2 * Copyright (c) 2008 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Eric Anholt <eric@anholt.net>
25 * Keith Packard <keithp@keithp.com>
26 * Mika Kuoppala <mika.kuoppala@intel.com>
30 #include <generated/utsrelease.h>
33 static const char *yesno(int v)
35 return v ? "yes" : "no";
38 static const char *ring_str(int ring)
41 case RCS: return "render";
42 case VCS: return "bsd";
43 case BCS: return "blt";
44 case VECS: return "vebox";
45 case VCS2: return "bsd2";
50 static const char *pin_flag(int pinned)
60 static const char *tiling_flag(int tiling)
64 case I915_TILING_NONE: return "";
65 case I915_TILING_X: return " X";
66 case I915_TILING_Y: return " Y";
70 static const char *dirty_flag(int dirty)
72 return dirty ? " dirty" : "";
75 static const char *purgeable_flag(int purgeable)
77 return purgeable ? " purgeable" : "";
80 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
83 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
88 if (e->bytes == e->size - 1 || e->err)
94 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
97 if (e->pos + len <= e->start) {
102 /* First vsnprintf needs to fit in its entirety for memmove */
103 if (len >= e->size) {
111 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
114 /* If this is first printf in this window, adjust it so that
115 * start position matches start of the buffer
118 if (e->pos < e->start) {
119 const size_t off = e->start - e->pos;
121 /* Should not happen but be paranoid */
122 if (off > len || e->bytes) {
127 memmove(e->buf, e->buf + off, len - off);
128 e->bytes = len - off;
137 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
138 const char *f, va_list args)
142 if (!__i915_error_ok(e))
145 /* Seek the first printf which is hits start position */
146 if (e->pos < e->start) {
150 len = vsnprintf(NULL, 0, f, tmp);
153 if (!__i915_error_seek(e, len))
157 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
158 if (len >= e->size - e->bytes)
159 len = e->size - e->bytes - 1;
161 __i915_error_advance(e, len);
164 static void i915_error_puts(struct drm_i915_error_state_buf *e,
169 if (!__i915_error_ok(e))
174 /* Seek the first printf which is hits start position */
175 if (e->pos < e->start) {
176 if (!__i915_error_seek(e, len))
180 if (len >= e->size - e->bytes)
181 len = e->size - e->bytes - 1;
182 memcpy(e->buf + e->bytes, str, len);
184 __i915_error_advance(e, len);
187 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
188 #define err_puts(e, s) i915_error_puts(e, s)
190 static void print_error_buffers(struct drm_i915_error_state_buf *m,
192 struct drm_i915_error_buffer *err,
195 err_printf(m, "%s [%d]:\n", name, count);
198 err_printf(m, " %08x %8u %02x %02x %x %x",
203 err->rseqno, err->wseqno);
204 err_puts(m, pin_flag(err->pinned));
205 err_puts(m, tiling_flag(err->tiling));
206 err_puts(m, dirty_flag(err->dirty));
207 err_puts(m, purgeable_flag(err->purgeable));
208 err_puts(m, err->userptr ? " userptr" : "");
209 err_puts(m, err->ring != -1 ? " " : "");
210 err_puts(m, ring_str(err->ring));
211 err_puts(m, i915_cache_level_str(err->cache_level));
214 err_printf(m, " (name: %d)", err->name);
215 if (err->fence_reg != I915_FENCE_REG_NONE)
216 err_printf(m, " (fence: %d)", err->fence_reg);
223 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
230 case HANGCHECK_ACTIVE:
232 case HANGCHECK_ACTIVE_LOOP:
233 return "active (loop)";
243 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
244 struct drm_device *dev,
245 struct drm_i915_error_ring *ring)
250 err_printf(m, " HEAD: 0x%08x\n", ring->head);
251 err_printf(m, " TAIL: 0x%08x\n", ring->tail);
