Merge tag 'drm-intel-next-2014-02-07' of ssh://git.freedesktop.org/git/drm-intel...
[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         default: return "";
46         }
47 }
48
49 static const char *pin_flag(int pinned)
50 {
51         if (pinned > 0)
52                 return " P";
53         else if (pinned < 0)
54                 return " p";
55         else
56                 return "";
57 }
58
59 static const char *tiling_flag(int tiling)
60 {
61         switch (tiling) {
62         default:
63         case I915_TILING_NONE: return "";
64         case I915_TILING_X: return " X";
65         case I915_TILING_Y: return " Y";
66         }
67 }
68
69 static const char *dirty_flag(int dirty)
70 {
71         return dirty ? " dirty" : "";
72 }
73
74 static const char *purgeable_flag(int purgeable)
75 {
76         return purgeable ? " purgeable" : "";
77 }
78
79 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
80 {
81
82         if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
83                 e->err = -ENOSPC;
84                 return false;
85         }
86
87         if (e->bytes == e->size - 1 || e->err)
88                 return false;
89
90         return true;
91 }
92
93 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
94                               unsigned len)
95 {
96         if (e->pos + len <= e->start) {
97                 e->pos += len;
98                 return false;
99         }
100
101         /* First vsnprintf needs to fit in its entirety for memmove */
102         if (len >= e->size) {
103                 e->err = -EIO;
104                 return false;
105         }
106
107         return true;
108 }
109
110 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
111                                  unsigned len)
112 {
113         /* If this is first printf in this window, adjust it so that
114          * start position matches start of the buffer
115          */
116
117         if (e->pos < e->start) {
118                 const size_t off = e->start - e->pos;
119
120                 /* Should not happen but be paranoid */
121                 if (off > len || e->bytes) {
122                         e->err = -EIO;
123                         return;
124                 }
125
126                 memmove(e->buf, e->buf + off, len - off);
127                 e->bytes = len - off;
128                 e->pos = e->start;
129                 return;
130         }
131
132         e->bytes += len;
133         e->pos += len;
134 }
135
136 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
137                                const char *f, va_list args)
138 {
139         unsigned len;
140
141         if (!__i915_error_ok(e))
142                 return;
143
144         /* Seek the first printf which is hits start position */
145         if (e->pos < e->start) {
146                 va_list tmp;
147
148                 va_copy(tmp, args);
149                 len = vsnprintf(NULL, 0, f, tmp);
150                 va_end(tmp);
151
152                 if (!__i915_error_seek(e, len))
153                         return;
154         }
155
156         len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
157         if (len >= e->size - e->bytes)
158                 len = e->size - e->bytes - 1;
159
160         __i915_error_advance(e, len);
161 }
162
163 static void i915_error_puts(struct drm_i915_error_state_buf *e,
164                             const char *str)
165 {
166         unsigned len;
167
168         if (!__i915_error_ok(e))
169                 return;
170
171         len = strlen(str);
172
173         /* Seek the first printf which is hits start position */
174         if (e->pos < e->start) {
175                 if (!__i915_error_seek(e, len))
176                         return;
177         }
178
179         if (len >= e->size - e->bytes)
180                 len = e->size - e->bytes - 1;
181         memcpy(e->buf + e->bytes, str, len);
182
183         __i915_error_advance(e, len);
184 }
185
186 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
187 #define err_puts(e, s) i915_error_puts(e, s)
188
189 static void print_error_buffers(struct drm_i915_error_state_buf *m,
190                                 const char *name,
191                                 struct drm_i915_error_buffer *err,
192                                 int count)
193 {
194         err_printf(m, "%s [%d]:\n", name, count);
195
196         while (count--) {
197                 err_printf(m, "  %08x %8u %02x %02x %x %x",
198                            err->gtt_offset,
199                            err->size,
200                            err->read_domains,
201                            err->write_domain,
202                            err->rseqno, err->wseqno);
203                 err_puts(m, pin_flag(err->pinned));
204                 err_puts(m, tiling_flag(err->tiling));
205                 err_puts(m, dirty_flag(err->dirty));
206                 err_puts(m, purgeable_flag(err->purgeable));
207                 err_puts(m, err->ring != -1 ? " " : "");
208                 err_puts(m, ring_str(err->ring));
209                 err_puts(m, i915_cache_level_str(err->cache_level));
210
211                 if (err->name)
212                         err_printf(m, " (name: %d)", err->name);
213                 if (err->fence_reg != I915_FENCE_REG_NONE)
214                         err_printf(m, " (fence: %d)", err->fence_reg);
215
216                 err_puts(m, "\n");
217                 err++;
218         }
219 }
220
221 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
222 {
223         switch (a) {
224         case HANGCHECK_IDLE:
225                 return "idle";
226         case HANGCHECK_WAIT:
227                 return "wait";
228         case HANGCHECK_ACTIVE:
229                 return "active";
230         case HANGCHECK_KICK:
231                 return "kick";
232         case HANGCHECK_HUNG:
233                 return "hung";
234         }
235
236         return "unknown";
237 }
238
239 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
240                                   struct drm_device *dev,
241                                   struct drm_i915_error_ring *ring)
242 {
243         if (!ring->valid)
244                 return;
245
246         err_printf(m, "  HEAD: 0x%08x\n", ring->head);
247         err_printf(m, "  TAIL: 0x%08x\n", ring->tail);
248         err_printf(m, "  CTL: 0x%08x\n", ring->ctl);
249         err_printf(m, "  HWS: 0x%08x\n", ring->hws);
250         err_printf(m, "  ACTHD: 0x%08x\n", ring->acthd);
251         err_printf(m, "  IPEIR: 0x%08x\n", ring->ipeir);
252         err_printf(m, "  IPEHR: 0x%08x\n", ring->ipehr);
