i915_gpu_error.c 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053
  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. #include <generated/utsrelease.h>
  30. #include "i915_drv.h"
  31. static const char *yesno(int v)
  32. {
  33. return v ? "yes" : "no";
  34. }
  35. static const char *ring_str(int ring)
  36. {
  37. switch (ring) {
  38. case RCS: return "render";
  39. case VCS: return "bsd";
  40. case BCS: return "blt";
  41. case VECS: return "vebox";
  42. default: return "";
  43. }
  44. }
  45. static const char *pin_flag(int pinned)
  46. {
  47. if (pinned > 0)
  48. return " P";
  49. else if (pinned < 0)
  50. return " p";
  51. else
  52. return "";
  53. }
  54. static const char *tiling_flag(int tiling)
  55. {
  56. switch (tiling) {
  57. default:
  58. case I915_TILING_NONE: return "";
  59. case I915_TILING_X: return " X";
  60. case I915_TILING_Y: return " Y";
  61. }
  62. }
  63. static const char *dirty_flag(int dirty)
  64. {
  65. return dirty ? " dirty" : "";
  66. }
  67. static const char *purgeable_flag(int purgeable)
  68. {
  69. return purgeable ? " purgeable" : "";
  70. }
  71. static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
  72. {
  73. if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
  74. e->err = -ENOSPC;
  75. return false;
  76. }
  77. if (e->bytes == e->size - 1 || e->err)
  78. return false;
  79. return true;
  80. }
  81. static 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. /* First vsnprintf needs to fit in its entirety for memmove */
  89. if (len >= e->size) {
  90. e->err = -EIO;
  91. return false;
  92. }
  93. return true;
  94. }
  95. static void __i915_error_advance(struct drm_i915_error_state_buf *e,
  96. unsigned len)
  97. {
  98. /* If this is first printf in this window, adjust it so that
  99. * start position matches start of the buffer
  100. */
  101. if (e->pos < e->start) {
  102. const size_t off = e->start - e->pos;
  103. /* Should not happen but be paranoid */
  104. if (off > len || e->bytes) {
  105. e->err = -EIO;
  106. return;
  107. }
  108. memmove(e->buf, e->buf + off, len - off);
  109. e->bytes = len - off;
  110. e->pos = e->start;
  111. return;
  112. }
  113. e->bytes += len;
  114. e->pos += len;
  115. }
  116. static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
  117. const char *f, va_list args)
  118. {
  119. unsigned len;
  120. if (!__i915_error_ok(e))
  121. return;
  122. /* Seek the first printf which is hits start position */
  123. if (e->pos < e->start) {
  124. va_list tmp;
  125. va_copy(tmp, args);
  126. if (!__i915_error_seek(e, vsnprintf(NULL, 0, f, tmp)))
  127. return;
  128. }
  129. len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
  130. if (len >= e->size - e->bytes)
  131. len = e->size - e->bytes - 1;
  132. __i915_error_advance(e, len);
  133. }
  134. static void i915_error_puts(struct drm_i915_error_state_buf *e,
  135. const char *str)
  136. {
  137. unsigned len;
  138. if (!__i915_error_ok(e))
  139. return;
  140. len = strlen(str);
  141. /* Seek the first printf which is hits start position */
  142. if (e->pos < e->start) {
  143. if (!__i915_error_seek(e, len))
  144. return;
  145. }
  146. if (len >= e->size - e->bytes)
  147. len = e->size - e->bytes - 1;
  148. memcpy(e->buf + e->bytes, str, len);
  149. __i915_error_advance(e, len);
  150. }
  151. #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
  152. #define err_puts(e, s) i915_error_puts(e, s)
  153. static void print_error_buffers(struct drm_i915_error_state_buf *m,
  154. const char *name,
  155. struct drm_i915_error_buffer *err,
  156. int count)
  157. {
  158. err_printf(m, "%s [%d]:\n", name, count);
  159. while (count--) {
  160. err_printf(m, " %08x %8u %02x %02x %x %x",
  161. err->gtt_offset,
  162. err->size,
  163. err->read_domains,
  164. err->write_domain,
