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 u32 radeon_get_ib_value(struct radeon_cs_parser *p, int idx)
44 struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
45 u32 pg_idx, pg_offset;
49 pg_idx = (idx * 4) / PAGE_SIZE;
50 pg_offset = (idx * 4) % PAGE_SIZE;
52 if (ibc->kpage_idx[0] == pg_idx)
53 return ibc->kpage[0][pg_offset/4];
54 if (ibc->kpage_idx[1] == pg_idx)
55 return ibc->kpage[1][pg_offset/4];
57 new_page = radeon_cs_update_pages(p, pg_idx);
59 p->parser_error = new_page;
63 idx_value = ibc->kpage[new_page][pg_offset/4];
67 int radeon_ib_get(struct radeon_device *rdev, int ring,
68 struct radeon_ib **ib, unsigned size)
72 *ib = kmalloc(sizeof(struct radeon_ib), GFP_KERNEL);
76 r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &(*ib)->sa_bo, size, 256, true);
78 dev_err(rdev->dev, "failed to get a new IB (%d)\n", r);
83 r = radeon_fence_create(rdev, &(*ib)->fence, ring);
85 dev_err(rdev->dev, "failed to create fence for new IB (%d)\n", r);
86 radeon_sa_bo_free(rdev, &(*ib)->sa_bo, NULL);
92 (*ib)->ptr = radeon_sa_bo_cpu_addr((*ib)->sa_bo);
93 (*ib)->gpu_addr = radeon_sa_bo_gpu_addr((*ib)->sa_bo);
95 (*ib)->is_const_ib = false;
100 void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib)
102 struct radeon_ib *tmp = *ib;
108 radeon_sa_bo_free(rdev, &tmp->sa_bo, tmp->fence);
109 radeon_fence_unref(&tmp->fence);
113 int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
115 struct radeon_ring *ring = &rdev->ring[ib->fence->ring];
118 if (!ib->length_dw || !ring->ready) {
119 /* TODO: Nothings in the ib we should report. */
120 dev_err(rdev->dev, "couldn't schedule ib\n");
124 /* 64 dwords should be enough for fence too */
125 r = radeon_ring_lock(rdev, ring, 64);
127 dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
130 radeon_ring_ib_execute(rdev, ib->fence->ring, ib);
131 radeon_fence_emit(rdev, ib->fence);
132 radeon_ring_unlock_commit(rdev, ring);
136 int radeon_ib_pool_init(struct radeon_device *rdev)
140 if (rdev->ib_pool_ready) {
143 r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo,
144 RADEON_IB_POOL_SIZE*64*1024,
145 RADEON_GEM_DOMAIN_GTT);
149 rdev->ib_pool_ready = true;
150 if (radeon_debugfs_sa_init(rdev)) {
151 dev_err(rdev->dev, "failed to register debugfs file for SA\n");
156 void radeon_ib_pool_fini(struct radeon_device *rdev)
158 if (rdev->ib_pool_ready) {
159 radeon_sa_bo_manager_fini(rdev, &rdev->ring_tmp_bo);
160 rdev->ib_pool_ready = false;
164 int radeon_ib_pool_start(struct radeon_device *rdev)
166 return radeon_sa_bo_manager_start(rdev, &rdev->ring_tmp_bo);
169 int radeon_ib_pool_suspend(struct radeon_device *rdev)
171 return radeon_sa_bo_manager_suspend(rdev, &rdev->ring_tmp_bo);
174 int radeon_ib_ring_tests(struct radeon_device *rdev)
179 for (i = 0; i < RADEON_NUM_RINGS; ++i) {
180 struct radeon_ring *ring = &rdev->ring[i];
185 r = radeon_ib_test(rdev, i, ring);
189 if (i == RADEON_RING_TYPE_GFX_INDEX) {
190 /* oh, oh, that's really bad */
191 DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r);
192 rdev->accel_working = false;
196 /* still not good, but we can live with it */
197 DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r);
207 int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
209 void radeon_ring_write(struct radeon_ring *ring, uint32_t v)
212 if (ring->count_dw <= 0) {
213 DRM_ERROR("radeon: writting more dword to ring than expected !\n");
216 ring->ring[ring->wptr++] = v;
217 ring->wptr &= ring->ptr_mask;
219 ring->ring_free_dw--;
222 int radeon_ring_index(struct radeon_device *rdev, struct radeon_ring *ring)
224 /* r1xx-r5xx only has CP ring */
225 if (rdev->family < CHIP_R600)