253         err_printf(m, "  INSTDONE: 0x%08x\n", ring->instdone);
254         if (INTEL_INFO(dev)->gen >= 4) {
255                 err_printf(m, "  BBADDR: 0x%08llx\n", ring->bbaddr);
256                 err_printf(m, "  BB_STATE: 0x%08x\n", ring->bbstate);
257                 err_printf(m, "  INSTPS: 0x%08x\n", ring->instps);
258         }
259         err_printf(m, "  INSTPM: 0x%08x\n", ring->instpm);
260         err_printf(m, "  FADDR: 0x%08x\n", ring->faddr);
261         if (INTEL_INFO(dev)->gen >= 6) {
262                 err_printf(m, "  RC PSMI: 0x%08x\n", ring->rc_psmi);
263                 err_printf(m, "  FAULT_REG: 0x%08x\n", ring->fault_reg);
264                 err_printf(m, "  SYNC_0: 0x%08x [last synced 0x%08x]\n",
265                            ring->semaphore_mboxes[0],
266                            ring->semaphore_seqno[0]);
267                 err_printf(m, "  SYNC_1: 0x%08x [last synced 0x%08x]\n",
268                            ring->semaphore_mboxes[1],
269                            ring->semaphore_seqno[1]);
270                 if (HAS_VEBOX(dev)) {
271                         err_printf(m, "  SYNC_2: 0x%08x [last synced 0x%08x]\n",
272                                    ring->semaphore_mboxes[2],
273                                    ring->semaphore_seqno[2]);
274                 }
275         }
276         if (USES_PPGTT(dev)) {
277                 err_printf(m, "  GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
278
279                 if (INTEL_INFO(dev)->gen >= 8) {
280                         int i;
281                         for (i = 0; i < 4; i++)
282                                 err_printf(m, "  PDP%d: 0x%016llx\n",
283                                            i, ring->vm_info.pdp[i]);
284                 } else {
285                         err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
286                                    ring->vm_info.pp_dir_base);
287                 }
288         }
289         err_printf(m, "  seqno: 0x%08x\n", ring->seqno);
290         err_printf(m, "  waiting: %s\n", yesno(ring->waiting));
291         err_printf(m, "  ring->head: 0x%08x\n", ring->cpu_ring_head);
292         err_printf(m, "  ring->tail: 0x%08x\n", ring->cpu_ring_tail);
293         err_printf(m, "  hangcheck: %s [%d]\n",
294                    hangcheck_action_to_str(ring->hangcheck_action),
295                    ring->hangcheck_score);
296 }
297
298 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
299 {
300         va_list args;
301
302         va_start(args, f);
303         i915_error_vprintf(e, f, args);
304         va_end(args);
305 }
306
307 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
308                             const struct i915_error_state_file_priv *error_priv)
309 {
310         struct drm_device *dev = error_priv->dev;
311         drm_i915_private_t *dev_priv = dev->dev_private;
312         struct drm_i915_error_state *error = error_priv->error;
313         int i, j, page, offset, elt;
314
315         if (!error) {
316                 err_printf(m, "no error state collected\n");
317                 goto out;
318         }
319
320         err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
321                    error->time.tv_usec);
322         err_printf(m, "Kernel: " UTS_RELEASE "\n");
323         err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
324         err_printf(m, "EIR: 0x%08x\n", error->eir);
325         err_printf(m, "IER: 0x%08x\n", error->ier);
326         err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
327         err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
328         err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
329         err_printf(m, "CCID: 0x%08x\n", error->ccid);
330         err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
331
332         for (i = 0; i < dev_priv->num_fence_regs; i++)
333                 err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
334
335         for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
336                 err_printf(m, "  INSTDONE_%d: 0x%08x\n", i,
337                            error->extra_instdone[i]);
338
339         if (INTEL_INFO(dev)->gen >= 6) {
340                 err_printf(m, "ERROR: 0x%08x\n", error->error);
341                 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
342         }
343
344         if (INTEL_INFO(dev)->gen == 7)
345                 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
346
347         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
348                 err_printf(m, "%s command stream:\n", ring_str(i));
349                 i915_ring_error_state(m, dev, &error->ring[i]);
350         }
351
352         if (error->active_bo)
353                 print_error_buffers(m, "Active",
354                                     error->active_bo[0],
355                                     error->active_bo_count[0]);
356
357         if (error->pinned_bo)
358                 print_error_buffers(m, "Pinned",
359                                     error->pinned_bo[0],
360                                     error->pinned_bo_count[0]);
361
362         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
363                 struct drm_i915_error_object *obj;
364
365                 if ((obj = error->ring[i].batchbuffer)) {
366                         err_printf(m, "%s --- gtt_offset = 0x%08x\n",
367                                    dev_priv->ring[i].name,
368                                    obj->gtt_offset);
369                         offset = 0;
370                         for (page = 0; page < obj->page_count; page++) {
371                                 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
372                                         err_printf(m, "%08x :  %08x\n", offset,
373                                                    obj->pages[page][elt]);
374                                         offset += 4;
375                                 }
376                         }
377                 }
378
379                 if (error->ring[i].num_requests) {
380                         err_printf(m, "%s --- %d requests\n",
381                                    dev_priv->ring[i].name,
382                                    error->ring[i].num_requests);
383                         for (j = 0; j < error->ring[i].num_requests; j++) {
384                                 err_printf(m, "  seqno 0x%08x, emitted %ld, tail 0x%08x\n",
385                                            error->ring[i].requests[j].seqno,
386                                            error->ring[i].requests[j].jiffies,
