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 *ring_str(int ring)
36 case RCS: return "render";
37 case VCS: return "bsd";
38 case BCS: return "blt";
39 case VECS: return "vebox";
40 case VCS2: return "bsd2";
45 static const char *pin_flag(int pinned)
55 static const char *tiling_flag(int tiling)
59 case I915_TILING_NONE: return "";
60 case I915_TILING_X: return " X";
61 case I915_TILING_Y: return " Y";
65 static const char *dirty_flag(int dirty)
67 return dirty ? " dirty" : "";
70 static const char *purgeable_flag(int purgeable)
72 return purgeable ? " purgeable" : "";
75 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
78 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
83 if (e->bytes == e->size - 1 || e->err)
89 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
92 if (e->pos + len <= e->start) {
97 /* First vsnprintf needs to fit in its entirety for memmove */
106 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
109 /* If this is first printf in this window, adjust it so that
110 * start position matches start of the buffer
113 if (e->pos < e->start) {
114 const size_t off = e->start - e->pos;
116 /* Should not happen but be paranoid */
117 if (off > len || e->bytes) {
122 memmove(e->buf, e->buf + off, len - off);
123 e->bytes = len - off;
132 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
133 const char *f, va_list args)
137 if (!__i915_error_ok(e))
140 /* Seek the first printf which is hits start position */
141 if (e->pos < e->start) {
145 len = vsnprintf(NULL, 0, f, tmp);
148 if (!__i915_error_seek(e, len))
152 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
153 if (len >= e->size - e->bytes)
154 len = e->size - e->bytes - 1;
156 __i915_error_advance(e, len);
159 static void i915_error_puts(struct drm_i915_error_state_buf *e,
164 if (!__i915_error_ok(e))
169 /* Seek the first printf which is hits start position */
170 if (e->pos < e->start) {
171 if (!__i915_error_seek(e, len))
175 if (len >= e->size - e->bytes)
176 len = e->size - e->bytes - 1;
177 memcpy(e->buf + e->bytes, str, len);
179 __i915_error_advance(e, len);
182 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
183 #define err_puts(e, s) i915_error_puts(e, s)
185 static void print_error_buffers(struct drm_i915_error_state_buf *m,
187 struct drm_i915_error_buffer *err,
192 err_printf(m, " %s [%d]:\n", name, count);
195 err_printf(m, " %08x_%08x %8u %02x %02x [ ",
196 upper_32_bits(err->gtt_offset),
197 lower_32_bits(err->gtt_offset),
201 for (i = 0; i < I915_NUM_ENGINES; i++)
202 err_printf(m, "%02x ", err->rseqno[i]);
204 err_printf(m, "] %02x", err->wseqno);
205 err_puts(m, pin_flag(err->pinned));
206 err_puts(m, tiling_flag(err->tiling));
207 err_puts(m, dirty_flag(err->dirty));
208 err_puts(m, purgeable_flag(err->purgeable));
209 err_puts(m, err->userptr ? " userptr" : "");
210 err_puts(m, err->ring != -1 ? " " : "");
211 err_puts(m, ring_str(err->ring));
212 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
215 err_printf(m, " (name: %d)", err->name);
216 if (err->fence_reg != I915_FENCE_REG_NONE)
217 err_printf(m, " (fence: %d)", err->fence_reg);
224 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
231 case HANGCHECK_ACTIVE:
242 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
243 struct drm_device *dev,
244 struct drm_i915_error_state *error,
247 struct drm_i915_error_ring *ring = &error->ring[ring_idx];
252 err_printf(m, "%s command stream:\n", ring_str(ring_idx));
253 err_printf(m, " START: 0x%08x\n", ring->start);
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, "PCI Revision: 0x%02x\n", dev->pdev->revision);
367 err_printf(m, "PCI Subsystem: %04x:%04x\n",
368 dev->pdev->subsystem_vendor,
369 dev->pdev->subsystem_device);
370 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
373 struct intel_csr *csr = &dev_priv->csr;
375 err_printf(m, "DMC loaded: %s\n",
376 yesno(csr->dmc_payload != NULL));
377 err_printf(m, "DMC fw version: %d.%d\n",
378 CSR_VERSION_MAJOR(csr->version),
379 CSR_VERSION_MINOR(csr->version));
382 err_printf(m, "EIR: 0x%08x\n", error->eir);
383 err_printf(m, "IER: 0x%08x\n", error->ier);
384 if (INTEL_INFO(dev)->gen >= 8) {
385 for (i = 0; i < 4; i++)
386 err_printf(m, "GTIER gt %d: 0x%08x\n", i,
388 } else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
389 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
