radeon_ring.c 23 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841
  1. /*
  2. * Copyright 2008 Advanced Micro Devices, Inc.
  3. * Copyright 2008 Red Hat Inc.
  4. * Copyright 2009 Jerome Glisse.
  5. *
  6. * Permission is hereby granted, free of charge, to any person obtaining a
  7. * copy of this software and associated documentation files (the "Software"),
  8. * to deal in the Software without restriction, including without limitation
  9. * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10. * and/or sell copies of the Software, and to permit persons to whom the
  11. * Software is furnished to do so, subject to the following conditions:
  12. *
  13. * The above copyright notice and this permission notice shall be included in
  14. * all copies or substantial portions of the Software.
  15. *
  16. * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17. * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18. * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19. * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20. * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21. * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22. * OTHER DEALINGS IN THE SOFTWARE.
  23. *
  24. * Authors: Dave Airlie
  25. * Alex Deucher
  26. * Jerome Glisse
  27. * Christian König
  28. */
  29. #include <linux/seq_file.h>
  30. #include <linux/slab.h>
  31. #include "drmP.h"
  32. #include "radeon_drm.h"
  33. #include "radeon_reg.h"
  34. #include "radeon.h"
  35. #include "atom.h"
  36. /*
  37. * IB
  38. * IBs (Indirect Buffers) and areas of GPU accessible memory where
  39. * commands are stored. You can put a pointer to the IB in the
  40. * command ring and the hw will fetch the commands from the IB
  41. * and execute them. Generally userspace acceleration drivers
  42. * produce command buffers which are send to the kernel and
  43. * put in IBs for execution by the requested ring.
  44. */
  45. int radeon_debugfs_sa_init(struct radeon_device *rdev);
  46. /**
  47. * radeon_ib_get - request an IB (Indirect Buffer)
  48. *
  49. * @rdev: radeon_device pointer
  50. * @ring: ring index the IB is associated with
  51. * @ib: IB object returned
  52. * @size: requested IB size
  53. *
  54. * Request an IB (all asics). IBs are allocated using the
  55. * suballocator.
  56. * Returns 0 on success, error on failure.
  57. */
  58. int radeon_ib_get(struct radeon_device *rdev, int ring,
  59. struct radeon_ib *ib, struct radeon_vm *vm,
  60. unsigned size)
  61. {
  62. int i, r;
  63. r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true);
  64. if (r) {
  65. dev_err(rdev->dev, "failed to get a new IB (%d)\n", r);
  66. return r;
  67. }
  68. r = radeon_semaphore_create(rdev, &ib->semaphore);
  69. if (r) {
  70. return r;
  71. }
  72. ib->ring = ring;
  73. ib->fence = NULL;
  74. ib->ptr = radeon_sa_bo_cpu_addr(ib->sa_bo);
  75. ib->vm = vm;
  76. if (vm) {
  77. /* ib pool is bind at 0 in virtual address space,
  78. * so gpu_addr is the offset inside the pool bo
  79. */
  80. ib->gpu_addr = ib->sa_bo->soffset;
  81. } else {
  82. ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo);
  83. }
  84. ib->is_const_ib = false;
  85. for (i = 0; i < RADEON_NUM_RINGS; ++i)
  86. ib->sync_to[i] = NULL;
  87. return 0;
  88. }
  89. /**
  90. * radeon_ib_free - free an IB (Indirect Buffer)
  91. *
  92. * @rdev: radeon_device pointer
  93. * @ib: IB object to free
  94. *
  95. * Free an IB (all asics).
  96. */
  97. void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
  98. {
  99. radeon_semaphore_free(rdev, &ib->semaphore, ib->fence);
  100. radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence);
  101. radeon_fence_unref(&ib->fence);
  102. }
  103. /**
  104. * radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring
  105. *
  106. * @rdev: radeon_device pointer
  107. * @ib: IB object to schedule
  108. * @const_ib: Const IB to schedule (SI only)
  109. *
  110. * Schedule an IB on the associated ring (all asics).
  111. * Returns 0 on success, error on failure.
