Commit | Line | Data |
---|---|---|
771fe6b9 JG |
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 | */ | |
28 | #include <linux/seq_file.h> | |
5a0e3ad6 | 29 | #include <linux/slab.h> |
771fe6b9 JG |
30 | #include "drmP.h" |
31 | #include "radeon_drm.h" | |
32 | #include "radeon_reg.h" | |
33 | #include "radeon.h" | |
34 | #include "atom.h" | |
35 | ||
36 | int radeon_debugfs_ib_init(struct radeon_device *rdev); | |
37 | ||
9f93ed39 JG |
38 | void radeon_ib_bogus_cleanup(struct radeon_device *rdev) |
39 | { | |
40 | struct radeon_ib *ib, *n; | |
41 | ||
42 | list_for_each_entry_safe(ib, n, &rdev->ib_pool.bogus_ib, list) { | |
43 | list_del(&ib->list); | |
44 | vfree(ib->ptr); | |
45 | kfree(ib); | |
46 | } | |
47 | } | |
48 | ||
49 | void radeon_ib_bogus_add(struct radeon_device *rdev, struct radeon_ib *ib) | |
50 | { | |
51 | struct radeon_ib *bib; | |
52 | ||
53 | bib = kmalloc(sizeof(*bib), GFP_KERNEL); | |
54 | if (bib == NULL) | |
55 | return; | |
56 | bib->ptr = vmalloc(ib->length_dw * 4); | |
57 | if (bib->ptr == NULL) { | |
58 | kfree(bib); | |
59 | return; | |
60 | } | |
61 | memcpy(bib->ptr, ib->ptr, ib->length_dw * 4); | |
62 | bib->length_dw = ib->length_dw; | |
63 | mutex_lock(&rdev->ib_pool.mutex); | |
64 | list_add_tail(&bib->list, &rdev->ib_pool.bogus_ib); | |
65 | mutex_unlock(&rdev->ib_pool.mutex); | |
66 | } | |
67 | ||
771fe6b9 JG |
68 | /* |
69 | * IB. | |
70 | */ | |
71 | int radeon_ib_get(struct radeon_device *rdev, struct radeon_ib **ib) | |
72 | { | |
73 | struct radeon_fence *fence; | |
74 | struct radeon_ib *nib; | |
91cb91be | 75 | int r = 0, i, c; |
771fe6b9 JG |
76 | |
77 | *ib = NULL; | |
78 | r = radeon_fence_create(rdev, &fence); | |
79 | if (r) { | |
91cb91be | 80 | dev_err(rdev->dev, "failed to create fence for new IB\n"); |
771fe6b9 JG |
81 | return r; |
82 | } | |
83 | mutex_lock(&rdev->ib_pool.mutex); | |
91cb91be JG |
84 | for (i = rdev->ib_pool.head_id, c = 0, nib = NULL; c < RADEON_IB_POOL_SIZE; c++, i++) { |
85 | i &= (RADEON_IB_POOL_SIZE - 1); | |
86 | if (rdev->ib_pool.ibs[i].free) { | |
87 | nib = &rdev->ib_pool.ibs[i]; | |
88 | break; | |
89 | } | |
771fe6b9 | 90 | } |
91cb91be JG |
91 | if (nib == NULL) { |
92 | /* This should never happen, it means we allocated all | |
93 | * IB and haven't scheduled one yet, return EBUSY to | |
94 | * userspace hoping that on ioctl recall we get better | |
95 | * luck | |
96 | */ | |
97 | dev_err(rdev->dev, "no free indirect buffer !\n"); | |
ecb114a1 | 98 | mutex_unlock(&rdev->ib_pool.mutex); |
91cb91be JG |
99 | radeon_fence_unref(&fence); |
100 | return -EBUSY; | |
771fe6b9 | 101 | } |
91cb91be JG |
102 | rdev->ib_pool.head_id = (nib->idx + 1) & (RADEON_IB_POOL_SIZE - 1); |
103 | nib->free = false; | |
104 | if (nib->fence) { | |
ecb114a1 | 105 | mutex_unlock(&rdev->ib_pool.mutex); |
91cb91be JG |
106 | r = radeon_fence_wait(nib->fence, false); |
107 | if (r) { | |
108 | dev_err(rdev->dev, "error waiting fence of IB(%u:0x%016lX:%u)\n", | |
109 | nib->idx, (unsigned long)nib->gpu_addr, nib->length_dw); | |
