Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
[cascardo/linux.git] / drivers / gpu / drm / i915 / i915_gpu_error.c
1 /*
2  * Copyright (c) 2008 Intel Corporation
3  *
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:
10  *
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
13  * Software.
14  *
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
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Eric Anholt <eric@anholt.net>
25  *    Keith Packard <keithp@keithp.com>
26  *    Mika Kuoppala <mika.kuoppala@intel.com>
27  *
28  */
29
30 #include <generated/utsrelease.h>
31 #include "i915_drv.h"
32
33 static const char *yesno(int v)
34 {
35         return v ? "yes" : "no";
36 }
37
38 static const char *ring_str(int ring)
39 {
40         switch (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";
46         default: return "";
47         }
48 }
49
50 static const char *pin_flag(int pinned)
51 {
52         if (pinned > 0)
53                 return " P";
54         else if (pinned < 0)
55                 return " p";
56         else
57                 return "";
58 }
59
60 static const char *tiling_flag(int tiling)
61 {
62         switch (tiling) {
63         default:
64         case I915_TILING_NONE: return "";
65         case I915_TILING_X: return " X";
66         case I915_TILING_Y: return " Y";
67         }
68 }
69
70 static const char *dirty_flag(int dirty)
71 {
72         return dirty ? " dirty" : "";
73 }
74
75 static const char *purgeable_flag(int purgeable)
76 {
77         return purgeable ? " purgeable" : "";
78 }
79
80 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
81 {
82
83         if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
84                 e->err = -ENOSPC;
85                 return false;
86         }
87
88         if (e->bytes == e->size - 1 || e->err)
89                 return false;
90
91         return true;
92 }
93
94 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
95                               unsigned len)
96 {
97         if (e->pos + len <= e->start) {
98                 e->pos += len;
99                 return false;
100         }
101
102         /* First vsnprintf needs to fit in its entirety for memmove */
103         if (len >= e->size) {
104                 e->err = -EIO;
105                 return false;
106         }
107
108         return true;
109 }
110
111 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
112                                  unsigned len)
113 {
114         /* If this is first printf in this window, adjust it so that
115          * start position matches start of the buffer
116          */
117
118         if (e->pos < e->start) {
119                 const size_t off = e->start - e->pos;
120
121                 /* Should not happen but be paranoid */
122                 if (off > len || e->bytes) {
123                         e->err = -EIO;
124                         return;
125                 }
126
127                 memmove(e->buf, e->buf + off, len - off);
128                 e->bytes = len - off;
129                 e->pos = e->start;
130                 return;
131         }
132
133         e->bytes += len;
134         e->pos += len;
135 }
136
137 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
138                                const char *f, va_list args)
139 {
140         unsigned len;
141
142         if (!__i915_error_ok(e))
143                 return;
144
145         /* Seek the first printf which is hits start position */
146         if (e->pos < e->start) {
147                 va_list tmp;
148
149                 va_copy(tmp, args);
150                 len = vsnprintf(NULL, 0, f, tmp);
151                 va_end(tmp);
152
153                 if (!__i915_error_seek(e, len))
154                         return;
155         }
156
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;
160
161         __i915_error_advance(e, len);
162 }
163
164 static void i915_error_puts(struct drm_i915_error_state_buf *e,
165                             const char *str)
166 {
167         unsigned len;
168
169         if (!__i915_error_ok(e))
170                 return;
171
172         len = strlen(str);
173
174         /* Seek the first printf which is hits start position */
175         if (e->pos < e->start) {
176                 if (!__i915_error_seek(e, len))
177                         return;
178         }
179
180         if (len >= e->size - e->bytes)
181                 len = e->size - e->bytes - 1;
182         memcpy(e->buf + e->bytes, str, len);
183
184         __i915_error_advance(e, len);
185 }
186
187 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
188 #define err_puts(e, s) i915_error_puts(e, s)
189
190 static void print_error_buffers(struct drm_i915_error_state_buf *m,
191                                 const char *name,
192                                 struct drm_i915_error_buffer *err,
193                                 int count)
194 {
195         err_printf(m, "  %s [%d]:\n", name, count);
196
197         while (count--) {
198                 err_printf(m, "    %08x %8u %02x %02x %x %x",
199                            err->gtt_offset,
200                            err->size,
201                            err->read_domains,
202                            err->write_domain,
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(m->i915, err->cache_level));
212
213                 if (err->name)
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);
217
218                 err_puts(m, "\n");
219                 err++;
220         }
221 }
222
223 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
224 {
225         switch (a) {
226         case HANGCHECK_IDLE:
227                 return "idle";
228         case HANGCHECK_WAIT:
229                 return "wait";
230         case HANGCHECK_ACTIVE:
231                 return "active";
232         case HANGCHECK_ACTIVE_LOOP:
233                 return "active (loop)";
234         case HANGCHECK_KICK:
235                 return "kick";
236         case HANGCHECK_HUNG:
237                 return "hung";
238         }
239
240         return "unknown";
241 }
242
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)
246 {
247         if (!ring->valid)
248                 return;
249
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);
262         }
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]);
279                 }
280         }
281         if (USES_PPGTT(dev)) {
282                 err_printf(m, "  GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
283
284                 if (INTEL_INFO(dev)->gen >= 8) {
285                         int i;
286                         for (i = 0; i < 4; i++)
287                                 err_printf(m, "  PDP%d: 0x%016llx\n",
288                                            i, ring->vm_info.pdp[i]);
289                 } else {
290                         err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
291                                    ring->vm_info.pp_dir_base);
292                 }
293         }