  165. err->rseqno, err->wseqno);
  166. err_puts(m, pin_flag(err->pinned));
  167. err_puts(m, tiling_flag(err->tiling));
  168. err_puts(m, dirty_flag(err->dirty));
  169. err_puts(m, purgeable_flag(err->purgeable));
  170. err_puts(m, err->ring != -1 ? " " : "");
  171. err_puts(m, ring_str(err->ring));
  172. err_puts(m, i915_cache_level_str(err->cache_level));
  173. if (err->name)
  174. err_printf(m, " (name: %d)", err->name);
  175. if (err->fence_reg != I915_FENCE_REG_NONE)
  176. err_printf(m, " (fence: %d)", err->fence_reg);
  177. err_puts(m, "\n");
  178. err++;
  179. }
  180. }
  181. static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
  182. {
  183. switch (a) {
  184. case HANGCHECK_IDLE:
  185. return "idle";
  186. case HANGCHECK_WAIT:
  187. return "wait";
  188. case HANGCHECK_ACTIVE:
  189. return "active";
  190. case HANGCHECK_KICK:
  191. return "kick";
  192. case HANGCHECK_HUNG:
  193. return "hung";
  194. }
  195. return "unknown";
  196. }
  197. static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
  198. struct drm_device *dev,
  199. struct drm_i915_error_state *error,
  200. unsigned ring)
  201. {
  202. BUG_ON(ring >= I915_NUM_RINGS); /* shut up confused gcc */
  203. err_printf(m, "%s command stream:\n", ring_str(ring));
  204. err_printf(m, " HEAD: 0x%08x\n", error->head[ring]);
  205. err_printf(m, " TAIL: 0x%08x\n", error->tail[ring]);
  206. err_printf(m, " CTL: 0x%08x\n", error->ctl[ring]);
  207. err_printf(m, " ACTHD: 0x%08x\n", error->acthd[ring]);
  208. err_printf(m, " IPEIR: 0x%08x\n", error->ipeir[ring]);
  209. err_printf(m, " IPEHR: 0x%08x\n", error->ipehr[ring]);
  210. err_printf(m, " INSTDONE: 0x%08x\n", error->instdone[ring]);
  211. if (ring == RCS && INTEL_INFO(dev)->gen >= 4)
  212. err_printf(m, " BBADDR: 0x%08llx\n", error->bbaddr);
  213. if (INTEL_INFO(dev)->gen >= 4)
  214. err_printf(m, " BB_STATE: 0x%08x\n", error->bbstate[ring]);
  215. if (INTEL_INFO(dev)->gen >= 4)
  216. err_printf(m, " INSTPS: 0x%08x\n", error->instps[ring]);
  217. err_printf(m, " INSTPM: 0x%08x\n", error->instpm[ring]);
  218. err_printf(m, " FADDR: 0x%08x\n", error->faddr[ring]);
  219. if (INTEL_INFO(dev)->gen >= 6) {
  220. err_printf(m, " RC PSMI: 0x%08x\n", error->rc_psmi[ring]);
  221. err_printf(m, " FAULT_REG: 0x%08x\n", error->fault_reg[ring]);
  222. err_printf(m, " SYNC_0: 0x%08x [last synced 0x%08x]\n",
  223. error->semaphore_mboxes[ring][0],
  224. error->semaphore_seqno[ring][0]);
  225. err_printf(m, " SYNC_1: 0x%08x [last synced 0x%08x]\n",
  226. error->semaphore_mboxes[ring][1],
  227. error->semaphore_seqno[ring][1]);
  228. if (HAS_VEBOX(dev)) {
  229. err_printf(m, " SYNC_2: 0x%08x [last synced 0x%08x]\n",
  230. error->semaphore_mboxes[ring][2],
  231. error->semaphore_seqno[ring][2]);
  232. }
  233. }
  234. err_printf(m, " seqno: 0x%08x\n", error->seqno[ring]);
  235. err_printf(m, " waiting: %s\n", yesno(error->waiting[ring]));
  236. err_printf(m, " ring->head: 0x%08x\n", error->cpu_ring_head[ring]);
  237. err_printf(m, " ring->tail: 0x%08x\n", error->cpu_ring_tail[ring]);
  238. err_printf(m, " hangcheck: %s [%d]\n",
  239. hangcheck_action_to_str(error->hangcheck_action[ring]),
  240. error->hangcheck_score[ring]);
  241. }
  242. void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
  243. {
  244. va_list args;
  245. va_start(args, f);
  246. i915_error_vprintf(e, f, args);
  247. va_end(args);
  248. }
  249. int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
  250. const struct i915_error_state_file_priv *error_priv)
  251. {
  252. struct drm_device *dev = error_priv->dev;
  253. drm_i915_private_t *dev_priv = dev->dev_private;
  254. struct drm_i915_error_state *error = error_priv->error;
  255. struct intel_ring_buffer *ring;
  256. int i, j, page, offset, elt;
  257. if (!error) {
  258. err_printf(m, "no error state collected\n");