  112. *
  113. * On SI, there are two parallel engines fed from the primary ring,
  114. * the CE (Constant Engine) and the DE (Drawing Engine). Since
  115. * resource descriptors have moved to memory, the CE allows you to
  116. * prime the caches while the DE is updating register state so that
  117. * the resource descriptors will be already in cache when the draw is
  118. * processed. To accomplish this, the userspace driver submits two
  119. * IBs, one for the CE and one for the DE. If there is a CE IB (called
  120. * a CONST_IB), it will be put on the ring prior to the DE IB. Prior
  121. * to SI there was just a DE IB.
  122. */
  123. int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
  124. struct radeon_ib *const_ib)
  125. {
  126. struct radeon_ring *ring = &rdev->ring[ib->ring];
  127. bool need_sync = false;
  128. int i, r = 0;
  129. if (!ib->length_dw || !ring->ready) {
  130. /* TODO: Nothings in the ib we should report. */
  131. dev_err(rdev->dev, "couldn't schedule ib\n");
  132. return -EINVAL;
  133. }
  134. /* 64 dwords should be enough for fence too */
  135. r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_RINGS * 8);
  136. if (r) {
  137. dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
  138. return r;
  139. }
  140. for (i = 0; i < RADEON_NUM_RINGS; ++i) {
  141. struct radeon_fence *fence = ib->sync_to[i];
  142. if (radeon_fence_need_sync(fence, ib->ring)) {
  143. need_sync = true;
  144. radeon_semaphore_sync_rings(rdev, ib->semaphore,
  145. fence->ring, ib->ring);
  146. radeon_fence_note_sync(fence, ib->ring);
  147. }
  148. }
  149. /* immediately free semaphore when we don't need to sync */
  150. if (!need_sync) {
  151. radeon_semaphore_free(rdev, &ib->semaphore, NULL);
  152. }
  153. if (const_ib) {
  154. radeon_ring_ib_execute(rdev, const_ib->ring, const_ib);
  155. radeon_semaphore_free(rdev, &const_ib->semaphore, NULL);
  156. }
  157. radeon_ring_ib_execute(rdev, ib->ring, ib);
  158. r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
  159. if (r) {
  160. dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r);
  161. radeon_ring_unlock_undo(rdev, ring);
  162. return r;
  163. }
  164. if (const_ib) {
  165. const_ib->fence = radeon_fence_ref(ib->fence);
  166. }
  167. radeon_ring_unlock_commit(rdev, ring);
  168. return 0;
  169. }
  170. /**
  171. * radeon_ib_pool_init - Init the IB (Indirect Buffer) pool
  172. *
  173. * @rdev: radeon_device pointer
  174. *
  175. * Initialize the suballocator to manage a pool of memory
  176. * for use as IBs (all asics).
  177. * Returns 0 on success, error on failure.
  178. */
  179. int radeon_ib_pool_init(struct radeon_device *rdev)
  180. {
  181. int r;
  182. if (rdev->ib_pool_ready) {
  183. return 0;
  184. }
  185. r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo,
  186. RADEON_IB_POOL_SIZE*64*1024,
  187. RADEON_GEM_DOMAIN_GTT);
  188. if (r) {
  189. return r;
  190. }
  191. r = radeon_sa_bo_manager_start(rdev, &rdev->ring_tmp_bo);
  192. if (r) {
  193. return r;
  194. }
  195. rdev->ib_pool_ready = true;
  196. if (radeon_debugfs_sa_init(rdev)) {
  197. dev_err(rdev->dev, "failed to register debugfs file for SA\n");
  198. }
  199. return 0;
  200. }
  201. /**
  202. * radeon_ib_pool_fini - Free the IB (Indirect Buffer) pool
  203. *
  204. * @rdev: radeon_device pointer
  205. *
  206. * Tear down the suballocator managing the pool of memory
  207. * for use as IBs (all asics).
  208. */
  209. void radeon_ib_pool_fini(struct radeon_device *rdev)
  210. {
  211. if (rdev->ib_pool_ready) {
  212. radeon_sa_bo_manager_suspend(rdev, &rdev->ring_tmp_bo);
  213. radeon_sa_bo_manager_fini(rdev, &rdev->ring_tmp_bo);
  214. rdev->ib_pool_ready = false;
  215. }
  216. }
  217. /**
  218. * radeon_ib_ring_tests - test IBs on the rings
  219. *
  220. * @rdev: radeon_device pointer
  221. *
  222. * Test an IB (Indirect Buffer) on each ring.
  223. * If the test fails, disable the ring.
  224. * Returns 0 on success, error if the primary GFX ring
  225. * IB test fails.