226 return RADEON_RING_TYPE_GFX_INDEX;
228 if (rdev->family >= CHIP_CAYMAN) {
229 if (ring == &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX])
230 return CAYMAN_RING_TYPE_CP1_INDEX;
231 else if (ring == &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX])
232 return CAYMAN_RING_TYPE_CP2_INDEX;
234 return RADEON_RING_TYPE_GFX_INDEX;
237 void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
241 if (rdev->wb.enabled)
242 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
244 rptr = RREG32(ring->rptr_reg);
245 ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
246 /* This works because ring_size is a power of 2 */
247 ring->ring_free_dw = (ring->rptr + (ring->ring_size / 4));
248 ring->ring_free_dw -= ring->wptr;
249 ring->ring_free_dw &= ring->ptr_mask;
250 if (!ring->ring_free_dw) {
251 ring->ring_free_dw = ring->ring_size / 4;
256 int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
260 /* Align requested size with padding so unlock_commit can
262 ndw = (ndw + ring->align_mask) & ~ring->align_mask;
263 while (ndw > (ring->ring_free_dw - 1)) {
264 radeon_ring_free_size(rdev, ring);
265 if (ndw < ring->ring_free_dw) {
268 r = radeon_fence_wait_next_locked(rdev, radeon_ring_index(rdev, ring));
272 ring->count_dw = ndw;
273 ring->wptr_old = ring->wptr;
277 int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
281 mutex_lock(&rdev->ring_lock);
282 r = radeon_ring_alloc(rdev, ring, ndw);
284 mutex_unlock(&rdev->ring_lock);
290 void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring)
292 unsigned count_dw_pad;
295 /* We pad to match fetch size */
296 count_dw_pad = (ring->align_mask + 1) -
297 (ring->wptr & ring->align_mask);
298 for (i = 0; i < count_dw_pad; i++) {
299 radeon_ring_write(ring, ring->nop);
302 WREG32(ring->wptr_reg, (ring->wptr << ring->ptr_reg_shift) & ring->ptr_reg_mask);
303 (void)RREG32(ring->wptr_reg);
306 void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring)
308 radeon_ring_commit(rdev, ring);
309 mutex_unlock(&rdev->ring_lock);
312 void radeon_ring_undo(struct radeon_ring *ring)
314 ring->wptr = ring->wptr_old;
317 void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
319 radeon_ring_undo(ring);
320 mutex_unlock(&rdev->ring_lock);
323 void radeon_ring_force_activity(struct radeon_device *rdev, struct radeon_ring *ring)
327 radeon_ring_free_size(rdev, ring);
328 if (ring->rptr == ring->wptr) {
329 r = radeon_ring_alloc(rdev, ring, 1);
331 radeon_ring_write(ring, ring->nop);
332 radeon_ring_commit(rdev, ring);
337 void radeon_ring_lockup_update(struct radeon_ring *ring)
339 ring->last_rptr = ring->rptr;
340 ring->last_activity = jiffies;
344 * radeon_ring_test_lockup() - check if ring is lockedup by recording information
345 * @rdev: radeon device structure
346 * @ring: radeon_ring structure holding ring information
348 * We don't need to initialize the lockup tracking information as we will either
349 * have CP rptr to a different value of jiffies wrap around which will force
350 * initialization of the lockup tracking informations.
352 * A possible false positivie is if we get call after while and last_cp_rptr ==
353 * the current CP rptr, even if it's unlikely it might happen. To avoid this
354 * if the elapsed time since last call is bigger than 2 second than we return
355 * false and update the tracking information. Due to this the caller must call
356 * radeon_ring_test_lockup several time in less than 2sec for lockup to be reported
357 * the fencing code should be cautious about that.
359 * Caller should write to the ring to force CP to do something so we don't get
360 * false positive when CP is just gived nothing to do.