  259. goto out;
  260. }
  261. err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
  262. error->time.tv_usec);
  263. err_printf(m, "Kernel: " UTS_RELEASE "\n");
  264. err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
  265. err_printf(m, "EIR: 0x%08x\n", error->eir);
  266. err_printf(m, "IER: 0x%08x\n", error->ier);
  267. err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
  268. err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
  269. err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
  270. err_printf(m, "CCID: 0x%08x\n", error->ccid);
  271. err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
  272. for (i = 0; i < dev_priv->num_fence_regs; i++)
  273. err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]);
  274. for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
  275. err_printf(m, " INSTDONE_%d: 0x%08x\n", i,
  276. error->extra_instdone[i]);
  277. if (INTEL_INFO(dev)->gen >= 6) {
  278. err_printf(m, "ERROR: 0x%08x\n", error->error);
  279. err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
  280. }
  281. if (INTEL_INFO(dev)->gen == 7)
  282. err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
  283. for_each_ring(ring, dev_priv, i)
  284. i915_ring_error_state(m, dev, error, i);
  285. if (error->active_bo)
  286. print_error_buffers(m, "Active",
  287. error->active_bo[0],
  288. error->active_bo_count[0]);
  289. if (error->pinned_bo)
  290. print_error_buffers(m, "Pinned",
  291. error->pinned_bo[0],
  292. error->pinned_bo_count[0]);
  293. for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
  294. struct drm_i915_error_object *obj;
  295. if ((obj = error->ring[i].batchbuffer)) {
  296. err_printf(m, "%s --- gtt_offset = 0x%08x\n",
  297. dev_priv->ring[i].name,
  298. obj->gtt_offset);
  299. offset = 0;
  300. for (page = 0; page < obj->page_count; page++) {
  301. for (elt = 0; elt < PAGE_SIZE/4; elt++) {
  302. err_printf(m, "%08x : %08x\n", offset,
  303. obj->pages[page][elt]);
  304. offset += 4;
  305. }
  306. }
  307. }
  308. if (error->ring[i].num_requests) {
  309. err_printf(m, "%s --- %d requests\n",
  310. dev_priv->ring[i].name,
  311. error->ring[i].num_requests);
  312. for (j = 0; j < error->ring[i].num_requests; j++) {
  313. err_printf(m, " seqno 0x%08x, emitted %ld, tail 0x%08x\n",
  314. error->ring[i].requests[j].seqno,
  315. error->ring[i].requests[j].jiffies,
  316. error->ring[i].requests[j].tail);
  317. }
  318. }
  319. if ((obj = error->ring[i].ringbuffer)) {
  320. err_printf(m, "%s --- ringbuffer = 0x%08x\n",
  321. dev_priv->ring[i].name,
  322. obj->gtt_offset);
  323. offset = 0;
  324. for (page = 0; page < obj->page_count; page++) {
  325. for (elt = 0; elt < PAGE_SIZE/4; elt++) {
  326. err_printf(m, "%08x : %08x\n",
  327. offset,
  328. obj->pages[page][elt]);
  329. offset += 4;
  330. }
  331. }
  332. }
  333. obj = error->ring[i].ctx;
  334. if (obj) {
  335. err_printf(m, "%s --- HW Context = 0x%08x\n",
  336. dev_priv->ring[i].name,
  337. obj->gtt_offset);
  338. offset = 0;
  339. for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
  340. err_printf(m, "[%04x] %08x %08x %08x %08x\n",
  341. offset,
  342. obj->pages[0][elt],
  343. obj->pages[0][elt+1],
  344. obj->pages[0][elt+2],
  345. obj->pages[0][elt+3]);
  346. offset += 16;
  347. }
  348. }
  349. }
  350. if (error->overlay)
  351. intel_overlay_print_error_state(m, error->overlay);
  352. if (error->display)
  353. intel_display_print_error_state(m, dev, error->display);
  354. out:
  355. if (m->bytes == 0 && m->err)
  356. return m->err;
  357. return 0;
  358. }
  359. int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
  360. size_t count, loff_t pos)
  361. {
  362. memset(ebuf, 0, sizeof(*ebuf));
  363. /* We need to have enough room to store any i915_error_state printf
  364. * so that we can move it to start position.
