Loading...
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 <linux/stop_machine.h>
32#include <linux/zlib.h>
33#include "i915_drv.h"
34
35static const char *engine_str(int engine)
36{
37 switch (engine) {
38 case RCS: return "render";
39 case VCS: return "bsd";
40 case BCS: return "blt";
41 case VECS: return "vebox";
42 case VCS2: return "bsd2";
43 default: return "";
44 }
45}
46
47static const char *tiling_flag(int tiling)
48{
49 switch (tiling) {
50 default:
51 case I915_TILING_NONE: return "";
52 case I915_TILING_X: return " X";
53 case I915_TILING_Y: return " Y";
54 }
55}
56
57static const char *dirty_flag(int dirty)
58{
59 return dirty ? " dirty" : "";
60}
61
62static const char *purgeable_flag(int purgeable)
63{
64 return purgeable ? " purgeable" : "";
65}
66
67static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
68{
69
70 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
71 e->err = -ENOSPC;
72 return false;
73 }
74
75 if (e->bytes == e->size - 1 || e->err)
76 return false;
77
78 return true;
79}
80
81static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
82 unsigned len)
83{
84 if (e->pos + len <= e->start) {
85 e->pos += len;
86 return false;
87 }
88
89 /* First vsnprintf needs to fit in its entirety for memmove */
90 if (len >= e->size) {
91 e->err = -EIO;
92 return false;
93 }
94
95 return true;
96}
97
98static void __i915_error_advance(struct drm_i915_error_state_buf *e,
99 unsigned len)
100{
101 /* If this is first printf in this window, adjust it so that
102 * start position matches start of the buffer
103 */
104
105 if (e->pos < e->start) {
106 const size_t off = e->start - e->pos;
107
108 /* Should not happen but be paranoid */
109 if (off > len || e->bytes) {
110 e->err = -EIO;
111 return;
112 }
113
114 memmove(e->buf, e->buf + off, len - off);
115 e->bytes = len - off;
116 e->pos = e->start;
117 return;
118 }
119
120 e->bytes += len;
121 e->pos += len;
122}
123
124static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
125 const char *f, va_list args)
126{
127 unsigned len;
128
129 if (!__i915_error_ok(e))
130 return;
131
132 /* Seek the first printf which is hits start position */
133 if (e->pos < e->start) {
134 va_list tmp;
135
136 va_copy(tmp, args);
137 len = vsnprintf(NULL, 0, f, tmp);
138 va_end(tmp);
139
140 if (!__i915_error_seek(e, len))
141 return;
142 }
143
144 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
145 if (len >= e->size - e->bytes)
146 len = e->size - e->bytes - 1;
147
148 __i915_error_advance(e, len);
149}
150
151static void i915_error_puts(struct drm_i915_error_state_buf *e,
152 const char *str)
153{
154 unsigned len;
155
156 if (!__i915_error_ok(e))
157 return;
158
159 len = strlen(str);
160
161 /* Seek the first printf which is hits start position */
162 if (e->pos < e->start) {
163 if (!__i915_error_seek(e, len))
164 return;
165 }
166
167 if (len >= e->size - e->bytes)
168 len = e->size - e->bytes - 1;
169 memcpy(e->buf + e->bytes, str, len);
170
171 __i915_error_advance(e, len);
172}
173
174#define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
175#define err_puts(e, s) i915_error_puts(e, s)
176
177#ifdef CONFIG_DRM_I915_COMPRESS_ERROR
178
179static bool compress_init(struct z_stream_s *zstream)
180{
181 memset(zstream, 0, sizeof(*zstream));
182
183 zstream->workspace =
184 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
185 GFP_ATOMIC | __GFP_NOWARN);
186 if (!zstream->workspace)
187 return false;
188
189 if (zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) != Z_OK) {
190 kfree(zstream->workspace);
191 return false;
192 }
193
194 return true;
195}
196
197static int compress_page(struct z_stream_s *zstream,
198 void *src,
199 struct drm_i915_error_object *dst)
200{
201 zstream->next_in = src;
202 zstream->avail_in = PAGE_SIZE;
203
204 do {
205 if (zstream->avail_out == 0) {
206 unsigned long page;
207
208 page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
209 if (!page)
210 return -ENOMEM;
211
212 dst->pages[dst->page_count++] = (void *)page;
213
214 zstream->next_out = (void *)page;
215 zstream->avail_out = PAGE_SIZE;
216 }
217
218 if (zlib_deflate(zstream, Z_SYNC_FLUSH) != Z_OK)
219 return -EIO;
220 } while (zstream->avail_in);
221
222 /* Fallback to uncompressed if we increase size? */
223 if (0 && zstream->total_out > zstream->total_in)
224 return -E2BIG;
225
226 return 0;
227}
228
229static void compress_fini(struct z_stream_s *zstream,
230 struct drm_i915_error_object *dst)
231{
232 if (dst) {
233 zlib_deflate(zstream, Z_FINISH);
234 dst->unused = zstream->avail_out;
235 }
236
237 zlib_deflateEnd(zstream);
238 kfree(zstream->workspace);
239}
240
241static void err_compression_marker(struct drm_i915_error_state_buf *m)
242{
243 err_puts(m, ":");
244}
245
246#else
247
248static bool compress_init(struct z_stream_s *zstream)
249{
250 return true;
251}
252
253static int compress_page(struct z_stream_s *zstream,
254 void *src,
255 struct drm_i915_error_object *dst)
256{
257 unsigned long page;
258
259 page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
260 if (!page)
261 return -ENOMEM;
262
263 dst->pages[dst->page_count++] =
264 memcpy((void *)page, src, PAGE_SIZE);
265
266 return 0;
267}
268
269static void compress_fini(struct z_stream_s *zstream,
270 struct drm_i915_error_object *dst)
271{
272}
273
274static void err_compression_marker(struct drm_i915_error_state_buf *m)
275{
276 err_puts(m, "~");
277}
278
279#endif
280
281static void print_error_buffers(struct drm_i915_error_state_buf *m,
282 const char *name,
283 struct drm_i915_error_buffer *err,
284 int count)
285{
286 int i;
287
288 err_printf(m, "%s [%d]:\n", name, count);
289
290 while (count--) {
291 err_printf(m, " %08x_%08x %8u %02x %02x [ ",
292 upper_32_bits(err->gtt_offset),
293 lower_32_bits(err->gtt_offset),
294 err->size,
295 err->read_domains,
296 err->write_domain);
297 for (i = 0; i < I915_NUM_ENGINES; i++)
298 err_printf(m, "%02x ", err->rseqno[i]);
299
300 err_printf(m, "] %02x", err->wseqno);
301 err_puts(m, tiling_flag(err->tiling));
302 err_puts(m, dirty_flag(err->dirty));
303 err_puts(m, purgeable_flag(err->purgeable));
304 err_puts(m, err->userptr ? " userptr" : "");
305 err_puts(m, err->engine != -1 ? " " : "");
306 err_puts(m, engine_str(err->engine));
307 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
308
309 if (err->name)
310 err_printf(m, " (name: %d)", err->name);
311 if (err->fence_reg != I915_FENCE_REG_NONE)
312 err_printf(m, " (fence: %d)", err->fence_reg);
313
314 err_puts(m, "\n");
315 err++;
316 }
317}
318
319static const char *hangcheck_action_to_str(enum intel_engine_hangcheck_action a)
320{
321 switch (a) {
322 case HANGCHECK_IDLE:
323 return "idle";
324 case HANGCHECK_WAIT:
325 return "wait";
326 case HANGCHECK_ACTIVE:
327 return "active";
328 case HANGCHECK_KICK:
329 return "kick";
330 case HANGCHECK_HUNG:
331 return "hung";
332 }
333
334 return "unknown";
335}
336
337static void error_print_instdone(struct drm_i915_error_state_buf *m,
338 struct drm_i915_error_engine *ee)
339{
340 int slice;
341 int subslice;
342
343 err_printf(m, " INSTDONE: 0x%08x\n",
344 ee->instdone.instdone);
345
346 if (ee->engine_id != RCS || INTEL_GEN(m->i915) <= 3)
347 return;
348
349 err_printf(m, " SC_INSTDONE: 0x%08x\n",
350 ee->instdone.slice_common);
351
352 if (INTEL_GEN(m->i915) <= 6)
353 return;
354
355 for_each_instdone_slice_subslice(m->i915, slice, subslice)
356 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
357 slice, subslice,
358 ee->instdone.sampler[slice][subslice]);
359
360 for_each_instdone_slice_subslice(m->i915, slice, subslice)
361 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n",
362 slice, subslice,
363 ee->instdone.row[slice][subslice]);
364}
365
366static void error_print_request(struct drm_i915_error_state_buf *m,
367 const char *prefix,
368 struct drm_i915_error_request *erq)
369{
370 if (!erq->seqno)
371 return;
372
373 err_printf(m, "%s pid %d, seqno %8x:%08x, emitted %dms ago, head %08x, tail %08x\n",
374 prefix, erq->pid,
375 erq->context, erq->seqno,
376 jiffies_to_msecs(jiffies - erq->jiffies),
377 erq->head, erq->tail);
378}
379
380static void error_print_engine(struct drm_i915_error_state_buf *m,
381 struct drm_i915_error_engine *ee)
382{
383 err_printf(m, "%s command stream:\n", engine_str(ee->engine_id));
384 err_printf(m, " START: 0x%08x\n", ee->start);
385 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head);
386 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n",
387 ee->tail, ee->rq_post, ee->rq_tail);
388 err_printf(m, " CTL: 0x%08x\n", ee->ctl);
389 err_printf(m, " MODE: 0x%08x\n", ee->mode);
390 err_printf(m, " HWS: 0x%08x\n", ee->hws);
391 err_printf(m, " ACTHD: 0x%08x %08x\n",
392 (u32)(ee->acthd>>32), (u32)ee->acthd);
393 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir);
394 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr);
395
396 error_print_instdone(m, ee);
397
398 if (ee->batchbuffer) {
399 u64 start = ee->batchbuffer->gtt_offset;
400 u64 end = start + ee->batchbuffer->gtt_size;
401
402 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n",
403 upper_32_bits(start), lower_32_bits(start),
404 upper_32_bits(end), lower_32_bits(end));
405 }
406 if (INTEL_GEN(m->i915) >= 4) {
407 err_printf(m, " BBADDR: 0x%08x_%08x\n",
408 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
409 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate);
410 err_printf(m, " INSTPS: 0x%08x\n", ee->instps);
411 }
412 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm);
413 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
414 lower_32_bits(ee->faddr));
415 if (INTEL_GEN(m->i915) >= 6) {
416 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi);
417 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg);
418 err_printf(m, " SYNC_0: 0x%08x\n",
419 ee->semaphore_mboxes[0]);
420 err_printf(m, " SYNC_1: 0x%08x\n",
421 ee->semaphore_mboxes[1]);
422 if (HAS_VEBOX(m->i915))
423 err_printf(m, " SYNC_2: 0x%08x\n",
424 ee->semaphore_mboxes[2]);
425 }
426 if (USES_PPGTT(m->i915)) {
427 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
428
429 if (INTEL_GEN(m->i915) >= 8) {
430 int i;
431 for (i = 0; i < 4; i++)
432 err_printf(m, " PDP%d: 0x%016llx\n",
433 i, ee->vm_info.pdp[i]);
434 } else {
435 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
436 ee->vm_info.pp_dir_base);
437 }
438 }
439 err_printf(m, " seqno: 0x%08x\n", ee->seqno);
440 err_printf(m, " last_seqno: 0x%08x\n", ee->last_seqno);
441 err_printf(m, " waiting: %s\n", yesno(ee->waiting));
442 err_printf(m, " ring->head: 0x%08x\n", ee->cpu_ring_head);
443 err_printf(m, " ring->tail: 0x%08x\n", ee->cpu_ring_tail);
444 err_printf(m, " hangcheck: %s [%d]\n",
445 hangcheck_action_to_str(ee->hangcheck_action),
446 ee->hangcheck_score);
447 error_print_request(m, " ELSP[0]: ", &ee->execlist[0]);
448 error_print_request(m, " ELSP[1]: ", &ee->execlist[1]);
449}
450
451void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
452{
453 va_list args;
454
455 va_start(args, f);
456 i915_error_vprintf(e, f, args);
457 va_end(args);
458}
459
460static int
461ascii85_encode_len(int len)
462{
463 return DIV_ROUND_UP(len, 4);
464}
465
466static bool
467ascii85_encode(u32 in, char *out)
468{
469 int i;
470
471 if (in == 0)
472 return false;
473
474 out[5] = '\0';
475 for (i = 5; i--; ) {
476 out[i] = '!' + in % 85;
477 in /= 85;
478 }
479
480 return true;
481}
482
483static void print_error_obj(struct drm_i915_error_state_buf *m,
484 struct intel_engine_cs *engine,
485 const char *name,
486 struct drm_i915_error_object *obj)
487{
488 char out[6];
489 int page;
490
491 if (!obj)
492 return;
493
494 if (name) {
495 err_printf(m, "%s --- %s = 0x%08x %08x\n",
496 engine ? engine->name : "global", name,
497 upper_32_bits(obj->gtt_offset),
498 lower_32_bits(obj->gtt_offset));
499 }
500
501 err_compression_marker(m);
502 for (page = 0; page < obj->page_count; page++) {
503 int i, len;
504
505 len = PAGE_SIZE;
506 if (page == obj->page_count - 1)
507 len -= obj->unused;
508 len = ascii85_encode_len(len);
509
510 for (i = 0; i < len; i++) {
511 if (ascii85_encode(obj->pages[page][i], out))
512 err_puts(m, out);
513 else
514 err_puts(m, "z");
515 }
516 }
517 err_puts(m, "\n");
