]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/gpu/drm/i915/i915_gpu_error.c
drm/i915/bxt: Fix off-by-one error in Broxton PLL IDs
[mirror_ubuntu-bionic-kernel.git] / drivers / gpu / drm / i915 / i915_gpu_error.c
CommitLineData
84734a04
MK
1/*
2 * Copyright (c) 2008 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 * Keith Packard <keithp@keithp.com>
26 * Mika Kuoppala <mika.kuoppala@intel.com>
27 *
28 */
29
30#include <generated/utsrelease.h>
31#include "i915_drv.h"
32
84734a04
MK
33static const char *ring_str(int ring)
34{
35 switch (ring) {
36 case RCS: return "render";
37 case VCS: return "bsd";
38 case BCS: return "blt";
39 case VECS: return "vebox";
845f74a7 40 case VCS2: return "bsd2";
84734a04
MK
41 default: return "";
42 }
43}
44
45static const char *pin_flag(int pinned)
46{
47 if (pinned > 0)
48 return " P";
49 else if (pinned < 0)
50 return " p";
51 else
52 return "";
53}
54
55static const char *tiling_flag(int tiling)
56{
57 switch (tiling) {
58 default:
59 case I915_TILING_NONE: return "";
60 case I915_TILING_X: return " X";
61 case I915_TILING_Y: return " Y";
62 }
63}
64
65static const char *dirty_flag(int dirty)
66{
67 return dirty ? " dirty" : "";
68}
69
70static const char *purgeable_flag(int purgeable)
71{
72 return purgeable ? " purgeable" : "";
73}
74
75static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
76{
77
78 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
79 e->err = -ENOSPC;
80 return false;
81 }
82
83 if (e->bytes == e->size - 1 || e->err)
84 return false;
85
86 return true;
87}
88
89static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
90 unsigned len)
91{
92 if (e->pos + len <= e->start) {
93 e->pos += len;
94 return false;
95 }
96
97 /* First vsnprintf needs to fit in its entirety for memmove */
98 if (len >= e->size) {
99 e->err = -EIO;
100 return false;
101 }
102
103 return true;
104}
105
106static void __i915_error_advance(struct drm_i915_error_state_buf *e,
107 unsigned len)
108{
109 /* If this is first printf in this window, adjust it so that
110 * start position matches start of the buffer
111 */
112
113 if (e->pos < e->start) {
114 const size_t off = e->start - e->pos;
115
116 /* Should not happen but be paranoid */
117 if (off > len || e->bytes) {
118 e->err = -EIO;
119 return;
120 }
121
122 memmove(e->buf, e->buf + off, len - off);
123 e->bytes = len - off;
124 e->pos = e->start;
125 return;
126 }
127
128 e->bytes += len;
129 e->pos += len;
130}
131
132static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
133 const char *f, va_list args)
134{
135 unsigned len;
136
137 if (!__i915_error_ok(e))
138 return;
139
140 /* Seek the first printf which is hits start position */
141 if (e->pos < e->start) {
e29bb4eb
CW
142 va_list tmp;
143
144 va_copy(tmp, args);
1d2cb9a5
MK
145 len = vsnprintf(NULL, 0, f, tmp);
146 va_end(tmp);
147
148 if (!__i915_error_seek(e, len))
84734a04
MK
149 return;
150 }
151
152 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
153 if (len >= e->size - e->bytes)
154 len = e->size - e->bytes - 1;
155
156 __i915_error_advance(e, len);
157}
158
159static void i915_error_puts(struct drm_i915_error_state_buf *e,
160 const char *str)
161{
162 unsigned len;
163
164 if (!__i915_error_ok(e))
165 return;
166
167 len = strlen(str);
168
169 /* Seek the first printf which is hits start position */
170 if (e->pos < e->start) {
171 if (!__i915_error_seek(e, len))
172 return;
173 }
174
175 if (len >= e->size - e->bytes)
176 len = e->size - e->bytes - 1;
177 memcpy(e->buf + e->bytes, str, len);
178
179 __i915_error_advance(e, len);
180}
181
182#define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
183#define err_puts(e, s) i915_error_puts(e, s)
184
185static void print_error_buffers(struct drm_i915_error_state_buf *m,
186 const char *name,
187 struct drm_i915_error_buffer *err,
188 int count)
189{
b4716185
CW
190 int i;
191
3a448734 192 err_printf(m, " %s [%d]:\n", name, count);
84734a04
MK
193
194 while (count--) {
e1f12325
MT
195 err_printf(m, " %08x_%08x %8u %02x %02x [ ",
196 upper_32_bits(err->gtt_offset),
197 lower_32_bits(err->gtt_offset),
84734a04
MK
198 err->size,
199 err->read_domains,
b4716185
CW
200 err->write_domain);
201 for (i = 0; i < I915_NUM_RINGS; i++)
202 err_printf(m, "%02x ", err->rseqno[i]);
203
204 err_printf(m, "] %02x", err->wseqno);
84734a04
MK
205 err_puts(m, pin_flag(err->pinned));
206 err_puts(m, tiling_flag(err->tiling));
207 err_puts(m, dirty_flag(err->dirty));
208 err_puts(m, purgeable_flag(err->purgeable));
5cc9ed4b 209 err_puts(m, err->userptr ? " userptr" : "");
84734a04
MK
210 err_puts(m, err->ring != -1 ? " " : "");
211 err_puts(m, ring_str(err->ring));
0a4cd7c8 212 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
84734a04
MK
213
214 if (err->name)
215 err_printf(m, " (name: %d)", err->name);
216 if (err->fence_reg != I915_FENCE_REG_NONE)
217 err_printf(m, " (fence: %d)", err->fence_reg);
218
219 err_puts(m, "\n");
220 err++;
221 }
222}
223
da661464
MK
224static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
225{
226 switch (a) {
227 case HANGCHECK_IDLE:
228 return "idle";
229 case HANGCHECK_WAIT:
230 return "wait";
231 case HANGCHECK_ACTIVE:
232 return "active";
233 case HANGCHECK_KICK:
234 return "kick";
235 case HANGCHECK_HUNG:
236 return "hung";
237 }
238
239 return "unknown";
240}
241
84734a04
MK
242static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
243 struct drm_device *dev,
77c1aa84
DV
244 struct drm_i915_error_state *error,
245 int ring_idx)
84734a04 246{
77c1aa84
DV