387                                            error->ring[i].requests[j].tail);
388                         }
389                 }
390
391                 if ((obj = error->ring[i].ringbuffer)) {
392                         err_printf(m, "%s --- ringbuffer = 0x%08x\n",
393                                    dev_priv->ring[i].name,
394                                    obj->gtt_offset);
395                         offset = 0;
396                         for (page = 0; page < obj->page_count; page++) {
397                                 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
398                                         err_printf(m, "%08x :  %08x\n",
399                                                    offset,
400                                                    obj->pages[page][elt]);
401                                         offset += 4;
402                                 }
403                         }
404                 }
405
406                 if ((obj = error->ring[i].hws_page)) {
407                         err_printf(m, "%s --- HW Status = 0x%08x\n",
408                                    dev_priv->ring[i].name,
409                                    obj->gtt_offset);
410                         offset = 0;
411                         for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
412                                 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
413                                            offset,
414                                            obj->pages[0][elt],
415                                            obj->pages[0][elt+1],
416                                            obj->pages[0][elt+2],
417                                            obj->pages[0][elt+3]);
418                                         offset += 16;
419                         }
420                 }
421
422                 if ((obj = error->ring[i].ctx)) {
423                         err_printf(m, "%s --- HW Context = 0x%08x\n",
424                                    dev_priv->ring[i].name,
425                                    obj->gtt_offset);
426                         offset = 0;
427                         for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
428                                 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
429                                            offset,
430                                            obj->pages[0][elt],
431                                            obj->pages[0][elt+1],
432                                            obj->pages[0][elt+2],
433                                            obj->pages[0][elt+3]);
434                                         offset += 16;
435                         }
436                 }
437         }
438
439         if (error->overlay)
440                 intel_overlay_print_error_state(m, error->overlay);
441
442         if (error->display)
443                 intel_display_print_error_state(m, dev, error->display);
444
445 out:
446         if (m->bytes == 0 && m->err)
447                 return m->err;
448
449         return 0;
450 }
451
452 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
453                               size_t count, loff_t pos)
454 {
455         memset(ebuf, 0, sizeof(*ebuf));
456
457         /* We need to have enough room to store any i915_error_state printf
458          * so that we can move it to start position.
459          */
460         ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
461         ebuf->buf = kmalloc(ebuf->size,
462                                 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
463
464         if (ebuf->buf == NULL) {
465                 ebuf->size = PAGE_SIZE;
466                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
467         }
468
469         if (ebuf->buf == NULL) {
470                 ebuf->size = 128;
471                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
472         }
473
474         if (ebuf->buf == NULL)
475                 return -ENOMEM;
476
477         ebuf->start = pos;
478
479         return 0;
480 }
481
482 static void i915_error_object_free(struct drm_i915_error_object *obj)
483 {
484         int page;
485
486         if (obj == NULL)
487                 return;
488
489         for (page = 0; page < obj->page_count; page++)
490                 kfree(obj->pages[page]);
491
492         kfree(obj);
493 }
494
495 static void i915_error_state_free(struct kref *error_ref)
496 {
497         struct drm_i915_error_state *error = container_of(error_ref,
498                                                           typeof(*error), ref);
499         int i;
500
501         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
502                 i915_error_object_free(error->ring[i].batchbuffer);
503                 i915_error_object_free(error->ring[i].ringbuffer);
504                 i915_error_object_free(error->ring[i].hws_page);
505                 i915_error_object_free(error->ring[i].ctx);
506                 kfree(error->ring[i].requests);
507         }
508
509         kfree(error->active_bo);
510         kfree(error->overlay);
511         kfree(error->display);
512         kfree(error);
513 }
514
515 static struct drm_i915_error_object *
516 i915_error_object_create_sized(struct drm_i915_private *dev_priv,
517                                struct drm_i915_gem_object *src,
518                                struct i915_address_space *vm,
519                                const int num_pages)
520 {
521         struct drm_i915_error_object *dst;
522         int i;
523         u32 reloc_offset;
524
525         if (src == NULL || src->pages == NULL)
526                 return NULL;
527
528         dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
529         if (dst == NULL)
530                 return NULL;
531
532         reloc_offset = dst->gtt_offset = i915_gem_obj_offset(src, vm);
533         for (i = 0; i < num_pages; i++) {
534                 unsigned long flags;
535                 void *d;
536
537                 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
538                 if (d == NULL)
539                         goto unwind;
540
541                 local_irq_save(flags);
542                 if (src->cache_level == I915_CACHE_NONE &&
543                     reloc_offset < dev_priv->gtt.mappable_end &&
544                     src->has_global_gtt_mapping &&
545                     i915_is_ggtt(vm)) {
546                         void __iomem *s;
547
548                         /* Simply ignore tiling or any overlapping fence.
