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 if (USES_PPGTT(dev)) {
274 err_printf(m, " GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
276 if (INTEL_INFO(dev)->gen >= 8) {
278 for (i = 0; i < 4; i++)
279 err_printf(m, " PDP%d: 0x%016llx\n",
280 i, ring->vm_info.pdp[i]);
282 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
283 ring->vm_info.pp_dir_base);
286 err_printf(m, " seqno: 0x%08x\n", ring->seqno);
287 err_printf(m, " waiting: %s\n", yesno(ring->waiting));
288 err_printf(m, " ring->head: 0x%08x\n", ring->cpu_ring_head);
289 err_printf(m, " ring->tail: 0x%08x\n", ring->cpu_ring_tail);
290 err_printf(m, " hangcheck: %s [%d]\n",
291 hangcheck_action_to_str(ring->hangcheck_action),
292 ring->hangcheck_score);
295 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
300 i915_error_vprintf(e, f, args);
304 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
305 const struct i915_error_state_file_priv *error_priv)
307 struct drm_device *dev = error_priv->dev;
308 drm_i915_private_t *dev_priv = dev->dev_private;
309 struct drm_i915_error_state *error = error_priv->error;
310 int i, j, page, offset, elt;
313 err_printf(m, "no error state collected\n");
317 err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
318 error->time.tv_usec);
319 err_printf(m, "Kernel: " UTS_RELEASE "\n");
320 err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
321 err_printf(m, "EIR: 0x%08x\n", error->eir);
322 err_printf(m, "IER: 0x%08x\n", error->ier);
323 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
324 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
325 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
326 err_printf(m, "CCID: 0x%08x\n", error->ccid);
327 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
329 for (i = 0; i < dev_priv->num_fence_regs; i++)
330 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
332 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
333 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
334 error->extra_instdone[i]);
336 if (INTEL_INFO(dev)->gen >= 6) {
337 err_printf(m, "ERROR: 0x%08x\n", error->error);
338 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
341 if (INTEL_INFO(dev)->gen == 7)
342 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
344 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
345 err_printf(m, "%s command stream:\n", ring_str(i));
346 i915_ring_error_state(m, dev, &error->ring[i]);
349 if (error->active_bo)
350 print_error_buffers(m, "Active",
352 error->active_bo_count[0]);
354 if (error->pinned_bo)
355 print_error_buffers(m, "Pinned",
357 error->pinned_bo_count[0]);
359 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
360 struct drm_i915_error_object *obj;
362 if ((obj = error->ring[i].batchbuffer)) {
363 err_printf(m, "%s --- gtt_offset = 0x%08x\n",
364 dev_priv->ring[i].name,
367 for (page = 0; page < obj->page_count; page++) {
368 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
369 err_printf(m, "%08x : %08x\n", offset,
370 obj->pages[page][elt]);
376 if (error->ring[i].num_requests) {
377 err_printf(m, "%s --- %d requests\n",
378 dev_priv->ring[i].name,
379 error->ring[i].num_requests);
380 for (j = 0; j < error->ring[i].num_requests; j++) {
381 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
382 error->ring[i].requests[j].seqno,
383 error->ring[i].requests[j].jiffies,
384 error->ring[i].requests[j].tail);
388 if ((obj = error->ring[i].ringbuffer)) {
389 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
390 dev_priv->ring[i].name,
393 for (page = 0; page < obj->page_count; page++) {
394 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
395 err_printf(m, "%08x : %08x\n",
397 obj->pages[page][elt]);
403 if ((obj = error->ring[i].hws_page)) {
404 err_printf(m, "%s --- HW Status = 0x%08x\n",
405 dev_priv->ring[i].name,
408 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
409 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
412 obj->pages[0][elt+1],
413 obj->pages[0][elt+2],
414 obj->pages[0][elt+3]);
419 if ((obj = error->ring[i].ctx)) {
420 err_printf(m, "%s --- HW Context = 0x%08x\n",
421 dev_priv->ring[i].name,
424 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
425 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
428 obj->pages[0][elt+1],
429 obj->pages[0][elt+2],
430 obj->pages[0][elt+3]);
437 intel_overlay_print_error_state(m, error->overlay);
440 intel_display_print_error_state(m, dev, error->display);
443 if (m->bytes == 0 && m->err)
449 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
450 size_t count, loff_t pos)
452 memset(ebuf, 0, sizeof(*ebuf));
454 /* We need to have enough room to store any i915_error_state printf
455 * so that we can move it to start position.
