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 (HAS_PCH_SPLIT(dev
) || IS_VALLEYVIEW(dev
))
365 err_printf(m
, "GTIER: 0x%08x\n", error
->gtier
);
366 err_printf(m
, "PGTBL_ER: 0x%08x\n", error
->pgtbl_er
);
367 err_printf(m
, "FORCEWAKE: 0x%08x\n", error
->forcewake
);
368 err_printf(m
, "DERRMR: 0x%08x\n", error
->derrmr
);
369 err_printf(m
, "CCID: 0x%08x\n", error
->ccid
);
370 err_printf(m
, "Missed interrupts: 0x%08lx\n", dev_priv
->gpu_error
.missed_irq_rings
);
372 for (i
= 0; i
< dev_priv
->num_fence_regs
; i
++)
373 err_printf(m
, " fence[%d] = %08llx\n", i
, error
->fence
[i
]);
375 for (i
= 0; i
< ARRAY_SIZE(error
->extra_instdone
); i
++)
376 err_printf(m
, " INSTDONE_%d: 0x%08x\n", i
,
377 error
->extra_instdone
[i
]);
379 if (INTEL_INFO(dev
)->gen
>= 6) {
380 err_printf(m
, "ERROR: 0x%08x\n", error
->error
);
381 err_printf(m
, "DONE_REG: 0x%08x\n", error
->done_reg
);
384 if (INTEL_INFO(dev
)->gen
== 7)
385 err_printf(m
, "ERR_INT: 0x%08x\n", error
->err_int
);
387 for (i
= 0; i
< ARRAY_SIZE(error
->ring
); i
++) {
388 err_printf(m
, "%s command stream:\n", ring_str(i
));
389 i915_ring_error_state(m
, dev
, &error
->ring
[i
]);
392 if (error
->active_bo
)
393 print_error_buffers(m
, "Active",
395 error
->active_bo_count
[0]);
397 if (error
->pinned_bo
)
398 print_error_buffers(m
, "Pinned",
400 error
->pinned_bo_count
[0]);
402 for (i
= 0; i
< ARRAY_SIZE(error
->ring
); i
++) {
403 obj
= error
->ring
[i
].batchbuffer
;
405 err_puts(m
, dev_priv
->ring
[i
].name
);
406 if (error
->ring
[i
].pid
!= -1)
407 err_printf(m
, " (submitted by %s [%d])",
410 err_printf(m
, " --- gtt_offset = 0x%08x\n",
412 print_error_obj(m
, obj
);
415 obj
= error
->ring
[i
].wa_batchbuffer
;
417 err_printf(m
, "%s (w/a) --- gtt_offset = 0x%08x\n",
418 dev_priv
->ring
[i
].name
, obj
->gtt_offset
);
419 print_error_obj(m
, obj
);
422 if (error
->ring
[i
].num_requests
) {
423 err_printf(m
, "%s --- %d requests\n",
424 dev_priv
->ring
[i
].name
,
425 error
->ring
[i
].num_requests
);
426 for (j
= 0; j
< error
->ring
[i
].num_requests
; j
++) {
427 err_printf(m
, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
428 error
->ring
[i
].requests
[j
].seqno
,
429 error
->ring
[i
].requests
[j
].jiffies
,
430 error
->ring
[i
].requests
[j
].tail
);
434 if ((obj
= error
->ring
[i
].ringbuffer
)) {
435 err_printf(m
, "%s --- ringbuffer = 0x%08x\n",
436 dev_priv
->ring
[i
].name
,
438 print_error_obj(m
, obj
);
441 if ((obj
= error
->ring
[i
].hws_page
)) {
442 err_printf(m
, "%s --- HW Status = 0x%08x\n",
443 dev_priv
->ring
[i
].name
,
446 for (elt
= 0; elt
< PAGE_SIZE
/16; elt
+= 4) {
447 err_printf(m
, "[%04x] %08x %08x %08x %08x\n",
450 obj
->pages
[0][elt
+1],
451 obj
->pages
[0][elt
+2],
452 obj
->pages
[0][elt
+3]);
457 if ((obj
= error
->ring
[i
].ctx
)) {
458 err_printf(m
, "%s --- HW Context = 0x%08x\n",
459 dev_priv
->ring
[i
].name
,
461 print_error_obj(m
, obj
);
465 if ((obj
= error
->semaphore_obj
)) {
466 err_printf(m
, "Semaphore page = 0x%08x\n", obj
->gtt_offset
);
467 for (elt
= 0; elt
< PAGE_SIZE
/16; elt
+= 4) {
468 err_printf(m
, "[%04x] %08x %08x %08x %08x\n",
471 obj
->pages
[0][elt
+1],
472 obj
->pages
[0][elt
+2],
473 obj
->pages
[0][elt
+3]);
478 intel_overlay_print_error_state(m
, error
->overlay
);
481 intel_display_print_error_state(m
, dev
, error
->display
);
484 if (m
->bytes
== 0 && m
->err
)
490 int i915_error_state_buf_init(struct drm_i915_error_state_buf
*ebuf
,
491 size_t count
, loff_t pos
)
493 memset(ebuf
, 0, sizeof(*ebuf
));
495 /* We need to have enough room to store any i915_error_state printf
496 * so that we can move it to start position.
