blob: 5c0f9a49da0e9ef6d3e3a2dc119ab8ddcc67e5d0 [file] [log] [blame]
Michal Wajdeczkof9cda042017-01-13 17:41:57 +00001/*
2 * Copyright © 2014-2017 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 */
24#include <linux/debugfs.h>
25#include <linux/relay.h>
26#include "i915_drv.h"
27
28static void guc_log_capture_logs(struct intel_guc *guc);
29
30/**
31 * DOC: GuC firmware log
32 *
33 * Firmware log is enabled by setting i915.guc_log_level to non-negative level.
34 * Log data is printed out via reading debugfs i915_guc_log_dump. Reading from
35 * i915_guc_load_status will print out firmware loading status and scratch
36 * registers value.
37 *
38 */
39
40static int guc_log_flush_complete(struct intel_guc *guc)
41{
42 u32 action[] = {
43 INTEL_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE
44 };
45
46 return intel_guc_send(guc, action, ARRAY_SIZE(action));
47}
48
49static int guc_log_flush(struct intel_guc *guc)
50{
51 u32 action[] = {
52 INTEL_GUC_ACTION_FORCE_LOG_BUFFER_FLUSH,
53 0
54 };
55
56 return intel_guc_send(guc, action, ARRAY_SIZE(action));
57}
58
59static int guc_log_control(struct intel_guc *guc, u32 control_val)
60{
61 u32 action[] = {
62 INTEL_GUC_ACTION_UK_LOG_ENABLE_LOGGING,
63 control_val
64 };
65
66 return intel_guc_send(guc, action, ARRAY_SIZE(action));
67}
68
69
70/*
71 * Sub buffer switch callback. Called whenever relay has to switch to a new
72 * sub buffer, relay stays on the same sub buffer if 0 is returned.
73 */
74static int subbuf_start_callback(struct rchan_buf *buf,
75 void *subbuf,
76 void *prev_subbuf,
77 size_t prev_padding)
78{
79 /* Use no-overwrite mode by default, where relay will stop accepting
80 * new data if there are no empty sub buffers left.
81 * There is no strict synchronization enforced by relay between Consumer
82 * and Producer. In overwrite mode, there is a possibility of getting
83 * inconsistent/garbled data, the producer could be writing on to the
84 * same sub buffer from which Consumer is reading. This can't be avoided
85 * unless Consumer is fast enough and can always run in tandem with
86 * Producer.
87 */
88 if (relay_buf_full(buf))
89 return 0;
90
91 return 1;
92}
93
94/*
95 * file_create() callback. Creates relay file in debugfs.
96 */
97static struct dentry *create_buf_file_callback(const char *filename,
98 struct dentry *parent,
99 umode_t mode,
100 struct rchan_buf *buf,
101 int *is_global)
102{
103 struct dentry *buf_file;
104
105 /* This to enable the use of a single buffer for the relay channel and
106 * correspondingly have a single file exposed to User, through which
107 * it can collect the logs in order without any post-processing.
108 * Need to set 'is_global' even if parent is NULL for early logging.
109 */
110 *is_global = 1;
111
112 if (!parent)
113 return NULL;
114
115 /* Not using the channel filename passed as an argument, since for each
116 * channel relay appends the corresponding CPU number to the filename
117 * passed in relay_open(). This should be fine as relay just needs a
118 * dentry of the file associated with the channel buffer and that file's
119 * name need not be same as the filename passed as an argument.
120 */
121 buf_file = debugfs_create_file("guc_log", mode,
122 parent, buf, &relay_file_operations);
123 return buf_file;
124}
125
126/*
127 * file_remove() default callback. Removes relay file in debugfs.
