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
ring_str(int ring)33 static const char *ring_str(int ring)
34 {
35 switch (ring) {
36 case RCS: return "render";
37 case VCS: return "bsd";
38 case BCS: return "blt";
39 case VECS: return "vebox";
40 case VCS2: return "bsd2";
41 default: return "";
42 }
43 }
44
pin_flag(int pinned)45 static const char *pin_flag(int pinned)
46 {
47 if (pinned > 0)
48 return " P";
49 else if (pinned < 0)
50 return " p";
51 else
52 return "";
53 }
54
tiling_flag(int tiling)55 static const char *tiling_flag(int tiling)
56 {
57 switch (tiling) {
58 default:
59 case I915_TILING_NONE: return "";
60 case I915_TILING_X: return " X";
61 case I915_TILING_Y: return " Y";
62 }
63 }
64
dirty_flag(int dirty)65 static const char *dirty_flag(int dirty)
66 {
67 return dirty ? " dirty" : "";
68 }
69
purgeable_flag(int purgeable)70 static const char *purgeable_flag(int purgeable)
71 {
72 return purgeable ? " purgeable" : "";
73 }
74
__i915_error_ok(struct drm_i915_error_state_buf * e)75 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
76 {
77
78 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
79 e->err = -ENOSPC;
80 return false;
81 }
82
83 if (e->bytes == e->size - 1 || e->err)
84 return false;
85
86 return true;
87 }
88
__i915_error_seek(struct drm_i915_error_state_buf * e,unsigned len)89 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
90 unsigned len)
91 {
92 if (e->pos + len <= e->start) {
93 e->pos += len;
94 return false;
95 }
96
97 /* First vsnprintf needs to fit in its entirety for memmove */
98 if (len >= e->size) {
99 e->err = -EIO;
100 return false;
101 }
102
103 return true;
104 }
105
__i915_error_advance(struct drm_i915_error_state_buf * e,unsigned len)106 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
107 unsigned len)
108 {
109 /* If this is first printf in this window, adjust it so that
110 * start position matches start of the buffer
111 */
112
113 if (e->pos < e->start) {
114 const size_t off = e->start - e->pos;
115
116 /* Should not happen but be paranoid */
117 if (off > len || e->bytes) {
118 e->err = -EIO;
119 return;
120 }
121
122 memmove(e->buf, e->buf + off, len - off);
123 e->bytes = len - off;
124 e->pos = e->start;
125 return;
126 }
127
128 e->bytes += len;
129 e->pos += len;
130 }
131
i915_error_vprintf(struct drm_i915_error_state_buf * e,const char * f,va_list args)132 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
133 const char *f, va_list args)
134 {
135 unsigned len;
136
137 if (!__i915_error_ok(e))
138 return;
139
140 /* Seek the first printf which is hits start position */
141 if (e->pos < e->start) {
142 va_list tmp;
143
144 va_copy(tmp, args);
145 len = vsnprintf(NULL, 0, f, tmp);
146 va_end(tmp);
147
148 if (!__i915_error_seek(e, len))
149 return;
150 }
151
152 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
153 if (len >= e->size - e->bytes)
154 len = e->size - e->bytes - 1;
155
156 __i915_error_advance(e, len);
157 }
158
i915_error_puts(struct drm_i915_error_state_buf * e,const char * str)159 static void i915_error_puts(struct drm_i915_error_state_buf *e,
160 const char *str)
161 {
162 unsigned len;
163
164 if (!__i915_error_ok(e))
165 return;
166
167 len = strlen(str);
168
169 /* Seek the first printf which is hits start position */
170 if (e->pos < e->start) {
171 if (!__i915_error_seek(e, len))
172 return;
173 }
174
175 if (len >= e->size - e->bytes)
176 len = e->size - e->bytes - 1;
177 memcpy(e->buf + e->bytes, str, len);
178
179 __i915_error_advance(e, len);
180 }
181
182 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
183 #define err_puts(e, s) i915_error_puts(e, s)
184
print_error_buffers(struct drm_i915_error_state_buf * m,const char * name,struct drm_i915_error_buffer * err,int count)185 static void print_error_buffers(struct drm_i915_error_state_buf *m,
186 const char *name,
187 struct drm_i915_error_buffer *err,
188 int count)
189 {
190 int i;
191
192 err_printf(m, " %s [%d]:\n", name, count);
193
194 while (count--) {
195 err_printf(m, " %08x_%08x %8u %02x %02x [ ",
196 upper_32_bits(err->gtt_offset),
197 lower_32_bits(err->gtt_offset),
198 err->size,
199 err->read_domains,
200 err->write_domain);
201 for (i = 0; i < I915_NUM_RINGS; i++)
202 err_printf(m, "%02x ", err->rseqno[i]);
203
204 err_printf(m, "] %02x", err->wseqno);
205 err_puts(m, pin_flag(err->pinned));
206 err_puts(m, tiling_flag(err->tiling));
207 err_puts(m, dirty_flag(err->dirty));
208 err_puts(m, purgeable_flag(err->purgeable));
209 err_puts(m, err->userptr ? " userptr" : "");
210 err_puts(m, err->ring != -1 ? " " : "");
211 err_puts(m, ring_str(err->ring));
212 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
213
214 if (err->name)
215 err_printf(m, " (name: %d)", err->name);
216 if (err->fence_reg != I915_FENCE_REG_NONE)
217 err_printf(m, " (fence: %d)", err->fence_reg);
218
219 err_puts(m, "\n");
220 err++;
221 }
222 }
223
hangcheck_action_to_str(enum intel_ring_hangcheck_action a)224 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