518}
519
520static void err_print_capabilities(struct drm_i915_error_state_buf *m,
521 const struct intel_device_info *info)
522{
523#define PRINT_FLAG(x) err_printf(m, #x ": %s\n", yesno(info->x))
524 DEV_INFO_FOR_EACH_FLAG(PRINT_FLAG);
525#undef PRINT_FLAG
526}
527
528int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
529 const struct i915_error_state_file_priv *error_priv)
530{
531 struct drm_i915_private *dev_priv = to_i915(error_priv->dev);
532 struct pci_dev *pdev = dev_priv->drm.pdev;
533 struct drm_i915_error_state *error = error_priv->error;
534 struct drm_i915_error_object *obj;
535 int max_hangcheck_score;
536 int i, j;
537
538 if (!error) {
539 err_printf(m, "no error state collected\n");
540 goto out;
541 }
542
543 err_printf(m, "%s\n", error->error_msg);
544 err_printf(m, "Kernel: " UTS_RELEASE "\n");
545 err_printf(m, "Time: %ld s %ld us\n",
546 error->time.tv_sec, error->time.tv_usec);
547 err_printf(m, "Boottime: %ld s %ld us\n",
548 error->boottime.tv_sec, error->boottime.tv_usec);
549 err_printf(m, "Uptime: %ld s %ld us\n",
550 error->uptime.tv_sec, error->uptime.tv_usec);
551 err_print_capabilities(m, &error->device_info);
552 max_hangcheck_score = 0;
553 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
554 if (error->engine[i].hangcheck_score > max_hangcheck_score)
555 max_hangcheck_score = error->engine[i].hangcheck_score;
556 }
557 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
558 if (error->engine[i].hangcheck_score == max_hangcheck_score &&
559 error->engine[i].pid != -1) {
560 err_printf(m, "Active process (on ring %s): %s [%d]\n",
561 engine_str(i),
562 error->engine[i].comm,
563 error->engine[i].pid);
564 }
565 }
566 err_printf(m, "Reset count: %u\n", error->reset_count);
567 err_printf(m, "Suspend count: %u\n", error->suspend_count);
568 err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
569 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
570 err_printf(m, "PCI Subsystem: %04x:%04x\n",
571 pdev->subsystem_vendor,
572 pdev->subsystem_device);
573 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
574
575 if (HAS_CSR(dev_priv)) {
576 struct intel_csr *csr = &dev_priv->csr;
577
578 err_printf(m, "DMC loaded: %s\n",
579 yesno(csr->dmc_payload != NULL));
580 err_printf(m, "DMC fw version: %d.%d\n",
581 CSR_VERSION_MAJOR(csr->version),
582 CSR_VERSION_MINOR(csr->version));
583 }
584
585 err_printf(m, "EIR: 0x%08x\n", error->eir);
586 err_printf(m, "IER: 0x%08x\n", error->ier);
587 if (INTEL_GEN(dev_priv) >= 8) {
588 for (i = 0; i < 4; i++)
589 err_printf(m, "GTIER gt %d: 0x%08x\n", i,
590 error->gtier[i]);
591 } else if (HAS_PCH_SPLIT(dev_priv) || IS_VALLEYVIEW(dev_priv))
592 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
593 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
594 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
595 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
596 err_printf(m, "CCID: 0x%08x\n", error->ccid);
597 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
598
599 for (i = 0; i < dev_priv->num_fence_regs; i++)
600 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
601
602 if (INTEL_GEN(dev_priv) >= 6) {
603 err_printf(m, "ERROR: 0x%08x\n", error->error);
604
605 if (INTEL_GEN(dev_priv) >= 8)
606 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
607 error->fault_data1, error->fault_data0);
608
609 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
610 }
611
612 if (IS_GEN7(dev_priv))
613 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
614
615 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
616 if (error->engine[i].engine_id != -1)
617 error_print_engine(m, &error->engine[i]);
618 }
619
620 for (i = 0; i < ARRAY_SIZE(error->active_vm); i++) {
621 char buf[128];
622 int len, first = 1;
623
624 if (!error->active_vm[i])
625 break;
626
627 len = scnprintf(buf, sizeof(buf), "Active (");
628 for (j = 0; j < ARRAY_SIZE(error->engine); j++) {
629 if (error->engine[j].vm != error->active_vm[i])
630 continue;
631
632 len += scnprintf(buf + len, sizeof(buf), "%s%s",
633 first ? "" : ", ",
634 dev_priv->engine[j]->name);
635 first = 0;
636 }
637 scnprintf(buf + len, sizeof(buf), ")");
638 print_error_buffers(m, buf,
639 error->active_bo[i],
640 error->active_bo_count[i]);
641 }
642
643 print_error_buffers(m, "Pinned (global)",
644 error->pinned_bo,
645 error->pinned_bo_count);
646
647 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
648 struct drm_i915_error_engine *ee = &error->engine[i];
649
650 obj = ee->batchbuffer;
651 if (obj) {
652 err_puts(m, dev_priv->engine[i]->name);
653 if (ee->pid != -1)
654 err_printf(m, " (submitted by %s [%d])",
655 ee->comm,
656 ee->pid);
657 err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
658 upper_32_bits(obj->gtt_offset),
659 lower_32_bits(obj->gtt_offset));
660 print_error_obj(m, dev_priv->engine[i], NULL, obj);
661 }
662
663 if (ee->num_requests) {
664 err_printf(m, "%s --- %d requests\n",
665 dev_priv->engine[i]->name,
666 ee->num_requests);
667 for (j = 0; j < ee->num_requests; j++)
668 error_print_request(m, " ", &ee->requests[j]);
669 }
670
671 if (IS_ERR(ee->waiters)) {
672 err_printf(m, "%s --- ? waiters [unable to acquire spinlock]\n",
673 dev_priv->engine[i]->name);
674 } else if (ee->num_waiters) {
675 err_printf(m, "%s --- %d waiters\n",
676 dev_priv->engine[i]->name,
677 ee->num_waiters);
678 for (j = 0; j < ee->num_waiters; j++) {
679 err_printf(m, " seqno 0x%08x for %s [%d]\n",
680 ee->waiters[j].seqno,
681 ee->waiters[j].comm,
682 ee->waiters[j].pid);
683 }
684 }
685
686 print_error_obj(m, dev_priv->engine[i],
687 "ringbuffer", ee->ringbuffer);
688
689 print_error_obj(m, dev_priv->engine[i],
690 "HW Status", ee->hws_page);
691
692 print_error_obj(m, dev_priv->engine[i],
693 "HW context", ee->ctx);
694
695 print_error_obj(m, dev_priv->engine[i],
696 "WA context", ee->wa_ctx);
697
698 print_error_obj(m, dev_priv->engine[i],
699 "WA batchbuffer", ee->wa_batchbuffer);
700 }
701
702 print_error_obj(m, NULL, "Semaphores", error->semaphore);
703
704 print_error_obj(m, NULL, "GuC log buffer", error->guc_log);
705
706 if (error->overlay)
707 intel_overlay_print_error_state(m, error->overlay);
708
709 if (error->display)
710 intel_display_print_error_state(m, dev_priv, error->display);
711
712out:
713 if (m->bytes == 0 && m->err)
714 return m->err;
715
716 return 0;
717}
718
719int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
720 struct drm_i915_private *i915,
721 size_t count, loff_t pos)
722{
723 memset(ebuf, 0, sizeof(*ebuf));
724 ebuf->i915 = i915;
725
726 /* We need to have enough room to store any i915_error_state printf
727 * so that we can move it to start position.
728 */
729 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
730 ebuf->buf = kmalloc(ebuf->size,
731 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
732
733 if (ebuf->buf == NULL) {
734 ebuf->size = PAGE_SIZE;
735 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
736 }
737
738 if (ebuf->buf == NULL) {
739 ebuf->size = 128;
740 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
741 }
742
743 if (ebuf->buf == NULL)
744 return -ENOMEM;
745
746 ebuf->start = pos;
747
748 return 0;
749}
750
751static void i915_error_object_free(struct drm_i915_error_object *obj)
752{
753 int page;
754
755 if (obj == NULL)
756 return;
757
758 for (page = 0; page < obj->page_count; page++)
759 free_page((unsigned long)obj->pages[page]);
760
761 kfree(obj);
762}
763
764static void i915_error_state_free(struct kref *error_ref)
765{
766 struct drm_i915_error_state *error = container_of(error_ref,
767 typeof(*error), ref);
768 int i;
769
770 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
771 struct drm_i915_error_engine *ee = &error->engine[i];
772
773 i915_error_object_free(ee->batchbuffer);
774 i915_error_object_free(ee->wa_batchbuffer);
775 i915_error_object_free(ee->ringbuffer);
776 i915_error_object_free(ee->hws_page);
777 i915_error_object_free(ee->ctx);
778 i915_error_object_free(ee->wa_ctx);
779
780 kfree(ee->requests);
781 if (!IS_ERR_OR_NULL(ee->waiters))
782 kfree(ee->waiters);
783 }
784
785 i915_error_object_free(error->semaphore);
786 i915_error_object_free(error->guc_log);
787
788 for (i = 0; i < ARRAY_SIZE(error->active_bo); i++)
789 kfree(error->active_bo[i]);
790 kfree(error->pinned_bo);
791
792 kfree(error->overlay);
793 kfree(error->display);
794 kfree(error);
795}
796
797static struct drm_i915_error_object *
798i915_error_object_create(struct drm_i915_private *i915,
799 struct i915_vma *vma)
800{
801 struct i915_ggtt *ggtt = &i915->ggtt;
802 const u64 slot = ggtt->error_capture.start;
803 struct drm_i915_error_object *dst;
804 struct z_stream_s zstream;
805 unsigned long num_pages;
806 struct sgt_iter iter;
807 dma_addr_t dma;
808
809 if (!vma)
810 return NULL;
811
812 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
813 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
814 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *),
815 GFP_ATOMIC | __GFP_NOWARN);
816 if (!dst)
817 return NULL;
818
819 dst->gtt_offset = vma->node.start;
820 dst->gtt_size = vma->node.size;
821 dst->page_count = 0;
822 dst->unused = 0;
823
824 if (!compress_init(&zstream)) {
825 kfree(dst);
826 return NULL;
827 }
828
829 for_each_sgt_dma(dma, iter, vma->pages) {
830 void __iomem *s;
831 int ret;
832
833 ggtt->base.insert_page(&ggtt->base, dma, slot,
834 I915_CACHE_NONE, 0);
835
836 s = io_mapping_map_atomic_wc(&ggtt->mappable, slot);
837 ret = compress_page(&zstream, (void __force *)s, dst);
838 io_mapping_unmap_atomic(s);
839
840 if (ret)
841 goto unwind;
842 }
843 goto out;
844
845unwind:
846 while (dst->page_count--)
847 free_page((unsigned long)dst->pages[dst->page_count]);
848 kfree(dst);
849 dst = NULL;
850
851out:
852 compress_fini(&zstream, dst);
853 ggtt->base.clear_range(&ggtt->base, slot, PAGE_SIZE);
854 return dst;
855}
856
857/* The error capture is special as tries to run underneath the normal
858 * locking rules - so we use the raw version of the i915_gem_active lookup.
859 */
860static inline uint32_t
861__active_get_seqno(struct i915_gem_active *active)
862{
863 struct drm_i915_gem_request *request;
864
865 request = __i915_gem_active_peek(active);
866 return request ? request->global_seqno : 0;
867}
868
869static inline int
870__active_get_engine_id(struct i915_gem_active *active)
871{
872 struct drm_i915_gem_request *request;
873
874 request = __i915_gem_active_peek(active);
875 return request ? request->engine->id : -1;
876}
877
878static void capture_bo(struct drm_i915_error_buffer *err,
879 struct i915_vma *vma)
880{
881 struct drm_i915_gem_object *obj = vma->obj;
882 int i;
883
884 err->size = obj->base.size;
885 err->name = obj->base.name;
886
887 for (i = 0; i < I915_NUM_ENGINES; i++)
888 err->rseqno[i] = __active_get_seqno(&vma->last_read[i]);
889 err->wseqno = __active_get_seqno(&obj->frontbuffer_write);
890 err->engine = __active_get_engine_id(&obj->frontbuffer_write);
891
892 err->gtt_offset = vma->node.start;
893 err->read_domains = obj->base.read_domains;
894 err->write_domain = obj->base.write_domain;
895 err->fence_reg = vma->fence ? vma->fence->id : -1;
896 err->tiling = i915_gem_object_get_tiling(obj);
897 err->dirty = obj->mm.dirty;
898 err->purgeable = obj->mm.madv != I915_MADV_WILLNEED;
899 err->userptr = obj->userptr.mm != NULL;
900 err->cache_level = obj->cache_level;
901}
902
903static u32 capture_error_bo(struct drm_i915_error_buffer *err,
904 int count, struct list_head *head,
905 bool pinned_only)
906{
907 struct i915_vma *vma;
908 int i = 0;
909
910 list_for_each_entry(vma, head, vm_link) {
911 if (pinned_only && !i915_vma_is_pinned(vma))
912 continue;
913
914 capture_bo(err++, vma);
915 if (++i == count)
916 break;
917 }
918
919 return i;
920}
921
922/* Generate a semi-unique error code. The code is not meant to have meaning, The
923 * code's only purpose is to try to prevent false duplicated bug reports by
924 * grossly estimating a GPU error state.
925 *
926 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
927 * the hang if we could strip the GTT offset information from it.
928 *
929 * It's only a small step better than a random number in its current form.