128 */
129static int remove_buf_file_callback(struct dentry *dentry)
130{
131 debugfs_remove(dentry);
132 return 0;
133}
134
135/* relay channel callbacks */
136static struct rchan_callbacks relay_callbacks = {
137 .subbuf_start = subbuf_start_callback,
138 .create_buf_file = create_buf_file_callback,
139 .remove_buf_file = remove_buf_file_callback,
140};
141
142static void guc_log_remove_relay_file(struct intel_guc *guc)
143{
144 relay_close(guc->log.relay_chan);
145}
146
147static int guc_log_create_relay_channel(struct intel_guc *guc)
148{
149 struct drm_i915_private *dev_priv = guc_to_i915(guc);
150 struct rchan *guc_log_relay_chan;
151 size_t n_subbufs, subbuf_size;
152
153 /* Keep the size of sub buffers same as shared log buffer */
154 subbuf_size = guc->log.vma->obj->base.size;
155
156 /* Store up to 8 snapshots, which is large enough to buffer sufficient
157 * boot time logs and provides enough leeway to User, in terms of
158 * latency, for consuming the logs from relay. Also doesn't take
159 * up too much memory.
160 */
161 n_subbufs = 8;
162
163 guc_log_relay_chan = relay_open(NULL, NULL, subbuf_size,
164 n_subbufs, &relay_callbacks, dev_priv);
165 if (!guc_log_relay_chan) {
166 DRM_ERROR("Couldn't create relay chan for GuC logging\n");
167 return -ENOMEM;
168 }
169
170 GEM_BUG_ON(guc_log_relay_chan->subbuf_size < subbuf_size);
171 guc->log.relay_chan = guc_log_relay_chan;
172 return 0;
173}
174
175static int guc_log_create_relay_file(struct intel_guc *guc)
176{
177 struct drm_i915_private *dev_priv = guc_to_i915(guc);
178 struct dentry *log_dir;
179 int ret;
180
181 /* For now create the log file in /sys/kernel/debug/dri/0 dir */
182 log_dir = dev_priv->drm.primary->debugfs_root;
183
184 /* If /sys/kernel/debug/dri/0 location do not exist, then debugfs is
185 * not mounted and so can't create the relay file.
186 * The relay API seems to fit well with debugfs only, for availing relay
187 * there are 3 requirements which can be met for debugfs file only in a
188 * straightforward/clean manner :-
189 * i) Need the associated dentry pointer of the file, while opening the
190 * relay channel.
191 * ii) Should be able to use 'relay_file_operations' fops for the file.
192 * iii) Set the 'i_private' field of file's inode to the pointer of
193 * relay channel buffer.
194 */
195 if (!log_dir) {
196 DRM_ERROR("Debugfs dir not available yet for GuC log file\n");
197 return -ENODEV;
198 }
199
200 ret = relay_late_setup_files(guc->log.relay_chan, "guc_log", log_dir);
201 if (ret) {
202 DRM_ERROR("Couldn't associate relay chan with file %d\n", ret);
203 return ret;
204 }
205
206 return 0;
207}
208
209static void guc_move_to_next_buf(struct intel_guc *guc)
210{
211 /* Make sure the updates made in the sub buffer are visible when
212 * Consumer sees the following update to offset inside the sub buffer.
213 */
214 smp_wmb();
215
216 /* All data has been written, so now move the offset of sub buffer. */
217 relay_reserve(guc->log.relay_chan, guc->log.vma->obj->base.size);
218
219 /* Switch to the next sub buffer */
220 relay_flush(guc->log.relay_chan);
221}
222
223static void *guc_get_write_buffer(struct intel_guc *guc)
224{
225 if (!guc->log.relay_chan)
226 return NULL;
227
228 /* Just get the base address of a new sub buffer and copy data into it
229 * ourselves. NULL will be returned in no-overwrite mode, if all sub
230 * buffers are full. Could have used the relay_write() to indirectly
231 * copy the data, but that would have been bit convoluted, as we need to
232 * write to only certain locations inside a sub buffer which cannot be
233 * done without using relay_reserve() along with relay_write(). So its
234 * better to use relay_reserve() alone.