252 err_printf(m, " CTL: 0x%08x\n", ring->ctl);
253 err_printf(m, " HWS: 0x%08x\n", ring->hws);
254 err_printf(m, " ACTHD: 0x%08x %08x\n", (u32)(ring->acthd>>32), (u32)ring->acthd);
255 err_printf(m, " IPEIR: 0x%08x\n", ring->ipeir);
256 err_printf(m, " IPEHR: 0x%08x\n", ring->ipehr);
257 err_printf(m, " INSTDONE: 0x%08x\n", ring->instdone);
258 if (INTEL_INFO(dev)->gen >= 4) {
259 err_printf(m, " BBADDR: 0x%08x %08x\n", (u32)(ring->bbaddr>>32), (u32)ring->bbaddr);
260 err_printf(m, " BB_STATE: 0x%08x\n", ring->bbstate);
261 err_printf(m, " INSTPS: 0x%08x\n", ring->instps);
263 err_printf(m, " INSTPM: 0x%08x\n", ring->instpm);
264 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ring->faddr),
265 lower_32_bits(ring->faddr));
266 if (INTEL_INFO(dev)->gen >= 6) {
267 err_printf(m, " RC PSMI: 0x%08x\n", ring->rc_psmi);
268 err_printf(m, " FAULT_REG: 0x%08x\n", ring->fault_reg);
269 err_printf(m, " SYNC_0: 0x%08x [last synced 0x%08x]\n",
270 ring->semaphore_mboxes[0],
271 ring->semaphore_seqno[0]);
272 err_printf(m, " SYNC_1: 0x%08x [last synced 0x%08x]\n",
273 ring->semaphore_mboxes[1],
274 ring->semaphore_seqno[1]);
275 if (HAS_VEBOX(dev)) {
276 err_printf(m, " SYNC_2: 0x%08x [last synced 0x%08x]\n",
277 ring->semaphore_mboxes[2],
278 ring->semaphore_seqno[2]);
281 if (USES_PPGTT(dev)) {
282 err_printf(m, " GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
284 if (INTEL_INFO(dev)->gen >= 8) {
286 for (i = 0; i < 4; i++)
287 err_printf(m, " PDP%d: 0x%016llx\n",
288 i, ring->vm_info.pdp[i]);
290 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
291 ring->vm_info.pp_dir_base);
294 err_printf(m, " seqno: 0x%08x\n", ring->seqno);
295 err_printf(m, " waiting: %s\n", yesno(ring->waiting));
296 err_printf(m, " ring->head: 0x%08x\n", ring->cpu_ring_head);
297 err_printf(m, " ring->tail: 0x%08x\n", ring->cpu_ring_tail);
298 err_printf(m, " hangcheck: %s [%d]\n",
299 hangcheck_action_to_str(ring->hangcheck_action),
300 ring->hangcheck_score);
303 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
308 i915_error_vprintf(e, f, args);
312 static void print_error_obj(struct drm_i915_error_state_buf *m,
313 struct drm_i915_error_object *obj)
315 int page, offset, elt;
317 for (page = offset = 0; page < obj->page_count; page++) {
318 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
319 err_printf(m, "%08x : %08x\n", offset,
320 obj->pages[page][elt]);
326 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
327 const struct i915_error_state_file_priv *error_priv)
329 struct drm_device *dev = error_priv->dev;
330 struct drm_i915_private *dev_priv = dev->dev_private;
331 struct drm_i915_error_state *error = error_priv->error;
332 struct drm_i915_error_object *obj;
333 int i, j, offset, elt;
334 int max_hangcheck_score;
337 err_printf(m, "no error state collected\n");
341 err_printf(m, "%s\n", error->error_msg);
342 err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
343 error->time.tv_usec);
344 err_printf(m, "Kernel: " UTS_RELEASE "\n");
345 max_hangcheck_score = 0;
346 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
347 if (error->ring[i].hangcheck_score > max_hangcheck_score)
348 max_hangcheck_score = error->ring[i].hangcheck_score;
350 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
351 if (error->ring[i].hangcheck_score == max_hangcheck_score &&
352 error->ring[i].pid != -1) {
353 err_printf(m, "Active process (on ring %s): %s [%d]\n",
359 err_printf(m, "Reset count: %u\n", error->reset_count);
360 err_printf(m, "Suspend count: %u\n", error->suspend_count);
361 err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
