2 * Copyright (c) 2008 Intel Corporation
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:
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
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
24 * Eric Anholt <eric@anholt.net>
25 * Keith Packard <keithp@keithp.com>
26 * Mika Kuoppala <mika.kuoppala@intel.com>
30 #include <generated/utsrelease.h>
33 static const char *yesno(int v)
35 return v ? "yes" : "no";
38 static const char *ring_str(int ring)
41 case RCS: return "render";
42 case VCS: return "bsd";
43 case BCS: return "blt";
44 case VECS: return "vebox";
45 case VCS2: return "bsd2";
50 static const char *pin_flag(int pinned)
60 static const char *tiling_flag(int tiling)
64 case I915_TILING_NONE: return "";
65 case I915_TILING_X: return " X";
66 case I915_TILING_Y: return " Y";
70 static const char *dirty_flag(int dirty)
72 return dirty ? " dirty" : "";
75 static const char *purgeable_flag(int purgeable)
77 return purgeable ? " purgeable" : "";
80 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
83 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
88 if (e->bytes == e->size - 1 || e->err)
94 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
97 if (e->pos + len <= e->start) {
102 /* First vsnprintf needs to fit in its entirety for memmove */
103 if (len >= e->size) {
111 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
114 /* If this is first printf in this window, adjust it so that
115 * start position matches start of the buffer
118 if (e->pos < e->start) {
119 const size_t off = e->start - e->pos;
121 /* Should not happen but be paranoid */
122 if (off > len || e->bytes) {
127 memmove(e->buf, e->buf + off, len - off);
128 e->bytes = len - off;
137 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
138 const char *f, va_list args)
142 if (!__i915_error_ok(e))
145 /* Seek the first printf which is hits start position */
146 if (e->pos < e->start) {
150 len = vsnprintf(NULL, 0, f, tmp);
153 if (!__i915_error_seek(e, len))
157 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
158 if (len >= e->size - e->bytes)
159 len = e->size - e->bytes - 1;
161 __i915_error_advance(e, len);
164 static void i915_error_puts(struct drm_i915_error_state_buf *e,
169 if (!__i915_error_ok(e))
174 /* Seek the first printf which is hits start position */
175 if (e->pos < e->start) {
176 if (!__i915_error_seek(e, len))
180 if (len >= e->size - e->bytes)
181 len = e->size - e->bytes - 1;
182 memcpy(e->buf + e->bytes, str, len);
184 __i915_error_advance(e, len);
187 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
188 #define err_puts(e, s) i915_error_puts(e, s)
190 static void print_error_buffers(struct drm_i915_error_state_buf *m,
192 struct drm_i915_error_buffer *err,
195 err_printf(m, " %s [%d]:\n", name, count);
198 err_printf(m, " %08x %8u %02x %02x %x %x",
203 err->rseqno, err->wseqno);
204 err_puts(m, pin_flag(err->pinned));
205 err_puts(m, tiling_flag(err->tiling));
206 err_puts(m, dirty_flag(err->dirty));
207 err_puts(m, purgeable_flag(err->purgeable));
208 err_puts(m, err->userptr ? " userptr" : "");
209 err_puts(m, err->ring != -1 ? " " : "");
210 err_puts(m, ring_str(err->ring));
211 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
214 err_printf(m, " (name: %d)", err->name);
215 if (err->fence_reg != I915_FENCE_REG_NONE)
216 err_printf(m, " (fence: %d)", err->fence_reg);
223 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
230 case HANGCHECK_ACTIVE:
232 case HANGCHECK_ACTIVE_LOOP:
233 return "active (loop)";
243 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
244 struct drm_device *dev,
245 struct drm_i915_error_state *error,
248 struct drm_i915_error_ring *ring = &error->ring[ring_idx];