390 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
391 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
392 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
393 err_printf(m, "CCID: 0x%08x\n", error->ccid);
394 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
396 for (i = 0; i < dev_priv->num_fence_regs; i++)
397 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
399 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
400 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
401 error->extra_instdone[i]);
403 if (INTEL_INFO(dev)->gen >= 6) {
404 err_printf(m, "ERROR: 0x%08x\n", error->error);
406 if (INTEL_INFO(dev)->gen >= 8)
407 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
408 error->fault_data1, error->fault_data0);
410 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
413 if (INTEL_INFO(dev)->gen == 7)
414 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
416 for (i = 0; i < ARRAY_SIZE(error->ring); i++)
417 i915_ring_error_state(m, dev, error, i);
419 for (i = 0; i < error->vm_count; i++) {
420 err_printf(m, "vm[%d]\n", i);
422 print_error_buffers(m, "Active",
424 error->active_bo_count[i]);
426 print_error_buffers(m, "Pinned",
428 error->pinned_bo_count[i]);
431 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
432 obj = error->ring[i].batchbuffer;
434 err_puts(m, dev_priv->engine[i].name);
435 if (error->ring[i].pid != -1)
436 err_printf(m, " (submitted by %s [%d])",
439 err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
440 upper_32_bits(obj->gtt_offset),
441 lower_32_bits(obj->gtt_offset));
442 print_error_obj(m, obj);
445 obj = error->ring[i].wa_batchbuffer;
447 err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
448 dev_priv->engine[i].name,
449 lower_32_bits(obj->gtt_offset));
450 print_error_obj(m, obj);
453 if (error->ring[i].num_requests) {
454 err_printf(m, "%s --- %d requests\n",
455 dev_priv->engine[i].name,
456 error->ring[i].num_requests);
457 for (j = 0; j < error->ring[i].num_requests; j++) {
458 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
459 error->ring[i].requests[j].seqno,
460 error->ring[i].requests[j].jiffies,
461 error->ring[i].requests[j].tail);
465 if ((obj = error->ring[i].ringbuffer)) {
466 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
467 dev_priv->engine[i].name,
468 lower_32_bits(obj->gtt_offset));
469 print_error_obj(m, obj);
472 if ((obj = error->ring[i].hws_page)) {
473 u64 hws_offset = obj->gtt_offset;
474 u32 *hws_page = &obj->pages[0][0];
476 if (i915.enable_execlists) {
477 hws_offset += LRC_PPHWSP_PN * PAGE_SIZE;
478 hws_page = &obj->pages[LRC_PPHWSP_PN][0];
480 err_printf(m, "%s --- HW Status = 0x%08llx\n",
481 dev_priv->engine[i].name, hws_offset);
483 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
484 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
494 obj = error->ring[i].wa_ctx;
496 u64 wa_ctx_offset = obj->gtt_offset;
497 u32 *wa_ctx_page = &obj->pages[0][0];
498 struct intel_engine_cs *engine = &dev_priv->engine[RCS];
499 u32 wa_ctx_size = (engine->wa_ctx.indirect_ctx.size +
500 engine->wa_ctx.per_ctx.size);
502 err_printf(m, "%s --- WA ctx batch buffer = 0x%08llx\n",
503 dev_priv->engine[i].name, wa_ctx_offset);
505 for (elt = 0; elt < wa_ctx_size; elt += 4) {
506 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
508 wa_ctx_page[elt + 0],
509 wa_ctx_page[elt + 1],
510 wa_ctx_page[elt + 2],
511 wa_ctx_page[elt + 3]);
516 if ((obj = error->ring[i].ctx)) {
517 err_printf(m, "%s --- HW Context = 0x%08x\n",
518 dev_priv->engine[i].name,
519 lower_32_bits(obj->gtt_offset));
520 print_error_obj(m, obj);
524 if ((obj = error->semaphore_obj)) {
525 err_printf(m, "Semaphore page = 0x%08x\n",
526 lower_32_bits(obj->gtt_offset));
527 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
528 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
531 obj->pages[0][elt+1],
532 obj->pages[0][elt+2],
533 obj->pages[0][elt+3]);
538 intel_overlay_print_error_state(m, error->overlay);
541 intel_display_print_error_state(m, dev, error->display);
544 if (m->bytes == 0 && m->err)
550 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
551 struct drm_i915_private *i915,
552 size_t count, loff_t pos)
554 memset(ebuf, 0, sizeof(*ebuf));
557 /* We need to have enough room to store any i915_error_state printf
558 * so that we can move it to start position.