498 ebuf
->size
= count
+ 1 > PAGE_SIZE
? count
+ 1 : PAGE_SIZE
;
499 ebuf
->buf
= kmalloc(ebuf
->size
,
500 GFP_TEMPORARY
| __GFP_NORETRY
| __GFP_NOWARN
);
502 if (ebuf
->buf
== NULL
) {
503 ebuf
->size
= PAGE_SIZE
;
504 ebuf
->buf
= kmalloc(ebuf
->size
, GFP_TEMPORARY
);
507 if (ebuf
->buf
== NULL
) {
509 ebuf
->buf
= kmalloc(ebuf
->size
, GFP_TEMPORARY
);
512 if (ebuf
->buf
== NULL
)
520 static void i915_error_object_free(struct drm_i915_error_object
*obj
)
527 for (page
= 0; page
< obj
->page_count
; page
++)
528 kfree(obj
->pages
[page
]);
533 static void i915_error_state_free(struct kref
*error_ref
)
535 struct drm_i915_error_state
*error
= container_of(error_ref
,
536 typeof(*error
), ref
);
539 for (i
= 0; i
< ARRAY_SIZE(error
->ring
); i
++) {
540 i915_error_object_free(error
->ring
[i
].batchbuffer
);
541 i915_error_object_free(error
->ring
[i
].ringbuffer
);
542 i915_error_object_free(error
->ring
[i
].hws_page
);
543 i915_error_object_free(error
->ring
[i
].ctx
);
544 kfree(error
->ring
[i
].requests
);
547 i915_error_object_free(error
->semaphore_obj
);
548 kfree(error
->active_bo
);
549 kfree(error
->overlay
);
550 kfree(error
->display
);
554 static struct drm_i915_error_object
*
555 i915_error_object_create_sized(struct drm_i915_private
*dev_priv
,
556 struct drm_i915_gem_object
*src
,
557 struct i915_address_space
*vm
,
560 struct drm_i915_error_object
*dst
;
564 if (src
== NULL
|| src
->pages
== NULL
)
567 dst
= kmalloc(sizeof(*dst
) + num_pages
* sizeof(u32
*), GFP_ATOMIC
);
571 reloc_offset
= dst
->gtt_offset
= i915_gem_obj_offset(src
, vm
);
572 for (i
= 0; i
< num_pages
; i
++) {
576 d
= kmalloc(PAGE_SIZE
, GFP_ATOMIC
);
580 local_irq_save(flags
);
581 if (src
->cache_level
== I915_CACHE_NONE
&&
582 reloc_offset
< dev_priv
->gtt
.mappable_end
&&
583 src
->has_global_gtt_mapping
&&
587 /* Simply ignore tiling or any overlapping fence.
588 * It's part of the error state, and this hopefully
589 * captures what the GPU read.