362 err_printf(m, "EIR: 0x%08x\n", error->eir);
363 err_printf(m, "IER: 0x%08x\n", error->ier);
364 if (INTEL_INFO(dev)->gen >= 8) {
365 for (i = 0; i < 4; i++)
366 err_printf(m, "GTIER gt %d: 0x%08x\n", i,
368 } else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
369 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
370 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
371 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
372 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
373 err_printf(m, "CCID: 0x%08x\n", error->ccid);
374 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
376 for (i = 0; i < dev_priv->num_fence_regs; i++)
377 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
379 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
380 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
381 error->extra_instdone[i]);
383 if (INTEL_INFO(dev)->gen >= 6) {
384 err_printf(m, "ERROR: 0x%08x\n", error->error);
385 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
388 if (INTEL_INFO(dev)->gen == 7)
389 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
391 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
392 err_printf(m, "%s command stream:\n", ring_str(i));
393 i915_ring_error_state(m, dev, &error->ring[i]);
396 if (error->active_bo)
397 print_error_buffers(m, "Active",
399 error->active_bo_count[0]);
401 if (error->pinned_bo)
402 print_error_buffers(m, "Pinned",
404 error->pinned_bo_count[0]);
406 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
407 obj = error->ring[i].batchbuffer;
409 err_puts(m, dev_priv->ring[i].name);
410 if (error->ring[i].pid != -1)
411 err_printf(m, " (submitted by %s [%d])",
414 err_printf(m, " --- gtt_offset = 0x%08x\n",
416 print_error_obj(m, obj);
419 obj = error->ring[i].wa_batchbuffer;
421 err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
422 dev_priv->ring[i].name, obj->gtt_offset);
423 print_error_obj(m, obj);
426 if (error->ring[i].num_requests) {
427 err_printf(m, "%s --- %d requests\n",
428 dev_priv->ring[i].name,
429 error->ring[i].num_requests);
430 for (j = 0; j < error->ring[i].num_requests; j++) {
431 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
432 error->ring[i].requests[j].seqno,
433 error->ring[i].requests[j].jiffies,
434 error->ring[i].requests[j].tail);
438 if ((obj = error->ring[i].ringbuffer)) {
439 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
440 dev_priv->ring[i].name,
442 print_error_obj(m, obj);
445 if ((obj = error->ring[i].hws_page)) {
446 err_printf(m, "%s --- HW Status = 0x%08x\n",
447 dev_priv->ring[i].name,
450 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
451 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
454 obj->pages[0][elt+1],
455 obj->pages[0][elt+2],
456 obj->pages[0][elt+3]);
461 if ((obj = error->ring[i].ctx)) {
462 err_printf(m, "%s --- HW Context = 0x%08x\n",
463 dev_priv->ring[i].name,
465 print_error_obj(m, obj);
469 if ((obj = error->semaphore_obj)) {
470 err_printf(m, "Semaphore page = 0x%08x\n", obj->gtt_offset);
471 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
472 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
475 obj->pages[0][elt+1],
476 obj->pages[0][elt+2],
477 obj->pages[0][elt+3]);
482 intel_overlay_print_error_state(m, error->overlay);
485 intel_display_print_error_state(m, dev, error->display);
488 if (m->bytes == 0 && m->err)
494 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
495 size_t count, loff_t pos)
497 memset(ebuf, 0, sizeof(*ebuf));
499 /* We need to have enough room to store any i915_error_state printf
500 * so that we can move it to start position.
