1 // SPDX-License-Identifier: MIT
3 * Copyright © 2014-2019 Intel Corporation
6 #include <linux/debugfs.h>
8 #include "gt/intel_gt.h"
10 #include "i915_memcpy.h"
11 #include "intel_guc_log.h"
13 static void guc_log_capture_logs(struct intel_guc_log *log);
16 * DOC: GuC firmware log
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
24 static int guc_action_flush_log_complete(struct intel_guc *guc)
27 INTEL_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE
30 return intel_guc_send(guc, action, ARRAY_SIZE(action));
33 static int guc_action_flush_log(struct intel_guc *guc)
36 INTEL_GUC_ACTION_FORCE_LOG_BUFFER_FLUSH,
40 return intel_guc_send(guc, action, ARRAY_SIZE(action));
43 static int guc_action_control_log(struct intel_guc *guc, bool enable,
44 bool default_logging, u32 verbosity)
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)
53 GEM_BUG_ON(verbosity > GUC_LOG_VERBOSITY_MAX);
55 return intel_guc_send(guc, action, ARRAY_SIZE(action));
58 static void guc_log_enable_flush_events(struct intel_guc_log *log)
60 intel_guc_enable_msg(log_to_guc(log),
61 INTEL_GUC_RECV_MSG_FLUSH_LOG_BUFFER |
62 INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED);
65 static void guc_log_disable_flush_events(struct intel_guc_log *log)
67 intel_guc_disable_msg(log_to_guc(log),
68 INTEL_GUC_RECV_MSG_FLUSH_LOG_BUFFER |
69 INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED);
73 * Sub buffer switch callback. Called whenever relay has to switch to a new
74 * sub buffer, relay stays on the same sub buffer if 0 is returned.
76 static int subbuf_start_callback(struct rchan_buf *buf,
82 * Use no-overwrite mode by default, where relay will stop accepting
83 * new data if there are no empty sub buffers left.
84 * There is no strict synchronization enforced by relay between Consumer
85 * and Producer. In overwrite mode, there is a possibility of getting
86 * inconsistent/garbled data, the producer could be writing on to the
87 * same sub buffer from which Consumer is reading. This can't be avoided
88 * unless Consumer is fast enough and can always run in tandem with
91 if (relay_buf_full(buf))
98 * file_create() callback. Creates relay file in debugfs.
100 static struct dentry *create_buf_file_callback(const char *filename,
101 struct dentry *parent,
103 struct rchan_buf *buf,
106 struct dentry *buf_file;
109 * This to enable the use of a single buffer for the relay channel and
110 * correspondingly have a single file exposed to User, through which
111 * it can collect the logs in order without any post-processing.
112 * Need to set 'is_global' even if parent is NULL for early logging.
119 buf_file = debugfs_create_file(filename, mode,
120 parent, buf, &relay_file_operations);
121 if (IS_ERR(buf_file))
128 * file_remove() default callback. Removes relay file in debugfs.
130 static int remove_buf_file_callback(struct dentry *dentry)
132 debugfs_remove(dentry);
136 /* relay channel callbacks */
137 static struct rchan_callbacks relay_callbacks = {
138 .subbuf_start = subbuf_start_callback,
139 .create_buf_file = create_buf_file_callback,
140 .remove_buf_file = remove_buf_file_callback,
143 static void guc_move_to_next_buf(struct intel_guc_log *log)
146 * Make sure the updates made in the sub buffer are visible when
147 * Consumer sees the following update to offset inside the sub buffer.
151 /* All data has been written, so now move the offset of sub buffer. */
152 relay_reserve(log->relay.channel, log->vma->obj->base.size);
154 /* Switch to the next sub buffer */
155 relay_flush(log->relay.channel);
158 static void *guc_get_write_buffer(struct intel_guc_log *log)
161 * Just get the base address of a new sub buffer and copy data into it
162 * ourselves. NULL will be returned in no-overwrite mode, if all sub
163 * buffers are full. Could have used the relay_write() to indirectly
164 * copy the data, but that would have been bit convoluted, as we need to
165 * write to only certain locations inside a sub buffer which cannot be
166 * done without using relay_reserve() along with relay_write(). So its
167 * better to use relay_reserve() alone.