253 err_printf(m, "%s command stream:\n", ring_str(ring_idx));
254 err_printf(m, " HEAD: 0x%08x\n", ring->head);
255 err_printf(m, " TAIL: 0x%08x\n", ring->tail);
256 err_printf(m, " CTL: 0x%08x\n", ring->ctl);
257 err_printf(m, " HWS: 0x%08x\n", ring->hws);
258 err_printf(m, " ACTHD: 0x%08x %08x\n", (u32)(ring->acthd>>32), (u32)ring->acthd);
259 err_printf(m, " IPEIR: 0x%08x\n", ring->ipeir);
260 err_printf(m, " IPEHR: 0x%08x\n", ring->ipehr);
261 err_printf(m, " INSTDONE: 0x%08x\n", ring->instdone);
262 if (INTEL_INFO(dev)->gen >= 4) {
263 err_printf(m, " BBADDR: 0x%08x %08x\n", (u32)(ring->bbaddr>>32), (u32)ring->bbaddr);
264 err_printf(m, " BB_STATE: 0x%08x\n", ring->bbstate);
265 err_printf(m, " INSTPS: 0x%08x\n", ring->instps);
267 err_printf(m, " INSTPM: 0x%08x\n", ring->instpm);
268 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ring->faddr),
269 lower_32_bits(ring->faddr));
270 if (INTEL_INFO(dev)->gen >= 6) {
271 err_printf(m, " RC PSMI: 0x%08x\n", ring->rc_psmi);
272 err_printf(m, " FAULT_REG: 0x%08x\n", ring->fault_reg);
273 err_printf(m, " SYNC_0: 0x%08x [last synced 0x%08x]\n",
274 ring->semaphore_mboxes[0],
275 ring->semaphore_seqno[0]);
276 err_printf(m, " SYNC_1: 0x%08x [last synced 0x%08x]\n",
277 ring->semaphore_mboxes[1],
278 ring->semaphore_seqno[1]);
279 if (HAS_VEBOX(dev)) {
280 err_printf(m, " SYNC_2: 0x%08x [last synced 0x%08x]\n",
281 ring->semaphore_mboxes[2],
282 ring->semaphore_seqno[2]);
285 if (USES_PPGTT(dev)) {
286 err_printf(m, " GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
288 if (INTEL_INFO(dev)->gen >= 8) {
290 for (i = 0; i < 4; i++)
291 err_printf(m, " PDP%d: 0x%016llx\n",
292 i, ring->vm_info.pdp[i]);
294 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
295 ring->vm_info.pp_dir_base);
298 err_printf(m, " seqno: 0x%08x\n", ring->seqno);
299 err_printf(m, " waiting: %s\n", yesno(ring->waiting));
300 err_printf(m, " ring->head: 0x%08x\n", ring->cpu_ring_head);
301 err_printf(m, " ring->tail: 0x%08x\n", ring->cpu_ring_tail);
302 err_printf(m, " hangcheck: %s [%d]\n",
303 hangcheck_action_to_str(ring->hangcheck_action),
304 ring->hangcheck_score);
307 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
312 i915_error_vprintf(e, f, args);
316 static void print_error_obj(struct drm_i915_error_state_buf *m,
317 struct drm_i915_error_object *obj)
319 int page, offset, elt;
321 for (page = offset = 0; page < obj->page_count; page++) {
322 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
323 err_printf(m, "%08x : %08x\n", offset,
324 obj->pages[page][elt]);
330 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
331 const struct i915_error_state_file_priv *error_priv)
333 struct drm_device *dev = error_priv->dev;
334 struct drm_i915_private *dev_priv = dev->dev_private;
335 struct drm_i915_error_state *error = error_priv->error;
336 struct drm_i915_error_object *obj;
337 int i, j, offset, elt;
338 int max_hangcheck_score;
341 err_printf(m, "no error state collected\n");
345 err_printf(m, "%s\n", error->error_msg);
346 err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
347 error->time.tv_usec);
348 err_printf(m, "Kernel: " UTS_RELEASE "\n");
349 max_hangcheck_score = 0;
350 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
351 if (error->ring[i].hangcheck_score > max_hangcheck_score)
352 max_hangcheck_score = error->ring[i].hangcheck_score;
354 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
355 if (error->ring[i].hangcheck_score == max_hangcheck_score &&
356 error->ring[i].pid != -1) {
357 err_printf(m, "Active process (on ring %s): %s [%d]\n",
363 err_printf(m, "Reset count: %u\n", error->reset_count);
364 err_printf(m, "Suspend count: %u\n", error->suspend_count);