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);
301 }
302
303 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
304 {
305         va_list args;
306
307         va_start(args, f);
308         i915_error_vprintf(e, f, args);
309         va_end(args);
310 }
311
312 static void print_error_obj(struct drm_i915_error_state_buf *m,
313                             struct drm_i915_error_object *obj)
314 {
315         int page, offset, elt;
316
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]);
321                         offset += 4;
322                 }
323         }
324 }
325
326 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
327                             const struct i915_error_state_file_priv *error_priv)
328 {
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;
335
336         if (!error) {
337                 err_printf(m, "no error state collected\n");
338                 goto out;
339         }
340
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;
349         }
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",
354                                    ring_str(i),
355                                    error->ring[i].comm,
356                                    error->ring[i].pid);
357                 }
358         }
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,
367                                    error->gtier[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);
375
376         for (i = 0; i < dev_priv->num_fence_regs; i++)
377                 err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
378
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]);
382
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);
386         }
387
388         if (INTEL_INFO(dev)->gen == 7)
389                 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
390
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]);
394         }
395
396         for (i = 0; i < error->vm_count; i++) {
397                 err_printf(m, "vm[%d]\n", i);
398
399                 print_error_buffers(m, "Active",
400                                     error->active_bo[i],
401                                     error->active_bo_count[i]);
402
403                 print_error_buffers(m, "Pinned",
404                                     error->pinned_bo[i],
405                                     error->pinned_bo_count[i]);
406         }
407
408         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
409                 obj = error->ring[i].batchbuffer;
410                 if (obj) {
411                         err_puts(m, dev_priv->ring[i].name);
412                         if (error->ring[i].pid != -1)
413                                 err_printf(m, " (submitted by %s [%d])",
414                                            error->ring[i].comm,
415                                            error->ring[i].pid);
416                         err_printf(m, " --- gtt_offset = 0x%08x\n",
417                                    obj->gtt_offset);
418                         print_error_obj(m, obj);
419                 }
420
421                 obj = error->ring[i].wa_batchbuffer;
422                 if (obj) {
423                         err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
424                                    dev_priv->ring[i].name, obj->gtt_offset);
425                         print_error_obj(m, obj);
426                 }
427
428                 if (error->ring[i].num_requests) {
429                         err_printf(m, "%s --- %d requests\n",
430                                    dev_priv->ring[i].name,
431                                    error->ring[i].num_requests);
432                         for (j = 0; j < error->ring[i].num_requests; j++) {
433                                 err_printf(m, "  seqno 0x%08x, emitted %ld, tail 0x%08x\n",
434                                            error->ring[i].requests[j].seqno,
435                                            error->ring[i].requests[j].jiffies,
436                                            error->ring[i].requests[j].tail);
437                         }
438                 }
439
440                 if ((obj = error->ring[i].ringbuffer)) {
441                         err_printf(m, "%s --- ringbuffer = 0x%08x\n",
442                                    dev_priv->ring[i].name,
443                                    obj->gtt_offset);
444                         print_error_obj(m, obj);
445                 }
446
447                 if ((obj = error->ring[i].hws_page)) {
448                         err_printf(m, "%s --- HW Status = 0x%08x\n",
449                                    dev_priv->ring[i].name,
450                                    obj->gtt_offset);
451                         offset = 0;
452                         for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
453                                 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
454                                            offset,
455                                            obj->pages[0][elt],
456                                            obj->pages[0][elt+1],
457                                            obj->pages[0][elt+2],
458                                            obj->pages[0][elt+3]);
459                                         offset += 16;
460                         }
461                 }
462
463                 if ((obj = error->ring[i].ctx)) {
464                         err_printf(m, "%s --- HW Context = 0x%08x\n",
465                                    dev_priv->ring[i].name,
466                                    obj->gtt_offset);
467                         print_error_obj(m, obj);
468                 }
469         }
470
471         if ((obj = error->semaphore_obj)) {
472                 err_printf(m, "Semaphore page = 0x%08x\n", obj->gtt_offset);
473                 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
474                         err_printf(m, "[%04x] %08x %08x %08x %08x\n",
475                                    elt * 4,
476                                    obj->pages[0][elt],
477                                    obj->pages[0][elt+1],
478                                    obj->pages[0][elt+2],
479                                    obj->pages[0][elt+3]);
480                 }
481         }
482
483         if (error->overlay)
484                 intel_overlay_print_error_state(m, error->overlay);
485
486         if (error->display)
487                 intel_display_print_error_state(m, dev, error->display);
488
489 out:
490         if (m->bytes == 0 && m->err)
491                 return m->err;
492
493         return 0;
494 }
495
496 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
497                               struct drm_i915_private *i915,
498                               size_t count, loff_t pos)
499 {
500         memset(ebuf, 0, sizeof(*ebuf));
501         ebuf->i915 = i915;
502
503         /* We need to have enough room to store any i915_error_state printf
504          * so that we can move it to start position.