  226. */
  227. int radeon_ib_ring_tests(struct radeon_device *rdev)
  228. {
  229. unsigned i;
  230. int r;
  231. for (i = 0; i < RADEON_NUM_RINGS; ++i) {
  232. struct radeon_ring *ring = &rdev->ring[i];
  233. if (!ring->ready)
  234. continue;
  235. r = radeon_ib_test(rdev, i, ring);
  236. if (r) {
  237. ring->ready = false;
  238. if (i == RADEON_RING_TYPE_GFX_INDEX) {
  239. /* oh, oh, that's really bad */
  240. DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r);
  241. rdev->accel_working = false;
  242. return r;
  243. } else {
  244. /* still not good, but we can live with it */
  245. DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r);
  246. }
  247. }
  248. }
  249. return 0;
  250. }
  251. /*
  252. * Rings
  253. * Most engines on the GPU are fed via ring buffers. Ring
  254. * buffers are areas of GPU accessible memory that the host
  255. * writes commands into and the GPU reads commands out of.
  256. * There is a rptr (read pointer) that determines where the
  257. * GPU is currently reading, and a wptr (write pointer)
  258. * which determines where the host has written. When the
  259. * pointers are equal, the ring is idle. When the host
  260. * writes commands to the ring buffer, it increments the
  261. * wptr. The GPU then starts fetching commands and executes
  262. * them until the pointers are equal again.
  263. */
  264. int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
  265. /**
  266. * radeon_ring_write - write a value to the ring
  267. *
  268. * @ring: radeon_ring structure holding ring information
  269. * @v: dword (dw) value to write
  270. *
  271. * Write a value to the requested ring buffer (all asics).
  272. */
  273. void radeon_ring_write(struct radeon_ring *ring, uint32_t v)
  274. {
  275. #if DRM_DEBUG_CODE
  276. if (ring->count_dw <= 0) {
  277. DRM_ERROR("radeon: writting more dword to ring than expected !\n");
  278. }
  279. #endif
  280. ring->ring[ring->wptr++] = v;
  281. ring->wptr &= ring->ptr_mask;
  282. ring->count_dw--;
  283. ring->ring_free_dw--;
  284. }
  285. /**
  286. * radeon_ring_supports_scratch_reg - check if the ring supports
  287. * writing to scratch registers
  288. *
  289. * @rdev: radeon_device pointer
  290. * @ring: radeon_ring structure holding ring information
  291. *
  292. * Check if a specific ring supports writing to scratch registers (all asics).
  293. * Returns true if the ring supports writing to scratch regs, false if not.
  294. */
  295. bool radeon_ring_supports_scratch_reg(struct radeon_device *rdev,
  296. struct radeon_ring *ring)
  297. {
  298. switch (ring->idx) {
  299. case RADEON_RING_TYPE_GFX_INDEX:
  300. case CAYMAN_RING_TYPE_CP1_INDEX:
  301. case CAYMAN_RING_TYPE_CP2_INDEX:
  302. return true;
  303. default:
  304. return false;
  305. }
  306. }
  307. /**
  308. * radeon_ring_free_size - update the free size
  309. *
  310. * @rdev: radeon_device pointer
  311. * @ring: radeon_ring structure holding ring information
  312. *
  313. * Update the free dw slots in the ring buffer (all asics).
  314. */
  315. void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
  316. {
  317. u32 rptr;
  318. if (rdev->wb.enabled)
  319. rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
  320. else
  321. rptr = RREG32(ring->rptr_reg);
  322. ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
  323. /* This works because ring_size is a power of 2 */
  324. ring->ring_free_dw = (ring->rptr + (ring->ring_size / 4));
  325. ring->ring_free_dw -= ring->wptr;
  326. ring->ring_free_dw &= ring->ptr_mask;
  327. if (!ring->ring_free_dw) {
  328. ring->ring_free_dw = ring->ring_size / 4;
  329. }
  330. }
  331. /**
  332. * radeon_ring_alloc - allocate space on the ring buffer
  333. *
  334. * @rdev: radeon_device pointer
  335. * @ring: radeon_ring structure holding ring information
  336. * @ndw: number of dwords to allocate in the ring buffer
  337. *
  338. * Allocate @ndw dwords in the ring buffer (all asics).
  339. * Returns 0 on success, error on failure.