235 */
236 return relay_reserve(guc->log.relay_chan, 0);
237}
238
239static bool guc_check_log_buf_overflow(struct intel_guc *guc,
240 enum guc_log_buffer_type type,
241 unsigned int full_cnt)
242{
243 unsigned int prev_full_cnt = guc->log.prev_overflow_count[type];
244 bool overflow = false;
245
246 if (full_cnt != prev_full_cnt) {
247 overflow = true;
248
249 guc->log.prev_overflow_count[type] = full_cnt;
250 guc->log.total_overflow_count[type] += full_cnt - prev_full_cnt;
251
252 if (full_cnt < prev_full_cnt) {
253 /* buffer_full_cnt is a 4 bit counter */
254 guc->log.total_overflow_count[type] += 16;
255 }
256 DRM_ERROR_RATELIMITED("GuC log buffer overflow\n");
257 }
258
259 return overflow;
260}
261
262static unsigned int guc_get_log_buffer_size(enum guc_log_buffer_type type)
263{
264 switch (type) {
265 case GUC_ISR_LOG_BUFFER:
266 return (GUC_LOG_ISR_PAGES + 1) * PAGE_SIZE;
267 case GUC_DPC_LOG_BUFFER:
268 return (GUC_LOG_DPC_PAGES + 1) * PAGE_SIZE;
269 case GUC_CRASH_DUMP_LOG_BUFFER:
270 return (GUC_LOG_CRASH_PAGES + 1) * PAGE_SIZE;
271 default:
272 MISSING_CASE(type);
273 }
274
275 return 0;
276}
277
278static void guc_read_update_log_buffer(struct intel_guc *guc)
279{
280 unsigned int buffer_size, read_offset, write_offset, bytes_to_copy, full_cnt;
281 struct guc_log_buffer_state *log_buf_state, *log_buf_snapshot_state;
282 struct guc_log_buffer_state log_buf_state_local;
283 enum guc_log_buffer_type type;
284 void *src_data, *dst_data;
285 bool new_overflow;
286
287 if (WARN_ON(!guc->log.buf_addr))
288 return;
289
290 /* Get the pointer to shared GuC log buffer */
291 log_buf_state = src_data = guc->log.buf_addr;
292
293 /* Get the pointer to local buffer to store the logs */
294 log_buf_snapshot_state = dst_data = guc_get_write_buffer(guc);
295
296 /* Actual logs are present from the 2nd page */
297 src_data += PAGE_SIZE;
298 dst_data += PAGE_SIZE;
299
300 for (type = GUC_ISR_LOG_BUFFER; type < GUC_MAX_LOG_BUFFER; type++) {
301 /* Make a copy of the state structure, inside GuC log buffer
302 * (which is uncached mapped), on the stack to avoid reading
303 * from it multiple times.
304 */
305 memcpy(&log_buf_state_local, log_buf_state,
306 sizeof(struct guc_log_buffer_state));
307 buffer_size = guc_get_log_buffer_size(type);
308 read_offset = log_buf_state_local.read_ptr;
309 write_offset = log_buf_state_local.sampled_write_ptr;
310 full_cnt = log_buf_state_local.buffer_full_cnt;
311
312 /* Bookkeeping stuff */
313 guc->log.flush_count[type] += log_buf_state_local.flush_to_file;
314 new_overflow = guc_check_log_buf_overflow(guc, type, full_cnt);
315
316 /* Update the state of shared log buffer */
317 log_buf_state->read_ptr = write_offset;
318 log_buf_state->flush_to_file = 0;
319 log_buf_state++;
320
321 if (unlikely(!log_buf_snapshot_state))
322 continue;
323
324 /* First copy the state structure in snapshot buffer */
325 memcpy(log_buf_snapshot_state, &log_buf_state_local,
326 sizeof(struct guc_log_buffer_state));
327
328 /* The write pointer could have been updated by GuC firmware,
329 * after sending the flush interrupt to Host, for consistency
330 * set write pointer value to same value of sampled_write_ptr
331 * in the snapshot buffer.
