444 GUC_LOG_LEVEL_MAX : GUC_LOG_LEVEL_DISABLED;
445 }
446
447 GEM_BUG_ON(i915_modparams.guc_log_level < GUC_LOG_LEVEL_DISABLED);
448 GEM_BUG_ON(i915_modparams.guc_log_level > GUC_LOG_LEVEL_MAX);
449 return i915_modparams.guc_log_level;
450 }
451
intel_guc_log_create(struct intel_guc_log * log)452 int intel_guc_log_create(struct intel_guc_log *log)
453 {
454 struct intel_guc *guc = log_to_guc(log);
455 struct i915_vma *vma;
456 u32 guc_log_size;
457 int ret;
458
459 GEM_BUG_ON(log->vma);
460
461 /*
462 * GuC Log buffer Layout
463 *
464 * +===============================+ 00B
465 * | Crash dump state header |
466 * +-------------------------------+ 32B
467 * | DPC state header |
468 * +-------------------------------+ 64B
469 * | ISR state header |
470 * +-------------------------------+ 96B
471 * | |
472 * +===============================+ PAGE_SIZE (4KB)
473 * | Crash Dump logs |
474 * +===============================+ + CRASH_SIZE
475 * | DPC logs |
476 * +===============================+ + DPC_SIZE
477 * | ISR logs |
478 * +===============================+ + ISR_SIZE
479 */
480 guc_log_size = PAGE_SIZE + CRASH_BUFFER_SIZE + DPC_BUFFER_SIZE +
481 ISR_BUFFER_SIZE;
482
483 vma = intel_guc_allocate_vma(guc, guc_log_size);
484 if (IS_ERR(vma)) {
485 ret = PTR_ERR(vma);
486 goto err;
487 }
488
489 log->vma = vma;
490
491 log->level = __get_default_log_level(log);
492 DRM_DEBUG_DRIVER("guc_log_level=%d (%s, verbose:%s, verbosity:%d)\n",
493 log->level, enableddisabled(log->level),
494 yesno(GUC_LOG_LEVEL_IS_VERBOSE(log->level)),
495 GUC_LOG_LEVEL_TO_VERBOSITY(log->level));
496
497 return 0;
498
499 err:
500 DRM_ERROR("Failed to allocate GuC log buffer. %d\n", ret);
501 return ret;
502 }
503
intel_guc_log_destroy(struct intel_guc_log * log)504 void intel_guc_log_destroy(struct intel_guc_log *log)
505 {
506 i915_vma_unpin_and_release(&log->vma, 0);
507 }
508
intel_guc_log_set_level(struct intel_guc_log * log,u32 level)509 int intel_guc_log_set_level(struct intel_guc_log *log, u32 level)
510 {
511 struct intel_guc *guc = log_to_guc(log);
512 struct drm_i915_private *dev_priv = guc_to_gt(guc)->i915;
513 intel_wakeref_t wakeref;
514 int ret = 0;
515
516 BUILD_BUG_ON(GUC_LOG_VERBOSITY_MIN != 0);
517 GEM_BUG_ON(!log->vma);
518
519 /*
520 * GuC is recognizing log levels starting from 0 to max, we're using 0
521 * as indication that logging should be disabled.
522 */
523 if (level < GUC_LOG_LEVEL_DISABLED || level > GUC_LOG_LEVEL_MAX)
524 return -EINVAL;
525
526 mutex_lock(&dev_priv->drm.struct_mutex);
527
528 if (log->level == level)
529 goto out_unlock;
530
531 with_intel_runtime_pm(&dev_priv->runtime_pm, wakeref)
532 ret = guc_action_control_log(guc,
533 GUC_LOG_LEVEL_IS_VERBOSE(level),
534 GUC_LOG_LEVEL_IS_ENABLED(level),
535 GUC_LOG_LEVEL_TO_VERBOSITY(level));
536 if (ret) {
537 DRM_DEBUG_DRIVER("guc_log_control action failed %d\n", ret);
538 goto out_unlock;
539 }
540
541 log->level = level;
542
543 out_unlock:
544 mutex_unlock(&dev_priv->drm.struct_mutex);
545
546 return ret;
547 }
548
intel_guc_log_relay_enabled(const struct intel_guc_log * log)549 bool intel_guc_log_relay_enabled(const struct intel_guc_log *log)
550 {
551 return log->relay.buf_addr;
552 }
553
intel_guc_log_relay_open(struct intel_guc_log * log)554 int intel_guc_log_relay_open(struct intel_guc_log *log)
555 {
556 int ret;
557
558 if (!log->vma)
559 return -ENODEV;
560
561 mutex_lock(&log->relay.lock);
562
563 if (intel_guc_log_relay_enabled(log)) {
564 ret = -EEXIST;
565 goto out_unlock;
566 }
567
568 /*
569 * We require SSE 4.1 for fast reads from the GuC log buffer and
570 * it should be present on the chipsets supporting GuC based
571 * submisssions.
572 */
573 if (!i915_has_memcpy_from_wc()) {
574 ret = -ENXIO;
575 goto out_unlock;
576 }
577
578 ret = guc_log_relay_create(log);
579 if (ret)
580 goto out_unlock;
581
582 ret = guc_log_map(log);
583 if (ret)
584 goto out_relay;
585
586 mutex_unlock(&log->relay.lock);
587
588 guc_log_enable_flush_events(log);
589
590 /*
591 * When GuC is logging without us relaying to userspace, we're ignoring
592 * the flush notification. This means that we need to unconditionally
593 * flush on relay enabling, since GuC only notifies us once.
594 */
595 queue_work(system_highpri_wq, &log->relay.flush_work);
596
597 return 0;
598
599 out_relay:
600 guc_log_relay_destroy(log);
601 out_unlock:
602 mutex_unlock(&log->relay.lock);
603
604 return ret;
605 }
606
intel_guc_log_relay_flush(struct intel_guc_log * log)607 void intel_guc_log_relay_flush(struct intel_guc_log *log)
608 {
609 struct intel_guc *guc = log_to_guc(log);
610 struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
611 intel_wakeref_t wakeref;
612
613 /*
614 * Before initiating the forceful flush, wait for any pending/ongoing
615 * flush to complete otherwise forceful flush may not actually happen.
616 */
617 flush_work(&log->relay.flush_work);
618
619 with_intel_runtime_pm(&i915->runtime_pm, wakeref)
620 guc_action_flush_log(guc);
621
622 /* GuC would have updated log buffer by now, so capture it */
623 guc_log_capture_logs(log);
624 }
625
intel_guc_log_relay_close(struct intel_guc_log * log)626 void intel_guc_log_relay_close(struct intel_guc_log *log)
627 {
628 struct intel_guc *guc = log_to_guc(log);
629 struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
630
631 guc_log_disable_flush_events(log);
632 intel_synchronize_irq(i915);
633
634 flush_work(&log->relay.flush_work);
635
636 mutex_lock(&log->relay.lock);
637 GEM_BUG_ON(!intel_guc_log_relay_enabled(log));
638 guc_log_unmap(log);
639 guc_log_relay_destroy(log);
640 mutex_unlock(&log->relay.lock);
641 }
642
intel_guc_log_handle_flush_event(struct intel_guc_log * log)643 void intel_guc_log_handle_flush_event(struct intel_guc_log *log)
644 {
645 queue_work(system_highpri_wq, &log->relay.flush_work);
646 }
647