  340. */
  341. int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
  342. {
  343. int r;
  344. /* Align requested size with padding so unlock_commit can
  345. * pad safely */
  346. ndw = (ndw + ring->align_mask) & ~ring->align_mask;
  347. while (ndw > (ring->ring_free_dw - 1)) {
  348. radeon_ring_free_size(rdev, ring);
  349. if (ndw < ring->ring_free_dw) {
  350. break;
  351. }
  352. r = radeon_fence_wait_next_locked(rdev, ring->idx);
  353. if (r)
  354. return r;
  355. }
  356. ring->count_dw = ndw;
  357. ring->wptr_old = ring->wptr;
  358. return 0;
  359. }
  360. /**
  361. * radeon_ring_lock - lock the ring and allocate space on it
  362. *
  363. * @rdev: radeon_device pointer
  364. * @ring: radeon_ring structure holding ring information
  365. * @ndw: number of dwords to allocate in the ring buffer
  366. *
  367. * Lock the ring and allocate @ndw dwords in the ring buffer
  368. * (all asics).
  369. * Returns 0 on success, error on failure.
  370. */
  371. int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
  372. {
  373. int r;
  374. mutex_lock(&rdev->ring_lock);
  375. r = radeon_ring_alloc(rdev, ring, ndw);
  376. if (r) {
  377. mutex_unlock(&rdev->ring_lock);
  378. return r;
  379. }
  380. return 0;
  381. }
  382. /**
  383. * radeon_ring_commit - tell the GPU to execute the new
  384. * commands on the ring buffer
  385. *
  386. * @rdev: radeon_device pointer
  387. * @ring: radeon_ring structure holding ring information
  388. *
  389. * Update the wptr (write pointer) to tell the GPU to
  390. * execute new commands on the ring buffer (all asics).
  391. */
  392. void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring)
  393. {
  394. /* We pad to match fetch size */
  395. while (ring->wptr & ring->align_mask) {
  396. radeon_ring_write(ring, ring->nop);
  397. }
  398. DRM_MEMORYBARRIER();
  399. WREG32(ring->wptr_reg, (ring->wptr << ring->ptr_reg_shift) & ring->ptr_reg_mask);
  400. (void)RREG32(ring->wptr_reg);
  401. }
  402. /**
  403. * radeon_ring_unlock_commit - tell the GPU to execute the new
  404. * commands on the ring buffer and unlock it
  405. *
  406. * @rdev: radeon_device pointer
  407. * @ring: radeon_ring structure holding ring information
  408. *
  409. * Call radeon_ring_commit() then unlock the ring (all asics).
  410. */
  411. void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring)
  412. {
  413. radeon_ring_commit(rdev, ring);
  414. mutex_unlock(&rdev->ring_lock);
  415. }
  416. /**
  417. * radeon_ring_undo - reset the wptr
  418. *
  419. * @ring: radeon_ring structure holding ring information
  420. *
  421. * Reset the driver's copy of the wtpr (all asics).
  422. */
  423. void radeon_ring_undo(struct radeon_ring *ring)
  424. {
  425. ring->wptr = ring->wptr_old;
  426. }
  427. /**
  428. * radeon_ring_unlock_undo - reset the wptr and unlock the ring
  429. *
  430. * @ring: radeon_ring structure holding ring information
  431. *
  432. * Call radeon_ring_undo() then unlock the ring (all asics).
  433. */
  434. void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
  435. {
  436. radeon_ring_undo(ring);
  437. mutex_unlock(&rdev->ring_lock);
  438. }
  439. /**
  440. * radeon_ring_force_activity - add some nop packets to the ring
  441. *
  442. * @rdev: radeon_device pointer
  443. * @ring: radeon_ring structure holding ring information
  444. *
  445. * Add some nop packets to the ring to force activity (all asics).
  446. * Used for lockup detection to see if the rptr is advancing.
  447. */
  448. void radeon_ring_force_activity(struct radeon_device *rdev, struct radeon_ring *ring)
  449. {
  450. int r;
  451. radeon_ring_free_size(rdev, ring);
  452. if (ring->rptr == ring->wptr) {
  453. r = radeon_ring_alloc(rdev, ring, 1);
  454. if (!r) {
  455. radeon_ring_write(ring, ring->nop);
  456. radeon_ring_commit(rdev, ring);
  457. }
  458. }
  459. }
  460. /**
  461. * radeon_ring_force_activity - update lockup variables
  462. *
  463. * @ring: radeon_ring structure holding ring information
  464. *
  465. * Update the last rptr value and timestamp (all asics).