332 */
333 log_buf_snapshot_state->write_ptr = write_offset;
334 log_buf_snapshot_state++;
335
336 /* Now copy the actual logs. */
337 if (unlikely(new_overflow)) {
338 /* copy the whole buffer in case of overflow */
339 read_offset = 0;
340 write_offset = buffer_size;
341 } else if (unlikely((read_offset > buffer_size) ||
342 (write_offset > buffer_size))) {
343 DRM_ERROR("invalid log buffer state\n");
344 /* copy whole buffer as offsets are unreliable */
345 read_offset = 0;
346 write_offset = buffer_size;
347 }
348
349 /* Just copy the newly written data */
350 if (read_offset > write_offset) {
351 i915_memcpy_from_wc(dst_data, src_data, write_offset);
352 bytes_to_copy = buffer_size - read_offset;
353 } else {
354 bytes_to_copy = write_offset - read_offset;
355 }
356 i915_memcpy_from_wc(dst_data + read_offset,
357 src_data + read_offset, bytes_to_copy);
358
359 src_data += buffer_size;
360 dst_data += buffer_size;
361 }
362
363 if (log_buf_snapshot_state)
364 guc_move_to_next_buf(guc);
365 else {
366 /* Used rate limited to avoid deluge of messages, logs might be
367 * getting consumed by User at a slow rate.
368 */
369 DRM_ERROR_RATELIMITED("no sub-buffer to capture logs\n");
370 guc->log.capture_miss_count++;
371 }
372}
373
374static void guc_log_cleanup(struct intel_guc *guc)
375{
376 struct drm_i915_private *dev_priv = guc_to_i915(guc);
377
378 lockdep_assert_held(&dev_priv->drm.struct_mutex);
379
380 /* First disable the flush interrupt */
381 gen9_disable_guc_interrupts(dev_priv);
382
383 if (guc->log.flush_wq)
384 destroy_workqueue(guc->log.flush_wq);
385
386 guc->log.flush_wq = NULL;
387
388 if (guc->log.relay_chan)
389 guc_log_remove_relay_file(guc);
390
391 guc->log.relay_chan = NULL;
392
393 if (guc->log.buf_addr)
394 i915_gem_object_unpin_map(guc->log.vma->obj);
395
396 guc->log.buf_addr = NULL;
397}
398
399static void capture_logs_work(struct work_struct *work)
400{
401 struct intel_guc *guc =
402 container_of(work, struct intel_guc, log.flush_work);
403
404 guc_log_capture_logs(guc);
405}
406
407static int guc_log_create_extras(struct intel_guc *guc)
408{
409 struct drm_i915_private *dev_priv = guc_to_i915(guc);
410 void *vaddr;
411 int ret;
412
413 lockdep_assert_held(&dev_priv->drm.struct_mutex);
414
415 /* Nothing to do */
416 if (i915.guc_log_level < 0)
417 return 0;
418
419 if (!guc->log.buf_addr) {
420 /* Create a WC (Uncached for read) vmalloc mapping of log
421 * buffer pages, so that we can directly get the data
422 * (up-to-date) from memory.
423 */
424 vaddr = i915_gem_object_pin_map(guc->log.vma->obj, I915_MAP_WC);
425 if (IS_ERR(vaddr)) {
426 ret = PTR_ERR(vaddr);
427 DRM_ERROR("Couldn't map log buffer pages %d\n", ret);
428 return ret;
429 }
430
431 guc->log.buf_addr = vaddr;
432 }
433
434 if (!guc->log.relay_chan) {
435 /* Create a relay channel, so that we have buffers for storing
436 * the GuC firmware logs, the channel will be linked with a file
437 * later on when debugfs is registered.