247 struct drm_i915_error_ring *ring = &error->ring[ring_idx];
248
362b8af7 249 if (!ring->valid)
372fbb8e
CW
250 return;
251
77c1aa84 252 err_printf(m, "%s command stream:\n", ring_str(ring_idx));
94f8cf10
CW
253 err_printf(m, " START: 0x%08x\n", ring->start);
254 err_printf(m, " HEAD: 0x%08x\n", ring->head);
255 err_printf(m, " TAIL: 0x%08x\n", ring->tail);
256 err_printf(m, " CTL: 0x%08x\n", ring->ctl);
257 err_printf(m, " HWS: 0x%08x\n", ring->hws);
e3243d16 258 err_printf(m, " ACTHD: 0x%08x %08x\n", (u32)(ring->acthd>>32), (u32)ring->acthd);
362b8af7
BW
259 err_printf(m, " IPEIR: 0x%08x\n", ring->ipeir);
260 err_printf(m, " IPEHR: 0x%08x\n", ring->ipehr);
261 err_printf(m, " INSTDONE: 0x%08x\n", ring->instdone);
3dda20a9 262 if (INTEL_INFO(dev)->gen >= 4) {
e3243d16 263 err_printf(m, " BBADDR: 0x%08x %08x\n", (u32)(ring->bbaddr>>32), (u32)ring->bbaddr);
362b8af7
BW
264 err_printf(m, " BB_STATE: 0x%08x\n", ring->bbstate);
265 err_printf(m, " INSTPS: 0x%08x\n", ring->instps);
3dda20a9 266 }
362b8af7 267 err_printf(m, " INSTPM: 0x%08x\n", ring->instpm);
13ffadd1
BW
268 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ring->faddr),
269 lower_32_bits(ring->faddr));
84734a04 270 if (INTEL_INFO(dev)->gen >= 6) {
362b8af7
BW
271 err_printf(m, " RC PSMI: 0x%08x\n", ring->rc_psmi);
272 err_printf(m, " FAULT_REG: 0x%08x\n", ring->fault_reg);
84734a04 273 err_printf(m, " SYNC_0: 0x%08x [last synced 0x%08x]\n",
362b8af7
BW
274 ring->semaphore_mboxes[0],
275 ring->semaphore_seqno[0]);
84734a04 276 err_printf(m, " SYNC_1: 0x%08x [last synced 0x%08x]\n",
362b8af7
BW
277 ring->semaphore_mboxes[1],
278 ring->semaphore_seqno[1]);
4e5aabfd
BW
279 if (HAS_VEBOX(dev)) {
280 err_printf(m, " SYNC_2: 0x%08x [last synced 0x%08x]\n",
362b8af7
BW
281 ring->semaphore_mboxes[2],
282 ring->semaphore_seqno[2]);
4e5aabfd 283 }
84734a04 284 }
6c7a01ec
BW
285 if (USES_PPGTT(dev)) {
286 err_printf(m, " GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
287
288 if (INTEL_INFO(dev)->gen >= 8) {
289 int i;
290 for (i = 0; i < 4; i++)
291 err_printf(m, " PDP%d: 0x%016llx\n",
292 i, ring->vm_info.pdp[i]);
293 } else {
294 err_printf(m, " PP_DIR_BASE: 0x%08x\n",
295 ring->vm_info.pp_dir_base);
296 }
297 }
362b8af7
BW
298 err_printf(m, " seqno: 0x%08x\n", ring->seqno);
299 err_printf(m, " waiting: %s\n", yesno(ring->waiting));
300 err_printf(m, " ring->head: 0x%08x\n", ring->cpu_ring_head);
301 err_printf(m, " ring->tail: 0x%08x\n", ring->cpu_ring_tail);
da661464 302 err_printf(m, " hangcheck: %s [%d]\n",
362b8af7
BW
303 hangcheck_action_to_str(ring->hangcheck_action),
304 ring->hangcheck_score);
84734a04
MK
305}
306
307void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
308{
309 va_list args;
310
311 va_start(args, f);
312 i915_error_vprintf(e, f, args);
313 va_end(args);
314}
315
ab0e7ff9
CW
316static void print_error_obj(struct drm_i915_error_state_buf *m,
317 struct drm_i915_error_object *obj)
318{
319 int page, offset, elt;
320
321 for (page = offset = 0; page < obj->page_count; page++) {
322 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
323 err_printf(m, "%08x : %08x\n", offset,
324 obj->pages[page][elt]);
325 offset += 4;
326 }
327 }
328}
329
84734a04
MK
330int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
331 const struct i915_error_state_file_priv *error_priv)
332{
333 struct drm_device *dev = error_priv->dev;
50227e1c 334 struct drm_i915_private *dev_priv = dev->dev_private;
84734a04 335 struct drm_i915_error_state *error = error_priv->error;
0ca36d78 336 struct drm_i915_error_object *obj;
ab0e7ff9
CW
337 int i, j, offset, elt;
338 int max_hangcheck_score;
84734a04
MK
339
340 if (!error) {
341 err_printf(m, "no error state collected\n");
342 goto out;
343 }
344
cb383002 345 err_printf(m, "%s\n", error->error_msg);
84734a04
MK
346 err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
347 error->time.tv_usec);
348 err_printf(m, "Kernel: " UTS_RELEASE "\n");
ab0e7ff9
CW
349 max_hangcheck_score = 0;
350 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
351 if (error->ring[i].hangcheck_score > max_hangcheck_score)
352 max_hangcheck_score = error->ring[i].hangcheck_score;
353 }
354 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
355 if (error->ring[i].hangcheck_score == max_hangcheck_score &&
356 error->ring[i].pid != -1) {
357 err_printf(m, "Active process (on ring %s): %s [%d]\n",
358 ring_str(i),
359 error->ring[i].comm,
360 error->ring[i].pid);
361 }
362 }
48b031e3 363 err_printf(m, "Reset count: %u\n", error->reset_count);
62d5d69b 364 err_printf(m, "Suspend count: %u\n", error->suspend_count);
ffbab09b 365 err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
06e6ff8f
AS
366 err_printf(m, "PCI Revision: 0x%02x\n", dev->pdev->revision);
367 err_printf(m, "PCI Subsystem: %04x:%04x\n",
368 dev->pdev->subsystem_vendor,
369 dev->pdev->subsystem_device);
eb5be9d0 370 err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
0ac7655c
MK
371
372 if (HAS_CSR(dev)) {
373 struct intel_csr *csr = &dev_priv->csr;
374
375 err_printf(m, "DMC loaded: %s\n",
376 yesno(csr->dmc_payload != NULL));
377 err_printf(m, "DMC fw version: %d.%d\n",
378 CSR_VERSION_MAJOR(csr->version),
379 CSR_VERSION_MINOR(csr->version));
380 }
381
84734a04
MK
382 err_printf(m, "EIR: 0x%08x\n", error->eir);
383 err_printf(m, "IER: 0x%08x\n", error->ier);
885ea5a8
RV
384 if (INTEL_INFO(dev)->gen >= 8) {
385 for (i = 0; i < 4; i++)
386 err_printf(m, "GTIER gt %d: 0x%08x\n", i,
387 error->gtier[i]);
388 } else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