502 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
503 ebuf->buf = kmalloc(ebuf->size,
504 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
506 if (ebuf->buf == NULL) {
507 ebuf->size = PAGE_SIZE;
508 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
511 if (ebuf->buf == NULL) {
513 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
516 if (ebuf->buf == NULL)
524 static void i915_error_object_free(struct drm_i915_error_object *obj)
531 for (page = 0; page < obj->page_count; page++)
532 kfree(obj->pages[page]);
537 static void i915_error_state_free(struct kref *error_ref)
539 struct drm_i915_error_state *error = container_of(error_ref,
540 typeof(*error), ref);
543 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
544 i915_error_object_free(error->ring[i].batchbuffer);
545 i915_error_object_free(error->ring[i].ringbuffer);
546 i915_error_object_free(error->ring[i].hws_page);
547 i915_error_object_free(error->ring[i].ctx);
548 kfree(error->ring[i].requests);
551 i915_error_object_free(error->semaphore_obj);
552 kfree(error->active_bo);
553 kfree(error->overlay);
554 kfree(error->display);
558 static struct drm_i915_error_object *
559 i915_error_object_create_sized(struct drm_i915_private *dev_priv,
560 struct drm_i915_gem_object *src,
561 struct i915_address_space *vm,
564 struct drm_i915_error_object *dst;
568 if (src == NULL || src->pages == NULL)
571 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
575 reloc_offset = dst->gtt_offset = i915_gem_obj_offset(src, vm);
576 for (i = 0; i < num_pages; i++) {
580 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
584 local_irq_save(flags);
585 if (src->cache_level == I915_CACHE_NONE &&
586 reloc_offset < dev_priv->gtt.mappable_end &&
587 src->has_global_gtt_mapping &&
591 /* Simply ignore tiling or any overlapping fence.
592 * It's part of the error state, and this hopefully
593 * captures what the GPU read.
596 s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
598 memcpy_fromio(d, s, PAGE_SIZE);
599 io_mapping_unmap_atomic(s);
600 } else if (src->stolen) {
601 unsigned long offset;
603 offset = dev_priv->mm.stolen_base;
604 offset += src->stolen->start;
605 offset += i << PAGE_SHIFT;
607 memcpy_fromio(d, (void __iomem *) offset, PAGE_SIZE);
612 page = i915_gem_object_get_page(src, i);
614 drm_clflush_pages(&page, 1);
616 s = kmap_atomic(page);
617 memcpy(d, s, PAGE_SIZE);
620 drm_clflush_pages(&page, 1);
622 local_irq_restore(flags);
626 reloc_offset += PAGE_SIZE;
628 dst->page_count = num_pages;
634 kfree(dst->pages[i]);
638 #define i915_error_object_create(dev_priv, src, vm) \
639 i915_error_object_create_sized((dev_priv), (src), (vm), \
640 (src)->base.size>>PAGE_SHIFT)
642 #define i915_error_ggtt_object_create(dev_priv, src) \
643 i915_error_object_create_sized((dev_priv), (src), &(dev_priv)->gtt.base, \
644 (src)->base.size>>PAGE_SHIFT)
646 static void capture_bo(struct drm_i915_error_buffer *err,
647 struct drm_i915_gem_object *obj)
649 err->size = obj->base.size;
650 err->name = obj->base.name;
651 err->rseqno = obj->last_read_seqno;
652 err->wseqno = obj->last_write_seqno;
653 err->gtt_offset = i915_gem_obj_ggtt_offset(obj);
654 err->read_domains = obj->base.read_domains;
655 err->write_domain = obj->base.write_domain;
656 err->fence_reg = obj->fence_reg;
658 if (i915_gem_obj_is_pinned(obj))
660 if (obj->user_pin_count > 0)
662 err->tiling = obj->tiling_mode;
663 err->dirty = obj->dirty;
664 err->purgeable = obj->madv != I915_MADV_WILLNEED;
665 err->userptr = obj->userptr.mm != NULL;
666 err->ring = obj->ring ? obj->ring->id : -1;
667 err->cache_level = obj->cache_level;
670 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
671 int count, struct list_head *head)
673 struct i915_vma *vma;
676 list_for_each_entry(vma, head, mm_list) {
677 capture_bo(err++, vma->obj);
685 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
686 int count, struct list_head *head)
688 struct drm_i915_gem_object *obj;
691 list_for_each_entry(obj, head, global_list) {
692 if (!i915_gem_obj_is_pinned(obj))
695 capture_bo(err++, obj);
703 /* Generate a semi-unique error code. The code is not meant to have meaning, The
704 * code's only purpose is to try to prevent false duplicated bug reports by
705 * grossly estimating a GPU error state.