457 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
458 ebuf->buf = kmalloc(ebuf->size,
459 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
461 if (ebuf->buf == NULL) {
462 ebuf->size = PAGE_SIZE;
463 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
466 if (ebuf->buf == NULL) {
468 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
471 if (ebuf->buf == NULL)
479 static void i915_error_object_free(struct drm_i915_error_object *obj)
486 for (page = 0; page < obj->page_count; page++)
487 kfree(obj->pages[page]);
492 static void i915_error_state_free(struct kref *error_ref)
494 struct drm_i915_error_state *error = container_of(error_ref,
495 typeof(*error), ref);
498 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
499 i915_error_object_free(error->ring[i].batchbuffer);
500 i915_error_object_free(error->ring[i].ringbuffer);
501 i915_error_object_free(error->ring[i].hws_page);
502 i915_error_object_free(error->ring[i].ctx);
503 kfree(error->ring[i].requests);
506 kfree(error->active_bo);
507 kfree(error->overlay);
508 kfree(error->display);
512 static struct drm_i915_error_object *
513 i915_error_object_create_sized(struct drm_i915_private *dev_priv,
514 struct drm_i915_gem_object *src,
515 struct i915_address_space *vm,
518 struct drm_i915_error_object *dst;
522 if (src == NULL || src->pages == NULL)
525 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
529 reloc_offset = dst->gtt_offset = i915_gem_obj_offset(src, vm);
530 for (i = 0; i < num_pages; i++) {
534 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
538 local_irq_save(flags);
539 if (src->cache_level == I915_CACHE_NONE &&
540 reloc_offset < dev_priv->gtt.mappable_end &&
541 src->has_global_gtt_mapping &&
545 /* Simply ignore tiling or any overlapping fence.
546 * It's part of the error state, and this hopefully
547 * captures what the GPU read.
550 s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
552 memcpy_fromio(d, s, PAGE_SIZE);
553 io_mapping_unmap_atomic(s);
554 } else if (src->stolen) {
555 unsigned long offset;
557 offset = dev_priv->mm.stolen_base;
558 offset += src->stolen->start;
559 offset += i << PAGE_SHIFT;
561 memcpy_fromio(d, (void __iomem *) offset, PAGE_SIZE);
566 page = i915_gem_object_get_page(src, i);
568 drm_clflush_pages(&page, 1);
570 s = kmap_atomic(page);
571 memcpy(d, s, PAGE_SIZE);
574 drm_clflush_pages(&page, 1);
576 local_irq_restore(flags);
580 reloc_offset += PAGE_SIZE;
582 dst->page_count = num_pages;
588 kfree(dst->pages[i]);
592 #define i915_error_object_create(dev_priv, src, vm) \
593 i915_error_object_create_sized((dev_priv), (src), (vm), \
594 (src)->base.size>>PAGE_SHIFT)
596 #define i915_error_ggtt_object_create(dev_priv, src) \
597 i915_error_object_create_sized((dev_priv), (src), &(dev_priv)->gtt.base, \
598 (src)->base.size>>PAGE_SHIFT)
600 static void capture_bo(struct drm_i915_error_buffer *err,
601 struct drm_i915_gem_object *obj)
603 err->size = obj->base.size;
604 err->name = obj->base.name;
605 err->rseqno = obj->last_read_seqno;
606 err->wseqno = obj->last_write_seqno;
607 err->gtt_offset = i915_gem_obj_ggtt_offset(obj);
608 err->read_domains = obj->base.read_domains;
609 err->write_domain = obj->base.write_domain;
610 err->fence_reg = obj->fence_reg;
