Rev 5271 | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5078 | serge | 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 |
||
30 | #include "radeon.h" |
||
31 | |||
32 | /* |
||
33 | * IB |
||
34 | * IBs (Indirect Buffers) and areas of GPU accessible memory where |
||
35 | * commands are stored. You can put a pointer to the IB in the |
||
36 | * command ring and the hw will fetch the commands from the IB |
||
37 | * and execute them. Generally userspace acceleration drivers |
||
38 | * produce command buffers which are send to the kernel and |
||
39 | * put in IBs for execution by the requested ring. |
||
40 | */ |
||
41 | static int radeon_debugfs_sa_init(struct radeon_device *rdev); |
||
42 | |||
43 | /** |
||
44 | * radeon_ib_get - request an IB (Indirect Buffer) |
||
45 | * |
||
46 | * @rdev: radeon_device pointer |
||
47 | * @ring: ring index the IB is associated with |
||
48 | * @ib: IB object returned |
||
49 | * @size: requested IB size |
||
50 | * |
||
51 | * Request an IB (all asics). IBs are allocated using the |
||
52 | * suballocator. |
||
53 | * Returns 0 on success, error on failure. |
||
54 | */ |
||
55 | int radeon_ib_get(struct radeon_device *rdev, int ring, |
||
56 | struct radeon_ib *ib, struct radeon_vm *vm, |
||
57 | unsigned size) |
||
58 | { |
||
59 | int r; |
||
60 | |||
61 | r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256); |
||
62 | if (r) { |
||
63 | dev_err(rdev->dev, "failed to get a new IB (%d)\n", r); |
||
64 | return r; |
||
65 | } |
||
66 | |||
5271 | serge | 67 | radeon_sync_create(&ib->sync); |
5078 | serge | 68 | |
69 | ib->ring = ring; |
||
70 | ib->fence = NULL; |
||
71 | ib->ptr = radeon_sa_bo_cpu_addr(ib->sa_bo); |
||
72 | ib->vm = vm; |
||
73 | if (vm) { |
||
74 | /* ib pool is bound at RADEON_VA_IB_OFFSET in virtual address |
||
75 | * space and soffset is the offset inside the pool bo |
||
76 | */ |
||
77 | ib->gpu_addr = ib->sa_bo->soffset + RADEON_VA_IB_OFFSET; |
||
78 | } else { |
||
79 | ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo); |
||
80 | } |
||
81 | ib->is_const_ib = false; |
||
82 | |||
83 | return 0; |
||
84 | } |
||
85 | |||
86 | /** |
||
87 | * radeon_ib_free - free an IB (Indirect Buffer) |
||
88 | * |
||
89 | * @rdev: radeon_device pointer |
||
90 | * @ib: IB object to free |
||
91 | * |
||
92 | * Free an IB (all asics). |
||
93 | */ |
||
94 | void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib) |
||
95 | { |
||
5271 | serge | 96 | radeon_sync_free(rdev, &ib->sync, ib->fence); |
5078 | serge | 97 | radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence); |
98 | radeon_fence_unref(&ib->fence); |
||
99 | } |
||
100 | |||
101 | /** |
||
102 | * radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring |
||
103 | * |
||
104 | * @rdev: radeon_device pointer |
||
105 | * @ib: IB object to schedule |
||
106 | * @const_ib: Const IB to schedule (SI only) |
||
107 | * @hdp_flush: Whether or not to perform an HDP cache flush |
||
108 | * |
||
109 | * Schedule an IB on the associated ring (all asics). |
||
110 | * Returns 0 on success, error on failure. |
||
111 | * |
||
112 | * On SI, there are two parallel engines fed from the primary ring, |
||
113 | * the CE (Constant Engine) and the DE (Drawing Engine). Since |
||
114 | * resource descriptors have moved to memory, the CE allows you to |
||
115 | * prime the caches while the DE is updating register state so that |
||
116 | * the resource descriptors will be already in cache when the draw is |
