2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 * Authors: Dave Airlie
29 #include <linux/seq_file.h>
30 #include <linux/slab.h>
32 #include "radeon_drm.h"
33 #include "radeon_reg.h"
40 int radeon_debugfs_sa_init(struct radeon_device *rdev);
42 int radeon_ib_get(struct radeon_device *rdev, int ring,
43 struct radeon_ib *ib, unsigned size)
47 r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true);
49 dev_err(rdev->dev, "failed to get a new IB (%d)\n", r);
53 r = radeon_semaphore_create(rdev, &ib->semaphore);
60 ib->ptr = radeon_sa_bo_cpu_addr(ib->sa_bo);
61 ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo);
63 ib->is_const_ib = false;
64 for (i = 0; i < RADEON_NUM_RINGS; ++i)
65 ib->sync_to[i] = NULL;
70 void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
72 radeon_semaphore_free(rdev, &ib->semaphore, ib->fence);
73 radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence);
74 radeon_fence_unref(&ib->fence);
77 int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
79 struct radeon_ring *ring = &rdev->ring[ib->ring];
80 bool need_sync = false;
83 if (!ib->length_dw || !ring->ready) {
84 /* TODO: Nothings in the ib we should report. */
85 dev_err(rdev->dev, "couldn't schedule ib\n");
89 /* 64 dwords should be enough for fence too */
90 r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_RINGS * 8);
92 dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
95 for (i = 0; i < RADEON_NUM_RINGS; ++i) {
96 struct radeon_fence *fence = ib->sync_to[i];
97 if (radeon_fence_need_sync(fence, ib->ring)) {
99 radeon_semaphore_sync_rings(rdev, ib->semaphore,
100 fence->ring, ib->ring);
101 radeon_fence_note_sync(fence, ib->ring);
104 /* immediately free semaphore when we don't need to sync */
106 radeon_semaphore_free(rdev, &ib->semaphore, NULL);
108 radeon_ring_ib_execute(rdev, ib->ring, ib);
109 r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
111 dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r);
112 radeon_ring_unlock_undo(rdev, ring);
115 radeon_ring_unlock_commit(rdev, ring);
119 int radeon_ib_pool_init(struct radeon_device *rdev)
123 if (rdev->ib_pool_ready) {
126 r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo,
127 RADEON_IB_POOL_SIZE*64*1024,
128 RADEON_GEM_DOMAIN_GTT);
132 rdev->ib_pool_ready = true;
133 if (radeon_debugfs_sa_init(rdev)) {
134 dev_err(rdev->dev, "failed to register debugfs file for SA\n");
139 void radeon_ib_pool_fini(struct radeon_device *rdev)
141 if (rdev->ib_pool_ready) {
142 radeon_sa_bo_manager_fini(rdev, &rdev->ring_tmp_bo);
143 rdev->ib_pool_ready = false;
147 int radeon_ib_pool_start(struct radeon_device *rdev)
149 return radeon_sa_bo_manager_start(rdev, &rdev->ring_tmp_bo);
152 int radeon_ib_pool_suspend(struct radeon_device *rdev)
154 return radeon_sa_bo_manager_suspend(rdev, &rdev->ring_tmp_bo);
157 int radeon_ib_ring_tests(struct radeon_device *rdev)
162 for (i = 0; i < RADEON_NUM_RINGS; ++i) {
163 struct radeon_ring *ring = &rdev->ring[i];
168 r = radeon_ib_test(rdev, i, ring);
172 if (i == RADEON_RING_TYPE_GFX_INDEX) {
173 /* oh, oh, that's really bad */
174 DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r);
175 rdev->accel_working = false;
179 /* still not good, but we can live with it */
180 DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r);
190 int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
192 void radeon_ring_write(struct radeon_ring *ring, uint32_t v)
195 if (ring->count_dw <= 0) {
196 DRM_ERROR("radeon: writting more dword to ring than expected !\n");
199 ring->ring[ring->wptr++] = v;
200 ring->wptr &= ring->ptr_mask;
202 ring->ring_free_dw--;
205 int radeon_ring_index(struct radeon_device *rdev, struct radeon_ring *ring)