612 if (i915_gem_obj_is_pinned(obj))
614 if (obj->user_pin_count > 0)
616 err->tiling = obj->tiling_mode;
617 err->dirty = obj->dirty;
618 err->purgeable = obj->madv != I915_MADV_WILLNEED;
619 err->ring = obj->ring ? obj->ring->id : -1;
620 err->cache_level = obj->cache_level;
623 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
624 int count, struct list_head *head)
626 struct i915_vma *vma;
629 list_for_each_entry(vma, head, mm_list) {
630 capture_bo(err++, vma->obj);
638 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
639 int count, struct list_head *head)
641 struct drm_i915_gem_object *obj;
644 list_for_each_entry(obj, head, global_list) {
645 if (!i915_gem_obj_is_pinned(obj))
648 capture_bo(err++, obj);
656 static void i915_gem_record_fences(struct drm_device *dev,
657 struct drm_i915_error_state *error)
659 struct drm_i915_private *dev_priv = dev->dev_private;
663 switch (INTEL_INFO(dev)->gen) {
667 for (i = 0; i < dev_priv->num_fence_regs; i++)
668 error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
672 for (i = 0; i < 16; i++)
673 error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
676 if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
677 for (i = 0; i < 8; i++)
678 error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
680 for (i = 0; i < 8; i++)
681 error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
689 /* This assumes all batchbuffers are executed from the PPGTT. It might have to
690 * change in the future. */
691 static bool is_active_vm(struct i915_address_space *vm,
692 struct intel_ring_buffer *ring)
694 struct drm_device *dev = vm->dev;
695 struct drm_i915_private *dev_priv = dev->dev_private;
696 struct i915_hw_ppgtt *ppgtt;
698 if (INTEL_INFO(dev)->gen < 7)
699 return i915_is_ggtt(vm);
701 /* FIXME: This ignores that the global gtt vm is also on this list. */
702 ppgtt = container_of(vm, struct i915_hw_ppgtt, base);
704 if (INTEL_INFO(dev)->gen >= 8) {
705 u64 pdp0 = (u64)I915_READ(GEN8_RING_PDP_UDW(ring, 0)) << 32;
706 pdp0 |= I915_READ(GEN8_RING_PDP_LDW(ring, 0));
707 return pdp0 == ppgtt->pd_dma_addr[0];
710 pp_db = I915_READ(RING_PP_DIR_BASE(ring));
711 return (pp_db >> 10) == ppgtt->pd_offset;
715 static struct drm_i915_error_object *
716 i915_error_first_batchbuffer(struct drm_i915_private *dev_priv,
717 struct intel_ring_buffer *ring)
719 struct i915_address_space *vm;
720 struct i915_vma *vma;
721 struct drm_i915_gem_object *obj;
722 bool found_active = false;
725 if (!ring->get_seqno)
728 if (HAS_BROKEN_CS_TLB(dev_priv->dev)) {
729 u32 acthd = I915_READ(ACTHD);
731 if (WARN_ON(ring->id != RCS))
734 obj = ring->scratch.obj;
736 acthd >= i915_gem_obj_ggtt_offset(obj) &&
737 acthd < i915_gem_obj_ggtt_offset(obj) + obj->base.size)
738 return i915_error_ggtt_object_create(dev_priv, obj);
741 seqno = ring->get_seqno(ring, false);
742 list_for_each_entry(vm, &dev_priv->vm_list, global_link) {
743 if (!is_active_vm(vm, ring))
748 list_for_each_entry(vma, &vm->active_list, mm_list) {
750 if (obj->ring != ring)
753 if (i915_seqno_passed(seqno, obj->last_read_seqno))
756 if ((obj->base.read_domains & I915_GEM_DOMAIN_COMMAND) == 0)
759 /* We need to copy these to an anonymous buffer as the simplest
760 * method to avoid being overwritten by userspace.