930 */
931static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
932 struct drm_i915_error_state *error,
933 int *engine_id)
934{
935 uint32_t error_code = 0;
936 int i;
937
938 /* IPEHR would be an ideal way to detect errors, as it's the gross
939 * measure of "the command that hung." However, has some very common
940 * synchronization commands which almost always appear in the case
941 * strictly a client bug. Use instdone to differentiate those some.
942 */
943 for (i = 0; i < I915_NUM_ENGINES; i++) {
944 if (error->engine[i].hangcheck_action == HANGCHECK_HUNG) {
945 if (engine_id)
946 *engine_id = i;
947
948 return error->engine[i].ipehr ^
949 error->engine[i].instdone.instdone;
950 }
951 }
952
953 return error_code;
954}
955
956static void i915_gem_record_fences(struct drm_i915_private *dev_priv,
957 struct drm_i915_error_state *error)
958{
959 int i;
960
961 if (IS_GEN3(dev_priv) || IS_GEN2(dev_priv)) {
962 for (i = 0; i < dev_priv->num_fence_regs; i++)
963 error->fence[i] = I915_READ(FENCE_REG(i));
964 } else if (IS_GEN5(dev_priv) || IS_GEN4(dev_priv)) {
965 for (i = 0; i < dev_priv->num_fence_regs; i++)
966 error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
967 } else if (INTEL_GEN(dev_priv) >= 6) {
968 for (i = 0; i < dev_priv->num_fence_regs; i++)
969 error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
970 }
971}
972
973static inline u32
974gen8_engine_sync_index(struct intel_engine_cs *engine,
975 struct intel_engine_cs *other)
976{
977 int idx;
978
979 /*
980 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
981 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
982 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
983 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
984 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
985 */
986
987 idx = (other - engine) - 1;
988 if (idx < 0)
989 idx += I915_NUM_ENGINES;
990
991 return idx;
992}
993
994static void gen8_record_semaphore_state(struct drm_i915_error_state *error,
995 struct intel_engine_cs *engine,
996 struct drm_i915_error_engine *ee)
997{
998 struct drm_i915_private *dev_priv = engine->i915;
999 struct intel_engine_cs *to;
1000 enum intel_engine_id id;
1001
1002 if (!error->semaphore)
1003 return;
1004
1005 for_each_engine(to, dev_priv, id) {
1006 int idx;
1007 u16 signal_offset;
1008 u32 *tmp;
1009
1010 if (engine == to)
1011 continue;
1012
1013 signal_offset =
1014 (GEN8_SIGNAL_OFFSET(engine, id) & (PAGE_SIZE - 1)) / 4;
1015 tmp = error->semaphore->pages[0];
1016 idx = gen8_engine_sync_index(engine, to);
1017
1018 ee->semaphore_mboxes[idx] = tmp[signal_offset];
1019 }
1020}
1021
1022static void gen6_record_semaphore_state(struct intel_engine_cs *engine,
1023 struct drm_i915_error_engine *ee)
1024{
1025 struct drm_i915_private *dev_priv = engine->i915;
1026
1027 ee->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base));
1028 ee->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base));
1029 if (HAS_VEBOX(dev_priv))
1030 ee->semaphore_mboxes[2] =
1031 I915_READ(RING_SYNC_2(engine->mmio_base));
1032}
1033
1034static void error_record_engine_waiters(struct intel_engine_cs *engine,
1035 struct drm_i915_error_engine *ee)
1036{
1037 struct intel_breadcrumbs *b = &engine->breadcrumbs;
1038 struct drm_i915_error_waiter *waiter;
1039 struct rb_node *rb;
1040 int count;
1041
1042 ee->num_waiters = 0;
1043 ee->waiters = NULL;
1044
1045 if (RB_EMPTY_ROOT(&b->waiters))
1046 return;
1047
1048 if (!spin_trylock_irq(&b->lock)) {
1049 ee->waiters = ERR_PTR(-EDEADLK);
1050 return;
1051 }
1052
1053 count = 0;
1054 for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb))
1055 count++;
1056 spin_unlock_irq(&b->lock);
1057
1058 waiter = NULL;
1059 if (count)
1060 waiter = kmalloc_array(count,
1061 sizeof(struct drm_i915_error_waiter),
1062 GFP_ATOMIC);
1063 if (!waiter)
1064 return;
1065
1066 if (!spin_trylock_irq(&b->lock)) {
1067 kfree(waiter);
1068 ee->waiters = ERR_PTR(-EDEADLK);
1069 return;
1070 }
1071
1072 ee->waiters = waiter;
1073 for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
1074 struct intel_wait *w = container_of(rb, typeof(*w), node);
1075
1076 strcpy(waiter->comm, w->tsk->comm);
1077 waiter->pid = w->tsk->pid;
1078 waiter->seqno = w->seqno;
1079 waiter++;
1080
1081 if (++ee->num_waiters == count)
1082 break;
1083 }
1084 spin_unlock_irq(&b->lock);
1085}
1086
1087static void error_record_engine_registers(struct drm_i915_error_state *error,
1088 struct intel_engine_cs *engine,
1089 struct drm_i915_error_engine *ee)
1090{
1091 struct drm_i915_private *dev_priv = engine->i915;
1092
1093 if (INTEL_GEN(dev_priv) >= 6) {
1094 ee->rc_psmi = I915_READ(RING_PSMI_CTL(engine->mmio_base));
1095 ee->fault_reg = I915_READ(RING_FAULT_REG(engine));
1096 if (INTEL_GEN(dev_priv) >= 8)
1097 gen8_record_semaphore_state(error, engine, ee);
1098 else
1099 gen6_record_semaphore_state(engine, ee);
1100 }
1101
1102 if (INTEL_GEN(dev_priv) >= 4) {
1103 ee->faddr = I915_READ(RING_DMA_FADD(engine->mmio_base));
1104 ee->ipeir = I915_READ(RING_IPEIR(engine->mmio_base));
1105 ee->ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
1106 ee->instps = I915_READ(RING_INSTPS(engine->mmio_base));
1107 ee->bbaddr = I915_READ(RING_BBADDR(engine->mmio_base));
1108 if (INTEL_GEN(dev_priv) >= 8) {
1109 ee->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(engine->mmio_base)) << 32;
1110 ee->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(engine->mmio_base)) << 32;
1111 }
1112 ee->bbstate = I915_READ(RING_BBSTATE(engine->mmio_base));
1113 } else {
1114 ee->faddr = I915_READ(DMA_FADD_I8XX);
1115 ee->ipeir = I915_READ(IPEIR);
1116 ee->ipehr = I915_READ(IPEHR);
1117 }
1118
1119 intel_engine_get_instdone(engine, &ee->instdone);
1120
1121 ee->waiting = intel_engine_has_waiter(engine);
1122 ee->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
1123 ee->acthd = intel_engine_get_active_head(engine);
1124 ee->seqno = intel_engine_get_seqno(engine);
1125 ee->last_seqno = intel_engine_last_submit(engine);
1126 ee->start = I915_READ_START(engine);
1127 ee->head = I915_READ_HEAD(engine);
1128 ee->tail = I915_READ_TAIL(engine);
1129 ee->ctl = I915_READ_CTL(engine);
1130 if (INTEL_GEN(dev_priv) > 2)
1131 ee->mode = I915_READ_MODE(engine);
1132
1133 if (!HWS_NEEDS_PHYSICAL(dev_priv)) {
1134 i915_reg_t mmio;
1135
1136 if (IS_GEN7(dev_priv)) {
1137 switch (engine->id) {
1138 default:
1139 case RCS:
1140 mmio = RENDER_HWS_PGA_GEN7;
1141 break;
1142 case BCS:
1143 mmio = BLT_HWS_PGA_GEN7;
1144 break;
1145 case VCS:
1146 mmio = BSD_HWS_PGA_GEN7;
1147 break;
1148 case VECS:
1149 mmio = VEBOX_HWS_PGA_GEN7;
1150 break;
1151 }
1152 } else if (IS_GEN6(engine->i915)) {
1153 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1154 } else {
1155 /* XXX: gen8 returns to sanity */
1156 mmio = RING_HWS_PGA(engine->mmio_base);
1157 }
1158
1159 ee->hws = I915_READ(mmio);
1160 }
1161
1162 ee->hangcheck_score = engine->hangcheck.score;
1163 ee->hangcheck_action = engine->hangcheck.action;
1164
1165 if (USES_PPGTT(dev_priv)) {
1166 int i;
1167
1168 ee->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(engine));
1169
1170 if (IS_GEN6(dev_priv))
1171 ee->vm_info.pp_dir_base =
1172 I915_READ(RING_PP_DIR_BASE_READ(engine));
1173 else if (IS_GEN7(dev_priv))
1174 ee->vm_info.pp_dir_base =
1175 I915_READ(RING_PP_DIR_BASE(engine));
1176 else if (INTEL_GEN(dev_priv) >= 8)
1177 for (i = 0; i < 4; i++) {
1178 ee->vm_info.pdp[i] =
1179 I915_READ(GEN8_RING_PDP_UDW(engine, i));
1180 ee->vm_info.pdp[i] <<= 32;
1181 ee->vm_info.pdp[i] |=
1182 I915_READ(GEN8_RING_PDP_LDW(engine, i));
1183 }
1184 }
1185}
1186
1187static void record_request(struct drm_i915_gem_request *request,
1188 struct drm_i915_error_request *erq)
1189{
1190 erq->context = request->ctx->hw_id;
1191 erq->seqno = request->global_seqno;
1192 erq->jiffies = request->emitted_jiffies;
1193 erq->head = request->head;
1194 erq->tail = request->tail;
1195
1196 rcu_read_lock();
1197 erq->pid = request->ctx->pid ? pid_nr(request->ctx->pid) : 0;
1198 rcu_read_unlock();
1199}
1200
1201static void engine_record_requests(struct intel_engine_cs *engine,
1202 struct drm_i915_gem_request *first,
1203 struct drm_i915_error_engine *ee)
1204{
1205 struct drm_i915_gem_request *request;
1206 int count;
1207
1208 count = 0;
1209 request = first;
1210 list_for_each_entry_from(request, &engine->timeline->requests, link)
1211 count++;
1212 if (!count)
1213 return;
1214
1215 ee->requests = kcalloc(count, sizeof(*ee->requests), GFP_ATOMIC);
1216 if (!ee->requests)
1217 return;
1218
1219 ee->num_requests = count;
1220
1221 count = 0;
1222 request = first;
1223 list_for_each_entry_from(request, &engine->timeline->requests, link) {
1224 if (count >= ee->num_requests) {
1225 /*
1226 * If the ring request list was changed in
1227 * between the point where the error request
1228 * list was created and dimensioned and this
1229 * point then just exit early to avoid crashes.
1230 *
1231 * We don't need to communicate that the
1232 * request list changed state during error
1233 * state capture and that the error state is
1234 * slightly incorrect as a consequence since we
1235 * are typically only interested in the request
1236 * list state at the point of error state
1237 * capture, not in any changes happening during
1238 * the capture.
1239 */
1240 break;
1241 }
1242
1243 record_request(request, &ee->requests[count++]);
1244 }
1245 ee->num_requests = count;
1246}
1247
1248static void error_record_engine_execlists(struct intel_engine_cs *engine,
1249 struct drm_i915_error_engine *ee)
1250{
1251 unsigned int n;
1252
1253 for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++)
1254 if (engine->execlist_port[n].request)
1255 record_request(engine->execlist_port[n].request,
1256 &ee->execlist[n]);
1257}
1258
1259static void i915_gem_record_rings(struct drm_i915_private *dev_priv,
1260 struct drm_i915_error_state *error)
1261{
1262 struct i915_ggtt *ggtt = &dev_priv->ggtt;
1263 int i;
1264
1265 error->semaphore =
1266 i915_error_object_create(dev_priv, dev_priv->semaphore);
1267
1268 for (i = 0; i < I915_NUM_ENGINES; i++) {
1269 struct intel_engine_cs *engine = dev_priv->engine[i];
1270 struct drm_i915_error_engine *ee = &error->engine[i];
1271 struct drm_i915_gem_request *request;
1272
1273 ee->pid = -1;
1274 ee->engine_id = -1;
1275
1276 if (!engine)
1277 continue;
1278
1279 ee->engine_id = i;
1280
1281 error_record_engine_registers(error, engine, ee);
1282 error_record_engine_waiters(engine, ee);
1283 error_record_engine_execlists(engine, ee);
1284
1285 request = i915_gem_find_active_request(engine);
1286 if (request) {
1287 struct intel_ring *ring;
1288 struct pid *pid;
1289
1290 ee->vm = request->ctx->ppgtt ?
1291 &request->ctx->ppgtt->base : &ggtt->base;
1292
1293 /* We need to copy these to an anonymous buffer
1294 * as the simplest method to avoid being overwritten
1295 * by userspace.