560 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
561 ebuf->buf = kmalloc(ebuf->size,
562 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
564 if (ebuf->buf == NULL) {
565 ebuf->size = PAGE_SIZE;
566 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
569 if (ebuf->buf == NULL) {
571 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
574 if (ebuf->buf == NULL)
582 static void i915_error_object_free(struct drm_i915_error_object *obj)
589 for (page = 0; page < obj->page_count; page++)
590 kfree(obj->pages[page]);
595 static void i915_error_state_free(struct kref *error_ref)
597 struct drm_i915_error_state *error = container_of(error_ref,
598 typeof(*error), ref);
601 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
602 i915_error_object_free(error->ring[i].batchbuffer);
603 i915_error_object_free(error->ring[i].wa_batchbuffer);
604 i915_error_object_free(error->ring[i].ringbuffer);
605 i915_error_object_free(error->ring[i].hws_page);
606 i915_error_object_free(error->ring[i].ctx);
607 kfree(error->ring[i].requests);
608 i915_error_object_free(error->ring[i].wa_ctx);
611 i915_error_object_free(error->semaphore_obj);
613 for (i = 0; i < error->vm_count; i++)
614 kfree(error->active_bo[i]);
616 kfree(error->active_bo);
617 kfree(error->active_bo_count);
618 kfree(error->pinned_bo);
619 kfree(error->pinned_bo_count);
620 kfree(error->overlay);
621 kfree(error->display);
625 static struct drm_i915_error_object *
626 i915_error_object_create(struct drm_i915_private *dev_priv,
627 struct drm_i915_gem_object *src,
628 struct i915_address_space *vm)
630 struct drm_i915_error_object *dst;
631 struct i915_vma *vma = NULL;
637 if (src == NULL || src->pages == NULL)
640 num_pages = src->base.size >> PAGE_SHIFT;
642 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
646 if (i915_gem_obj_bound(src, vm))
647 dst->gtt_offset = i915_gem_obj_offset(src, vm);
649 dst->gtt_offset = -1;
651 reloc_offset = dst->gtt_offset;
652 if (i915_is_ggtt(vm))
653 vma = i915_gem_obj_to_ggtt(src);
654 use_ggtt = (src->cache_level == I915_CACHE_NONE &&
655 vma && (vma->bound & GLOBAL_BIND) &&
656 reloc_offset + num_pages * PAGE_SIZE <= dev_priv->ggtt.mappable_end);
658 /* Cannot access stolen address directly, try to use the aperture */
662 if (!(vma && vma->bound & GLOBAL_BIND))
665 reloc_offset = i915_gem_obj_ggtt_offset(src);
666 if (reloc_offset + num_pages * PAGE_SIZE > dev_priv->ggtt.mappable_end)
670 /* Cannot access snooped pages through the aperture */
671 if (use_ggtt && src->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv->dev))
674 dst->page_count = num_pages;
675 while (num_pages--) {
679 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
683 local_irq_save(flags);
687 /* Simply ignore tiling or any overlapping fence.
688 * It's part of the error state, and this hopefully
689 * captures what the GPU read.