707 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
708 * the hang if we could strip the GTT offset information from it.
710 * It's only a small step better than a random number in its current form.
712 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
713 struct drm_i915_error_state *error,
716 uint32_t error_code = 0;
719 /* IPEHR would be an ideal way to detect errors, as it's the gross
720 * measure of "the command that hung." However, has some very common
721 * synchronization commands which almost always appear in the case
722 * strictly a client bug. Use instdone to differentiate those some.
724 for (i = 0; i < I915_NUM_RINGS; i++) {
725 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
729 return error->ring[i].ipehr ^ error->ring[i].instdone;
736 static void i915_gem_record_fences(struct drm_device *dev,
737 struct drm_i915_error_state *error)
739 struct drm_i915_private *dev_priv = dev->dev_private;
743 switch (INTEL_INFO(dev)->gen) {
747 for (i = 0; i < dev_priv->num_fence_regs; i++)
748 error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
752 for (i = 0; i < 16; i++)
753 error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
756 if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
757 for (i = 0; i < 8; i++)
758 error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
760 for (i = 0; i < 8; i++)
761 error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
770 static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
771 struct drm_i915_error_state *error,
772 struct intel_engine_cs *ring,
773 struct drm_i915_error_ring *ering)
775 struct intel_engine_cs *to;
778 if (!i915_semaphore_is_enabled(dev_priv->dev))
781 if (!error->semaphore_obj)
782 error->semaphore_obj =
783 i915_error_object_create(dev_priv,
784 dev_priv->semaphore_obj,
785 &dev_priv->gtt.base);
787 for_each_ring(to, dev_priv, i) {
795 signal_offset = (GEN8_SIGNAL_OFFSET(ring, i) & (PAGE_SIZE - 1))
797 tmp = error->semaphore_obj->pages[0];
798 idx = intel_ring_sync_index(ring, to);
800 ering->semaphore_mboxes[idx] = tmp[signal_offset];
801 ering->semaphore_seqno[idx] = ring->semaphore.sync_seqno[idx];
805 static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
806 struct intel_engine_cs *ring,
807 struct drm_i915_error_ring *ering)
809 ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(ring->mmio_base));
810 ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(ring->mmio_base));
811 ering->semaphore_seqno[0] = ring->semaphore.sync_seqno[0];
812 ering->semaphore_seqno[1] = ring->semaphore.sync_seqno[1];
814 if (HAS_VEBOX(dev_priv->dev)) {
815 ering->semaphore_mboxes[2] =
816 I915_READ(RING_SYNC_2(ring->mmio_base));
817 ering->semaphore_seqno[2] = ring->semaphore.sync_seqno[2];
821 static void i915_record_ring_state(struct drm_device *dev,
822 struct drm_i915_error_state *error,
823 struct intel_engine_cs *ring,
824 struct drm_i915_error_ring *ering)
826 struct drm_i915_private *dev_priv = dev->dev_private;
828 if (INTEL_INFO(dev)->gen >= 6) {
829 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
830 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
831 if (INTEL_INFO(dev)->gen >= 8)
832 gen8_record_semaphore_state(dev_priv, error, ring, ering);
834 gen6_record_semaphore_state(dev_priv, ring, ering);
837 if (INTEL_INFO(dev)->gen >= 4) {
838 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
839 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
840 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
841 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
842 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
843 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
844 if (INTEL_INFO(dev)->gen >= 8) {
845 ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(ring->mmio_base)) << 32;
846 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
848 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
850 ering->faddr = I915_READ(DMA_FADD_I8XX);
851 ering->ipeir = I915_READ(IPEIR);