  365. */
  366. ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
  367. ebuf->buf = kmalloc(ebuf->size,
  368. GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
  369. if (ebuf->buf == NULL) {
  370. ebuf->size = PAGE_SIZE;
  371. ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
  372. }
  373. if (ebuf->buf == NULL) {
  374. ebuf->size = 128;
  375. ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
  376. }
  377. if (ebuf->buf == NULL)
  378. return -ENOMEM;
  379. ebuf->start = pos;
  380. return 0;
  381. }
  382. static void i915_error_object_free(struct drm_i915_error_object *obj)
  383. {
  384. int page;
  385. if (obj == NULL)
  386. return;
  387. for (page = 0; page < obj->page_count; page++)
  388. kfree(obj->pages[page]);
  389. kfree(obj);
  390. }
  391. static void i915_error_state_free(struct kref *error_ref)
  392. {
  393. struct drm_i915_error_state *error = container_of(error_ref,
  394. typeof(*error), ref);
  395. int i;
  396. for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
  397. i915_error_object_free(error->ring[i].batchbuffer);
  398. i915_error_object_free(error->ring[i].ringbuffer);
  399. i915_error_object_free(error->ring[i].ctx);
  400. kfree(error->ring[i].requests);
  401. }
  402. kfree(error->active_bo);
  403. kfree(error->overlay);
  404. kfree(error->display);
  405. kfree(error);
  406. }
  407. static struct drm_i915_error_object *
  408. i915_error_object_create_sized(struct drm_i915_private *dev_priv,
  409. struct drm_i915_gem_object *src,
  410. const int num_pages)
  411. {
  412. struct drm_i915_error_object *dst;
  413. int i;
  414. u32 reloc_offset;
  415. if (src == NULL || src->pages == NULL)
  416. return NULL;
  417. dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
  418. if (dst == NULL)
  419. return NULL;
  420. reloc_offset = dst->gtt_offset = i915_gem_obj_ggtt_offset(src);
  421. for (i = 0; i < num_pages; i++) {
  422. unsigned long flags;
  423. void *d;
  424. d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
  425. if (d == NULL)
  426. goto unwind;
  427. local_irq_save(flags);
  428. if (reloc_offset < dev_priv->gtt.mappable_end &&
  429. src->has_global_gtt_mapping) {
  430. void __iomem *s;
  431. /* Simply ignore tiling or any overlapping fence.
  432. * It's part of the error state, and this hopefully
  433. * captures what the GPU read.
  434. */
  435. s = io_mapping_map_atomic_wc(dev_priv->gtt.mappable,
  436. reloc_offset);
  437. memcpy_fromio(d, s, PAGE_SIZE);
  438. io_mapping_unmap_atomic(s);
  439. } else if (src->stolen) {
  440. unsigned long offset;
  441. offset = dev_priv->mm.stolen_base;
  442. offset += src->stolen->start;
  443. offset += i << PAGE_SHIFT;
  444. memcpy_fromio(d, (void __iomem *) offset, PAGE_SIZE);
  445. } else {
  446. struct page *page;
  447. void *s;
  448. page = i915_gem_object_get_page(src, i);
  449. drm_clflush_pages(&page, 1);
  450. s = kmap_atomic(page);
  451. memcpy(d, s, PAGE_SIZE);
  452. kunmap_atomic(s);
  453. drm_clflush_pages(&page, 1);
  454. }
  455. local_irq_restore(flags);
  456. dst->pages[i] = d;
  457. reloc_offset += PAGE_SIZE;
  458. }
  459. dst->page_count = num_pages;
  460. return dst;
  461. unwind:
  462. while (i--)
  463. kfree(dst->pages[i]);
  464. kfree(dst);
  465. return NULL;
  466. }
  467. #define i915_error_object_create(dev_priv, src) \
  468. i915_error_object_create_sized((dev_priv), (src), \
  469. (src)->base.size>>PAGE_SHIFT)
  470. static void capture_bo(struct drm_i915_error_buffer *err,
  471. struct drm_i915_gem_object *obj)
