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
28 #include <linux/seq_file.h>
29 #include <linux/slab.h>
31 #include "radeon_drm.h"
32 #include "radeon_reg.h"
36 int radeon_debugfs_ib_init(struct radeon_device *rdev);
37 int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
39 u32 radeon_get_ib_value(struct radeon_cs_parser *p, int idx)
41 struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
42 u32 pg_idx, pg_offset;
46 pg_idx = (idx * 4) / PAGE_SIZE;
47 pg_offset = (idx * 4) % PAGE_SIZE;
49 if (ibc->kpage_idx[0] == pg_idx)
50 return ibc->kpage[0][pg_offset/4];
51 if (ibc->kpage_idx[1] == pg_idx)
52 return ibc->kpage[1][pg_offset/4];
54 new_page = radeon_cs_update_pages(p, pg_idx);
56 p->parser_error = new_page;
60 idx_value = ibc->kpage[new_page][pg_offset/4];
64 void radeon_ring_write(struct radeon_ring *ring, uint32_t v)
67 if (ring->count_dw <= 0) {
68 DRM_ERROR("radeon: writting more dword to ring than expected !\n");
71 ring->ring[ring->wptr++] = v;
72 ring->wptr &= ring->ptr_mask;
80 bool radeon_ib_try_free(struct radeon_device *rdev, struct radeon_ib *ib)
84 /* only free ib which have been emited */
85 if (ib->fence && ib->fence->emitted) {
86 if (radeon_fence_signaled(ib->fence)) {
87 radeon_fence_unref(&ib->fence);
88 radeon_sa_bo_free(rdev, &ib->sa_bo);
95 int radeon_ib_get(struct radeon_device *rdev, int ring,
96 struct radeon_ib **ib, unsigned size)
98 struct radeon_fence *fence;
103 /* align size on 256 bytes */
104 size = ALIGN(size, 256);
106 r = radeon_fence_create(rdev, &fence, ring);
108 dev_err(rdev->dev, "failed to create fence for new IB\n");
112 radeon_mutex_lock(&rdev->ib_pool.mutex);
113 idx = rdev->ib_pool.head_id;
116 dev_err(rdev->dev, "failed to get an ib after 5 retry\n");
117 radeon_mutex_unlock(&rdev->ib_pool.mutex);
118 radeon_fence_unref(&fence);
122 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
123 radeon_ib_try_free(rdev, &rdev->ib_pool.ibs[idx]);
124 if (rdev->ib_pool.ibs[idx].fence == NULL) {
125 r = radeon_sa_bo_new(rdev, &rdev->ib_pool.sa_manager,
126 &rdev->ib_pool.ibs[idx].sa_bo,
129 *ib = &rdev->ib_pool.ibs[idx];
130 (*ib)->ptr = rdev->ib_pool.sa_manager.cpu_ptr;
131 (*ib)->ptr += ((*ib)->sa_bo.offset >> 2);
132 (*ib)->gpu_addr = rdev->ib_pool.sa_manager.gpu_addr;
133 (*ib)->gpu_addr += (*ib)->sa_bo.offset;
134 (*ib)->fence = fence;
136 (*ib)->is_const_ib = false;
137 /* ib are most likely to be allocated in a ring fashion
138 * thus rdev->ib_pool.head_id should be the id of the
141 rdev->ib_pool.head_id = (1 + idx);
142 rdev->ib_pool.head_id &= (RADEON_IB_POOL_SIZE - 1);
143 radeon_mutex_unlock(&rdev->ib_pool.mutex);
147 idx = (idx + 1) & (RADEON_IB_POOL_SIZE - 1);
149 /* this should be rare event, ie all ib scheduled none signaled yet.