||
117 | * processed. To accomplish this, the userspace driver submits two |
||
118 | * IBs, one for the CE and one for the DE. If there is a CE IB (called |
||
119 | * a CONST_IB), it will be put on the ring prior to the DE IB. Prior |
||
120 | * to SI there was just a DE IB. |
||
121 | */ |
||
122 | int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib, |
||
123 | struct radeon_ib *const_ib, bool hdp_flush) |
||
124 | { |
||
125 | struct radeon_ring *ring = &rdev->ring[ib->ring]; |
||
126 | int r = 0; |
||
127 | |||
128 | if (!ib->length_dw || !ring->ready) { |
||
129 | /* TODO: Nothings in the ib we should report. */ |
||
130 | dev_err(rdev->dev, "couldn't schedule ib\n"); |
||
131 | return -EINVAL; |
||
132 | } |
||
133 | |||
134 | /* 64 dwords should be enough for fence too */ |
||
135 | r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_SYNCS * 8); |
||
136 | if (r) { |
||
137 | dev_err(rdev->dev, "scheduling IB failed (%d).\n", r); |
||
138 | return r; |
||
139 | } |
||
140 | |||
141 | /* grab a vm id if necessary */ |
||
142 | if (ib->vm) { |
||
143 | struct radeon_fence *vm_id_fence; |
||
144 | vm_id_fence = radeon_vm_grab_id(rdev, ib->vm, ib->ring); |
||
5271 | serge | 145 | radeon_sync_fence(&ib->sync, vm_id_fence); |
5078 | serge | 146 | } |
147 | |||
148 | /* sync with other rings */ |
||
5271 | serge | 149 | r = radeon_sync_rings(rdev, &ib->sync, ib->ring); |
5078 | serge | 150 | if (r) { |
151 | dev_err(rdev->dev, "failed to sync rings (%d)\n", r); |
||
152 | radeon_ring_unlock_undo(rdev, ring); |
||
153 | return r; |
||
154 | } |
||
155 | |||
156 | if (ib->vm) |
||
5271 | serge | 157 | radeon_vm_flush(rdev, ib->vm, ib->ring, |
158 | ib->sync.last_vm_update); |
||
5078 | serge | 159 | |
160 | if (const_ib) { |
||
161 | radeon_ring_ib_execute(rdev, const_ib->ring, const_ib); |
||
5271 | serge | 162 | radeon_sync_free(rdev, &const_ib->sync, NULL); |
5078 | serge | 163 | } |
164 | radeon_ring_ib_execute(rdev, ib->ring, ib); |
||
165 | r = radeon_fence_emit(rdev, &ib->fence, ib->ring); |
||
166 | if (r) { |
||
167 | dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r); |
||
168 | radeon_ring_unlock_undo(rdev, ring); |
||
169 | return r; |
||
170 | } |
||
171 | if (const_ib) { |
||
172 | const_ib->fence = radeon_fence_ref(ib->fence); |
||
173 | } |
||
174 | |||
175 | if (ib->vm) |
||
176 | radeon_vm_fence(rdev, ib->vm, ib->fence); |
||
177 | |||
178 | radeon_ring_unlock_commit(rdev, ring, hdp_flush); |
||
179 | return 0; |
||
180 | } |
||
181 | |||
182 | /** |
||
183 | * radeon_ib_pool_init - Init the IB (Indirect Buffer) pool |
||
184 | * |
||
185 | * @rdev: radeon_device pointer |
||
186 | * |
||
187 | * Initialize the suballocator to manage a pool of memory |
||
188 | * for use as IBs (all asics). |
||
189 | * Returns 0 on success, error on failure. |
||
190 | */ |
||
191 | int radeon_ib_pool_init(struct radeon_device *rdev) |
||
192 | { |
||
193 | int r; |
||
194 | |||
195 | if (rdev->ib_pool_ready) { |
||
196 | return 0; |
||
197 | } |
||
198 | |||
199 | if (rdev->family >= CHIP_BONAIRE) { |
||
200 | r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo, |
||
201 | RADEON_IB_POOL_SIZE*64*1024, |
||
202 | RADEON_GPU_PAGE_SIZE, |
||
203 | RADEON_GEM_DOMAIN_GTT, |
||
204 | RADEON_GEM_GTT_WC); |
||
205 | } else { |
||
206 | /* Before CIK, it's better to stick to cacheable GTT due |
||
207 | * to the command stream checking |
||
208 | */ |
||
209 | r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo, |