  472. {
  473. err->size = obj->base.size;
  474. err->name = obj->base.name;
  475. err->rseqno = obj->last_read_seqno;
  476. err->wseqno = obj->last_write_seqno;
  477. err->gtt_offset = i915_gem_obj_ggtt_offset(obj);
  478. err->read_domains = obj->base.read_domains;
  479. err->write_domain = obj->base.write_domain;
  480. err->fence_reg = obj->fence_reg;
  481. err->pinned = 0;
  482. if (obj->pin_count > 0)
  483. err->pinned = 1;
  484. if (obj->user_pin_count > 0)
  485. err->pinned = -1;
  486. err->tiling = obj->tiling_mode;
  487. err->dirty = obj->dirty;
  488. err->purgeable = obj->madv != I915_MADV_WILLNEED;
  489. err->ring = obj->ring ? obj->ring->id : -1;
  490. err->cache_level = obj->cache_level;
  491. }
  492. static u32 capture_active_bo(struct drm_i915_error_buffer *err,
  493. int count, struct list_head *head)
  494. {
  495. struct i915_vma *vma;
  496. int i = 0;
  497. list_for_each_entry(vma, head, mm_list) {
  498. capture_bo(err++, vma->obj);
  499. if (++i == count)
  500. break;
  501. }
  502. return i;
  503. }
  504. static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
  505. int count, struct list_head *head)
  506. {
  507. struct drm_i915_gem_object *obj;
  508. int i = 0;
  509. list_for_each_entry(obj, head, global_list) {
  510. if (obj->pin_count == 0)
  511. continue;
  512. capture_bo(err++, obj);
  513. if (++i == count)
  514. break;
  515. }
  516. return i;
  517. }
  518. static void i915_gem_record_fences(struct drm_device *dev,
  519. struct drm_i915_error_state *error)
  520. {
  521. struct drm_i915_private *dev_priv = dev->dev_private;
  522. int i;
  523. /* Fences */
  524. switch (INTEL_INFO(dev)->gen) {
  525. case 7:
  526. case 6:
  527. for (i = 0; i < dev_priv->num_fence_regs; i++)
  528. error->fence[i] = I915_READ64(FENCE_REG_SANDYBRIDGE_0 + (i * 8));
  529. break;
  530. case 5:
  531. case 4:
  532. for (i = 0; i < 16; i++)
  533. error->fence[i] = I915_READ64(FENCE_REG_965_0 + (i * 8));
  534. break;
  535. case 3:
  536. if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
  537. for (i = 0; i < 8; i++)
  538. error->fence[i+8] = I915_READ(FENCE_REG_945_8 + (i * 4));
  539. case 2:
  540. for (i = 0; i < 8; i++)
  541. error->fence[i] = I915_READ(FENCE_REG_830_0 + (i * 4));
  542. break;
  543. default:
  544. BUG();
  545. }
  546. }
  547. static struct drm_i915_error_object *
  548. i915_error_first_batchbuffer(struct drm_i915_private *dev_priv,
  549. struct intel_ring_buffer *ring)
  550. {
  551. struct i915_address_space *vm;
  552. struct i915_vma *vma;
  553. struct drm_i915_gem_object *obj;
  554. u32 seqno;
  555. if (!ring->get_seqno)
  556. return NULL;
  557. if (HAS_BROKEN_CS_TLB(dev_priv->dev)) {
  558. u32 acthd = I915_READ(ACTHD);
  559. if (WARN_ON(ring->id != RCS))
  560. return NULL;
  561. obj = ring->scratch.obj;
  562. if (acthd >= i915_gem_obj_ggtt_offset(obj) &&
  563. acthd < i915_gem_obj_ggtt_offset(obj) + obj->base.size)
  564. return i915_error_object_create(dev_priv, obj);
  565. }
  566. seqno = ring->get_seqno(ring, false);
  567. list_for_each_entry(vm, &dev_priv->vm_list, global_link) {
  568. list_for_each_entry(vma, &vm->active_list, mm_list) {
  569. obj = vma->obj;
  570. if (obj->ring != ring)
  571. continue;
  572. if (i915_seqno_passed(seqno, obj->last_read_seqno))
  573. continue;
  574. if ((obj->base.read_domains & I915_GEM_DOMAIN_COMMAND) == 0)
  575. continue;
  576. /* We need to copy these to an anonymous buffer as the simplest
  577. * method to avoid being overwritten by userspace.