692 s = io_mapping_map_atomic_wc(dev_priv->ggtt.mappable,
694 memcpy_fromio(d, s, PAGE_SIZE);
695 io_mapping_unmap_atomic(s);
700 page = i915_gem_object_get_page(src, i);
702 drm_clflush_pages(&page, 1);
704 s = kmap_atomic(page);
705 memcpy(d, s, PAGE_SIZE);
708 drm_clflush_pages(&page, 1);
710 local_irq_restore(flags);
713 reloc_offset += PAGE_SIZE;
720 kfree(dst->pages[i]);
724 #define i915_error_ggtt_object_create(dev_priv, src) \
725 i915_error_object_create((dev_priv), (src), &(dev_priv)->ggtt.base)
727 static void capture_bo(struct drm_i915_error_buffer *err,
728 struct i915_vma *vma)
730 struct drm_i915_gem_object *obj = vma->obj;
733 err->size = obj->base.size;
734 err->name = obj->base.name;
735 for (i = 0; i < I915_NUM_ENGINES; i++)
736 err->rseqno[i] = i915_gem_request_get_seqno(obj->last_read_req[i]);
737 err->wseqno = i915_gem_request_get_seqno(obj->last_write_req);
738 err->gtt_offset = vma->node.start;
739 err->read_domains = obj->base.read_domains;
740 err->write_domain = obj->base.write_domain;
741 err->fence_reg = obj->fence_reg;
743 if (i915_gem_obj_is_pinned(obj))
745 err->tiling = obj->tiling_mode;
746 err->dirty = obj->dirty;
747 err->purgeable = obj->madv != I915_MADV_WILLNEED;
748 err->userptr = obj->userptr.mm != NULL;
749 err->ring = obj->last_write_req ?
750 i915_gem_request_get_engine(obj->last_write_req)->id : -1;
751 err->cache_level = obj->cache_level;
754 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
755 int count, struct list_head *head)
757 struct i915_vma *vma;
760 list_for_each_entry(vma, head, vm_link) {
761 capture_bo(err++, vma);
769 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
770 int count, struct list_head *head,
771 struct i915_address_space *vm)
773 struct drm_i915_gem_object *obj;
774 struct drm_i915_error_buffer * const first = err;
775 struct drm_i915_error_buffer * const last = err + count;
777 list_for_each_entry(obj, head, global_list) {
778 struct i915_vma *vma;
783 list_for_each_entry(vma, &obj->vma_list, obj_link)
784 if (vma->vm == vm && vma->pin_count > 0)
785 capture_bo(err++, vma);
791 /* Generate a semi-unique error code. The code is not meant to have meaning, The
792 * code's only purpose is to try to prevent false duplicated bug reports by
793 * grossly estimating a GPU error state.
795 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
796 * the hang if we could strip the GTT offset information from it.
798 * It's only a small step better than a random number in its current form.
800 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
801 struct drm_i915_error_state *error,
804 uint32_t error_code = 0;
807 /* IPEHR would be an ideal way to detect errors, as it's the gross
808 * measure of "the command that hung." However, has some very common
809 * synchronization commands which almost always appear in the case
810 * strictly a client bug. Use instdone to differentiate those some.
812 for (i = 0; i < I915_NUM_ENGINES; i++) {
813 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
817 return error->ring[i].ipehr ^ error->ring[i].instdone;
824 static void i915_gem_record_fences(struct drm_device *dev,
825 struct drm_i915_error_state *error)
827 struct drm_i915_private *dev_priv = dev->dev_private;
830 if (IS_GEN3(dev) || IS_GEN2(dev)) {
831 for (i = 0; i < dev_priv->num_fence_regs; i++)
832 error->fence[i] = I915_READ(FENCE_REG(i));
833 } else if (IS_GEN5(dev) || IS_GEN4(dev)) {
834 for (i = 0; i < dev_priv->num_fence_regs; i++)
835 error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
836 } else if (INTEL_INFO(dev)->gen >= 6) {
837 for (i = 0; i < dev_priv->num_fence_regs; i++)
838 error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
843 static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
844 struct drm_i915_error_state *error,
845 struct intel_engine_cs *engine,
846 struct drm_i915_error_ring *ering)
848 struct intel_engine_cs *to;
851 if (!i915_semaphore_is_enabled(dev_priv->dev))
854 if (!error->semaphore_obj)
855 error->semaphore_obj =
856 i915_error_ggtt_object_create(dev_priv,
857 dev_priv->semaphore_obj);
859 for_each_engine(to, dev_priv, i) {
867 signal_offset = (GEN8_SIGNAL_OFFSET(engine, i) & (PAGE_SIZE - 1))
869 tmp = error->semaphore_obj->pages[0];
870 idx = intel_ring_sync_index(engine, to);
872 ering->semaphore_mboxes[idx] = tmp[signal_offset];
873 ering->semaphore_seqno[idx] = engine->semaphore.sync_seqno[idx];
877 static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
878 struct intel_engine_cs *engine,
