Rev 3764 | Rev 5271 | Go to most recent revision | Show entire file | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 3764 | Rev 5078 | ||
---|---|---|---|
Line 22... | Line 22... | ||
22 | * OTHER DEALINGS IN THE SOFTWARE. |
22 | * OTHER DEALINGS IN THE SOFTWARE. |
23 | * |
23 | * |
24 | * Authors: Dave Airlie |
24 | * Authors: Dave Airlie |
25 | * Alex Deucher |
25 | * Alex Deucher |
26 | * Jerome Glisse |
26 | * Jerome Glisse |
27 | * Christian Konig |
27 | * Christian König |
28 | */ |
28 | */ |
29 | #include |
- | |
30 | #include |
- | |
31 | #include |
29 | #include |
32 | #include |
- | |
33 | #include "radeon_reg.h" |
- | |
34 | #include "radeon.h" |
30 | #include "radeon.h" |
35 | #include "atom.h" |
- | |
36 | - | ||
37 | /* |
- | |
38 | * IB |
- | |
39 | * IBs (Indirect Buffers) and areas of GPU accessible memory where |
- | |
40 | * commands are stored. You can put a pointer to the IB in the |
- | |
41 | * command ring and the hw will fetch the commands from the IB |
- | |
42 | * and execute them. Generally userspace acceleration drivers |
- | |
43 | * produce command buffers which are send to the kernel and |
- | |
44 | * put in IBs for execution by the requested ring. |
- | |
45 | */ |
- | |
46 | static int radeon_debugfs_sa_init(struct radeon_device *rdev); |
- | |
47 | - | ||
48 | /** |
- | |
49 | * radeon_ib_get - request an IB (Indirect Buffer) |
- | |
50 | * |
- | |
51 | * @rdev: radeon_device pointer |
- | |
52 | * @ring: ring index the IB is associated with |
- | |
53 | * @ib: IB object returned |
- | |
54 | * @size: requested IB size |
- | |
55 | * |
- | |
56 | * Request an IB (all asics). IBs are allocated using the |
- | |
57 | * suballocator. |
- | |
58 | * Returns 0 on success, error on failure. |
- | |
59 | */ |
- | |
60 | int radeon_ib_get(struct radeon_device *rdev, int ring, |
- | |
61 | struct radeon_ib *ib, struct radeon_vm *vm, |
- | |
62 | unsigned size) |
- | |
63 | { |
- | |
64 | int i, r; |
- | |
65 | - | ||
66 | r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true); |
- | |
67 | if (r) { |
- | |
68 | dev_err(rdev->dev, "failed to get a new IB (%d)\n", r); |
- | |
69 | return r; |
- | |
70 | } |
- | |
71 | - | ||
72 | r = radeon_semaphore_create(rdev, &ib->semaphore); |
- | |
73 | if (r) { |
- | |
74 | return r; |
- | |
75 | } |
- | |
76 | - | ||
77 | ib->ring = ring; |
- | |
78 | ib->fence = NULL; |
- | |
79 | ib->ptr = radeon_sa_bo_cpu_addr(ib->sa_bo); |
- | |
80 | ib->vm = vm; |
- | |
81 | if (vm) { |
- | |
82 | /* ib pool is bound at RADEON_VA_IB_OFFSET in virtual address |
- | |
83 | * space and soffset is the offset inside the pool bo |
- | |
84 | */ |
- | |
85 | ib->gpu_addr = ib->sa_bo->soffset + RADEON_VA_IB_OFFSET; |
- | |
86 | } else { |
- | |
87 | ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo); |
- | |
88 | } |
- | |
89 | ib->is_const_ib = false; |
- | |
90 | for (i = 0; i < RADEON_NUM_RINGS; ++i) |
- | |
91 | ib->sync_to[i] = NULL; |
- | |
92 | - | ||
93 | return 0; |
- | |
94 | } |
- | |
95 | - | ||
96 | /** |
- | |
97 | * radeon_ib_free - free an IB (Indirect Buffer) |
- | |
98 | * |
- | |
99 | * @rdev: radeon_device pointer |
- | |
100 | * @ib: IB object to free |
- | |
101 | * |
- | |
102 | * Free an IB (all asics). |
- | |
103 | */ |
- | |
104 | void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib) |
- | |
105 | { |
- | |
106 | radeon_semaphore_free(rdev, &ib->semaphore, ib->fence); |
- | |
107 | radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence); |
- | |
108 | radeon_fence_unref(&ib->fence); |
- | |
109 | } |
- | |
110 | - | ||
111 | /** |
- | |
112 | * radeon_ib_sync_to - sync to fence before executing the IB |
- | |
113 | * |
- | |
114 | * @ib: IB object to add fence to |
- | |
115 | * @fence: fence to sync to |
- | |
116 | * |
- | |
117 | * Sync to the fence before executing the IB |
- | |
118 | */ |
- | |
119 | void radeon_ib_sync_to(struct radeon_ib *ib, struct radeon_fence *fence) |
- | |
120 | { |
- | |
121 | struct radeon_fence *other; |
- | |
122 | - | ||
123 | if (!fence) |
- | |
124 | return; |
- | |
125 | - | ||
126 | other = ib->sync_to[fence->ring]; |
- | |
127 | ib->sync_to[fence->ring] = radeon_fence_later(fence, other); |