1296 */
1297 ee->batchbuffer =
1298 i915_error_object_create(dev_priv,
1299 request->batch);
1300
1301 if (HAS_BROKEN_CS_TLB(dev_priv))
1302 ee->wa_batchbuffer =
1303 i915_error_object_create(dev_priv,
1304 engine->scratch);
1305
1306 ee->ctx =
1307 i915_error_object_create(dev_priv,
1308 request->ctx->engine[i].state);
1309
1310 pid = request->ctx->pid;
1311 if (pid) {
1312 struct task_struct *task;
1313
1314 rcu_read_lock();
1315 task = pid_task(pid, PIDTYPE_PID);
1316 if (task) {
1317 strcpy(ee->comm, task->comm);
1318 ee->pid = task->pid;
1319 }
1320 rcu_read_unlock();
1321 }
1322
1323 error->simulated |=
1324 request->ctx->flags & CONTEXT_NO_ERROR_CAPTURE;
1325
1326 ee->rq_head = request->head;
1327 ee->rq_post = request->postfix;
1328 ee->rq_tail = request->tail;
1329
1330 ring = request->ring;
1331 ee->cpu_ring_head = ring->head;
1332 ee->cpu_ring_tail = ring->tail;
1333 ee->ringbuffer =
1334 i915_error_object_create(dev_priv, ring->vma);
1335
1336 engine_record_requests(engine, request, ee);
1337 }
1338
1339 ee->hws_page =
1340 i915_error_object_create(dev_priv,
1341 engine->status_page.vma);
1342
1343 ee->wa_ctx =
1344 i915_error_object_create(dev_priv, engine->wa_ctx.vma);
1345 }
1346}
1347
1348static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1349 struct drm_i915_error_state *error,
1350 struct i915_address_space *vm,
1351 int idx)
1352{
1353 struct drm_i915_error_buffer *active_bo;
1354 struct i915_vma *vma;
1355 int count;
1356
1357 count = 0;
1358 list_for_each_entry(vma, &vm->active_list, vm_link)
1359 count++;
1360
1361 active_bo = NULL;
1362 if (count)
1363 active_bo = kcalloc(count, sizeof(*active_bo), GFP_ATOMIC);
1364 if (active_bo)
1365 count = capture_error_bo(active_bo, count, &vm->active_list, false);
1366 else
1367 count = 0;
1368
1369 error->active_vm[idx] = vm;
1370 error->active_bo[idx] = active_bo;
1371 error->active_bo_count[idx] = count;
1372}
1373
1374static void i915_capture_active_buffers(struct drm_i915_private *dev_priv,
1375 struct drm_i915_error_state *error)
1376{
1377 int cnt = 0, i, j;
1378
1379 BUILD_BUG_ON(ARRAY_SIZE(error->engine) > ARRAY_SIZE(error->active_bo));
1380 BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_vm));
1381 BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_bo_count));
1382
1383 /* Scan each engine looking for unique active contexts/vm */
1384 for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
1385 struct drm_i915_error_engine *ee = &error->engine[i];
1386 bool found;
1387
1388 if (!ee->vm)
1389 continue;
1390
1391 found = false;
1392 for (j = 0; j < i && !found; j++)
1393 found = error->engine[j].vm == ee->vm;
1394 if (!found)
1395 i915_gem_capture_vm(dev_priv, error, ee->vm, cnt++);
1396 }
1397}
1398
1399static void i915_capture_pinned_buffers(struct drm_i915_private *dev_priv,
1400 struct drm_i915_error_state *error)
1401{
1402 struct i915_address_space *vm = &dev_priv->ggtt.base;
1403 struct drm_i915_error_buffer *bo;
1404 struct i915_vma *vma;
1405 int count_inactive, count_active;
1406
1407 count_inactive = 0;
1408 list_for_each_entry(vma, &vm->active_list, vm_link)
1409 count_inactive++;
1410
1411 count_active = 0;
1412 list_for_each_entry(vma, &vm->inactive_list, vm_link)
1413 count_active++;
1414
1415 bo = NULL;
1416 if (count_inactive + count_active)
1417 bo = kcalloc(count_inactive + count_active,
1418 sizeof(*bo), GFP_ATOMIC);
1419 if (!bo)
1420 return;
1421
1422 count_inactive = capture_error_bo(bo, count_inactive,
1423 &vm->active_list, true);
1424 count_active = capture_error_bo(bo + count_inactive, count_active,
1425 &vm->inactive_list, true);
1426 error->pinned_bo_count = count_inactive + count_active;
1427 error->pinned_bo = bo;
1428}
1429
1430static void i915_gem_capture_guc_log_buffer(struct drm_i915_private *dev_priv,
1431 struct drm_i915_error_state *error)
1432{
1433 /* Capturing log buf contents won't be useful if logging was disabled */
1434 if (!dev_priv->guc.log.vma || (i915.guc_log_level < 0))
1435 return;
1436
1437 error->guc_log = i915_error_object_create(dev_priv,
1438 dev_priv->guc.log.vma);
1439}
1440
1441/* Capture all registers which don't fit into another category. */
1442static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1443 struct drm_i915_error_state *error)
1444{
1445 int i;
1446
1447 /* General organization
1448 * 1. Registers specific to a single generation
1449 * 2. Registers which belong to multiple generations
1450 * 3. Feature specific registers.
1451 * 4. Everything else
1452 * Please try to follow the order.
1453 */
1454
1455 /* 1: Registers specific to a single generation */
1456 if (IS_VALLEYVIEW(dev_priv)) {
1457 error->gtier[0] = I915_READ(GTIER);
1458 error->ier = I915_READ(VLV_IER);
1459 error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
1460 }
1461
1462 if (IS_GEN7(dev_priv))
1463 error->err_int = I915_READ(GEN7_ERR_INT);
1464
1465 if (INTEL_GEN(dev_priv) >= 8) {
1466 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1467 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1468 }
1469
1470 if (IS_GEN6(dev_priv)) {
1471 error->forcewake = I915_READ_FW(FORCEWAKE);
1472 error->gab_ctl = I915_READ(GAB_CTL);
1473 error->gfx_mode = I915_READ(GFX_MODE);
1474 }
1475
1476 /* 2: Registers which belong to multiple generations */
1477 if (INTEL_GEN(dev_priv) >= 7)
1478 error->forcewake = I915_READ_FW(FORCEWAKE_MT);
1479
1480 if (INTEL_GEN(dev_priv) >= 6) {
1481 error->derrmr = I915_READ(DERRMR);
1482 error->error = I915_READ(ERROR_GEN6);
1483 error->done_reg = I915_READ(DONE_REG);
1484 }
1485
1486 /* 3: Feature specific registers */
1487 if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv)) {
1488 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1489 error->gac_eco = I915_READ(GAC_ECO_BITS);
1490 }
1491
1492 /* 4: Everything else */
1493 if (HAS_HW_CONTEXTS(dev_priv))
1494 error->ccid = I915_READ(CCID);
1495
1496 if (INTEL_GEN(dev_priv) >= 8) {
1497 error->ier = I915_READ(GEN8_DE_MISC_IER);
1498 for (i = 0; i < 4; i++)
1499 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1500 } else if (HAS_PCH_SPLIT(dev_priv)) {
1501 error->ier = I915_READ(DEIER);
1502 error->gtier[0] = I915_READ(GTIER);
1503 } else if (IS_GEN2(dev_priv)) {
1504 error->ier = I915_READ16(IER);
1505 } else if (!IS_VALLEYVIEW(dev_priv)) {
1506 error->ier = I915_READ(IER);
1507 }
1508 error->eir = I915_READ(EIR);
1509 error->pgtbl_er = I915_READ(PGTBL_ER);
1510}
1511
1512static void i915_error_capture_msg(struct drm_i915_private *dev_priv,
1513 struct drm_i915_error_state *error,
1514 u32 engine_mask,
1515 const char *error_msg)
1516{
1517 u32 ecode;
1518 int engine_id = -1, len;
1519
1520 ecode = i915_error_generate_code(dev_priv, error, &engine_id);
1521
1522 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1523 "GPU HANG: ecode %d:%d:0x%08x",
1524 INTEL_GEN(dev_priv), engine_id, ecode);
1525
1526 if (engine_id != -1 && error->engine[engine_id].pid != -1)
1527 len += scnprintf(error->error_msg + len,
1528 sizeof(error->error_msg) - len,
1529 ", in %s [%d]",
1530 error->engine[engine_id].comm,
1531 error->engine[engine_id].pid);
1532
1533 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1534 ", reason: %s, action: %s",
1535 error_msg,
1536 engine_mask ? "reset" : "continue");
1537}
1538
1539static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1540 struct drm_i915_error_state *error)
1541{
1542 error->iommu = -1;
1543#ifdef CONFIG_INTEL_IOMMU
1544 error->iommu = intel_iommu_gfx_mapped;
1545#endif
1546 error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1547 error->suspend_count = dev_priv->suspend_count;
1548
1549 memcpy(&error->device_info,
1550 INTEL_INFO(dev_priv),
1551 sizeof(error->device_info));
1552}
1553
1554static int capture(void *data)
1555{
1556 struct drm_i915_error_state *error = data;
1557
1558 i915_capture_gen_state(error->i915, error);
1559 i915_capture_reg_state(error->i915, error);
1560 i915_gem_record_fences(error->i915, error);
1561 i915_gem_record_rings(error->i915, error);
1562 i915_capture_active_buffers(error->i915, error);
1563 i915_capture_pinned_buffers(error->i915, error);
1564 i915_gem_capture_guc_log_buffer(error->i915, error);
1565
1566 do_gettimeofday(&error->time);
1567 error->boottime = ktime_to_timeval(ktime_get_boottime());
1568 error->uptime =
1569 ktime_to_timeval(ktime_sub(ktime_get(),
1570 error->i915->gt.last_init_time));
1571
1572 error->overlay = intel_overlay_capture_error_state(error->i915);
1573 error->display = intel_display_capture_error_state(error->i915);
1574
1575 return 0;
1576}
1577
1578#define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1579
1580/**
1581 * i915_capture_error_state - capture an error record for later analysis
1582 * @dev: drm device
1583 *
1584 * Should be called when an error is detected (either a hang or an error
1585 * interrupt) to capture error state from the time of the error. Fills
1586 * out a structure which becomes available in debugfs for user level tools
1587 * to pick up.
1588 */
1589void i915_capture_error_state(struct drm_i915_private *dev_priv,
1590 u32 engine_mask,
1591 const char *error_msg)
1592{
1593 static bool warned;
1594 struct drm_i915_error_state *error;
1595 unsigned long flags;
1596
1597 if (!i915.error_capture)
1598 return;
1599
1600 if (READ_ONCE(dev_priv->gpu_error.first_error))
1601 return;
1602
1603 /* Account for pipe specific data like PIPE*STAT */
1604 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1605 if (!error) {
1606 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1607 return;
1608 }
1609
1610 kref_init(&error->ref);
1611 error->i915 = dev_priv;
1612
1613 stop_machine(capture, error, NULL);
1614
1615 i915_error_capture_msg(dev_priv, error, engine_mask, error_msg);
1616 DRM_INFO("%s\n", error->error_msg);
1617
1618 if (!error->simulated) {
1619 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1620 if (!dev_priv->gpu_error.first_error) {
1621 dev_priv->gpu_error.first_error = error;
1622 error = NULL;
1623 }
1624 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1625 }
1626
1627 if (error) {
1628 i915_error_state_free(&error->ref);
1629 return;
1630 }
1631
1632 if (!warned &&
1633 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1634 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1635 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1636 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1637 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1638 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1639 dev_priv->drm.primary->index);
1640 warned = true;
1641 }
1642}
1643
1644void i915_error_state_get(struct drm_device *dev,
1645 struct i915_error_state_file_priv *error_priv)
1646{
1647 struct drm_i915_private *dev_priv = to_i915(dev);
1648
1649 spin_lock_irq(&dev_priv->gpu_error.lock);
1650 error_priv->error = dev_priv->gpu_error.first_error;
1651 if (error_priv->error)
1652 kref_get(&error_priv->error->ref);
1653 spin_unlock_irq(&dev_priv->gpu_error.lock);
1654}
1655
1656void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1657{
1658 if (error_priv->error)
1659 kref_put(&error_priv->error->ref, i915_error_state_free);
1660}
1661
1662void i915_destroy_error_state(struct drm_device *dev)
1663{
1664 struct drm_i915_private *dev_priv = to_i915(dev);
1665 struct drm_i915_error_state *error;
1666
1667 spin_lock_irq(&dev_priv->gpu_error.lock);
1668 error = dev_priv->gpu_error.first_error;
1669 dev_priv->gpu_error.first_error = NULL;
1670 spin_unlock_irq(&dev_priv->gpu_error.lock);
1671
1672 if (error)
1673 kref_put(&error->ref, i915_error_state_free);
1674}
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 <linux/ascii85.h>
31#include <linux/nmi.h>
32#include <linux/pagevec.h>
33#include <linux/scatterlist.h>
34#include <linux/utsname.h>
35#include <linux/zlib.h>
36
37#include <drm/drm_print.h>
38
39#include "display/intel_dmc.h"
40#include "display/intel_overlay.h"
41
42#include "gem/i915_gem_context.h"
43#include "gem/i915_gem_lmem.h"
44#include "gt/intel_gt.h"
45#include "gt/intel_gt_pm.h"
46
47#include "i915_drv.h"
48#include "i915_gpu_error.h"
49#include "i915_memcpy.h"
50#include "i915_scatterlist.h"
51
52#define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
53#define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
54
55static void __sg_set_buf(struct scatterlist *sg,
56 void *addr, unsigned int len, loff_t it)
57{
58 sg->page_link = (unsigned long)virt_to_page(addr);
59 sg->offset = offset_in_page(addr);
60 sg->length = len;
61 sg->dma_address = it;
62}
63
64static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
65{
66 if (!len)
67 return false;
68
69 if (e->bytes + len + 1 <= e->size)
70 return true;
71
72 if (e->bytes) {
73 __sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
74 e->iter += e->bytes;
75 e->buf = NULL;
76 e->bytes = 0;
77 }
78
79 if (e->cur == e->end) {
80 struct scatterlist *sgl;
81
82 sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
83 if (!sgl) {
84 e->err = -ENOMEM;
85 return false;
86 }
87
88 if (e->cur) {
89 e->cur->offset = 0;
90 e->cur->length = 0;
91 e->cur->page_link =
92 (unsigned long)sgl | SG_CHAIN;
93 } else {
94 e->sgl = sgl;
95 }
96
97 e->cur = sgl;
98 e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
99 }
100
101 e->size = ALIGN(len + 1, SZ_64K);
102 e->buf = kmalloc(e->size, ALLOW_FAIL);
103 if (!e->buf) {
104 e->size = PAGE_ALIGN(len + 1);
105 e->buf = kmalloc(e->size, GFP_KERNEL);
106 }
107 if (!e->buf) {
108 e->err = -ENOMEM;
109 return false;
110 }
111
112 return true;
113}
114
115__printf(2, 0)
116static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
117 const char *fmt, va_list args)
118{
119 va_list ap;
120 int len;
121
122 if (e->err)
123 return;
124
125 va_copy(ap, args);
126 len = vsnprintf(NULL, 0, fmt, ap);
127 va_end(ap);
128 if (len <= 0) {
129 e->err = len;
130 return;
131 }
132
133 if (!__i915_error_grow(e, len))
134 return;
135
136 GEM_BUG_ON(e->bytes >= e->size);
137 len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
138 if (len < 0) {
139 e->err = len;
140 return;
141 }
142 e->bytes += len;
143}
144
145static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
146{
147 unsigned len;
148
149 if (e->err || !str)
150 return;
151
152 len = strlen(str);
153 if (!__i915_error_grow(e, len))
154 return;
155
156 GEM_BUG_ON(e->bytes + len > e->size);
157 memcpy(e->buf + e->bytes, str, len);
158 e->bytes += len;
159}
160
161#define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
162#define err_puts(e, s) i915_error_puts(e, s)
163
164static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
165{
166 i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
167}
168
169static inline struct drm_printer
170i915_error_printer(struct drm_i915_error_state_buf *e)
171{
172 struct drm_printer p = {
173 .printfn = __i915_printfn_error,
174 .arg = e,
175 };