879 struct drm_i915_error_ring *ering)
881 ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base));
882 ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base));
883 ering->semaphore_seqno[0] = engine->semaphore.sync_seqno[0];
884 ering->semaphore_seqno[1] = engine->semaphore.sync_seqno[1];
886 if (HAS_VEBOX(dev_priv->dev)) {
887 ering->semaphore_mboxes[2] =
888 I915_READ(RING_SYNC_2(engine->mmio_base));
889 ering->semaphore_seqno[2] = engine->semaphore.sync_seqno[2];
893 static void i915_record_ring_state(struct drm_device *dev,
894 struct drm_i915_error_state *error,
895 struct intel_engine_cs *engine,
896 struct drm_i915_error_ring *ering)
898 struct drm_i915_private *dev_priv = dev->dev_private;
900 if (INTEL_INFO(dev)->gen >= 6) {
901 ering->rc_psmi = I915_READ(RING_PSMI_CTL(engine->mmio_base));
902 ering->fault_reg = I915_READ(RING_FAULT_REG(engine));
903 if (INTEL_INFO(dev)->gen >= 8)
904 gen8_record_semaphore_state(dev_priv, error, engine,
907 gen6_record_semaphore_state(dev_priv, engine, ering);
910 if (INTEL_INFO(dev)->gen >= 4) {
911 ering->faddr = I915_READ(RING_DMA_FADD(engine->mmio_base));
912 ering->ipeir = I915_READ(RING_IPEIR(engine->mmio_base));
913 ering->ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
914 ering->instdone = I915_READ(RING_INSTDONE(engine->mmio_base));
915 ering->instps = I915_READ(RING_INSTPS(engine->mmio_base));
916 ering->bbaddr = I915_READ(RING_BBADDR(engine->mmio_base));
917 if (INTEL_INFO(dev)->gen >= 8) {
918 ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(engine->mmio_base)) << 32;
919 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(engine->mmio_base)) << 32;
921 ering->bbstate = I915_READ(RING_BBSTATE(engine->mmio_base));
923 ering->faddr = I915_READ(DMA_FADD_I8XX);
924 ering->ipeir = I915_READ(IPEIR);
925 ering->ipehr = I915_READ(IPEHR);
926 ering->instdone = I915_READ(GEN2_INSTDONE);
929 ering->waiting = waitqueue_active(&engine->irq_queue);
930 ering->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
931 ering->seqno = engine->get_seqno(engine, false);
932 ering->acthd = intel_ring_get_active_head(engine);
933 ering->start = I915_READ_START(engine);
934 ering->head = I915_READ_HEAD(engine);
935 ering->tail = I915_READ_TAIL(engine);
936 ering->ctl = I915_READ_CTL(engine);
938 if (I915_NEED_GFX_HWS(dev)) {
942 switch (engine->id) {
945 mmio = RENDER_HWS_PGA_GEN7;
948 mmio = BLT_HWS_PGA_GEN7;
951 mmio = BSD_HWS_PGA_GEN7;
954 mmio = VEBOX_HWS_PGA_GEN7;
957 } else if (IS_GEN6(engine->dev)) {
958 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
960 /* XXX: gen8 returns to sanity */
961 mmio = RING_HWS_PGA(engine->mmio_base);
964 ering->hws = I915_READ(mmio);
967 ering->hangcheck_score = engine->hangcheck.score;
968 ering->hangcheck_action = engine->hangcheck.action;
970 if (USES_PPGTT(dev)) {
973 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(engine));
976 ering->vm_info.pp_dir_base =
977 I915_READ(RING_PP_DIR_BASE_READ(engine));
978 else if (IS_GEN7(dev))
979 ering->vm_info.pp_dir_base =
980 I915_READ(RING_PP_DIR_BASE(engine));
981 else if (INTEL_INFO(dev)->gen >= 8)
982 for (i = 0; i < 4; i++) {
983 ering->vm_info.pdp[i] =
984 I915_READ(GEN8_RING_PDP_UDW(engine, i));
985 ering->vm_info.pdp[i] <<= 32;
986 ering->vm_info.pdp[i] |=
987 I915_READ(GEN8_RING_PDP_LDW(engine, i));
993 static void i915_gem_record_active_context(struct intel_engine_cs *engine,
994 struct drm_i915_error_state *error,
995 struct drm_i915_error_ring *ering)
997 struct drm_i915_private *dev_priv = engine->dev->dev_private;
998 struct drm_i915_gem_object *obj;
1000 /* Currently render ring is the only HW context user */
1001 if (engine->id != RCS || !error->ccid)
1004 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1005 if (!i915_gem_obj_ggtt_bound(obj))
1008 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
1009 ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
1015 static void i915_gem_record_rings(struct drm_device *dev,
1016 struct drm_i915_error_state *error)
1018 struct drm_i915_private *dev_priv = dev->dev_private;
1019 struct drm_i915_gem_request *request;
1022 for (i = 0; i < I915_NUM_ENGINES; i++) {
1023 struct intel_engine_cs *engine = &dev_priv->engine[i];
1024 struct intel_ringbuffer *rbuf;
1026 error->ring[i].pid = -1;
1028 if (engine->dev == NULL)
1031 error->ring[i].valid = true;
1033 i915_record_ring_state(dev, error, engine, &error->ring[i]);
1035 request = i915_gem_find_active_request(engine);
1037 struct i915_address_space *vm;
1039 vm = request->ctx && request->ctx->ppgtt ?