- | |
128 | } |
- | |
129 | - | ||
130 | /** |
- | |
131 | * radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring |
- | |
132 | * |
- | |
133 | * @rdev: radeon_device pointer |
- | |
134 | * @ib: IB object to schedule |
- | |
135 | * @const_ib: Const IB to schedule (SI only) |
- | |
136 | * |
- | |
137 | * Schedule an IB on the associated ring (all asics). |
- | |
138 | * Returns 0 on success, error on failure. |
- | |
139 | * |
- | |
140 | * On SI, there are two parallel engines fed from the primary ring, |
- | |
141 | * the CE (Constant Engine) and the DE (Drawing Engine). Since |
- | |
142 | * resource descriptors have moved to memory, the CE allows you to |
- | |
143 | * prime the caches while the DE is updating register state so that |
- | |
144 | * the resource descriptors will be already in cache when the draw is |
- | |
145 | * processed. To accomplish this, the userspace driver submits two |
- | |
146 | * IBs, one for the CE and one for the DE. If there is a CE IB (called |
- | |
147 | * a CONST_IB), it will be put on the ring prior to the DE IB. Prior |
- | |
148 | * to SI there was just a DE IB. |
- | |
149 | */ |
- | |
150 | int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib, |
- | |
151 | struct radeon_ib *const_ib) |
- | |
152 | { |
- | |
153 | struct radeon_ring *ring = &rdev->ring[ib->ring]; |
- | |
154 | bool need_sync = false; |
- | |
155 | int i, r = 0; |
- | |
156 | - | ||
157 | if (!ib->length_dw || !ring->ready) { |
- | |
158 | /* TODO: Nothings in the ib we should report. */ |
- | |
159 | dev_err(rdev->dev, "couldn't schedule ib\n"); |
- | |
160 | return -EINVAL; |
- | |
161 | } |
- | |
162 | - | ||
163 | /* 64 dwords should be enough for fence too */ |
- | |
164 | r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_RINGS * 8); |
- | |
165 | if (r) { |
- | |
166 | dev_err(rdev->dev, "scheduling IB failed (%d).\n", r); |
- | |
167 | return r; |
- | |
168 | } |
- | |
169 | for (i = 0; i < RADEON_NUM_RINGS; ++i) { |
- | |
170 | struct radeon_fence *fence = ib->sync_to[i]; |
- | |
171 | if (radeon_fence_need_sync(fence, ib->ring)) { |
- | |
172 | need_sync = true; |
- | |
173 | radeon_semaphore_sync_rings(rdev, ib->semaphore, |
- | |
174 | fence->ring, ib->ring); |
- | |
175 | radeon_fence_note_sync(fence, ib->ring); |
- | |
176 | } |
- | |
177 | } |
- | |
178 | /* immediately free semaphore when we don't need to sync */ |
- | |
179 | if (!need_sync) { |
- | |
180 | radeon_semaphore_free(rdev, &ib->semaphore, NULL); |
- | |
181 | } |
- | |
182 | /* if we can't remember our last VM flush then flush now! */ |
- | |
183 | /* XXX figure out why we have to flush for every IB */ |
- | |
184 | if (ib->vm /*&& !ib->vm->last_flush*/) { |
- | |
185 | radeon_ring_vm_flush(rdev, ib->ring, ib->vm); |
- | |
186 | } |
- | |
187 | if (const_ib) { |
- | |
188 | radeon_ring_ib_execute(rdev, const_ib->ring, const_ib); |
- | |
189 | radeon_semaphore_free(rdev, &const_ib->semaphore, NULL); |
- | |
190 | } |
- | |
191 | radeon_ring_ib_execute(rdev, ib->ring, ib); |
- | |
192 | r = radeon_fence_emit(rdev, &ib->fence, ib->ring); |
- | |
193 | if (r) { |
- | |
194 | dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r); |
- | |
195 | radeon_ring_unlock_undo(rdev, ring); |
- | |
196 | return r; |
- | |
197 | } |
- | |
198 | if (const_ib) { |
- | |
199 | const_ib->fence = radeon_fence_ref(ib->fence); |
- | |
200 | } |
- | |
201 | /* we just flushed the VM, remember that */ |
- | |
202 | if (ib->vm && !ib->vm->last_flush) { |
- | |
203 | ib->vm->last_flush = radeon_fence_ref(ib->fence); |
- | |
204 | } |
- | |
205 | radeon_ring_unlock_commit(rdev, ring); |
- | |
206 | return 0; |
- | |
207 | } |
- | |
208 | - | ||
209 | /** |
- | |
210 | * radeon_ib_pool_init - Init the IB (Indirect Buffer) pool |
- | |
211 | * |
- | |
212 | * @rdev: radeon_device pointer |
- | |
213 | * |
- | |
214 | * Initialize the suballocator to manage a pool of memory |
- | |
215 | * for use as IBs (all asics). |
- | |
216 | * Returns 0 on success, error on failure. |
- | |
217 | */ |
- | |
218 | int radeon_ib_pool_init(struct radeon_device *rdev) |
- | |
219 | { |
- | |
220 | int r; |
- | |
221 | - | ||
222 | if (rdev->ib_pool_ready) { |