389 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
84734a04
MK
390 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
391 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
392 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
393 err_printf(m, "CCID: 0x%08x\n", error->ccid);
094f9a54 394 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
84734a04
MK
395
396 for (i = 0; i < dev_priv->num_fence_regs; i++)
397 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
398
399 for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
400 err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
401 error->extra_instdone[i]);
402
403 if (INTEL_INFO(dev)->gen >= 6) {
404 err_printf(m, "ERROR: 0x%08x\n", error->error);
6c826f34
MK
405
406 if (INTEL_INFO(dev)->gen >= 8)
407 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
408 error->fault_data1, error->fault_data0);
409
84734a04
MK
410 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
411 }
412
413 if (INTEL_INFO(dev)->gen == 7)
414 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
415
77c1aa84
DV
416 for (i = 0; i < ARRAY_SIZE(error->ring); i++)
417 i915_ring_error_state(m, dev, error, i);
84734a04 418
3a448734
CW
419 for (i = 0; i < error->vm_count; i++) {
420 err_printf(m, "vm[%d]\n", i);
421
84734a04 422 print_error_buffers(m, "Active",
3a448734
CW
423 error->active_bo[i],
424 error->active_bo_count[i]);
84734a04 425
84734a04 426 print_error_buffers(m, "Pinned",
3a448734
CW
427 error->pinned_bo[i],
428 error->pinned_bo_count[i]);
429 }
84734a04
MK
430
431 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
ab0e7ff9
CW
432 obj = error->ring[i].batchbuffer;
433 if (obj) {
434 err_puts(m, dev_priv->ring[i].name);
435 if (error->ring[i].pid != -1)
436 err_printf(m, " (submitted by %s [%d])",
437 error->ring[i].comm,
438 error->ring[i].pid);
e1f12325
MT
439 err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
440 upper_32_bits(obj->gtt_offset),
441 lower_32_bits(obj->gtt_offset));
ab0e7ff9
CW
442 print_error_obj(m, obj);
443 }
444
445 obj = error->ring[i].wa_batchbuffer;
446 if (obj) {
447 err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
e1f12325
MT
448 dev_priv->ring[i].name,
449 lower_32_bits(obj->gtt_offset));
ab0e7ff9 450 print_error_obj(m, obj);
84734a04
MK
451 }
452
453 if (error->ring[i].num_requests) {
454 err_printf(m, "%s --- %d requests\n",
455 dev_priv->ring[i].name,
456 error->ring[i].num_requests);
457 for (j = 0; j < error->ring[i].num_requests; j++) {
458 err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
459 error->ring[i].requests[j].seqno,
460 error->ring[i].requests[j].jiffies,
461 error->ring[i].requests[j].tail);
462 }
463 }
464
465 if ((obj = error->ring[i].ringbuffer)) {
466 err_printf(m, "%s --- ringbuffer = 0x%08x\n",
467 dev_priv->ring[i].name,
e1f12325 468 lower_32_bits(obj->gtt_offset));
ab0e7ff9 469 print_error_obj(m, obj);
84734a04
MK
470 }
471
362b8af7 472 if ((obj = error->ring[i].hws_page)) {
3a5a0393
JB
473 u64 hws_offset = obj->gtt_offset;
474 u32 *hws_page = &obj->pages[0][0];
475
476 if (i915.enable_execlists) {
477 hws_offset += LRC_PPHWSP_PN * PAGE_SIZE;
478 hws_page = &obj->pages[LRC_PPHWSP_PN][0];
479 }
d1675198 480 err_printf(m, "%s --- HW Status = 0x%08llx\n",
3a5a0393 481 dev_priv->ring[i].name, hws_offset);
f3ce3821
CW
482 offset = 0;
483 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
484 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
485 offset,
3a5a0393
JB
486 hws_page[elt],
487 hws_page[elt+1],
488 hws_page[elt+2],
489 hws_page[elt+3]);
f3ce3821
CW
490 offset += 16;
491 }
492 }
493
f85db059 494 obj = error->ring[i].wa_ctx;
495 if (obj) {
496 u64 wa_ctx_offset = obj->gtt_offset;
497 u32 *wa_ctx_page = &obj->pages[0][0];
498 struct intel_engine_cs *ring = &dev_priv->ring[RCS];
499 u32 wa_ctx_size = (ring->wa_ctx.indirect_ctx.size +
500 ring->wa_ctx.per_ctx.size);
501
502 err_printf(m, "%s --- WA ctx batch buffer = 0x%08llx\n",
503 dev_priv->ring[i].name, wa_ctx_offset);
504 offset = 0;
505 for (elt = 0; elt < wa_ctx_size; elt += 4) {
506 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
507 offset,
508 wa_ctx_page[elt + 0],
509 wa_ctx_page[elt + 1],
510 wa_ctx_page[elt + 2],
511 wa_ctx_page[elt + 3]);
512 offset += 16;
513 }
514 }
515
372fbb8e 516 if ((obj = error->ring[i].ctx)) {
84734a04
MK
517 err_printf(m, "%s --- HW Context = 0x%08x\n",
518 dev_priv->ring[i].name,
e1f12325 519 lower_32_bits(obj->gtt_offset));
17d36749 520 print_error_obj(m, obj);
84734a04
MK
521 }
522 }
523
0ca36d78 524 if ((obj = error->semaphore_obj)) {
e1f12325
MT
525 err_printf(m, "Semaphore page = 0x%08x\n",
526 lower_32_bits(obj->gtt_offset));
0ca36d78
BW
527 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
528 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
529 elt * 4,
530 obj->pages[0][elt],
531 obj->pages[0][elt+1],
532 obj->pages[0][elt+2],
533 obj->pages[0][elt+3]);
534 }
535 }
536
84734a04
MK
537 if (error->overlay)
538 intel_overlay_print_error_state(m, error->overlay);
539
540 if (error->display)
541 intel_display_print_error_state(m, dev, error->display);
542
543out:
544 if (m->bytes == 0 && m->err)
545 return m->err;
546
547 return 0;
548}
549
550int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
0a4cd7c8 551 struct drm_i915_private *i915,
84734a04
MK
552 size_t count, loff_t pos)
553{
554 memset(ebuf, 0, sizeof(*ebuf));
0a4cd7c8 555 ebuf->i915 = i915;
84734a04
MK
556
557 /* We need to have enough room to store any i915_error_state printf
558 * so that we can move it to start position.
