Commit | Line | Data |
---|---|---|
d38ceaf9 AD |
1 | /* |
2 | * Copyright 2008 Advanced Micro Devices, Inc. | |
3 | * Copyright 2008 Red Hat Inc. | |
4 | * Copyright 2009 Jerome Glisse. | |
5 | * | |
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: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
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. | |
23 | * | |
24 | * Authors: Dave Airlie | |
25 | * Alex Deucher | |
26 | * Jerome Glisse | |
27 | * Christian König | |
28 | */ | |
29 | #include <linux/seq_file.h> | |
30 | #include <linux/slab.h> | |
31 | #include <drm/drmP.h> | |
32 | #include <drm/amdgpu_drm.h> | |
33 | #include "amdgpu.h" | |
34 | #include "atom.h" | |
35 | ||
36 | /* | |
37 | * Rings | |
38 | * Most engines on the GPU are fed via ring buffers. Ring | |
39 | * buffers are areas of GPU accessible memory that the host | |
40 | * writes commands into and the GPU reads commands out of. | |
41 | * There is a rptr (read pointer) that determines where the | |
42 | * GPU is currently reading, and a wptr (write pointer) | |
43 | * which determines where the host has written. When the | |
44 | * pointers are equal, the ring is idle. When the host | |
45 | * writes commands to the ring buffer, it increments the | |
46 | * wptr. The GPU then starts fetching commands and executes | |
47 | * them until the pointers are equal again. | |
48 | */ | |
49 | static int amdgpu_debugfs_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring); | |
50 | ||
d38ceaf9 AD |
51 | /** |
52 | * amdgpu_ring_alloc - allocate space on the ring buffer | |
53 | * | |
54 | * @adev: amdgpu_device pointer | |
55 | * @ring: amdgpu_ring structure holding ring information | |
56 | * @ndw: number of dwords to allocate in the ring buffer | |
57 | * | |
58 | * Allocate @ndw dwords in the ring buffer (all asics). | |
59 | * Returns 0 on success, error on failure. | |
60 | */ | |
61 | int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw) | |
62 | { | |
d38ceaf9 AD |
63 | /* Align requested size with padding so unlock_commit can |
64 | * pad safely */ | |
d38ceaf9 | 65 | ndw = (ndw + ring->align_mask) & ~ring->align_mask; |
c7e6be23 CK |
66 | |
67 | /* Make sure we aren't trying to allocate more space | |
68 | * than the maximum for one submission | |
69 | */ | |
70 | if (WARN_ON_ONCE(ndw > ring->max_dw)) | |
71 | return -ENOMEM; | |
72 | ||
d38ceaf9 AD |
73 | ring->count_dw = ndw; |
74 | ring->wptr_old = ring->wptr; | |
75 | return 0; | |
76 | } | |
77 | ||
edff0e28 JZ |
78 | /** amdgpu_ring_insert_nop - insert NOP packets |
79 | * | |
80 | * @ring: amdgpu_ring structure holding ring information | |
81 | * @count: the number of NOP packets to insert | |
82 | * | |
83 | * This is the generic insert_nop function for rings except SDMA | |
84 | */ | |
85 | void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) | |
86 | { | |
87 | int i; | |
88 | ||
89 | for (i = 0; i < count; i++) | |
90 | amdgpu_ring_write(ring, ring->nop); | |
91 | } | |
92 | ||
9e5d5309 CK |
93 | /** amdgpu_ring_generic_pad_ib - pad IB with NOP packets |
94 | * | |
95 | * @ring: amdgpu_ring structure holding ring information | |
96 | * @ib: IB to add NOP packets to | |
97 | * | |
98 | * This is the generic pad_ib function for rings except SDMA | |
99 | */ | |
100 | void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) | |
101 | { | |