365 err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
366 err_printf(m, "EIR: 0x%08x\n", error->eir);
367 err_printf(m, "IER: 0x%08x\n", error->ier);
368 if (INTEL_INFO(dev)->gen >= 8) {
369 for (i = 0; i < 4; i++)
370 err_printf(m, "GTIER gt %d: 0x%08x\n", i,
372 } else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
373 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
374 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
375 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
376 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
377 err_printf(m, "CCID: 0x%08x\n", error->ccid);
378 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
380 for (i = 0; i < dev_priv->num_fence_regs; i++)
381 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
383 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
384 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
385 error->extra_instdone[i]);
387 if (INTEL_INFO(dev)->gen >= 6) {
388 err_printf(m, "ERROR: 0x%08x\n", error->error);
389 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
392 if (INTEL_INFO(dev)->gen == 7)
393 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
395 for (i = 0; i < ARRAY_SIZE(error->ring); i++)
396 i915_ring_error_state(m, dev, error, i);
398 for (i = 0; i < error->vm_count; i++) {
399 err_printf(m, "vm[%d]\n", i);
401 print_error_buffers(m, "Active",
403 error->active_bo_count[i]);
405 print_error_buffers(m, "Pinned",
407 error->pinned_bo_count[i]);
410 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
411 obj = error->ring[i].batchbuffer;
413 err_puts(m, dev_priv->ring[i].name);
414 if (error->ring[i].pid != -1)
415 err_printf(m, " (submitted by %s [%d])",
418 err_printf(m, " --- gtt_offset = 0x%08x\n",
420 print_error_obj(m, obj);
423 obj = error->ring[i].wa_batchbuffer;
425 err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
426 dev_priv->ring[i].name, obj->gtt_offset);
427 print_error_obj(m, obj);
430 if (error->ring[i].num_requests) {
431 err_printf(m, "%s --- %d requests\n",
432 dev_priv->ring[i].name,
433 error->ring[i].num_requests);
434 for (j = 0; j < error->ring[i].num_requests; j++) {
435 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
436 error->ring[i].requests[j].seqno,
437 error->ring[i].requests[j].jiffies,
438 error->ring[i].requests[j].tail);
442 if ((obj = error->ring[i].ringbuffer)) {
443 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
444 dev_priv->ring[i].name,
446 print_error_obj(m, obj);
449 if ((obj = error->ring[i].hws_page)) {
450 err_printf(m, "%s --- HW Status = 0x%08x\n",
451 dev_priv->ring[i].name,
454 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
455 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
458 obj->pages[0][elt+1],
459 obj->pages[0][elt+2],
460 obj->pages[0][elt+3]);
465 if ((obj = error->ring[i].ctx)) {
466 err_printf(m, "%s --- HW Context = 0x%08x\n",
467 dev_priv->ring[i].name,
469 print_error_obj(m, obj);
473 if ((obj = error->semaphore_obj)) {
474 err_printf(m, "Semaphore page = 0x%08x\n", obj->gtt_offset);
475 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
476 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
479 obj->pages[0][elt+1],
480 obj->pages[0][elt+2],
481 obj->pages[0][elt+3]);
486 intel_overlay_print_error_state(m, error->overlay);
489 intel_display_print_error_state(m, dev, error->display);
492 if (m->bytes == 0 && m->err)
498 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
499 struct drm_i915_private *i915,
500 size_t count, loff_t pos)
502 memset(ebuf, 0, sizeof(*ebuf));
505 /* We need to have enough room to store any i915_error_state printf
506 * so that we can move it to start position.