549                          * It's part of the error state, and this hopefully
550                          * captures what the GPU read.
551                          */
552
553                         s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
554                                                      reloc_offset);
555                         memcpy_fromio(d, s, PAGE_SIZE);
556                         io_mapping_unmap_atomic(s);
557                 } else if (src->stolen) {
558                         unsigned long offset;
559
560                         offset = dev_priv->mm.stolen_base;
561                         offset += src->stolen->start;
562                         offset += i << PAGE_SHIFT;
563
564                         memcpy_fromio(d, (void __iomem *) offset, PAGE_SIZE);
565                 } else {
566                         struct page *page;
567                         void *s;
568
569                         page = i915_gem_object_get_page(src, i);
570
571                         drm_clflush_pages(&page, 1);
572
573                         s = kmap_atomic(page);
574                         memcpy(d, s, PAGE_SIZE);
575                         kunmap_atomic(s);
576
577                         drm_clflush_pages(&page, 1);
578                 }
579                 local_irq_restore(flags);
580
581                 dst->pages[i] = d;
582
583                 reloc_offset += PAGE_SIZE;
584         }
585         dst->page_count = num_pages;
586
587         return dst;
588
589 unwind:
590         while (i--)
591                 kfree(dst->pages[i]);
592         kfree(dst);
593         return NULL;
594 }
595 #define i915_error_object_create(dev_priv, src, vm) \
596         i915_error_object_create_sized((dev_priv), (src), (vm), \
597                                        (src)->base.size>>PAGE_SHIFT)
598
599 #define i915_error_ggtt_object_create(dev_priv, src) \
600         i915_error_object_create_sized((dev_priv), (src), &(dev_priv)->gtt.base, \
601                                        (src)->base.size>>PAGE_SHIFT)
602
603 static void capture_bo(struct drm_i915_error_buffer *err,
604                        struct drm_i915_gem_object *obj)
605 {
606         err->size = obj->base.size;
607         err->name = obj->base.name;
608         err->rseqno = obj->last_read_seqno;
609         err->wseqno = obj->last_write_seqno;
610         err->gtt_offset = i915_gem_obj_ggtt_offset(obj);
611         err->read_domains = obj->base.read_domains;
612         err->write_domain = obj->base.write_domain;
613         err->fence_reg = obj->fence_reg;
614         err->pinned = 0;
615         if (i915_gem_obj_is_pinned(obj))
616                 err->pinned = 1;
617         if (obj->user_pin_count > 0)
618                 err->pinned = -1;
619         err->tiling = obj->tiling_mode;
620         err->dirty = obj->dirty;
621         err->purgeable = obj->madv != I915_MADV_WILLNEED;
622         err->ring = obj->ring ? obj->ring->id : -1;
623         err->cache_level = obj->cache_level;
624 }
625
626 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
627                              int count, struct list_head *head)
628 {
629         struct i915_vma *vma;
630         int i = 0;
631
632         list_for_each_entry(vma, head, mm_list) {
633                 capture_bo(err++, vma->obj);
634                 if (++i == count)
635                         break;
636         }
637
638         return i;
639 }
640
641 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
642                              int count, struct list_head *head)
643 {
644         struct drm_i915_gem_object *obj;
645         int i = 0;
646
647         list_for_each_entry(obj, head, global_list) {
648                 if (!i915_gem_obj_is_pinned(obj))
649                         continue;
650
651                 capture_bo(err++, obj);
652                 if (++i == count)
653                         break;
654         }
655
656         return i;
657 }
658
659 /* Generate a semi-unique error code. The code is not meant to have meaning, The
660  * code's only purpose is to try to prevent false duplicated bug reports by
661  * grossly estimating a GPU error state.
662  *
663  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
664  * the hang if we could strip the GTT offset information from it.