102 | while (ib->length_dw & ring->align_mask) | |
103 | ib->ptr[ib->length_dw++] = ring->nop; | |
104 | } | |
105 | ||
d38ceaf9 AD |
106 | /** |
107 | * amdgpu_ring_commit - tell the GPU to execute the new | |
108 | * commands on the ring buffer | |
109 | * | |
110 | * @adev: amdgpu_device pointer | |
111 | * @ring: amdgpu_ring structure holding ring information | |
112 | * | |
113 | * Update the wptr (write pointer) to tell the GPU to | |
114 | * execute new commands on the ring buffer (all asics). | |
115 | */ | |
116 | void amdgpu_ring_commit(struct amdgpu_ring *ring) | |
117 | { | |
edff0e28 JZ |
118 | uint32_t count; |
119 | ||
d38ceaf9 | 120 | /* We pad to match fetch size */ |
edff0e28 JZ |
121 | count = ring->align_mask + 1 - (ring->wptr & ring->align_mask); |
122 | count %= ring->align_mask + 1; | |
123 | ring->funcs->insert_nop(ring, count); | |
124 | ||
d38ceaf9 AD |
125 | mb(); |
126 | amdgpu_ring_set_wptr(ring); | |
127 | } | |
128 | ||
d38ceaf9 AD |
129 | /** |
130 | * amdgpu_ring_undo - reset the wptr | |
131 | * | |
132 | * @ring: amdgpu_ring structure holding ring information | |
133 | * | |
134 | * Reset the driver's copy of the wptr (all asics). | |
135 | */ | |
136 | void amdgpu_ring_undo(struct amdgpu_ring *ring) | |
137 | { | |
138 | ring->wptr = ring->wptr_old; | |
139 | } | |
140 | ||
d38ceaf9 AD |
141 | /** |
142 | * amdgpu_ring_backup - Back up the content of a ring | |
143 | * | |
144 | * @ring: the ring we want to back up | |
145 | * | |
146 | * Saves all unprocessed commits from a ring, returns the number of dwords saved. | |
147 | */ | |
148 | unsigned amdgpu_ring_backup(struct amdgpu_ring *ring, | |
149 | uint32_t **data) | |
150 | { | |
151 | unsigned size, ptr, i; | |
152 | ||
d38ceaf9 AD |
153 | *data = NULL; |
154 | ||
a27de35c | 155 | if (ring->ring_obj == NULL) |
d38ceaf9 | 156 | return 0; |
d38ceaf9 AD |
157 | |
158 | /* it doesn't make sense to save anything if all fences are signaled */ | |
a27de35c | 159 | if (!amdgpu_fence_count_emitted(ring)) |
d38ceaf9 | 160 | return 0; |
d38ceaf9 AD |
161 | |
162 | ptr = le32_to_cpu(*ring->next_rptr_cpu_addr); | |
163 | ||
164 | size = ring->wptr + (ring->ring_size / 4); | |
165 | size -= ptr; | |
166 | size &= ring->ptr_mask; | |
a27de35c | 167 | if (size == 0) |
d38ceaf9 | 168 | return 0; |
d38ceaf9 AD |
169 | |
170 | /* and then save the content of the ring */ | |
171 | *data = kmalloc_array(size, sizeof(uint32_t), GFP_KERNEL); | |
a27de35c | 172 | if (!*data) |
d38ceaf9 | 173 | return 0; |
d38ceaf9 AD |
174 | for (i = 0; i < size; ++i) { |
175 | (*data)[i] = ring->ring[ptr++]; | |
176 | ptr &= ring->ptr_mask; | |
177 | } | |
178 | ||
d38ceaf9 AD |
179 | return size; |
180 | } | |
181 | ||
182 | /** | |
183 | * amdgpu_ring_restore - append saved commands to the ring again | |
184 | * | |
185 | * @ring: ring to append commands to | |
186 | * @size: number of dwords we want to write | |
187 | * @data: saved commands | |
188 | * | |
189 | * Allocates space on the ring and restore the previously saved commands. | |
190 | */ | |
191 | int amdgpu_ring_restore(struct amdgpu_ring *ring, | |
192 | unsigned size, uint32_t *data) | |
193 | { | |
194 | int i, r; | |
195 | ||
196 | if (!size || !data) | |
197 | return 0; | |
198 | ||
199 | /* restore the saved ring content */ | |
a27de35c | 200 | r = amdgpu_ring_alloc(ring, size); |