762 return i915_error_object_create(dev_priv, obj, vm);
766 WARN_ON(!found_active);
770 static void i915_record_ring_state(struct drm_device *dev,
771 struct intel_ring_buffer *ring,
772 struct drm_i915_error_ring *ering)
774 struct drm_i915_private *dev_priv = dev->dev_private;
776 if (INTEL_INFO(dev)->gen >= 6) {
777 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
778 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
779 ering->semaphore_mboxes[0]
780 = I915_READ(RING_SYNC_0(ring->mmio_base));
781 ering->semaphore_mboxes[1]
782 = I915_READ(RING_SYNC_1(ring->mmio_base));
783 ering->semaphore_seqno[0] = ring->sync_seqno[0];
784 ering->semaphore_seqno[1] = ring->sync_seqno[1];
787 if (HAS_VEBOX(dev)) {
788 ering->semaphore_mboxes[2] =
789 I915_READ(RING_SYNC_2(ring->mmio_base));
790 ering->semaphore_seqno[2] = ring->sync_seqno[2];
793 if (INTEL_INFO(dev)->gen >= 4) {
794 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
795 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
796 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
797 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
798 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
799 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
800 if (INTEL_INFO(dev)->gen >= 8)
801 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
802 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
804 ering->faddr = I915_READ(DMA_FADD_I8XX);
805 ering->ipeir = I915_READ(IPEIR);
806 ering->ipehr = I915_READ(IPEHR);
807 ering->instdone = I915_READ(INSTDONE);
810 ering->waiting = waitqueue_active(&ring->irq_queue);
811 ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
812 ering->seqno = ring->get_seqno(ring, false);
813 ering->acthd = intel_ring_get_active_head(ring);
814 ering->head = I915_READ_HEAD(ring);
815 ering->tail = I915_READ_TAIL(ring);
816 ering->ctl = I915_READ_CTL(ring);
818 if (I915_NEED_GFX_HWS(dev)) {
825 mmio = RENDER_HWS_PGA_GEN7;
828 mmio = BLT_HWS_PGA_GEN7;
831 mmio = BSD_HWS_PGA_GEN7;
834 mmio = VEBOX_HWS_PGA_GEN7;
837 } else if (IS_GEN6(ring->dev)) {
838 mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
840 /* XXX: gen8 returns to sanity */
841 mmio = RING_HWS_PGA(ring->mmio_base);
844 ering->hws = I915_READ(mmio);
847 ering->cpu_ring_head = ring->head;
848 ering->cpu_ring_tail = ring->tail;
850 ering->hangcheck_score = ring->hangcheck.score;
851 ering->hangcheck_action = ring->hangcheck.action;
853 if (USES_PPGTT(dev)) {
856 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
858 switch (INTEL_INFO(dev)->gen) {
860 for (i = 0; i < 4; i++) {
861 ering->vm_info.pdp[i] =
862 I915_READ(GEN8_RING_PDP_UDW(ring, i));
863 ering->vm_info.pdp[i] <<= 32;
864 ering->vm_info.pdp[i] |=
865 I915_READ(GEN8_RING_PDP_LDW(ring, i));
869 ering->vm_info.pp_dir_base = RING_PP_DIR_BASE(ring);
872 ering->vm_info.pp_dir_base = RING_PP_DIR_BASE_READ(ring);
879 static void i915_gem_record_active_context(struct intel_ring_buffer *ring,
880 struct drm_i915_error_state *error,
881 struct drm_i915_error_ring *ering)
883 struct drm_i915_private *dev_priv = ring->dev->dev_private;
884 struct drm_i915_gem_object *obj;
886 /* Currently render ring is the only HW context user */
887 if (ring->id != RCS || !error->ccid)
890 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
891 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