592 s
= io_mapping_map_atomic_wc(dev_priv
->gtt
.mappable
,
594 memcpy_fromio(d
, s
, PAGE_SIZE
);
595 io_mapping_unmap_atomic(s
);
596 } else if (src
->stolen
) {
597 unsigned long offset
;
599 offset
= dev_priv
->mm
.stolen_base
;
600 offset
+= src
->stolen
->start
;
601 offset
+= i
<< PAGE_SHIFT
;
603 memcpy_fromio(d
, (void __iomem
*) offset
, PAGE_SIZE
);
608 page
= i915_gem_object_get_page(src
, i
);
610 drm_clflush_pages(&page
, 1);
612 s
= kmap_atomic(page
);
613 memcpy(d
, s
, PAGE_SIZE
);
616 drm_clflush_pages(&page
, 1);
618 local_irq_restore(flags
);
622 reloc_offset
+= PAGE_SIZE
;
624 dst
->page_count
= num_pages
;
630 kfree(dst
->pages
[i
]);
634 #define i915_error_object_create(dev_priv, src, vm) \
635 i915_error_object_create_sized((dev_priv), (src), (vm), \
636 (src)->base.size>>PAGE_SHIFT)
638 #define i915_error_ggtt_object_create(dev_priv, src) \
639 i915_error_object_create_sized((dev_priv), (src), &(dev_priv)->gtt.base, \
640 (src)->base.size>>PAGE_SHIFT)
642 static void capture_bo(struct drm_i915_error_buffer
*err
,
643 struct drm_i915_gem_object
*obj
)
645 err
->size
= obj
->base
.size
;
646 err
->name
= obj
->base
.name
;
647 err
->rseqno
= obj
->last_read_seqno
;
648 err
->wseqno
= obj
->last_write_seqno
;
649 err
->gtt_offset
= i915_gem_obj_ggtt_offset(obj
);
650 err
->read_domains
= obj
->base
.read_domains
;
651 err
->write_domain
= obj
->base
.write_domain
;
652 err
->fence_reg
= obj
->fence_reg
;
654 if (i915_gem_obj_is_pinned(obj
))
656 if (obj
->user_pin_count
> 0)
658 err
->tiling
= obj
->tiling_mode
;
659 err
->dirty
= obj
->dirty
;
660 err
->purgeable
= obj
->madv
!= I915_MADV_WILLNEED
;
661 err
->userptr
= obj
->userptr
.mm
!= NULL
;
662 err
->ring
= obj
->ring
? obj
->ring
->id
: -1;
663 err
->cache_level
= obj
->cache_level
;
666 static u32
capture_active_bo(struct drm_i915_error_buffer
*err
,
667 int count
, struct list_head
*head
)
669 struct i915_vma
*vma
;
672 list_for_each_entry(vma
, head
, mm_list
) {
673 capture_bo(err
++, vma
->obj
);
681 static u32
capture_pinned_bo(struct drm_i915_error_buffer
*err
,
682 int count
, struct list_head
*head
)
684 struct drm_i915_gem_object
*obj
;
687 list_for_each_entry(obj
, head
, global_list
) {
688 if (!i915_gem_obj_is_pinned(obj
))
691 capture_bo(err
++, obj
);
699 /* Generate a semi-unique error code. The code is not meant to have meaning, The
700 * code's only purpose is to try to prevent false duplicated bug reports by
701 * grossly estimating a GPU error state.
703 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
704 * the hang if we could strip the GTT offset information from it.
706 * It's only a small step better than a random number in its current form.
708 static uint32_t i915_error_generate_code(struct drm_i915_private
*dev_priv
,
709 struct drm_i915_error_state
*error
,
712 uint32_t error_code
= 0;
715 /* IPEHR would be an ideal way to detect errors, as it's the gross
716 * measure of "the command that hung." However, has some very common
717 * synchronization commands which almost always appear in the case
718 * strictly a client bug. Use instdone to differentiate those some.