- | |
223 | return 0; |
- | |
224 | } |
- | |
225 | r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo, |
- | |
226 | RADEON_IB_POOL_SIZE*64*1024, |
- | |
227 | RADEON_GEM_DOMAIN_GTT); |
- | |
228 | if (r) { |
- | |
229 | return r; |
- | |
230 | } |
- | |
231 | - | ||
232 | r = radeon_sa_bo_manager_start(rdev, &rdev->ring_tmp_bo); |
- | |
233 | if (r) { |
- | |
234 | return r; |
- | |
235 | } |
- | |
236 | - | ||
237 | rdev->ib_pool_ready = true; |
- | |
238 | if (radeon_debugfs_sa_init(rdev)) { |
- | |
239 | dev_err(rdev->dev, "failed to register debugfs file for SA\n"); |
- | |
240 | } |
- | |
241 | return 0; |
- | |
242 | } |
- | |
243 | - | ||
244 | /** |
- | |
245 | * radeon_ib_pool_fini - Free the IB (Indirect Buffer) pool |
- | |
246 | * |
- | |
247 | * @rdev: radeon_device pointer |
- | |
248 | * |
- | |
249 | * Tear down the suballocator managing the pool of memory |
- | |
250 | * for use as IBs (all asics). |
- | |
251 | */ |
- | |
252 | void radeon_ib_pool_fini(struct radeon_device *rdev) |
- | |
253 | { |
- | |
254 | if (rdev->ib_pool_ready) { |
- | |
255 | radeon_sa_bo_manager_suspend(rdev, &rdev->ring_tmp_bo); |
- | |
256 | radeon_sa_bo_manager_fini(rdev, &rdev->ring_tmp_bo); |
- | |
257 | rdev->ib_pool_ready = false; |
- | |
258 | } |
- | |
259 | } |
- | |
260 | - | ||
261 | /** |
- | |
262 | * radeon_ib_ring_tests - test IBs on the rings |
- | |
263 | * |
- | |
264 | * @rdev: radeon_device pointer |
- | |
265 | * |
- | |
266 | * Test an IB (Indirect Buffer) on each ring. |
- | |
267 | * If the test fails, disable the ring. |
- | |
268 | * Returns 0 on success, error if the primary GFX ring |
- | |
269 | * IB test fails. |
- | |
270 | */ |
- | |
271 | int radeon_ib_ring_tests(struct radeon_device *rdev) |
- | |
272 | { |
- | |
273 | unsigned i; |
- | |
274 | int r; |
- | |
275 | - | ||
276 | for (i = 0; i < RADEON_NUM_RINGS; ++i) { |
- | |
277 | struct radeon_ring *ring = &rdev->ring[i]; |
- | |
278 | - | ||
279 | if (!ring->ready) |
- | |
280 | continue; |
- | |
281 | - | ||
282 | r = radeon_ib_test(rdev, i, ring); |
- | |
283 | if (r) { |
- | |
284 | ring->ready = false; |
- | |
285 | - | ||
286 | if (i == RADEON_RING_TYPE_GFX_INDEX) { |
- | |
287 | /* oh, oh, that's really bad */ |
- | |
288 | DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r); |
- | |
289 | rdev->accel_working = false; |
- | |
290 | return r; |
- | |
291 | - | ||
292 | } else { |
- | |
293 | /* still not good, but we can live with it */ |
- | |
294 | DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r); |
- | |
295 | } |
- | |
296 | } |
- | |
297 | } |
- | |
298 | return 0; |
- | |
299 | } |
- | |
Line 300... | Line 31... | ||
300 | 31 | ||
301 | /* |
32 | /* |
302 | * Rings |
33 | * Rings |
303 | * Most engines on the GPU are fed via ring buffers. Ring |
34 | * Most engines on the GPU are fed via ring buffers. Ring |
Line 365... | Line 96... | ||
365 | * |
96 | * |
366 | * Update the free dw slots in the ring buffer (all asics). |
97 | * Update the free dw slots in the ring buffer (all asics). |
367 | */ |
98 | */ |
368 | void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring) |
99 | void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring) |
369 | { |
100 | { |
370 | u32 rptr; |
101 | uint32_t rptr = radeon_ring_get_rptr(rdev, ring); |
Line 371... | Line -... | ||
371 | - | ||
372 | if (rdev->wb.enabled && ring != &rdev->ring[R600_RING_TYPE_UVD_INDEX]) |
- | |
373 | rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]); |
- | |
374 | else |
- | |
375 | rptr = RREG32(ring->rptr_reg); |
- | |
376 | ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift; |
102 | |
377 | /* This works because ring_size is a power of 2 */ |
103 | /* This works because ring_size is a power of 2 */ |
378 | ring->ring_free_dw = (ring->rptr + (ring->ring_size / 4)); |
104 | ring->ring_free_dw = rptr + (ring->ring_size / 4); |
379 | ring->ring_free_dw -= ring->wptr; |
105 | ring->ring_free_dw -= ring->wptr; |
380 | ring->ring_free_dw &= ring->ptr_mask; |
106 | ring->ring_free_dw &= ring->ptr_mask; |
- | 107 | if (!ring->ring_free_dw) { |
|
381 | if (!ring->ring_free_dw) { |
108 | /* this is an empty ring */ |
- | 109 | ring->ring_free_dw = ring->ring_size / 4; |