1040 &request->ctx->ppgtt->base :
1041 &dev_priv->ggtt.base;
1043 /* We need to copy these to an anonymous buffer
1044 * as the simplest method to avoid being overwritten
1047 error->ring[i].batchbuffer =
1048 i915_error_object_create(dev_priv,
1052 if (HAS_BROKEN_CS_TLB(dev_priv->dev))
1053 error->ring[i].wa_batchbuffer =
1054 i915_error_ggtt_object_create(dev_priv,
1055 engine->scratch.obj);
1058 struct task_struct *task;
1061 task = pid_task(request->pid, PIDTYPE_PID);
1063 strcpy(error->ring[i].comm, task->comm);
1064 error->ring[i].pid = task->pid;
1070 if (i915.enable_execlists) {
1071 /* TODO: This is only a small fix to keep basic error
1072 * capture working, but we need to add more information
1073 * for it to be useful (e.g. dump the context being
1077 rbuf = request->ctx->engine[engine->id].ringbuf;
1079 rbuf = dev_priv->kernel_context->engine[engine->id].ringbuf;
1081 rbuf = engine->buffer;
1083 error->ring[i].cpu_ring_head = rbuf->head;
1084 error->ring[i].cpu_ring_tail = rbuf->tail;
1086 error->ring[i].ringbuffer =
1087 i915_error_ggtt_object_create(dev_priv, rbuf->obj);
1089 error->ring[i].hws_page =
1090 i915_error_ggtt_object_create(dev_priv,
1091 engine->status_page.obj);
1093 if (engine->wa_ctx.obj) {
1094 error->ring[i].wa_ctx =
1095 i915_error_ggtt_object_create(dev_priv,
1096 engine->wa_ctx.obj);
1099 i915_gem_record_active_context(engine, error, &error->ring[i]);
1102 list_for_each_entry(request, &engine->request_list, list)
1105 error->ring[i].num_requests = count;
1106 error->ring[i].requests =
1107 kcalloc(count, sizeof(*error->ring[i].requests),
1109 if (error->ring[i].requests == NULL) {
1110 error->ring[i].num_requests = 0;
1115 list_for_each_entry(request, &engine->request_list, list) {
1116 struct drm_i915_error_request *erq;
1118 if (count >= error->ring[i].num_requests) {
1120 * If the ring request list was changed in
1121 * between the point where the error request
1122 * list was created and dimensioned and this
1123 * point then just exit early to avoid crashes.
