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";
49 static const char *pin_flag(int pinned)
59 static const char *tiling_flag(int tiling)
63 case I915_TILING_NONE: return "";
64 case I915_TILING_X: return " X";
65 case I915_TILING_Y: return " Y";
69 static const char *dirty_flag(int dirty)
71 return dirty ? " dirty" : "";
74 static const char *purgeable_flag(int purgeable)
76 return purgeable ? " purgeable" : "";
79 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
82 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
87 if (e->bytes == e->size - 1 || e->err)
93 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
96 if (e->pos + len <= e->start) {
101 /* First vsnprintf needs to fit in its entirety for memmove */
102 if (len >= e->size) {
110 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
113 /* If this is first printf in this window, adjust it so that
114 * start position matches start of the buffer
117 if (e->pos < e->start) {
118 const size_t off = e->start - e->pos;
120 /* Should not happen but be paranoid */
121 if (off > len || e->bytes) {
126 memmove(e->buf, e->buf + off, len - off);
127 e->bytes = len - off;
136 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
137 const char *f, va_list args)
141 if (!__i915_error_ok(e))
144 /* Seek the first printf which is hits start position */
145 if (e->pos < e->start) {
149 if (!__i915_error_seek(e, vsnprintf(NULL, 0, f, tmp)))
153 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
154 if (len >= e->size - e->bytes)
155 len = e->size - e->bytes - 1;
157 __i915_error_advance(e, len);
160 static void i915_error_puts(struct drm_i915_error_state_buf *e,
165 if (!__i915_error_ok(e))
170 /* Seek the first printf which is hits start position */
171 if (e->pos < e->start) {
172 if (!__i915_error_seek(e, len))
176 if (len >= e->size - e->bytes)
177 len = e->size - e->bytes - 1;
178 memcpy(e->buf + e->bytes, str, len);
180 __i915_error_advance(e, len);
183 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
184 #define err_puts(e, s) i915_error_puts(e, s)
186 static void print_error_buffers(struct drm_i915_error_state_buf *m,
188 struct drm_i915_error_buffer *err,
191 err_printf(m, "%s [%d]:\n", name, count);
194 err_printf(m, " %08x %8u %02x %02x %x %x",
199 err->rseqno, err->wseqno);
200 err_puts(m, pin_flag(err->pinned));
201 err_puts(m, tiling_flag(err->tiling));
202 err_puts(m, dirty_flag(err->dirty));
203 err_puts(m, purgeable_flag(err->purgeable));
204 err_puts(m, err->ring != -1 ? " " : "");
205 err_puts(m, ring_str(err->ring));
206 err_puts(m, i915_cache_level_str(err->cache_level));
209 err_printf(m, " (name: %d)", err->name);
210 if (err->fence_reg != I915_FENCE_REG_NONE)
211 err_printf(m, " (fence: %d)", err->fence_reg);
218 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
225 case HANGCHECK_ACTIVE:
236 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
237 struct drm_device *dev,
238 struct drm_i915_error_ring *ring)
243 err_printf(m, " HEAD: 0x%08x\n", ring->head);
244 err_printf(m, " TAIL: 0x%08x\n", ring->tail);
245 err_printf(m, " CTL: 0x%08x\n", ring->ctl);
246 err_printf(m, " HWS: 0x%08x\n", ring->hws);
247 err_printf(m, " ACTHD: 0x%08x\n", ring->acthd);
248 err_printf(m, " IPEIR: 0x%08x\n", ring->ipeir);
249 err_printf(m, " IPEHR: 0x%08x\n", ring->ipehr);
250 err_printf(m, " INSTDONE: 0x%08x\n", ring->instdone);
251 if (INTEL_INFO(dev)->gen >= 4) {
252 err_printf(m, " BBADDR: 0x%08llx\n", ring->bbaddr);