110 | mutex_lock(&rdev->ib_pool.mutex); | |
111 | nib->free = true; | |
112 | mutex_unlock(&rdev->ib_pool.mutex); | |
113 | radeon_fence_unref(&fence); | |
114 | return r; | |
115 | } | |
116 | mutex_lock(&rdev->ib_pool.mutex); | |
771fe6b9 JG |
117 | } |
118 | radeon_fence_unref(&nib->fence); | |
91cb91be | 119 | nib->fence = fence; |
771fe6b9 | 120 | nib->length_dw = 0; |
ecb114a1 | 121 | mutex_unlock(&rdev->ib_pool.mutex); |
771fe6b9 | 122 | *ib = nib; |
91cb91be | 123 | return 0; |
771fe6b9 JG |
124 | } |
125 | ||
126 | void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib) | |
127 | { | |
128 | struct radeon_ib *tmp = *ib; | |
129 | ||
130 | *ib = NULL; | |
131 | if (tmp == NULL) { | |
132 | return; | |
133 | } | |
7d404c7b JG |
134 | if (!tmp->fence->emited) |
135 | radeon_fence_unref(&tmp->fence); | |
771fe6b9 | 136 | mutex_lock(&rdev->ib_pool.mutex); |
91cb91be | 137 | tmp->free = true; |
771fe6b9 JG |
138 | mutex_unlock(&rdev->ib_pool.mutex); |
139 | } | |
140 | ||
771fe6b9 JG |
141 | int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib) |
142 | { | |
143 | int r = 0; | |
144 | ||
771fe6b9 JG |
145 | if (!ib->length_dw || !rdev->cp.ready) { |
146 | /* TODO: Nothings in the ib we should report. */ | |
91cb91be | 147 | DRM_ERROR("radeon: couldn't schedule IB(%u).\n", ib->idx); |
771fe6b9 JG |
148 | return -EINVAL; |
149 | } | |
ecb114a1 | 150 | |
6cdf6585 | 151 | /* 64 dwords should be enough for fence too */ |
771fe6b9 JG |
152 | r = radeon_ring_lock(rdev, 64); |
153 | if (r) { | |
154 | DRM_ERROR("radeon: scheduling IB failled (%d).\n", r); | |
771fe6b9 JG |
155 | return r; |
156 | } | |
3ce0a23d | 157 | radeon_ring_ib_execute(rdev, ib); |
771fe6b9 | 158 | radeon_fence_emit(rdev, ib->fence); |
ecb114a1 | 159 | mutex_lock(&rdev->ib_pool.mutex); |
91cb91be JG |
160 | /* once scheduled IB is considered free and protected by the fence */ |
161 | ib->free = true; | |
771fe6b9 | 162 | mutex_unlock(&rdev->ib_pool.mutex); |
ecb114a1 | 163 | radeon_ring_unlock_commit(rdev); |
771fe6b9 JG |
164 | return 0; |
165 | } | |
166 | ||
167 | int radeon_ib_pool_init(struct radeon_device *rdev) | |
168 | { | |
169 | void *ptr; | |
170 | uint64_t gpu_addr; | |
171 | int i; | |
172 | int r = 0; | |
173 | ||
9f022ddf JG |
174 | if (rdev->ib_pool.robj) |
175 | return 0; | |
9f93ed39 | 176 | INIT_LIST_HEAD(&rdev->ib_pool.bogus_ib); |
771fe6b9 | 177 | /* Allocate 1M object buffer */ |
4c788679 JG |
178 | r = radeon_bo_create(rdev, NULL, RADEON_IB_POOL_SIZE*64*1024, |
179 | true, RADEON_GEM_DOMAIN_GTT, | |
180 | &rdev->ib_pool.robj); | |
771fe6b9 JG |
181 | if (r) { |
182 | DRM_ERROR("radeon: failed to ib pool (%d).\n", r); | |
183 | return r; | |
184 | } | |
4c788679 JG |
185 | r = radeon_bo_reserve(rdev->ib_pool.robj, false); |
186 | if (unlikely(r != 0)) | |
187 | return r; | |
188 | r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr); | |
771fe6b9 | 189 | if (r) { |
4c788679 | 190 | radeon_bo_unreserve(rdev->ib_pool.robj); |
771fe6b9 JG |
191 | DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r); |
