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 | * IB | |
38 | * IBs (Indirect Buffers) and areas of GPU accessible memory where | |
39 | * commands are stored. You can put a pointer to the IB in the | |
40 | * command ring and the hw will fetch the commands from the IB | |
41 | * and execute them. Generally userspace acceleration drivers | |
42 | * produce command buffers which are send to the kernel and | |
43 | * put in IBs for execution by the requested ring. | |
44 | */ | |
45 | static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev); | |
46 | ||
47 | /** | |
48 | * amdgpu_ib_get - request an IB (Indirect Buffer) | |
49 | * | |
50 | * @ring: ring index the IB is associated with | |
51 | * @size: requested IB size | |
52 | * @ib: IB object returned | |
53 | * | |
54 | * Request an IB (all asics). IBs are allocated using the | |
55 | * suballocator. | |
56 | * Returns 0 on success, error on failure. | |
57 | */ | |
b07c60c0 | 58 | int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, |
d38ceaf9 AD |
59 | unsigned size, struct amdgpu_ib *ib) |
60 | { | |
d38ceaf9 AD |
61 | int r; |
62 | ||
63 | if (size) { | |
bbf0b345 | 64 | r = amdgpu_sa_bo_new(&adev->ring_tmp_bo, |
d38ceaf9 AD |
65 | &ib->sa_bo, size, 256); |
66 | if (r) { | |
67 | dev_err(adev->dev, "failed to get a new IB (%d)\n", r); | |
68 | return r; | |
69 | } | |
70 | ||
71 | ib->ptr = amdgpu_sa_bo_cpu_addr(ib->sa_bo); | |
72 | ||
73 | if (!vm) | |
74 | ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo); | |
d38ceaf9 AD |
75 | } |
76 | ||
4ff37a83 | 77 | ib->vm_id = 0; |
d38ceaf9 AD |
78 | |
79 | return 0; | |
80 | } | |
81 | ||
82 | /** | |
83 | * amdgpu_ib_free - free an IB (Indirect Buffer) | |
84 | * | |
85 | * @adev: amdgpu_device pointer | |
86 | * @ib: IB object to free | |
cc55c45d | 87 | * @f: the fence SA bo need wait on for the ib alloation |
d38ceaf9 AD |
88 | * |
89 | * Free an IB (all asics). | |
90 | */ | |
4d9c514d CK |
91 | void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib, |
92 | struct fence *f) | |
d38ceaf9 | 93 | { |
cc55c45d | 94 | amdgpu_sa_bo_free(adev, &ib->sa_bo, f); |
d38ceaf9 AD |
95 | } |
96 | ||
97 | /** | |
98 | * amdgpu_ib_schedule - schedule an IB (Indirect Buffer) on the ring | |
99 | * | |
100 | * @adev: amdgpu_device pointer | |
101 | * @num_ibs: number of IBs to schedule | |
102 | * @ibs: IB objects to schedule | |
ec72b800 | 103 | * @f: fence created during this submission |
d38ceaf9 AD |
104 | * |
105 | * Schedule an IB on the associated ring (all asics). | |
106 | * Returns 0 on success, error on failure. | |
107 | * | |
108 | * On SI, there are two parallel engines fed from the primary ring, | |
109 | * the CE (Constant Engine) and the DE (Drawing Engine). Since | |
110 | * resource descriptors have moved to memory, the CE allows you to | |
111 | * prime the caches while the DE is updating register state so that | |
112 | * the resource descriptors will be already in cache when the draw is | |
113 | * processed. To accomplish this, the userspace driver submits two | |
114 | * IBs, one for the CE and one for the DE. If there is a CE IB (called | |