  466. */
  467. void radeon_ring_lockup_update(struct radeon_ring *ring)
  468. {
  469. ring->last_rptr = ring->rptr;
  470. ring->last_activity = jiffies;
  471. }
  472. /**
  473. * radeon_ring_test_lockup() - check if ring is lockedup by recording information
  474. * @rdev: radeon device structure
  475. * @ring: radeon_ring structure holding ring information
  476. *
  477. * We don't need to initialize the lockup tracking information as we will either
  478. * have CP rptr to a different value of jiffies wrap around which will force
  479. * initialization of the lockup tracking informations.
  480. *
  481. * A possible false positivie is if we get call after while and last_cp_rptr ==
  482. * the current CP rptr, even if it's unlikely it might happen. To avoid this
  483. * if the elapsed time since last call is bigger than 2 second than we return
  484. * false and update the tracking information. Due to this the caller must call
  485. * radeon_ring_test_lockup several time in less than 2sec for lockup to be reported
  486. * the fencing code should be cautious about that.
  487. *
  488. * Caller should write to the ring to force CP to do something so we don't get
  489. * false positive when CP is just gived nothing to do.
  490. *
  491. **/
  492. bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
  493. {
  494. unsigned long cjiffies, elapsed;
  495. uint32_t rptr;
  496. cjiffies = jiffies;
  497. if (!time_after(cjiffies, ring->last_activity)) {
  498. /* likely a wrap around */
  499. radeon_ring_lockup_update(ring);
  500. return false;
  501. }
  502. rptr = RREG32(ring->rptr_reg);
  503. ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
  504. if (ring->rptr != ring->last_rptr) {
  505. /* CP is still working no lockup */
  506. radeon_ring_lockup_update(ring);
  507. return false;
  508. }
  509. elapsed = jiffies_to_msecs(cjiffies - ring->last_activity);
  510. if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
  511. dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed);
  512. return true;
  513. }
  514. /* give a chance to the GPU ... */
  515. return false;
  516. }
  517. /**
  518. * radeon_ring_backup - Back up the content of a ring
  519. *
  520. * @rdev: radeon_device pointer
  521. * @ring: the ring we want to back up
  522. *
  523. * Saves all unprocessed commits from a ring, returns the number of dwords saved.
  524. */
  525. unsigned radeon_ring_backup(struct radeon_device *rdev, struct radeon_ring *ring,
  526. uint32_t **data)
  527. {
  528. unsigned size, ptr, i;
  529. /* just in case lock the ring */
  530. mutex_lock(&rdev->ring_lock);
  531. *data = NULL;
  532. if (ring->ring_obj == NULL) {
  533. mutex_unlock(&rdev->ring_lock);
  534. return 0;
  535. }
  536. /* it doesn't make sense to save anything if all fences are signaled */
  537. if (!radeon_fence_count_emitted(rdev, ring->idx)) {
  538. mutex_unlock(&rdev->ring_lock);
  539. return 0;
  540. }
  541. /* calculate the number of dw on the ring */
  542. if (ring->rptr_save_reg)
  543. ptr = RREG32(ring->rptr_save_reg);
  544. else if (rdev->wb.enabled)
  545. ptr = le32_to_cpu(*ring->next_rptr_cpu_addr);
  546. else {
  547. /* no way to read back the next rptr */
  548. mutex_unlock(&rdev->ring_lock);
  549. return 0;
  550. }
  551. size = ring->wptr + (ring->ring_size / 4);
  552. size -= ptr;
  553. size &= ring->ptr_mask;
  554. if (size == 0) {
  555. mutex_unlock(&rdev->ring_lock);
  556. return 0;
  557. }
  558. /* and then save the content of the ring */
  559. *data = kmalloc_array(size, sizeof(uint32_t), GFP_KERNEL);
  560. if (!*data) {
  561. mutex_unlock(&rdev->ring_lock);
  562. return 0;
  563. }
  564. for (i = 0; i < size; ++i) {
  565. (*data)[i] = ring->ring[ptr++];
  566. ptr &= ring->ptr_mask;
  567. }
  568. mutex_unlock(&rdev->ring_lock);
  569. return size;
  570. }
  571. /**
  572. * radeon_ring_restore - append saved commands to the ring again
  573. *
  574. * @rdev: radeon_device pointer
  575. * @ring: ring to append commands to
  576. * @size: number of dwords we want to write
  577. * @data: saved commands
  578. *
  579. * Allocates space on the ring and restore the previously saved commands.