559 */
560 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
561 ebuf->buf = kmalloc(ebuf->size,
562 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
563
564 if (ebuf->buf == NULL) {
565 ebuf->size = PAGE_SIZE;
566 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
567 }
568
569 if (ebuf->buf == NULL) {
570 ebuf->size = 128;
571 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
572 }
573
574 if (ebuf->buf == NULL)
575 return -ENOMEM;
576
577 ebuf->start = pos;
578
579 return 0;
580}
581
582static void i915_error_object_free(struct drm_i915_error_object *obj)
583{
584 int page;
585
586 if (obj == NULL)
587 return;
588
589 for (page = 0; page < obj->page_count; page++)
590 kfree(obj->pages[page]);
591
592 kfree(obj);
593}
594
595static void i915_error_state_free(struct kref *error_ref)
596{
597 struct drm_i915_error_state *error = container_of(error_ref,
598 typeof(*error), ref);
599 int i;
600
601 for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
602 i915_error_object_free(error->ring[i].batchbuffer);
b3da4a62 603 i915_error_object_free(error->ring[i].wa_batchbuffer);
84734a04 604 i915_error_object_free(error->ring[i].ringbuffer);
362b8af7 605 i915_error_object_free(error->ring[i].hws_page);
84734a04
MK
606 i915_error_object_free(error->ring[i].ctx);
607 kfree(error->ring[i].requests);
f85db059 608 i915_error_object_free(error->ring[i].wa_ctx);
84734a04
MK
609 }
610
0ca36d78 611 i915_error_object_free(error->semaphore_obj);
0b37a9a9
MT
612
613 for (i = 0; i < error->vm_count; i++)
614 kfree(error->active_bo[i]);
615
84734a04 616 kfree(error->active_bo);
0b37a9a9
MT
617 kfree(error->active_bo_count);
618 kfree(error->pinned_bo);
619 kfree(error->pinned_bo_count);
84734a04
MK
620 kfree(error->overlay);
621 kfree(error->display);
622 kfree(error);
623}
624
625static struct drm_i915_error_object *
8ae62dc6
CW
626i915_error_object_create(struct drm_i915_private *dev_priv,
627 struct drm_i915_gem_object *src,
628 struct i915_address_space *vm)
84734a04
MK
629{
630 struct drm_i915_error_object *dst;
aff43766 631 struct i915_vma *vma = NULL;
8ae62dc6 632 int num_pages;
b3c3f5e6
CW
633 bool use_ggtt;
634 int i = 0;
e1f12325 635 u64 reloc_offset;
84734a04
MK
636
637 if (src == NULL || src->pages == NULL)
638 return NULL;
639
8ae62dc6
CW
640 num_pages = src->base.size >> PAGE_SHIFT;
641
84734a04
MK
642 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
643 if (dst == NULL)
644 return NULL;
645
87a01e82
CW
646 if (i915_gem_obj_bound(src, vm))
647 dst->gtt_offset = i915_gem_obj_offset(src, vm);
648 else
649 dst->gtt_offset = -1;
b3c3f5e6
CW
650
651 reloc_offset = dst->gtt_offset;
aff43766
TU
652 if (i915_is_ggtt(vm))
653 vma = i915_gem_obj_to_ggtt(src);
b3c3f5e6 654 use_ggtt = (src->cache_level == I915_CACHE_NONE &&
aff43766
TU
655 vma && (vma->bound & GLOBAL_BIND) &&
656 reloc_offset + num_pages * PAGE_SIZE <= dev_priv->gtt.mappable_end);
b3c3f5e6
CW
657
658 /* Cannot access stolen address directly, try to use the aperture */
659 if (src->stolen) {
660 use_ggtt = true;
661
aff43766 662 if (!(vma && vma->bound & GLOBAL_BIND))
b3c3f5e6
CW
663 goto unwind;
664
665 reloc_offset = i915_gem_obj_ggtt_offset(src);
666 if (reloc_offset + num_pages * PAGE_SIZE > dev_priv->gtt.mappable_end)
667 goto unwind;
668 }
669
670 /* Cannot access snooped pages through the aperture */
671 if (use_ggtt && src->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv->dev))
672 goto unwind;
673
674 dst->page_count = num_pages;
675 while (num_pages--) {
84734a04
MK
676 unsigned long flags;
677 void *d;
678
679 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
680 if (d == NULL)
681 goto unwind;
682
683 local_irq_save(flags);
b3c3f5e6 684 if (use_ggtt) {
84734a04
MK
685 void __iomem *s;
686
687 /* Simply ignore tiling or any overlapping fence.
688 * It's part of the error state, and this hopefully
689 * captures what the GPU read.
690 */
691
692 s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
693 reloc_offset);
694 memcpy_fromio(d, s, PAGE_SIZE);
695 io_mapping_unmap_atomic(s);
84734a04
MK
696 } else {
697 struct page *page;
698 void *s;
699
700 page = i915_gem_object_get_page(src, i);
701
702 drm_clflush_pages(&page, 1);
703
704 s = kmap_atomic(page);
705 memcpy(d, s, PAGE_SIZE);
706 kunmap_atomic(s);
707
708 drm_clflush_pages(&page, 1);
709 }
710 local_irq_restore(flags);
711
b3c3f5e6 712 dst->pages[i++] = d;
84734a04
MK
713 reloc_offset += PAGE_SIZE;
714 }
84734a04
MK
715
716 return dst;
717
718unwind:
719 while (i--)
720 kfree(dst->pages[i]);
721 kfree(dst);
722 return NULL;
723}
a7b91078 724#define i915_error_ggtt_object_create(dev_priv, src) \
8ae62dc6 725 i915_error_object_create((dev_priv), (src), &(dev_priv)->gtt.base)
84734a04
MK
726
727static void capture_bo(struct drm_i915_error_buffer *err,
3a448734 728 struct i915_vma *vma)
84734a04 729{
3a448734 730 struct drm_i915_gem_object *obj = vma->obj;
b4716185 731 int i;
3a448734 732
84734a04
MK
733 err->size = obj->base.size;
734 err->name = obj->base.name;
b4716185
CW
735 for (i = 0; i < I915_NUM_RINGS; i++)
736 err->rseqno[i] = i915_gem_request_get_seqno(obj->last_read_req[i]);
97b2a6a1 737 err->wseqno = i915_gem_request_get_seqno(obj->last_write_req);
3a448734 738 err->gtt_offset = vma->node.start;
84734a04
MK
739 err->read_domains = obj->base.read_domains;
740 err->write_domain = obj->base.write_domain;
741 err->fence_reg = obj->fence_reg;
742 err->pinned = 0;
d7f46fc4 743 if (i915_gem_obj_is_pinned(obj))
84734a04 744 err->pinned = 1;
84734a04
MK
745 err->tiling = obj->tiling_mode;
746 err->dirty = obj->dirty;
747 err->purgeable = obj->madv != I915_MADV_WILLNEED;
5cc9ed4b 748 err->userptr = obj->userptr.mm != NULL;
b4716185
CW
749 err->ring = obj->last_write_req ?
750 i915_gem_request_get_ring(obj->last_write_req)->id : -1;
84734a04
MK
751 err->cache_level = obj->cache_level;
752}
753
754static u32 capture_active_bo(struct drm_i915_error_buffer *err,
755 int count, struct list_head *head)
756{
ca191b13 757 struct i915_vma *vma;
84734a04
MK
758 int i = 0;
759
1c7f4bca 760 list_for_each_entry(vma, head, vm_link) {
3a448734 761 capture_bo(err++, vma);
84734a04
MK
762 if (++i == count)
763 break;
764 }
765
766 return i;
767}
768
769static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
3a448734
CW
770 int count, struct list_head *head,
771 struct i915_address_space *vm)
84734a04
MK
772{
773 struct drm_i915_gem_object *obj;
3a448734
CW
774 struct drm_i915_error_buffer * const first = err;
775 struct drm_i915_error_buffer * const last = err + count;
84734a04
MK
776
777 list_for_each_entry(obj, head, global_list) {
3a448734 778 struct i915_vma *vma;
84734a04 779
3a448734 780 if (err == last)
84734a04 781 break;
3a448734 782
1c7f4bca 783 list_for_each_entry(vma, &obj->vma_list, obj_link)
fe14d5f4 784 if (vma->vm == vm && vma->pin_count > 0)
3a448734 785 capture_bo(err++, vma);
84734a04
MK
786 }
787
3a448734 788 return err - first;
84734a04
MK
789}
790
011cf577
BW
791/* Generate a semi-unique error code. The code is not meant to have meaning, The
792 * code's only purpose is to try to prevent false duplicated bug reports by
793 * grossly estimating a GPU error state.