115 | * a CONST_IB), it will be put on the ring prior to the DE IB. Prior | |
116 | * to SI there was just a DE IB. | |
117 | */ | |
b07c60c0 | 118 | int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, |
336d1f5e | 119 | struct amdgpu_ib *ibs, struct fence *last_vm_update, |
c5637837 | 120 | struct amdgpu_job *job, struct fence **f) |
d38ceaf9 | 121 | { |
b07c60c0 | 122 | struct amdgpu_device *adev = ring->adev; |
d38ceaf9 | 123 | struct amdgpu_ib *ib = &ibs[0]; |
aa3b73f6 | 124 | uint64_t ctx, old_ctx; |
73cfa5f5 | 125 | struct fence *hwf; |
c5637837 | 126 | struct amdgpu_vm *vm = NULL; |
03ccf481 ML |
127 | unsigned i, patch_offset = ~0; |
128 | ||
d38ceaf9 | 129 | int r = 0; |
d38ceaf9 AD |
130 | |
131 | if (num_ibs == 0) | |
132 | return -EINVAL; | |
133 | ||
3cb485f3 | 134 | ctx = ibs->ctx; |
c5637837 ML |
135 | if (job) /* for domain0 job like ring test, ibs->job is not assigned */ |
136 | vm = job->vm; | |
d919ad49 | 137 | |
d38ceaf9 AD |
138 | if (!ring->ready) { |
139 | dev_err(adev->dev, "couldn't schedule ib\n"); | |
140 | return -EINVAL; | |
141 | } | |
be86c606 | 142 | |
4ff37a83 | 143 | if (vm && !ibs->vm_id) { |
8d0a7cea CK |
144 | dev_err(adev->dev, "VM IB without ID\n"); |
145 | return -EINVAL; | |
146 | } | |
147 | ||
867d0517 | 148 | r = amdgpu_ring_alloc(ring, 256 * num_ibs); |
d38ceaf9 AD |
149 | if (r) { |
150 | dev_err(adev->dev, "scheduling IB failed (%d).\n", r); | |
151 | return r; | |
152 | } | |
153 | ||
03ccf481 ML |
154 | if (ring->type == AMDGPU_RING_TYPE_SDMA && ring->funcs->init_cond_exec) |
155 | patch_offset = amdgpu_ring_init_cond_exec(ring); | |
156 | ||
d38ceaf9 AD |
157 | if (vm) { |
158 | /* do context switch */ | |
41d9eb2c CK |
159 | r = amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr, |
160 | ib->gds_base, ib->gds_size, | |
161 | ib->gws_base, ib->gws_size, | |
162 | ib->oa_base, ib->oa_size); | |
163 | if (r) { | |
164 | amdgpu_ring_undo(ring); | |
165 | return r; | |
166 | } | |
e722b71a | 167 | } |
d2edb07b | 168 | |
794ff571 ML |
169 | if (ring->funcs->emit_hdp_flush) |
170 | amdgpu_ring_emit_hdp_flush(ring); | |
171 | ||
128cff1a ML |
172 | /* always set cond_exec_polling to CONTINUE */ |
173 | *ring->cond_exe_cpu_addr = 1; | |
174 | ||
3cb485f3 | 175 | old_ctx = ring->current_ctx; |
d38ceaf9 AD |
176 | for (i = 0; i < num_ibs; ++i) { |
177 | ib = &ibs[i]; | |
d38ceaf9 | 178 | amdgpu_ring_emit_ib(ring, ib); |
3cb485f3 | 179 | ring->current_ctx = ctx; |
d38ceaf9 AD |
180 | } |
181 | ||
794ff571 ML |
182 | if (ring->funcs->emit_hdp_invalidate) |
183 | amdgpu_ring_emit_hdp_invalidate(ring); | |
11afbde8 | 184 | |
73cfa5f5 | 185 | r = amdgpu_fence_emit(ring, &hwf); |
d38ceaf9 AD |
186 | if (r) { |
187 | dev_err(adev->dev, "failed to emit fence (%d)\n", r); | |
3cb485f3 | 188 | ring->current_ctx = old_ctx; |
971fe9a9 CK |
189 | if (ib->vm_id) |
190 | amdgpu_vm_reset_id(adev, ib->vm_id); | |
a27de35c | 191 | amdgpu_ring_undo(ring); |
d38ceaf9 AD |
192 | return r; |
193 | } | |
194 | ||
195 | /* wrap the last IB with fence */ | |
196 | if (ib->user) { | |