  580. */
  581. int radeon_ring_restore(struct radeon_device *rdev, struct radeon_ring *ring,
  582. unsigned size, uint32_t *data)
  583. {
  584. int i, r;
  585. if (!size || !data)
  586. return 0;
  587. /* restore the saved ring content */
  588. r = radeon_ring_lock(rdev, ring, size);
  589. if (r)
  590. return r;
  591. for (i = 0; i < size; ++i) {
  592. radeon_ring_write(ring, data[i]);
  593. }
  594. radeon_ring_unlock_commit(rdev, ring);
  595. kfree(data);
  596. return 0;
  597. }
  598. /**
  599. * radeon_ring_init - init driver ring struct.
  600. *
  601. * @rdev: radeon_device pointer
  602. * @ring: radeon_ring structure holding ring information
  603. * @ring_size: size of the ring
  604. * @rptr_offs: offset of the rptr writeback location in the WB buffer
  605. * @rptr_reg: MMIO offset of the rptr register
  606. * @wptr_reg: MMIO offset of the wptr register
  607. * @ptr_reg_shift: bit offset of the rptr/wptr values
  608. * @ptr_reg_mask: bit mask of the rptr/wptr values
  609. * @nop: nop packet for this ring
  610. *
  611. * Initialize the driver information for the selected ring (all asics).
  612. * Returns 0 on success, error on failure.
  613. */
  614. int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
  615. unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg,
  616. u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop)
  617. {
  618. int r;
  619. ring->ring_size = ring_size;
  620. ring->rptr_offs = rptr_offs;
  621. ring->rptr_reg = rptr_reg;
  622. ring->wptr_reg = wptr_reg;
  623. ring->ptr_reg_shift = ptr_reg_shift;
  624. ring->ptr_reg_mask = ptr_reg_mask;
  625. ring->nop = nop;
  626. /* Allocate ring buffer */
  627. if (ring->ring_obj == NULL) {
  628. r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
  629. RADEON_GEM_DOMAIN_GTT,
  630. NULL, &ring->ring_obj);
  631. if (r) {
  632. dev_err(rdev->dev, "(%d) ring create failed\n", r);
  633. return r;
  634. }
  635. r = radeon_bo_reserve(ring->ring_obj, false);
  636. if (unlikely(r != 0))
  637. return r;
  638. r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
  639. &ring->gpu_addr);
  640. if (r) {
  641. radeon_bo_unreserve(ring->ring_obj);
  642. dev_err(rdev->dev, "(%d) ring pin failed\n", r);
  643. return r;
  644. }
  645. r = radeon_bo_kmap(ring->ring_obj,
  646. (void **)&ring->ring);
  647. radeon_bo_unreserve(ring->ring_obj);
  648. if (r) {
  649. dev_err(rdev->dev, "(%d) ring map failed\n", r);
  650. return r;
  651. }
  652. }
  653. ring->ptr_mask = (ring->ring_size / 4) - 1;
  654. ring->ring_free_dw = ring->ring_size / 4;
  655. if (rdev->wb.enabled) {
  656. u32 index = RADEON_WB_RING0_NEXT_RPTR + (ring->idx * 4);
  657. ring->next_rptr_gpu_addr = rdev->wb.gpu_addr + index;
  658. ring->next_rptr_cpu_addr = &rdev->wb.wb[index/4];
  659. }
  660. if (radeon_debugfs_ring_init(rdev, ring)) {
  661. DRM_ERROR("Failed to register debugfs file for rings !\n");
  662. }
  663. radeon_ring_lockup_update(ring);
  664. return 0;
  665. }
  666. /**
  667. * radeon_ring_fini - tear down the driver ring struct.
  668. *
  669. * @rdev: radeon_device pointer
  670. * @ring: radeon_ring structure holding ring information
  671. *
  672. * Tear down the driver information for the selected ring (all asics).