505          */
506         ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
507         ebuf->buf = kmalloc(ebuf->size,
508                                 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
509
510         if (ebuf->buf == NULL) {
511                 ebuf->size = PAGE_SIZE;
512                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
513         }
514
515         if (ebuf->buf == NULL) {
516                 ebuf->size = 128;
517                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
518         }
519
520         if (ebuf->buf == NULL)
521                 return -ENOMEM;
522
523         ebuf->start = pos;
524
525         return 0;
526 }
527
528 static void i915_error_object_free(struct drm_i915_error_object *obj)
529 {
530         int page;
531
532         if (obj == NULL)
533                 return;
534
535         for (page = 0; page < obj->page_count; page++)
536                 kfree(obj->pages[page]);
537
538         kfree(obj);
539 }
540
541 static void i915_error_state_free(struct kref *error_ref)
542 {
543         struct drm_i915_error_state *error = container_of(error_ref,
544                                                           typeof(*error), ref);
545         int i;
546
547         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
548                 i915_error_object_free(error->ring[i].batchbuffer);
549                 i915_error_object_free(error->ring[i].ringbuffer);
550                 i915_error_object_free(error->ring[i].hws_page);
551                 i915_error_object_free(error->ring[i].ctx);
552                 kfree(error->ring[i].requests);
553         }
554
555         i915_error_object_free(error->semaphore_obj);
556         kfree(error->active_bo);
557         kfree(error->overlay);
558         kfree(error->display);
559         kfree(error);
560 }
561
562 static struct drm_i915_error_object *
563 i915_error_object_create(struct drm_i915_private *dev_priv,
564                          struct drm_i915_gem_object *src,
565                          struct i915_address_space *vm)
566 {
567         struct drm_i915_error_object *dst;
568         int num_pages;
569         bool use_ggtt;
570         int i = 0;
571         u32 reloc_offset;
572
573         if (src == NULL || src->pages == NULL)
574                 return NULL;
575
576         num_pages = src->base.size >> PAGE_SHIFT;
577
578         dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
579         if (dst == NULL)
580                 return NULL;
581
582         if (i915_gem_obj_bound(src, vm))
583                 dst->gtt_offset = i915_gem_obj_offset(src, vm);
584         else
585                 dst->gtt_offset = -1;
586
587         reloc_offset = dst->gtt_offset;
588         use_ggtt = (src->cache_level == I915_CACHE_NONE &&
589                     i915_is_ggtt(vm) &&
590                     src->has_global_gtt_mapping &&
591                     reloc_offset + num_pages * PAGE_SIZE <= dev_priv->gtt.mappable_end);
592
593         /* Cannot access stolen address directly, try to use the aperture */
594         if (src->stolen) {
595                 use_ggtt = true;
596
597                 if (!src->has_global_gtt_mapping)
598                         goto unwind;
599
600                 reloc_offset = i915_gem_obj_ggtt_offset(src);
601                 if (reloc_offset + num_pages * PAGE_SIZE > dev_priv->gtt.mappable_end)
602                         goto unwind;
603         }
604
605         /* Cannot access snooped pages through the aperture */
606         if (use_ggtt && src->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv->dev))
607                 goto unwind;
608
609         dst->page_count = num_pages;
610         while (num_pages--) {
611                 unsigned long flags;
612                 void *d;
613
614                 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
615                 if (d == NULL)
616                         goto unwind;
617
618                 local_irq_save(flags);
619                 if (use_ggtt) {
620                         void __iomem *s;
621
622                         /* Simply ignore tiling or any overlapping fence.
623                          * It's part of the error state, and this hopefully
624                          * captures what the GPU read.