192 | return r; | |
193 | } | |
4c788679 JG |
194 | r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr); |
195 | radeon_bo_unreserve(rdev->ib_pool.robj); | |
771fe6b9 JG |
196 | if (r) { |
197 | DRM_ERROR("radeon: failed to map ib poll (%d).\n", r); | |
198 | return r; | |
199 | } | |
200 | for (i = 0; i < RADEON_IB_POOL_SIZE; i++) { | |
201 | unsigned offset; | |
202 | ||
203 | offset = i * 64 * 1024; | |
204 | rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset; | |
205 | rdev->ib_pool.ibs[i].ptr = ptr + offset; | |
206 | rdev->ib_pool.ibs[i].idx = i; | |
207 | rdev->ib_pool.ibs[i].length_dw = 0; | |
91cb91be | 208 | rdev->ib_pool.ibs[i].free = true; |
771fe6b9 | 209 | } |
91cb91be | 210 | rdev->ib_pool.head_id = 0; |
771fe6b9 JG |
211 | rdev->ib_pool.ready = true; |
212 | DRM_INFO("radeon: ib pool ready.\n"); | |
213 | if (radeon_debugfs_ib_init(rdev)) { | |
214 | DRM_ERROR("Failed to register debugfs file for IB !\n"); | |
215 | } | |
216 | return r; | |
217 | } | |
218 | ||
219 | void radeon_ib_pool_fini(struct radeon_device *rdev) | |
220 | { | |
4c788679 JG |
221 | int r; |
222 | ||
771fe6b9 JG |
223 | if (!rdev->ib_pool.ready) { |
224 | return; | |
225 | } | |
226 | mutex_lock(&rdev->ib_pool.mutex); | |
9f93ed39 | 227 | radeon_ib_bogus_cleanup(rdev); |
eb6b6d7c | 228 | |
771fe6b9 | 229 | if (rdev->ib_pool.robj) { |
4c788679 JG |
230 | r = radeon_bo_reserve(rdev->ib_pool.robj, false); |
231 | if (likely(r == 0)) { | |
232 | radeon_bo_kunmap(rdev->ib_pool.robj); | |
233 | radeon_bo_unpin(rdev->ib_pool.robj); | |
234 | radeon_bo_unreserve(rdev->ib_pool.robj); | |
235 | } | |
236 | radeon_bo_unref(&rdev->ib_pool.robj); | |
771fe6b9 JG |
237 | rdev->ib_pool.robj = NULL; |
238 | } | |
239 | mutex_unlock(&rdev->ib_pool.mutex); | |
240 | } | |
241 | ||
771fe6b9 JG |
242 | |
243 | /* | |
244 | * Ring. | |
245 | */ | |
246 | void radeon_ring_free_size(struct radeon_device *rdev) | |
247 | { | |
3ce0a23d JG |
248 | if (rdev->family >= CHIP_R600) |
249 | rdev->cp.rptr = RREG32(R600_CP_RB_RPTR); | |
250 | else | |
251 | rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR); | |
771fe6b9 JG |
252 | /* This works because ring_size is a power of 2 */ |
253 | rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4)); | |
254 | rdev->cp.ring_free_dw -= rdev->cp.wptr; | |
255 | rdev->cp.ring_free_dw &= rdev->cp.ptr_mask; | |
256 | if (!rdev->cp.ring_free_dw) { | |
257 | rdev->cp.ring_free_dw = rdev->cp.ring_size / 4; | |
258 | } | |
259 | } | |
260 | ||
261 | int radeon_ring_lock(struct radeon_device *rdev, unsigned ndw) | |
262 | { | |
263 | int r; | |
264 | ||
265 | /* Align requested size with padding so unlock_commit can | |
266 | * pad safely */ | |
267 | ndw = (ndw + rdev->cp.align_mask) & ~rdev->cp.align_mask; | |
268 | mutex_lock(&rdev->cp.mutex); | |
269 | while (ndw > (rdev->cp.ring_free_dw - 1)) { | |
270 | radeon_ring_free_size(rdev); | |
271 | if (ndw < rdev->cp.ring_free_dw) { | |
272 | break; | |
273 | } | |
274 | r = radeon_fence_wait_next(rdev); | |
275 | if (r) { | |