d38ceaf9 AD |
201 | if (r) |
202 | return r; | |
203 | ||
204 | for (i = 0; i < size; ++i) { | |
205 | amdgpu_ring_write(ring, data[i]); | |
206 | } | |
207 | ||
a27de35c | 208 | amdgpu_ring_commit(ring); |
d38ceaf9 AD |
209 | kfree(data); |
210 | return 0; | |
211 | } | |
212 | ||
213 | /** | |
214 | * amdgpu_ring_init - init driver ring struct. | |
215 | * | |
216 | * @adev: amdgpu_device pointer | |
217 | * @ring: amdgpu_ring structure holding ring information | |
218 | * @ring_size: size of the ring | |
219 | * @nop: nop packet for this ring | |
220 | * | |
221 | * Initialize the driver information for the selected ring (all asics). | |
222 | * Returns 0 on success, error on failure. | |
223 | */ | |
224 | int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, | |
225 | unsigned ring_size, u32 nop, u32 align_mask, | |
226 | struct amdgpu_irq_src *irq_src, unsigned irq_type, | |
227 | enum amdgpu_ring_type ring_type) | |
228 | { | |
229 | u32 rb_bufsz; | |
230 | int r; | |
231 | ||
232 | if (ring->adev == NULL) { | |
233 | if (adev->num_rings >= AMDGPU_MAX_RINGS) | |
234 | return -EINVAL; | |
235 | ||
236 | ring->adev = adev; | |
237 | ring->idx = adev->num_rings++; | |
238 | adev->rings[ring->idx] = ring; | |
e6151a08 CK |
239 | r = amdgpu_fence_driver_init_ring(ring, |
240 | amdgpu_sched_hw_submission); | |
4f839a24 CK |
241 | if (r) |
242 | return r; | |
d38ceaf9 AD |
243 | } |
244 | ||
245 | r = amdgpu_wb_get(adev, &ring->rptr_offs); | |
246 | if (r) { | |
247 | dev_err(adev->dev, "(%d) ring rptr_offs wb alloc failed\n", r); | |
248 | return r; | |
249 | } | |
250 | ||
251 | r = amdgpu_wb_get(adev, &ring->wptr_offs); | |
252 | if (r) { | |
253 | dev_err(adev->dev, "(%d) ring wptr_offs wb alloc failed\n", r); | |
254 | return r; | |
255 | } | |
256 | ||
257 | r = amdgpu_wb_get(adev, &ring->fence_offs); | |
258 | if (r) { | |
259 | dev_err(adev->dev, "(%d) ring fence_offs wb alloc failed\n", r); | |
260 | return r; | |
261 | } | |
262 | ||
263 | r = amdgpu_wb_get(adev, &ring->next_rptr_offs); | |
264 | if (r) { | |
265 | dev_err(adev->dev, "(%d) ring next_rptr wb alloc failed\n", r); | |
266 | return r; | |
267 | } | |
268 | ring->next_rptr_gpu_addr = adev->wb.gpu_addr + (ring->next_rptr_offs * 4); | |
269 | ring->next_rptr_cpu_addr = &adev->wb.wb[ring->next_rptr_offs]; | |
128cff1a ML |
270 | |
271 | r = amdgpu_wb_get(adev, &ring->cond_exe_offs); | |
272 | if (r) { | |
273 | dev_err(adev->dev, "(%d) ring cond_exec_polling wb alloc failed\n", r); | |
274 | return r; | |
275 | } | |
276 | ring->cond_exe_gpu_addr = adev->wb.gpu_addr + (ring->cond_exe_offs * 4); | |
277 | ring->cond_exe_cpu_addr = &adev->wb.wb[ring->cond_exe_offs]; | |
278 | ||
176e1ab1 | 279 | spin_lock_init(&ring->fence_lock); |
d38ceaf9 AD |
280 | r = amdgpu_fence_driver_start_ring(ring, irq_src, irq_type); |
281 | if (r) { | |
282 | dev_err(adev->dev, "failed initializing fences (%d).\n", r); | |
283 | return r; | |
284 | } | |
285 | ||
d38ceaf9 AD |
286 | /* Align ring size */ |
287 | rb_bufsz = order_base_2(ring_size / 8); | |
288 | ring_size = (1 << (rb_bufsz + 1)) * 4; | |
289 | ring->ring_size = ring_size; | |
290 | ring->align_mask = align_mask; | |
291 | ring->nop = nop; | |
292 | ring->type = ring_type; | |
293 | ||
294 | /* Allocate ring buffer */ | |
295 | if (ring->ring_obj == NULL) { | |