625                          */
626
627                         s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
628                                                      reloc_offset);
629                         memcpy_fromio(d, s, PAGE_SIZE);
630                         io_mapping_unmap_atomic(s);
631                 } else {
632                         struct page *page;
633                         void *s;
634
635                         page = i915_gem_object_get_page(src, i);
636
637                         drm_clflush_pages(&page, 1);
638
639                         s = kmap_atomic(page);
640                         memcpy(d, s, PAGE_SIZE);
641                         kunmap_atomic(s);
642
643                         drm_clflush_pages(&page, 1);
644                 }
645                 local_irq_restore(flags);
646
647                 dst->pages[i++] = d;
648                 reloc_offset += PAGE_SIZE;
649         }
650
651         return dst;
652
653 unwind:
654         while (i--)
655                 kfree(dst->pages[i]);
656         kfree(dst);
657         return NULL;
658 }
659 #define i915_error_ggtt_object_create(dev_priv, src) \
660         i915_error_object_create((dev_priv), (src), &(dev_priv)->gtt.base)
661
662 static void capture_bo(struct drm_i915_error_buffer *err,
663                        struct i915_vma *vma)
664 {
665         struct drm_i915_gem_object *obj = vma->obj;
666
667         err->size = obj->base.size;
668         err->name = obj->base.name;
669         err->rseqno = obj->last_read_seqno;
670         err->wseqno = obj->last_write_seqno;
671         err->gtt_offset = vma->node.start;
672         err->read_domains = obj->base.read_domains;
673         err->write_domain = obj->base.write_domain;
674         err->fence_reg = obj->fence_reg;
675         err->pinned = 0;
676         if (i915_gem_obj_is_pinned(obj))
677                 err->pinned = 1;
678         if (obj->user_pin_count > 0)
679                 err->pinned = -1;
680         err->tiling = obj->tiling_mode;
681         err->dirty = obj->dirty;
682         err->purgeable = obj->madv != I915_MADV_WILLNEED;
683         err->userptr = obj->userptr.mm != NULL;
684         err->ring = obj->ring ? obj->ring->id : -1;
685         err->cache_level = obj->cache_level;
686 }
687
688 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
689                              int count, struct list_head *head)
690 {
691         struct i915_vma *vma;
692         int i = 0;
693
694         list_for_each_entry(vma, head, mm_list) {
695                 capture_bo(err++, vma);
696                 if (++i == count)
697                         break;
698         }
699
700         return i;
701 }
702
703 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
704                              int count, struct list_head *head,
705                              struct i915_address_space *vm)
706 {
707         struct drm_i915_gem_object *obj;
708         struct drm_i915_error_buffer * const first = err;
709         struct drm_i915_error_buffer * const last = err + count;
710
711         list_for_each_entry(obj, head, global_list) {
712                 struct i915_vma *vma;
713
714                 if (err == last)
715                         break;
716
717                 list_for_each_entry(vma, &obj->vma_list, vma_link)
718                         if (vma->vm == vm && vma->pin_count > 0) {
719                                 capture_bo(err++, vma);
720                                 break;
721                         }
722         }
723
724         return err - first;
725 }
726
727 /* Generate a semi-unique error code. The code is not meant to have meaning, The
728  * code's only purpose is to try to prevent false duplicated bug reports by
729  * grossly estimating a GPU error state.
730  *
731  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
732  * the hang if we could strip the GTT offset information from it.
733  *
734  * It's only a small step better than a random number in its current form.
735  */
736 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
737                                          struct drm_i915_error_state *error,
738                                          int *ring_id)
739 {
740         uint32_t error_code = 0;
741         int i;
742
743         /* IPEHR would be an ideal way to detect errors, as it's the gross
744          * measure of "the command that hung." However, has some very common
745          * synchronization commands which almost always appear in the case
746          * strictly a client bug. Use instdone to differentiate those some.
747          */
748         for (i = 0; i < I915_NUM_RINGS; i++) {
749                 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
750                         if (ring_id)
751                                 *ring_id = i;
752
753                         return error->ring[i].ipehr ^ error->ring[i].instdone;
754                 }
755         }
756
757         return error_code;
758 }
759
760 static void i915_gem_record_fences(struct drm_device *dev,
761                                    struct drm_i915_error_state *error)
762 {
763         struct drm_i915_private *dev_priv = dev->dev_private;
764         int i;
765
766         /* Fences */
767         switch (INTEL_INFO(dev)->gen) {
768         case 8:
769         case 7:
770         case 6:
771                 for (i = 0; i < dev_priv->num_fence_regs; i++)
772                         error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
773                 break;
774         case 5:
775         case 4:
776                 for (i = 0; i < 16; i++)
777                         error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
778                 break;
779         case 3:
780                 if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
781                         for (i = 0; i < 8; i++)
782                                 error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
783         case 2:
784                 for (i = 0; i < 8; i++)
785                         error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
786                 break;
787
788         default:
789                 BUG();
790         }
791 }
792
793
794 static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
795                                         struct drm_i915_error_state *error,
796                                         struct intel_engine_cs *ring,
797                                         struct drm_i915_error_ring *ering)
798 {
799         struct intel_engine_cs *to;
800         int i;
801
802         if (!i915_semaphore_is_enabled(dev_priv->dev))
803                 return;
804
805         if (!error->semaphore_obj)
806                 error->semaphore_obj =
807                         i915_error_object_create(dev_priv,
808                                                  dev_priv->semaphore_obj,
809                                                  &dev_priv->gtt.base);
810
811         for_each_ring(to, dev_priv, i) {
812                 int idx;
813                 u16 signal_offset;
814                 u32 *tmp;
815
816                 if (ring == to)
817                         continue;
818
819                 signal_offset = (GEN8_SIGNAL_OFFSET(ring, i) & (PAGE_SIZE - 1))
820                                 / 4;
821                 tmp = error->semaphore_obj->pages[0];
822                 idx = intel_ring_sync_index(ring, to);
823
824                 ering->semaphore_mboxes[idx] = tmp[signal_offset];