253 err_printf(m, " BB_STATE: 0x%08x\n", ring->bbstate);
254 err_printf(m, " INSTPS: 0x%08x\n", ring->instps);
256 err_printf(m, " INSTPM: 0x%08x\n", ring->instpm);
257 err_printf(m, " FADDR: 0x%08x\n", ring->faddr);
258 if (INTEL_INFO(dev)->gen >= 6) {
259 err_printf(m, " RC PSMI: 0x%08x\n", ring->rc_psmi);
260 err_printf(m, " FAULT_REG: 0x%08x\n", ring->fault_reg);
261 err_printf(m, " SYNC_0: 0x%08x [last synced 0x%08x]\n",
262 ring->semaphore_mboxes[0],
263 ring->semaphore_seqno[0]);
264 err_printf(m, " SYNC_1: 0x%08x [last synced 0x%08x]\n",
265 ring->semaphore_mboxes[1],
266 ring->semaphore_seqno[1]);
267 if (HAS_VEBOX(dev)) {
268 err_printf(m, " SYNC_2: 0x%08x [last synced 0x%08x]\n",
269 ring->semaphore_mboxes[2],
270 ring->semaphore_seqno[2]);
273 err_printf(m, " seqno: 0x%08x\n", ring->seqno);
274 err_printf(m, " waiting: %s\n", yesno(ring->waiting));
275 err_printf(m, " ring->head: 0x%08x\n", ring->cpu_ring_head);
276 err_printf(m, " ring->tail: 0x%08x\n", ring->cpu_ring_tail);
277 err_printf(m, " hangcheck: %s [%d]\n",
278 hangcheck_action_to_str(ring->hangcheck_action),
279 ring->hangcheck_score);
282 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
287 i915_error_vprintf(e, f, args);
291 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
292 const struct i915_error_state_file_priv *error_priv)
294 struct drm_device *dev = error_priv->dev;
295 drm_i915_private_t *dev_priv = dev->dev_private;
296 struct drm_i915_error_state *error = error_priv->error;
297 int i, j, page, offset, elt;
300 err_printf(m, "no error state collected\n");
304 err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
305 error->time.tv_usec);
306 err_printf(m, "Kernel: " UTS_RELEASE "\n");
307 err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
308 err_printf(m, "EIR: 0x%08x\n", error->eir);
309 err_printf(m, "IER: 0x%08x\n", error->ier);
310 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
311 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
312 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
313 err_printf(m, "CCID: 0x%08x\n", error->ccid);
314 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
316 for (i = 0; i < dev_priv->num_fence_regs; i++)
317 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
319 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
320 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
321 error->extra_instdone[i]);
323 if (INTEL_INFO(dev)->gen >= 6) {
324 err_printf(m, "ERROR: 0x%08x\n", error->error);
325 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
328 if (INTEL_INFO(dev)->gen == 7)
329 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
331 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
332 err_printf(m, "%s command stream:\n", ring_str(i));
333 i915_ring_error_state(m, dev, &error->ring[i]);
336 if (error->active_bo)
337 print_error_buffers(m, "Active",
339 error->active_bo_count[0]);
341 if (error->pinned_bo)
342 print_error_buffers(m, "Pinned",
344 error->pinned_bo_count[0]);
346 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
347 struct drm_i915_error_object *obj;
349 if ((obj = error->ring[i].batchbuffer)) {
350 err_printf(m, "%s --- gtt_offset = 0x%08x\n",
351 dev_priv->ring[i].name,
354 for (page = 0; page < obj->page_count; page++) {
355 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
356 err_printf(m, "%08x : %08x\n", offset,
357 obj->pages[page][elt]);
363 if (error->ring[i].num_requests) {