720 for (i
= 0; i
< I915_NUM_RINGS
; i
++) {
721 if (error
->ring
[i
].hangcheck_action
== HANGCHECK_HUNG
) {
725 return error
->ring
[i
].ipehr
^ error
->ring
[i
].instdone
;
732 static void i915_gem_record_fences(struct drm_device
*dev
,
733 struct drm_i915_error_state
*error
)
735 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
739 switch (INTEL_INFO(dev
)->gen
) {
743 for (i
= 0; i
< dev_priv
->num_fence_regs
; i
++)
744 error
->fence
[i
] = I915_READ64(FENCE_REG_SANDYBRIDGE_0
+ (i
* 8));
748 for (i
= 0; i
< 16; i
++)
749 error
->fence
[i
] = I915_READ64(FENCE_REG_965_0
+ (i
* 8));
752 if (IS_I945G(dev
) || IS_I945GM(dev
) || IS_G33(dev
))
753 for (i
= 0; i
< 8; i
++)
754 error
->fence
[i
+8] = I915_READ(FENCE_REG_945_8
+ (i
* 4));
756 for (i
= 0; i
< 8; i
++)
757 error
->fence
[i
] = I915_READ(FENCE_REG_830_0
+ (i
* 4));
766 static void gen8_record_semaphore_state(struct drm_i915_private
*dev_priv
,
767 struct drm_i915_error_state
*error
,
768 struct intel_engine_cs
*ring
,
769 struct drm_i915_error_ring
*ering
)
771 struct intel_engine_cs
*to
;
774 if (!i915_semaphore_is_enabled(dev_priv
->dev
))
777 if (!error
->semaphore_obj
)
778 error
->semaphore_obj
=
779 i915_error_object_create(dev_priv
,
780 dev_priv
->semaphore_obj
,
781 &dev_priv
->gtt
.base
);
783 for_each_ring(to
, dev_priv
, i
) {
791 signal_offset
= (GEN8_SIGNAL_OFFSET(ring
, i
) & (PAGE_SIZE
- 1))
793 tmp
= error
->semaphore_obj
->pages
[0];
794 idx
= intel_ring_sync_index(ring
, to
);
796 ering
->semaphore_mboxes
[idx
] = tmp
[signal_offset
];
797 ering
->semaphore_seqno
[idx
] = ring
->semaphore
.sync_seqno
[idx
];
801 static void gen6_record_semaphore_state(struct drm_i915_private
*dev_priv
,
802 struct intel_engine_cs
*ring
,
803 struct drm_i915_error_ring
*ering
)
805 ering
->semaphore_mboxes
[0] = I915_READ(RING_SYNC_0(ring
->mmio_base
));
806 ering
->semaphore_mboxes
[1] = I915_READ(RING_SYNC_1(ring
->mmio_base
));
807 ering
->semaphore_seqno
[0] = ring
->semaphore
.sync_seqno
[0];
808 ering
->semaphore_seqno
[1] = ring
->semaphore
.sync_seqno
[1];
810 if (HAS_VEBOX(dev_priv
->dev
)) {
811 ering
->semaphore_mboxes
[2] =
812 I915_READ(RING_SYNC_2(ring
->mmio_base
));
813 ering
->semaphore_seqno
[2] = ring
->semaphore
.sync_seqno
[2];
817 static void i915_record_ring_state(struct drm_device
*dev
,
818 struct drm_i915_error_state
*error
,
819 struct intel_engine_cs
*ring
,
820 struct drm_i915_error_ring
*ering
)
822 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
824 if (INTEL_INFO(dev
)->gen
>= 6) {
825 ering
->rc_psmi
= I915_READ(ring
->mmio_base
+ 0x50);
826 ering
->fault_reg
= I915_READ(RING_FAULT_REG(ring
));
827 if (INTEL_INFO(dev
)->gen
>= 8)
828 gen8_record_semaphore_state(dev_priv
, error
, ring
, ering
);
830 gen6_record_semaphore_state(dev_priv
, ring
, ering
);
833 if (INTEL_INFO(dev
)->gen
>= 4) {
834 ering
->faddr
= I915_READ(RING_DMA_FADD(ring
->mmio_base
));
835 ering
->ipeir
= I915_READ(RING_IPEIR(ring
->mmio_base
));
836 ering
->ipehr
= I915_READ(RING_IPEHR(ring
->mmio_base
));
837 ering
->instdone
= I915_READ(RING_INSTDONE(ring
->mmio_base
));
838 ering
->instps
= I915_READ(RING_INSTPS(ring
->mmio_base
));
839 ering
->bbaddr
= I915_READ(RING_BBADDR(ring
->mmio_base
));
840 if (INTEL_INFO(dev
)->gen
>= 8) {
841 ering
->faddr
|= (u64