276 | mutex_unlock(&rdev->cp.mutex); | |
277 | return r; | |
278 | } | |
279 | } | |
280 | rdev->cp.count_dw = ndw; | |
281 | rdev->cp.wptr_old = rdev->cp.wptr; | |
282 | return 0; | |
283 | } | |
284 | ||
285 | void radeon_ring_unlock_commit(struct radeon_device *rdev) | |
286 | { | |
287 | unsigned count_dw_pad; | |
288 | unsigned i; | |
289 | ||
290 | /* We pad to match fetch size */ | |
291 | count_dw_pad = (rdev->cp.align_mask + 1) - | |
292 | (rdev->cp.wptr & rdev->cp.align_mask); | |
293 | for (i = 0; i < count_dw_pad; i++) { | |
3ce0a23d | 294 | radeon_ring_write(rdev, 2 << 30); |
771fe6b9 JG |
295 | } |
296 | DRM_MEMORYBARRIER(); | |
3ce0a23d | 297 | radeon_cp_commit(rdev); |
771fe6b9 JG |
298 | mutex_unlock(&rdev->cp.mutex); |
299 | } | |
300 | ||
301 | void radeon_ring_unlock_undo(struct radeon_device *rdev) | |
302 | { | |
303 | rdev->cp.wptr = rdev->cp.wptr_old; | |
304 | mutex_unlock(&rdev->cp.mutex); | |
305 | } | |
306 | ||
771fe6b9 JG |
307 | int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size) |
308 | { | |
309 | int r; | |
310 | ||
311 | rdev->cp.ring_size = ring_size; | |
312 | /* Allocate ring buffer */ | |
313 | if (rdev->cp.ring_obj == NULL) { | |
4c788679 JG |
314 | r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true, |
315 | RADEON_GEM_DOMAIN_GTT, | |
316 | &rdev->cp.ring_obj); | |
771fe6b9 | 317 | if (r) { |
4c788679 | 318 | dev_err(rdev->dev, "(%d) ring create failed\n", r); |
771fe6b9 JG |
319 | return r; |
320 | } | |
4c788679 JG |
321 | r = radeon_bo_reserve(rdev->cp.ring_obj, false); |
322 | if (unlikely(r != 0)) | |
323 | return r; | |
324 | r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT, | |
325 | &rdev->cp.gpu_addr); | |
771fe6b9 | 326 | if (r) { |
4c788679 JG |
327 | radeon_bo_unreserve(rdev->cp.ring_obj); |
328 | dev_err(rdev->dev, "(%d) ring pin failed\n", r); | |
771fe6b9 JG |
329 | return r; |
330 | } | |
4c788679 | 331 | r = radeon_bo_kmap(rdev->cp.ring_obj, |
771fe6b9 | 332 | (void **)&rdev->cp.ring); |
4c788679 | 333 | radeon_bo_unreserve(rdev->cp.ring_obj); |
771fe6b9 | 334 | if (r) { |
4c788679 | 335 | dev_err(rdev->dev, "(%d) ring map failed\n", r); |
771fe6b9 JG |
336 | return r; |
337 | } | |
338 | } | |
339 | rdev->cp.ptr_mask = (rdev->cp.ring_size / 4) - 1; | |
340 | rdev->cp.ring_free_dw = rdev->cp.ring_size / 4; | |
341 | return 0; | |
342 | } | |
343 | ||
344 | void radeon_ring_fini(struct radeon_device *rdev) | |
345 | { | |
4c788679 JG |
346 | int r; |
347 | ||
771fe6b9 JG |
348 | mutex_lock(&rdev->cp.mutex); |
349 | if (rdev->cp.ring_obj) { | |
4c788679 JG |
350 | r = radeon_bo_reserve(rdev->cp.ring_obj, false); |
351 | if (likely(r == 0)) { | |
352 | radeon_bo_kunmap(rdev->cp.ring_obj); | |
353 | radeon_bo_unpin(rdev->cp.ring_obj); | |
354 | radeon_bo_unreserve(rdev->cp.ring_obj); | |
355 | } | |
356 | radeon_bo_unref(&rdev->cp.ring_obj); | |
771fe6b9 JG |
357 | rdev->cp.ring = NULL; |
358 | rdev->cp.ring_obj = NULL; | |
359 | } | |
360 | mutex_unlock(&rdev->cp.mutex); | |
361 | } | |
362 | ||
363 | ||
364 | /* | |