207 /* r1xx-r5xx only has CP ring */
208 if (rdev->family < CHIP_R600)
209 return RADEON_RING_TYPE_GFX_INDEX;
211 if (rdev->family >= CHIP_CAYMAN) {
212 if (ring == &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX])
213 return CAYMAN_RING_TYPE_CP1_INDEX;
214 else if (ring == &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX])
215 return CAYMAN_RING_TYPE_CP2_INDEX;
217 return RADEON_RING_TYPE_GFX_INDEX;
220 void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
224 if (rdev->wb.enabled)
225 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
227 rptr = RREG32(ring->rptr_reg);
228 ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
229 /* This works because ring_size is a power of 2 */
230 ring->ring_free_dw = (ring->rptr + (ring->ring_size / 4));
231 ring->ring_free_dw -= ring->wptr;
232 ring->ring_free_dw &= ring->ptr_mask;
233 if (!ring->ring_free_dw) {
234 ring->ring_free_dw = ring->ring_size / 4;
239 int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
243 /* Align requested size with padding so unlock_commit can
245 ndw = (ndw + ring->align_mask) & ~ring->align_mask;
246 while (ndw > (ring->ring_free_dw - 1)) {
247 radeon_ring_free_size(rdev, ring);
248 if (ndw < ring->ring_free_dw) {
251 r = radeon_fence_wait_next_locked(rdev, radeon_ring_index(rdev, ring));
255 ring->count_dw = ndw;
256 ring->wptr_old = ring->wptr;
260 int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
264 mutex_lock(&rdev->ring_lock);
265 r = radeon_ring_alloc(rdev, ring, ndw);
267 mutex_unlock(&rdev->ring_lock);
273 void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring)
275 unsigned count_dw_pad;
278 /* We pad to match fetch size */
279 count_dw_pad = (ring->align_mask + 1) -
280 (ring->wptr & ring->align_mask);
281 for (i = 0; i < count_dw_pad; i++) {
282 radeon_ring_write(ring, ring->nop);
285 WREG32(ring->wptr_reg, (ring->wptr << ring->ptr_reg_shift) & ring->ptr_reg_mask);
286 (void)RREG32(ring->wptr_reg);
289 void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring)
291 radeon_ring_commit(rdev, ring);
292 mutex_unlock(&rdev->ring_lock);
295 void radeon_ring_undo(struct radeon_ring *ring)
297 ring->wptr = ring->wptr_old;
300 void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
302 radeon_ring_undo(ring);
303 mutex_unlock(&rdev->ring_lock);
306 void radeon_ring_force_activity(struct radeon_device *rdev, struct radeon_ring *ring)
310 radeon_ring_free_size(rdev, ring);
311 if (ring->rptr == ring->wptr) {
312 r = radeon_ring_alloc(rdev, ring, 1);
314 radeon_ring_write(ring, ring->nop);
315 radeon_ring_commit(rdev, ring);
320 void radeon_ring_lockup_update(struct radeon_ring *ring)
322 ring->last_rptr = ring->rptr;
323 ring->last_activity = jiffies;
327 * radeon_ring_test_lockup() - check if ring is lockedup by recording information
328 * @rdev: radeon device structure
329 * @ring: radeon_ring structure holding ring information
331 * We don't need to initialize the lockup tracking information as we will either
332 * have CP rptr to a different value of jiffies wrap around which will force
333 * initialization of the lockup tracking informations.
335 * A possible false positivie is if we get call after while and last_cp_rptr ==
336 * the current CP rptr, even if it's unlikely it might happen. To avoid this
337 * if the elapsed time since last call is bigger than 2 second than we return
338 * false and update the tracking information. Due to this the caller must call
339 * radeon_ring_test_lockup several time in less than 2sec for lockup to be reported
340 * the fencing code should be cautious about that.
342 * Caller should write to the ring to force CP to do something so we don't get
343 * false positive when CP is just gived nothing to do.