  673. */
  674. void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
  675. {
  676. int r;
  677. struct radeon_bo *ring_obj;
  678. mutex_lock(&rdev->ring_lock);
  679. ring_obj = ring->ring_obj;
  680. ring->ready = false;
  681. ring->ring = NULL;
  682. ring->ring_obj = NULL;
  683. mutex_unlock(&rdev->ring_lock);
  684. if (ring_obj) {
  685. r = radeon_bo_reserve(ring_obj, false);
  686. if (likely(r == 0)) {
  687. radeon_bo_kunmap(ring_obj);
  688. radeon_bo_unpin(ring_obj);
  689. radeon_bo_unreserve(ring_obj);
  690. }
  691. radeon_bo_unref(&ring_obj);
  692. }
  693. }
  694. /*
  695. * Debugfs info
  696. */
  697. #if defined(CONFIG_DEBUG_FS)
  698. static int radeon_debugfs_ring_info(struct seq_file *m, void *data)
  699. {
  700. struct drm_info_node *node = (struct drm_info_node *) m->private;
  701. struct drm_device *dev = node->minor->dev;
  702. struct radeon_device *rdev = dev->dev_private;
  703. int ridx = *(int*)node->info_ent->data;
  704. struct radeon_ring *ring = &rdev->ring[ridx];
  705. unsigned count, i, j;
  706. radeon_ring_free_size(rdev, ring);
  707. count = (ring->ring_size / 4) - ring->ring_free_dw;
  708. seq_printf(m, "wptr(0x%04x): 0x%08x\n", ring->wptr_reg, RREG32(ring->wptr_reg));
  709. seq_printf(m, "rptr(0x%04x): 0x%08x\n", ring->rptr_reg, RREG32(ring->rptr_reg));
  710. if (ring->rptr_save_reg) {
  711. seq_printf(m, "rptr next(0x%04x): 0x%08x\n", ring->rptr_save_reg,
  712. RREG32(ring->rptr_save_reg));
  713. }
  714. seq_printf(m, "driver's copy of the wptr: 0x%08x\n", ring->wptr);
  715. seq_printf(m, "driver's copy of the rptr: 0x%08x\n", ring->rptr);
  716. seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
  717. seq_printf(m, "%u dwords in ring\n", count);
  718. i = ring->rptr;
  719. for (j = 0; j <= count; j++) {
  720. seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
  721. i = (i + 1) & ring->ptr_mask;
  722. }
  723. return 0;
  724. }
  725. static int radeon_ring_type_gfx_index = RADEON_RING_TYPE_GFX_INDEX;
  726. static int cayman_ring_type_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX;
  727. static int cayman_ring_type_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX;
  728. static struct drm_info_list radeon_debugfs_ring_info_list[] = {
  729. {"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_ring_type_gfx_index},
  730. {"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp1_index},
  731. {"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp2_index},
  732. };
  733. static int radeon_debugfs_sa_info(struct seq_file *m, void *data)
  734. {
  735. struct drm_info_node *node = (struct drm_info_node *) m->private;
  736. struct drm_device *dev = node->minor->dev;
  737. struct radeon_device *rdev = dev->dev_private;
  738. radeon_sa_bo_dump_debug_info(&rdev->ring_tmp_bo, m);
  739. return 0;
  740. }
  741. static struct drm_info_list radeon_debugfs_sa_list[] = {
  742. {"radeon_sa_info", &radeon_debugfs_sa_info, 0, NULL},
  743. };
  744. #endif
  745. int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
  746. {
  747. #if defined(CONFIG_DEBUG_FS)
  748. unsigned i;
  749. for (i = 0; i < ARRAY_SIZE(radeon_debugfs_ring_info_list); ++i) {
  750. struct drm_info_list *info = &radeon_debugfs_ring_info_list[i];
  751. int ridx = *(int*)radeon_debugfs_ring_info_list[i].data;
  752. unsigned r;
  753. if (&rdev->ring[ridx] != ring)
  754. continue;
  755. r = radeon_debugfs_add_files(rdev, info, 1);
  756. if (r)
  757. return r;
  758. }
  759. #endif
  760. return 0;
  761. }
  762. int radeon_debugfs_sa_init(struct radeon_device *rdev)
  763. {
  764. #if defined(CONFIG_DEBUG_FS)
  765. return radeon_debugfs_add_files(rdev, radeon_debugfs_sa_list, 1);
  766. #else
  767. return 0;
  768. #endif
  769. }