|
- | 110 | /* update lockup info to avoid false positive */ |
|
382 | ring->ring_free_dw = ring->ring_size / 4; |
111 | radeon_ring_lockup_update(rdev, ring); |
383 | } |
112 | } |
Line 384... | Line 113... | ||
384 | } |
113 | } |
385 | 114 | ||
Line 401... | Line 130... | ||
401 | if (ndw > (ring->ring_size / 4)) |
130 | if (ndw > (ring->ring_size / 4)) |
402 | return -ENOMEM; |
131 | return -ENOMEM; |
403 | /* Align requested size with padding so unlock_commit can |
132 | /* Align requested size with padding so unlock_commit can |
404 | * pad safely */ |
133 | * pad safely */ |
405 | radeon_ring_free_size(rdev, ring); |
134 | radeon_ring_free_size(rdev, ring); |
406 | if (ring->ring_free_dw == (ring->ring_size / 4)) { |
- | |
407 | /* This is an empty ring update lockup info to avoid |
- | |
408 | * false positive. |
- | |
409 | */ |
- | |
410 | radeon_ring_lockup_update(ring); |
- | |
411 | } |
- | |
412 | ndw = (ndw + ring->align_mask) & ~ring->align_mask; |
135 | ndw = (ndw + ring->align_mask) & ~ring->align_mask; |
413 | while (ndw > (ring->ring_free_dw - 1)) { |
136 | while (ndw > (ring->ring_free_dw - 1)) { |
414 | radeon_ring_free_size(rdev, ring); |
137 | radeon_ring_free_size(rdev, ring); |
415 | if (ndw < ring->ring_free_dw) { |
138 | if (ndw < ring->ring_free_dw) { |
416 | break; |
139 | break; |
417 | } |
140 | } |
418 | r = radeon_fence_wait_next_locked(rdev, ring->idx); |
141 | r = radeon_fence_wait_next(rdev, ring->idx); |
419 | if (r) |
142 | if (r) |
420 | return r; |
143 | return r; |
421 | } |
144 | } |
422 | ring->count_dw = ndw; |
145 | ring->count_dw = ndw; |
423 | ring->wptr_old = ring->wptr; |
146 | ring->wptr_old = ring->wptr; |
Line 452... | Line 175... | ||
452 | * radeon_ring_commit - tell the GPU to execute the new |
175 | * radeon_ring_commit - tell the GPU to execute the new |
453 | * commands on the ring buffer |
176 | * commands on the ring buffer |
454 | * |
177 | * |
455 | * @rdev: radeon_device pointer |
178 | * @rdev: radeon_device pointer |
456 | * @ring: radeon_ring structure holding ring information |
179 | * @ring: radeon_ring structure holding ring information |
- | 180 | * @hdp_flush: Whether or not to perform an HDP cache flush |
|
457 | * |
181 | * |
458 | * Update the wptr (write pointer) to tell the GPU to |
182 | * Update the wptr (write pointer) to tell the GPU to |
459 | * execute new commands on the ring buffer (all asics). |
183 | * execute new commands on the ring buffer (all asics). |
460 | */ |
184 | */ |
461 | void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring) |
185 | void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring, |
- | 186 | bool hdp_flush) |
|
462 | { |
187 | { |
- | 188 | /* If we are emitting the HDP flush via the ring buffer, we need to |
|
- | 189 | * do it before padding. |
|
- | 190 | */ |
|
- | 191 | if (hdp_flush && rdev->asic->ring[ring->idx]->hdp_flush) |
|
- | 192 | rdev->asic->ring[ring->idx]->hdp_flush(rdev, ring); |
|
463 | /* We pad to match fetch size */ |
193 | /* We pad to match fetch size */ |
464 | while (ring->wptr & ring->align_mask) { |
194 | while (ring->wptr & ring->align_mask) { |
465 | radeon_ring_write(ring, ring->nop); |
195 | radeon_ring_write(ring, ring->nop); |
466 | } |
196 | } |
467 | DRM_MEMORYBARRIER(); |
197 | mb(); |
468 | WREG32(ring->wptr_reg, (ring->wptr << ring->ptr_reg_shift) & ring->ptr_reg_mask); |
198 | /* If we are emitting the HDP flush via MMIO, we need to do it after |
- | 199 | * all CPU writes to VRAM finished. |
|
- | 200 | */ |
|
- | 201 | if (hdp_flush && rdev->asic->mmio_hdp_flush) |
|
- | 202 | rdev->asic->mmio_hdp_flush(rdev); |
|
469 | (void)RREG32(ring->wptr_reg); |
203 | radeon_ring_set_wptr(rdev, ring); |
470 | } |
204 | } |
Line 471... | Line 205... | ||
471 | 205 | ||
472 | /** |
206 | /** |
473 | * radeon_ring_unlock_commit - tell the GPU to execute the new |
207 | * radeon_ring_unlock_commit - tell the GPU to execute the new |
474 | * commands on the ring buffer and unlock it |
208 | * commands on the ring buffer and unlock it |
475 | * |
209 | * |
476 | * @rdev: radeon_device pointer |
210 | * @rdev: radeon_device pointer |