169 return relay_reserve(log->relay.channel, 0);
172 static bool guc_check_log_buf_overflow(struct intel_guc_log *log,
173 enum guc_log_buffer_type type,
174 unsigned int full_cnt)
176 unsigned int prev_full_cnt = log->stats[type].sampled_overflow;
177 bool overflow = false;
179 if (full_cnt != prev_full_cnt) {
182 log->stats[type].overflow = full_cnt;
183 log->stats[type].sampled_overflow += full_cnt - prev_full_cnt;
185 if (full_cnt < prev_full_cnt) {
186 /* buffer_full_cnt is a 4 bit counter */
187 log->stats[type].sampled_overflow += 16;
190 dev_notice_ratelimited(guc_to_gt(log_to_guc(log))->i915->drm.dev,
191 "GuC log buffer overflow\n");
197 static unsigned int guc_get_log_buffer_size(enum guc_log_buffer_type type)
200 case GUC_ISR_LOG_BUFFER:
201 return ISR_BUFFER_SIZE;
202 case GUC_DPC_LOG_BUFFER:
203 return DPC_BUFFER_SIZE;
204 case GUC_CRASH_DUMP_LOG_BUFFER:
205 return CRASH_BUFFER_SIZE;
213 static void guc_read_update_log_buffer(struct intel_guc_log *log)
215 unsigned int buffer_size, read_offset, write_offset, bytes_to_copy, full_cnt;
216 struct guc_log_buffer_state *log_buf_state, *log_buf_snapshot_state;
217 struct guc_log_buffer_state log_buf_state_local;
218 enum guc_log_buffer_type type;
219 void *src_data, *dst_data;
222 mutex_lock(&log->relay.lock);
224 if (WARN_ON(!intel_guc_log_relay_created(log)))
227 /* Get the pointer to shared GuC log buffer */
228 log_buf_state = src_data = log->relay.buf_addr;
230 /* Get the pointer to local buffer to store the logs */
231 log_buf_snapshot_state = dst_data = guc_get_write_buffer(log);
233 if (unlikely(!log_buf_snapshot_state)) {
235 * Used rate limited to avoid deluge of messages, logs might be
236 * getting consumed by User at a slow rate.
238 DRM_ERROR_RATELIMITED("no sub-buffer to capture logs\n");
239 log->relay.full_count++;
244 /* Actual logs are present from the 2nd page */
245 src_data += PAGE_SIZE;
246 dst_data += PAGE_SIZE;
248 for (type = GUC_ISR_LOG_BUFFER; type < GUC_MAX_LOG_BUFFER; type++) {
250 * Make a copy of the state structure, inside GuC log buffer
251 * (which is uncached mapped), on the stack to avoid reading
252 * from it multiple times.
254 memcpy(&log_buf_state_local, log_buf_state,
255 sizeof(struct guc_log_buffer_state));
256 buffer_size = guc_get_log_buffer_size(type);
257 read_offset = log_buf_state_local.read_ptr;
258 write_offset = log_buf_state_local.sampled_write_ptr;
259 full_cnt = log_buf_state_local.buffer_full_cnt;
261 /* Bookkeeping stuff */
262 log->stats[type].flush += log_buf_state_local.flush_to_file;
263 new_overflow = guc_check_log_buf_overflow(log, type, full_cnt);
265 /* Update the state of shared log buffer */
266 log_buf_state->read_ptr = write_offset;
267 log_buf_state->flush_to_file = 0;
270 /* First copy the state structure in snapshot buffer */
271 memcpy(log_buf_snapshot_state, &log_buf_state_local,
272 sizeof(struct guc_log_buffer_state));
275 * The write pointer could have been updated by GuC firmware,
276 * after sending the flush interrupt to Host, for consistency
277 * set write pointer value to same value of sampled_write_ptr
278 * in the snapshot buffer.