364 err_printf(m, "%s --- %d requests\n",
365 dev_priv->ring[i].name,
366 error->ring[i].num_requests);
367 for (j = 0; j < error->ring[i].num_requests; j++) {
368 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
369 error->ring[i].requests[j].seqno,
370 error->ring[i].requests[j].jiffies,
371 error->ring[i].requests[j].tail);
375 if ((obj = error->ring[i].ringbuffer)) {
376 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
377 dev_priv->ring[i].name,
380 for (page = 0; page < obj->page_count; page++) {
381 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
382 err_printf(m, "%08x : %08x\n",
384 obj->pages[page][elt]);
390 if ((obj = error->ring[i].hws_page)) {
391 err_printf(m, "%s --- HW Status = 0x%08x\n",
392 dev_priv->ring[i].name,
395 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
396 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
399 obj->pages[0][elt+1],
400 obj->pages[0][elt+2],
401 obj->pages[0][elt+3]);
406 if ((obj = error->ring[i].ctx)) {
407 err_printf(m, "%s --- HW Context = 0x%08x\n",
408 dev_priv->ring[i].name,
411 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
412 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
415 obj->pages[0][elt+1],
416 obj->pages[0][elt+2],
417 obj->pages[0][elt+3]);
424 intel_overlay_print_error_state(m, error->overlay);
427 intel_display_print_error_state(m, dev, error->display);
430 if (m->bytes == 0 && m->err)
436 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
437 size_t count, loff_t pos)
439 memset(ebuf, 0, sizeof(*ebuf));
441 /* We need to have enough room to store any i915_error_state printf
442 * so that we can move it to start position.
444 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
445 ebuf->buf = kmalloc(ebuf->size,
446 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
448 if (ebuf->buf == NULL) {
449 ebuf->size = PAGE_SIZE;
450 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
453 if (ebuf->buf == NULL) {
455 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
458 if (ebuf->buf == NULL)
466 static void i915_error_object_free(struct drm_i915_error_object *obj)
473 for (page = 0; page < obj->page_count; page++)
474 kfree(obj->pages[page]);
479 static void i915_error_state_free(struct kref *error_ref)
481 struct drm_i915_error_state *error = container_of(error_ref,
482 typeof(*error), ref);
485 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
486 i915_error_object_free(error->ring[i].batchbuffer);
487 i915_error_object_free(error->ring[i].ringbuffer);
488 i915_error_object_free(error->ring[i].hws_page);
489 i915_error_object_free(error->ring[i].ctx);
490 kfree(error->ring[i].requests);
493 kfree(error->active_bo);
494 kfree(error->overlay);
495 kfree(error->display);
499 static struct drm_i915_error_object *
500 i915_error_object_create_sized(struct drm_i915_private *dev_priv,
501 struct drm_i915_gem_object *src,
502 struct i915_address_space *vm,
505 struct drm_i915_error_object *dst;
509 if (src == NULL || src->pages == NULL)
512 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
516 reloc_offset = dst->gtt_offset = i915_gem_obj_offset(src, vm);
517 for (i = 0; i < num_pages; i++) {
521 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
525 local_irq_save(flags);
526 if (reloc_offset < dev_priv->gtt.mappable_end &&
527 src->has_global_gtt_mapping &&
531 /* Simply ignore tiling or any overlapping fence.
532 * It's part of the error state, and this hopefully
533 * captures what the GPU read.