296 | r = amdgpu_bo_create(adev, ring->ring_size, PAGE_SIZE, true, | |
297 | AMDGPU_GEM_DOMAIN_GTT, 0, | |
72d7668b | 298 | NULL, NULL, &ring->ring_obj); |
d38ceaf9 AD |
299 | if (r) { |
300 | dev_err(adev->dev, "(%d) ring create failed\n", r); | |
301 | return r; | |
302 | } | |
303 | r = amdgpu_bo_reserve(ring->ring_obj, false); | |
304 | if (unlikely(r != 0)) | |
305 | return r; | |
306 | r = amdgpu_bo_pin(ring->ring_obj, AMDGPU_GEM_DOMAIN_GTT, | |
307 | &ring->gpu_addr); | |
308 | if (r) { | |
309 | amdgpu_bo_unreserve(ring->ring_obj); | |
310 | dev_err(adev->dev, "(%d) ring pin failed\n", r); | |
311 | return r; | |
312 | } | |
313 | r = amdgpu_bo_kmap(ring->ring_obj, | |
314 | (void **)&ring->ring); | |
315 | amdgpu_bo_unreserve(ring->ring_obj); | |
316 | if (r) { | |
317 | dev_err(adev->dev, "(%d) ring map failed\n", r); | |
318 | return r; | |
319 | } | |
320 | } | |
321 | ring->ptr_mask = (ring->ring_size / 4) - 1; | |
c7e6be23 CK |
322 | ring->max_dw = DIV_ROUND_UP(ring->ring_size / 4, |
323 | amdgpu_sched_hw_submission); | |
d38ceaf9 AD |
324 | |
325 | if (amdgpu_debugfs_ring_init(adev, ring)) { | |
326 | DRM_ERROR("Failed to register debugfs file for rings !\n"); | |
327 | } | |
d38ceaf9 AD |
328 | return 0; |
329 | } | |
330 | ||
331 | /** | |
332 | * amdgpu_ring_fini - tear down the driver ring struct. | |
333 | * | |
334 | * @adev: amdgpu_device pointer | |
335 | * @ring: amdgpu_ring structure holding ring information | |
336 | * | |
337 | * Tear down the driver information for the selected ring (all asics). | |
338 | */ | |
339 | void amdgpu_ring_fini(struct amdgpu_ring *ring) | |
340 | { | |
341 | int r; | |
342 | struct amdgpu_bo *ring_obj; | |
343 | ||
d38ceaf9 AD |
344 | ring_obj = ring->ring_obj; |
345 | ring->ready = false; | |
346 | ring->ring = NULL; | |
347 | ring->ring_obj = NULL; | |
d38ceaf9 AD |
348 | |
349 | amdgpu_wb_free(ring->adev, ring->fence_offs); | |
350 | amdgpu_wb_free(ring->adev, ring->rptr_offs); | |
351 | amdgpu_wb_free(ring->adev, ring->wptr_offs); | |
352 | amdgpu_wb_free(ring->adev, ring->next_rptr_offs); | |
353 | ||
354 | if (ring_obj) { | |
355 | r = amdgpu_bo_reserve(ring_obj, false); | |
356 | if (likely(r == 0)) { | |
357 | amdgpu_bo_kunmap(ring_obj); | |
358 | amdgpu_bo_unpin(ring_obj); | |
359 | amdgpu_bo_unreserve(ring_obj); | |
360 | } | |
361 | amdgpu_bo_unref(&ring_obj); | |
362 | } | |
363 | } | |
364 | ||
365 | /* | |
366 | * Debugfs info | |
367 | */ | |
368 | #if defined(CONFIG_DEBUG_FS) | |
369 | ||
370 | static int amdgpu_debugfs_ring_info(struct seq_file *m, void *data) | |
371 | { | |
372 | struct drm_info_node *node = (struct drm_info_node *) m->private; | |
373 | struct drm_device *dev = node->minor->dev; | |
374 | struct amdgpu_device *adev = dev->dev_private; | |
375 | int roffset = *(int*)node->info_ent->data; | |
376 | struct amdgpu_ring *ring = (void *)(((uint8_t*)adev) + roffset); | |
377 | ||
378 | uint32_t rptr, wptr, rptr_next; | |
c7e6be23 | 379 | unsigned i; |
d38ceaf9 AD |
380 | |
381 | wptr = amdgpu_ring_get_wptr(ring); | |
c7e6be23 | 382 | seq_printf(m, "wptr: 0x%08x [%5d]\n", wptr, wptr); |
d38ceaf9 AD |
383 | |
384 | rptr = amdgpu_ring_get_rptr(ring); | |
41f2d990 | 385 | rptr_next = le32_to_cpu(*ring->next_rptr_cpu_addr); |
d38ceaf9 | 386 | |
c7e6be23 CK |
387 | seq_printf(m, "rptr: 0x%08x [%5d]\n", rptr, rptr); |
388 | ||
d38ceaf9 AD |