794 *
795 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
796 * the hang if we could strip the GTT offset information from it.
797 *
798 * It's only a small step better than a random number in its current form.
799 */
800static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
cb383002
MK
801 struct drm_i915_error_state *error,
802 int *ring_id)
011cf577
BW
803{
804 uint32_t error_code = 0;
805 int i;
806
807 /* IPEHR would be an ideal way to detect errors, as it's the gross
808 * measure of "the command that hung." However, has some very common
809 * synchronization commands which almost always appear in the case
810 * strictly a client bug. Use instdone to differentiate those some.
811 */
cb383002
MK
812 for (i = 0; i < I915_NUM_RINGS; i++) {
813 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
814 if (ring_id)
815 *ring_id = i;
816
011cf577 817 return error->ring[i].ipehr ^ error->ring[i].instdone;
cb383002
MK
818 }
819 }
011cf577
BW
820
821 return error_code;
822}
823
84734a04
MK
824static void i915_gem_record_fences(struct drm_device *dev,
825 struct drm_i915_error_state *error)
826{
827 struct drm_i915_private *dev_priv = dev->dev_private;
828 int i;
829
ce38ab05 830 if (IS_GEN3(dev) || IS_GEN2(dev)) {
ce38ab05 831 for (i = 0; i < dev_priv->num_fence_regs; i++)
eecf613a
VS
832 error->fence[i] = I915_READ(FENCE_REG(i));
833 } else if (IS_GEN5(dev) || IS_GEN4(dev)) {
834 for (i = 0; i < dev_priv->num_fence_regs; i++)
835 error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
836 } else if (INTEL_INFO(dev)->gen >= 6) {
837 for (i = 0; i < dev_priv->num_fence_regs; i++)
838 error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
839 }
84734a04
MK
840}
841
87f85ebc 842
0ca36d78
BW
843static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
844 struct drm_i915_error_state *error,
845 struct intel_engine_cs *ring,
846 struct drm_i915_error_ring *ering)
847{
b4558b46 848 struct intel_engine_cs *to;
0ca36d78
BW
849 int i;
850
851 if (!i915_semaphore_is_enabled(dev_priv->dev))
852 return;
853
854 if (!error->semaphore_obj)
855 error->semaphore_obj =
cc1df8a3
DV
856 i915_error_ggtt_object_create(dev_priv,
857 dev_priv->semaphore_obj);
0ca36d78 858
b4558b46
RV
859 for_each_ring(to, dev_priv, i) {
860 int idx;
861 u16 signal_offset;
862 u32 *tmp;
0ca36d78 863
b4558b46
RV
864 if (ring == to)
865 continue;
866
864c6181
RV
867 signal_offset = (GEN8_SIGNAL_OFFSET(ring, i) & (PAGE_SIZE - 1))
868 / 4;
b4558b46
RV
869 tmp = error->semaphore_obj->pages[0];
870 idx = intel_ring_sync_index(ring, to);
871
872 ering->semaphore_mboxes[idx] = tmp[signal_offset];
873 ering->semaphore_seqno[idx] = ring->semaphore.sync_seqno[idx];
0ca36d78
BW
874 }
875}
876
87f85ebc
BW
877static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
878 struct intel_engine_cs *ring,
879 struct drm_i915_error_ring *ering)
880{
881 ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(ring->mmio_base));
882 ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(ring->mmio_base));
883 ering->semaphore_seqno[0] = ring->semaphore.sync_seqno[0];
884 ering->semaphore_seqno[1] = ring->semaphore.sync_seqno[1];
885
886 if (HAS_VEBOX(dev_priv->dev)) {
887 ering->semaphore_mboxes[2] =
888 I915_READ(RING_SYNC_2(ring->mmio_base));
889 ering->semaphore_seqno[2] = ring->semaphore.sync_seqno[2];
890 }
891}
892
84734a04 893static void i915_record_ring_state(struct drm_device *dev,
0ca36d78 894 struct drm_i915_error_state *error,
a4872ba6 895 struct intel_engine_cs *ring,
362b8af7 896 struct drm_i915_error_ring *ering)
84734a04
MK
897{
898 struct drm_i915_private *dev_priv = dev->dev_private;
899
900 if (INTEL_INFO(dev)->gen >= 6) {
3613cf1d 901 ering->rc_psmi = I915_READ(RING_PSMI_CTL(ring->mmio_base));
362b8af7 902 ering->fault_reg = I915_READ(RING_FAULT_REG(ring));
0ca36d78
BW
903 if (INTEL_INFO(dev)->gen >= 8)
904 gen8_record_semaphore_state(dev_priv, error, ring, ering);
905 else
906 gen6_record_semaphore_state(dev_priv, ring, ering);
4e5aabfd
BW
907 }
908
84734a04 909 if (INTEL_INFO(dev)->gen >= 4) {
362b8af7
BW
910 ering->faddr = I915_READ(RING_DMA_FADD(ring->mmio_base));
911 ering->ipeir = I915_READ(RING_IPEIR(ring->mmio_base));
912 ering->ipehr = I915_READ(RING_IPEHR(ring->mmio_base));
913 ering->instdone = I915_READ(RING_INSTDONE(ring->mmio_base));
914 ering->instps = I915_READ(RING_INSTPS(ring->mmio_base));
915 ering->bbaddr = I915_READ(RING_BBADDR(ring->mmio_base));
13ffadd1
BW
916 if (INTEL_INFO(dev)->gen >= 8) {
917 ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(ring->mmio_base)) << 32;
362b8af7 918 ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(ring->mmio_base)) << 32;
13ffadd1 919 }
362b8af7 920 ering->bbstate = I915_READ(RING_BBSTATE(ring->mmio_base));
84734a04 921 } else {
362b8af7
BW
922 ering->faddr = I915_READ(DMA_FADD_I8XX);
923 ering->ipeir = I915_READ(IPEIR);
924 ering->ipehr = I915_READ(IPEHR);
bd93a50e 925 ering->instdone = I915_READ(GEN2_INSTDONE);
84734a04
MK
926 }
927
362b8af7
BW
928 ering->waiting = waitqueue_active(&ring->irq_queue);
929 ering->instpm = I915_READ(RING_INSTPM(ring->mmio_base));
930 ering->seqno = ring->get_seqno(ring, false);
931 ering->acthd = intel_ring_get_active_head(ring);
94f8cf10 932 ering->start = I915_READ_START(ring);
362b8af7
BW
933 ering->head = I915_READ_HEAD(ring);
934 ering->tail = I915_READ_TAIL(ring);
935 ering->ctl = I915_READ_CTL(ring);
84734a04 936
f3ce3821 937 if (I915_NEED_GFX_HWS(dev)) {
f0f59a00 938 i915_reg_t mmio;
f3ce3821
CW
939
940 if (IS_GEN7(dev)) {