1125 * We don't need to communicate that the
1126 * request list changed state during error
1127 * state capture and that the error state is
1128 * slightly incorrect as a consequence since we
1129 * are typically only interested in the request
1130 * list state at the point of error state
1131 * capture, not in any changes happening during
1137 erq = &error->ring[i].requests[count++];
1138 erq->seqno = request->seqno;
1139 erq->jiffies = request->emitted_jiffies;
1140 erq->tail = request->postfix;
1145 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1148 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1149 struct drm_i915_error_state *error,
1150 struct i915_address_space *vm,
1153 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1154 struct drm_i915_gem_object *obj;
1155 struct i915_vma *vma;
1159 list_for_each_entry(vma, &vm->active_list, vm_link)
1161 error->active_bo_count[ndx] = i;
1163 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1164 list_for_each_entry(vma, &obj->vma_list, obj_link)
1165 if (vma->vm == vm && vma->pin_count > 0)
1168 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1171 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1173 pinned_bo = active_bo + error->active_bo_count[ndx];
1177 error->active_bo_count[ndx] =
1178 capture_active_bo(active_bo,
1179 error->active_bo_count[ndx],
1183 error->pinned_bo_count[ndx] =
1184 capture_pinned_bo(pinned_bo,
1185 error->pinned_bo_count[ndx],
1186 &dev_priv->mm.bound_list, vm);
1187 error->active_bo[ndx] = active_bo;
1188 error->pinned_bo[ndx] = pinned_bo;
1191 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1192 struct drm_i915_error_state *error)
1194 struct i915_address_space *vm;
1197 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1200 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1201 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1202 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1204 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1207 if (error->active_bo == NULL ||
1208 error->pinned_bo == NULL ||
1209 error->active_bo_count == NULL ||
1210 error->pinned_bo_count == NULL) {
1211 kfree(error->active_bo);
1212 kfree(error->active_bo_count);
1213 kfree(error->pinned_bo);
1214 kfree(error->pinned_bo_count);
1216 error->active_bo = NULL;
1217 error->active_bo_count = NULL;
1218 error->pinned_bo = NULL;
1219 error->pinned_bo_count = NULL;
1221 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1222 i915_gem_capture_vm(dev_priv, error, vm, i++);
1224 error->vm_count = cnt;
1228 /* Capture all registers which don't fit into another category. */
1229 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1230 struct drm_i915_error_state *error)
1232 struct drm_device *dev = dev_priv->dev;
1235 /* General organization
1236 * 1. Registers specific to a single generation
1237 * 2. Registers which belong to multiple generations
1238 * 3. Feature specific registers.
1239 * 4. Everything else
1240 * Please try to follow the order.
1243 /* 1: Registers specific to a single generation */
1244 if (IS_VALLEYVIEW(dev)) {
1245 error->gtier[0] = I915_READ(GTIER);
1246 error->ier = I915_READ(VLV_IER);
1247 error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
1251 error->err_int = I915_READ(GEN7_ERR_INT);
1253 if (INTEL_INFO(dev)->gen >= 8) {
1254 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1255 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1259 error->forcewake = I915_READ_FW(FORCEWAKE);
1260 error->gab_ctl = I915_READ(GAB_CTL);
1261 error->gfx_mode = I915_READ(GFX_MODE);
1264 /* 2: Registers which belong to multiple generations */
1265 if (INTEL_INFO(dev)->gen >= 7)
1266 error->forcewake = I915_READ_FW(FORCEWAKE_MT);
1268 if (INTEL_INFO(dev)->gen >= 6) {
1269 error->derrmr = I915_READ(DERRMR);
1270 error->error = I915_READ(ERROR_GEN6);
1271 error->done_reg = I915_READ(DONE_REG);
1274 /* 3: Feature specific registers */
1275 if (IS_GEN6(dev) || IS_GEN7(dev)) {
1276 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1277 error->gac_eco = I915_READ(GAC_ECO_BITS);
1280 /* 4: Everything else */
1281 if (HAS_HW_CONTEXTS(dev))
1282 error->ccid = I915_READ(CCID);
1284 if (INTEL_INFO(dev)->gen >= 8) {
1285 error->ier = I915_READ(GEN8_DE_MISC_IER);
1286 for (i = 0; i < 4; i++)
1287 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1288 } else if (HAS_PCH_SPLIT(dev)) {
1289 error->ier = I915_READ(DEIER);
1290 error->gtier[0] = I915_READ(GTIER);
1291 } else if (IS_GEN2(dev)) {
1292 error->ier = I915_READ16(IER);
1293 } else if (!IS_VALLEYVIEW(dev)) {
1294 error->ier = I915_READ(IER);
1296 error->eir = I915_READ(EIR);
1297 error->pgtbl_er = I915_READ(PGTBL_ER);
1299 i915_get_extra_instdone(dev, error->extra_instdone);