825                 ering->semaphore_seqno[idx] = ring->semaphore.sync_seqno[idx];
826         }
827 }
828
829 static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
830                                         struct intel_engine_cs *ring,
831                                         struct drm_i915_error_ring *ering)
832 {
833         ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(ring->mmio_base));
834         ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(ring->mmio_base));
835         ering->semaphore_seqno[0] = ring->semaphore.sync_seqno[0];
836         ering->semaphore_seqno[1] = ring->semaphore.sync_seqno[1];
837
838         if (HAS_VEBOX(dev_priv->dev)) {
839                 ering->semaphore_mboxes[2] =
840                         I915_READ(RING_SYNC_2(ring->mmio_base));
841                 ering->semaphore_seqno[2] = ring->semaphore.sync_seqno[2];
842         }
843 }
844
845 static void i915_record_ring_state(struct drm_device *dev,
846                                    struct drm_i915_error_state *error,
847                                    struct intel_engine_cs *ring,
848                                    struct drm_i915_error_ring *ering)
849 {
850         struct drm_i915_private *dev_priv = dev->dev_private;
851
852         if (INTEL_INFO(dev)->gen >= 6) {
853                 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
854                 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
855                 if (INTEL_INFO(dev)->gen >= 8)
856                         gen8_record_semaphore_state(dev_priv, error, ring, ering);
857                 else
858                         gen6_record_semaphore_state(dev_priv, ring, ering);
859         }
860
861         if (INTEL_INFO(dev)->gen >= 4) {
862                 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
863                 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
864                 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
865                 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
866                 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
867                 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
868                 if (INTEL_INFO(dev)->gen >= 8) {
869                         ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(ring->mmio_base)) << 32;
870                         ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
871                 }
872                 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
873         } else {
874                 ering->faddr = I915_READ(DMA_FADD_I8XX);
875                 ering->ipeir = I915_READ(IPEIR);
876                 ering->ipehr = I915_READ(IPEHR);
877                 ering->instdone = I915_READ(INSTDONE);
878         }
879
880         ering->waiting = waitqueue_active(&ring->irq_queue);
881         ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
882         ering->seqno = ring->get_seqno(ring, false);
883         ering->acthd = intel_ring_get_active_head(ring);
884         ering->head = I915_READ_HEAD(ring);
885         ering->tail = I915_READ_TAIL(ring);
886         ering->ctl = I915_READ_CTL(ring);
887
888         if (I915_NEED_GFX_HWS(dev)) {
889                 int mmio;
890
891                 if (IS_GEN7(dev)) {
892                         switch (ring->id) {
893                         default:
894                         case RCS:
895                                 mmio = RENDER_HWS_PGA_GEN7;
896                                 break;
897                         case BCS:
898                                 mmio = BLT_HWS_PGA_GEN7;
899                                 break;
900                         case VCS:
901                                 mmio = BSD_HWS_PGA_GEN7;
902                                 break;
903                         case VECS:
904                                 mmio = VEBOX_HWS_PGA_GEN7;
905                                 break;
906                         }
907                 } else if (IS_GEN6(ring->dev)) {
908                         mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
909                 } else {
910                         /* XXX: gen8 returns to sanity */
911                         mmio = RING_HWS_PGA(ring->mmio_base);
912                 }
913
914                 ering->hws = I915_READ(mmio);
915         }
916
917         ering->hangcheck_score = ring->hangcheck.score;
918         ering->hangcheck_action = ring->hangcheck.action;
919
920         if (USES_PPGTT(dev)) {
921                 int i;
922
923                 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
924
925                 switch (INTEL_INFO(dev)->gen) {
926                 case 8:
927                         for (i = 0; i < 4; i++) {
928                                 ering->vm_info.pdp[i] =
929                                         I915_READ(GEN8_RING_PDP_UDW(ring, i));
930                                 ering->vm_info.pdp[i] <<= 32;
931                                 ering->vm_info.pdp[i] |=
932                                         I915_READ(GEN8_RING_PDP_LDW(ring, i));
933                         }
934                         break;
935                 case 7:
936                         ering->vm_info.pp_dir_base =
937                                 I915_READ(RING_PP_DIR_BASE(ring));
938                         break;
939                 case 6:
940                         ering->vm_info.pp_dir_base =
941                                 I915_READ(RING_PP_DIR_BASE_READ(ring));
942                         break;
943                 }
944         }
945 }
946
947
948 static void i915_gem_record_active_context(struct intel_engine_cs *ring,
949                                            struct drm_i915_error_state *error,
950                                            struct drm_i915_error_ring *ering)
951 {
952         struct drm_i915_private *dev_priv = ring->dev->dev_private;
953         struct drm_i915_gem_object *obj;
954
955         /* Currently render ring is the only HW context user */
956         if (ring->id != RCS || !error->ccid)
957                 return;
958
959         list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
960                 if (!i915_gem_obj_ggtt_bound(obj))
961                         continue;
962
963                 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
964                         ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
965                         break;
966                 }
967         }
968 }
969
970 static void i915_gem_record_rings(struct drm_device *dev,
971                                   struct drm_i915_error_state *error)
972 {
973         struct drm_i915_private *dev_priv = dev->dev_private;
974         struct drm_i915_gem_request *request;
975         int i, count;
976
977         for (i = 0; i < I915_NUM_RINGS; i++) {
978                 struct intel_engine_cs *ring = &dev_priv->ring[i];
979                 struct intel_ringbuffer *rbuf;
980
981                 error->ring[i].pid = -1;
982
983                 if (ring->dev == NULL)
984                         continue;
985
986                 error->ring[i].valid = true;
987
988                 i915_record_ring_state(dev, error, ring, &error->ring[i]);
989
990                 request = i915_gem_find_active_request(ring);
991                 if (request) {
992                         struct i915_address_space *vm;
993
994                         vm = request->ctx && request->ctx->ppgtt ?