941 switch (ring->id) {
942 default:
943 case RCS:
944 mmio = RENDER_HWS_PGA_GEN7;
945 break;
946 case BCS:
947 mmio = BLT_HWS_PGA_GEN7;
948 break;
949 case VCS:
950 mmio = BSD_HWS_PGA_GEN7;
951 break;
952 case VECS:
953 mmio = VEBOX_HWS_PGA_GEN7;
954 break;
955 }
956 } else if (IS_GEN6(ring->dev)) {
957 mmio = RING_HWS_PGA_GEN6(ring->mmio_base);
958 } else {
959 /* XXX: gen8 returns to sanity */
960 mmio = RING_HWS_PGA(ring->mmio_base);
961 }
962
362b8af7 963 ering->hws = I915_READ(mmio);
f3ce3821
CW
964 }
965
362b8af7
BW
966 ering->hangcheck_score = ring->hangcheck.score;
967 ering->hangcheck_action = ring->hangcheck.action;
6c7a01ec
BW
968
969 if (USES_PPGTT(dev)) {
970 int i;
971
972 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(ring));
973
74745b09
RV
974 if (IS_GEN6(dev))
975 ering->vm_info.pp_dir_base =
976 I915_READ(RING_PP_DIR_BASE_READ(ring));
977 else if (IS_GEN7(dev))
978 ering->vm_info.pp_dir_base =
979 I915_READ(RING_PP_DIR_BASE(ring));
980 else if (INTEL_INFO(dev)->gen >= 8)
6c7a01ec
BW
981 for (i = 0; i < 4; i++) {
982 ering->vm_info.pdp[i] =
983 I915_READ(GEN8_RING_PDP_UDW(ring, i));
984 ering->vm_info.pdp[i] <<= 32;
985 ering->vm_info.pdp[i] |=
986 I915_READ(GEN8_RING_PDP_LDW(ring, i));
987 }
6c7a01ec 988 }
84734a04
MK
989}
990
991
a4872ba6 992static void i915_gem_record_active_context(struct intel_engine_cs *ring,
84734a04
MK
993 struct drm_i915_error_state *error,
994 struct drm_i915_error_ring *ering)
995{
996 struct drm_i915_private *dev_priv = ring->dev->dev_private;
997 struct drm_i915_gem_object *obj;
998
999 /* Currently render ring is the only HW context user */
1000 if (ring->id != RCS || !error->ccid)
1001 return;
1002
1003 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
36362ad3
BW
1004 if (!i915_gem_obj_ggtt_bound(obj))
1005 continue;
1006
84734a04 1007 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
17d36749 1008 ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
84734a04
MK
1009 break;
1010 }
1011 }
1012}
1013
1014static void i915_gem_record_rings(struct drm_device *dev,
1015 struct drm_i915_error_state *error)
1016{
1017 struct drm_i915_private *dev_priv = dev->dev_private;
84734a04
MK
1018 struct drm_i915_gem_request *request;
1019 int i, count;
1020
372fbb8e 1021 for (i = 0; i < I915_NUM_RINGS; i++) {
a4872ba6 1022 struct intel_engine_cs *ring = &dev_priv->ring[i];
9075e52f 1023 struct intel_ringbuffer *rbuf;
372fbb8e 1024
eee73b46
CW
1025 error->ring[i].pid = -1;
1026
372fbb8e
CW
1027 if (ring->dev == NULL)
1028 continue;
1029
1030 error->ring[i].valid = true;
1031
0ca36d78 1032 i915_record_ring_state(dev, error, ring, &error->ring[i]);
84734a04 1033
ab0e7ff9
CW
1034 request = i915_gem_find_active_request(ring);
1035 if (request) {
ae6c4806
DV
1036 struct i915_address_space *vm;
1037
1038 vm = request->ctx && request->ctx->ppgtt ?
1039 &request->ctx->ppgtt->base :
1040 &dev_priv->gtt.base;
1041
ab0e7ff9
CW
1042 /* We need to copy these to an anonymous buffer
1043 * as the simplest method to avoid being overwritten
1044 * by userspace.
1045 */
1046 error->ring[i].batchbuffer =
1047 i915_error_object_create(dev_priv,
1048 request->batch_obj,
ae6c4806 1049 vm);
ab0e7ff9 1050
8ae62dc6 1051 if (HAS_BROKEN_CS_TLB(dev_priv->dev))
ab0e7ff9
CW
1052 error->ring[i].wa_batchbuffer =
1053 i915_error_ggtt_object_create(dev_priv,
1054 ring->scratch.obj);
1055
071c92de 1056 if (request->pid) {
ab0e7ff9
CW
1057 struct task_struct *task;
1058
1059 rcu_read_lock();
071c92de 1060 task = pid_task(request->pid, PIDTYPE_PID);
ab0e7ff9
CW
1061 if (task) {
1062 strcpy(error->ring[i].comm, task->comm);
1063 error->ring[i].pid = task->pid;
1064 }
1065 rcu_read_unlock();
1066 }
1067 }
84734a04 1068
9075e52f
OM
1069 if (i915.enable_execlists) {
1070 /* TODO: This is only a small fix to keep basic error
1071 * capture working, but we need to add more information
1072 * for it to be useful (e.g. dump the context being
1073 * executed).
1074 */
1075 if (request)
1076 rbuf = request->ctx->engine[ring->id].ringbuf;
1077 else
ed54c1a1 1078 rbuf = dev_priv->kernel_context->engine[ring->id].ringbuf;
9075e52f
OM
1079 } else
1080 rbuf = ring->buffer;
1081
1082 error->ring[i].cpu_ring_head = rbuf->head;
1083 error->ring[i].cpu_ring_tail = rbuf->tail;
1084
84734a04 1085 error->ring[i].ringbuffer =
9075e52f 1086 i915_error_ggtt_object_create(dev_priv, rbuf->obj);
84734a04 1087
8ae62dc6
CW
1088 error->ring[i].hws_page =
1089 i915_error_ggtt_object_create(dev_priv, ring->status_page.obj);
84734a04 1090
f85db059 1091 if (ring->wa_ctx.obj) {
1092 error->ring[i].wa_ctx =
1093 i915_error_ggtt_object_create(dev_priv,
1094 ring->wa_ctx.obj);
1095 }
1096
84734a04
MK
1097 i915_gem_record_active_context(ring, error, &error->ring[i]);
1098
1099 count = 0;
1100 list_for_each_entry(request, &ring->request_list, list)
1101 count++;
1102
1103 error->ring[i].num_requests = count;
1104 error->ring[i].requests =
a1e22653 1105 kcalloc(count, sizeof(*error->ring[i].requests),
84734a04
MK
1106 GFP_ATOMIC);
1107 if (error->ring[i].requests == NULL) {
1108 error->ring[i].num_requests = 0;
1109 continue;
1110 }
1111
1112 count = 0;
1113 list_for_each_entry(request, &ring->request_list, list) {
1114 struct drm_i915_error_request *erq;
1115
9c8e1bdb
TE
1116 if (count >= error->ring[i].num_requests) {
1117 /*
1118 * If the ring request list was changed in
1119 * between the point where the error request
1120 * list was created and dimensioned and this
1121 * point then just exit early to avoid crashes.