438 */
439 ret = guc_log_create_relay_channel(guc);
440 if (ret)
441 return ret;
442 }
443
444 if (!guc->log.flush_wq) {
445 INIT_WORK(&guc->log.flush_work, capture_logs_work);
446
447 /*
448 * GuC log buffer flush work item has to do register access to
449 * send the ack to GuC and this work item, if not synced before
450 * suspend, can potentially get executed after the GFX device is
451 * suspended.
452 * By marking the WQ as freezable, we don't have to bother about
453 * flushing of this work item from the suspend hooks, the pending
454 * work item if any will be either executed before the suspend
455 * or scheduled later on resume. This way the handling of work
456 * item can be kept same between system suspend & rpm suspend.
457 */
458 guc->log.flush_wq = alloc_ordered_workqueue("i915-guc_log",
459 WQ_HIGHPRI | WQ_FREEZABLE);
460 if (guc->log.flush_wq == NULL) {
461 DRM_ERROR("Couldn't allocate the wq for GuC logging\n");
462 return -ENOMEM;
463 }
464 }
465
466 return 0;
467}
468
469void intel_guc_log_create(struct intel_guc *guc)
470{
471 struct i915_vma *vma;
472 unsigned long offset;
473 uint32_t size, flags;
474
475 if (i915.guc_log_level > GUC_LOG_VERBOSITY_MAX)
476 i915.guc_log_level = GUC_LOG_VERBOSITY_MAX;
477
478 /* The first page is to save log buffer state. Allocate one
479 * extra page for others in case for overlap */
480 size = (1 + GUC_LOG_DPC_PAGES + 1 +
481 GUC_LOG_ISR_PAGES + 1 +
482 GUC_LOG_CRASH_PAGES + 1) << PAGE_SHIFT;
483
484 vma = guc->log.vma;
485 if (!vma) {
486 /* We require SSE 4.1 for fast reads from the GuC log buffer and
487 * it should be present on the chipsets supporting GuC based
488 * submisssions.
489 */
490 if (WARN_ON(!i915_has_memcpy_from_wc())) {
491 /* logging will not be enabled */
492 i915.guc_log_level = -1;
493 return;
494 }
495
496 vma = intel_guc_allocate_vma(guc, size);
497 if (IS_ERR(vma)) {
498 /* logging will be off */
499 i915.guc_log_level = -1;
500 return;
501 }
502
503 guc->log.vma = vma;
504
505 if (guc_log_create_extras(guc)) {
506 guc_log_cleanup(guc);
507 i915_vma_unpin_and_release(&guc->log.vma);
508 i915.guc_log_level = -1;
509 return;
510 }
511 }
512
513 /* each allocated unit is a page */
514 flags = GUC_LOG_VALID | GUC_LOG_NOTIFY_ON_HALF_FULL |
515 (GUC_LOG_DPC_PAGES << GUC_LOG_DPC_SHIFT) |
516 (GUC_LOG_ISR_PAGES << GUC_LOG_ISR_SHIFT) |
517 (GUC_LOG_CRASH_PAGES << GUC_LOG_CRASH_SHIFT);
518
519 offset = guc_ggtt_offset(vma) >> PAGE_SHIFT; /* in pages */
520 guc->log.flags = (offset << GUC_LOG_BUF_ADDR_SHIFT) | flags;
521}
522
523static int guc_log_late_setup(struct intel_guc *guc)
524{
525 struct drm_i915_private *dev_priv = guc_to_i915(guc);
526 int ret;
527
528 lockdep_assert_held(&dev_priv->drm.struct_mutex);
529
530 if (i915.guc_log_level < 0)
531 return -EINVAL;
532
533 /* If log_level was set as -1 at boot time, then setup needed to
534 * handle log buffer flush interrupts would not have been done yet,
535 * so do that now.