665  *
666  * It's only a small step better than a random number in its current form.
667  */
668 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
669                                          struct drm_i915_error_state *error)
670 {
671         uint32_t error_code = 0;
672         int i;
673
674         /* IPEHR would be an ideal way to detect errors, as it's the gross
675          * measure of "the command that hung." However, has some very common
676          * synchronization commands which almost always appear in the case
677          * strictly a client bug. Use instdone to differentiate those some.
678          */
679         for (i = 0; i < I915_NUM_RINGS; i++)
680                 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG)
681                         return error->ring[i].ipehr ^ error->ring[i].instdone;
682
683         return error_code;
684 }
685
686 static void i915_gem_record_fences(struct drm_device *dev,
687                                    struct drm_i915_error_state *error)
688 {
689         struct drm_i915_private *dev_priv = dev->dev_private;
690         int i;
691
692         /* Fences */
693         switch (INTEL_INFO(dev)->gen) {
694         case 8:
695         case 7:
696         case 6:
697                 for (i = 0; i < dev_priv->num_fence_regs; i++)
698                         error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
699                 break;
700         case 5:
701         case 4:
702                 for (i = 0; i < 16; i++)
703                         error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
704                 break;
705         case 3:
706                 if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
707                         for (i = 0; i < 8; i++)
708                                 error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
709         case 2:
710                 for (i = 0; i < 8; i++)
711                         error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
712                 break;
713
714         default:
715                 BUG();
716         }
717 }
718
719 /* This assumes all batchbuffers are executed from the PPGTT. It might have to
720  * change in the future. */
721 static bool is_active_vm(struct i915_address_space *vm,
722                          struct intel_ring_buffer *ring)
723 {
724         struct drm_device *dev = vm->dev;
725         struct drm_i915_private *dev_priv = dev->dev_private;
726         struct i915_hw_ppgtt *ppgtt;
727
728         if (INTEL_INFO(dev)->gen < 7)
729                 return i915_is_ggtt(vm);
730
731         /* FIXME: This ignores that the global gtt vm is also on this list. */
732         ppgtt = container_of(vm, struct i915_hw_ppgtt, base);
733
734         if (INTEL_INFO(dev)->gen >= 8) {
735                 u64 pdp0 = (u64)I915_READ(GEN8_RING_PDP_UDW(ring, 0)) << 32;
736                 pdp0 |=  I915_READ(GEN8_RING_PDP_LDW(ring, 0));
737                 return pdp0 == ppgtt->pd_dma_addr[0];
738         } else {
739                 u32 pp_db;
740                 pp_db = I915_READ(RING_PP_DIR_BASE(ring));
741                 return (pp_db >> 10) == ppgtt->pd_offset;
742         }
743 }
744
745 static struct drm_i915_error_object *
746 i915_error_first_batchbuffer(struct drm_i915_private *dev_priv,
747                              struct intel_ring_buffer *ring)
748 {
749         struct i915_address_space *vm;
750         struct i915_vma *vma;
751         struct drm_i915_gem_object *obj;
752         bool found_active = false;
753         u32 seqno;
754
755         if (!ring->get_seqno)
756                 return NULL;
757
758         if (HAS_BROKEN_CS_TLB(dev_priv->dev)) {
759                 u32 acthd = I915_READ(ACTHD);
760
761                 if (WARN_ON(ring->id != RCS))
762                         return NULL;
763
764                 obj = ring->scratch.obj;
765                 if (obj != NULL &&
766                     acthd >= i915_gem_obj_ggtt_offset(obj) &&
767                     acthd < i915_gem_obj_ggtt_offset(obj) + obj->base.size)
768                         return i915_error_ggtt_object_create(dev_priv, obj);
769         }
770
771         seqno = ring->get_seqno(ring, false);
772         list_for_each_entry(vm, &dev_priv->vm_list, global_link) {
773                 if (!is_active_vm(vm, ring))
774                         continue;
775
776                 found_active = true;
777
778                 list_for_each_entry(vma, &vm->active_list, mm_list) {
779                         obj = vma->obj;
780                         if (obj->ring != ring)
781                                 continue;
782
783                         if (i915_seqno_passed(seqno, obj->last_read_seqno))
784                                 continue;
785
786                         if ((obj->base.read_domains & I915_GEM_DOMAIN_COMMAND) == 0)
787                                 continue;
788
789                         /* We need to copy these to an anonymous buffer as the simplest
790                          * method to avoid being overwritten by userspace.