- | 211 | * @ring: radeon_ring structure holding ring information |
|
477 | * @ring: radeon_ring structure holding ring information |
212 | * @hdp_flush: Whether or not to perform an HDP cache flush |
478 | * |
213 | * |
479 | * Call radeon_ring_commit() then unlock the ring (all asics). |
214 | * Call radeon_ring_commit() then unlock the ring (all asics). |
480 | */ |
215 | */ |
- | 216 | void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring, |
|
481 | void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring) |
217 | bool hdp_flush) |
482 | { |
218 | { |
483 | radeon_ring_commit(rdev, ring); |
219 | radeon_ring_commit(rdev, ring, hdp_flush); |
484 | mutex_unlock(&rdev->ring_lock); |
220 | mutex_unlock(&rdev->ring_lock); |
Line 485... | Line 221... | ||
485 | } |
221 | } |
486 | 222 | ||
Line 508... | Line 244... | ||
508 | radeon_ring_undo(ring); |
244 | radeon_ring_undo(ring); |
509 | mutex_unlock(&rdev->ring_lock); |
245 | mutex_unlock(&rdev->ring_lock); |
510 | } |
246 | } |
Line 511... | Line 247... | ||
511 | 247 | ||
512 | /** |
- | |
513 | * radeon_ring_force_activity - add some nop packets to the ring |
- | |
514 | * |
- | |
515 | * @rdev: radeon_device pointer |
- | |
516 | * @ring: radeon_ring structure holding ring information |
- | |
517 | * |
- | |
518 | * Add some nop packets to the ring to force activity (all asics). |
- | |
519 | * Used for lockup detection to see if the rptr is advancing. |
- | |
520 | */ |
- | |
521 | void radeon_ring_force_activity(struct radeon_device *rdev, struct radeon_ring *ring) |
- | |
522 | { |
- | |
523 | int r; |
- | |
524 | - | ||
525 | radeon_ring_free_size(rdev, ring); |
- | |
526 | if (ring->rptr == ring->wptr) { |
- | |
527 | r = radeon_ring_alloc(rdev, ring, 1); |
- | |
528 | if (!r) { |
- | |
529 | radeon_ring_write(ring, ring->nop); |
- | |
530 | radeon_ring_commit(rdev, ring); |
- | |
531 | } |
- | |
532 | } |
- | |
533 | } |
- | |
534 | - | ||
535 | /** |
248 | /** |
536 | * radeon_ring_lockup_update - update lockup variables |
249 | * radeon_ring_lockup_update - update lockup variables |
537 | * |
250 | * |
538 | * @ring: radeon_ring structure holding ring information |
251 | * @ring: radeon_ring structure holding ring information |
539 | * |
252 | * |
540 | * Update the last rptr value and timestamp (all asics). |
253 | * Update the last rptr value and timestamp (all asics). |
541 | */ |
254 | */ |
- | 255 | void radeon_ring_lockup_update(struct radeon_device *rdev, |
|
542 | void radeon_ring_lockup_update(struct radeon_ring *ring) |
256 | struct radeon_ring *ring) |
543 | { |
257 | { |
544 | ring->last_rptr = ring->rptr; |
258 | atomic_set(&ring->last_rptr, radeon_ring_get_rptr(rdev, ring)); |
545 | ring->last_activity = GetTimerTicks(); |
259 | atomic64_set(&ring->last_activity, jiffies_64); |
Line 546... | Line 260... | ||
546 | } |
260 | } |
547 | 261 | ||
548 | /** |
262 | /** |
549 | * radeon_ring_test_lockup() - check if ring is lockedup by recording information |
263 | * radeon_ring_test_lockup() - check if ring is lockedup by recording information |
550 | * @rdev: radeon device structure |
264 | * @rdev: radeon device structure |
551 | * @ring: radeon_ring structure holding ring information |
- | |
552 | * |
- | |
553 | * We don't need to initialize the lockup tracking information as we will either |
- | |
554 | * have CP rptr to a different value of jiffies wrap around which will force |
- | |
555 | * initialization of the lockup tracking informations. |
- | |
556 | * |
- | |
557 | * A possible false positivie is if we get call after while and last_cp_rptr == |
- | |
558 | * the current CP rptr, even if it's unlikely it might happen. To avoid this |
- | |
559 | * if the elapsed time since last call is bigger than 2 second than we return |
- | |
560 | * false and update the tracking information. Due to this the caller must call |
- | |
561 | * radeon_ring_test_lockup several time in less than 2sec for lockup to be reported |
- | |
562 | * the fencing code should be cautious about that. |
- | |
563 | * |
- | |
564 | * Caller should write to the ring to force CP to do something so we don't get |
- | |
565 | * false positive when CP is just gived nothing to do. |