892 ering->ctx = i915_error_object_create_sized(dev_priv,
901 static void i915_gem_record_rings(struct drm_device *dev,
902 struct drm_i915_error_state *error)
904 struct drm_i915_private *dev_priv = dev->dev_private;
905 struct drm_i915_gem_request *request;
908 for (i = 0; i < I915_NUM_RINGS; i++) {
909 struct intel_ring_buffer *ring = &dev_priv->ring[i];
911 if (ring->dev == NULL)
914 error->ring[i].valid = true;
916 i915_record_ring_state(dev, ring, &error->ring[i]);
918 error->ring[i].batchbuffer =
919 i915_error_first_batchbuffer(dev_priv, ring);
921 error->ring[i].ringbuffer =
922 i915_error_ggtt_object_create(dev_priv, ring->obj);
924 if (ring->status_page.obj)
925 error->ring[i].hws_page =
926 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
928 i915_gem_record_active_context(ring, error, &error->ring[i]);
931 list_for_each_entry(request, &ring->request_list, list)
934 error->ring[i].num_requests = count;
935 error->ring[i].requests =
936 kcalloc(count, sizeof(*error->ring[i].requests),
938 if (error->ring[i].requests == NULL) {
939 error->ring[i].num_requests = 0;
944 list_for_each_entry(request, &ring->request_list, list) {
945 struct drm_i915_error_request *erq;
947 erq = &error->ring[i].requests[count++];
948 erq->seqno = request->seqno;
949 erq->jiffies = request->emitted_jiffies;
950 erq->tail = request->tail;
955 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
958 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
959 struct drm_i915_error_state *error,
960 struct i915_address_space *vm,
963 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
964 struct drm_i915_gem_object *obj;
965 struct i915_vma *vma;
969 list_for_each_entry(vma, &vm->active_list, mm_list)
971 error->active_bo_count[ndx] = i;
972 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
973 if (i915_gem_obj_is_pinned(obj))
975 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
978 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
980 pinned_bo = active_bo + error->active_bo_count[ndx];
984 error->active_bo_count[ndx] =
985 capture_active_bo(active_bo,
986 error->active_bo_count[ndx],
990 error->pinned_bo_count[ndx] =
991 capture_pinned_bo(pinned_bo,
992 error->pinned_bo_count[ndx],
993 &dev_priv->mm.bound_list);
994 error->active_bo[ndx] = active_bo;
995 error->pinned_bo[ndx] = pinned_bo;
998 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
999 struct drm_i915_error_state *error)
1001 struct i915_address_space *vm;
1004 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1007 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1008 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1009 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1011 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1014 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1015 i915_gem_capture_vm(dev_priv, error, vm, i++);
1018 /* Capture all registers which don't fit into another category. */
1019 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1020 struct drm_i915_error_state *error)
1022 struct drm_device *dev = dev_priv->dev;
1025 /* General organization
1026 * 1. Registers specific to a single generation
1027 * 2. Registers which belong to multiple generations
1028 * 3. Feature specific registers.
1029 * 4. Everything else
1030 * Please try to follow the order.