) I915_READ(RING_DMA_FADD_UDW(ring
->mmio_base
)) << 32;
842 ering
->bbaddr
|= (u64
) I915_READ(RING_BBADDR_UDW(ring
->mmio_base
)) << 32;
844 ering
->bbstate
= I915_READ(RING_BBSTATE(ring
->mmio_base
));
846 ering
->faddr
= I915_READ(DMA_FADD_I8XX
);
847 ering
->ipeir
= I915_READ(IPEIR
);
848 ering
->ipehr
= I915_READ(IPEHR
);
849 ering
->instdone
= I915_READ(INSTDONE
);
852 ering
->waiting
= waitqueue_active(&ring
->irq_queue
);
853 ering
->instpm
= I915_READ(RING_INSTPM(ring
->mmio_base
));
854 ering
->seqno
= ring
->get_seqno(ring
, false);
855 ering
->acthd
= intel_ring_get_active_head(ring
);
856 ering
->head
= I915_READ_HEAD(ring
);
857 ering
->tail
= I915_READ_TAIL(ring
);
858 ering
->ctl
= I915_READ_CTL(ring
);
860 if (I915_NEED_GFX_HWS(dev
)) {
867 mmio
= RENDER_HWS_PGA_GEN7
;
870 mmio
= BLT_HWS_PGA_GEN7
;
873 mmio
= BSD_HWS_PGA_GEN7
;
876 mmio
= VEBOX_HWS_PGA_GEN7
;
879 } else if (IS_GEN6(ring
->dev
)) {
880 mmio
= RING_HWS_PGA_GEN6(ring
->mmio_base
);
882 /* XXX: gen8 returns to sanity */
883 mmio
= RING_HWS_PGA(ring
->mmio_base
);
886 ering
->hws
= I915_READ(mmio
);
889 ering
->cpu_ring_head
= ring
->buffer
->head
;
890 ering
->cpu_ring_tail
= ring
->buffer
->tail
;
892 ering
->hangcheck_score
= ring
->hangcheck
.score
;
893 ering
->hangcheck_action
= ring
->hangcheck
.action
;
895 if (USES_PPGTT(dev
)) {
898 ering
->vm_info
.gfx_mode
= I915_READ(RING_MODE_GEN7(ring
));
900 switch (INTEL_INFO(dev
)->gen
) {
902 for (i
= 0; i
< 4; i
++) {
903 ering
->vm_info
.pdp
[i
] =
904 I915_READ(GEN8_RING_PDP_UDW(ring
, i
));
905 ering
->vm_info
.pdp
[i
] <<= 32;
906 ering
->vm_info
.pdp
[i
] |=
907 I915_READ(GEN8_RING_PDP_LDW(ring
, i
));
911 ering
->vm_info
.pp_dir_base
=
912 I915_READ(RING_PP_DIR_BASE(ring
));
915 ering
->vm_info
.pp_dir_base
=
916 I915_READ(RING_PP_DIR_BASE_READ(ring
));
923 static void i915_gem_record_active_context(struct intel_engine_cs
*ring
,
924 struct drm_i915_error_state
*error
,
925 struct drm_i915_error_ring
*ering
)
927 struct drm_i915_private
*dev_priv
= ring
->dev
->dev_private
;
928 struct drm_i915_gem_object
*obj
;
930 /* Currently render ring is the only HW context user */
931 if (ring
->id
!= RCS
|| !error
->ccid
)
934 list_for_each_entry(obj
, &dev_priv
->mm
.bound_list
, global_list
) {
935 if (!i915_gem_obj_ggtt_bound(obj
))
938 if ((error
->ccid
& PAGE_MASK
) == i915_gem_obj_ggtt_offset(obj
)) {
939 ering
->ctx
= i915_error_ggtt_object_create(dev_priv
, obj
);
945 static void i915_gem_record_rings(struct drm_device
*dev
,
946 struct drm_i915_error_state
*error
)
948 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
949 struct drm_i915_gem_request
*request
;
952 for (i
= 0; i
< I915_NUM_RINGS
; i
++) {
953 struct intel_engine_cs
*ring
= &dev_priv
->ring
[i
];
955 error
->ring
[i
].pid
= -1;
957 if (ring
->dev
== NULL
)
960 error
->ring
[i
].valid
= true;
962 i915_record_ring_state(dev
, error
, ring
, &error
->ring
[i
]);
964 request
= i915_gem_find_active_request(ring
);
966 /* We need to copy these to an anonymous buffer
967 * as the simplest method to avoid being overwritten
970 error
->ring
[i
].batchbuffer
=
971 i915_error_object_create(dev_priv
,
975 &dev_priv
->gtt
.base
);
977 if (HAS_BROKEN_CS_TLB(dev_priv
->dev
) &&
979 error
->ring
[i
].wa_batchbuffer
=
980 i915_error_ggtt_object_create(dev_priv