852 ering->ipehr = I915_READ(IPEHR);
853 ering->instdone = I915_READ(INSTDONE);
856 ering->waiting = waitqueue_active(&ring->irq_queue);
857 ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
858 ering->seqno = ring->get_seqno(ring, false);
859 ering->acthd = intel_ring_get_active_head(ring);
860 ering->head = I915_READ_HEAD(ring);
861 ering->tail = I915_READ_TAIL(ring);
862 ering->ctl = I915_READ_CTL(ring);
864 if (I915_NEED_GFX_HWS(dev)) {
871 mmio = RENDER_HWS_PGA_GEN7;
874 mmio = BLT_HWS_PGA_GEN7;
877 mmio = BSD_HWS_PGA_GEN7;
880 mmio = VEBOX_HWS_PGA_GEN7;
883 } else if (IS_GEN6(ring->dev)) {
884 mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
886 /* XXX: gen8 returns to sanity */
887 mmio = RING_HWS_PGA(ring->mmio_base);
890 ering->hws = I915_READ(mmio);
893 ering->cpu_ring_head = ring->buffer->head;
894 ering->cpu_ring_tail = ring->buffer->tail;
896 ering->hangcheck_score = ring->hangcheck.score;
897 ering->hangcheck_action = ring->hangcheck.action;
899 if (USES_PPGTT(dev)) {
902 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
904 switch (INTEL_INFO(dev)->gen) {
906 for (i = 0; i < 4; i++) {
907 ering->vm_info.pdp[i] =
908 I915_READ(GEN8_RING_PDP_UDW(ring, i));
909 ering->vm_info.pdp[i] <<= 32;
910 ering->vm_info.pdp[i] |=
911 I915_READ(GEN8_RING_PDP_LDW(ring, i));
915 ering->vm_info.pp_dir_base =
916 I915_READ(RING_PP_DIR_BASE(ring));
919 ering->vm_info.pp_dir_base =
920 I915_READ(RING_PP_DIR_BASE_READ(ring));
927 static void i915_gem_record_active_context(struct intel_engine_cs *ring,
928 struct drm_i915_error_state *error,
929 struct drm_i915_error_ring *ering)
931 struct drm_i915_private *dev_priv = ring->dev->dev_private;
932 struct drm_i915_gem_object *obj;
934 /* Currently render ring is the only HW context user */
935 if (ring->id != RCS || !error->ccid)
938 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
939 if (!i915_gem_obj_ggtt_bound(obj))
942 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
943 ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
949 static void i915_gem_record_rings(struct drm_device *dev,
950 struct drm_i915_error_state *error)
952 struct drm_i915_private *dev_priv = dev->dev_private;
953 struct drm_i915_gem_request *request;
956 for (i = 0; i < I915_NUM_RINGS; i++) {
957 struct intel_engine_cs *ring = &dev_priv->ring[i];
959 error->ring[i].pid = -1;
961 if (ring->dev == NULL)
964 error->ring[i].valid = true;
966 i915_record_ring_state(dev, error, ring, &error->ring[i]);
968 request = i915_gem_find_active_request(ring);
970 /* We need to copy these to an anonymous buffer
971 * as the simplest method to avoid being overwritten
974 error->ring[i].batchbuffer =
975 i915_error_object_create(dev_priv,
979 &dev_priv->gtt.base);
981 if (HAS_BROKEN_CS_TLB(dev_priv->dev) &&
983 error->ring[i].wa_batchbuffer =
984 i915_error_ggtt_object_create(dev_priv,
987 if (request->file_priv) {
988 struct task_struct *task;
991 task = pid_task(request->file_priv->file->pid,
994 strcpy(error->ring[i].comm, task->comm);
995 error->ring[i].pid = task->pid;
1001 error->ring[i].ringbuffer =
1002 i915_error_ggtt_object_create(dev_priv, ring->buffer->obj);
1004 if (ring->status_page.obj)
1005 error->ring[i].hws_page =
1006 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
1008 i915_gem_record_active_context(ring, error, &error->ring[i]);
1011 list_for_each_entry(request, &ring->request_list, list)
1014 error->ring[i].num_requests = count;
1015 error->ring[i].requests =
1016 kcalloc(count, sizeof(*error->ring[i].requests),
1018 if (error->ring[i].requests == NULL) {
1019 error->ring[i].num_requests = 0;
1024 list_for_each_entry(request, &ring->request_list, list) {
1025 struct drm_i915_error_request *erq;
1027 erq = &error->ring[i].requests[count++];
1028 erq->seqno = request->seqno;
1029 erq->jiffies = request->emitted_jiffies;
1030 erq->tail = request->tail;
1035 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1038 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1039 struct drm_i915_error_state *error,