365 | * Debugfs info | |
366 | */ | |
367 | #if defined(CONFIG_DEBUG_FS) | |
368 | static int radeon_debugfs_ib_info(struct seq_file *m, void *data) | |
369 | { | |
370 | struct drm_info_node *node = (struct drm_info_node *) m->private; | |
371 | struct radeon_ib *ib = node->info_ent->data; | |
372 | unsigned i; | |
373 | ||
374 | if (ib == NULL) { | |
375 | return 0; | |
376 | } | |
91cb91be | 377 | seq_printf(m, "IB %04u\n", ib->idx); |
771fe6b9 JG |
378 | seq_printf(m, "IB fence %p\n", ib->fence); |
379 | seq_printf(m, "IB size %05u dwords\n", ib->length_dw); | |
380 | for (i = 0; i < ib->length_dw; i++) { | |
381 | seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]); | |
382 | } | |
383 | return 0; | |
384 | } | |
385 | ||
9f93ed39 JG |
386 | static int radeon_debugfs_ib_bogus_info(struct seq_file *m, void *data) |
387 | { | |
388 | struct drm_info_node *node = (struct drm_info_node *) m->private; | |
389 | struct radeon_device *rdev = node->info_ent->data; | |
390 | struct radeon_ib *ib; | |
391 | unsigned i; | |
392 | ||
393 | mutex_lock(&rdev->ib_pool.mutex); | |
394 | if (list_empty(&rdev->ib_pool.bogus_ib)) { | |
395 | mutex_unlock(&rdev->ib_pool.mutex); | |
396 | seq_printf(m, "no bogus IB recorded\n"); | |
397 | return 0; | |
398 | } | |
399 | ib = list_first_entry(&rdev->ib_pool.bogus_ib, struct radeon_ib, list); | |
400 | list_del_init(&ib->list); | |
401 | mutex_unlock(&rdev->ib_pool.mutex); | |
402 | seq_printf(m, "IB size %05u dwords\n", ib->length_dw); | |
403 | for (i = 0; i < ib->length_dw; i++) { | |
404 | seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]); | |
405 | } | |
406 | vfree(ib->ptr); | |
407 | kfree(ib); | |
408 | return 0; | |
409 | } | |
410 | ||
771fe6b9 JG |
411 | static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE]; |
412 | static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32]; | |
9f93ed39 JG |
413 | |
414 | static struct drm_info_list radeon_debugfs_ib_bogus_info_list[] = { | |
415 | {"radeon_ib_bogus", radeon_debugfs_ib_bogus_info, 0, NULL}, | |
416 | }; | |
771fe6b9 JG |
417 | #endif |
418 | ||
419 | int radeon_debugfs_ib_init(struct radeon_device *rdev) | |
420 | { | |
421 | #if defined(CONFIG_DEBUG_FS) | |
422 | unsigned i; | |
9f93ed39 | 423 | int r; |
771fe6b9 | 424 | |
9f93ed39 JG |
425 | radeon_debugfs_ib_bogus_info_list[0].data = rdev; |
426 | r = radeon_debugfs_add_files(rdev, radeon_debugfs_ib_bogus_info_list, 1); | |
427 | if (r) | |
428 | return r; | |
771fe6b9 JG |
429 | for (i = 0; i < RADEON_IB_POOL_SIZE; i++) { |
430 | sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i); | |
431 | radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i]; | |
432 | radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info; | |
433 | radeon_debugfs_ib_list[i].driver_features = 0; | |
434 | radeon_debugfs_ib_list[i].data = &rdev->ib_pool.ibs[i]; | |
435 | } | |
436 | return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list, | |
437 | RADEON_IB_POOL_SIZE); | |
438 | #else | |
439 | return 0; | |
440 | #endif | |
441 | } |