,
983 if (request
->file_priv
) {
984 struct task_struct
*task
;
987 task
= pid_task(request
->file_priv
->file
->pid
,
990 strcpy(error
->ring
[i
].comm
, task
->comm
);
991 error
->ring
[i
].pid
= task
->pid
;
997 error
->ring
[i
].ringbuffer
=
998 i915_error_ggtt_object_create(dev_priv
, ring
->buffer
->obj
);
1000 if (ring
->status_page
.obj
)
1001 error
->ring
[i
].hws_page
=
1002 i915_error_ggtt_object_create(dev_priv
, ring
->status_page
.obj
);
1004 i915_gem_record_active_context(ring
, error
, &error
->ring
[i
]);
1007 list_for_each_entry(request
, &ring
->request_list
, list
)
1010 error
->ring
[i
].num_requests
= count
;
1011 error
->ring
[i
].requests
=
1012 kcalloc(count
, sizeof(*error
->ring
[i
].requests
),
1014 if (error
->ring
[i
].requests
== NULL
) {
1015 error
->ring
[i
].num_requests
= 0;
1020 list_for_each_entry(request
, &ring
->request_list
, list
) {
1021 struct drm_i915_error_request
*erq
;
1023 erq
= &error
->ring
[i
].requests
[count
++];
1024 erq
->seqno
= request
->seqno
;
1025 erq
->jiffies
= request
->emitted_jiffies
;
1026 erq
->tail
= request
->tail
;
1031 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1034 static void i915_gem_capture_vm(struct drm_i915_private
*dev_priv
,
1035 struct drm_i915_error_state
*error
,
1036 struct i915_address_space
*vm
,
1039 struct drm_i915_error_buffer
*active_bo
= NULL
, *pinned_bo
= NULL
;
1040 struct drm_i915_gem_object
*obj
;
1041 struct i915_vma
*vma
;
1045 list_for_each_entry(vma
, &vm
->active_list
, mm_list
)
1047 error
->active_bo_count
[ndx
] = i
;
1048 list_for_each_entry(obj
, &dev_priv
->mm
.bound_list
, global_list
)
1049 if (i915_gem_obj_is_pinned(obj
))
1051 error
->pinned_bo_count
[ndx
] = i
- error
->active_bo_count
[ndx
];
1054 active_bo
= kcalloc(i
, sizeof(*active_bo
), GFP_ATOMIC
);
1056 pinned_bo
= active_bo
+ error
->active_bo_count
[ndx
];
1060 error
->active_bo_count
[ndx
] =
1061 capture_active_bo(active_bo
,
1062 error
->active_bo_count
[ndx
],
1066 error
->pinned_bo_count
[ndx
] =
1067 capture_pinned_bo(pinned_bo
,
1068 error
->pinned_bo_count
[ndx
],
1069 &dev_priv
->mm
.bound_list
);
1070 error
->active_bo
[ndx
] = active_bo
;
1071 error
->pinned_bo
[ndx
] = pinned_bo
;
1074 static void i915_gem_capture_buffers(struct drm_i915_private
*dev_priv
,
1075 struct drm_i915_error_state
*error
)
1077 struct i915_address_space
*vm
;
1080 list_for_each_entry(vm
, &dev_priv
->vm_list
, global_link
)
1083 error
->active_bo
= kcalloc(cnt
, sizeof(*error
->active_bo
), GFP_ATOMIC
);
1084 error
->pinned_bo
= kcalloc(cnt
, sizeof(*error
->pinned_bo
), GFP_ATOMIC
);
1085 error
->active_bo_count
= kcalloc(cnt
, sizeof(*error
->active_bo_count
),
1087 error
->pinned_bo_count
= kcalloc(cnt
, sizeof(*error
->pinned_bo_count
),
1090 list_for_each_entry(vm
, &dev_priv
->vm_list
, global_link
)
1091 i915_gem_capture_vm(dev_priv
, error
, vm
, i
++);
1094 /* Capture all registers which don't fit into another category. */
1095 static void i915_capture_reg_state(struct drm_i915_private
*dev_priv
,
1096 struct drm_i915_error_state
*error
)
1098 struct drm_device
*dev
= dev_priv
->dev
;
1100 /* General organization
1101 * 1. Registers specific to a single generation
1102 * 2. Registers which belong to multiple generations
1103 * 3. Feature specific registers.