  578. */
  579. return i915_error_object_create(dev_priv, obj);
  580. }
  581. }
  582. return NULL;
  583. }
  584. static void i915_record_ring_state(struct drm_device *dev,
  585. struct drm_i915_error_state *error,
  586. struct intel_ring_buffer *ring)
  587. {
  588. struct drm_i915_private *dev_priv = dev->dev_private;
  589. if (INTEL_INFO(dev)->gen >= 6) {
  590. error->rc_psmi[ring->id] = I915_READ(ring->mmio_base + 0x50);
  591. error->fault_reg[ring->id] = I915_READ(RING_FAULT_REG(ring));
  592. error->semaphore_mboxes[ring->id][0]
  593. = I915_READ(RING_SYNC_0(ring->mmio_base));
  594. error->semaphore_mboxes[ring->id][1]
  595. = I915_READ(RING_SYNC_1(ring->mmio_base));
  596. error->semaphore_seqno[ring->id][0] = ring->sync_seqno[0];
  597. error->semaphore_seqno[ring->id][1] = ring->sync_seqno[1];
  598. }
  599. if (HAS_VEBOX(dev)) {
  600. error->semaphore_mboxes[ring->id][2] =
  601. I915_READ(RING_SYNC_2(ring->mmio_base));
  602. error->semaphore_seqno[ring->id][2] = ring->sync_seqno[2];
  603. }
  604. if (INTEL_INFO(dev)->gen >= 4) {
  605. error->faddr[ring->id] = I915_READ(RING_DMA_FADD(ring->mmio_base));
  606. error->ipeir[ring->id] = I915_READ(RING_IPEIR(ring->mmio_base));
  607. error->ipehr[ring->id] = I915_READ(RING_IPEHR(ring->mmio_base));
  608. error->instdone[ring->id] = I915_READ(RING_INSTDONE(ring->mmio_base));
  609. error->instps[ring->id] = I915_READ(RING_INSTPS(ring->mmio_base));
  610. if (ring->id == RCS)
  611. error->bbaddr = I915_READ64(BB_ADDR);
  612. error->bbstate[ring->id] = I915_READ(RING_BBSTATE(ring->mmio_base));
  613. } else {
  614. error->faddr[ring->id] = I915_READ(DMA_FADD_I8XX);
  615. error->ipeir[ring->id] = I915_READ(IPEIR);
  616. error->ipehr[ring->id] = I915_READ(IPEHR);
  617. error->instdone[ring->id] = I915_READ(INSTDONE);
  618. }
  619. error->waiting[ring->id] = waitqueue_active(&ring->irq_queue);
  620. error->instpm[ring->id] = I915_READ(RING_INSTPM(ring->mmio_base));
  621. error->seqno[ring->id] = ring->get_seqno(ring, false);
  622. error->acthd[ring->id] = intel_ring_get_active_head(ring);
  623. error->head[ring->id] = I915_READ_HEAD(ring);
  624. error->tail[ring->id] = I915_READ_TAIL(ring);
  625. error->ctl[ring->id] = I915_READ_CTL(ring);
  626. error->cpu_ring_head[ring->id] = ring->head;
  627. error->cpu_ring_tail[ring->id] = ring->tail;
  628. error->hangcheck_score[ring->id] = ring->hangcheck.score;
  629. error->hangcheck_action[ring->id] = ring->hangcheck.action;
  630. }
  631. static void i915_gem_record_active_context(struct intel_ring_buffer *ring,
  632. struct drm_i915_error_state *error,
  633. struct drm_i915_error_ring *ering)
  634. {
  635. struct drm_i915_private *dev_priv = ring->dev->dev_private;
  636. struct drm_i915_gem_object *obj;
  637. /* Currently render ring is the only HW context user */
  638. if (ring->id != RCS || !error->ccid)
  639. return;
  640. list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
  641. if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
  642. ering->ctx = i915_error_object_create_sized(dev_priv,
  643. obj, 1);
  644. break;
  645. }
  646. }
  647. }
  648. static void i915_gem_record_rings(struct drm_device *dev,
  649. struct drm_i915_error_state *error)
  650. {
  651. struct drm_i915_private *dev_priv = dev->dev_private;
  652. struct intel_ring_buffer *ring;
  653. struct drm_i915_gem_request *request;
  654. int i, count;
  655. for_each_ring(ring, dev_priv, i) {
  656. i915_record_ring_state(dev, error, ring);
  657. error->ring[i].batchbuffer =
  658. i915_error_first_batchbuffer(dev_priv, ring);
  659. error->ring[i].ringbuffer =
  660. i915_error_object_create(dev_priv, ring->obj);
  661. i915_gem_record_active_context(ring, error, &error->ring[i]);
  662. count = 0;
  663. list_for_each_entry(request, &ring->request_list, list)
  664. count++;
  665. error->ring[i].num_requests = count;
  666. error->ring[i].requests =
  667. kcalloc(count, sizeof(*error->ring[i].requests),
  668. GFP_ATOMIC);
  669. if (error->ring[i].requests == NULL) {
  670. error->ring[i].num_requests = 0;
  671. continue;
  672. }
  673. count = 0;
  674. list_for_each_entry(request, &ring->request_list, list) {
  675. struct drm_i915_error_request *erq;
  676. erq = &error->ring[i].requests[count++];
  677. erq->seqno = request->seqno;
  678. erq->jiffies = request->emitted_jiffies;
  679. erq->tail = request->tail;
  680. }
  681. }
  682. }
  683. /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
  684. * VM.