508 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
509 ebuf->buf = kmalloc(ebuf->size,
510 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
512 if (ebuf->buf == NULL) {
513 ebuf->size = PAGE_SIZE;
514 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
517 if (ebuf->buf == NULL) {
519 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
522 if (ebuf->buf == NULL)
530 static void i915_error_object_free(struct drm_i915_error_object *obj)
537 for (page = 0; page < obj->page_count; page++)
538 kfree(obj->pages[page]);
543 static void i915_error_state_free(struct kref *error_ref)
545 struct drm_i915_error_state *error = container_of(error_ref,
546 typeof(*error), ref);
549 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
550 i915_error_object_free(error->ring[i].batchbuffer);
551 i915_error_object_free(error->ring[i].ringbuffer);
552 i915_error_object_free(error->ring[i].hws_page);
553 i915_error_object_free(error->ring[i].ctx);
554 kfree(error->ring[i].requests);
557 i915_error_object_free(error->semaphore_obj);
558 kfree(error->active_bo);
559 kfree(error->overlay);
560 kfree(error->display);
564 static struct drm_i915_error_object *
565 i915_error_object_create(struct drm_i915_private *dev_priv,
566 struct drm_i915_gem_object *src,
567 struct i915_address_space *vm)
569 struct drm_i915_error_object *dst;
570 struct i915_vma *vma = NULL;
576 if (src == NULL || src->pages == NULL)
579 num_pages = src->base.size >> PAGE_SHIFT;
581 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
585 if (i915_gem_obj_bound(src, vm))
586 dst->gtt_offset = i915_gem_obj_offset(src, vm);
588 dst->gtt_offset = -1;
590 reloc_offset = dst->gtt_offset;
591 if (i915_is_ggtt(vm))
592 vma = i915_gem_obj_to_ggtt(src);
593 use_ggtt = (src->cache_level == I915_CACHE_NONE &&
594 vma && (vma->bound & GLOBAL_BIND) &&
595 reloc_offset + num_pages * PAGE_SIZE <= dev_priv->gtt.mappable_end);
597 /* Cannot access stolen address directly, try to use the aperture */
601 if (!(vma && vma->bound & GLOBAL_BIND))
604 reloc_offset = i915_gem_obj_ggtt_offset(src);
605 if (reloc_offset + num_pages * PAGE_SIZE > dev_priv->gtt.mappable_end)
609 /* Cannot access snooped pages through the aperture */
610 if (use_ggtt && src->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv->dev))
613 dst->page_count = num_pages;
614 while (num_pages--) {
618 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
622 local_irq_save(flags);
626 /* Simply ignore tiling or any overlapping fence.
627 * It's part of the error state, and this hopefully
628 * captures what the GPU read.
631 s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
633 memcpy_fromio(d, s, PAGE_SIZE);
634 io_mapping_unmap_atomic(s);
639 page = i915_gem_object_get_page(src, i);
641 drm_clflush_pages(&page, 1);
643 s = kmap_atomic(page);
644 memcpy(d, s, PAGE_SIZE);
647 drm_clflush_pages(&page, 1);
649 local_irq_restore(flags);
652 reloc_offset += PAGE_SIZE;
659 kfree(dst->pages[i]);
663 #define i915_error_ggtt_object_create(dev_priv, src) \
664 i915_error_object_create((dev_priv), (src), &(dev_priv)->gtt.base)
666 static void capture_bo(struct drm_i915_error_buffer *err,
667 struct i915_vma *vma)
669 struct drm_i915_gem_object *obj = vma->obj;
671 err->size = obj->base.size;
672 err->name = obj->base.name;
673 err->rseqno = obj->last_read_seqno;
674 err->wseqno = obj->last_write_seqno;
675 err->gtt_offset = vma->node.start;
676 err->read_domains = obj->base.read_domains;
677 err->write_domain = obj->base.write_domain;
678 err->fence_reg = obj->fence_reg;
680 if (i915_gem_obj_is_pinned(obj))
682 if (obj->user_pin_count > 0)
684 err->tiling = obj->tiling_mode;
685 err->dirty = obj->dirty;
686 err->purgeable = obj->madv != I915_MADV_WILLNEED;
687 err->userptr = obj->userptr.mm != NULL;
688 err->ring = obj->ring ? obj->ring->id : -1;
689 err->cache_level = obj->cache_level;
692 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
693 int count, struct list_head *head)
695 struct i915_vma *vma;
698 list_for_each_entry(vma, head, mm_list) {
699 capture_bo(err++, vma);
707 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
708 int count, struct list_head *head,
709 struct i915_address_space *vm)
711 struct drm_i915_gem_object *obj;
712 struct drm_i915_error_buffer * const first = err;
713 struct drm_i915_error_buffer * const last = err + count;
715 list_for_each_entry(obj, head, global_list) {
716 struct i915_vma *vma;
721 list_for_each_entry(vma, &obj->vma_list, vma_link)
722 if (vma->vm == vm && vma->pin_count > 0) {
723 capture_bo(err++, vma);
731 /* Generate a semi-unique error code. The code is not meant to have meaning, The
732 * code's only purpose is to try to prevent false duplicated bug reports by
733 * grossly estimating a GPU error state.
735 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
736 * the hang if we could strip the GTT offset information from it.
738 * It's only a small step better than a random number in its current form.