225 {
226 switch (a) {
227 case HANGCHECK_IDLE:
228 return "idle";
229 case HANGCHECK_WAIT:
230 return "wait";
231 case HANGCHECK_ACTIVE:
232 return "active";
233 case HANGCHECK_ACTIVE_LOOP:
234 return "active (loop)";
235 case HANGCHECK_KICK:
236 return "kick";
237 case HANGCHECK_HUNG:
238 return "hung";
239 }
240
241 return "unknown";
242 }
243
i915_ring_error_state(struct drm_i915_error_state_buf * m,struct drm_device * dev,struct drm_i915_error_state * error,int ring_idx)244 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
245 struct drm_device *dev,
246 struct drm_i915_error_state *error,
247 int ring_idx)
248 {
249 struct drm_i915_error_ring *ring = &error->ring[ring_idx];
250
251 if (!ring->valid)
252 return;
253
254 err_printf(m, "%s command stream:\n", ring_str(ring_idx));
255 err_printf(m, " START: 0x%08x\n", ring->start);
256 err_printf(m, " HEAD: 0x%08x\n", ring->head);
257 err_printf(m, " TAIL: 0x%08x\n", ring->tail);
258 err_printf(m, " CTL: 0x%08x\n", ring->ctl);
259 err_printf(m, " HWS: 0x%08x\n", ring->hws);
260 err_printf(m, " ACTHD: 0x%08x %08x\n", (u32)(ring->acthd>>32), (u32)ring->acthd);
261 err_printf(m, " IPEIR: 0x%08x\n", ring->ipeir);
262 err_printf(m, " IPEHR: 0x%08x\n", ring->ipehr);
263 err_printf(m, " INSTDONE: 0x%08x\n", ring->instdone);
264 if (INTEL_INFO(dev)->gen >= 4) {
265 err_printf(m, " BBADDR: 0x%08x %08x\n", (u32)(ring->bbaddr>>32), (u32)ring->bbaddr);
266 err_printf(m, " BB_STATE: 0x%08x\n", ring->bbstate);
267 err_printf(m, " INSTPS: 0x%08x\n", ring->instps);
268 }
269 err_printf(m, " INSTPM: 0x%08x\n", ring->instpm);
270 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ring->faddr),
271 lower_32_bits(ring->faddr));
272 if (INTEL_INFO(dev)->gen >= 6) {
273 err_printf(m, " RC PSMI: 0x%08x\n", ring->rc_psmi);
274 err_printf(m, " FAULT_REG: 0x%08x\n", ring->fault_reg);
275 err_printf(m, " SYNC_0: 0x%08x [last synced 0x%08x]\n",
276 ring->semaphore_mboxes[0],
277 ring->semaphore_seqno[0]);
278 err_printf(m, " SYNC_1: 0x%08x [last synced 0x%08x]\n",
279 ring->semaphore_mboxes[1],
280 ring->semaphore_seqno[1]);
281 if (HAS_VEBOX(dev)) {
282 err_printf(m, " SYNC_2: 0x%08x [last synced 0x%08x]\n",
283 ring->semaphore_mboxes[2],
284 ring->semaphore_seqno[2]);
285 }
286 }
287 if (USES_PPGTT(dev)) {
288 err_printf(m, " GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
289
290 if (INTEL_INFO(dev)->gen >= 8) {
291 int i;
292 for (i = 0; i < 4; i++)
293 err_printf(m, " PDP%d: 0x%016llx\n",
294 i, ring->vm_info.pdp[i]);
295 } else {
296 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
297 ring->vm_info.pp_dir_base);
298 }
299 }
300 err_printf(m, " seqno: 0x%08x\n", ring->seqno);
301 err_printf(m, " waiting: %s\n", yesno(ring->waiting));
302 err_printf(m, " ring->head: 0x%08x\n", ring->cpu_ring_head);
303 err_printf(m, " ring->tail: 0x%08x\n", ring->cpu_ring_tail);
304 err_printf(m, " hangcheck: %s [%d]\n",
305 hangcheck_action_to_str(ring->hangcheck_action),
306 ring->hangcheck_score);
307 }
308
i915_error_printf(struct drm_i915_error_state_buf * e,const char * f,...)309 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
310 {
311 va_list args;
312
313 va_start(args, f);
314 i915_error_vprintf(e, f, args);
315 va_end(args);
316 }
317
print_error_obj(struct drm_i915_error_state_buf * m,struct drm_i915_error_object * obj)318 static void print_error_obj(struct drm_i915_error_state_buf *m,
319 struct drm_i915_error_object *obj)
320 {
321 int page, offset, elt;
322
323 for (page = offset = 0; page < obj->page_count; page++) {
324 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
325 err_printf(m, "%08x : %08x\n", offset,
326 obj->pages[page][elt]);
327 offset += 4;
328 }
329 }
330 }
331
i915_error_state_to_str(struct drm_i915_error_state_buf * m,const struct i915_error_state_file_priv * error_priv)332 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
333 const struct i915_error_state_file_priv *error_priv)
334 {
335 struct drm_device *dev = error_priv->dev;
336 struct drm_i915_private *dev_priv = dev->dev_private;
337 struct drm_i915_error_state *error = error_priv->error;
338 struct drm_i915_error_object *obj;
339 int i, j, offset, elt;
340 int max_hangcheck_score;
341
342 if (!error) {
343 err_printf(m, "no error state collected\n");
344 goto out;
345 }
346
347 err_printf(m, "%s\n", error->error_msg);
348 err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
349 error->time.tv_usec);
350 err_printf(m, "Kernel: " UTS_RELEASE "\n");
351 max_hangcheck_score = 0;
352 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
353 if (error->ring[i].hangcheck_score > max_hangcheck_score)
354 max_hangcheck_score = error->ring[i].hangcheck_score;
355 }
356 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
357 if (error->ring[i].hangcheck_score == max_hangcheck_score &&
358 error->ring[i].pid != -1) {
359 err_printf(m, "Active process (on ring %s): %s [%d]\n",
360 ring_str(i),
361 error->ring[i].comm,
362 error->ring[i].pid);
363 }
364 }
365 err_printf(m, "Reset count: %u\n", error->reset_count);