995                                 &request->ctx->ppgtt->base :
996                                 &dev_priv->gtt.base;
997
998                         /* We need to copy these to an anonymous buffer
999                          * as the simplest method to avoid being overwritten
1000                          * by userspace.
1001                          */
1002                         error->ring[i].batchbuffer =
1003                                 i915_error_object_create(dev_priv,
1004                                                          request->batch_obj,
1005                                                          vm);
1006
1007                         if (HAS_BROKEN_CS_TLB(dev_priv->dev))
1008                                 error->ring[i].wa_batchbuffer =
1009                                         i915_error_ggtt_object_create(dev_priv,
1010                                                              ring->scratch.obj);
1011
1012                         if (request->file_priv) {
1013                                 struct task_struct *task;
1014
1015                                 rcu_read_lock();
1016                                 task = pid_task(request->file_priv->file->pid,
1017                                                 PIDTYPE_PID);
1018                                 if (task) {
1019                                         strcpy(error->ring[i].comm, task->comm);
1020                                         error->ring[i].pid = task->pid;
1021                                 }
1022                                 rcu_read_unlock();
1023                         }
1024                 }
1025
1026                 if (i915.enable_execlists) {
1027                         /* TODO: This is only a small fix to keep basic error
1028                          * capture working, but we need to add more information
1029                          * for it to be useful (e.g. dump the context being
1030                          * executed).
1031                          */
1032                         if (request)
1033                                 rbuf = request->ctx->engine[ring->id].ringbuf;
1034                         else
1035                                 rbuf = ring->default_context->engine[ring->id].ringbuf;
1036                 } else
1037                         rbuf = ring->buffer;
1038
1039                 error->ring[i].cpu_ring_head = rbuf->head;
1040                 error->ring[i].cpu_ring_tail = rbuf->tail;
1041
1042                 error->ring[i].ringbuffer =
1043                         i915_error_ggtt_object_create(dev_priv, rbuf->obj);
1044
1045                 error->ring[i].hws_page =
1046                         i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
1047
1048                 i915_gem_record_active_context(ring, error, &error->ring[i]);
1049
1050                 count = 0;
1051                 list_for_each_entry(request, &ring->request_list, list)
1052                         count++;
1053
1054                 error->ring[i].num_requests = count;
1055                 error->ring[i].requests =
1056                         kcalloc(count, sizeof(*error->ring[i].requests),
1057                                 GFP_ATOMIC);
1058                 if (error->ring[i].requests == NULL) {
1059                         error->ring[i].num_requests = 0;
1060                         continue;
1061                 }
1062
1063                 count = 0;
1064                 list_for_each_entry(request, &ring->request_list, list) {
1065                         struct drm_i915_error_request *erq;
1066
1067                         erq = &error->ring[i].requests[count++];
1068                         erq->seqno = request->seqno;
1069                         erq->jiffies = request->emitted_jiffies;
1070                         erq->tail = request->tail;
1071                 }
1072         }
1073 }
1074
1075 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1076  * VM.