363 bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
365 unsigned long cjiffies, elapsed;
369 if (!time_after(cjiffies, ring->last_activity)) {
370 /* likely a wrap around */
371 radeon_ring_lockup_update(ring);
374 rptr = RREG32(ring->rptr_reg);
375 ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
376 if (ring->rptr != ring->last_rptr) {
377 /* CP is still working no lockup */
378 radeon_ring_lockup_update(ring);
381 elapsed = jiffies_to_msecs(cjiffies - ring->last_activity);
382 if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
383 dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed);
386 /* give a chance to the GPU ... */
390 int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
391 unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg,
392 u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop)
396 ring->ring_size = ring_size;
397 ring->rptr_offs = rptr_offs;
398 ring->rptr_reg = rptr_reg;
399 ring->wptr_reg = wptr_reg;
400 ring->ptr_reg_shift = ptr_reg_shift;
401 ring->ptr_reg_mask = ptr_reg_mask;
403 /* Allocate ring buffer */
404 if (ring->ring_obj == NULL) {
405 r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
406 RADEON_GEM_DOMAIN_GTT,
409 dev_err(rdev->dev, "(%d) ring create failed\n", r);
412 r = radeon_bo_reserve(ring->ring_obj, false);
413 if (unlikely(r != 0))
415 r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
418 radeon_bo_unreserve(ring->ring_obj);
419 dev_err(rdev->dev, "(%d) ring pin failed\n", r);
422 r = radeon_bo_kmap(ring->ring_obj,
423 (void **)&ring->ring);
424 radeon_bo_unreserve(ring->ring_obj);
426 dev_err(rdev->dev, "(%d) ring map failed\n", r);
430 ring->ptr_mask = (ring->ring_size / 4) - 1;
431 ring->ring_free_dw = ring->ring_size / 4;
432 if (radeon_debugfs_ring_init(rdev, ring)) {
433 DRM_ERROR("Failed to register debugfs file for rings !\n");
438 void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
441 struct radeon_bo *ring_obj;
443 mutex_lock(&rdev->ring_lock);
444 ring_obj = ring->ring_obj;
447 ring->ring_obj = NULL;
448 mutex_unlock(&rdev->ring_lock);
451 r = radeon_bo_reserve(ring_obj, false);
452 if (likely(r == 0)) {
453 radeon_bo_kunmap(ring_obj);
454 radeon_bo_unpin(ring_obj);
455 radeon_bo_unreserve(ring_obj);
457 radeon_bo_unref(&ring_obj);
464 #if defined(CONFIG_DEBUG_FS)
466 static int radeon_debugfs_ring_info(struct seq_file *m, void *data)
468 struct drm_info_node *node = (struct drm_info_node *) m->private;
469 struct drm_device *dev = node->minor->dev;
470 struct radeon_device *rdev = dev->dev_private;
471 int ridx = *(int*)node->info_ent->data;
472 struct radeon_ring *ring = &rdev->ring[ridx];
473 unsigned count, i, j;
475 radeon_ring_free_size(rdev, ring);
476 count = (ring->ring_size / 4) - ring->ring_free_dw;
477 seq_printf(m, "wptr(0x%04x): 0x%08x\n", ring->wptr_reg, RREG32(ring->wptr_reg));
478 seq_printf(m, "rptr(0x%04x): 0x%08x\n", ring->rptr_reg, RREG32(ring->rptr_reg));
479 seq_printf(m, "driver's copy of the wptr: 0x%08x\n", ring->wptr);
480 seq_printf(m, "driver's copy of the rptr: 0x%08x\n", ring->rptr);
481 seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
482 seq_printf(m, "%u dwords in ring\n", count);
484 for (j = 0; j <= count; j++) {
485 seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
486 i = (i + 1) & ring->ptr_mask;
491 static int radeon_ring_type_gfx_index = RADEON_RING_TYPE_GFX_INDEX;
492 static int cayman_ring_type_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX;
493 static int cayman_ring_type_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX;
495 static struct drm_info_list radeon_debugfs_ring_info_list[] = {
496 {"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_ring_type_gfx_index},
497 {"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp1_index},
498 {"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_ring_type_cp2_index},
501 static int radeon_debugfs_sa_info(struct seq_file *m, void *data)
503 struct drm_info_node *node = (struct drm_info_node *) m->private;
504 struct drm_device *dev = node->minor->dev;
505 struct radeon_device *rdev = dev->dev_private;
507 radeon_sa_bo_dump_debug_info(&rdev->ring_tmp_bo, m);
513 static struct drm_info_list radeon_debugfs_sa_list[] = {
514 {"radeon_sa_info", &radeon_debugfs_sa_info, 0, NULL},
519 int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
521 #if defined(CONFIG_DEBUG_FS)
523 for (i = 0; i < ARRAY_SIZE(radeon_debugfs_ring_info_list); ++i) {
524 struct drm_info_list *info = &radeon_debugfs_ring_info_list[i];
525 int ridx = *(int*)radeon_debugfs_ring_info_list[i].data;
528 if (&rdev->ring[ridx] != ring)
531 r = radeon_debugfs_add_files(rdev, info, 1);
539 int radeon_debugfs_sa_init(struct radeon_device *rdev)
541 #if defined(CONFIG_DEBUG_FS)
542 return radeon_debugfs_add_files(rdev, radeon_debugfs_sa_list, 1);