366 err_printf(m, "Suspend count: %u\n", error->suspend_count);
367 err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
368 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
369 err_printf(m, "EIR: 0x%08x\n", error->eir);
370 err_printf(m, "IER: 0x%08x\n", error->ier);
371 if (INTEL_INFO(dev)->gen >= 8) {
372 for (i = 0; i < 4; i++)
373 err_printf(m, "GTIER gt %d: 0x%08x\n", i,
374 error->gtier[i]);
375 } else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
376 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
377 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
378 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
379 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
380 err_printf(m, "CCID: 0x%08x\n", error->ccid);
381 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
382
383 for (i = 0; i < dev_priv->num_fence_regs; i++)
384 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
385
386 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
387 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
388 error->extra_instdone[i]);
389
390 if (INTEL_INFO(dev)->gen >= 6) {
391 err_printf(m, "ERROR: 0x%08x\n", error->error);
392
393 if (INTEL_INFO(dev)->gen >= 8)
394 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
395 error->fault_data1, error->fault_data0);
396
397 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
398 }
399
400 if (INTEL_INFO(dev)->gen == 7)
401 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
402
403 for (i = 0; i < ARRAY_SIZE(error->ring); i++)
404 i915_ring_error_state(m, dev, error, i);
405
406 for (i = 0; i < error->vm_count; i++) {
407 err_printf(m, "vm[%d]\n", i);
408
409 print_error_buffers(m, "Active",
410 error->active_bo[i],
411 error->active_bo_count[i]);
412
413 print_error_buffers(m, "Pinned",
414 error->pinned_bo[i],
415 error->pinned_bo_count[i]);
416 }
417
418 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
419 obj = error->ring[i].batchbuffer;
420 if (obj) {
421 err_puts(m, dev_priv->ring[i].name);
422 if (error->ring[i].pid != -1)
423 err_printf(m, " (submitted by %s [%d])",
424 error->ring[i].comm,
425 error->ring[i].pid);
426 err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
427 upper_32_bits(obj->gtt_offset),
428 lower_32_bits(obj->gtt_offset));
429 print_error_obj(m, obj);
430 }
431
432 obj = error->ring[i].wa_batchbuffer;
433 if (obj) {
434 err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
435 dev_priv->ring[i].name,
436 lower_32_bits(obj->gtt_offset));
437 print_error_obj(m, obj);
438 }
439
440 if (error->ring[i].num_requests) {
441 err_printf(m, "%s --- %d requests\n",
442 dev_priv->ring[i].name,
443 error->ring[i].num_requests);
444 for (j = 0; j < error->ring[i].num_requests; j++) {
445 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
446 error->ring[i].requests[j].seqno,
447 error->ring[i].requests[j].jiffies,
448 error->ring[i].requests[j].tail);
449 }
450 }
451
452 if ((obj = error->ring[i].ringbuffer)) {
453 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
454 dev_priv->ring[i].name,
455 lower_32_bits(obj->gtt_offset));
456 print_error_obj(m, obj);
457 }
458
459 if ((obj = error->ring[i].hws_page)) {
460 u64 hws_offset = obj->gtt_offset;
461 u32 *hws_page = &obj->pages[0][0];
462
463 if (i915.enable_execlists) {
464 hws_offset += LRC_PPHWSP_PN * PAGE_SIZE;
465 hws_page = &obj->pages[LRC_PPHWSP_PN][0];
466 }
467 err_printf(m, "%s --- HW Status = 0x%08llx\n",
468 dev_priv->ring[i].name, hws_offset);
469 offset = 0;
470 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
471 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
472 offset,
473 hws_page[elt],
474 hws_page[elt+1],
475 hws_page[elt+2],
476 hws_page[elt+3]);
477 offset += 16;
478 }
479 }
480
481 if ((obj = error->ring[i].ctx)) {
482 err_printf(m, "%s --- HW Context = 0x%08x\n",
483 dev_priv->ring[i].name,
484 lower_32_bits(obj->gtt_offset));
485 print_error_obj(m, obj);
486 }
487 }
488
489 if ((obj = error->semaphore_obj)) {
490 err_printf(m, "Semaphore page = 0x%08x\n",
491 lower_32_bits(obj->gtt_offset));
492 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
493 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
494 elt * 4,
495 obj->pages[0][elt],
496 obj->pages[0][elt+1],
497 obj->pages[0][elt+2],
498 obj->pages[0][elt+3]);
499 }
500 }
501
502 if (error->overlay)
503 intel_overlay_print_error_state(m, error->overlay);
504
505 if (error->display)
506 intel_display_print_error_state(m, dev, error->display);
507
508 out:
509 if (m->bytes == 0 && m->err)
510 return m->err;
511
512 return 0;
513 }
514
i915_error_state_buf_init(struct drm_i915_error_state_buf * ebuf,struct drm_i915_private * i915,size_t count,loff_t pos)515 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
516 struct drm_i915_private *i915,
517 size_t count, loff_t pos)
518 {
519 memset(ebuf, 0, sizeof(*ebuf));
520 ebuf->i915 = i915;
521
522 /* We need to have enough room to store any i915_error_state printf
523 * so that we can move it to start position.