151 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
152 if (rdev->ib_pool.ibs[idx].fence && rdev->ib_pool.ibs[idx].fence->emitted) {
153 r = radeon_fence_wait(rdev->ib_pool.ibs[idx].fence, false);
157 /* an error happened */
160 idx = (idx + 1) & (RADEON_IB_POOL_SIZE - 1);
162 radeon_mutex_unlock(&rdev->ib_pool.mutex);
163 radeon_fence_unref(&fence);
167 void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib)
169 struct radeon_ib *tmp = *ib;
175 radeon_mutex_lock(&rdev->ib_pool.mutex);
176 if (tmp->fence && !tmp->fence->emitted) {
177 radeon_sa_bo_free(rdev, &tmp->sa_bo);
178 radeon_fence_unref(&tmp->fence);
180 radeon_mutex_unlock(&rdev->ib_pool.mutex);
183 int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
185 struct radeon_ring *ring = &rdev->ring[ib->fence->ring];
188 if (!ib->length_dw || !ring->ready) {
189 /* TODO: Nothings in the ib we should report. */
190 DRM_ERROR("radeon: couldn't schedule IB(%u).\n", ib->idx);
194 /* 64 dwords should be enough for fence too */
195 r = radeon_ring_lock(rdev, ring, 64);
197 DRM_ERROR("radeon: scheduling IB failed (%d).\n", r);
200 radeon_ring_ib_execute(rdev, ib->fence->ring, ib);
201 radeon_fence_emit(rdev, ib->fence);
202 radeon_ring_unlock_commit(rdev, ring);
206 int radeon_ib_pool_init(struct radeon_device *rdev)
208 struct radeon_sa_manager tmp;
211 r = radeon_sa_bo_manager_init(rdev, &tmp,
212 RADEON_IB_POOL_SIZE*64*1024,
213 RADEON_GEM_DOMAIN_GTT);
218 radeon_mutex_lock(&rdev->ib_pool.mutex);
219 if (rdev->ib_pool.ready) {
220 radeon_mutex_unlock(&rdev->ib_pool.mutex);
221 radeon_sa_bo_manager_fini(rdev, &tmp);
225 rdev->ib_pool.sa_manager = tmp;
226 INIT_LIST_HEAD(&rdev->ib_pool.sa_manager.sa_bo);
227 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
228 rdev->ib_pool.ibs[i].fence = NULL;
229 rdev->ib_pool.ibs[i].idx = i;
230 rdev->ib_pool.ibs[i].length_dw = 0;
231 INIT_LIST_HEAD(&rdev->ib_pool.ibs[i].sa_bo.list);
233 rdev->ib_pool.head_id = 0;
234 rdev->ib_pool.ready = true;
235 DRM_INFO("radeon: ib pool ready.\n");
237 if (radeon_debugfs_ib_init(rdev)) {
238 DRM_ERROR("Failed to register debugfs file for IB !\n");
240 radeon_mutex_unlock(&rdev->ib_pool.mutex);
244 void radeon_ib_pool_fini(struct radeon_device *rdev)
248 radeon_mutex_lock(&rdev->ib_pool.mutex);
249 if (rdev->ib_pool.ready) {
250 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
251 radeon_sa_bo_free(rdev, &rdev->ib_pool.ibs[i].sa_bo);
252 radeon_fence_unref(&rdev->ib_pool.ibs[i].fence);
254 radeon_sa_bo_manager_fini(rdev, &rdev->ib_pool.sa_manager);
255 rdev->ib_pool.ready = false;
257 radeon_mutex_unlock(&rdev->ib_pool.mutex);
260 int radeon_ib_pool_start(struct radeon_device *rdev)
262 return radeon_sa_bo_manager_start(rdev, &rdev->ib_pool.sa_manager);
265 int radeon_ib_pool_suspend(struct radeon_device *rdev)
267 return radeon_sa_bo_manager_suspend(rdev, &rdev->ib_pool.sa_manager);
270 int radeon_ib_ring_tests(struct radeon_device *rdev)
275 for (i = 0; i < RADEON_NUM_RINGS; ++i) {
276 struct radeon_ring *ring = &rdev->ring[i];
281 r = radeon_ib_test(rdev, i, ring);
285 if (i == RADEON_RING_TYPE_GFX_INDEX) {
286 /* oh, oh, that's really bad */
287 DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r);
288 rdev->accel_working = false;
292 /* still not good, but we can live with it */
293 DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r);
303 int radeon_ring_index(struct radeon_device *rdev, struct radeon_ring *ring)
305 /* r1xx-r5xx only has CP ring */
306 if (rdev->family < CHIP_R600)
307 return RADEON_RING_TYPE_GFX_INDEX;
309 if (rdev->family >= CHIP_CAYMAN) {