  685. */
  686. static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
  687. struct drm_i915_error_state *error,
  688. struct i915_address_space *vm,
  689. const int ndx)
  690. {
  691. struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
  692. struct drm_i915_gem_object *obj;
  693. struct i915_vma *vma;
  694. int i;
  695. i = 0;
  696. list_for_each_entry(vma, &vm->active_list, mm_list)
  697. i++;
  698. error->active_bo_count[ndx] = i;
  699. list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
  700. if (obj->pin_count)
  701. i++;
  702. error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
  703. if (i) {
  704. active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
  705. if (active_bo)
  706. pinned_bo = active_bo + error->active_bo_count[ndx];
  707. }
  708. if (active_bo)
  709. error->active_bo_count[ndx] =
  710. capture_active_bo(active_bo,
  711. error->active_bo_count[ndx],
  712. &vm->active_list);
  713. if (pinned_bo)
  714. error->pinned_bo_count[ndx] =
  715. capture_pinned_bo(pinned_bo,
  716. error->pinned_bo_count[ndx],
  717. &dev_priv->mm.bound_list);
  718. error->active_bo[ndx] = active_bo;
  719. error->pinned_bo[ndx] = pinned_bo;
  720. }
  721. static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
  722. struct drm_i915_error_state *error)
  723. {
  724. struct i915_address_space *vm;
  725. int cnt = 0, i = 0;
  726. list_for_each_entry(vm, &dev_priv->vm_list, global_link)
  727. cnt++;
  728. if (WARN(cnt > 1, "Multiple VMs not yet supported\n"))
  729. cnt = 1;
  730. vm = &dev_priv->gtt.base;
  731. error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
  732. error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
  733. error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
  734. GFP_ATOMIC);
  735. error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
  736. GFP_ATOMIC);
  737. list_for_each_entry(vm, &dev_priv->vm_list, global_link)
  738. i915_gem_capture_vm(dev_priv, error, vm, i++);
  739. }
  740. /**
  741. * i915_capture_error_state - capture an error record for later analysis
  742. * @dev: drm device
  743. *
  744. * Should be called when an error is detected (either a hang or an error
  745. * interrupt) to capture error state from the time of the error. Fills
  746. * out a structure which becomes available in debugfs for user level tools
  747. * to pick up.