536 s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
538 memcpy_fromio(d, s, PAGE_SIZE);
539 io_mapping_unmap_atomic(s);
540 } else if (src->stolen) {
541 unsigned long offset;
543 offset = dev_priv->mm.stolen_base;
544 offset += src->stolen->start;
545 offset += i << PAGE_SHIFT;
547 memcpy_fromio(d, (void __iomem *) offset, PAGE_SIZE);
552 page = i915_gem_object_get_page(src, i);
554 drm_clflush_pages(&page, 1);
556 s = kmap_atomic(page);
557 memcpy(d, s, PAGE_SIZE);
560 drm_clflush_pages(&page, 1);
562 local_irq_restore(flags);
566 reloc_offset += PAGE_SIZE;
568 dst->page_count = num_pages;
574 kfree(dst->pages[i]);
578 #define i915_error_object_create(dev_priv, src, vm) \
579 i915_error_object_create_sized((dev_priv), (src), (vm), \
580 (src)->base.size>>PAGE_SHIFT)
582 #define i915_error_ggtt_object_create(dev_priv, src) \
583 i915_error_object_create_sized((dev_priv), (src), &(dev_priv)->gtt.base, \
584 (src)->base.size>>PAGE_SHIFT)
586 static void capture_bo(struct drm_i915_error_buffer *err,
587 struct drm_i915_gem_object *obj)
589 err->size = obj->base.size;
590 err->name = obj->base.name;
591 err->rseqno = obj->last_read_seqno;
592 err->wseqno = obj->last_write_seqno;
593 err->gtt_offset = i915_gem_obj_ggtt_offset(obj);
594 err->read_domains = obj->base.read_domains;
595 err->write_domain = obj->base.write_domain;
596 err->fence_reg = obj->fence_reg;
598 if (i915_gem_obj_is_pinned(obj))
600 if (obj->user_pin_count > 0)
602 err->tiling = obj->tiling_mode;
603 err->dirty = obj->dirty;
604 err->purgeable = obj->madv != I915_MADV_WILLNEED;
605 err->ring = obj->ring ? obj->ring->id : -1;
606 err->cache_level = obj->cache_level;
609 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
610 int count, struct list_head *head)
612 struct i915_vma *vma;
615 list_for_each_entry(vma, head, mm_list) {
616 capture_bo(err++, vma->obj);
624 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
625 int count, struct list_head *head)
627 struct drm_i915_gem_object *obj;
630 list_for_each_entry(obj, head, global_list) {
631 if (!i915_gem_obj_is_pinned(obj))
634 capture_bo(err++, obj);
642 static void i915_gem_record_fences(struct drm_device *dev,
643 struct drm_i915_error_state *error)
645 struct drm_i915_private *dev_priv = dev->dev_private;
649 switch (INTEL_INFO(dev)->gen) {
653 for (i = 0; i < dev_priv->num_fence_regs; i++)
654 error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
658 for (i = 0; i < 16; i++)
659 error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
662 if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
663 for (i = 0; i < 8; i++)
664 error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
666 for (i = 0; i < 8; i++)
667 error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
675 /* This assumes all batchbuffers are executed from the PPGTT. It might have to
676 * change in the future. */
677 static bool is_active_vm(struct i915_address_space *vm,
678 struct intel_ring_buffer *ring)
680 struct drm_device *dev = vm->dev;
681 struct drm_i915_private *dev_priv = dev->dev_private;
682 struct i915_hw_ppgtt *ppgtt;
684 if (INTEL_INFO(dev)->gen < 7)
685 return i915_is_ggtt(vm);
687 /* FIXME: This ignores that the global gtt vm is also on this list. */
688 ppgtt = container_of(vm, struct i915_hw_ppgtt, base);
690 if (INTEL_INFO(dev)->gen >= 8) {
691 u64 pdp0 = (u64)I915_READ(GEN8_RING_PDP_UDW(ring, 0)) << 32;
692 pdp0 |= I915_READ(GEN8_RING_PDP_LDW(ring, 0));
693 return pdp0 == ppgtt->pd_dma_addr[0];
696 pp_db = I915_READ(RING_PP_DIR_BASE(ring));
697 return (pp_db >> 10) == ppgtt->pd_offset;
701 static struct drm_i915_error_object *
702 i915_error_first_batchbuffer(struct drm_i915_private *dev_priv,
703 struct intel_ring_buffer *ring)
705 struct i915_address_space *vm;
706 struct i915_vma *vma;
707 struct drm_i915_gem_object *obj;
708 bool found_active = false;
711 if (!ring->get_seqno)
714 if (HAS_BROKEN_CS_TLB(dev_priv->dev)) {
715 u32 acthd = I915_READ(ACTHD);
717 if (WARN_ON(ring->id != RCS))
720 obj = ring->scratch.obj;
722 acthd >= i915_gem_obj_ggtt_offset(obj) &&
723 acthd < i915_gem_obj_ggtt_offset(obj) + obj->base.size)
724 return i915_error_ggtt_object_create(dev_priv, obj);
727 seqno = ring->get_seqno(ring, false);
728 list_for_each_entry(vm, &dev_priv->vm_list, global_link) {
729 if (!is_active_vm(vm, ring))
734 list_for_each_entry(vma, &vm->active_list, mm_list) {
736 if (obj->ring != ring)
739 if (i915_seqno_passed(seqno, obj->last_read_seqno))
742 if ((obj->base.read_domains & I915_GEM_DOMAIN_COMMAND) == 0)
745 /* We need to copy these to an anonymous buffer as the simplest
746 * method to avoid being overwritten by userspace.