524 */
525 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
526 ebuf->buf = kmalloc(ebuf->size,
527 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
528
529 if (ebuf->buf == NULL) {
530 ebuf->size = PAGE_SIZE;
531 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
532 }
533
534 if (ebuf->buf == NULL) {
535 ebuf->size = 128;
536 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
537 }
538
539 if (ebuf->buf == NULL)
540 return -ENOMEM;
541
542 ebuf->start = pos;
543
544 return 0;
545 }
546
i915_error_object_free(struct drm_i915_error_object * obj)547 static void i915_error_object_free(struct drm_i915_error_object *obj)
548 {
549 int page;
550
551 if (obj == NULL)
552 return;
553
554 for (page = 0; page < obj->page_count; page++)
555 kfree(obj->pages[page]);
556
557 kfree(obj);
558 }
559
i915_error_state_free(struct kref * error_ref)560 static void i915_error_state_free(struct kref *error_ref)
561 {
562 struct drm_i915_error_state *error = container_of(error_ref,
563 typeof(*error), ref);
564 int i;
565
566 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
567 i915_error_object_free(error->ring[i].batchbuffer);
568 i915_error_object_free(error->ring[i].wa_batchbuffer);
569 i915_error_object_free(error->ring[i].ringbuffer);
570 i915_error_object_free(error->ring[i].hws_page);
571 i915_error_object_free(error->ring[i].ctx);
572 kfree(error->ring[i].requests);
573 }
574
575 i915_error_object_free(error->semaphore_obj);
576
577 for (i = 0; i < error->vm_count; i++)
578 kfree(error->active_bo[i]);
579
580 kfree(error->active_bo);
581 kfree(error->active_bo_count);
582 kfree(error->pinned_bo);
583 kfree(error->pinned_bo_count);
584 kfree(error->overlay);
585 kfree(error->display);
586 kfree(error);
587 }
588
589 static struct drm_i915_error_object *
i915_error_object_create(struct drm_i915_private * dev_priv,struct drm_i915_gem_object * src,struct i915_address_space * vm)590 i915_error_object_create(struct drm_i915_private *dev_priv,
591 struct drm_i915_gem_object *src,
592 struct i915_address_space *vm)
593 {
594 struct drm_i915_error_object *dst;
595 struct i915_vma *vma = NULL;
596 int num_pages;
597 bool use_ggtt;
598 int i = 0;
599 u64 reloc_offset;
600
601 if (src == NULL || src->pages == NULL)
602 return NULL;
603
604 num_pages = src->base.size >> PAGE_SHIFT;
605
606 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
607 if (dst == NULL)
608 return NULL;
609
610 if (i915_gem_obj_bound(src, vm))
611 dst->gtt_offset = i915_gem_obj_offset(src, vm);
612 else
613 dst->gtt_offset = -1;
614
615 reloc_offset = dst->gtt_offset;
616 if (i915_is_ggtt(vm))
617 vma = i915_gem_obj_to_ggtt(src);
618 use_ggtt = (src->cache_level == I915_CACHE_NONE &&
619 vma && (vma->bound & GLOBAL_BIND) &&
620 reloc_offset + num_pages * PAGE_SIZE <= dev_priv->gtt.mappable_end);
621
622 /* Cannot access stolen address directly, try to use the aperture */
623 if (src->stolen) {
624 use_ggtt = true;
625
626 if (!(vma && vma->bound & GLOBAL_BIND))
627 goto unwind;
628
629 reloc_offset = i915_gem_obj_ggtt_offset(src);
630 if (reloc_offset + num_pages * PAGE_SIZE > dev_priv->gtt.mappable_end)
631 goto unwind;
632 }
633
634 /* Cannot access snooped pages through the aperture */
635 if (use_ggtt && src->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv->dev))
636 goto unwind;
637
638 dst->page_count = num_pages;
639 while (num_pages--) {
640 unsigned long flags;
641 void *d;
642
643 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
644 if (d == NULL)
645 goto unwind;
646
647 local_irq_save(flags);
648 if (use_ggtt) {
649 void __iomem *s;
650
651 /* Simply ignore tiling or any overlapping fence.
652 * It's part of the error state, and this hopefully
653 * captures what the GPU read.
654 */
655
656 s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
657 reloc_offset);
658 memcpy_fromio(d, s, PAGE_SIZE);
659 io_mapping_unmap_atomic(s);
660 } else {
661 struct page *page;
662 void *s;
663
664 page = i915_gem_object_get_page(src, i);
665
666 drm_clflush_pages(&page, 1);
667
668 s = kmap_atomic(page);
669 memcpy(d, s, PAGE_SIZE);
670 kunmap_atomic(s);
671
672 drm_clflush_pages(&page, 1);
673 }
674 local_irq_restore(flags);
675
676 dst->pages[i++] = d;
677 reloc_offset += PAGE_SIZE;
678 }
679
680 return dst;
681
682 unwind:
683 while (i--)
684 kfree(dst->pages[i]);
685 kfree(dst);
686 return NULL;
687 }
688 #define i915_error_ggtt_object_create(dev_priv, src) \
689 i915_error_object_create((dev_priv), (src), &(dev_priv)->gtt.base)
690
capture_bo(struct drm_i915_error_buffer * err,struct i915_vma * vma)691 static void capture_bo(struct drm_i915_error_buffer *err,
692 struct i915_vma *vma)
693 {
694 struct drm_i915_gem_object *obj = vma->obj;
695 int i;
696
697 err->size = obj->base.size;
698 err->name = obj->base.name;
699 for (i = 0; i < I915_NUM_RINGS; i++)
700 err->rseqno[i] = i915_gem_request_get_seqno(obj->last_read_req[i]);
701 err->wseqno = i915_gem_request_get_seqno(obj->last_write_req);
702 err->gtt_offset = vma->node.start;
703 err->read_domains = obj->base.read_domains;
704 err->write_domain = obj->base.write_domain;
705 err->fence_reg = obj->fence_reg;
706 err->pinned = 0;
707 if (i915_gem_obj_is_pinned(obj))
708 err->pinned = 1;
709 err->tiling = obj->tiling_mode;
710 err->dirty = obj->dirty;
711 err->purgeable = obj->madv != I915_MADV_WILLNEED;
712 err->userptr = obj->userptr.mm != NULL;
713 err->ring = obj->last_write_req ?
714 i915_gem_request_get_ring(obj->last_write_req)->id : -1;
715 err->cache_level = obj->cache_level;
716 }
717
capture_active_bo(struct drm_i915_error_buffer * err,int count,struct list_head * head)718 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
719 int count, struct list_head *head)
720 {
721 struct i915_vma *vma;
722 int i = 0;
723
724 list_for_each_entry(vma, head, mm_list) {
725 capture_bo(err++, vma);
726 if (++i == count)
727 break;
728 }
729
730 return i;
731 }
732
capture_pinned_bo(struct drm_i915_error_buffer * err,int count,struct list_head * head,struct i915_address_space * vm)733 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
734 int count, struct list_head *head,
735 struct i915_address_space *vm)
736 {
737 struct drm_i915_gem_object *obj;
738 struct drm_i915_error_buffer * const first = err;
739 struct drm_i915_error_buffer * const last = err + count;
740
741 list_for_each_entry(obj, head, global_list) {
742 struct i915_vma *vma;
743
744 if (err == last)
745 break;
746
747 list_for_each_entry(vma, &obj->vma_list, vma_link)
748 if (vma->vm == vm && vma->pin_count > 0)
749 capture_bo(err++, vma);
750 }
751
752 return err - first;
753 }
754
755 /* Generate a semi-unique error code. The code is not meant to have meaning, The
756 * code's only purpose is to try to prevent false duplicated bug reports by
757 * grossly estimating a GPU error state.