310 if (ring == &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX])
311 return CAYMAN_RING_TYPE_CP1_INDEX;
312 else if (ring == &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX])
313 return CAYMAN_RING_TYPE_CP2_INDEX;
315 return RADEON_RING_TYPE_GFX_INDEX;
318 void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
322 if (rdev->wb.enabled)
323 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
325 rptr = RREG32(ring->rptr_reg);
326 ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
327 /* This works because ring_size is a power of 2 */
328 ring->ring_free_dw = (ring->rptr + (ring->ring_size / 4));
329 ring->ring_free_dw -= ring->wptr;
330 ring->ring_free_dw &= ring->ptr_mask;
331 if (!ring->ring_free_dw) {
332 ring->ring_free_dw = ring->ring_size / 4;
337 int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
341 /* Align requested size with padding so unlock_commit can
343 ndw = (ndw + ring->align_mask) & ~ring->align_mask;
344 while (ndw > (ring->ring_free_dw - 1)) {
345 radeon_ring_free_size(rdev, ring);
346 if (ndw < ring->ring_free_dw) {
349 mutex_unlock(&ring->mutex);
350 r = radeon_fence_wait_next(rdev, radeon_ring_index(rdev, ring));
351 mutex_lock(&ring->mutex);
355 ring->count_dw = ndw;
356 ring->wptr_old = ring->wptr;
360 int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
364 mutex_lock(&ring->mutex);
365 r = radeon_ring_alloc(rdev, ring, ndw);
367 mutex_unlock(&ring->mutex);
373 void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring)
375 unsigned count_dw_pad;
378 /* We pad to match fetch size */
379 count_dw_pad = (ring->align_mask + 1) -
380 (ring->wptr & ring->align_mask);
381 for (i = 0; i < count_dw_pad; i++) {
382 radeon_ring_write(ring, ring->nop);
385 WREG32(ring->wptr_reg, (ring->wptr << ring->ptr_reg_shift) & ring->ptr_reg_mask);
386 (void)RREG32(ring->wptr_reg);
389 void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring)
391 radeon_ring_commit(rdev, ring);
392 mutex_unlock(&ring->mutex);
395 void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
397 ring->wptr = ring->wptr_old;
398 mutex_unlock(&ring->mutex);
401 void radeon_ring_force_activity(struct radeon_device *rdev, struct radeon_ring *ring)
405 mutex_lock(&ring->mutex);
406 radeon_ring_free_size(rdev, ring);
407 if (ring->rptr == ring->wptr) {
408 r = radeon_ring_alloc(rdev, ring, 1);
410 radeon_ring_write(ring, ring->nop);
411 radeon_ring_commit(rdev, ring);
414 mutex_unlock(&ring->mutex);
417 void radeon_ring_lockup_update(struct radeon_ring *ring)
419 ring->last_rptr = ring->rptr;
420 ring->last_activity = jiffies;
424 * radeon_ring_test_lockup() - check if ring is lockedup by recording information
425 * @rdev: radeon device structure
426 * @ring: radeon_ring structure holding ring information
428 * We don't need to initialize the lockup tracking information as we will either
429 * have CP rptr to a different value of jiffies wrap around which will force
430 * initialization of the lockup tracking informations.
432 * A possible false positivie is if we get call after while and last_cp_rptr ==
433 * the current CP rptr, even if it's unlikely it might happen. To avoid this
434 * if the elapsed time since last call is bigger than 2 second than we return
435 * false and update the tracking information. Due to this the caller must call
436 * radeon_ring_test_lockup several time in less than 2sec for lockup to be reported
437 * the fencing code should be cautious about that.
439 * Caller should write to the ring to force CP to do something so we don't get
440 * false positive when CP is just gived nothing to do.