748 return i915_error_object_create(dev_priv, obj, vm);
752 WARN_ON(!found_active);
756 static void i915_record_ring_state(struct drm_device *dev,
757 struct intel_ring_buffer *ring,
758 struct drm_i915_error_ring *ering)
760 struct drm_i915_private *dev_priv = dev->dev_private;
762 if (INTEL_INFO(dev)->gen >= 6) {
763 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
764 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
765 ering->semaphore_mboxes[0]
766 = I915_READ(RING_SYNC_0(ring->mmio_base));
767 ering->semaphore_mboxes[1]
768 = I915_READ(RING_SYNC_1(ring->mmio_base));
769 ering->semaphore_seqno[0] = ring->sync_seqno[0];
770 ering->semaphore_seqno[1] = ring->sync_seqno[1];
773 if (HAS_VEBOX(dev)) {
774 ering->semaphore_mboxes[2] =
775 I915_READ(RING_SYNC_2(ring->mmio_base));
776 ering->semaphore_seqno[2] = ring->sync_seqno[2];
779 if (INTEL_INFO(dev)->gen >= 4) {
780 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
781 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
782 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
783 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
784 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
785 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
786 if (INTEL_INFO(dev)->gen >= 8)
787 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
788 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
790 ering->faddr = I915_READ(DMA_FADD_I8XX);
791 ering->ipeir = I915_READ(IPEIR);
792 ering->ipehr = I915_READ(IPEHR);
793 ering->instdone = I915_READ(INSTDONE);
796 ering->waiting = waitqueue_active(&ring->irq_queue);
797 ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
798 ering->seqno = ring->get_seqno(ring, false);
799 ering->acthd = intel_ring_get_active_head(ring);
800 ering->head = I915_READ_HEAD(ring);
801 ering->tail = I915_READ_TAIL(ring);
802 ering->ctl = I915_READ_CTL(ring);
804 if (I915_NEED_GFX_HWS(dev)) {
811 mmio = RENDER_HWS_PGA_GEN7;
814 mmio = BLT_HWS_PGA_GEN7;
817 mmio = BSD_HWS_PGA_GEN7;
820 mmio = VEBOX_HWS_PGA_GEN7;
823 } else if (IS_GEN6(ring->dev)) {
824 mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
826 /* XXX: gen8 returns to sanity */
827 mmio = RING_HWS_PGA(ring->mmio_base);
830 ering->hws = I915_READ(mmio);
833 ering->cpu_ring_head = ring->head;
834 ering->cpu_ring_tail = ring->tail;
836 ering->hangcheck_score = ring->hangcheck.score;
837 ering->hangcheck_action = ring->hangcheck.action;
841 static void i915_gem_record_active_context(struct intel_ring_buffer *ring,
842 struct drm_i915_error_state *error,
843 struct drm_i915_error_ring *ering)
845 struct drm_i915_private *dev_priv = ring->dev->dev_private;
846 struct drm_i915_gem_object *obj;
848 /* Currently render ring is the only HW context user */
849 if (ring->id != RCS || !error->ccid)
852 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
853 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
854 ering->ctx = i915_error_object_create_sized(dev_priv,
863 static void i915_gem_record_rings(struct drm_device *dev,
864 struct drm_i915_error_state *error)
866 struct drm_i915_private *dev_priv = dev->dev_private;