758 *
759 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
760 * the hang if we could strip the GTT offset information from it.
761 *
762 * It's only a small step better than a random number in its current form.
763 */
i915_error_generate_code(struct drm_i915_private * dev_priv,struct drm_i915_error_state * error,int * ring_id)764 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
765 struct drm_i915_error_state *error,
766 int *ring_id)
767 {
768 uint32_t error_code = 0;
769 int i;
770
771 /* IPEHR would be an ideal way to detect errors, as it's the gross
772 * measure of "the command that hung." However, has some very common
773 * synchronization commands which almost always appear in the case
774 * strictly a client bug. Use instdone to differentiate those some.
775 */
776 for (i = 0; i < I915_NUM_RINGS; i++) {
777 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
778 if (ring_id)
779 *ring_id = i;
780
781 return error->ring[i].ipehr ^ error->ring[i].instdone;
782 }
783 }
784
785 return error_code;
786 }
787
i915_gem_record_fences(struct drm_device * dev,struct drm_i915_error_state * error)788 static void i915_gem_record_fences(struct drm_device *dev,
789 struct drm_i915_error_state *error)
790 {
791 struct drm_i915_private *dev_priv = dev->dev_private;
792 int i;
793
794 if (IS_GEN3(dev) || IS_GEN2(dev)) {
795 for (i = 0; i < dev_priv->num_fence_regs; i++)
796 error->fence[i] = I915_READ(FENCE_REG(i));
797 } else if (IS_GEN5(dev) || IS_GEN4(dev)) {
798 for (i = 0; i < dev_priv->num_fence_regs; i++)
799 error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
800 } else if (INTEL_INFO(dev)->gen >= 6) {
801 for (i = 0; i < dev_priv->num_fence_regs; i++)
802 error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
803 }
804 }
805
806
gen8_record_semaphore_state(struct drm_i915_private * dev_priv,struct drm_i915_error_state * error,struct intel_engine_cs * ring,struct drm_i915_error_ring * ering)807 static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
808 struct drm_i915_error_state *error,
809 struct intel_engine_cs *ring,
810 struct drm_i915_error_ring *ering)
811 {
812 struct intel_engine_cs *to;
813 int i;
814
815 if (!i915_semaphore_is_enabled(dev_priv->dev))
816 return;
817
818 if (!error->semaphore_obj)
819 error->semaphore_obj =
820 i915_error_ggtt_object_create(dev_priv,
821 dev_priv->semaphore_obj);
822
823 for_each_ring(to, dev_priv, i) {
824 int idx;
825 u16 signal_offset;
826 u32 *tmp;
827
828 if (ring == to)
829 continue;
830
831 signal_offset = (GEN8_SIGNAL_OFFSET(ring, i) & (PAGE_SIZE - 1))
832 / 4;
833 tmp = error->semaphore_obj->pages[0];
834 idx = intel_ring_sync_index(ring, to);
835
836 ering->semaphore_mboxes[idx] = tmp[signal_offset];
837 ering->semaphore_seqno[idx] = ring->semaphore.sync_seqno[idx];
838 }
839 }
840
gen6_record_semaphore_state(struct drm_i915_private * dev_priv,struct intel_engine_cs * ring,struct drm_i915_error_ring * ering)841 static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
842 struct intel_engine_cs *ring,
843 struct drm_i915_error_ring *ering)
844 {
845 ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(ring->mmio_base));
846 ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(ring->mmio_base));
847 ering->semaphore_seqno[0] = ring->semaphore.sync_seqno[0];
848 ering->semaphore_seqno[1] = ring->semaphore.sync_seqno[1];
849
850 if (HAS_VEBOX(dev_priv->dev)) {
851 ering->semaphore_mboxes[2] =
852 I915_READ(RING_SYNC_2(ring->mmio_base));
853 ering->semaphore_seqno[2] = ring->semaphore.sync_seqno[2];
854 }
855 }
856
i915_record_ring_state(struct drm_device * dev,struct drm_i915_error_state * error,struct intel_engine_cs * ring,struct drm_i915_error_ring * ering)857 static void i915_record_ring_state(struct drm_device *dev,
858 struct drm_i915_error_state *error,
859 struct intel_engine_cs *ring,
860 struct drm_i915_error_ring *ering)
861 {
862 struct drm_i915_private *dev_priv = dev->dev_private;
863
864 if (INTEL_INFO(dev)->gen >= 6) {
865 ering->rc_psmi = I915_READ(ring->mmio_base + 0x50);
866 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
867 if (INTEL_INFO(dev)->gen >= 8)
868 gen8_record_semaphore_state(dev_priv, error, ring, ering);
869 else
870 gen6_record_semaphore_state(dev_priv, ring, ering);
871 }
872
873 if (INTEL_INFO(dev)->gen >= 4) {
874 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
875 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
876 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
877 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
878 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
879 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
880 if (INTEL_INFO(dev)->gen >= 8) {
881 ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(ring->mmio_base)) << 32;
882 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
883 }
884 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
885 } else {
886 ering->faddr = I915_READ(DMA_FADD_I8XX);
887 ering->ipeir = I915_READ(IPEIR);
888 ering->ipehr = I915_READ(IPEHR);
889 ering->instdone = I915_READ(GEN2_INSTDONE);
890 }
891
892 ering->waiting = waitqueue_active(&ring->irq_queue);
893 ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
894 ering->seqno = ring->get_seqno(ring, false);
895 ering->acthd = intel_ring_get_active_head(ring);