176 return p;
177}
178
179/* single threaded page allocator with a reserved stash for emergencies */
180static void pool_fini(struct pagevec *pv)
181{
182 pagevec_release(pv);
183}
184
185static int pool_refill(struct pagevec *pv, gfp_t gfp)
186{
187 while (pagevec_space(pv)) {
188 struct page *p;
189
190 p = alloc_page(gfp);
191 if (!p)
192 return -ENOMEM;
193
194 pagevec_add(pv, p);
195 }
196
197 return 0;
198}
199
200static int pool_init(struct pagevec *pv, gfp_t gfp)
201{
202 int err;
203
204 pagevec_init(pv);
205
206 err = pool_refill(pv, gfp);
207 if (err)
208 pool_fini(pv);
209
210 return err;
211}
212
213static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
214{
215 struct page *p;
216
217 p = alloc_page(gfp);
218 if (!p && pagevec_count(pv))
219 p = pv->pages[--pv->nr];
220
221 return p ? page_address(p) : NULL;
222}
223
224static void pool_free(struct pagevec *pv, void *addr)
225{
226 struct page *p = virt_to_page(addr);
227
228 if (pagevec_space(pv))
229 pagevec_add(pv, p);
230 else
231 __free_page(p);
232}
233
234#ifdef CONFIG_DRM_I915_COMPRESS_ERROR
235
236struct i915_vma_compress {
237 struct pagevec pool;
238 struct z_stream_s zstream;
239 void *tmp;
240};
241
242static bool compress_init(struct i915_vma_compress *c)
243{
244 struct z_stream_s *zstream = &c->zstream;
245
246 if (pool_init(&c->pool, ALLOW_FAIL))
247 return false;
248
249 zstream->workspace =
250 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
251 ALLOW_FAIL);
252 if (!zstream->workspace) {
253 pool_fini(&c->pool);
254 return false;
255 }
256
257 c->tmp = NULL;
258 if (i915_has_memcpy_from_wc())
259 c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
260
261 return true;
262}
263
264static bool compress_start(struct i915_vma_compress *c)
265{
266 struct z_stream_s *zstream = &c->zstream;
267 void *workspace = zstream->workspace;
268
269 memset(zstream, 0, sizeof(*zstream));
270 zstream->workspace = workspace;
271
272 return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
273}
274
275static void *compress_next_page(struct i915_vma_compress *c,
276 struct i915_vma_coredump *dst)
277{
278 void *page;
279
280 if (dst->page_count >= dst->num_pages)
281 return ERR_PTR(-ENOSPC);
282
283 page = pool_alloc(&c->pool, ALLOW_FAIL);
284 if (!page)
285 return ERR_PTR(-ENOMEM);
286
287 return dst->pages[dst->page_count++] = page;
288}
289
290static int compress_page(struct i915_vma_compress *c,
291 void *src,
292 struct i915_vma_coredump *dst,
293 bool wc)
294{
295 struct z_stream_s *zstream = &c->zstream;
296
297 zstream->next_in = src;
298 if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
299 zstream->next_in = c->tmp;
300 zstream->avail_in = PAGE_SIZE;
301
302 do {
303 if (zstream->avail_out == 0) {
304 zstream->next_out = compress_next_page(c, dst);
305 if (IS_ERR(zstream->next_out))
306 return PTR_ERR(zstream->next_out);
307
308 zstream->avail_out = PAGE_SIZE;
309 }
310
311 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
312 return -EIO;
313
314 cond_resched();
315 } while (zstream->avail_in);
316
317 /* Fallback to uncompressed if we increase size? */
318 if (0 && zstream->total_out > zstream->total_in)
319 return -E2BIG;
320
321 return 0;
322}
323
324static int compress_flush(struct i915_vma_compress *c,
325 struct i915_vma_coredump *dst)
326{
327 struct z_stream_s *zstream = &c->zstream;
328
329 do {
330 switch (zlib_deflate(zstream, Z_FINISH)) {
331 case Z_OK: /* more space requested */
332 zstream->next_out = compress_next_page(c, dst);
333 if (IS_ERR(zstream->next_out))
334 return PTR_ERR(zstream->next_out);
335
336 zstream->avail_out = PAGE_SIZE;
337 break;
338
339 case Z_STREAM_END:
340 goto end;
341
342 default: /* any error */
343 return -EIO;
344 }
345 } while (1);
346
347end:
348 memset(zstream->next_out, 0, zstream->avail_out);
349 dst->unused = zstream->avail_out;
350 return 0;
351}
352
353static void compress_finish(struct i915_vma_compress *c)
354{
355 zlib_deflateEnd(&c->zstream);
356}
357
358static void compress_fini(struct i915_vma_compress *c)
359{
360 kfree(c->zstream.workspace);
361 if (c->tmp)
362 pool_free(&c->pool, c->tmp);
363 pool_fini(&c->pool);
364}
365
366static void err_compression_marker(struct drm_i915_error_state_buf *m)
367{
368 err_puts(m, ":");
369}
370
371#else
372
373struct i915_vma_compress {
374 struct pagevec pool;
375};
376
377static bool compress_init(struct i915_vma_compress *c)
378{
379 return pool_init(&c->pool, ALLOW_FAIL) == 0;
380}
381
382static bool compress_start(struct i915_vma_compress *c)
383{
384 return true;
385}
386
387static int compress_page(struct i915_vma_compress *c,
388 void *src,
389 struct i915_vma_coredump *dst,
390 bool wc)
391{
392 void *ptr;
393
394 ptr = pool_alloc(&c->pool, ALLOW_FAIL);
395 if (!ptr)
396 return -ENOMEM;
397
398 if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
399 memcpy(ptr, src, PAGE_SIZE);
400 dst->pages[dst->page_count++] = ptr;
401 cond_resched();
402
403 return 0;
404}
405
406static int compress_flush(struct i915_vma_compress *c,
407 struct i915_vma_coredump *dst)
408{
409 return 0;
410}
411
412static void compress_finish(struct i915_vma_compress *c)
413{
414}
415
416static void compress_fini(struct i915_vma_compress *c)
417{
418 pool_fini(&c->pool);
419}
420
421static void err_compression_marker(struct drm_i915_error_state_buf *m)
422{
423 err_puts(m, "~");
424}
425
426#endif
427
428static void error_print_instdone(struct drm_i915_error_state_buf *m,
429 const struct intel_engine_coredump *ee)
430{
431 const struct sseu_dev_info *sseu = &ee->engine->gt->info.sseu;
432 int slice;
433 int subslice;
434
435 err_printf(m, " INSTDONE: 0x%08x\n",
436 ee->instdone.instdone);
437
438 if (ee->engine->class != RENDER_CLASS || GRAPHICS_VER(m->i915) <= 3)
439 return;
440
441 err_printf(m, " SC_INSTDONE: 0x%08x\n",
442 ee->instdone.slice_common);
443
444 if (GRAPHICS_VER(m->i915) <= 6)
445 return;
446
447 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
448 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
449 slice, subslice,
450 ee->instdone.sampler[slice][subslice]);
451
452 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
453 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n",
454 slice, subslice,
455 ee->instdone.row[slice][subslice]);
456
457 if (GRAPHICS_VER(m->i915) < 12)
458 return;
459
460 err_printf(m, " SC_INSTDONE_EXTRA: 0x%08x\n",
461 ee->instdone.slice_common_extra[0]);
462 err_printf(m, " SC_INSTDONE_EXTRA2: 0x%08x\n",
463 ee->instdone.slice_common_extra[1]);
464}
465
466static void error_print_request(struct drm_i915_error_state_buf *m,
467 const char *prefix,
468 const struct i915_request_coredump *erq)
469{
470 if (!erq->seqno)
471 return;
472
473 err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n",
474 prefix, erq->pid, erq->context, erq->seqno,
475 test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
476 &erq->flags) ? "!" : "",
477 test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
478 &erq->flags) ? "+" : "",
479 erq->sched_attr.priority,
480 erq->head, erq->tail);
481}
482
483static void error_print_context(struct drm_i915_error_state_buf *m,
484 const char *header,
485 const struct i915_gem_context_coredump *ctx)
486{
487 const u32 period = m->i915->gt.clock_period_ns;
488
489 err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
490 header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
491 ctx->guilty, ctx->active,
492 ctx->total_runtime * period,
493 mul_u32_u32(ctx->avg_runtime, period));
494}
495
496static struct i915_vma_coredump *
497__find_vma(struct i915_vma_coredump *vma, const char *name)
498{
499 while (vma) {
500 if (strcmp(vma->name, name) == 0)
501 return vma;
502 vma = vma->next;
503 }
504
505 return NULL;
506}
507
508static struct i915_vma_coredump *
509find_batch(const struct intel_engine_coredump *ee)
510{
511 return __find_vma(ee->vma, "batch");
512}
513
514static void error_print_engine(struct drm_i915_error_state_buf *m,
515 const struct intel_engine_coredump *ee)
516{
517 struct i915_vma_coredump *batch;
518 int n;
519
520 err_printf(m, "%s command stream:\n", ee->engine->name);
521 err_printf(m, " CCID: 0x%08x\n", ee->ccid);
522 err_printf(m, " START: 0x%08x\n", ee->start);
523 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head);
524 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n",
525 ee->tail, ee->rq_post, ee->rq_tail);
526 err_printf(m, " CTL: 0x%08x\n", ee->ctl);
527 err_printf(m, " MODE: 0x%08x\n", ee->mode);
528 err_printf(m, " HWS: 0x%08x\n", ee->hws);
529 err_printf(m, " ACTHD: 0x%08x %08x\n",
530 (u32)(ee->acthd>>32), (u32)ee->acthd);
531 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir);
532 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr);
533 err_printf(m, " ESR: 0x%08x\n", ee->esr);
534
535 error_print_instdone(m, ee);
536
537 batch = find_batch(ee);
538 if (batch) {
539 u64 start = batch->gtt_offset;
540 u64 end = start + batch->gtt_size;
541
542 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n",
543 upper_32_bits(start), lower_32_bits(start),
544 upper_32_bits(end), lower_32_bits(end));
545 }
546 if (GRAPHICS_VER(m->i915) >= 4) {
547 err_printf(m, " BBADDR: 0x%08x_%08x\n",
548 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
549 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate);
550 err_printf(m, " INSTPS: 0x%08x\n", ee->instps);
551 }
552 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm);
553 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
554 lower_32_bits(ee->faddr));
555 if (GRAPHICS_VER(m->i915) >= 6) {
556 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi);
557 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg);
558 }
559 if (HAS_PPGTT(m->i915)) {
560 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
561
562 if (GRAPHICS_VER(m->i915) >= 8) {
563 int i;
564 for (i = 0; i < 4; i++)
565 err_printf(m, " PDP%d: 0x%016llx\n",
566 i, ee->vm_info.pdp[i]);
567 } else {
568 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
569 ee->vm_info.pp_dir_base);
570 }
571 }
572 err_printf(m, " hung: %u\n", ee->hung);
573 err_printf(m, " engine reset count: %u\n", ee->reset_count);
574
575 for (n = 0; n < ee->num_ports; n++) {
576 err_printf(m, " ELSP[%d]:", n);
577 error_print_request(m, " ", &ee->execlist[n]);
578 }
579
580 error_print_context(m, " Active context: ", &ee->context);
581}
582
583void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
584{
585 va_list args;
586
587 va_start(args, f);
588 i915_error_vprintf(e, f, args);
589 va_end(args);
590}
591
592static void print_error_vma(struct drm_i915_error_state_buf *m,
593 const struct intel_engine_cs *engine,
594 const struct i915_vma_coredump *vma)
595{
596 char out[ASCII85_BUFSZ];
597 int page;
598
599 if (!vma)
600 return;
601
602 err_printf(m, "%s --- %s = 0x%08x %08x\n",
603 engine ? engine->name : "global", vma->name,
604 upper_32_bits(vma->gtt_offset),
605 lower_32_bits(vma->gtt_offset));
606
607 if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
608 err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
609
610 err_compression_marker(m);
611 for (page = 0; page < vma->page_count; page++) {
612 int i, len;
613
614 len = PAGE_SIZE;
615 if (page == vma->page_count - 1)
616 len -= vma->unused;
617 len = ascii85_encode_len(len);
618
619 for (i = 0; i < len; i++)
620 err_puts(m, ascii85_encode(vma->pages[page][i], out));
621 }
622 err_puts(m, "\n");
623}
624
625static void err_print_capabilities(struct drm_i915_error_state_buf *m,
626 struct i915_gpu_coredump *error)
627{
628 struct drm_printer p = i915_error_printer(m);
629
630 intel_device_info_print_static(&error->device_info, &p);
631 intel_device_info_print_runtime(&error->runtime_info, &p);
632 intel_driver_caps_print(&error->driver_caps, &p);
633}
634
635static void err_print_params(struct drm_i915_error_state_buf *m,
636 const struct i915_params *params)
637{
638 struct drm_printer p = i915_error_printer(m);
639
640 i915_params_dump(params, &p);
641}
642
643static void err_print_pciid(struct drm_i915_error_state_buf *m,
644 struct drm_i915_private *i915)
645{
646 struct pci_dev *pdev = to_pci_dev(i915->drm.dev);
647
648 err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
649 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
650 err_printf(m, "PCI Subsystem: %04x:%04x\n",
651 pdev->subsystem_vendor,
652 pdev->subsystem_device);
653}
654
655static void err_print_uc(struct drm_i915_error_state_buf *m,
656 const struct intel_uc_coredump *error_uc)
657{
658 struct drm_printer p = i915_error_printer(m);
659
660 intel_uc_fw_dump(&error_uc->guc_fw, &p);
661 intel_uc_fw_dump(&error_uc->huc_fw, &p);
662 print_error_vma(m, NULL, error_uc->guc_log);
663}
664
665static void err_free_sgl(struct scatterlist *sgl)
666{
667 while (sgl) {
668 struct scatterlist *sg;
669
670 for (sg = sgl; !sg_is_chain(sg); sg++) {
671 kfree(sg_virt(sg));
672 if (sg_is_last(sg))
673 break;
674 }
675
676 sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
677 free_page((unsigned long)sgl);
678 sgl = sg;
679 }
680}
681
682static void err_print_gt_info(struct drm_i915_error_state_buf *m,
683 struct intel_gt_coredump *gt)
684{
685 struct drm_printer p = i915_error_printer(m);
686
687 intel_gt_info_print(>->info, &p);
688 intel_sseu_print_topology(>->info.sseu, &p);
689}
690
691static void err_print_gt(struct drm_i915_error_state_buf *m,
692 struct intel_gt_coredump *gt)
693{
694 const struct intel_engine_coredump *ee;
695 int i;
696
697 err_printf(m, "GT awake: %s\n", yesno(gt->awake));
698 err_printf(m, "EIR: 0x%08x\n", gt->eir);
699 err_printf(m, "IER: 0x%08x\n", gt->ier);
700 for (i = 0; i < gt->ngtier; i++)
701 err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
702 err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
703 err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
704 err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
705
706 for (i = 0; i < gt->nfence; i++)
707 err_printf(m, " fence[%d] = %08llx\n", i, gt->fence[i]);
708
709 if (IS_GRAPHICS_VER(m->i915, 6, 11)) {
710 err_printf(m, "ERROR: 0x%08x\n", gt->error);
711 err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
712 }
713
714 if (GRAPHICS_VER(m->i915) >= 8)
715 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
716 gt->fault_data1, gt->fault_data0);
717
718 if (GRAPHICS_VER(m->i915) == 7)
719 err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
720
721 if (IS_GRAPHICS_VER(m->i915, 8, 11))
722 err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
723
724 if (GRAPHICS_VER(m->i915) == 12)
725 err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
726
727 if (GRAPHICS_VER(m->i915) >= 12) {
728 int i;
729
730 for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
731 /*
732 * SFC_DONE resides in the VD forcewake domain, so it
733 * only exists if the corresponding VCS engine is
734 * present.
