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
30 #include <linux/debugfs.h>
32 #include <drm/drm_device.h>
33 #include <drm/drm_file.h>
39 * Most engines on the GPU are fed via ring buffers. Ring
40 * buffers are areas of GPU accessible memory that the host
41 * writes commands into and the GPU reads commands out of.
42 * There is a rptr (read pointer) that determines where the
43 * GPU is currently reading, and a wptr (write pointer)
44 * which determines where the host has written. When the
45 * pointers are equal, the ring is idle. When the host
46 * writes commands to the ring buffer, it increments the
47 * wptr. The GPU then starts fetching commands and executes
48 * them until the pointers are equal again.
50 static void radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
53 * radeon_ring_supports_scratch_reg - check if the ring supports
54 * writing to scratch registers
56 * @rdev: radeon_device pointer
57 * @ring: radeon_ring structure holding ring information
59 * Check if a specific ring supports writing to scratch registers (all asics).
60 * Returns true if the ring supports writing to scratch regs, false if not.
62 bool radeon_ring_supports_scratch_reg(struct radeon_device *rdev,
63 struct radeon_ring *ring)
66 case RADEON_RING_TYPE_GFX_INDEX:
67 case CAYMAN_RING_TYPE_CP1_INDEX:
68 case CAYMAN_RING_TYPE_CP2_INDEX:
76 * radeon_ring_free_size - update the free size
78 * @rdev: radeon_device pointer
79 * @ring: radeon_ring structure holding ring information
81 * Update the free dw slots in the ring buffer (all asics).
83 void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
85 uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
87 /* This works because ring_size is a power of 2 */
88 ring->ring_free_dw = rptr + (ring->ring_size / 4);
89 ring->ring_free_dw -= ring->wptr;
90 ring->ring_free_dw &= ring->ptr_mask;
91 if (!ring->ring_free_dw) {
92 /* this is an empty ring */
93 ring->ring_free_dw = ring->ring_size / 4;
94 /* update lockup info to avoid false positive */
95 radeon_ring_lockup_update(rdev, ring);
100 * radeon_ring_alloc - allocate space on the ring buffer
102 * @rdev: radeon_device pointer
103 * @ring: radeon_ring structure holding ring information
104 * @ndw: number of dwords to allocate in the ring buffer
106 * Allocate @ndw dwords in the ring buffer (all asics).
107 * Returns 0 on success, error on failure.
109 int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
113 /* make sure we aren't trying to allocate more space than there is on the ring */
114 if (ndw > (ring->ring_size / 4))
116 /* Align requested size with padding so unlock_commit can
118 radeon_ring_free_size(rdev, ring);
119 ndw = (ndw + ring->align_mask) & ~ring->align_mask;
120 while (ndw > (ring->ring_free_dw - 1)) {
121 radeon_ring_free_size(rdev, ring);
122 if (ndw < ring->ring_free_dw) {
125 r = radeon_fence_wait_next(rdev, ring->idx);
129 ring->count_dw = ndw;
130 ring->wptr_old = ring->wptr;
135 * radeon_ring_lock - lock the ring and allocate space on it
137 * @rdev: radeon_device pointer
138 * @ring: radeon_ring structure holding ring information
139 * @ndw: number of dwords to allocate in the ring buffer
141 * Lock the ring and allocate @ndw dwords in the ring buffer
143 * Returns 0 on success, error on failure.
145 int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
149 mutex_lock(&rdev->ring_lock);
150 r = radeon_ring_alloc(rdev, ring, ndw);
152 mutex_unlock(&rdev->ring_lock);
159 * radeon_ring_commit - tell the GPU to execute the new
160 * commands on the ring buffer
162 * @rdev: radeon_device pointer
163 * @ring: radeon_ring structure holding ring information
164 * @hdp_flush: Whether or not to perform an HDP cache flush
166 * Update the wptr (write pointer) to tell the GPU to
167 * execute new commands on the ring buffer (all asics).
169 void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring,
172 /* If we are emitting the HDP flush via the ring buffer, we need to
173 * do it before padding.
175 if (hdp_flush && rdev->asic->ring[ring->idx]->hdp_flush)
176 rdev->asic->ring[ring->idx]->hdp_flush(rdev, ring);
177 /* We pad to match fetch size */
178 while (ring->wptr & ring->align_mask) {
179 radeon_ring_write(ring, ring->nop);
182 /* If we are emitting the HDP flush via MMIO, we need to do it after
183 * all CPU writes to VRAM finished.
185 if (hdp_flush && rdev->asic->mmio_hdp_flush)
186 rdev->asic->mmio_hdp_flush(rdev);
187 radeon_ring_set_wptr(rdev, ring);
191 * radeon_ring_unlock_commit - tell the GPU to execute the new
192 * commands on the ring buffer and unlock it
194 * @rdev: radeon_device pointer
195 * @ring: radeon_ring structure holding ring information
196 * @hdp_flush: Whether or not to perform an HDP cache flush
198 * Call radeon_ring_commit() then unlock the ring (all asics).