740 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
741 struct drm_i915_error_state *error,
744 uint32_t error_code = 0;
747 /* IPEHR would be an ideal way to detect errors, as it's the gross
748 * measure of "the command that hung." However, has some very common
749 * synchronization commands which almost always appear in the case
750 * strictly a client bug. Use instdone to differentiate those some.
752 for (i = 0; i < I915_NUM_RINGS; i++) {
753 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
757 return error->ring[i].ipehr ^ error->ring[i].instdone;
764 static void i915_gem_record_fences(struct drm_device *dev,
765 struct drm_i915_error_state *error)
767 struct drm_i915_private *dev_priv = dev->dev_private;
771 switch (INTEL_INFO(dev)->gen) {
776 for (i = 0; i < dev_priv->num_fence_regs; i++)
777 error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
781 for (i = 0; i < 16; i++)
782 error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
785 if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
786 for (i = 0; i < 8; i++)
787 error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
789 for (i = 0; i < 8; i++)
790 error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
799 static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
800 struct drm_i915_error_state *error,
801 struct intel_engine_cs *ring,
802 struct drm_i915_error_ring *ering)
804 struct intel_engine_cs *to;
807 if (!i915_semaphore_is_enabled(dev_priv->dev))
810 if (!error->semaphore_obj)
811 error->semaphore_obj =
812 i915_error_ggtt_object_create(dev_priv,
813 dev_priv->semaphore_obj);
815 for_each_ring(to, dev_priv, i) {
823 signal_offset = (GEN8_SIGNAL_OFFSET(ring, i) & (PAGE_SIZE - 1))
825 tmp = error->semaphore_obj->pages[0];
826 idx = intel_ring_sync_index(ring, to);
828 ering->semaphore_mboxes[idx] = tmp[signal_offset];
829 ering->semaphore_seqno[idx] = ring->semaphore.sync_seqno[idx];
833 static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
834 struct intel_engine_cs *ring,
835 struct drm_i915_error_ring *ering)
837 ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(ring->mmio_base));
838 ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(ring->mmio_base));
839 ering->semaphore_seqno[0] = ring->semaphore.sync_seqno[0];
840 ering->semaphore_seqno[1] = ring->semaphore.sync_seqno[1];
842 if (HAS_VEBOX(dev_priv->dev)) {
843 ering->semaphore_mboxes[2] =
844 I915_READ(RING_SYNC_2(ring->mmio_base));
845 ering->semaphore_seqno[2] = ring->semaphore.sync_seqno[2];
849 static void i915_record_ring_state(struct drm_device *dev,
850 struct drm_i915_error_state *error,
851 struct intel_engine_cs *ring,
852 struct drm_i915_error_ring *ering)
854 struct drm_i915_private *dev_priv = dev->dev_private;
856 if (INTEL_INFO(dev)->gen >= 6) {
857 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
858 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
859 if (INTEL_INFO(dev)->gen >= 8)
860 gen8_record_semaphore_state(dev_priv, error, ring, ering);
862 gen6_record_semaphore_state(dev_priv, ring, ering);
865 if (INTEL_INFO(dev)->gen >= 4) {
866 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
867 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
868 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
869 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
870 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
871 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
872 if (INTEL_INFO(dev)->gen >= 8) {
873 ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(ring->mmio_base)) << 32;
874 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
876 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
878 ering->faddr = I915_READ(DMA_FADD_I8XX);
879 ering->ipeir = I915_READ(IPEIR);
880 ering->ipehr = I915_READ(IPEHR);
881 ering->instdone = I915_READ(INSTDONE);
884 ering->waiting = waitqueue_active(&ring->irq_queue);
885 ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
886 ering->seqno = ring->get_seqno(ring, false);
887 ering->acthd = intel_ring_get_active_head(ring);
888 ering->head = I915_READ_HEAD(ring);
889 ering->tail = I915_READ_TAIL(ring);
890 ering->ctl = I915_READ_CTL(ring);