1077  */
1078 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1079                                 struct drm_i915_error_state *error,
1080                                 struct i915_address_space *vm,
1081                                 const int ndx)
1082 {
1083         struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1084         struct drm_i915_gem_object *obj;
1085         struct i915_vma *vma;
1086         int i;
1087
1088         i = 0;
1089         list_for_each_entry(vma, &vm->active_list, mm_list)
1090                 i++;
1091         error->active_bo_count[ndx] = i;
1092
1093         list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1094                 list_for_each_entry(vma, &obj->vma_list, vma_link)
1095                         if (vma->vm == vm && vma->pin_count > 0) {
1096                                 i++;
1097                                 break;
1098                         }
1099         }
1100         error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1101
1102         if (i) {
1103                 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1104                 if (active_bo)
1105                         pinned_bo = active_bo + error->active_bo_count[ndx];
1106         }
1107
1108         if (active_bo)
1109                 error->active_bo_count[ndx] =
1110                         capture_active_bo(active_bo,
1111                                           error->active_bo_count[ndx],
1112                                           &vm->active_list);
1113
1114         if (pinned_bo)
1115                 error->pinned_bo_count[ndx] =
1116                         capture_pinned_bo(pinned_bo,
1117                                           error->pinned_bo_count[ndx],
1118                                           &dev_priv->mm.bound_list, vm);
1119         error->active_bo[ndx] = active_bo;
1120         error->pinned_bo[ndx] = pinned_bo;
1121 }
1122
1123 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1124                                      struct drm_i915_error_state *error)
1125 {
1126         struct i915_address_space *vm;
1127         int cnt = 0, i = 0;
1128
1129         list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1130                 cnt++;
1131
1132         error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1133         error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1134         error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1135                                          GFP_ATOMIC);
1136         error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1137                                          GFP_ATOMIC);
1138
1139         if (error->active_bo == NULL ||
1140             error->pinned_bo == NULL ||
1141             error->active_bo_count == NULL ||
1142             error->pinned_bo_count == NULL) {
1143                 kfree(error->active_bo);
1144                 kfree(error->active_bo_count);
1145                 kfree(error->pinned_bo);
1146                 kfree(error->pinned_bo_count);
1147
1148                 error->active_bo = NULL;
1149                 error->active_bo_count = NULL;
1150                 error->pinned_bo = NULL;
1151                 error->pinned_bo_count = NULL;
1152         } else {
1153                 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1154                         i915_gem_capture_vm(dev_priv, error, vm, i++);
1155
1156                 error->vm_count = cnt;
1157         }
1158 }
1159
1160 /* Capture all registers which don't fit into another category. */
1161 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1162                                    struct drm_i915_error_state *error)
1163 {
1164         struct drm_device *dev = dev_priv->dev;
1165         int i;
1166
1167         /* General organization
1168          * 1. Registers specific to a single generation
1169          * 2. Registers which belong to multiple generations
1170          * 3. Feature specific registers.
1171          * 4. Everything else
1172          * Please try to follow the order.
1173          */
1174
1175         /* 1: Registers specific to a single generation */
1176         if (IS_VALLEYVIEW(dev)) {
1177                 error->gtier[0] = I915_READ(GTIER);
1178                 error->ier = I915_READ(VLV_IER);
1179                 error->forcewake = I915_READ(FORCEWAKE_VLV);
1180         }
1181
1182         if (IS_GEN7(dev))
1183                 error->err_int = I915_READ(GEN7_ERR_INT);
1184
1185         if (IS_GEN6(dev)) {
1186                 error->forcewake = I915_READ(FORCEWAKE);
1187                 error->gab_ctl = I915_READ(GAB_CTL);
1188                 error->gfx_mode = I915_READ(GFX_MODE);
1189         }
1190
1191         /* 2: Registers which belong to multiple generations */
1192         if (INTEL_INFO(dev)->gen >= 7)
1193                 error->forcewake = I915_READ(FORCEWAKE_MT);
1194
1195         if (INTEL_INFO(dev)->gen >= 6) {
1196                 error->derrmr = I915_READ(DERRMR);
1197                 error->error = I915_READ(ERROR_GEN6);
1198                 error->done_reg = I915_READ(DONE_REG);
1199         }
1200
1201         /* 3: Feature specific registers */
1202         if (IS_GEN6(dev) || IS_GEN7(dev)) {
1203                 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1204                 error->gac_eco = I915_READ(GAC_ECO_BITS);
1205         }
1206
1207         /* 4: Everything else */
1208         if (HAS_HW_CONTEXTS(dev))
1209                 error->ccid = I915_READ(CCID);
1210
1211         if (INTEL_INFO(dev)->gen >= 8) {
1212                 error->ier = I915_READ(GEN8_DE_MISC_IER);
1213                 for (i = 0; i < 4; i++)
1214                         error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1215         } else if (HAS_PCH_SPLIT(dev)) {
1216                 error->ier = I915_READ(DEIER);
1217                 error->gtier[0] = I915_READ(GTIER);
1218         } else if (IS_GEN2(dev)) {
1219                 error->ier = I915_READ16(IER);
1220         } else if (!IS_VALLEYVIEW(dev)) {
1221                 error->ier = I915_READ(IER);
1222         }
1223         error->eir = I915_READ(EIR);
1224         error->pgtbl_er = I915_READ(PGTBL_ER);
1225
1226         i915_get_extra_instdone(dev, error->extra_instdone);
1227 }
1228
1229 static void i915_error_capture_msg(struct drm_device *dev,
1230                                    struct drm_i915_error_state *error,
1231                                    bool wedged,
1232                                    const char *error_msg)
1233 {
1234         struct drm_i915_private *dev_priv = dev->dev_private;
1235         u32 ecode;
1236         int ring_id = -1, len;
1237
1238         ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1239
1240         len = scnprintf(error->error_msg, sizeof(error->error_msg),
1241                         "GPU HANG: ecode %d:0x%08x", ring_id, ecode);
1242
1243         if (ring_id != -1 && error->ring[ring_id].pid != -1)
1244                 len += scnprintf(error->error_msg + len,
1245                                  sizeof(error->error_msg) - len,
1246                                  ", in %s [%d]",
1247                                  error->ring[ring_id].comm,
1248                                  error->ring[ring_id].pid);
1249
1250         scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1251                   ", reason: %s, action: %s",
1252                   error_msg,
1253                   wedged ? "reset" : "continue");
1254 }
1255
1256 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1257                                    struct drm_i915_error_state *error)
1258 {
1259         error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1260         error->suspend_count = dev_priv->suspend_count;
1261 }
1262
1263 /**
1264  * i915_capture_error_state - capture an error record for later analysis
1265  * @dev: drm device
1266  *
1267  * Should be called when an error is detected (either a hang or an error
1268  * interrupt) to capture error state from the time of the error.  Fills
1269  * out a structure which becomes available in debugfs for user level tools
1270  * to pick up.