867 struct drm_i915_gem_request *request;
870 for (i = 0; i < I915_NUM_RINGS; i++) {
871 struct intel_ring_buffer *ring = &dev_priv->ring[i];
873 if (ring->dev == NULL)
876 error->ring[i].valid = true;
878 i915_record_ring_state(dev, ring, &error->ring[i]);
880 error->ring[i].batchbuffer =
881 i915_error_first_batchbuffer(dev_priv, ring);
883 error->ring[i].ringbuffer =
884 i915_error_ggtt_object_create(dev_priv, ring->obj);
886 if (ring->status_page.obj)
887 error->ring[i].hws_page =
888 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
890 i915_gem_record_active_context(ring, error, &error->ring[i]);
893 list_for_each_entry(request, &ring->request_list, list)
896 error->ring[i].num_requests = count;
897 error->ring[i].requests =
898 kcalloc(count, sizeof(*error->ring[i].requests),
900 if (error->ring[i].requests == NULL) {
901 error->ring[i].num_requests = 0;
906 list_for_each_entry(request, &ring->request_list, list) {
907 struct drm_i915_error_request *erq;
909 erq = &error->ring[i].requests[count++];
910 erq->seqno = request->seqno;
911 erq->jiffies = request->emitted_jiffies;
912 erq->tail = request->tail;
917 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
920 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
921 struct drm_i915_error_state *error,
922 struct i915_address_space *vm,
925 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
926 struct drm_i915_gem_object *obj;
927 struct i915_vma *vma;
931 list_for_each_entry(vma, &vm->active_list, mm_list)
933 error->active_bo_count[ndx] = i;
934 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
935 if (i915_gem_obj_is_pinned(obj))
937 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
940 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
942 pinned_bo = active_bo + error->active_bo_count[ndx];
946 error->active_bo_count[ndx] =
947 capture_active_bo(active_bo,
948 error->active_bo_count[ndx],
952 error->pinned_bo_count[ndx] =
953 capture_pinned_bo(pinned_bo,
954 error->pinned_bo_count[ndx],
955 &dev_priv->mm.bound_list);
956 error->active_bo[ndx] = active_bo;
957 error->pinned_bo[ndx] = pinned_bo;
960 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
961 struct drm_i915_error_state *error)
963 struct i915_address_space *vm;
966 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
969 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
970 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
971 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
973 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
976 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
977 i915_gem_capture_vm(dev_priv, error, vm, i++);
980 /* Capture all registers which don't fit into another category. */
981 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
982 struct drm_i915_error_state *error)
984 struct drm_device *dev = dev_priv->dev;
987 /* General organization
988 * 1. Registers specific to a single generation
989 * 2. Registers which belong to multiple generations
990 * 3. Feature specific registers.
992 * Please try to follow the order.