1122 *
1123 * We don't need to communicate that the
1124 * request list changed state during error
1125 * state capture and that the error state is
1126 * slightly incorrect as a consequence since we
1127 * are typically only interested in the request
1128 * list state at the point of error state
1129 * capture, not in any changes happening during
1130 * the capture.
1131 */
1132 break;
1133 }
1134
84734a04
MK
1135 erq = &error->ring[i].requests[count++];
1136 erq->seqno = request->seqno;
1137 erq->jiffies = request->emitted_jiffies;
72f95afa 1138 erq->tail = request->postfix;
84734a04
MK
1139 }
1140 }
1141}
1142
95f5301d
BW
1143/* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1144 * VM.
1145 */
1146static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1147 struct drm_i915_error_state *error,
1148 struct i915_address_space *vm,
1149 const int ndx)
84734a04 1150{
95f5301d 1151 struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
84734a04 1152 struct drm_i915_gem_object *obj;
95f5301d 1153 struct i915_vma *vma;
84734a04
MK
1154 int i;
1155
1156 i = 0;
1c7f4bca 1157 list_for_each_entry(vma, &vm->active_list, vm_link)
84734a04 1158 i++;
95f5301d 1159 error->active_bo_count[ndx] = i;
3a448734
CW
1160
1161 list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1c7f4bca 1162 list_for_each_entry(vma, &obj->vma_list, obj_link)
fe14d5f4 1163 if (vma->vm == vm && vma->pin_count > 0)
3a448734 1164 i++;
3a448734 1165 }
95f5301d 1166 error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
84734a04
MK
1167
1168 if (i) {
a1e22653 1169 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
95f5301d
BW
1170 if (active_bo)
1171 pinned_bo = active_bo + error->active_bo_count[ndx];
84734a04
MK
1172 }
1173
95f5301d
BW
1174 if (active_bo)
1175 error->active_bo_count[ndx] =
1176 capture_active_bo(active_bo,
1177 error->active_bo_count[ndx],
5cef07e1 1178 &vm->active_list);
84734a04 1179
95f5301d
BW
1180 if (pinned_bo)
1181 error->pinned_bo_count[ndx] =
1182 capture_pinned_bo(pinned_bo,
1183 error->pinned_bo_count[ndx],
3a448734 1184 &dev_priv->mm.bound_list, vm);
95f5301d
BW
1185 error->active_bo[ndx] = active_bo;
1186 error->pinned_bo[ndx] = pinned_bo;
1187}
1188
1189static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1190 struct drm_i915_error_state *error)
1191{
1192 struct i915_address_space *vm;
1193 int cnt = 0, i = 0;
1194
1195 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1196 cnt++;
1197
95f5301d
BW
1198 error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1199 error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1200 error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1201 GFP_ATOMIC);
1202 error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1203 GFP_ATOMIC);
1204
3a448734
CW
1205 if (error->active_bo == NULL ||
1206 error->pinned_bo == NULL ||
1207 error->active_bo_count == NULL ||
1208 error->pinned_bo_count == NULL) {
1209 kfree(error->active_bo);
1210 kfree(error->active_bo_count);
1211 kfree(error->pinned_bo);
1212 kfree(error->pinned_bo_count);
1213
1214 error->active_bo = NULL;
1215 error->active_bo_count = NULL;
1216 error->pinned_bo = NULL;
1217 error->pinned_bo_count = NULL;
1218 } else {
1219 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1220 i915_gem_capture_vm(dev_priv, error, vm, i++);
1221
1222 error->vm_count = cnt;
1223 }
84734a04
MK
1224}
1225
1d762aad
BW
1226/* Capture all registers which don't fit into another category. */
1227static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1228 struct drm_i915_error_state *error)
84734a04 1229{
1d762aad 1230 struct drm_device *dev = dev_priv->dev;
885ea5a8 1231 int i;
84734a04 1232
654c90c6
BW
1233 /* General organization
1234 * 1. Registers specific to a single generation
1235 * 2. Registers which belong to multiple generations
1236 * 3. Feature specific registers.
1237 * 4. Everything else
1238 * Please try to follow the order.
1239 */
84734a04 1240
654c90c6
BW
1241 /* 1: Registers specific to a single generation */
1242 if (IS_VALLEYVIEW(dev)) {
885ea5a8 1243 error->gtier[0] = I915_READ(GTIER);
843db716 1244 error->ier = I915_READ(VLV_IER);
40181697 1245 error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
654c90c6 1246 }
84734a04 1247
654c90c6
BW
1248 if (IS_GEN7(dev))
1249 error->err_int = I915_READ(GEN7_ERR_INT);
84734a04 1250
6c826f34
MK
1251 if (INTEL_INFO(dev)->gen >= 8) {
1252 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1253 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1254 }
1255
91ec5d11 1256 if (IS_GEN6(dev)) {
40181697 1257 error->forcewake = I915_READ_FW(FORCEWAKE);
91ec5d11
BW
1258 error->gab_ctl = I915_READ(GAB_CTL);
1259 error->gfx_mode = I915_READ(GFX_MODE);
1260 }
84734a04 1261
654c90c6
BW
1262 /* 2: Registers which belong to multiple generations */
1263 if (INTEL_INFO(dev)->gen >= 7)
40181697 1264 error->forcewake = I915_READ_FW(FORCEWAKE_MT);
84734a04
MK
1265
1266 if (INTEL_INFO(dev)->gen >= 6) {
654c90c6 1267 error->derrmr = I915_READ(DERRMR);
84734a04
MK
1268 error->error = I915_READ(ERROR_GEN6);
1269 error->done_reg = I915_READ(DONE_REG);
1270 }
1271
654c90c6 1272 /* 3: Feature specific registers */
91ec5d11
BW
1273 if (IS_GEN6(dev) || IS_GEN7(dev)) {
1274 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1275 error->gac_eco = I915_READ(GAC_ECO_BITS);
1276 }
1277
1278 /* 4: Everything else */
654c90c6
BW
1279 if (HAS_HW_CONTEXTS(dev))
1280 error->ccid = I915_READ(CCID);
1281
885ea5a8
RV
1282 if (INTEL_INFO(dev)->gen >= 8) {
1283 error->ier = I915_READ(GEN8_DE_MISC_IER);
1284 for (i = 0; i < 4; i++)
1285 error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1286 } else if (HAS_PCH_SPLIT(dev)) {
843db716 1287 error->ier = I915_READ(DEIER);
885ea5a8 1288 error->gtier[0] = I915_READ(GTIER);
843db716
RV
1289 } else if (IS_GEN2(dev)) {
1290 error->ier = I915_READ16(IER);
1291 } else if (!IS_VALLEYVIEW(dev)) {
1292 error->ier = I915_READ(IER);
654c90c6 1293 }
654c90c6
BW
1294 error->eir = I915_READ(EIR);
1295 error->pgtbl_er = I915_READ(PGTBL_ER);
84734a04
MK
1296