280 log_buf_snapshot_state->write_ptr = write_offset;
281 log_buf_snapshot_state++;
283 /* Now copy the actual logs. */
284 if (unlikely(new_overflow)) {
285 /* copy the whole buffer in case of overflow */
287 write_offset = buffer_size;
288 } else if (unlikely((read_offset > buffer_size) ||
289 (write_offset > buffer_size))) {
290 DRM_ERROR("invalid log buffer state\n");
291 /* copy whole buffer as offsets are unreliable */
293 write_offset = buffer_size;
296 /* Just copy the newly written data */
297 if (read_offset > write_offset) {
298 i915_memcpy_from_wc(dst_data, src_data, write_offset);
299 bytes_to_copy = buffer_size - read_offset;
301 bytes_to_copy = write_offset - read_offset;
303 i915_memcpy_from_wc(dst_data + read_offset,
304 src_data + read_offset, bytes_to_copy);
306 src_data += buffer_size;
307 dst_data += buffer_size;
310 guc_move_to_next_buf(log);
313 mutex_unlock(&log->relay.lock);
316 static void capture_logs_work(struct work_struct *work)
318 struct intel_guc_log *log =
319 container_of(work, struct intel_guc_log, relay.flush_work);
321 guc_log_capture_logs(log);
324 static int guc_log_map(struct intel_guc_log *log)
328 lockdep_assert_held(&log->relay.lock);
334 * Create a WC (Uncached for read) vmalloc mapping of log
335 * buffer pages, so that we can directly get the data
336 * (up-to-date) from memory.
338 vaddr = i915_gem_object_pin_map(log->vma->obj, I915_MAP_WC);
340 return PTR_ERR(vaddr);
342 log->relay.buf_addr = vaddr;
347 static void guc_log_unmap(struct intel_guc_log *log)
349 lockdep_assert_held(&log->relay.lock);
351 i915_gem_object_unpin_map(log->vma->obj);
352 log->relay.buf_addr = NULL;
355 void intel_guc_log_init_early(struct intel_guc_log *log)
357 mutex_init(&log->relay.lock);
358 INIT_WORK(&log->relay.flush_work, capture_logs_work);
359 log->relay.started = false;
362 static int guc_log_relay_create(struct intel_guc_log *log)
364 struct intel_guc *guc = log_to_guc(log);
365 struct drm_i915_private *dev_priv = guc_to_gt(guc)->i915;
366 struct rchan *guc_log_relay_chan;
367 size_t n_subbufs, subbuf_size;
370 lockdep_assert_held(&log->relay.lock);
371 GEM_BUG_ON(!log->vma);
373 /* Keep the size of sub buffers same as shared log buffer */
374 subbuf_size = log->vma->size;
377 * Store up to 8 snapshots, which is large enough to buffer sufficient
378 * boot time logs and provides enough leeway to User, in terms of
379 * latency, for consuming the logs from relay. Also doesn't take
380 * up too much memory.
384 guc_log_relay_chan = relay_open("guc_log",
385 dev_priv->drm.primary->debugfs_root,
386 subbuf_size, n_subbufs,
387 &relay_callbacks, dev_priv);
388 if (!guc_log_relay_chan) {
389 DRM_ERROR("Couldn't create relay chan for GuC logging\n");
395 GEM_BUG_ON(guc_log_relay_chan->subbuf_size < subbuf_size);
396 log->relay.channel = guc_log_relay_chan;
401 static void guc_log_relay_destroy(struct intel_guc_log *log)
403 lockdep_assert_held(&log->relay.lock);
405 relay_close(log->relay.channel);
406 log->relay.channel = NULL;
409 static void guc_log_capture_logs(struct intel_guc_log *log)
411 struct intel_guc *guc = log_to_guc(log);
412 struct drm_i915_private *dev_priv = guc_to_gt(guc)->i915;
413 intel_wakeref_t wakeref;
415 guc_read_update_log_buffer(log);
418 * Generally device is expected to be active only at this
419 * time, so get/put should be really quick.
421 with_intel_runtime_pm(&dev_priv->runtime_pm, wakeref)
422 guc_action_flush_log_complete(guc);
425 static u32 __get_default_log_level(struct intel_guc_log *log)
427 struct intel_guc *guc = log_to_guc(log);