791                          */
792                         return i915_error_object_create(dev_priv, obj, vm);
793                 }
794         }
795
796         WARN_ON(!found_active);
797         return NULL;
798 }
799
800 static void i915_record_ring_state(struct drm_device *dev,
801                                    struct intel_ring_buffer *ring,
802                                    struct drm_i915_error_ring *ering)
803 {
804         struct drm_i915_private *dev_priv = dev->dev_private;
805
806         if (INTEL_INFO(dev)->gen >= 6) {
807                 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
808                 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
809                 ering->semaphore_mboxes[0]
810                         = I915_READ(RING_SYNC_0(ring->mmio_base));
811                 ering->semaphore_mboxes[1]
812                         = I915_READ(RING_SYNC_1(ring->mmio_base));
813                 ering->semaphore_seqno[0] = ring->sync_seqno[0];
814                 ering->semaphore_seqno[1] = ring->sync_seqno[1];
815         }
816
817         if (HAS_VEBOX(dev)) {
818                 ering->semaphore_mboxes[2] =
819                         I915_READ(RING_SYNC_2(ring->mmio_base));
820                 ering->semaphore_seqno[2] = ring->sync_seqno[2];
821         }
822
823         if (INTEL_INFO(dev)->gen >= 4) {
824                 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
825                 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
826                 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
827                 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
828                 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
829                 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
830                 if (INTEL_INFO(dev)->gen >= 8)
831                         ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
832                 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
833         } else {
834                 ering->faddr = I915_READ(DMA_FADD_I8XX);
835                 ering->ipeir = I915_READ(IPEIR);
836                 ering->ipehr = I915_READ(IPEHR);
837                 ering->instdone = I915_READ(INSTDONE);
838         }
839
840         ering->waiting = waitqueue_active(&ring->irq_queue);
841         ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
842         ering->seqno = ring->get_seqno(ring, false);
843         ering->acthd = intel_ring_get_active_head(ring);
844         ering->head = I915_READ_HEAD(ring);
845         ering->tail = I915_READ_TAIL(ring);
846         ering->ctl = I915_READ_CTL(ring);
847
848         if (I915_NEED_GFX_HWS(dev)) {
849                 int mmio;
850
851                 if (IS_GEN7(dev)) {
852                         switch (ring->id) {
853                         default:
854                         case RCS:
855                                 mmio = RENDER_HWS_PGA_GEN7;
856                                 break;
857                         case BCS:
858                                 mmio = BLT_HWS_PGA_GEN7;
859                                 break;
860                         case VCS:
861                                 mmio = BSD_HWS_PGA_GEN7;
862                                 break;
863                         case VECS:
864                                 mmio = VEBOX_HWS_PGA_GEN7;
865                                 break;
866                         }
867                 } else if (IS_GEN6(ring->dev)) {
868                         mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
869                 } else {
870                         /* XXX: gen8 returns to sanity */
871                         mmio = RING_HWS_PGA(ring->mmio_base);
872                 }
873
874                 ering->hws = I915_READ(mmio);
875         }
876
877         ering->cpu_ring_head = ring->head;
878         ering->cpu_ring_tail = ring->tail;
879
880         ering->hangcheck_score = ring->hangcheck.score;
881         ering->hangcheck_action = ring->hangcheck.action;
882
883         if (USES_PPGTT(dev)) {
884                 int i;
885
886                 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
887
888                 switch (INTEL_INFO(dev)->gen) {
889                 case 8:
890                         for (i = 0; i < 4; i++) {
891                                 ering->vm_info.pdp[i] =
892                                         I915_READ(GEN8_RING_PDP_UDW(ring, i));
893                                 ering->vm_info.pdp[i] <<= 32;
894                                 ering->vm_info.pdp[i] |=
895                                         I915_READ(GEN8_RING_PDP_LDW(ring, i));
896                         }
897                         break;
898                 case 7:
899                         ering->vm_info.pp_dir_base = RING_PP_DIR_BASE(ring);
900                         break;
901                 case 6:
902                         ering->vm_info.pp_dir_base = RING_PP_DIR_BASE_READ(ring);
903                         break;
904                 }
905         }
906 }
907
908
909 static void i915_gem_record_active_context(struct intel_ring_buffer *ring,
910                                            struct drm_i915_error_state *error,
911                                            struct drm_i915_error_ring *ering)
912 {
913         struct drm_i915_private *dev_priv = ring->dev->dev_private;
914         struct drm_i915_gem_object *obj;
915
916         /* Currently render ring is the only HW context user */
917         if (ring->id != RCS || !error->ccid)
918                 return;
919
920         list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
921                 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
922                         ering->ctx = i915_error_object_create_sized(dev_priv,
923                                                                     obj,
924                                                                     &dev_priv->gtt.base,
925                                                                     1);
926                         break;
927                 }
928         }
929 }
930
931 static void i915_gem_record_rings(struct drm_device *dev,
932                                   struct drm_i915_error_state *error)
933 {
934         struct drm_i915_private *dev_priv = dev->dev_private;
935         struct drm_i915_gem_request *request;
936         int i, count;
937
938         for (i = 0; i < I915_NUM_RINGS; i++) {
939                 struct intel_ring_buffer *ring = &dev_priv->ring[i];
940
941                 if (ring->dev == NULL)
942                         continue;
943
944                 error->ring[i].valid = true;
945
946                 i915_record_ring_state(dev, ring, &error->ring[i]);
947
948                 error->ring[i].batchbuffer =
949                         i915_error_first_batchbuffer(dev_priv, ring);
950
951                 error->ring[i].ringbuffer =
952                         i915_error_ggtt_object_create(dev_priv, ring->obj);
953
954                 if (ring->status_page.obj)
955                         error->ring[i].hws_page =
956                                 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
957
958                 i915_gem_record_active_context(ring, error, &error->ring[i]);
959
960                 count = 0;
961                 list_for_each_entry(request, &ring->request_list, list)
962                         count++;
963
964                 error->ring[i].num_requests = count;
965                 error->ring[i].requests =
966                         kcalloc(count, sizeof(*error->ring[i].requests),
967                                 GFP_ATOMIC);
968                 if (error->ring[i].requests == NULL) {
969                         error->ring[i].num_requests = 0;
970                         continue;
971                 }
972
973                 count = 0;
974                 list_for_each_entry(request, &ring->request_list, list) {
975                         struct drm_i915_error_request *erq;
976
977                         erq = &error->ring[i].requests[count++];
978                         erq->seqno = request->seqno;
979                         erq->jiffies = request->emitted_jiffies;
980                         erq->tail = request->tail;
981                 }
982         }
983 }
984
985 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
986  * VM.