1297 i915_get_extra_instdone(dev, error->extra_instdone);
1d762aad
BW
1298}
1299
cb383002 1300static void i915_error_capture_msg(struct drm_device *dev,
58174462
MK
1301 struct drm_i915_error_state *error,
1302 bool wedged,
1303 const char *error_msg)
cb383002
MK
1304{
1305 struct drm_i915_private *dev_priv = dev->dev_private;
1306 u32 ecode;
58174462 1307 int ring_id = -1, len;
cb383002
MK
1308
1309 ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1310
58174462 1311 len = scnprintf(error->error_msg, sizeof(error->error_msg),
0b5492d6
MK
1312 "GPU HANG: ecode %d:%d:0x%08x",
1313 INTEL_INFO(dev)->gen, ring_id, ecode);
58174462
MK
1314
1315 if (ring_id != -1 && error->ring[ring_id].pid != -1)
1316 len += scnprintf(error->error_msg + len,
1317 sizeof(error->error_msg) - len,
1318 ", in %s [%d]",
1319 error->ring[ring_id].comm,
1320 error->ring[ring_id].pid);
1321
1322 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1323 ", reason: %s, action: %s",
1324 error_msg,
1325 wedged ? "reset" : "continue");
cb383002
MK
1326}
1327
48b031e3
MK
1328static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1329 struct drm_i915_error_state *error)
1330{
eb5be9d0
CW
1331 error->iommu = -1;
1332#ifdef CONFIG_INTEL_IOMMU
1333 error->iommu = intel_iommu_gfx_mapped;
1334#endif
48b031e3 1335 error->reset_count = i915_reset_count(&dev_priv->gpu_error);
62d5d69b 1336 error->suspend_count = dev_priv->suspend_count;
48b031e3
MK
1337}
1338
1d762aad
BW
1339/**
1340 * i915_capture_error_state - capture an error record for later analysis
1341 * @dev: drm device
1342 *
1343 * Should be called when an error is detected (either a hang or an error
1344 * interrupt) to capture error state from the time of the error. Fills
1345 * out a structure which becomes available in debugfs for user level tools
1346 * to pick up.
1347 */
58174462
MK
1348void i915_capture_error_state(struct drm_device *dev, bool wedged,
1349 const char *error_msg)
1d762aad 1350{
53a4c6b2 1351 static bool warned;
1d762aad
BW
1352 struct drm_i915_private *dev_priv = dev->dev_private;
1353 struct drm_i915_error_state *error;
1354 unsigned long flags;
1d762aad
BW
1355
1356 /* Account for pipe specific data like PIPE*STAT */
1357 error = kzalloc(sizeof(*error), GFP_ATOMIC);
1358 if (!error) {
1359 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1360 return;
1361 }
1362
011cf577
BW
1363 kref_init(&error->ref);
1364
48b031e3 1365 i915_capture_gen_state(dev_priv, error);
011cf577
BW
1366 i915_capture_reg_state(dev_priv, error);
1367 i915_gem_capture_buffers(dev_priv, error);
1368 i915_gem_record_fences(dev, error);
1369 i915_gem_record_rings(dev, error);
1d762aad 1370
84734a04
MK
1371 do_gettimeofday(&error->time);
1372
1373 error->overlay = intel_overlay_capture_error_state(dev);
1374 error->display = intel_display_capture_error_state(dev);
1375
58174462 1376 i915_error_capture_msg(dev, error, wedged, error_msg);
cb383002
MK
1377 DRM_INFO("%s\n", error->error_msg);
1378
84734a04
MK
1379 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1380 if (dev_priv->gpu_error.first_error == NULL) {
1381 dev_priv->gpu_error.first_error = error;
1382 error = NULL;
1383 }
1384 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1385
cb383002 1386 if (error) {
84734a04 1387 i915_error_state_free(&error->ref);
cb383002
MK
1388 return;
1389 }
1390
1391 if (!warned) {
1392 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1393 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1394 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1395 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1396 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1397 warned = true;
1398 }
84734a04
MK
1399}
1400
1401void i915_error_state_get(struct drm_device *dev,
1402 struct i915_error_state_file_priv *error_priv)
1403{
1404 struct drm_i915_private *dev_priv = dev->dev_private;
84734a04 1405
5b254c59 1406 spin_lock_irq(&dev_priv->gpu_error.lock);
84734a04
MK
1407 error_priv->error = dev_priv->gpu_error.first_error;
1408 if (error_priv->error)
1409 kref_get(&error_priv->error->ref);
5b254c59 1410 spin_unlock_irq(&dev_priv->gpu_error.lock);
84734a04
MK
1411
1412}
1413
1414void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1415{
1416 if (error_priv->error)
1417 kref_put(&error_priv->error->ref, i915_error_state_free);
1418}
1419
1420void i915_destroy_error_state(struct drm_device *dev)
1421{
1422 struct drm_i915_private *dev_priv = dev->dev_private;
1423 struct drm_i915_error_state *error;
84734a04 1424
5b254c59 1425 spin_lock_irq(&dev_priv->gpu_error.lock);
84734a04
MK
1426 error = dev_priv->gpu_error.first_error;
1427 dev_priv->gpu_error.first_error = NULL;
5b254c59 1428 spin_unlock_irq(&dev_priv->gpu_error.lock);
84734a04
MK
1429
1430 if (error)
1431 kref_put(&error->ref, i915_error_state_free);
1432}
1433
0a4cd7c8 1434const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
84734a04
MK
1435{
1436 switch (type) {
1437 case I915_CACHE_NONE: return " uncached";
0a4cd7c8 1438 case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
350ec881 1439 case I915_CACHE_L3_LLC: return " L3+LLC";
f56383cb 1440 case I915_CACHE_WT: return " WT";
84734a04
MK
1441 default: return "";
1442 }
1443}
1444
1445/* NB: please notice the memset */
1446void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1447{
1448 struct drm_i915_private *dev_priv = dev->dev_private;
1449 memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1450
563f94f6 1451 if (IS_GEN2(dev) || IS_GEN3(dev))
bd93a50e 1452 instdone[0] = I915_READ(GEN2_INSTDONE);
563f94f6 1453 else if (IS_GEN4(dev) || IS_GEN5(dev) || IS_GEN6(dev)) {
f1d54348 1454 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
13d70b81 1455 instdone[1] = I915_READ(GEN4_INSTDONE1);
563f94f6 1456 } else if (INTEL_INFO(dev)->gen >= 7) {
f1d54348 1457 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
84734a04
MK
1458 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1459 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1460 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
84734a04
MK
1461 }
1462}