428 struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
430 /* A negative value means "use platform/config default" */
431 if (i915->params.guc_log_level < 0) {
432 return (IS_ENABLED(CONFIG_DRM_I915_DEBUG) ||
433 IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) ?
434 GUC_LOG_LEVEL_MAX : GUC_LOG_LEVEL_NON_VERBOSE;
437 if (i915->params.guc_log_level > GUC_LOG_LEVEL_MAX) {
438 DRM_WARN("Incompatible option detected: %s=%d, %s!\n",
439 "guc_log_level", i915->params.guc_log_level,
440 "verbosity too high");
441 return (IS_ENABLED(CONFIG_DRM_I915_DEBUG) ||
442 IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) ?
443 GUC_LOG_LEVEL_MAX : GUC_LOG_LEVEL_DISABLED;
446 GEM_BUG_ON(i915->params.guc_log_level < GUC_LOG_LEVEL_DISABLED);
447 GEM_BUG_ON(i915->params.guc_log_level > GUC_LOG_LEVEL_MAX);
448 return i915->params.guc_log_level;
451 int intel_guc_log_create(struct intel_guc_log *log)
453 struct intel_guc *guc = log_to_guc(log);
454 struct i915_vma *vma;
458 GEM_BUG_ON(log->vma);
461 * GuC Log buffer Layout
463 * +===============================+ 00B
464 * | Crash dump state header |
465 * +-------------------------------+ 32B
466 * | DPC state header |
467 * +-------------------------------+ 64B
468 * | ISR state header |
469 * +-------------------------------+ 96B
471 * +===============================+ PAGE_SIZE (4KB)
472 * | Crash Dump logs |
473 * +===============================+ + CRASH_SIZE
475 * +===============================+ + DPC_SIZE
477 * +===============================+ + ISR_SIZE
479 guc_log_size = PAGE_SIZE + CRASH_BUFFER_SIZE + DPC_BUFFER_SIZE +
482 vma = intel_guc_allocate_vma(guc, guc_log_size);
490 log->level = __get_default_log_level(log);
491 DRM_DEBUG_DRIVER("guc_log_level=%d (%s, verbose:%s, verbosity:%d)\n",
492 log->level, enableddisabled(log->level),
493 yesno(GUC_LOG_LEVEL_IS_VERBOSE(log->level)),
494 GUC_LOG_LEVEL_TO_VERBOSITY(log->level));
499 DRM_ERROR("Failed to allocate GuC log buffer. %d\n", ret);
503 void intel_guc_log_destroy(struct intel_guc_log *log)
505 i915_vma_unpin_and_release(&log->vma, 0);
508 int intel_guc_log_set_level(struct intel_guc_log *log, u32 level)
510 struct intel_guc *guc = log_to_guc(log);
511 struct drm_i915_private *dev_priv = guc_to_gt(guc)->i915;
512 intel_wakeref_t wakeref;
515 BUILD_BUG_ON(GUC_LOG_VERBOSITY_MIN != 0);
516 GEM_BUG_ON(!log->vma);
519 * GuC is recognizing log levels starting from 0 to max, we're using 0
520 * as indication that logging should be disabled.
522 if (level < GUC_LOG_LEVEL_DISABLED || level > GUC_LOG_LEVEL_MAX)
525 mutex_lock(&dev_priv->drm.struct_mutex);
527 if (log->level == level)
530 with_intel_runtime_pm(&dev_priv->runtime_pm, wakeref)
531 ret = guc_action_control_log(guc,
532 GUC_LOG_LEVEL_IS_VERBOSE(level),
533 GUC_LOG_LEVEL_IS_ENABLED(level),
534 GUC_LOG_LEVEL_TO_VERBOSITY(level));
536 DRM_DEBUG_DRIVER("guc_log_control action failed %d\n", ret);
543 mutex_unlock(&dev_priv->drm.struct_mutex);
548 bool intel_guc_log_relay_created(const struct intel_guc_log *log)
550 return log->relay.buf_addr;
553 int intel_guc_log_relay_open(struct intel_guc_log *log)
560 mutex_lock(&log->relay.lock);
562 if (intel_guc_log_relay_created(log)) {
568 * We require SSE 4.1 for fast reads from the GuC log buffer and
569 * it should be present on the chipsets supporting GuC based
572 if (!i915_has_memcpy_from_wc()) {
577 ret = guc_log_relay_create(log);
581 ret = guc_log_map(log);
585 mutex_unlock(&log->relay.lock);
590 guc_log_relay_destroy(log);
592 mutex_unlock(&log->relay.lock);
597 int intel_guc_log_relay_start(struct intel_guc_log *log)
599 if (log->relay.started)
602 guc_log_enable_flush_events(log);
605 * When GuC is logging without us relaying to userspace, we're ignoring
606 * the flush notification. This means that we need to unconditionally
607 * flush on relay enabling, since GuC only notifies us once.