200 void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring,
203 radeon_ring_commit(rdev, ring, hdp_flush);
204 mutex_unlock(&rdev->ring_lock);
208 * radeon_ring_undo - reset the wptr
210 * @ring: radeon_ring structure holding ring information
212 * Reset the driver's copy of the wptr (all asics).
214 void radeon_ring_undo(struct radeon_ring *ring)
216 ring->wptr = ring->wptr_old;
220 * radeon_ring_unlock_undo - reset the wptr and unlock the ring
222 * @rdev: radeon device structure
223 * @ring: radeon_ring structure holding ring information
225 * Call radeon_ring_undo() then unlock the ring (all asics).
227 void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
229 radeon_ring_undo(ring);
230 mutex_unlock(&rdev->ring_lock);
234 * radeon_ring_lockup_update - update lockup variables
236 * @rdev: radeon device structure
237 * @ring: radeon_ring structure holding ring information
239 * Update the last rptr value and timestamp (all asics).
241 void radeon_ring_lockup_update(struct radeon_device *rdev,
242 struct radeon_ring *ring)
244 atomic_set(&ring->last_rptr, radeon_ring_get_rptr(rdev, ring));
245 atomic64_set(&ring->last_activity, jiffies_64);
249 * radeon_ring_test_lockup() - check if ring is lockedup by recording information
250 * @rdev: radeon device structure
251 * @ring: radeon_ring structure holding ring information
254 bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
256 uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
257 uint64_t last = atomic64_read(&ring->last_activity);
260 if (rptr != atomic_read(&ring->last_rptr)) {
261 /* ring is still working, no lockup */
262 radeon_ring_lockup_update(rdev, ring);
266 elapsed = jiffies_to_msecs(jiffies_64 - last);
267 if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
268 dev_err(rdev->dev, "ring %d stalled for more than %llumsec\n",
272 /* give a chance to the GPU ... */
277 * radeon_ring_backup - Back up the content of a ring
279 * @rdev: radeon_device pointer
280 * @ring: the ring we want to back up
281 * @data: placeholder for returned commit data
283 * Saves all unprocessed commits from a ring, returns the number of dwords saved.
285 unsigned radeon_ring_backup(struct radeon_device *rdev, struct radeon_ring *ring,
288 unsigned size, ptr, i;
290 /* just in case lock the ring */
291 mutex_lock(&rdev->ring_lock);
294 if (ring->ring_obj == NULL) {
295 mutex_unlock(&rdev->ring_lock);
299 /* it doesn't make sense to save anything if all fences are signaled */
300 if (!radeon_fence_count_emitted(rdev, ring->idx)) {
301 mutex_unlock(&rdev->ring_lock);
305 /* calculate the number of dw on the ring */
306 if (ring->rptr_save_reg)
307 ptr = RREG32(ring->rptr_save_reg);
308 else if (rdev->wb.enabled)
309 ptr = le32_to_cpu(*ring->next_rptr_cpu_addr);
311 /* no way to read back the next rptr */
312 mutex_unlock(&rdev->ring_lock);
316 size = ring->wptr + (ring->ring_size / 4);
318 size &= ring->ptr_mask;
320 mutex_unlock(&rdev->ring_lock);
324 /* and then save the content of the ring */
325 *data = kvmalloc_array(size, sizeof(uint32_t), GFP_KERNEL);
327 mutex_unlock(&rdev->ring_lock);
330 for (i = 0; i < size; ++i) {
331 (*data)[i] = ring->ring[ptr++];
332 ptr &= ring->ptr_mask;
335 mutex_unlock(&rdev->ring_lock);
340 * radeon_ring_restore - append saved commands to the ring again
342 * @rdev: radeon_device pointer
343 * @ring: ring to append commands to
344 * @size: number of dwords we want to write
345 * @data: saved commands
347 * Allocates space on the ring and restore the previously saved commands.
349 int radeon_ring_restore(struct radeon_device *rdev, struct radeon_ring *ring,
350 unsigned size, uint32_t *data)
357 /* restore the saved ring content */
358 r = radeon_ring_lock(rdev, ring, size);
362 for (i = 0; i < size; ++i) {
363 radeon_ring_write(ring, data[i]);
366 radeon_ring_unlock_commit(rdev, ring, false);
372 * radeon_ring_init - init driver ring struct.
374 * @rdev: radeon_device pointer
375 * @ring: radeon_ring structure holding ring information
376 * @ring_size: size of the ring
377 * @rptr_offs: offset of the rptr writeback location in the WB buffer
378 * @nop: nop packet for this ring
380 * Initialize the driver information for the selected ring (all asics).
381 * Returns 0 on success, error on failure.