896 ering->start = I915_READ_START(ring);
897 ering->head = I915_READ_HEAD(ring);
898 ering->tail = I915_READ_TAIL(ring);
899 ering->ctl = I915_READ_CTL(ring);
900
901 if (I915_NEED_GFX_HWS(dev)) {
902 int mmio;
903
904 if (IS_GEN7(dev)) {
905 switch (ring->id) {
906 default:
907 case RCS:
908 mmio = RENDER_HWS_PGA_GEN7;
909 break;
910 case BCS:
911 mmio = BLT_HWS_PGA_GEN7;
912 break;
913 case VCS:
914 mmio = BSD_HWS_PGA_GEN7;
915 break;
916 case VECS:
917 mmio = VEBOX_HWS_PGA_GEN7;
918 break;
919 }
920 } else if (IS_GEN6(ring->dev)) {
921 mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
922 } else {
923 /* XXX: gen8 returns to sanity */
924 mmio = RING_HWS_PGA(ring->mmio_base);
925 }
926
927 ering->hws = I915_READ(mmio);
928 }
929
930 ering->hangcheck_score = ring->hangcheck.score;
931 ering->hangcheck_action = ring->hangcheck.action;
932
933 if (USES_PPGTT(dev)) {
934 int i;
935
936 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
937
938 if (IS_GEN6(dev))
939 ering->vm_info.pp_dir_base =
940 I915_READ(RING_PP_DIR_BASE_READ(ring));
941 else if (IS_GEN7(dev))
942 ering->vm_info.pp_dir_base =
943 I915_READ(RING_PP_DIR_BASE(ring));
944 else if (INTEL_INFO(dev)->gen >= 8)
945 for (i = 0; i < 4; i++) {
946 ering->vm_info.pdp[i] =
947 I915_READ(GEN8_RING_PDP_UDW(ring, i));
948 ering->vm_info.pdp[i] <<= 32;
949 ering->vm_info.pdp[i] |=
950 I915_READ(GEN8_RING_PDP_LDW(ring, i));
951 }
952 }
953 }
954
955
i915_gem_record_active_context(struct intel_engine_cs * ring,struct drm_i915_error_state * error,struct drm_i915_error_ring * ering)956 static void i915_gem_record_active_context(struct intel_engine_cs *ring,
957 struct drm_i915_error_state *error,
958 struct drm_i915_error_ring *ering)
959 {
960 struct drm_i915_private *dev_priv = ring->dev->dev_private;
961 struct drm_i915_gem_object *obj;
962
963 /* Currently render ring is the only HW context user */
964 if (ring->id != RCS || !error->ccid)
965 return;
966
967 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
968 if (!i915_gem_obj_ggtt_bound(obj))
969 continue;
970
971 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
972 ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
973 break;
974 }
975 }
976 }
977
i915_gem_record_rings(struct drm_device * dev,struct drm_i915_error_state * error)978 static void i915_gem_record_rings(struct drm_device *dev,
979 struct drm_i915_error_state *error)
980 {
981 struct drm_i915_private *dev_priv = dev->dev_private;
982 struct drm_i915_gem_request *request;
983 int i, count;
984
985 for (i = 0; i < I915_NUM_RINGS; i++) {
986 struct intel_engine_cs *ring = &dev_priv->ring[i];
987 struct intel_ringbuffer *rbuf;
988
989 error->ring[i].pid = -1;
990
991 if (ring->dev == NULL)
992 continue;
993
994 error->ring[i].valid = true;
995
996 i915_record_ring_state(dev, error, ring, &error->ring[i]);
997
998 request = i915_gem_find_active_request(ring);
999 if (request) {
1000 struct i915_address_space *vm;
1001
1002 vm = request->ctx && request->ctx->ppgtt ?
1003 &request->ctx->ppgtt->base :
1004 &dev_priv->gtt.base;
1005
1006 /* We need to copy these to an anonymous buffer
1007 * as the simplest method to avoid being overwritten
1008 * by userspace.
1009 */
1010 error->ring[i].batchbuffer =
1011 i915_error_object_create(dev_priv,
1012 request->batch_obj,
1013 vm);
1014
1015 if (HAS_BROKEN_CS_TLB(dev_priv->dev))
1016 error->ring[i].wa_batchbuffer =
1017 i915_error_ggtt_object_create(dev_priv,
1018 ring->scratch.obj);
1019
1020 if (request->pid) {
1021 struct task_struct *task;
1022
1023 rcu_read_lock();
1024 task = pid_task(request->pid, PIDTYPE_PID);
1025 if (task) {
1026 strcpy(error->ring[i].comm, task->comm);
1027 error->ring[i].pid = task->pid;
1028 }
1029 rcu_read_unlock();
1030 }
1031 }
1032
1033 if (i915.enable_execlists) {
1034 /* TODO: This is only a small fix to keep basic error
1035 * capture working, but we need to add more information
1036 * for it to be useful (e.g. dump the context being
1037 * executed).
1038 */
1039 if (request)
1040 rbuf = request->ctx->engine[ring->id].ringbuf;
1041 else
1042 rbuf = ring->default_context->engine[ring->id].ringbuf;
1043 } else
1044 rbuf = ring->buffer;
1045
1046 error->ring[i].cpu_ring_head = rbuf->head;
1047 error->ring[i].cpu_ring_tail = rbuf->tail;
1048
1049 error->ring[i].ringbuffer =
1050 i915_error_ggtt_object_create(dev_priv, rbuf->obj);
1051
1052 error->ring[i].hws_page =
1053 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
1054
1055 i915_gem_record_active_context(ring, error, &error->ring[i]);
1056
1057 count = 0;
1058 list_for_each_entry(request, &ring->request_list, list)
1059 count++;
1060
1061 error->ring[i].num_requests = count;
1062 error->ring[i].requests =
1063 kcalloc(count, sizeof(*error->ring[i].requests),
1064 GFP_ATOMIC);
1065 if (error->ring[i].requests == NULL) {
1066 error->ring[i].num_requests = 0;
1067 continue;
1068 }
1069
1070 count = 0;
1071 list_for_each_entry(request, &ring->request_list, list) {
1072 struct drm_i915_error_request *erq;
1073
1074 erq = &error->ring[i].requests[count++];
1075 erq->seqno = request->seqno;
1076 erq->jiffies = request->emitted_jiffies;
1077 erq->tail = request->postfix;
1078 }
1079 }
1080 }
1081
1082 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1083 * VM.