1040 struct i915_address_space *vm,
1043 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1044 struct drm_i915_gem_object *obj;
1045 struct i915_vma *vma;
1049 list_for_each_entry(vma, &vm->active_list, mm_list)
1051 error->active_bo_count[ndx] = i;
1052 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
1053 if (i915_gem_obj_is_pinned(obj))
1055 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1058 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1060 pinned_bo = active_bo + error->active_bo_count[ndx];
1064 error->active_bo_count[ndx] =
1065 capture_active_bo(active_bo,
1066 error->active_bo_count[ndx],
1070 error->pinned_bo_count[ndx] =
1071 capture_pinned_bo(pinned_bo,
1072 error->pinned_bo_count[ndx],
1073 &dev_priv->mm.bound_list);
1074 error->active_bo[ndx] = active_bo;
1075 error->pinned_bo[ndx] = pinned_bo;
1078 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1079 struct drm_i915_error_state *error)
1081 struct i915_address_space *vm;
1084 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1087 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1088 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1089 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1091 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1094 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1095 i915_gem_capture_vm(dev_priv, error, vm, i++);
1098 /* Capture all registers which don't fit into another category. */
1099 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1100 struct drm_i915_error_state *error)
1102 struct drm_device *dev = dev_priv->dev;
1105 /* General organization
1106 * 1. Registers specific to a single generation
1107 * 2. Registers which belong to multiple generations
1108 * 3. Feature specific registers.
1109 * 4. Everything else
1110 * Please try to follow the order.
1113 /* 1: Registers specific to a single generation */
1114 if (IS_VALLEYVIEW(dev)) {
1115 error->gtier[0] = I915_READ(GTIER);
1116 error->ier = I915_READ(VLV_IER);
1117 error->forcewake = I915_READ(FORCEWAKE_VLV);
1121 error->err_int = I915_READ(GEN7_ERR_INT);
1124 error->forcewake = I915_READ(FORCEWAKE);
1125 error->gab_ctl = I915_READ(GAB_CTL);
1126 error->gfx_mode = I915_READ(GFX_MODE);
1129 /* 2: Registers which belong to multiple generations */
1130 if (INTEL_INFO(dev)->gen >= 7)
1131 error->forcewake = I915_READ(FORCEWAKE_MT);
1133 if (INTEL_INFO(dev)->gen >= 6) {
1134 error->derrmr = I915_READ(DERRMR);
1135 error->error = I915_READ(ERROR_GEN6);
1136 error->done_reg = I915_READ(DONE_REG);
1139 /* 3: Feature specific registers */
1140 if (IS_GEN6(dev) || IS_GEN7(dev)) {
1141 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1142 error->gac_eco = I915_READ(GAC_ECO_BITS);
1145 /* 4: Everything else */
1146 if (HAS_HW_CONTEXTS(dev))
1147 error->ccid = I915_READ(CCID);
1149 if (INTEL_INFO(dev)->gen >= 8) {
1150 error->ier = I915_READ(GEN8_DE_MISC_IER);
1151 for (i = 0; i < 4; i++)
1152 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1153 } else if (HAS_PCH_SPLIT(dev)) {
1154 error->ier = I915_READ(DEIER);
1155 error->gtier[0] = I915_READ(GTIER);
1156 } else if (IS_GEN2(dev)) {
1157 error->ier = I915_READ16(IER);
1158 } else if (!IS_VALLEYVIEW(dev)) {
1159 error->ier = I915_READ(IER);
1161 error->eir = I915_READ(EIR);
1162 error->pgtbl_er = I915_READ(PGTBL_ER);
1164 i915_get_extra_instdone(dev, error->extra_instdone);
1167 static void i915_error_capture_msg(struct drm_device *dev,
1168 struct drm_i915_error_state *error,
1170 const char *error_msg)
1172 struct drm_i915_private *dev_priv = dev->dev_private;
1174 int ring_id = -1, len;
1176 ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1178 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1179 "GPU HANG: ecode %d:0x%08x", ring_id, ecode);
1181 if (ring_id != -1 && error->ring[ring_id].pid != -1)
1182 len += scnprintf(error->error_msg + len,
1183 sizeof(error->error_msg) - len,
1185 error->ring[ring_id].comm,
1186 error->ring[ring_id].pid);