265 | * @ring: radeon_ring structure holding ring information |
566 | * |
266 | * |
567 | **/ |
267 | */ |
568 | bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring) |
268 | bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring) |
- | 269 | { |
|
569 | { |
270 | uint32_t rptr = radeon_ring_get_rptr(rdev, ring); |
570 | unsigned long cjiffies, elapsed; |
271 | uint64_t last = atomic64_read(&ring->last_activity); |
571 | uint32_t rptr; |
- | |
572 | 272 | uint64_t elapsed; |
|
573 | cjiffies = GetTimerTicks(); |
273 | |
574 | if (!time_after(cjiffies, ring->last_activity)) { |
274 | if (rptr != atomic_read(&ring->last_rptr)) { |
575 | /* likely a wrap around */ |
275 | /* ring is still working, no lockup */ |
576 | radeon_ring_lockup_update(ring); |
276 | radeon_ring_lockup_update(rdev, ring); |
577 | return false; |
- | |
578 | } |
- | |
579 | rptr = RREG32(ring->rptr_reg); |
- | |
580 | ring->rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift; |
- | |
581 | if (ring->rptr != ring->last_rptr) { |
- | |
582 | /* CP is still working no lockup */ |
- | |
583 | radeon_ring_lockup_update(ring); |
277 | return false; |
584 | return false; |
278 | } |
585 | } |
279 | |
586 | elapsed = jiffies_to_msecs(cjiffies - ring->last_activity); |
280 | elapsed = jiffies_to_msecs(jiffies_64 - last); |
- | 281 | if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) { |
|
587 | if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) { |
282 | dev_err(rdev->dev, "ring %d stalled for more than %llumsec\n", |
588 | dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed); |
283 | ring->idx, elapsed); |
589 | return true; |
284 | return true; |
590 | } |
285 | } |
591 | /* give a chance to the GPU ... */ |
286 | /* give a chance to the GPU ... */ |
Line 679... | Line 374... | ||
679 | 374 | ||
680 | for (i = 0; i < size; ++i) { |
375 | for (i = 0; i < size; ++i) { |
681 | radeon_ring_write(ring, data[i]); |
376 | radeon_ring_write(ring, data[i]); |
Line 682... | Line 377... | ||
682 | } |
377 | } |
683 | 378 | ||
684 | radeon_ring_unlock_commit(rdev, ring); |
379 | radeon_ring_unlock_commit(rdev, ring, false); |
685 | kfree(data); |
380 | kfree(data); |
Line 686... | Line 381... | ||
686 | return 0; |
381 | return 0; |
Line 691... | Line 386... | ||
691 | * |
386 | * |
692 | * @rdev: radeon_device pointer |
387 | * @rdev: radeon_device pointer |
693 | * @ring: radeon_ring structure holding ring information |
388 | * @ring: radeon_ring structure holding ring information |
694 | * @ring_size: size of the ring |
389 | * @ring_size: size of the ring |
695 | * @rptr_offs: offset of the rptr writeback location in the WB buffer |
390 | * @rptr_offs: offset of the rptr writeback location in the WB buffer |
696 | * @rptr_reg: MMIO offset of the rptr register |
- | |
697 | * @wptr_reg: MMIO offset of the wptr register |
- | |
698 | * @ptr_reg_shift: bit offset of the rptr/wptr values |
- | |
699 | * @ptr_reg_mask: bit mask of the rptr/wptr values |
- | |
700 | * @nop: nop packet for this ring |
391 | * @nop: nop packet for this ring |
701 | * |
392 | * |
702 | * Initialize the driver information for the selected ring (all asics). |
393 | * Initialize the driver information for the selected ring (all asics). |
703 | * Returns 0 on success, error on failure. |
394 | * Returns 0 on success, error on failure. |
704 | */ |
395 | */ |
705 | int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size, |
396 | int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size, |
706 | unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg, |
- | |
707 | u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop) |
397 | unsigned rptr_offs, u32 nop) |
708 | { |
398 | { |
709 | int r; |
399 | int r; |
Line 710... | Line 400... | ||
710 | 400 | ||
711 | ring->ring_size = ring_size; |
401 | ring->ring_size = ring_size; |
712 | ring->rptr_offs = rptr_offs; |
- | |
713 | ring->rptr_reg = rptr_reg; |
- | |
714 | ring->wptr_reg = wptr_reg; |
- | |
715 | ring->ptr_reg_shift = ptr_reg_shift; |
- | |
716 | ring->ptr_reg_mask = ptr_reg_mask; |
402 | ring->rptr_offs = rptr_offs; |
717 | ring->nop = nop; |
403 | ring->nop = nop; |
718 | /* Allocate ring buffer */ |
404 | /* Allocate ring buffer */ |