389 | seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n", |
390 | ring->wptr, ring->wptr); | |
d38ceaf9 AD |
391 | |
392 | if (!ring->ready) | |
393 | return 0; | |
394 | ||
395 | /* print 8 dw before current rptr as often it's the last executed | |
396 | * packet that is the root issue | |
397 | */ | |
398 | i = (rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask; | |
c7e6be23 CK |
399 | while (i != rptr) { |
400 | seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]); | |
401 | if (i == rptr) | |
402 | seq_puts(m, " *"); | |
403 | if (i == rptr_next) | |
404 | seq_puts(m, " #"); | |
405 | seq_puts(m, "\n"); | |
406 | i = (i + 1) & ring->ptr_mask; | |
407 | } | |
408 | while (i != wptr) { | |
d38ceaf9 | 409 | seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]); |
c7e6be23 | 410 | if (i == rptr) |
d38ceaf9 | 411 | seq_puts(m, " *"); |
c7e6be23 | 412 | if (i == rptr_next) |
d38ceaf9 AD |
413 | seq_puts(m, " #"); |
414 | seq_puts(m, "\n"); | |
415 | i = (i + 1) & ring->ptr_mask; | |
416 | } | |
417 | return 0; | |
418 | } | |
419 | ||
420 | /* TODO: clean this up !*/ | |
421 | static int amdgpu_gfx_index = offsetof(struct amdgpu_device, gfx.gfx_ring[0]); | |
422 | static int cayman_cp1_index = offsetof(struct amdgpu_device, gfx.compute_ring[0]); | |
423 | static int cayman_cp2_index = offsetof(struct amdgpu_device, gfx.compute_ring[1]); | |
c113ea1c AD |
424 | static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma.instance[0].ring); |
425 | static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma.instance[1].ring); | |
d38ceaf9 AD |
426 | static int r600_uvd_index = offsetof(struct amdgpu_device, uvd.ring); |
427 | static int si_vce1_index = offsetof(struct amdgpu_device, vce.ring[0]); | |
428 | static int si_vce2_index = offsetof(struct amdgpu_device, vce.ring[1]); | |
429 | ||
06ab6832 | 430 | static const struct drm_info_list amdgpu_debugfs_ring_info_list[] = { |
d38ceaf9 AD |
431 | {"amdgpu_ring_gfx", amdgpu_debugfs_ring_info, 0, &amdgpu_gfx_index}, |
432 | {"amdgpu_ring_cp1", amdgpu_debugfs_ring_info, 0, &cayman_cp1_index}, | |
433 | {"amdgpu_ring_cp2", amdgpu_debugfs_ring_info, 0, &cayman_cp2_index}, | |
434 | {"amdgpu_ring_dma1", amdgpu_debugfs_ring_info, 0, &amdgpu_dma1_index}, | |
435 | {"amdgpu_ring_dma2", amdgpu_debugfs_ring_info, 0, &amdgpu_dma2_index}, | |
436 | {"amdgpu_ring_uvd", amdgpu_debugfs_ring_info, 0, &r600_uvd_index}, | |
437 | {"amdgpu_ring_vce1", amdgpu_debugfs_ring_info, 0, &si_vce1_index}, | |
438 | {"amdgpu_ring_vce2", amdgpu_debugfs_ring_info, 0, &si_vce2_index}, | |
439 | }; | |
440 | ||
441 | #endif | |
442 | ||
443 | static int amdgpu_debugfs_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring) | |
444 | { | |
445 | #if defined(CONFIG_DEBUG_FS) | |
446 | unsigned i; | |
447 | for (i = 0; i < ARRAY_SIZE(amdgpu_debugfs_ring_info_list); ++i) { | |
06ab6832 | 448 | const struct drm_info_list *info = &amdgpu_debugfs_ring_info_list[i]; |
d38ceaf9 AD |
449 | int roffset = *(int*)amdgpu_debugfs_ring_info_list[i].data; |
450 | struct amdgpu_ring *other = (void *)(((uint8_t*)adev) + roffset); | |
451 | unsigned r; | |
452 | ||
453 | if (other != ring) | |
454 | continue; | |
455 | ||
456 | r = amdgpu_debugfs_add_files(adev, info, 1); | |
457 | if (r) | |
458 | return r; | |
459 | } | |
460 | #endif | |
461 | return 0; | |
462 | } |