1104 * 4. Everything else
1105 * Please try to follow the order.
1108 /* 1: Registers specific to a single generation */
1109 if (IS_VALLEYVIEW(dev
)) {
1110 error
->gtier
= I915_READ(GTIER
);
1111 error
->ier
= I915_READ(VLV_IER
);
1112 error
->forcewake
= I915_READ(FORCEWAKE_VLV
);
1116 error
->err_int
= I915_READ(GEN7_ERR_INT
);
1119 error
->forcewake
= I915_READ(FORCEWAKE
);
1120 error
->gab_ctl
= I915_READ(GAB_CTL
);
1121 error
->gfx_mode
= I915_READ(GFX_MODE
);
1124 /* 2: Registers which belong to multiple generations */
1125 if (INTEL_INFO(dev
)->gen
>= 7)
1126 error
->forcewake
= I915_READ(FORCEWAKE_MT
);
1128 if (INTEL_INFO(dev
)->gen
>= 6) {
1129 error
->derrmr
= I915_READ(DERRMR
);
1130 error
->error
= I915_READ(ERROR_GEN6
);
1131 error
->done_reg
= I915_READ(DONE_REG
);
1134 /* 3: Feature specific registers */
1135 if (IS_GEN6(dev
) || IS_GEN7(dev
)) {
1136 error
->gam_ecochk
= I915_READ(GAM_ECOCHK
);
1137 error
->gac_eco
= I915_READ(GAC_ECO_BITS
);
1140 /* 4: Everything else */
1141 if (HAS_HW_CONTEXTS(dev
))
1142 error
->ccid
= I915_READ(CCID
);
1144 if (HAS_PCH_SPLIT(dev
)) {
1145 error
->ier
= I915_READ(DEIER
);
1146 error
->gtier
= I915_READ(GTIER
);
1147 } else if (IS_GEN2(dev
)) {
1148 error
->ier
= I915_READ16(IER
);
1149 } else if (!IS_VALLEYVIEW(dev
)) {
1150 error
->ier
= I915_READ(IER
);
1152 error
->eir
= I915_READ(EIR
);
1153 error
->pgtbl_er
= I915_READ(PGTBL_ER
);
1155 i915_get_extra_instdone(dev
, error
->extra_instdone
);
1158 static void i915_error_capture_msg(struct drm_device
*dev
,
1159 struct drm_i915_error_state
*error
,
1161 const char *error_msg
)
1163 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1165 int ring_id
= -1, len
;
1167 ecode
= i915_error_generate_code(dev_priv
, error
, &ring_id
);
1169 len
= scnprintf(error
->error_msg
, sizeof(error
->error_msg
),
1170 "GPU HANG: ecode %d:0x%08x", ring_id
, ecode
);
1172 if (ring_id
!= -1 && error
->ring
[ring_id
].pid
!= -1)
1173 len
+= scnprintf(error
->error_msg
+ len
,
1174 sizeof(error
->error_msg
) - len
,
1176 error
->ring
[ring_id
].comm
,
1177 error
->ring
[ring_id
].pid
);
1179 scnprintf(error
->error_msg
+ len
, sizeof(error
->error_msg
) - len
,
1180 ", reason: %s, action: %s",
1182 wedged
? "reset" : "continue");
1185 static void i915_capture_gen_state(struct drm_i915_private
*dev_priv
,
1186 struct drm_i915_error_state
*error
)
1188 error
->reset_count
= i915_reset_count(&dev_priv
->gpu_error
);
1189 error
->suspend_count
= dev_priv
->suspend_count
;
1193 * i915_capture_error_state - capture an error record for later analysis
1196 * Should be called when an error is detected (either a hang or an error
1197 * interrupt) to capture error state from the time of the error. Fills
1198 * out a structure which becomes available in debugfs for user level tools
1201 void i915_capture_error_state(struct drm_device
*dev
, bool wedged
,
1202 const char *error_msg
)
1205 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1206 struct drm_i915_error_state
*error
;
1207 unsigned long flags
;
1209 /* Account for pipe specific data like PIPE*STAT */
1210 error
= kzalloc(sizeof(*error
), GFP_ATOMIC
);