1084 */
i915_gem_capture_vm(struct drm_i915_private * dev_priv,struct drm_i915_error_state * error,struct i915_address_space * vm,const int ndx)1085 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1086 struct drm_i915_error_state *error,
1087 struct i915_address_space *vm,
1088 const int ndx)
1089 {
1090 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1091 struct drm_i915_gem_object *obj;
1092 struct i915_vma *vma;
1093 int i;
1094
1095 i = 0;
1096 list_for_each_entry(vma, &vm->active_list, mm_list)
1097 i++;
1098 error->active_bo_count[ndx] = i;
1099
1100 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1101 list_for_each_entry(vma, &obj->vma_list, vma_link)
1102 if (vma->vm == vm && vma->pin_count > 0)
1103 i++;
1104 }
1105 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1106
1107 if (i) {
1108 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1109 if (active_bo)
1110 pinned_bo = active_bo + error->active_bo_count[ndx];
1111 }
1112
1113 if (active_bo)
1114 error->active_bo_count[ndx] =
1115 capture_active_bo(active_bo,
1116 error->active_bo_count[ndx],
1117 &vm->active_list);
1118
1119 if (pinned_bo)
1120 error->pinned_bo_count[ndx] =
1121 capture_pinned_bo(pinned_bo,
1122 error->pinned_bo_count[ndx],
1123 &dev_priv->mm.bound_list, vm);
1124 error->active_bo[ndx] = active_bo;
1125 error->pinned_bo[ndx] = pinned_bo;
1126 }
1127
i915_gem_capture_buffers(struct drm_i915_private * dev_priv,struct drm_i915_error_state * error)1128 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1129 struct drm_i915_error_state *error)
1130 {
1131 struct i915_address_space *vm;
1132 int cnt = 0, i = 0;
1133
1134 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1135 cnt++;
1136
1137 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1138 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1139 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1140 GFP_ATOMIC);
1141 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1142 GFP_ATOMIC);
1143
1144 if (error->active_bo == NULL ||
1145 error->pinned_bo == NULL ||
1146 error->active_bo_count == NULL ||
1147 error->pinned_bo_count == NULL) {
1148 kfree(error->active_bo);
1149 kfree(error->active_bo_count);
1150 kfree(error->pinned_bo);
1151 kfree(error->pinned_bo_count);
1152
1153 error->active_bo = NULL;
1154 error->active_bo_count = NULL;
1155 error->pinned_bo = NULL;
1156 error->pinned_bo_count = NULL;
1157 } else {
1158 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1159 i915_gem_capture_vm(dev_priv, error, vm, i++);
1160
1161 error->vm_count = cnt;
1162 }
1163 }
1164
1165 /* Capture all registers which don't fit into another category. */
i915_capture_reg_state(struct drm_i915_private * dev_priv,struct drm_i915_error_state * error)1166 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1167 struct drm_i915_error_state *error)
1168 {
1169 struct drm_device *dev = dev_priv->dev;
1170 int i;
1171
1172 /* General organization
1173 * 1. Registers specific to a single generation
1174 * 2. Registers which belong to multiple generations
1175 * 3. Feature specific registers.
1176 * 4. Everything else
1177 * Please try to follow the order.
1178 */
1179
1180 /* 1: Registers specific to a single generation */
1181 if (IS_VALLEYVIEW(dev)) {
1182 error->gtier[0] = I915_READ(GTIER);
1183 error->ier = I915_READ(VLV_IER);
1184 error->forcewake = I915_READ(FORCEWAKE_VLV);
1185 }
1186
1187 if (IS_GEN7(dev))
1188 error->err_int = I915_READ(GEN7_ERR_INT);
1189
1190 if (INTEL_INFO(dev)->gen >= 8) {
1191 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1192 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1193 }
1194
1195 if (IS_GEN6(dev)) {
1196 error->forcewake = I915_READ(FORCEWAKE);
1197 error->gab_ctl = I915_READ(GAB_CTL);
1198 error->gfx_mode = I915_READ(GFX_MODE);
1199 }
1200
1201 /* 2: Registers which belong to multiple generations */
1202 if (INTEL_INFO(dev)->gen >= 7)
1203 error->forcewake = I915_READ(FORCEWAKE_MT);
1204
1205 if (INTEL_INFO(dev)->gen >= 6) {
1206 error->derrmr = I915_READ(DERRMR);
1207 error->error = I915_READ(ERROR_GEN6);
1208 error->done_reg = I915_READ(DONE_REG);
1209 }
1210
1211 /* 3: Feature specific registers */
1212 if (IS_GEN6(dev) || IS_GEN7(dev)) {
1213 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1214 error->gac_eco = I915_READ(GAC_ECO_BITS);
1215 }
1216
1217 /* 4: Everything else */
1218 if (HAS_HW_CONTEXTS(dev))
1219 error->ccid = I915_READ(CCID);
1220
1221 if (INTEL_INFO(dev)->gen >= 8) {
1222 error->ier = I915_READ(GEN8_DE_MISC_IER);
1223 for (i = 0; i < 4; i++)
1224 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1225 } else if (HAS_PCH_SPLIT(dev)) {
1226 error->ier = I915_READ(DEIER);
1227 error->gtier[0] = I915_READ(GTIER);
1228 } else if (IS_GEN2(dev)) {
1229 error->ier = I915_READ16(IER);
1230 } else if (!IS_VALLEYVIEW(dev)) {
1231 error->ier = I915_READ(IER);
1232 }
1233 error->eir = I915_READ(EIR);
1234 error->pgtbl_er = I915_READ(PGTBL_ER);
1235
1236 i915_get_extra_instdone(dev, error->extra_instdone);
1237 }
1238
i915_error_capture_msg(struct drm_device * dev,struct drm_i915_error_state * error,bool wedged,const char * error_msg)1239 static void i915_error_capture_msg(struct drm_device *dev,
1240 struct drm_i915_error_state *error,
1241 bool wedged,
1242 const char *error_msg)
1243 {
1244 struct drm_i915_private *dev_priv = dev->dev_private;
1245 u32 ecode;
1246 int ring_id = -1, len;
1247
1248 ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1249
1250 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1251 "GPU HANG: ecode %d:%d:0x%08x",