735 */
736 if (!HAS_ENGINE(gt->_gt, _VCS(i * 2)))
737 continue;
738
739 err_printf(m, " SFC_DONE[%d]: 0x%08x\n", i,
740 gt->sfc_done[i]);
741 }
742
743 err_printf(m, " GAM_DONE: 0x%08x\n", gt->gam_done);
744 }
745
746 for (ee = gt->engine; ee; ee = ee->next) {
747 const struct i915_vma_coredump *vma;
748
749 error_print_engine(m, ee);
750 for (vma = ee->vma; vma; vma = vma->next)
751 print_error_vma(m, ee->engine, vma);
752 }
753
754 if (gt->uc)
755 err_print_uc(m, gt->uc);
756
757 err_print_gt_info(m, gt);
758}
759
760static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
761 struct i915_gpu_coredump *error)
762{
763 const struct intel_engine_coredump *ee;
764 struct timespec64 ts;
765
766 if (*error->error_msg)
767 err_printf(m, "%s\n", error->error_msg);
768 err_printf(m, "Kernel: %s %s\n",
769 init_utsname()->release,
770 init_utsname()->machine);
771 err_printf(m, "Driver: %s\n", DRIVER_DATE);
772 ts = ktime_to_timespec64(error->time);
773 err_printf(m, "Time: %lld s %ld us\n",
774 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
775 ts = ktime_to_timespec64(error->boottime);
776 err_printf(m, "Boottime: %lld s %ld us\n",
777 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
778 ts = ktime_to_timespec64(error->uptime);
779 err_printf(m, "Uptime: %lld s %ld us\n",
780 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
781 err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
782 error->capture, jiffies_to_msecs(jiffies - error->capture));
783
784 for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
785 err_printf(m, "Active process (on ring %s): %s [%d]\n",
786 ee->engine->name,
787 ee->context.comm,
788 ee->context.pid);
789
790 err_printf(m, "Reset count: %u\n", error->reset_count);
791 err_printf(m, "Suspend count: %u\n", error->suspend_count);
792 err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
793 err_printf(m, "Subplatform: 0x%x\n",
794 intel_subplatform(&error->runtime_info,
795 error->device_info.platform));
796 err_print_pciid(m, m->i915);
797
798 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
799
800 if (HAS_DMC(m->i915)) {
801 struct intel_dmc *dmc = &m->i915->dmc;
802
803 err_printf(m, "DMC loaded: %s\n",
804 yesno(intel_dmc_has_payload(m->i915) != 0));
805 err_printf(m, "DMC fw version: %d.%d\n",
806 DMC_VERSION_MAJOR(dmc->version),
807 DMC_VERSION_MINOR(dmc->version));
808 }
809
810 err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
811 err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
812
813 if (error->gt)
814 err_print_gt(m, error->gt);
815
816 if (error->overlay)
817 intel_overlay_print_error_state(m, error->overlay);
818
819 err_print_capabilities(m, error);
820 err_print_params(m, &error->params);
821}
822
823static int err_print_to_sgl(struct i915_gpu_coredump *error)
824{
825 struct drm_i915_error_state_buf m;
826
827 if (IS_ERR(error))
828 return PTR_ERR(error);
829
830 if (READ_ONCE(error->sgl))
831 return 0;
832
833 memset(&m, 0, sizeof(m));
834 m.i915 = error->i915;
835
836 __err_print_to_sgl(&m, error);
837
838 if (m.buf) {
839 __sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
840 m.bytes = 0;
841 m.buf = NULL;
842 }
843 if (m.cur) {
844 GEM_BUG_ON(m.end < m.cur);
845 sg_mark_end(m.cur - 1);
846 }
847 GEM_BUG_ON(m.sgl && !m.cur);
848
849 if (m.err) {
850 err_free_sgl(m.sgl);
851 return m.err;
852 }
853
854 if (cmpxchg(&error->sgl, NULL, m.sgl))
855 err_free_sgl(m.sgl);
856
857 return 0;
858}
859
860ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
861 char *buf, loff_t off, size_t rem)
862{
863 struct scatterlist *sg;
864 size_t count;
865 loff_t pos;
866 int err;
867
868 if (!error || !rem)
869 return 0;
870
871 err = err_print_to_sgl(error);
872 if (err)
873 return err;
874
875 sg = READ_ONCE(error->fit);
876 if (!sg || off < sg->dma_address)
877 sg = error->sgl;
878 if (!sg)
879 return 0;
880
881 pos = sg->dma_address;
882 count = 0;
883 do {
884 size_t len, start;
885
886 if (sg_is_chain(sg)) {
887 sg = sg_chain_ptr(sg);
888 GEM_BUG_ON(sg_is_chain(sg));
889 }
890
891 len = sg->length;
892 if (pos + len <= off) {
893 pos += len;
894 continue;
895 }
896
897 start = sg->offset;
898 if (pos < off) {
899 GEM_BUG_ON(off - pos > len);
900 len -= off - pos;
901 start += off - pos;
902 pos = off;
903 }
904
905 len = min(len, rem);
906 GEM_BUG_ON(!len || len > sg->length);
907
908 memcpy(buf, page_address(sg_page(sg)) + start, len);
909
910 count += len;
911 pos += len;
912
913 buf += len;
914 rem -= len;
915 if (!rem) {
916 WRITE_ONCE(error->fit, sg);
917 break;
918 }
919 } while (!sg_is_last(sg++));
920
921 return count;
922}
923
924static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
925{
926 while (vma) {
927 struct i915_vma_coredump *next = vma->next;
928 int page;
929
930 for (page = 0; page < vma->page_count; page++)
931 free_page((unsigned long)vma->pages[page]);
932
933 kfree(vma);
934 vma = next;
935 }
936}
937
938static void cleanup_params(struct i915_gpu_coredump *error)
939{
940 i915_params_free(&error->params);
941}
942
943static void cleanup_uc(struct intel_uc_coredump *uc)
944{
945 kfree(uc->guc_fw.path);
946 kfree(uc->huc_fw.path);
947 i915_vma_coredump_free(uc->guc_log);
948
949 kfree(uc);
950}
951
952static void cleanup_gt(struct intel_gt_coredump *gt)
953{
954 while (gt->engine) {
955 struct intel_engine_coredump *ee = gt->engine;
956
957 gt->engine = ee->next;
958
959 i915_vma_coredump_free(ee->vma);
960 kfree(ee);
961 }
962
963 if (gt->uc)
964 cleanup_uc(gt->uc);
965
966 kfree(gt);
967}
968
969void __i915_gpu_coredump_free(struct kref *error_ref)
970{
971 struct i915_gpu_coredump *error =
972 container_of(error_ref, typeof(*error), ref);
973
974 while (error->gt) {
975 struct intel_gt_coredump *gt = error->gt;
976
977 error->gt = gt->next;
978 cleanup_gt(gt);
979 }
980
981 kfree(error->overlay);
982
983 cleanup_params(error);
984
985 err_free_sgl(error->sgl);
986 kfree(error);
987}
988
989static struct i915_vma_coredump *
990i915_vma_coredump_create(const struct intel_gt *gt,
991 const struct i915_vma *vma,
992 const char *name,
993 struct i915_vma_compress *compress)
994{
995 struct i915_ggtt *ggtt = gt->ggtt;
996 const u64 slot = ggtt->error_capture.start;
997 struct i915_vma_coredump *dst;
998 unsigned long num_pages;
999 struct sgt_iter iter;
1000 int ret;
1001
1002 might_sleep();
1003
1004 if (!vma || !vma->pages || !compress)
1005 return NULL;
1006
1007 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
1008 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
1009 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
1010 if (!dst)
1011 return NULL;
1012
1013 if (!compress_start(compress)) {
1014 kfree(dst);
1015 return NULL;
1016 }
1017
1018 strcpy(dst->name, name);
1019 dst->next = NULL;
1020
1021 dst->gtt_offset = vma->node.start;
1022 dst->gtt_size = vma->node.size;
1023 dst->gtt_page_sizes = vma->page_sizes.gtt;
1024 dst->num_pages = num_pages;
1025 dst->page_count = 0;
1026 dst->unused = 0;
1027
1028 ret = -EINVAL;
1029 if (drm_mm_node_allocated(&ggtt->error_capture)) {
1030 void __iomem *s;
1031 dma_addr_t dma;
1032
1033 for_each_sgt_daddr(dma, iter, vma->pages) {
1034 mutex_lock(&ggtt->error_mutex);
1035 ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1036 I915_CACHE_NONE, 0);
1037 mb();
1038
1039 s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1040 ret = compress_page(compress,
1041 (void __force *)s, dst,
1042 true);
1043 io_mapping_unmap(s);
1044
1045 mb();
1046 ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1047 mutex_unlock(&ggtt->error_mutex);
1048 if (ret)
1049 break;
1050 }
1051 } else if (i915_gem_object_is_lmem(vma->obj)) {
1052 struct intel_memory_region *mem = vma->obj->mm.region;
1053 dma_addr_t dma;
1054
1055 for_each_sgt_daddr(dma, iter, vma->pages) {
1056 void __iomem *s;
1057
1058 s = io_mapping_map_wc(&mem->iomap,
1059 dma - mem->region.start,
1060 PAGE_SIZE);
1061 ret = compress_page(compress,
1062 (void __force *)s, dst,
1063 true);
1064 io_mapping_unmap(s);
1065 if (ret)
1066 break;
1067 }
1068 } else {
1069 struct page *page;
1070
1071 for_each_sgt_page(page, iter, vma->pages) {
1072 void *s;
1073
1074 drm_clflush_pages(&page, 1);
1075
1076 s = kmap(page);
1077 ret = compress_page(compress, s, dst, false);
1078 kunmap(page);
1079
1080 drm_clflush_pages(&page, 1);
1081
1082 if (ret)
1083 break;
1084 }
1085 }
1086
1087 if (ret || compress_flush(compress, dst)) {
1088 while (dst->page_count--)
1089 pool_free(&compress->pool, dst->pages[dst->page_count]);
1090 kfree(dst);
1091 dst = NULL;
1092 }
1093 compress_finish(compress);
1094
1095 return dst;
1096}
1097
1098static void gt_record_fences(struct intel_gt_coredump *gt)
1099{
1100 struct i915_ggtt *ggtt = gt->_gt->ggtt;
1101 struct intel_uncore *uncore = gt->_gt->uncore;
1102 int i;
1103
1104 if (GRAPHICS_VER(uncore->i915) >= 6) {
1105 for (i = 0; i < ggtt->num_fences; i++)
1106 gt->fence[i] =
1107 intel_uncore_read64(uncore,
1108 FENCE_REG_GEN6_LO(i));
1109 } else if (GRAPHICS_VER(uncore->i915) >= 4) {
1110 for (i = 0; i < ggtt->num_fences; i++)
1111 gt->fence[i] =
1112 intel_uncore_read64(uncore,
1113 FENCE_REG_965_LO(i));
1114 } else {
1115 for (i = 0; i < ggtt->num_fences; i++)
1116 gt->fence[i] =
1117 intel_uncore_read(uncore, FENCE_REG(i));
1118 }
1119 gt->nfence = i;
1120}
1121
1122static void engine_record_registers(struct intel_engine_coredump *ee)
1123{
1124 const struct intel_engine_cs *engine = ee->engine;
1125 struct drm_i915_private *i915 = engine->i915;
1126
1127 if (GRAPHICS_VER(i915) >= 6) {
1128 ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1129
1130 if (GRAPHICS_VER(i915) >= 12)
1131 ee->fault_reg = intel_uncore_read(engine->uncore,
1132 GEN12_RING_FAULT_REG);
1133 else if (GRAPHICS_VER(i915) >= 8)
1134 ee->fault_reg = intel_uncore_read(engine->uncore,
1135 GEN8_RING_FAULT_REG);
1136 else
1137 ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1138 }
1139
1140 if (GRAPHICS_VER(i915) >= 4) {