995 /* 1: Registers specific to a single generation */
996 if (IS_VALLEYVIEW(dev)) {
997 error->ier = I915_READ(GTIER) | I915_READ(VLV_IER);
998 error->forcewake = I915_READ(FORCEWAKE_VLV);
1002 error->err_int = I915_READ(GEN7_ERR_INT);
1005 error->forcewake = I915_READ(FORCEWAKE);
1008 error->ier = I915_READ16(IER);
1010 /* 2: Registers which belong to multiple generations */
1011 if (INTEL_INFO(dev)->gen >= 7)
1012 error->forcewake = I915_READ(FORCEWAKE_MT);
1014 if (INTEL_INFO(dev)->gen >= 6) {
1015 error->derrmr = I915_READ(DERRMR);
1016 error->error = I915_READ(ERROR_GEN6);
1017 error->done_reg = I915_READ(DONE_REG);
1020 /* 3: Feature specific registers */
1021 if (HAS_HW_CONTEXTS(dev))
1022 error->ccid = I915_READ(CCID);
1024 if (HAS_PCH_SPLIT(dev))
1025 error->ier = I915_READ(DEIER) | I915_READ(GTIER);
1027 error->ier = I915_READ(IER);
1029 error->pipestat[pipe] = I915_READ(PIPESTAT(pipe));
1032 /* 4: Everything else */
1033 error->eir = I915_READ(EIR);
1034 error->pgtbl_er = I915_READ(PGTBL_ER);
1036 i915_get_extra_instdone(dev, error->extra_instdone);
1040 * i915_capture_error_state - capture an error record for later analysis
1043 * Should be called when an error is detected (either a hang or an error
1044 * interrupt) to capture error state from the time of the error. Fills
1045 * out a structure which becomes available in debugfs for user level tools
1048 void i915_capture_error_state(struct drm_device *dev)
1050 struct drm_i915_private *dev_priv = dev->dev_private;
1051 struct drm_i915_error_state *error;
1052 unsigned long flags;
1054 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1055 error = dev_priv->gpu_error.first_error;
1056 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1060 /* Account for pipe specific data like PIPE*STAT */
1061 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1063 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1067 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1068 dev->primary->index);
1069 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1070 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1071 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1072 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1074 kref_init(&error->ref);
1076 i915_capture_reg_state(dev_priv, error);
1077 i915_gem_capture_buffers(dev_priv, error);
1078 i915_gem_record_fences(dev, error);
1079 i915_gem_record_rings(dev, error);
1081 do_gettimeofday(&error->time);
1083 error->overlay = intel_overlay_capture_error_state(dev);
1084 error->display = intel_display_capture_error_state(dev);
1086 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1087 if (dev_priv->gpu_error.first_error == NULL) {
1088 dev_priv->gpu_error.first_error = error;
1091 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1094 i915_error_state_free(&error->ref);
1097 void i915_error_state_get(struct drm_device *dev,
1098 struct i915_error_state_file_priv *error_priv)
1100 struct drm_i915_private *dev_priv = dev->dev_private;
1101 unsigned long flags;
1103 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1104 error_priv->error = dev_priv->gpu_error.first_error;
1105 if (error_priv->error)
1106 kref_get(&error_priv->error->ref);
1107 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1111 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1113 if (error_priv->error)
1114 kref_put(&error_priv->error->ref, i915_error_state_free);
1117 void i915_destroy_error_state(struct drm_device *dev)
1119 struct drm_i915_private *dev_priv = dev->dev_private;
1120 struct drm_i915_error_state *error;
1121 unsigned long flags;
1123 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1124 error = dev_priv->gpu_error.first_error;
1125 dev_priv->gpu_error.first_error = NULL;
1126 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1129 kref_put(&error->ref, i915_error_state_free);
1132 const char *i915_cache_level_str(int type)
1135 case I915_CACHE_NONE: return " uncached";
1136 case I915_CACHE_LLC: return " snooped or LLC";
1137 case I915_CACHE_L3_LLC: return " L3+LLC";
1138 case I915_CACHE_WT: return " WT";
1143 /* NB: please notice the memset */
1144 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1146 struct drm_i915_private *dev_priv = dev->dev_private;
1147 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1149 switch (INTEL_INFO(dev)->gen) {
1152 instdone[0] = I915_READ(INSTDONE);
1157 instdone[0] = I915_READ(INSTDONE_I965);
1158 instdone[1] = I915_READ(INSTDONE1);
1161 WARN_ONCE(1, "Unsupported platform\n");
1164 instdone[0] = I915_READ(GEN7_INSTDONE_1);
1165 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1166 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1167 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);