719 | if (ring->ring_obj == NULL) { |
405 | if (ring->ring_obj == NULL) { |
720 | r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true, |
406 | r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true, |
721 | RADEON_GEM_DOMAIN_GTT, |
407 | RADEON_GEM_DOMAIN_GTT, 0, |
722 | NULL, &ring->ring_obj); |
408 | NULL, &ring->ring_obj); |
723 | if (r) { |
409 | if (r) { |
724 | dev_err(rdev->dev, "(%d) ring create failed\n", r); |
410 | dev_err(rdev->dev, "(%d) ring create failed\n", r); |
725 | return r; |
411 | return r; |
Line 750... | Line 436... | ||
750 | ring->next_rptr_cpu_addr = &rdev->wb.wb[index/4]; |
436 | ring->next_rptr_cpu_addr = &rdev->wb.wb[index/4]; |
751 | } |
437 | } |
752 | if (radeon_debugfs_ring_init(rdev, ring)) { |
438 | if (radeon_debugfs_ring_init(rdev, ring)) { |
753 | DRM_ERROR("Failed to register debugfs file for rings !\n"); |
439 | DRM_ERROR("Failed to register debugfs file for rings !\n"); |
754 | } |
440 | } |
755 | radeon_ring_lockup_update(ring); |
441 | radeon_ring_lockup_update(rdev, ring); |
756 | return 0; |
442 | return 0; |
757 | } |
443 | } |
Line 758... | Line 444... | ||
758 | 444 | ||
759 | /** |
445 | /** |
Line 797... | Line 483... | ||
797 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
483 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
798 | struct drm_device *dev = node->minor->dev; |
484 | struct drm_device *dev = node->minor->dev; |
799 | struct radeon_device *rdev = dev->dev_private; |
485 | struct radeon_device *rdev = dev->dev_private; |
800 | int ridx = *(int*)node->info_ent->data; |
486 | int ridx = *(int*)node->info_ent->data; |
801 | struct radeon_ring *ring = &rdev->ring[ridx]; |
487 | struct radeon_ring *ring = &rdev->ring[ridx]; |
- | 488 | ||
- | 489 | uint32_t rptr, wptr, rptr_next; |
|
802 | unsigned count, i, j; |
490 | unsigned count, i, j; |
803 | u32 tmp; |
- | |
Line 804... | Line 491... | ||
804 | 491 | ||
805 | radeon_ring_free_size(rdev, ring); |
492 | radeon_ring_free_size(rdev, ring); |
- | 493 | count = (ring->ring_size / 4) - ring->ring_free_dw; |
|
806 | count = (ring->ring_size / 4) - ring->ring_free_dw; |
494 | |
807 | tmp = RREG32(ring->wptr_reg) >> ring->ptr_reg_shift; |
495 | wptr = radeon_ring_get_wptr(rdev, ring); |
- | 496 | seq_printf(m, "wptr: 0x%08x [%5d]\n", |
|
- | 497 | wptr, wptr); |
|
808 | seq_printf(m, "wptr(0x%04x): 0x%08x [%5d]\n", ring->wptr_reg, tmp, tmp); |
498 | |
809 | tmp = RREG32(ring->rptr_reg) >> ring->ptr_reg_shift; |
499 | rptr = radeon_ring_get_rptr(rdev, ring); |
- | 500 | seq_printf(m, "rptr: 0x%08x [%5d]\n", |
|
- | 501 | rptr, rptr); |
|
810 | seq_printf(m, "rptr(0x%04x): 0x%08x [%5d]\n", ring->rptr_reg, tmp, tmp); |
502 | |
- | 503 | if (ring->rptr_save_reg) { |
|
811 | if (ring->rptr_save_reg) { |
504 | rptr_next = RREG32(ring->rptr_save_reg); |
812 | seq_printf(m, "rptr next(0x%04x): 0x%08x\n", ring->rptr_save_reg, |
505 | seq_printf(m, "rptr next(0x%04x): 0x%08x [%5d]\n", |
- | 506 | ring->rptr_save_reg, rptr_next, rptr_next); |
|
- | 507 | } else |
|
813 | RREG32(ring->rptr_save_reg)); |
508 | rptr_next = ~0; |
814 | } |
509 | |
815 | seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n", ring->wptr, ring->wptr); |
510 | seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n", |
816 | seq_printf(m, "driver's copy of the rptr: 0x%08x [%5d]\n", ring->rptr, ring->rptr); |
511 | ring->wptr, ring->wptr); |
- | 512 | seq_printf(m, "last semaphore signal addr : 0x%016llx\n", |
|
817 | seq_printf(m, "last semaphore signal addr : 0x%016llx\n", ring->last_semaphore_signal_addr); |
513 | ring->last_semaphore_signal_addr); |
- | 514 | seq_printf(m, "last semaphore wait addr : 0x%016llx\n", |
|
818 | seq_printf(m, "last semaphore wait addr : 0x%016llx\n", ring->last_semaphore_wait_addr); |
515 | ring->last_semaphore_wait_addr); |
819 | seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw); |
516 | seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw); |
- | 517 | seq_printf(m, "%u dwords in ring\n", count); |
|
- | 518 | ||
- | 519 | if (!ring->ready) |
|
- | 520 | return 0; |
|
820 | seq_printf(m, "%u dwords in ring\n", count); |
521 | |
821 | /* print 8 dw before current rptr as often it's the last executed |