987  */
988 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
989                                 struct drm_i915_error_state *error,
990                                 struct i915_address_space *vm,
991                                 const int ndx)
992 {
993         struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
994         struct drm_i915_gem_object *obj;
995         struct i915_vma *vma;
996         int i;
997
998         i = 0;
999         list_for_each_entry(vma, &vm->active_list, mm_list)
1000                 i++;
1001         error->active_bo_count[ndx] = i;
1002         list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
1003                 if (i915_gem_obj_is_pinned(obj))
1004                         i++;
1005         error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1006
1007         if (i) {
1008                 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1009                 if (active_bo)
1010                         pinned_bo = active_bo + error->active_bo_count[ndx];
1011         }
1012
1013         if (active_bo)
1014                 error->active_bo_count[ndx] =
1015                         capture_active_bo(active_bo,
1016                                           error->active_bo_count[ndx],
1017                                           &vm->active_list);
1018
1019         if (pinned_bo)
1020                 error->pinned_bo_count[ndx] =
1021                         capture_pinned_bo(pinned_bo,
1022                                           error->pinned_bo_count[ndx],
1023                                           &dev_priv->mm.bound_list);
1024         error->active_bo[ndx] = active_bo;
1025         error->pinned_bo[ndx] = pinned_bo;
1026 }
1027
1028 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1029                                      struct drm_i915_error_state *error)
1030 {
1031         struct i915_address_space *vm;
1032         int cnt = 0, i = 0;
1033
1034         list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1035                 cnt++;
1036
1037         error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1038         error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1039         error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1040                                          GFP_ATOMIC);
1041         error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1042                                          GFP_ATOMIC);
1043
1044         list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1045                 i915_gem_capture_vm(dev_priv, error, vm, i++);
1046 }
1047
1048 /* Capture all registers which don't fit into another category. */
1049 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1050                                    struct drm_i915_error_state *error)
1051 {
1052         struct drm_device *dev = dev_priv->dev;
1053         int pipe;
1054
1055         /* General organization
1056          * 1. Registers specific to a single generation
1057          * 2. Registers which belong to multiple generations
1058          * 3. Feature specific registers.
1059          * 4. Everything else
1060          * Please try to follow the order.
1061          */
1062
1063         /* 1: Registers specific to a single generation */
1064         if (IS_VALLEYVIEW(dev)) {
1065                 error->ier = I915_READ(GTIER) | I915_READ(VLV_IER);
1066                 error->forcewake = I915_READ(FORCEWAKE_VLV);
1067         }
1068
1069         if (IS_GEN7(dev))
1070                 error->err_int = I915_READ(GEN7_ERR_INT);
1071
1072         if (IS_GEN6(dev)) {
1073                 error->forcewake = I915_READ(FORCEWAKE);
1074                 error->gab_ctl = I915_READ(GAB_CTL);
1075                 error->gfx_mode = I915_READ(GFX_MODE);
1076         }
1077
1078         if (IS_GEN2(dev))
1079                 error->ier = I915_READ16(IER);
1080
1081         /* 2: Registers which belong to multiple generations */
1082         if (INTEL_INFO(dev)->gen >= 7)
1083                 error->forcewake = I915_READ(FORCEWAKE_MT);
1084
1085         if (INTEL_INFO(dev)->gen >= 6) {
1086                 error->derrmr = I915_READ(DERRMR);
1087                 error->error = I915_READ(ERROR_GEN6);
1088                 error->done_reg = I915_READ(DONE_REG);
1089         }
1090
1091         /* 3: Feature specific registers */
1092         if (IS_GEN6(dev) || IS_GEN7(dev)) {
1093                 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1094                 error->gac_eco = I915_READ(GAC_ECO_BITS);
1095         }
1096
1097         /* 4: Everything else */
1098         if (HAS_HW_CONTEXTS(dev))
1099                 error->ccid = I915_READ(CCID);
1100
1101         if (HAS_PCH_SPLIT(dev))
1102                 error->ier = I915_READ(DEIER) | I915_READ(GTIER);
1103         else {
1104                 error->ier = I915_READ(IER);
1105                 for_each_pipe(pipe)
1106                         error->pipestat[pipe] = I915_READ(PIPESTAT(pipe));
1107         }
1108
1109         /* 4: Everything else */
1110         error->eir = I915_READ(EIR);
1111         error->pgtbl_er = I915_READ(PGTBL_ER);
1112
1113         i915_get_extra_instdone(dev, error->extra_instdone);
1114 }
1115
1116 /**
1117  * i915_capture_error_state - capture an error record for later analysis
1118  * @dev: drm device
1119  *
1120  * Should be called when an error is detected (either a hang or an error
1121  * interrupt) to capture error state from the time of the error.  Fills
1122  * out a structure which becomes available in debugfs for user level tools
1123  * to pick up.