1252 INTEL_INFO(dev)->gen, ring_id, ecode);
1253
1254 if (ring_id != -1 && error->ring[ring_id].pid != -1)
1255 len += scnprintf(error->error_msg + len,
1256 sizeof(error->error_msg) - len,
1257 ", in %s [%d]",
1258 error->ring[ring_id].comm,
1259 error->ring[ring_id].pid);
1260
1261 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1262 ", reason: %s, action: %s",
1263 error_msg,
1264 wedged ? "reset" : "continue");
1265 }
1266
i915_capture_gen_state(struct drm_i915_private * dev_priv,struct drm_i915_error_state * error)1267 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1268 struct drm_i915_error_state *error)
1269 {
1270 error->iommu = -1;
1271 #ifdef CONFIG_INTEL_IOMMU
1272 error->iommu = intel_iommu_gfx_mapped;
1273 #endif
1274 error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1275 error->suspend_count = dev_priv->suspend_count;
1276 }
1277
1278 /**
1279 * i915_capture_error_state - capture an error record for later analysis
1280 * @dev: drm device
1281 *
1282 * Should be called when an error is detected (either a hang or an error
1283 * interrupt) to capture error state from the time of the error. Fills
1284 * out a structure which becomes available in debugfs for user level tools
1285 * to pick up.
1286 */
i915_capture_error_state(struct drm_device * dev,bool wedged,const char * error_msg)1287 void i915_capture_error_state(struct drm_device *dev, bool wedged,
1288 const char *error_msg)
1289 {
1290 static bool warned;
1291 struct drm_i915_private *dev_priv = dev->dev_private;
1292 struct drm_i915_error_state *error;
1293 unsigned long flags;
1294
1295 /* Account for pipe specific data like PIPE*STAT */
1296 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1297 if (!error) {
1298 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1299 return;
1300 }
1301
1302 kref_init(&error->ref);
1303
1304 i915_capture_gen_state(dev_priv, error);
1305 i915_capture_reg_state(dev_priv, error);
1306 i915_gem_capture_buffers(dev_priv, error);
1307 i915_gem_record_fences(dev, error);
1308 i915_gem_record_rings(dev, error);
1309
1310 do_gettimeofday(&error->time);
1311
1312 error->overlay = intel_overlay_capture_error_state(dev);
1313 error->display = intel_display_capture_error_state(dev);
1314
1315 i915_error_capture_msg(dev, error, wedged, error_msg);
1316 DRM_INFO("%s\n", error->error_msg);
1317
1318 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1319 if (dev_priv->gpu_error.first_error == NULL) {
1320 dev_priv->gpu_error.first_error = error;
1321 error = NULL;
1322 }
1323 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1324
1325 if (error) {
1326 i915_error_state_free(&error->ref);
1327 return;
1328 }
1329
1330 if (!warned) {
1331 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1332 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1333 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1334 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1335 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1336 warned = true;
1337 }
1338 }
1339
i915_error_state_get(struct drm_device * dev,struct i915_error_state_file_priv * error_priv)1340 void i915_error_state_get(struct drm_device *dev,
1341 struct i915_error_state_file_priv *error_priv)
1342 {
1343 struct drm_i915_private *dev_priv = dev->dev_private;
1344
1345 spin_lock_irq(&dev_priv->gpu_error.lock);
1346 error_priv->error = dev_priv->gpu_error.first_error;
1347 if (error_priv->error)
1348 kref_get(&error_priv->error->ref);
1349 spin_unlock_irq(&dev_priv->gpu_error.lock);
1350
1351 }
1352
i915_error_state_put(struct i915_error_state_file_priv * error_priv)1353 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1354 {
1355 if (error_priv->error)
1356 kref_put(&error_priv->error->ref, i915_error_state_free);
1357 }
1358
i915_destroy_error_state(struct drm_device * dev)1359 void i915_destroy_error_state(struct drm_device *dev)
1360 {
1361 struct drm_i915_private *dev_priv = dev->dev_private;
1362 struct drm_i915_error_state *error;
1363
1364 spin_lock_irq(&dev_priv->gpu_error.lock);
1365 error = dev_priv->gpu_error.first_error;
1366 dev_priv->gpu_error.first_error = NULL;
1367 spin_unlock_irq(&dev_priv->gpu_error.lock);
1368
1369 if (error)
1370 kref_put(&error->ref, i915_error_state_free);
1371 }
1372
i915_cache_level_str(struct drm_i915_private * i915,int type)1373 const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
1374 {
1375 switch (type) {
1376 case I915_CACHE_NONE: return " uncached";
1377 case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
1378 case I915_CACHE_L3_LLC: return " L3+LLC";
1379 case I915_CACHE_WT: return " WT";
1380 default: return "";
1381 }
1382 }
1383
1384 /* NB: please notice the memset */
i915_get_extra_instdone(struct drm_device * dev,uint32_t * instdone)1385 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1386 {
1387 struct drm_i915_private *dev_priv = dev->dev_private;
1388 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1389
1390 if (IS_GEN2(dev) || IS_GEN3(dev))
1391 instdone[0] = I915_READ(GEN2_INSTDONE);
1392 else if (IS_GEN4(dev) || IS_GEN5(dev) || IS_GEN6(dev)) {
1393 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1394 instdone[1] = I915_READ(GEN4_INSTDONE1);
1395 } else if (INTEL_INFO(dev)->gen >= 7) {
1396 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1397 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1398 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1399 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
1400 }
1401 }
1402