536 */
537 ret = guc_log_create_extras(guc);
538 if (ret)
539 goto err;
540
541 ret = guc_log_create_relay_file(guc);
542 if (ret)
543 goto err;
544
545 return 0;
546err:
547 guc_log_cleanup(guc);
548 /* logging will remain off */
549 i915.guc_log_level = -1;
550 return ret;
551}
552
553static void guc_log_capture_logs(struct intel_guc *guc)
554{
555 struct drm_i915_private *dev_priv = guc_to_i915(guc);
556
557 guc_read_update_log_buffer(guc);
558
559 /* Generally device is expected to be active only at this
560 * time, so get/put should be really quick.
561 */
562 intel_runtime_pm_get(dev_priv);
563 guc_log_flush_complete(guc);
564 intel_runtime_pm_put(dev_priv);
565}
566
567static void guc_flush_logs(struct intel_guc *guc)
568{
569 struct drm_i915_private *dev_priv = guc_to_i915(guc);
570
571 if (!i915.enable_guc_submission || (i915.guc_log_level < 0))
572 return;
573
574 /* First disable the interrupts, will be renabled afterwards */
575 gen9_disable_guc_interrupts(dev_priv);
576
577 /* Before initiating the forceful flush, wait for any pending/ongoing
578 * flush to complete otherwise forceful flush may not actually happen.
579 */
580 flush_work(&guc->log.flush_work);
581
582 /* Ask GuC to update the log buffer state */
583 guc_log_flush(guc);
584
585 /* GuC would have updated log buffer by now, so capture it */
586 guc_log_capture_logs(guc);
587}
588
589int i915_guc_log_control(struct drm_i915_private *dev_priv, u64 control_val)
590{
591 struct intel_guc *guc = &dev_priv->guc;
592
593 union guc_log_control log_param;
594 int ret;
595
596 log_param.value = control_val;
597
598 if (log_param.verbosity < GUC_LOG_VERBOSITY_MIN ||
599 log_param.verbosity > GUC_LOG_VERBOSITY_MAX)
600 return -EINVAL;
601
602 /* This combination doesn't make sense & won't have any effect */
603 if (!log_param.logging_enabled && (i915.guc_log_level < 0))
604 return 0;
605
606 ret = guc_log_control(guc, log_param.value);
607 if (ret < 0) {
608 DRM_DEBUG_DRIVER("guc_logging_control action failed %d\n", ret);
609 return ret;
610 }
611
612 i915.guc_log_level = log_param.verbosity;
613
614 /* If log_level was set as -1 at boot time, then the relay channel file
615 * wouldn't have been created by now and interrupts also would not have
616 * been enabled.
617 */
618 if (!dev_priv->guc.log.relay_chan) {
619 ret = guc_log_late_setup(guc);
620 if (!ret)
621 gen9_enable_guc_interrupts(dev_priv);
622 } else if (!log_param.logging_enabled) {
623 /* Once logging is disabled, GuC won't generate logs & send an
624 * interrupt. But there could be some data in the log buffer
625 * which is yet to be captured. So request GuC to update the log
626 * buffer state and then collect the left over logs.
627 */
628 guc_flush_logs(guc);
629
630 /* As logging is disabled, update log level to reflect that */
631 i915.guc_log_level = -1;
632 } else {
633 /* In case interrupts were disabled, enable them now */
634 gen9_enable_guc_interrupts(dev_priv);
635 }
636
637 return ret;
638}
639
640void i915_guc_log_register(struct drm_i915_private *dev_priv)
641{
642 if (!i915.enable_guc_submission)
643 return;
644
645 mutex_lock(&dev_priv->drm.struct_mutex);
646 guc_log_late_setup(&dev_priv->guc);
647 mutex_unlock(&dev_priv->drm.struct_mutex);
648}
649
650void i915_guc_log_unregister(struct drm_i915_private *dev_priv)
651{
652 if (!i915.enable_guc_submission)
653 return;
654
655 mutex_lock(&dev_priv->drm.struct_mutex);
656 guc_log_cleanup(&dev_priv->guc);
657 mutex_unlock(&dev_priv->drm.struct_mutex);
658}