609 queue_work(system_highpri_wq, &log->relay.flush_work);
611 log->relay.started = true;
616 void intel_guc_log_relay_flush(struct intel_guc_log *log)
618 struct intel_guc *guc = log_to_guc(log);
619 intel_wakeref_t wakeref;
621 if (!log->relay.started)
625 * Before initiating the forceful flush, wait for any pending/ongoing
626 * flush to complete otherwise forceful flush may not actually happen.
628 flush_work(&log->relay.flush_work);
630 with_intel_runtime_pm(guc_to_gt(guc)->uncore->rpm, wakeref)
631 guc_action_flush_log(guc);
633 /* GuC would have updated log buffer by now, so capture it */
634 guc_log_capture_logs(log);
638 * Stops the relay log. Called from intel_guc_log_relay_close(), so no
639 * possibility of race with start/flush since relay_write cannot race
642 static void guc_log_relay_stop(struct intel_guc_log *log)
644 struct intel_guc *guc = log_to_guc(log);
645 struct drm_i915_private *i915 = guc_to_gt(guc)->i915;
647 if (!log->relay.started)
650 guc_log_disable_flush_events(log);
651 intel_synchronize_irq(i915);
653 flush_work(&log->relay.flush_work);
655 log->relay.started = false;
658 void intel_guc_log_relay_close(struct intel_guc_log *log)
660 guc_log_relay_stop(log);
662 mutex_lock(&log->relay.lock);
663 GEM_BUG_ON(!intel_guc_log_relay_created(log));
665 guc_log_relay_destroy(log);
666 mutex_unlock(&log->relay.lock);
669 void intel_guc_log_handle_flush_event(struct intel_guc_log *log)
671 queue_work(system_highpri_wq, &log->relay.flush_work);
675 stringify_guc_log_type(enum guc_log_buffer_type type)
678 case GUC_ISR_LOG_BUFFER:
680 case GUC_DPC_LOG_BUFFER:
682 case GUC_CRASH_DUMP_LOG_BUFFER:
692 * intel_guc_log_info - dump information about GuC log relay
694 * @p: the &drm_printer
696 * Pretty printer for GuC log info
698 void intel_guc_log_info(struct intel_guc_log *log, struct drm_printer *p)
700 enum guc_log_buffer_type type;
702 if (!intel_guc_log_relay_created(log)) {
703 drm_puts(p, "GuC log relay not created\n");
707 drm_puts(p, "GuC logging stats:\n");
709 drm_printf(p, "\tRelay full count: %u\n", log->relay.full_count);
711 for (type = GUC_ISR_LOG_BUFFER; type < GUC_MAX_LOG_BUFFER; type++) {
712 drm_printf(p, "\t%s:\tflush count %10u, overflow count %10u\n",
713 stringify_guc_log_type(type),
714 log->stats[type].flush,
715 log->stats[type].sampled_overflow);
720 * intel_guc_log_dump - dump the contents of the GuC log
722 * @p: the &drm_printer
723 * @dump_load_err: dump the log saved on GuC load error
725 * Pretty printer for the GuC log
727 int intel_guc_log_dump(struct intel_guc_log *log, struct drm_printer *p,
730 struct intel_guc *guc = log_to_guc(log);
731 struct intel_uc *uc = container_of(guc, struct intel_uc, guc);
732 struct drm_i915_gem_object *obj = NULL;
736 if (!intel_guc_is_supported(guc))
740 obj = uc->load_err_log;
741 else if (guc->log.vma)
742 obj = guc->log.vma->obj;
747 map = i915_gem_object_pin_map(obj, I915_MAP_WC);
749 DRM_DEBUG("Failed to pin object\n");
750 drm_puts(p, "(log data unaccessible)\n");
754 for (i = 0; i < obj->base.size / sizeof(u32); i += 4)
755 drm_printf(p, "0x%08x 0x%08x 0x%08x 0x%08x\n",
756 *(map + i), *(map + i + 1),
757 *(map + i + 2), *(map + i + 3));
761 i915_gem_object_unpin_map(obj);