892 if (I915_NEED_GFX_HWS(dev)) {
899 mmio = RENDER_HWS_PGA_GEN7;
902 mmio = BLT_HWS_PGA_GEN7;
905 mmio = BSD_HWS_PGA_GEN7;
908 mmio = VEBOX_HWS_PGA_GEN7;
911 } else if (IS_GEN6(ring->dev)) {
912 mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
914 /* XXX: gen8 returns to sanity */
915 mmio = RING_HWS_PGA(ring->mmio_base);
918 ering->hws = I915_READ(mmio);
921 ering->hangcheck_score = ring->hangcheck.score;
922 ering->hangcheck_action = ring->hangcheck.action;
924 if (USES_PPGTT(dev)) {
927 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
929 switch (INTEL_INFO(dev)->gen) {
932 for (i = 0; i < 4; i++) {
933 ering->vm_info.pdp[i] =
934 I915_READ(GEN8_RING_PDP_UDW(ring, i));
935 ering->vm_info.pdp[i] <<= 32;
936 ering->vm_info.pdp[i] |=
937 I915_READ(GEN8_RING_PDP_LDW(ring, i));
941 ering->vm_info.pp_dir_base =
942 I915_READ(RING_PP_DIR_BASE(ring));
945 ering->vm_info.pp_dir_base =
946 I915_READ(RING_PP_DIR_BASE_READ(ring));
953 static void i915_gem_record_active_context(struct intel_engine_cs *ring,
954 struct drm_i915_error_state *error,
955 struct drm_i915_error_ring *ering)
957 struct drm_i915_private *dev_priv = ring->dev->dev_private;
958 struct drm_i915_gem_object *obj;
960 /* Currently render ring is the only HW context user */
961 if (ring->id != RCS || !error->ccid)
964 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
965 if (!i915_gem_obj_ggtt_bound(obj))
968 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
969 ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
975 static void i915_gem_record_rings(struct drm_device *dev,
976 struct drm_i915_error_state *error)
978 struct drm_i915_private *dev_priv = dev->dev_private;
979 struct drm_i915_gem_request *request;
982 for (i = 0; i < I915_NUM_RINGS; i++) {
983 struct intel_engine_cs *ring = &dev_priv->ring[i];
984 struct intel_ringbuffer *rbuf;
986 error->ring[i].pid = -1;
988 if (ring->dev == NULL)
991 error->ring[i].valid = true;
993 i915_record_ring_state(dev, error, ring, &error->ring[i]);
995 request = i915_gem_find_active_request(ring);
997 struct i915_address_space *vm;
999 vm = request->ctx && request->ctx->ppgtt ?
1000 &request->ctx->ppgtt->base :
1001 &dev_priv->gtt.base;
1003 /* We need to copy these to an anonymous buffer
1004 * as the simplest method to avoid being overwritten
1007 error->ring[i].batchbuffer =
1008 i915_error_object_create(dev_priv,
1012 if (HAS_BROKEN_CS_TLB(dev_priv->dev))
1013 error->ring[i].wa_batchbuffer =
1014 i915_error_ggtt_object_create(dev_priv,
1017 if (request->file_priv) {
1018 struct task_struct *task;
1021 task = pid_task(request->file_priv->file->pid,
1024 strcpy(error->ring[i].comm, task->comm);
1025 error->ring[i].pid = task->pid;
1031 if (i915.enable_execlists) {
1032 /* TODO: This is only a small fix to keep basic error
1033 * capture working, but we need to add more information
1034 * for it to be useful (e.g. dump the context being
1038 rbuf = request->ctx->engine[ring->id].ringbuf;
1040 rbuf = ring->default_context->engine[ring->id].ringbuf;
1042 rbuf = ring->buffer;
1044 error->ring[i].cpu_ring_head = rbuf->head;
1045 error->ring[i].cpu_ring_tail = rbuf->tail;
1047 error->ring[i].ringbuffer =
1048 i915_error_ggtt_object_create(dev_priv, rbuf->obj);
1050 error->ring[i].hws_page =
1051 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
1053 i915_gem_record_active_context(ring, error, &error->ring[i]);
1056 list_for_each_entry(request, &ring->request_list, list)
1059 error->ring[i].num_requests = count;
1060 error->ring[i].requests =
1061 kcalloc(count, sizeof(*error->ring[i].requests),
1063 if (error->ring[i].requests == NULL) {
1064 error->ring[i].num_requests = 0;
1069 list_for_each_entry(request, &ring->request_list, list) {
1070 struct drm_i915_error_request *erq;
1072 erq = &error->ring[i].requests[count++];
1073 erq->seqno = request->seqno;
1074 erq->jiffies = request->emitted_jiffies;
1075 erq->tail = request->tail;
1080 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1083 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1084 struct drm_i915_error_state *error,
1085 struct i915_address_space *vm,
1088 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1089 struct drm_i915_gem_object *obj;