383 int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
384 unsigned rptr_offs, u32 nop)
388 ring->ring_size = ring_size;
389 ring->rptr_offs = rptr_offs;
392 /* Allocate ring buffer */
393 if (ring->ring_obj == NULL) {
394 r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
395 RADEON_GEM_DOMAIN_GTT, 0, NULL,
396 NULL, &ring->ring_obj);
398 dev_err(rdev->dev, "(%d) ring create failed\n", r);
401 r = radeon_bo_reserve(ring->ring_obj, false);
402 if (unlikely(r != 0))
404 r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
407 radeon_bo_unreserve(ring->ring_obj);
408 dev_err(rdev->dev, "(%d) ring pin failed\n", r);
411 r = radeon_bo_kmap(ring->ring_obj,
412 (void **)&ring->ring);
413 radeon_bo_unreserve(ring->ring_obj);
415 dev_err(rdev->dev, "(%d) ring map failed\n", r);
418 radeon_debugfs_ring_init(rdev, ring);
420 ring->ptr_mask = (ring->ring_size / 4) - 1;
421 ring->ring_free_dw = ring->ring_size / 4;
422 if (rdev->wb.enabled) {
423 u32 index = RADEON_WB_RING0_NEXT_RPTR + (ring->idx * 4);
424 ring->next_rptr_gpu_addr = rdev->wb.gpu_addr + index;
425 ring->next_rptr_cpu_addr = &rdev->wb.wb[index/4];
427 radeon_ring_lockup_update(rdev, ring);
432 * radeon_ring_fini - tear down the driver ring struct.
434 * @rdev: radeon_device pointer
435 * @ring: radeon_ring structure holding ring information
437 * Tear down the driver information for the selected ring (all asics).
439 void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
442 struct radeon_bo *ring_obj;
444 mutex_lock(&rdev->ring_lock);
445 ring_obj = ring->ring_obj;
448 ring->ring_obj = NULL;
449 mutex_unlock(&rdev->ring_lock);
452 r = radeon_bo_reserve(ring_obj, false);
453 if (likely(r == 0)) {
454 radeon_bo_kunmap(ring_obj);
455 radeon_bo_unpin(ring_obj);
456 radeon_bo_unreserve(ring_obj);
458 radeon_bo_unref(&ring_obj);
465 #if defined(CONFIG_DEBUG_FS)
467 static int radeon_debugfs_ring_info_show(struct seq_file *m, void *unused)
469 struct radeon_ring *ring = m->private;
470 struct radeon_device *rdev = ring->rdev;
472 uint32_t rptr, wptr, rptr_next;
473 unsigned count, i, j;
475 radeon_ring_free_size(rdev, ring);
476 count = (ring->ring_size / 4) - ring->ring_free_dw;
478 wptr = radeon_ring_get_wptr(rdev, ring);
479 seq_printf(m, "wptr: 0x%08x [%5d]\n",
482 rptr = radeon_ring_get_rptr(rdev, ring);
483 seq_printf(m, "rptr: 0x%08x [%5d]\n",
486 if (ring->rptr_save_reg) {
487 rptr_next = RREG32(ring->rptr_save_reg);
488 seq_printf(m, "rptr next(0x%04x): 0x%08x [%5d]\n",
489 ring->rptr_save_reg, rptr_next, rptr_next);
493 seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n",
494 ring->wptr, ring->wptr);
495 seq_printf(m, "last semaphore signal addr : 0x%016llx\n",
496 ring->last_semaphore_signal_addr);
497 seq_printf(m, "last semaphore wait addr : 0x%016llx\n",
498 ring->last_semaphore_wait_addr);
499 seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
500 seq_printf(m, "%u dwords in ring\n", count);
505 /* print 8 dw before current rptr as often it's the last executed
506 * packet that is the root issue
508 i = (rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask;
509 for (j = 0; j <= (count + 32); j++) {
510 seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]);
516 i = (i + 1) & ring->ptr_mask;
521 DEFINE_SHOW_ATTRIBUTE(radeon_debugfs_ring_info);
523 static const char *radeon_debugfs_ring_idx_to_name(uint32_t ridx)
526 case RADEON_RING_TYPE_GFX_INDEX:
527 return "radeon_ring_gfx";
528 case CAYMAN_RING_TYPE_CP1_INDEX:
529 return "radeon_ring_cp1";
530 case CAYMAN_RING_TYPE_CP2_INDEX:
531 return "radeon_ring_cp2";
532 case R600_RING_TYPE_DMA_INDEX:
533 return "radeon_ring_dma1";
534 case CAYMAN_RING_TYPE_DMA1_INDEX:
535 return "radeon_ring_dma2";
536 case R600_RING_TYPE_UVD_INDEX:
537 return "radeon_ring_uvd";
538 case TN_RING_TYPE_VCE1_INDEX:
539 return "radeon_ring_vce1";
540 case TN_RING_TYPE_VCE2_INDEX:
541 return "radeon_ring_vce2";
549 static void radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
551 #if defined(CONFIG_DEBUG_FS)
552 const char *ring_name = radeon_debugfs_ring_idx_to_name(ring->idx);
553 struct dentry *root = rdev_to_drm(rdev)->primary->debugfs_root;
556 debugfs_create_file(ring_name, 0444, root, ring,
557 &radeon_debugfs_ring_info_fops);