1141 ee->esr = ENGINE_READ(engine, RING_ESR);
1142 ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1143 ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1144 ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1145 ee->instps = ENGINE_READ(engine, RING_INSTPS);
1146 ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1147 ee->ccid = ENGINE_READ(engine, CCID);
1148 if (GRAPHICS_VER(i915) >= 8) {
1149 ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1150 ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1151 }
1152 ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1153 } else {
1154 ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1155 ee->ipeir = ENGINE_READ(engine, IPEIR);
1156 ee->ipehr = ENGINE_READ(engine, IPEHR);
1157 }
1158
1159 intel_engine_get_instdone(engine, &ee->instdone);
1160
1161 ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1162 ee->acthd = intel_engine_get_active_head(engine);
1163 ee->start = ENGINE_READ(engine, RING_START);
1164 ee->head = ENGINE_READ(engine, RING_HEAD);
1165 ee->tail = ENGINE_READ(engine, RING_TAIL);
1166 ee->ctl = ENGINE_READ(engine, RING_CTL);
1167 if (GRAPHICS_VER(i915) > 2)
1168 ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1169
1170 if (!HWS_NEEDS_PHYSICAL(i915)) {
1171 i915_reg_t mmio;
1172
1173 if (GRAPHICS_VER(i915) == 7) {
1174 switch (engine->id) {
1175 default:
1176 MISSING_CASE(engine->id);
1177 fallthrough;
1178 case RCS0:
1179 mmio = RENDER_HWS_PGA_GEN7;
1180 break;
1181 case BCS0:
1182 mmio = BLT_HWS_PGA_GEN7;
1183 break;
1184 case VCS0:
1185 mmio = BSD_HWS_PGA_GEN7;
1186 break;
1187 case VECS0:
1188 mmio = VEBOX_HWS_PGA_GEN7;
1189 break;
1190 }
1191 } else if (GRAPHICS_VER(engine->i915) == 6) {
1192 mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1193 } else {
1194 /* XXX: gen8 returns to sanity */
1195 mmio = RING_HWS_PGA(engine->mmio_base);
1196 }
1197
1198 ee->hws = intel_uncore_read(engine->uncore, mmio);
1199 }
1200
1201 ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1202
1203 if (HAS_PPGTT(i915)) {
1204 int i;
1205
1206 ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1207
1208 if (GRAPHICS_VER(i915) == 6) {
1209 ee->vm_info.pp_dir_base =
1210 ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1211 } else if (GRAPHICS_VER(i915) == 7) {
1212 ee->vm_info.pp_dir_base =
1213 ENGINE_READ(engine, RING_PP_DIR_BASE);
1214 } else if (GRAPHICS_VER(i915) >= 8) {
1215 u32 base = engine->mmio_base;
1216
1217 for (i = 0; i < 4; i++) {
1218 ee->vm_info.pdp[i] =
1219 intel_uncore_read(engine->uncore,
1220 GEN8_RING_PDP_UDW(base, i));
1221 ee->vm_info.pdp[i] <<= 32;
1222 ee->vm_info.pdp[i] |=
1223 intel_uncore_read(engine->uncore,
1224 GEN8_RING_PDP_LDW(base, i));
1225 }
1226 }
1227 }
1228}
1229
1230static void record_request(const struct i915_request *request,
1231 struct i915_request_coredump *erq)
1232{
1233 erq->flags = request->fence.flags;
1234 erq->context = request->fence.context;
1235 erq->seqno = request->fence.seqno;
1236 erq->sched_attr = request->sched.attr;
1237 erq->head = request->head;
1238 erq->tail = request->tail;
1239
1240 erq->pid = 0;
1241 rcu_read_lock();
1242 if (!intel_context_is_closed(request->context)) {
1243 const struct i915_gem_context *ctx;
1244
1245 ctx = rcu_dereference(request->context->gem_context);
1246 if (ctx)
1247 erq->pid = pid_nr(ctx->pid);
1248 }
1249 rcu_read_unlock();
1250}
1251
1252static void engine_record_execlists(struct intel_engine_coredump *ee)
1253{
1254 const struct intel_engine_execlists * const el = &ee->engine->execlists;
1255 struct i915_request * const *port = el->active;
1256 unsigned int n = 0;
1257
1258 while (*port)
1259 record_request(*port++, &ee->execlist[n++]);
1260
1261 ee->num_ports = n;
1262}
1263
1264static bool record_context(struct i915_gem_context_coredump *e,
1265 const struct i915_request *rq)
1266{
1267 struct i915_gem_context *ctx;
1268 struct task_struct *task;
1269 bool simulated;
1270
1271 rcu_read_lock();
1272 ctx = rcu_dereference(rq->context->gem_context);
1273 if (ctx && !kref_get_unless_zero(&ctx->ref))
1274 ctx = NULL;
1275 rcu_read_unlock();
1276 if (!ctx)
1277 return true;
1278
1279 rcu_read_lock();
1280 task = pid_task(ctx->pid, PIDTYPE_PID);
1281 if (task) {
1282 strcpy(e->comm, task->comm);
1283 e->pid = task->pid;
1284 }
1285 rcu_read_unlock();
1286
1287 e->sched_attr = ctx->sched;
1288 e->guilty = atomic_read(&ctx->guilty_count);
1289 e->active = atomic_read(&ctx->active_count);
1290
1291 e->total_runtime = rq->context->runtime.total;
1292 e->avg_runtime = ewma_runtime_read(&rq->context->runtime.avg);
1293
1294 simulated = i915_gem_context_no_error_capture(ctx);
1295
1296 i915_gem_context_put(ctx);
1297 return simulated;
1298}
1299
1300struct intel_engine_capture_vma {
1301 struct intel_engine_capture_vma *next;
1302 struct i915_vma *vma;
1303 char name[16];
1304};
1305
1306static struct intel_engine_capture_vma *
1307capture_vma(struct intel_engine_capture_vma *next,
1308 struct i915_vma *vma,
1309 const char *name,
1310 gfp_t gfp)
1311{
1312 struct intel_engine_capture_vma *c;
1313
1314 if (!vma)
1315 return next;
1316
1317 c = kmalloc(sizeof(*c), gfp);
1318 if (!c)
1319 return next;
1320
1321 if (!i915_active_acquire_if_busy(&vma->active)) {
1322 kfree(c);
1323 return next;
1324 }
1325
1326 strcpy(c->name, name);
1327 c->vma = vma; /* reference held while active */
1328
1329 c->next = next;
1330 return c;
1331}
1332
1333static struct intel_engine_capture_vma *
1334capture_user(struct intel_engine_capture_vma *capture,
1335 const struct i915_request *rq,
1336 gfp_t gfp)
1337{
1338 struct i915_capture_list *c;
1339
1340 for (c = rq->capture_list; c; c = c->next)
1341 capture = capture_vma(capture, c->vma, "user", gfp);
1342
1343 return capture;
1344}
1345
1346static void add_vma(struct intel_engine_coredump *ee,
1347 struct i915_vma_coredump *vma)
1348{
1349 if (vma) {
1350 vma->next = ee->vma;
1351 ee->vma = vma;
1352 }
1353}
1354
1355struct intel_engine_coredump *
1356intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
1357{
1358 struct intel_engine_coredump *ee;
1359
1360 ee = kzalloc(sizeof(*ee), gfp);
1361 if (!ee)
1362 return NULL;
1363
1364 ee->engine = engine;
1365
1366 engine_record_registers(ee);
1367 engine_record_execlists(ee);
1368
1369 return ee;
1370}
1371
1372struct intel_engine_capture_vma *
1373intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1374 struct i915_request *rq,
1375 gfp_t gfp)
1376{
1377 struct intel_engine_capture_vma *vma = NULL;
1378
1379 ee->simulated |= record_context(&ee->context, rq);
1380 if (ee->simulated)
1381 return NULL;
1382
1383 /*
1384 * We need to copy these to an anonymous buffer
1385 * as the simplest method to avoid being overwritten
1386 * by userspace.
1387 */
1388 vma = capture_vma(vma, rq->batch, "batch", gfp);
1389 vma = capture_user(vma, rq, gfp);
1390 vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1391 vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1392
1393 ee->rq_head = rq->head;
1394 ee->rq_post = rq->postfix;
1395 ee->rq_tail = rq->tail;
1396
1397 return vma;
1398}
1399
1400void
1401intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1402 struct intel_engine_capture_vma *capture,
1403 struct i915_vma_compress *compress)
1404{
1405 const struct intel_engine_cs *engine = ee->engine;
1406
1407 while (capture) {
1408 struct intel_engine_capture_vma *this = capture;
1409 struct i915_vma *vma = this->vma;
1410
1411 add_vma(ee,
1412 i915_vma_coredump_create(engine->gt,
1413 vma, this->name,
1414 compress));
1415
1416 i915_active_release(&vma->active);
1417
1418 capture = this->next;
1419 kfree(this);
1420 }
1421
1422 add_vma(ee,
1423 i915_vma_coredump_create(engine->gt,
1424 engine->status_page.vma,
1425 "HW Status",
1426 compress));
1427
1428 add_vma(ee,
1429 i915_vma_coredump_create(engine->gt,
1430 engine->wa_ctx.vma,
1431 "WA context",
1432 compress));
1433}
1434
1435static struct intel_engine_coredump *
1436capture_engine(struct intel_engine_cs *engine,
1437 struct i915_vma_compress *compress)
1438{
1439 struct intel_engine_capture_vma *capture = NULL;
1440 struct intel_engine_coredump *ee;
1441 struct i915_request *rq;
1442 unsigned long flags;
1443
1444 ee = intel_engine_coredump_alloc(engine, GFP_KERNEL);
1445 if (!ee)
1446 return NULL;
1447
1448 spin_lock_irqsave(&engine->active.lock, flags);
1449 rq = intel_engine_find_active_request(engine);
1450 if (rq)
1451 capture = intel_engine_coredump_add_request(ee, rq,
1452 ATOMIC_MAYFAIL);
1453 spin_unlock_irqrestore(&engine->active.lock, flags);
1454 if (!capture) {
1455 kfree(ee);
1456 return NULL;
1457 }
1458
1459 intel_engine_coredump_add_vma(ee, capture, compress);
1460
1461 return ee;
1462}
1463
1464static void
1465gt_record_engines(struct intel_gt_coredump *gt,
1466 intel_engine_mask_t engine_mask,
1467 struct i915_vma_compress *compress)
1468{
1469 struct intel_engine_cs *engine;
1470 enum intel_engine_id id;
1471
1472 for_each_engine(engine, gt->_gt, id) {
1473 struct intel_engine_coredump *ee;
1474
1475 /* Refill our page pool before entering atomic section */
1476 pool_refill(&compress->pool, ALLOW_FAIL);
1477
1478 ee = capture_engine(engine, compress);
1479 if (!ee)
1480 continue;
1481
1482 ee->hung = engine->mask & engine_mask;
1483
1484 gt->simulated |= ee->simulated;
1485 if (ee->simulated) {
1486 kfree(ee);
1487 continue;
1488 }
1489
1490 ee->next = gt->engine;
1491 gt->engine = ee;
1492 }
1493}
1494
1495static struct intel_uc_coredump *
1496gt_record_uc(struct intel_gt_coredump *gt,
1497 struct i915_vma_compress *compress)
1498{
1499 const struct intel_uc *uc = >->_gt->uc;
1500 struct intel_uc_coredump *error_uc;
1501
1502 error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1503 if (!error_uc)
1504 return NULL;
1505
1506 memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1507 memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1508
1509 /* Non-default firmware paths will be specified by the modparam.