1124  */
1125 void i915_capture_error_state(struct drm_device *dev)
1126 {
1127         static bool warned;
1128         struct drm_i915_private *dev_priv = dev->dev_private;
1129         struct drm_i915_error_state *error;
1130         unsigned long flags;
1131         uint32_t ecode;
1132
1133         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1134         error = dev_priv->gpu_error.first_error;
1135         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1136         if (error)
1137                 return;
1138
1139         /* Account for pipe specific data like PIPE*STAT */
1140         error = kzalloc(sizeof(*error), GFP_ATOMIC);
1141         if (!error) {
1142                 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1143                 return;
1144         }
1145
1146         DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1147                  dev->primary->index);
1148         kref_init(&error->ref);
1149
1150         i915_capture_reg_state(dev_priv, error);
1151         i915_gem_capture_buffers(dev_priv, error);
1152         i915_gem_record_fences(dev, error);
1153         i915_gem_record_rings(dev, error);
1154         ecode = i915_error_generate_code(dev_priv, error);
1155
1156         if (!warned) {
1157                 DRM_INFO("GPU HANG [%x]\n", ecode);
1158                 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1159                 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1160                 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1161                 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1162                 warned = true;
1163         }
1164
1165         do_gettimeofday(&error->time);
1166
1167         error->overlay = intel_overlay_capture_error_state(dev);
1168         error->display = intel_display_capture_error_state(dev);
1169
1170         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1171         if (dev_priv->gpu_error.first_error == NULL) {
1172                 dev_priv->gpu_error.first_error = error;
1173                 error = NULL;
1174         }
1175         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1176
1177         if (error)
1178                 i915_error_state_free(&error->ref);
1179 }
1180
1181 void i915_error_state_get(struct drm_device *dev,
1182                           struct i915_error_state_file_priv *error_priv)
1183 {
1184         struct drm_i915_private *dev_priv = dev->dev_private;
1185         unsigned long flags;
1186
1187         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1188         error_priv->error = dev_priv->gpu_error.first_error;
1189         if (error_priv->error)
1190                 kref_get(&error_priv->error->ref);
1191         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1192
1193 }
1194
1195 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1196 {
1197         if (error_priv->error)
1198                 kref_put(&error_priv->error->ref, i915_error_state_free);
1199 }
1200
1201 void i915_destroy_error_state(struct drm_device *dev)
1202 {
1203         struct drm_i915_private *dev_priv = dev->dev_private;
1204         struct drm_i915_error_state *error;
1205         unsigned long flags;
1206
1207         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1208         error = dev_priv->gpu_error.first_error;
1209         dev_priv->gpu_error.first_error = NULL;
1210         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1211
1212         if (error)
1213                 kref_put(&error->ref, i915_error_state_free);
1214 }
1215
1216 const char *i915_cache_level_str(int type)
1217 {
1218         switch (type) {
1219         case I915_CACHE_NONE: return " uncached";
1220         case I915_CACHE_LLC: return " snooped or LLC";
1221         case I915_CACHE_L3_LLC: return " L3+LLC";
1222         case I915_CACHE_WT: return " WT";
1223         default: return "";
1224         }
1225 }
1226
1227 /* NB: please notice the memset */
1228 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1229 {
1230         struct drm_i915_private *dev_priv = dev->dev_private;
1231         memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1232
1233         switch (INTEL_INFO(dev)->gen) {
1234         case 2:
1235         case 3:
1236                 instdone[0] = I915_READ(INSTDONE);
1237                 break;
1238         case 4:
1239         case 5:
1240         case 6:
1241                 instdone[0] = I915_READ(INSTDONE_I965);
1242                 instdone[1] = I915_READ(INSTDONE1);
1243                 break;
1244         default:
1245                 WARN_ONCE(1, "Unsupported platform\n");
1246         case 7:
1247         case 8:
1248                 instdone[0] = I915_READ(GEN7_INSTDONE_1);
1249                 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1250                 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1251                 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
1252                 break;
1253         }
1254 }