346 bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
348 unsigned long cjiffies, elapsed;
352 if (!time_after(cjiffies, ring->last_activity)) {
353 /* likely a wrap around */
354 radeon_ring_lockup_update(ring);
357 rptr = RREG32(ring->rptr_reg);
358 ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
359 if (ring->rptr != ring->last_rptr) {
360 /* CP is still working no lockup */
361 radeon_ring_lockup_update(ring);
364 elapsed = jiffies_to_msecs(cjiffies - ring->last_activity);
365 if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
366 dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed);
369 /* give a chance to the GPU ... */
373 int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
374 unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg,
375 u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop)
379 ring->ring_size = ring_size;
380 ring->rptr_offs = rptr_offs;
381 ring->rptr_reg = rptr_reg;
382 ring->wptr_reg = wptr_reg;
383 ring->ptr_reg_shift = ptr_reg_shift;
384 ring->ptr_reg_mask = ptr_reg_mask;
386 /* Allocate ring buffer */
387 if (ring->ring_obj == NULL) {
388 r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
389 RADEON_GEM_DOMAIN_GTT,
390 NULL, &ring->ring_obj);
392 dev_err(rdev->dev, "(%d) ring create failed\n", r);
395 r = radeon_bo_reserve(ring->ring_obj, false);
396 if (unlikely(r != 0))
398 r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
401 radeon_bo_unreserve(ring->ring_obj);
402 dev_err(rdev->dev, "(%d) ring pin failed\n", r);
405 r = radeon_bo_kmap(ring->ring_obj,
406 (void **)&ring->ring);
407 radeon_bo_unreserve(ring->ring_obj);
409 dev_err(rdev->dev, "(%d) ring map failed\n", r);
413 ring->ptr_mask = (ring->ring_size / 4) - 1;
414 ring->ring_free_dw = ring->ring_size / 4;
415 if (radeon_debugfs_ring_init(rdev, ring)) {
416 DRM_ERROR("Failed to register debugfs file for rings !\n");
421 void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
424 struct radeon_bo *ring_obj;
426 mutex_lock(&rdev->ring_lock);
427 ring_obj = ring->ring_obj;
430 ring->ring_obj = NULL;
431 mutex_unlock(&rdev->ring_lock);
434 r = radeon_bo_reserve(ring_obj, false);
435 if (likely(r == 0)) {
436 radeon_bo_kunmap(ring_obj);
437 radeon_bo_unpin(ring_obj);
438 radeon_bo_unreserve(ring_obj);
440 radeon_bo_unref(&ring_obj);
447 #if defined(CONFIG_DEBUG_FS)
449 static int radeon_debugfs_ring_info(struct seq_file *m, void *data)
451 struct drm_info_node *node = (struct drm_info_node *) m->private;
452 struct drm_device *dev = node->minor->dev;
453 struct radeon_device *rdev = dev->dev_private;
454 int ridx = *(int*)node->info_ent->data;
455 struct radeon_ring *ring = &rdev->ring[ridx];
456 unsigned count, i, j;
458 radeon_ring_free_size(rdev, ring);
459 count = (ring->ring_size / 4) - ring->ring_free_dw;
460 seq_printf(m, "wptr(0x%04x): 0x%08x\n", ring->wptr_reg, RREG32(ring->wptr_reg));
461 seq_printf(m, "rptr(0x%04x): 0x%08x\n", ring->rptr_reg, RREG32(ring->rptr_reg));
462 seq_printf(m, "driver's copy of the wptr: 0x%08x\n", ring->wptr);
463 seq_printf(m, "driver's copy of the rptr: 0x%08x\n", ring->rptr);
464 seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
465 seq_printf(m, "%u dwords in ring\n", count);
467 for (j = 0; j <= count; j++) {
468 seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
469 i = (i + 1) & ring->ptr_mask;
474 static int radeon_ring_type_gfx_index = RADEON_RING_TYPE_GFX_INDEX;
475 static int cayman_ring_type_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX;
476 static int cayman_ring_type_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX;
478 static struct drm_info_list radeon_debugfs_ring_info_list[] = {
479 {"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_ring_type_gfx_index},
480 {"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp1_index},
481 {"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp2_index},
484 static int radeon_debugfs_sa_info(struct seq_file *m, void *data)
486 struct drm_info_node *node = (struct drm_info_node *) m->private;
487 struct drm_device *dev = node->minor->dev;
488 struct radeon_device *rdev = dev->dev_private;
490 radeon_sa_bo_dump_debug_info(&rdev->ring_tmp_bo, m);
496 static struct drm_info_list radeon_debugfs_sa_list[] = {
497 {"radeon_sa_info", &radeon_debugfs_sa_info, 0, NULL},
502 int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
504 #if defined(CONFIG_DEBUG_FS)
506 for (i = 0; i < ARRAY_SIZE(radeon_debugfs_ring_info_list); ++i) {
507 struct drm_info_list *info = &radeon_debugfs_ring_info_list[i];
508 int ridx = *(int*)radeon_debugfs_ring_info_list[i].data;
511 if (&rdev->ring[ridx] != ring)
514 r = radeon_debugfs_add_files(rdev, info, 1);
522 int radeon_debugfs_sa_init(struct radeon_device *rdev)
524 #if defined(CONFIG_DEBUG_FS)
525 return radeon_debugfs_add_files(rdev, radeon_debugfs_sa_list, 1);