1212 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1216 kref_init(&error
->ref
);
1218 i915_capture_gen_state(dev_priv
, error
);
1219 i915_capture_reg_state(dev_priv
, error
);
1220 i915_gem_capture_buffers(dev_priv
, error
);
1221 i915_gem_record_fences(dev
, error
);
1222 i915_gem_record_rings(dev
, error
);
1224 do_gettimeofday(&error
->time
);
1226 error
->overlay
= intel_overlay_capture_error_state(dev
);
1227 error
->display
= intel_display_capture_error_state(dev
);
1229 i915_error_capture_msg(dev
, error
, wedged
, error_msg
);
1230 DRM_INFO("%s\n", error
->error_msg
);
1232 spin_lock_irqsave(&dev_priv
->gpu_error
.lock
, flags
);
1233 if (dev_priv
->gpu_error
.first_error
== NULL
) {
1234 dev_priv
->gpu_error
.first_error
= error
;
1237 spin_unlock_irqrestore(&dev_priv
->gpu_error
.lock
, flags
);
1240 i915_error_state_free(&error
->ref
);
1245 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1246 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1247 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1248 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1249 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev
->primary
->index
);
1254 void i915_error_state_get(struct drm_device
*dev
,
1255 struct i915_error_state_file_priv
*error_priv
)
1257 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1258 unsigned long flags
;
1260 spin_lock_irqsave(&dev_priv
->gpu_error
.lock
, flags
);
1261 error_priv
->error
= dev_priv
->gpu_error
.first_error
;
1262 if (error_priv
->error
)
1263 kref_get(&error_priv
->error
->ref
);
1264 spin_unlock_irqrestore(&dev_priv
->gpu_error
.lock
, flags
);
1268 void i915_error_state_put(struct i915_error_state_file_priv
*error_priv
)
1270 if (error_priv
->error
)
1271 kref_put(&error_priv
->error
->ref
, i915_error_state_free
);
1274 void i915_destroy_error_state(struct drm_device
*dev
)
1276 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1277 struct drm_i915_error_state
*error
;
1278 unsigned long flags
;
1280 spin_lock_irqsave(&dev_priv
->gpu_error
.lock
, flags
);
1281 error
= dev_priv
->gpu_error
.first_error
;
1282 dev_priv
->gpu_error
.first_error
= NULL
;
1283 spin_unlock_irqrestore(&dev_priv
->gpu_error
.lock
, flags
);
1286 kref_put(&error
->ref
, i915_error_state_free
);
1289 const char *i915_cache_level_str(int type
)
1292 case I915_CACHE_NONE
: return " uncached";
1293 case I915_CACHE_LLC
: return " snooped or LLC";
1294 case I915_CACHE_L3_LLC
: return " L3+LLC";
1295 case I915_CACHE_WT
: return " WT";
1300 /* NB: please notice the memset */
1301 void i915_get_extra_instdone(struct drm_device
*dev
, uint32_t *instdone
)
1303 struct drm_i915_private
*dev_priv
= dev
->dev_private
;
1304 memset(instdone
, 0, sizeof(*instdone
) * I915_NUM_INSTDONE_REG
);
1306 switch (INTEL_INFO(dev
)->gen
) {
1309 instdone
[0] = I915_READ(INSTDONE
);
1314 instdone
[0] = I915_READ(INSTDONE_I965
);
1315 instdone
[1] = I915_READ(INSTDONE1
);
1318 WARN_ONCE(1, "Unsupported platform\n");
1321 instdone
[0] = I915_READ(GEN7_INSTDONE_1
);
1322 instdone
[1] = I915_READ(GEN7_SC_INSTDONE
);
1323 instdone
[2] = I915_READ(GEN7_SAMPLER_INSTDONE
);
1324 instdone
[3] = I915_READ(GEN7_ROW_INSTDONE
);