1090 struct i915_vma *vma;
1094 list_for_each_entry(vma, &vm->active_list, mm_list)
1096 error->active_bo_count[ndx] = i;
1098 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1099 list_for_each_entry(vma, &obj->vma_list, vma_link)
1100 if (vma->vm == vm && vma->pin_count > 0) {
1105 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1108 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1110 pinned_bo = active_bo + error->active_bo_count[ndx];
1114 error->active_bo_count[ndx] =
1115 capture_active_bo(active_bo,
1116 error->active_bo_count[ndx],
1120 error->pinned_bo_count[ndx] =
1121 capture_pinned_bo(pinned_bo,
1122 error->pinned_bo_count[ndx],
1123 &dev_priv->mm.bound_list, vm);
1124 error->active_bo[ndx] = active_bo;
1125 error->pinned_bo[ndx] = pinned_bo;
1128 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1129 struct drm_i915_error_state *error)
1131 struct i915_address_space *vm;
1134 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1137 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1138 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1139 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1141 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1144 if (error->active_bo == NULL ||
1145 error->pinned_bo == NULL ||
1146 error->active_bo_count == NULL ||
1147 error->pinned_bo_count == NULL) {
1148 kfree(error->active_bo);
1149 kfree(error->active_bo_count);
1150 kfree(error->pinned_bo);
1151 kfree(error->pinned_bo_count);
1153 error->active_bo = NULL;
1154 error->active_bo_count = NULL;
1155 error->pinned_bo = NULL;
1156 error->pinned_bo_count = NULL;
1158 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1159 i915_gem_capture_vm(dev_priv, error, vm, i++);
1161 error->vm_count = cnt;
1165 /* Capture all registers which don't fit into another category. */
1166 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1167 struct drm_i915_error_state *error)
1169 struct drm_device *dev = dev_priv->dev;
1172 /* General organization
1173 * 1. Registers specific to a single generation
1174 * 2. Registers which belong to multiple generations
1175 * 3. Feature specific registers.
1176 * 4. Everything else
1177 * Please try to follow the order.
1180 /* 1: Registers specific to a single generation */
1181 if (IS_VALLEYVIEW(dev)) {
1182 error->gtier[0] = I915_READ(GTIER);
1183 error->ier = I915_READ(VLV_IER);
1184 error->forcewake = I915_READ(FORCEWAKE_VLV);
1188 error->err_int = I915_READ(GEN7_ERR_INT);
1191 error->forcewake = I915_READ(FORCEWAKE);
1192 error->gab_ctl = I915_READ(GAB_CTL);
1193 error->gfx_mode = I915_READ(GFX_MODE);
1196 /* 2: Registers which belong to multiple generations */
1197 if (INTEL_INFO(dev)->gen >= 7)
1198 error->forcewake = I915_READ(FORCEWAKE_MT);
1200 if (INTEL_INFO(dev)->gen >= 6) {
1201 error->derrmr = I915_READ(DERRMR);
1202 error->error = I915_READ(ERROR_GEN6);
1203 error->done_reg = I915_READ(DONE_REG);
1206 /* 3: Feature specific registers */
1207 if (IS_GEN6(dev) || IS_GEN7(dev)) {
1208 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1209 error->gac_eco = I915_READ(GAC_ECO_BITS);
1212 /* 4: Everything else */
1213 if (HAS_HW_CONTEXTS(dev))
1214 error->ccid = I915_READ(CCID);
1216 if (INTEL_INFO(dev)->gen >= 8) {
1217 error->ier = I915_READ(GEN8_DE_MISC_IER);
1218 for (i = 0; i < 4; i++)
1219 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1220 } else if (HAS_PCH_SPLIT(dev)) {
1221 error->ier = I915_READ(DEIER);
1222 error->gtier[0] = I915_READ(GTIER);
1223 } else if (IS_GEN2(dev)) {
1224 error->ier = I915_READ16(IER);
1225 } else if (!IS_VALLEYVIEW(dev)) {
1226 error->ier = I915_READ(IER);
1228 error->eir = I915_READ(EIR);
1229 error->pgtbl_er = I915_READ(PGTBL_ER);
1231 i915_get_extra_instdone(dev, error->extra_instdone);
1234 static void i915_error_capture_msg(struct drm_device *dev,
1235 struct drm_i915_error_state *error,
1237 const char *error_msg)
1239 struct drm_i915_private *dev_priv = dev->dev_private;
1241 int ring_id = -1, len;
1243 ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1245 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1246 "GPU HANG: ecode %d:%d:0x%08x",