1302 static void i915_error_capture_msg(struct drm_device *dev,
1303 struct drm_i915_error_state *error,
1305 const char *error_msg)
1307 struct drm_i915_private *dev_priv = dev->dev_private;
1309 int ring_id = -1, len;
1311 ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1313 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1314 "GPU HANG: ecode %d:%d:0x%08x",
1315 INTEL_INFO(dev)->gen, ring_id, ecode);
1317 if (ring_id != -1 && error->ring[ring_id].pid != -1)
1318 len += scnprintf(error->error_msg + len,
1319 sizeof(error->error_msg) - len,
1321 error->ring[ring_id].comm,
1322 error->ring[ring_id].pid);
1324 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1325 ", reason: %s, action: %s",
1327 wedged ? "reset" : "continue");
1330 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1331 struct drm_i915_error_state *error)
1334 #ifdef CONFIG_INTEL_IOMMU
1335 error->iommu = intel_iommu_gfx_mapped;
1337 error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1338 error->suspend_count = dev_priv->suspend_count;
1342 * i915_capture_error_state - capture an error record for later analysis
1345 * Should be called when an error is detected (either a hang or an error
1346 * interrupt) to capture error state from the time of the error. Fills
1347 * out a structure which becomes available in debugfs for user level tools
1350 void i915_capture_error_state(struct drm_device *dev, bool wedged,
1351 const char *error_msg)
1354 struct drm_i915_private *dev_priv = dev->dev_private;
1355 struct drm_i915_error_state *error;
1356 unsigned long flags;
1358 /* Account for pipe specific data like PIPE*STAT */
1359 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1361 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1365 kref_init(&error->ref);
1367 i915_capture_gen_state(dev_priv, error);
1368 i915_capture_reg_state(dev_priv, error);
1369 i915_gem_capture_buffers(dev_priv, error);
1370 i915_gem_record_fences(dev, error);
1371 i915_gem_record_rings(dev, error);
1373 do_gettimeofday(&error->time);
1375 error->overlay = intel_overlay_capture_error_state(dev);
1376 error->display = intel_display_capture_error_state(dev);
1378 i915_error_capture_msg(dev, error, wedged, error_msg);
1379 DRM_INFO("%s\n", error->error_msg);
1381 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1382 if (dev_priv->gpu_error.first_error == NULL) {
1383 dev_priv->gpu_error.first_error = error;
1386 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1389 i915_error_state_free(&error->ref);
1394 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1395 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1396 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1397 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1398 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1403 void i915_error_state_get(struct drm_device *dev,
1404 struct i915_error_state_file_priv *error_priv)
1406 struct drm_i915_private *dev_priv = dev->dev_private;
1408 spin_lock_irq(&dev_priv->gpu_error.lock);
1409 error_priv->error = dev_priv->gpu_error.first_error;
1410 if (error_priv->error)
1411 kref_get(&error_priv->error->ref);
1412 spin_unlock_irq(&dev_priv->gpu_error.lock);
1416 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1418 if (error_priv->error)
1419 kref_put(&error_priv->error->ref, i915_error_state_free);
1422 void i915_destroy_error_state(struct drm_device *dev)
1424 struct drm_i915_private *dev_priv = dev->dev_private;
1425 struct drm_i915_error_state *error;
1427 spin_lock_irq(&dev_priv->gpu_error.lock);
1428 error = dev_priv->gpu_error.first_error;
1429 dev_priv->gpu_error.first_error = NULL;
1430 spin_unlock_irq(&dev_priv->gpu_error.lock);
1433 kref_put(&error->ref, i915_error_state_free);
1436 const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
1439 case I915_CACHE_NONE: return " uncached";
1440 case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
1441 case I915_CACHE_L3_LLC: return " L3+LLC";
1442 case I915_CACHE_WT: return " WT";
1447 /* NB: please notice the memset */
1448 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1450 struct drm_i915_private *dev_priv = dev->dev_private;
1451 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1453 if (IS_GEN2(dev) || IS_GEN3(dev))
1454 instdone[0] = I915_READ(GEN2_INSTDONE);
1455 else if (IS_GEN4(dev) || IS_GEN5(dev) || IS_GEN6(dev)) {
1456 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1457 instdone[1] = I915_READ(GEN4_INSTDONE1);
1458 } else if (INTEL_INFO(dev)->gen >= 7) {
1459 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1460 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1461 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1462 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);