522 | /* print 8 dw before current rptr as often it's the last executed |
822 | * packet that is the root issue |
523 | * packet that is the root issue |
823 | */ |
524 | */ |
824 | i = (ring->rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask; |
525 | i = (rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask; |
825 | for (j = 0; j <= (count + 32); j++) { |
526 | for (j = 0; j <= (count + 32); j++) { |
- | 527 | seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]); |
|
- | 528 | if (rptr == i) |
|
- | 529 | seq_puts(m, " *"); |
|
- | 530 | if (rptr_next == i) |
|
- | 531 | seq_puts(m, " #"); |
|
826 | seq_printf(m, "r[%5d]=0x%08x\n", i, ring->ring[i]); |
532 | seq_puts(m, "\n"); |
827 | i = (i + 1) & ring->ptr_mask; |
533 | i = (i + 1) & ring->ptr_mask; |
828 | } |
534 | } |
829 | return 0; |
535 | return 0; |
Line 833... | Line 539... | ||
833 | static int cayman_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX; |
539 | static int cayman_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX; |
834 | static int cayman_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX; |
540 | static int cayman_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX; |
835 | static int radeon_dma1_index = R600_RING_TYPE_DMA_INDEX; |
541 | static int radeon_dma1_index = R600_RING_TYPE_DMA_INDEX; |
836 | static int radeon_dma2_index = CAYMAN_RING_TYPE_DMA1_INDEX; |
542 | static int radeon_dma2_index = CAYMAN_RING_TYPE_DMA1_INDEX; |
837 | static int r600_uvd_index = R600_RING_TYPE_UVD_INDEX; |
543 | static int r600_uvd_index = R600_RING_TYPE_UVD_INDEX; |
- | 544 | static int si_vce1_index = TN_RING_TYPE_VCE1_INDEX; |
|
- | 545 | static int si_vce2_index = TN_RING_TYPE_VCE2_INDEX; |
|
Line 838... | Line 546... | ||
838 | 546 | ||
839 | static struct drm_info_list radeon_debugfs_ring_info_list[] = { |
547 | static struct drm_info_list radeon_debugfs_ring_info_list[] = { |
840 | {"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_gfx_index}, |
548 | {"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_gfx_index}, |
841 | {"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_cp1_index}, |
549 | {"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_cp1_index}, |
842 | {"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_cp2_index}, |
550 | {"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_cp2_index}, |
843 | {"radeon_ring_dma1", radeon_debugfs_ring_info, 0, &radeon_dma1_index}, |
551 | {"radeon_ring_dma1", radeon_debugfs_ring_info, 0, &radeon_dma1_index}, |
844 | {"radeon_ring_dma2", radeon_debugfs_ring_info, 0, &radeon_dma2_index}, |
552 | {"radeon_ring_dma2", radeon_debugfs_ring_info, 0, &radeon_dma2_index}, |
845 | {"radeon_ring_uvd", radeon_debugfs_ring_info, 0, &r600_uvd_index}, |
- | |
846 | }; |
- | |
847 | 553 | {"radeon_ring_uvd", radeon_debugfs_ring_info, 0, &r600_uvd_index}, |
|
848 | static int radeon_debugfs_sa_info(struct seq_file *m, void *data) |
- | |
849 | { |
- | |
850 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
- | |
851 | struct drm_device *dev = node->minor->dev; |
- | |
852 | struct radeon_device *rdev = dev->dev_private; |
- | |
853 | - | ||
854 | radeon_sa_bo_dump_debug_info(&rdev->ring_tmp_bo, m); |
- | |
855 | - | ||
856 | return 0; |
- | |
857 | - | ||
858 | } |
- | |
859 | - | ||
860 | static struct drm_info_list radeon_debugfs_sa_list[] = { |
554 | {"radeon_ring_vce1", radeon_debugfs_ring_info, 0, &si_vce1_index}, |
861 | {"radeon_sa_info", &radeon_debugfs_sa_info, 0, NULL}, |
555 | {"radeon_ring_vce2", radeon_debugfs_ring_info, 0, &si_vce2_index}, |
Line 862... | Line 556... | ||
862 | }; |
556 | }; |
Line 863... | Line 557... | ||
863 | 557 | ||
Line 880... | Line 574... | ||
880 | return r; |
574 | return r; |
881 | } |
575 | } |
882 | #endif |
576 | #endif |
883 | return 0; |
577 | return 0; |
884 | } |
578 | }>=>>>>=> |
885 | - | ||
886 | static int radeon_debugfs_sa_init(struct radeon_device *rdev) |
- | |
887 | { |
- | |
888 | #if defined(CONFIG_DEBUG_FS) |
- | |
889 | return radeon_debugfs_add_files(rdev, radeon_debugfs_sa_list, 1); |
- | |
890 | #else |
- | |
891 | return 0; |
- | |
892 | #endif |
- | |
893 | }>=>>>><>>=>>>> |
- |