1247 INTEL_INFO(dev)->gen, ring_id, ecode);
1249 if (ring_id != -1 && error->ring[ring_id].pid != -1)
1250 len += scnprintf(error->error_msg + len,
1251 sizeof(error->error_msg) - len,
1253 error->ring[ring_id].comm,
1254 error->ring[ring_id].pid);
1256 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1257 ", reason: %s, action: %s",
1259 wedged ? "reset" : "continue");
1262 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1263 struct drm_i915_error_state *error)
1265 error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1266 error->suspend_count = dev_priv->suspend_count;
1270 * i915_capture_error_state - capture an error record for later analysis
1273 * Should be called when an error is detected (either a hang or an error
1274 * interrupt) to capture error state from the time of the error. Fills
1275 * out a structure which becomes available in debugfs for user level tools
1278 void i915_capture_error_state(struct drm_device *dev, bool wedged,
1279 const char *error_msg)
1282 struct drm_i915_private *dev_priv = dev->dev_private;
1283 struct drm_i915_error_state *error;
1284 unsigned long flags;
1286 /* Account for pipe specific data like PIPE*STAT */
1287 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1289 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1293 kref_init(&error->ref);
1295 i915_capture_gen_state(dev_priv, error);
1296 i915_capture_reg_state(dev_priv, error);
1297 i915_gem_capture_buffers(dev_priv, error);
1298 i915_gem_record_fences(dev, error);
1299 i915_gem_record_rings(dev, error);
1301 do_gettimeofday(&error->time);
1303 error->overlay = intel_overlay_capture_error_state(dev);
1304 error->display = intel_display_capture_error_state(dev);
1306 i915_error_capture_msg(dev, error, wedged, error_msg);
1307 DRM_INFO("%s\n", error->error_msg);
1309 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1310 if (dev_priv->gpu_error.first_error == NULL) {
1311 dev_priv->gpu_error.first_error = error;
1314 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1317 i915_error_state_free(&error->ref);
1322 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1323 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1324 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1325 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1326 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1331 void i915_error_state_get(struct drm_device *dev,
1332 struct i915_error_state_file_priv *error_priv)
1334 struct drm_i915_private *dev_priv = dev->dev_private;
1336 spin_lock_irq(&dev_priv->gpu_error.lock);
1337 error_priv->error = dev_priv->gpu_error.first_error;
1338 if (error_priv->error)
1339 kref_get(&error_priv->error->ref);
1340 spin_unlock_irq(&dev_priv->gpu_error.lock);
1344 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1346 if (error_priv->error)
1347 kref_put(&error_priv->error->ref, i915_error_state_free);
1350 void i915_destroy_error_state(struct drm_device *dev)
1352 struct drm_i915_private *dev_priv = dev->dev_private;
1353 struct drm_i915_error_state *error;
1355 spin_lock_irq(&dev_priv->gpu_error.lock);
1356 error = dev_priv->gpu_error.first_error;
1357 dev_priv->gpu_error.first_error = NULL;
1358 spin_unlock_irq(&dev_priv->gpu_error.lock);
1361 kref_put(&error->ref, i915_error_state_free);
1364 const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
1367 case I915_CACHE_NONE: return " uncached";
1368 case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
1369 case I915_CACHE_L3_LLC: return " L3+LLC";
1370 case I915_CACHE_WT: return " WT";
1375 /* NB: please notice the memset */
1376 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1378 struct drm_i915_private *dev_priv = dev->dev_private;
1379 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1381 switch (INTEL_INFO(dev)->gen) {
1384 instdone[0] = I915_READ(INSTDONE);
1389 instdone[0] = I915_READ(INSTDONE_I965);
1390 instdone[1] = I915_READ(INSTDONE1);
1393 WARN_ONCE(1, "Unsupported platform\n");
1397 instdone[0] = I915_READ(GEN7_INSTDONE_1);
1398 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1399 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1400 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);