443 bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
445 unsigned long cjiffies, elapsed;
449 if (!time_after(cjiffies, ring->last_activity)) {
450 /* likely a wrap around */
451 radeon_ring_lockup_update(ring);
454 rptr = RREG32(ring->rptr_reg);
455 ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
456 if (ring->rptr != ring->last_rptr) {
457 /* CP is still working no lockup */
458 radeon_ring_lockup_update(ring);
461 elapsed = jiffies_to_msecs(cjiffies - ring->last_activity);
462 if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
463 dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed);
466 /* give a chance to the GPU ... */
470 int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
471 unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg,
472 u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop)
476 ring->ring_size = ring_size;
477 ring->rptr_offs = rptr_offs;
478 ring->rptr_reg = rptr_reg;
479 ring->wptr_reg = wptr_reg;
480 ring->ptr_reg_shift = ptr_reg_shift;
481 ring->ptr_reg_mask = ptr_reg_mask;
483 /* Allocate ring buffer */
484 if (ring->ring_obj == NULL) {
485 r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
486 RADEON_GEM_DOMAIN_GTT,
489 dev_err(rdev->dev, "(%d) ring create failed\n", r);
492 r = radeon_bo_reserve(ring->ring_obj, false);
493 if (unlikely(r != 0))
495 r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
498 radeon_bo_unreserve(ring->ring_obj);
499 dev_err(rdev->dev, "(%d) ring pin failed\n", r);
502 r = radeon_bo_kmap(ring->ring_obj,
503 (void **)&ring->ring);
504 radeon_bo_unreserve(ring->ring_obj);
506 dev_err(rdev->dev, "(%d) ring map failed\n", r);
510 ring->ptr_mask = (ring->ring_size / 4) - 1;
511 ring->ring_free_dw = ring->ring_size / 4;
512 if (radeon_debugfs_ring_init(rdev, ring)) {
513 DRM_ERROR("Failed to register debugfs file for rings !\n");
518 void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
521 struct radeon_bo *ring_obj;
523 mutex_lock(&ring->mutex);
524 ring_obj = ring->ring_obj;
526 ring->ring_obj = NULL;
527 mutex_unlock(&ring->mutex);
530 r = radeon_bo_reserve(ring_obj, false);
531 if (likely(r == 0)) {
532 radeon_bo_kunmap(ring_obj);
533 radeon_bo_unpin(ring_obj);
534 radeon_bo_unreserve(ring_obj);
536 radeon_bo_unref(&ring_obj);
543 #if defined(CONFIG_DEBUG_FS)
545 static int radeon_debugfs_ring_info(struct seq_file *m, void *data)
547 struct drm_info_node *node = (struct drm_info_node *) m->private;
548 struct drm_device *dev = node->minor->dev;
549 struct radeon_device *rdev = dev->dev_private;
550 int ridx = *(int*)node->info_ent->data;
551 struct radeon_ring *ring = &rdev->ring[ridx];
552 unsigned count, i, j;
554 radeon_ring_free_size(rdev, ring);
555 count = (ring->ring_size / 4) - ring->ring_free_dw;
556 seq_printf(m, "wptr(0x%04x): 0x%08x\n", ring->wptr_reg, RREG32(ring->wptr_reg));
557 seq_printf(m, "rptr(0x%04x): 0x%08x\n", ring->rptr_reg, RREG32(ring->rptr_reg));
558 seq_printf(m, "driver's copy of the wptr: 0x%08x\n", ring->wptr);
559 seq_printf(m, "driver's copy of the rptr: 0x%08x\n", ring->rptr);
560 seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
561 seq_printf(m, "%u dwords in ring\n", count);
563 for (j = 0; j <= count; j++) {
564 seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
565 i = (i + 1) & ring->ptr_mask;
570 static int radeon_ring_type_gfx_index = RADEON_RING_TYPE_GFX_INDEX;
571 static int cayman_ring_type_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX;
572 static int cayman_ring_type_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX;
574 static struct drm_info_list radeon_debugfs_ring_info_list[] = {
575 {"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_ring_type_gfx_index},
576 {"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp1_index},
577 {"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp2_index},
580 static int radeon_debugfs_ib_info(struct seq_file *m, void *data)
582 struct drm_info_node *node = (struct drm_info_node *) m->private;
583 struct drm_device *dev = node->minor->dev;
584 struct radeon_device *rdev = dev->dev_private;
585 struct radeon_ib *ib = &rdev->ib_pool.ibs[*((unsigned*)node->info_ent->data)];
591 seq_printf(m, "IB %04u\n", ib->idx);
592 seq_printf(m, "IB fence %p\n", ib->fence);
593 seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
594 for (i = 0; i < ib->length_dw; i++) {
595 seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
600 static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE];
601 static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32];
602 static unsigned radeon_debugfs_ib_idx[RADEON_IB_POOL_SIZE];
605 int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
607 #if defined(CONFIG_DEBUG_FS)
609 for (i = 0; i < ARRAY_SIZE(radeon_debugfs_ring_info_list); ++i) {
610 struct drm_info_list *info = &radeon_debugfs_ring_info_list[i];
611 int ridx = *(int*)radeon_debugfs_ring_info_list[i].data;
614 if (&rdev->ring[ridx] != ring)
617 r = radeon_debugfs_add_files(rdev, info, 1);
625 int radeon_debugfs_ib_init(struct radeon_device *rdev)
627 #if defined(CONFIG_DEBUG_FS)
630 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
631 sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i);
632 radeon_debugfs_ib_idx[i] = i;
633 radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i];
634 radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info;
635 radeon_debugfs_ib_list[i].driver_features = 0;
636 radeon_debugfs_ib_list[i].data = &radeon_debugfs_ib_idx[i];
638 return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list,
639 RADEON_IB_POOL_SIZE);