||
210 | RADEON_IB_POOL_SIZE*64*1024, |
||
211 | RADEON_GPU_PAGE_SIZE, |
||
212 | RADEON_GEM_DOMAIN_GTT, 0); |
||
213 | } |
||
214 | if (r) { |
||
215 | return r; |
||
216 | } |
||
217 | |||
218 | r = radeon_sa_bo_manager_start(rdev, &rdev->ring_tmp_bo); |
||
219 | if (r) { |
||
220 | return r; |
||
221 | } |
||
222 | |||
223 | rdev->ib_pool_ready = true; |
||
224 | if (radeon_debugfs_sa_init(rdev)) { |
||
225 | dev_err(rdev->dev, "failed to register debugfs file for SA\n"); |
||
226 | } |
||
227 | return 0; |
||
228 | } |
||
229 | |||
230 | /** |
||
231 | * radeon_ib_pool_fini - Free the IB (Indirect Buffer) pool |
||
232 | * |
||
233 | * @rdev: radeon_device pointer |
||
234 | * |
||
235 | * Tear down the suballocator managing the pool of memory |
||
236 | * for use as IBs (all asics). |
||
237 | */ |
||
238 | void radeon_ib_pool_fini(struct radeon_device *rdev) |
||
239 | { |
||
240 | if (rdev->ib_pool_ready) { |
||
241 | radeon_sa_bo_manager_suspend(rdev, &rdev->ring_tmp_bo); |
||
242 | radeon_sa_bo_manager_fini(rdev, &rdev->ring_tmp_bo); |
||
243 | rdev->ib_pool_ready = false; |
||
244 | } |
||
245 | } |
||
246 | |||
247 | /** |
||
248 | * radeon_ib_ring_tests - test IBs on the rings |
||
249 | * |
||
250 | * @rdev: radeon_device pointer |
||
251 | * |
||
252 | * Test an IB (Indirect Buffer) on each ring. |
||
253 | * If the test fails, disable the ring. |
||
254 | * Returns 0 on success, error if the primary GFX ring |
||
255 | * IB test fails. |
||
256 | */ |
||
257 | int radeon_ib_ring_tests(struct radeon_device *rdev) |
||
258 | { |
||
259 | unsigned i; |
||
260 | int r; |
||
261 | |||
262 | for (i = 0; i < RADEON_NUM_RINGS; ++i) { |
||
263 | struct radeon_ring *ring = &rdev->ring[i]; |
||
264 | |||
265 | if (!ring->ready) |
||
266 | continue; |
||
267 | |||
268 | r = radeon_ib_test(rdev, i, ring); |
||
269 | if (r) { |
||
5271 | serge | 270 | radeon_fence_driver_force_completion(rdev, i); |
5078 | serge | 271 | ring->ready = false; |
272 | rdev->needs_reset = false; |
||
273 | |||
274 | if (i == RADEON_RING_TYPE_GFX_INDEX) { |
||
275 | /* oh, oh, that's really bad */ |
||
276 | DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r); |
||
7146 | serge | 277 | rdev->accel_working = false; |
5078 | serge | 278 | return r; |
279 | |||
280 | } else { |
||
281 | /* still not good, but we can live with it */ |
||
282 | DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r); |
||
283 | } |
||
284 | } |
||
285 | } |
||
286 | return 0; |
||
287 | } |
||
288 | |||
289 | /* |
||
290 | * Debugfs info |
||
291 | */ |
||
292 | #if defined(CONFIG_DEBUG_FS) |
||
293 | |||
294 | static int radeon_debugfs_sa_info(struct seq_file *m, void *data) |
||
295 | { |
||
296 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
||
297 | struct drm_device *dev = node->minor->dev; |
||
298 | struct radeon_device *rdev = dev->dev_private; |
||
299 | |||
300 | radeon_sa_bo_dump_debug_info(&rdev->ring_tmp_bo, m); |
||
301 | |||
302 | return 0; |
||
303 | |||
304 | } |
||
305 | |||
306 | static struct drm_info_list radeon_debugfs_sa_list[] = { |
||
7146 | serge | 307 | {"radeon_sa_info", &radeon_debugfs_sa_info, 0, NULL}, |
5078 | serge | 308 | }; |
309 | |||
310 | #endif |
||
311 | |||
312 | static int radeon_debugfs_sa_init(struct radeon_device *rdev) |
||
313 | { |
||
314 | #if defined(CONFIG_DEBUG_FS) |
||
315 | return radeon_debugfs_add_files(rdev, radeon_debugfs_sa_list, 1); |
||
316 | #else |
||
317 | return 0; |
||
318 | #endif |
||
319 | }> |