1188 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1189 ", reason: %s, action: %s",
1191 wedged ? "reset" : "continue");
1194 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1195 struct drm_i915_error_state *error)
1197 error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1198 error->suspend_count = dev_priv->suspend_count;
1202 * i915_capture_error_state - capture an error record for later analysis
1205 * Should be called when an error is detected (either a hang or an error
1206 * interrupt) to capture error state from the time of the error. Fills
1207 * out a structure which becomes available in debugfs for user level tools
1210 void i915_capture_error_state(struct drm_device *dev, bool wedged,
1211 const char *error_msg)
1214 struct drm_i915_private *dev_priv = dev->dev_private;
1215 struct drm_i915_error_state *error;
1216 unsigned long flags;
1218 /* Account for pipe specific data like PIPE*STAT */
1219 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1221 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1225 kref_init(&error->ref);
1227 i915_capture_gen_state(dev_priv, error);
1228 i915_capture_reg_state(dev_priv, error);
1229 i915_gem_capture_buffers(dev_priv, error);
1230 i915_gem_record_fences(dev, error);
1231 i915_gem_record_rings(dev, error);
1233 do_gettimeofday(&error->time);
1235 error->overlay = intel_overlay_capture_error_state(dev);
1236 error->display = intel_display_capture_error_state(dev);
1238 i915_error_capture_msg(dev, error, wedged, error_msg);
1239 DRM_INFO("%s\n", error->error_msg);
1241 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1242 if (dev_priv->gpu_error.first_error == NULL) {
1243 dev_priv->gpu_error.first_error = error;
1246 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1249 i915_error_state_free(&error->ref);
1254 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1255 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1256 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1257 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1258 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1263 void i915_error_state_get(struct drm_device *dev,
1264 struct i915_error_state_file_priv *error_priv)
1266 struct drm_i915_private *dev_priv = dev->dev_private;
1267 unsigned long flags;
1269 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1270 error_priv->error = dev_priv->gpu_error.first_error;
1271 if (error_priv->error)
1272 kref_get(&error_priv->error->ref);
1273 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1277 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1279 if (error_priv->error)
1280 kref_put(&error_priv->error->ref, i915_error_state_free);
1283 void i915_destroy_error_state(struct drm_device *dev)
1285 struct drm_i915_private *dev_priv = dev->dev_private;
1286 struct drm_i915_error_state *error;
1287 unsigned long flags;
1289 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1290 error = dev_priv->gpu_error.first_error;
1291 dev_priv->gpu_error.first_error = NULL;
1292 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1295 kref_put(&error->ref, i915_error_state_free);
1298 const char *i915_cache_level_str(int type)
1301 case I915_CACHE_NONE: return " uncached";
1302 case I915_CACHE_LLC: return " snooped or LLC";
1303 case I915_CACHE_L3_LLC: return " L3+LLC";
1304 case I915_CACHE_WT: return " WT";
1309 /* NB: please notice the memset */
1310 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1312 struct drm_i915_private *dev_priv = dev->dev_private;
1313 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1315 switch (INTEL_INFO(dev)->gen) {
1318 instdone[0] = I915_READ(INSTDONE);
1323 instdone[0] = I915_READ(INSTDONE_I965);
1324 instdone[1] = I915_READ(INSTDONE1);
1327 WARN_ONCE(1, "Unsupported platform\n");
1330 instdone[0] = I915_READ(GEN7_INSTDONE_1);
1331 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1332 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1333 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);