1033 /* 1: Registers specific to a single generation */
1034 if (IS_VALLEYVIEW(dev)) {
1035 error->ier = I915_READ(GTIER) | I915_READ(VLV_IER);
1036 error->forcewake = I915_READ(FORCEWAKE_VLV);
1040 error->err_int = I915_READ(GEN7_ERR_INT);
1043 error->forcewake = I915_READ(FORCEWAKE);
1044 error->gab_ctl = I915_READ(GAB_CTL);
1045 error->gfx_mode = I915_READ(GFX_MODE);
1049 error->ier = I915_READ16(IER);
1051 /* 2: Registers which belong to multiple generations */
1052 if (INTEL_INFO(dev)->gen >= 7)
1053 error->forcewake = I915_READ(FORCEWAKE_MT);
1055 if (INTEL_INFO(dev)->gen >= 6) {
1056 error->derrmr = I915_READ(DERRMR);
1057 error->error = I915_READ(ERROR_GEN6);
1058 error->done_reg = I915_READ(DONE_REG);
1061 /* 3: Feature specific registers */
1062 if (IS_GEN6(dev) || IS_GEN7(dev)) {
1063 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1064 error->gac_eco = I915_READ(GAC_ECO_BITS);
1067 /* 4: Everything else */
1068 if (HAS_HW_CONTEXTS(dev))
1069 error->ccid = I915_READ(CCID);
1071 if (HAS_PCH_SPLIT(dev))
1072 error->ier = I915_READ(DEIER) | I915_READ(GTIER);
1074 error->ier = I915_READ(IER);
1076 error->pipestat[pipe] = I915_READ(PIPESTAT(pipe));
1079 /* 4: Everything else */
1080 error->eir = I915_READ(EIR);
1081 error->pgtbl_er = I915_READ(PGTBL_ER);
1083 i915_get_extra_instdone(dev, error->extra_instdone);
1087 * i915_capture_error_state - capture an error record for later analysis
1090 * Should be called when an error is detected (either a hang or an error
1091 * interrupt) to capture error state from the time of the error. Fills
1092 * out a structure which becomes available in debugfs for user level tools
1095 void i915_capture_error_state(struct drm_device *dev)
1098 struct drm_i915_private *dev_priv = dev->dev_private;
1099 struct drm_i915_error_state *error;
1100 unsigned long flags;
1102 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1103 error = dev_priv->gpu_error.first_error;
1104 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1108 /* Account for pipe specific data like PIPE*STAT */
1109 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1111 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1115 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1116 dev->primary->index);
1118 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1119 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1120 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1121 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1125 kref_init(&error->ref);
1127 i915_capture_reg_state(dev_priv, error);
1128 i915_gem_capture_buffers(dev_priv, error);
1129 i915_gem_record_fences(dev, error);
1130 i915_gem_record_rings(dev, error);
1132 do_gettimeofday(&error->time);
1134 error->overlay = intel_overlay_capture_error_state(dev);
1135 error->display = intel_display_capture_error_state(dev);
1137 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1138 if (dev_priv->gpu_error.first_error == NULL) {
1139 dev_priv->gpu_error.first_error = error;
1142 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1145 i915_error_state_free(&error->ref);
1148 void i915_error_state_get(struct drm_device *dev,
1149 struct i915_error_state_file_priv *error_priv)
1151 struct drm_i915_private *dev_priv = dev->dev_private;
1152 unsigned long flags;
1154 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1155 error_priv->error = dev_priv->gpu_error.first_error;
1156 if (error_priv->error)
1157 kref_get(&error_priv->error->ref);
1158 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1162 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1164 if (error_priv->error)
1165 kref_put(&error_priv->error->ref, i915_error_state_free);
1168 void i915_destroy_error_state(struct drm_device *dev)
1170 struct drm_i915_private *dev_priv = dev->dev_private;
1171 struct drm_i915_error_state *error;
1172 unsigned long flags;
1174 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1175 error = dev_priv->gpu_error.first_error;
1176 dev_priv->gpu_error.first_error = NULL;
1177 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1180 kref_put(&error->ref, i915_error_state_free);
1183 const char *i915_cache_level_str(int type)
1186 case I915_CACHE_NONE: return " uncached";
1187 case I915_CACHE_LLC: return " snooped or LLC";
1188 case I915_CACHE_L3_LLC: return " L3+LLC";
1189 case I915_CACHE_WT: return " WT";
1194 /* NB: please notice the memset */
1195 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1197 struct drm_i915_private *dev_priv = dev->dev_private;
1198 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1200 switch (INTEL_INFO(dev)->gen) {
1203 instdone[0] = I915_READ(INSTDONE);
1208 instdone[0] = I915_READ(INSTDONE_I965);
1209 instdone[1] = I915_READ(INSTDONE1);
1212 WARN_ONCE(1, "Unsupported platform\n");
1215 instdone[0] = I915_READ(GEN7_INSTDONE_1);
1216 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1217 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1218 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);