  748. */
  749. void i915_capture_error_state(struct drm_device *dev)
  750. {
  751. struct drm_i915_private *dev_priv = dev->dev_private;
  752. struct drm_i915_error_state *error;
  753. unsigned long flags;
  754. int pipe;
  755. spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
  756. error = dev_priv->gpu_error.first_error;
  757. spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
  758. if (error)
  759. return;
  760. /* Account for pipe specific data like PIPE*STAT */
  761. error = kzalloc(sizeof(*error), GFP_ATOMIC);
  762. if (!error) {
  763. DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
  764. return;
  765. }
  766. DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
  767. dev->primary->index);
  768. DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
  769. DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
  770. DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
  771. DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
  772. kref_init(&error->ref);
  773. error->eir = I915_READ(EIR);
  774. error->pgtbl_er = I915_READ(PGTBL_ER);
  775. if (HAS_HW_CONTEXTS(dev))
  776. error->ccid = I915_READ(CCID);
  777. if (HAS_PCH_SPLIT(dev))
  778. error->ier = I915_READ(DEIER) | I915_READ(GTIER);
  779. else if (IS_VALLEYVIEW(dev))
  780. error->ier = I915_READ(GTIER) | I915_READ(VLV_IER);
  781. else if (IS_GEN2(dev))
  782. error->ier = I915_READ16(IER);
  783. else
  784. error->ier = I915_READ(IER);
  785. if (INTEL_INFO(dev)->gen >= 6)
  786. error->derrmr = I915_READ(DERRMR);
  787. if (IS_VALLEYVIEW(dev))
  788. error->forcewake = I915_READ(FORCEWAKE_VLV);
  789. else if (INTEL_INFO(dev)->gen >= 7)
  790. error->forcewake = I915_READ(FORCEWAKE_MT);
  791. else if (INTEL_INFO(dev)->gen == 6)
  792. error->forcewake = I915_READ(FORCEWAKE);
  793. if (!HAS_PCH_SPLIT(dev))
  794. for_each_pipe(pipe)
  795. error->pipestat[pipe] = I915_READ(PIPESTAT(pipe));
  796. if (INTEL_INFO(dev)->gen >= 6) {
  797. error->error = I915_READ(ERROR_GEN6);
  798. error->done_reg = I915_READ(DONE_REG);
  799. }
  800. if (INTEL_INFO(dev)->gen == 7)
  801. error->err_int = I915_READ(GEN7_ERR_INT);
  802. i915_get_extra_instdone(dev, error->extra_instdone);
  803. i915_gem_capture_buffers(dev_priv, error);
  804. i915_gem_record_fences(dev, error);
  805. i915_gem_record_rings(dev, error);
  806. do_gettimeofday(&error->time);
  807. error->overlay = intel_overlay_capture_error_state(dev);
  808. error->display = intel_display_capture_error_state(dev);
  809. spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
  810. if (dev_priv->gpu_error.first_error == NULL) {
  811. dev_priv->gpu_error.first_error = error;
  812. error = NULL;
  813. }
  814. spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
  815. if (error)
  816. i915_error_state_free(&error->ref);
  817. }
  818. void i915_error_state_get(struct drm_device *dev,
  819. struct i915_error_state_file_priv *error_priv)
  820. {
  821. struct drm_i915_private *dev_priv = dev->dev_private;
  822. unsigned long flags;
  823. spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
  824. error_priv->error = dev_priv->gpu_error.first_error;
  825. if (error_priv->error)
  826. kref_get(&error_priv->error->ref);
  827. spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
  828. }
  829. void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
  830. {
  831. if (error_priv->error)
  832. kref_put(&error_priv->error->ref, i915_error_state_free);
  833. }
  834. void i915_destroy_error_state(struct drm_device *dev)
  835. {
  836. struct drm_i915_private *dev_priv = dev->dev_private;
  837. struct drm_i915_error_state *error;
  838. unsigned long flags;
  839. spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
  840. error = dev_priv->gpu_error.first_error;
  841. dev_priv->gpu_error.first_error = NULL;
  842. spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
  843. if (error)
  844. kref_put(&error->ref, i915_error_state_free);
  845. }
  846. const char *i915_cache_level_str(int type)
  847. {
  848. switch (type) {
  849. case I915_CACHE_NONE: return " uncached";
  850. case I915_CACHE_LLC: return " snooped or LLC";
  851. case I915_CACHE_L3_LLC: return " L3+LLC";
  852. case I915_CACHE_WT: return " WT";
  853. default: return "";
  854. }
  855. }
  856. /* NB: please notice the memset */
  857. void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
  858. {
  859. struct drm_i915_private *dev_priv = dev->dev_private;
  860. memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
  861. switch (INTEL_INFO(dev)->gen) {
  862. case 2:
  863. case 3:
  864. instdone[0] = I915_READ(INSTDONE);
  865. break;
  866. case 4:
  867. case 5:
  868. case 6:
  869. instdone[0] = I915_READ(INSTDONE_I965);
  870. instdone[1] = I915_READ(INSTDONE1);
  871. break;
  872. default:
  873. WARN_ONCE(1, "Unsupported platform\n");
  874. case 7:
  875. instdone[0] = I915_READ(GEN7_INSTDONE_1);
  876. instdone[1] = I915_READ(GEN7_SC_INSTDONE);
  877. instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
  878. instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
  879. break;
  880. }
  881. }