1271  */
1272 void i915_capture_error_state(struct drm_device *dev, bool wedged,
1273                               const char *error_msg)
1274 {
1275         static bool warned;
1276         struct drm_i915_private *dev_priv = dev->dev_private;
1277         struct drm_i915_error_state *error;
1278         unsigned long flags;
1279
1280         /* Account for pipe specific data like PIPE*STAT */
1281         error = kzalloc(sizeof(*error), GFP_ATOMIC);
1282         if (!error) {
1283                 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1284                 return;
1285         }
1286
1287         kref_init(&error->ref);
1288
1289         i915_capture_gen_state(dev_priv, error);
1290         i915_capture_reg_state(dev_priv, error);
1291         i915_gem_capture_buffers(dev_priv, error);
1292         i915_gem_record_fences(dev, error);
1293         i915_gem_record_rings(dev, error);
1294
1295         do_gettimeofday(&error->time);
1296
1297         error->overlay = intel_overlay_capture_error_state(dev);
1298         error->display = intel_display_capture_error_state(dev);
1299
1300         i915_error_capture_msg(dev, error, wedged, error_msg);
1301         DRM_INFO("%s\n", error->error_msg);
1302
1303         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1304         if (dev_priv->gpu_error.first_error == NULL) {
1305                 dev_priv->gpu_error.first_error = error;
1306                 error = NULL;
1307         }
1308         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1309
1310         if (error) {
1311                 i915_error_state_free(&error->ref);
1312                 return;
1313         }
1314
1315         if (!warned) {
1316                 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1317                 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1318                 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1319                 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1320                 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1321                 warned = true;
1322         }
1323 }
1324
1325 void i915_error_state_get(struct drm_device *dev,
1326                           struct i915_error_state_file_priv *error_priv)
1327 {
1328         struct drm_i915_private *dev_priv = dev->dev_private;
1329         unsigned long flags;
1330
1331         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1332         error_priv->error = dev_priv->gpu_error.first_error;
1333         if (error_priv->error)
1334                 kref_get(&error_priv->error->ref);
1335         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1336
1337 }
1338
1339 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1340 {
1341         if (error_priv->error)
1342                 kref_put(&error_priv->error->ref, i915_error_state_free);
1343 }
1344
1345 void i915_destroy_error_state(struct drm_device *dev)
1346 {
1347         struct drm_i915_private *dev_priv = dev->dev_private;
1348         struct drm_i915_error_state *error;
1349         unsigned long flags;
1350
1351         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1352         error = dev_priv->gpu_error.first_error;
1353         dev_priv->gpu_error.first_error = NULL;
1354         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1355
1356         if (error)
1357                 kref_put(&error->ref, i915_error_state_free);
1358 }
1359
1360 const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
1361 {
1362         switch (type) {
1363         case I915_CACHE_NONE: return " uncached";
1364         case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
1365         case I915_CACHE_L3_LLC: return " L3+LLC";
1366         case I915_CACHE_WT: return " WT";
1367         default: return "";
1368         }
1369 }
1370
1371 /* NB: please notice the memset */
1372 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1373 {
1374         struct drm_i915_private *dev_priv = dev->dev_private;
1375         memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1376
1377         switch (INTEL_INFO(dev)->gen) {
1378         case 2:
1379         case 3:
1380                 instdone[0] = I915_READ(INSTDONE);
1381                 break;
1382         case 4:
1383         case 5:
1384         case 6:
1385                 instdone[0] = I915_READ(INSTDONE_I965);
1386                 instdone[1] = I915_READ(INSTDONE1);
1387                 break;
1388         default:
1389                 WARN_ONCE(1, "Unsupported platform\n");
1390         case 7:
1391         case 8:
1392                 instdone[0] = I915_READ(GEN7_INSTDONE_1);
1393                 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1394                 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1395                 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
1396                 break;
1397         }
1398 }