1510 * As modparams are generally accesible from the userspace make
1511 * explicit copies of the firmware paths.
1512 */
1513 error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1514 error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1515 error_uc->guc_log =
1516 i915_vma_coredump_create(gt->_gt,
1517 uc->guc.log.vma, "GuC log buffer",
1518 compress);
1519
1520 return error_uc;
1521}
1522
1523/* Capture all registers which don't fit into another category. */
1524static void gt_record_regs(struct intel_gt_coredump *gt)
1525{
1526 struct intel_uncore *uncore = gt->_gt->uncore;
1527 struct drm_i915_private *i915 = uncore->i915;
1528 int i;
1529
1530 /*
1531 * General organization
1532 * 1. Registers specific to a single generation
1533 * 2. Registers which belong to multiple generations
1534 * 3. Feature specific registers.
1535 * 4. Everything else
1536 * Please try to follow the order.
1537 */
1538
1539 /* 1: Registers specific to a single generation */
1540 if (IS_VALLEYVIEW(i915)) {
1541 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1542 gt->ier = intel_uncore_read(uncore, VLV_IER);
1543 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1544 }
1545
1546 if (GRAPHICS_VER(i915) == 7)
1547 gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1548
1549 if (GRAPHICS_VER(i915) >= 12) {
1550 gt->fault_data0 = intel_uncore_read(uncore,
1551 GEN12_FAULT_TLB_DATA0);
1552 gt->fault_data1 = intel_uncore_read(uncore,
1553 GEN12_FAULT_TLB_DATA1);
1554 } else if (GRAPHICS_VER(i915) >= 8) {
1555 gt->fault_data0 = intel_uncore_read(uncore,
1556 GEN8_FAULT_TLB_DATA0);
1557 gt->fault_data1 = intel_uncore_read(uncore,
1558 GEN8_FAULT_TLB_DATA1);
1559 }
1560
1561 if (GRAPHICS_VER(i915) == 6) {
1562 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1563 gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1564 gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1565 }
1566
1567 /* 2: Registers which belong to multiple generations */
1568 if (GRAPHICS_VER(i915) >= 7)
1569 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1570
1571 if (GRAPHICS_VER(i915) >= 6) {
1572 gt->derrmr = intel_uncore_read(uncore, DERRMR);
1573 if (GRAPHICS_VER(i915) < 12) {
1574 gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1575 gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1576 }
1577 }
1578
1579 /* 3: Feature specific registers */
1580 if (IS_GRAPHICS_VER(i915, 6, 7)) {
1581 gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1582 gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1583 }
1584
1585 if (IS_GRAPHICS_VER(i915, 8, 11))
1586 gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1587
1588 if (GRAPHICS_VER(i915) == 12)
1589 gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1590
1591 if (GRAPHICS_VER(i915) >= 12) {
1592 for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
1593 /*
1594 * SFC_DONE resides in the VD forcewake domain, so it
1595 * only exists if the corresponding VCS engine is
1596 * present.
1597 */
1598 if (!HAS_ENGINE(gt->_gt, _VCS(i * 2)))
1599 continue;
1600
1601 gt->sfc_done[i] =
1602 intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1603 }
1604
1605 gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1606 }
1607
1608 /* 4: Everything else */
1609 if (GRAPHICS_VER(i915) >= 11) {
1610 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1611 gt->gtier[0] =
1612 intel_uncore_read(uncore,
1613 GEN11_RENDER_COPY_INTR_ENABLE);
1614 gt->gtier[1] =
1615 intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1616 gt->gtier[2] =
1617 intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1618 gt->gtier[3] =
1619 intel_uncore_read(uncore,
1620 GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1621 gt->gtier[4] =
1622 intel_uncore_read(uncore,
1623 GEN11_CRYPTO_RSVD_INTR_ENABLE);
1624 gt->gtier[5] =
1625 intel_uncore_read(uncore,
1626 GEN11_GUNIT_CSME_INTR_ENABLE);
1627 gt->ngtier = 6;
1628 } else if (GRAPHICS_VER(i915) >= 8) {
1629 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1630 for (i = 0; i < 4; i++)
1631 gt->gtier[i] =
1632 intel_uncore_read(uncore, GEN8_GT_IER(i));
1633 gt->ngtier = 4;
1634 } else if (HAS_PCH_SPLIT(i915)) {
1635 gt->ier = intel_uncore_read(uncore, DEIER);
1636 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1637 gt->ngtier = 1;
1638 } else if (GRAPHICS_VER(i915) == 2) {
1639 gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1640 } else if (!IS_VALLEYVIEW(i915)) {
1641 gt->ier = intel_uncore_read(uncore, GEN2_IER);
1642 }
1643 gt->eir = intel_uncore_read(uncore, EIR);
1644 gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1645}
1646
1647static void gt_record_info(struct intel_gt_coredump *gt)
1648{
1649 memcpy(>->info, >->_gt->info, sizeof(struct intel_gt_info));
1650}
1651
1652/*
1653 * Generate a semi-unique error code. The code is not meant to have meaning, The
1654 * code's only purpose is to try to prevent false duplicated bug reports by
1655 * grossly estimating a GPU error state.
1656 *
1657 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1658 * the hang if we could strip the GTT offset information from it.
1659 *
1660 * It's only a small step better than a random number in its current form.
1661 */
1662static u32 generate_ecode(const struct intel_engine_coredump *ee)
1663{
1664 /*
1665 * IPEHR would be an ideal way to detect errors, as it's the gross
1666 * measure of "the command that hung." However, has some very common
1667 * synchronization commands which almost always appear in the case
1668 * strictly a client bug. Use instdone to differentiate those some.
1669 */
1670 return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1671}
1672
1673static const char *error_msg(struct i915_gpu_coredump *error)
1674{
1675 struct intel_engine_coredump *first = NULL;
1676 unsigned int hung_classes = 0;
1677 struct intel_gt_coredump *gt;
1678 int len;
1679
1680 for (gt = error->gt; gt; gt = gt->next) {
1681 struct intel_engine_coredump *cs;
1682
1683 for (cs = gt->engine; cs; cs = cs->next) {
1684 if (cs->hung) {
1685 hung_classes |= BIT(cs->engine->uabi_class);
1686 if (!first)
1687 first = cs;
1688 }
1689 }
1690 }
1691
1692 len = scnprintf(error->error_msg, sizeof(error->error_msg),
1693 "GPU HANG: ecode %d:%x:%08x",
1694 GRAPHICS_VER(error->i915), hung_classes,
1695 generate_ecode(first));
1696 if (first && first->context.pid) {
1697 /* Just show the first executing process, more is confusing */
1698 len += scnprintf(error->error_msg + len,
1699 sizeof(error->error_msg) - len,
1700 ", in %s [%d]",
1701 first->context.comm, first->context.pid);
1702 }
1703
1704 return error->error_msg;
1705}
1706
1707static void capture_gen(struct i915_gpu_coredump *error)
1708{
1709 struct drm_i915_private *i915 = error->i915;
1710
1711 error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1712 error->suspended = i915->runtime_pm.suspended;
1713
1714 error->iommu = -1;
1715#ifdef CONFIG_INTEL_IOMMU
1716 error->iommu = intel_iommu_gfx_mapped;
1717#endif
1718 error->reset_count = i915_reset_count(&i915->gpu_error);
1719 error->suspend_count = i915->suspend_count;
1720
1721 i915_params_copy(&error->params, &i915->params);
1722 memcpy(&error->device_info,
1723 INTEL_INFO(i915),
1724 sizeof(error->device_info));
1725 memcpy(&error->runtime_info,
1726 RUNTIME_INFO(i915),
1727 sizeof(error->runtime_info));
1728 error->driver_caps = i915->caps;
1729}
1730
1731struct i915_gpu_coredump *
1732i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1733{
1734 struct i915_gpu_coredump *error;
1735
1736 if (!i915->params.error_capture)
1737 return NULL;
1738
1739 error = kzalloc(sizeof(*error), gfp);
1740 if (!error)
1741 return NULL;
1742
1743 kref_init(&error->ref);
1744 error->i915 = i915;
1745
1746 error->time = ktime_get_real();
1747 error->boottime = ktime_get_boottime();
1748 error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1749 error->capture = jiffies;
1750
1751 capture_gen(error);
1752
1753 return error;
1754}
1755
1756#define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1757
1758struct intel_gt_coredump *
1759intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
1760{
1761 struct intel_gt_coredump *gc;
1762
1763 gc = kzalloc(sizeof(*gc), gfp);
1764 if (!gc)
1765 return NULL;
1766
1767 gc->_gt = gt;
1768 gc->awake = intel_gt_pm_is_awake(gt);
1769
1770 gt_record_regs(gc);
1771 gt_record_fences(gc);
1772
1773 return gc;
1774}
1775
1776struct i915_vma_compress *
1777i915_vma_capture_prepare(struct intel_gt_coredump *gt)
1778{
1779 struct i915_vma_compress *compress;
1780
1781 compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
1782 if (!compress)
1783 return NULL;
1784
1785 if (!compress_init(compress)) {
1786 kfree(compress);
1787 return NULL;
1788 }
1789
1790 return compress;
1791}
1792
1793void i915_vma_capture_finish(struct intel_gt_coredump *gt,
1794 struct i915_vma_compress *compress)
1795{
1796 if (!compress)
1797 return;
1798
1799 compress_fini(compress);
1800 kfree(compress);
1801}
1802
1803struct i915_gpu_coredump *
1804i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask)
1805{
1806 struct drm_i915_private *i915 = gt->i915;
1807 struct i915_gpu_coredump *error;
1808
1809 /* Check if GPU capture has been disabled */
1810 error = READ_ONCE(i915->gpu_error.first_error);
1811 if (IS_ERR(error))
1812 return error;
1813
1814 error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
1815 if (!error)
1816 return ERR_PTR(-ENOMEM);
1817
1818 error->gt = intel_gt_coredump_alloc(gt, ALLOW_FAIL);
1819 if (error->gt) {
1820 struct i915_vma_compress *compress;
1821
1822 compress = i915_vma_capture_prepare(error->gt);
1823 if (!compress) {
1824 kfree(error->gt);
1825 kfree(error);
1826 return ERR_PTR(-ENOMEM);
1827 }
1828
1829 gt_record_info(error->gt);
1830 gt_record_engines(error->gt, engine_mask, compress);
1831
1832 if (INTEL_INFO(i915)->has_gt_uc)
1833 error->gt->uc = gt_record_uc(error->gt, compress);
1834
1835 i915_vma_capture_finish(error->gt, compress);
1836
1837 error->simulated |= error->gt->simulated;
1838 }
1839
1840 error->overlay = intel_overlay_capture_error_state(i915);
1841
1842 return error;
1843}
1844
1845void i915_error_state_store(struct i915_gpu_coredump *error)
1846{
1847 struct drm_i915_private *i915;
1848 static bool warned;
1849
1850 if (IS_ERR_OR_NULL(error))
1851 return;
1852
1853 i915 = error->i915;
1854 drm_info(&i915->drm, "%s\n", error_msg(error));
1855
1856 if (error->simulated ||
1857 cmpxchg(&i915->gpu_error.first_error, NULL, error))
1858 return;
1859
1860 i915_gpu_coredump_get(error);
1861
1862 if (!xchg(&warned, true) &&
1863 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1864 pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1865 pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
1866 pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
1867 pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1868 pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1869 pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1870 i915->drm.primary->index);
1871 }
1872}
1873
1874/**
1875 * i915_capture_error_state - capture an error record for later analysis
1876 * @gt: intel_gt which originated the hang
1877 * @engine_mask: hung engines
1878 *
1879 *
1880 * Should be called when an error is detected (either a hang or an error
1881 * interrupt) to capture error state from the time of the error. Fills
1882 * out a structure which becomes available in debugfs for user level tools
1883 * to pick up.
1884 */
1885void i915_capture_error_state(struct intel_gt *gt,
1886 intel_engine_mask_t engine_mask)
1887{
1888 struct i915_gpu_coredump *error;
1889
1890 error = i915_gpu_coredump(gt, engine_mask);
1891 if (IS_ERR(error)) {
1892 cmpxchg(>->i915->gpu_error.first_error, NULL, error);
1893 return;
1894 }
1895
1896 i915_error_state_store(error);
1897 i915_gpu_coredump_put(error);
1898}
1899
1900struct i915_gpu_coredump *
1901i915_first_error_state(struct drm_i915_private *i915)
1902{
1903 struct i915_gpu_coredump *error;
1904
1905 spin_lock_irq(&i915->gpu_error.lock);
1906 error = i915->gpu_error.first_error;
1907 if (!IS_ERR_OR_NULL(error))
1908 i915_gpu_coredump_get(error);
1909 spin_unlock_irq(&i915->gpu_error.lock);
1910
1911 return error;
1912}
1913
1914void i915_reset_error_state(struct drm_i915_private *i915)
1915{
1916 struct i915_gpu_coredump *error;
1917
1918 spin_lock_irq(&i915->gpu_error.lock);
1919 error = i915->gpu_error.first_error;
1920 if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1921 i915->gpu_error.first_error = NULL;
1922 spin_unlock_irq(&i915->gpu_error.lock);
1923
1924 if (!IS_ERR_OR_NULL(error))
1925 i915_gpu_coredump_put(error);
1926}
1927
1928void i915_disable_error_state(struct drm_i915_private *i915, int err)
1929{
1930 spin_lock_irq(&i915->gpu_error.lock);
1931 if (!i915->gpu_error.first_error)
1932 i915->gpu_error.first_error = ERR_PTR(err);
1933 spin_unlock_irq(&i915->gpu_error.lock);
1934}