197 | uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo); | |
198 | addr += ib->user->offset; | |
5430a3ff | 199 | amdgpu_ring_emit_fence(ring, addr, ib->sequence, |
890ee23f | 200 | AMDGPU_FENCE_FLAG_64BIT); |
d38ceaf9 AD |
201 | } |
202 | ||
ec72b800 | 203 | if (f) |
73cfa5f5 | 204 | *f = fence_get(hwf); |
ec72b800 | 205 | |
03ccf481 ML |
206 | if (patch_offset != ~0 && ring->funcs->patch_cond_exec) |
207 | amdgpu_ring_patch_cond_exec(ring, patch_offset); | |
208 | ||
a27de35c | 209 | amdgpu_ring_commit(ring); |
d38ceaf9 AD |
210 | return 0; |
211 | } | |
212 | ||
213 | /** | |
214 | * amdgpu_ib_pool_init - Init the IB (Indirect Buffer) pool | |
215 | * | |
216 | * @adev: amdgpu_device pointer | |
217 | * | |
218 | * Initialize the suballocator to manage a pool of memory | |
219 | * for use as IBs (all asics). | |
220 | * Returns 0 on success, error on failure. | |
221 | */ | |
222 | int amdgpu_ib_pool_init(struct amdgpu_device *adev) | |
223 | { | |
224 | int r; | |
225 | ||
226 | if (adev->ib_pool_ready) { | |
227 | return 0; | |
228 | } | |
229 | r = amdgpu_sa_bo_manager_init(adev, &adev->ring_tmp_bo, | |
230 | AMDGPU_IB_POOL_SIZE*64*1024, | |
231 | AMDGPU_GPU_PAGE_SIZE, | |
232 | AMDGPU_GEM_DOMAIN_GTT); | |
233 | if (r) { | |
234 | return r; | |
235 | } | |
236 | ||
237 | r = amdgpu_sa_bo_manager_start(adev, &adev->ring_tmp_bo); | |
238 | if (r) { | |
239 | return r; | |
240 | } | |
241 | ||
242 | adev->ib_pool_ready = true; | |
243 | if (amdgpu_debugfs_sa_init(adev)) { | |
244 | dev_err(adev->dev, "failed to register debugfs file for SA\n"); | |
245 | } | |
246 | return 0; | |
247 | } | |
248 | ||
249 | /** | |
250 | * amdgpu_ib_pool_fini - Free the IB (Indirect Buffer) pool | |
251 | * | |
252 | * @adev: amdgpu_device pointer | |
253 | * | |
254 | * Tear down the suballocator managing the pool of memory | |
255 | * for use as IBs (all asics). | |
256 | */ | |
257 | void amdgpu_ib_pool_fini(struct amdgpu_device *adev) | |
258 | { | |
259 | if (adev->ib_pool_ready) { | |
260 | amdgpu_sa_bo_manager_suspend(adev, &adev->ring_tmp_bo); | |
261 | amdgpu_sa_bo_manager_fini(adev, &adev->ring_tmp_bo); | |
262 | adev->ib_pool_ready = false; | |
263 | } | |
264 | } | |
265 | ||
266 | /** | |
267 | * amdgpu_ib_ring_tests - test IBs on the rings | |
268 | * | |
269 | * @adev: amdgpu_device pointer | |
270 | * | |
271 | * Test an IB (Indirect Buffer) on each ring. | |
272 | * If the test fails, disable the ring. | |
273 | * Returns 0 on success, error if the primary GFX ring | |
274 | * IB test fails. | |
275 | */ | |
276 | int amdgpu_ib_ring_tests(struct amdgpu_device *adev) | |
277 | { | |
278 | unsigned i; | |
279 | int r; | |
280 | ||
281 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
282 | struct amdgpu_ring *ring = adev->rings[i]; | |
283 | ||
284 | if (!ring || !ring->ready) | |
285 | continue; | |
286 | ||
287 | r = amdgpu_ring_test_ib(ring); | |
288 | if (r) { | |
289 | ring->ready = false; | |
d38ceaf9 AD |
290 | |
291 | if (ring == &adev->gfx.gfx_ring[0]) { | |
292 | /* oh, oh, that's really bad */ | |
293 | DRM_ERROR("amdgpu: failed testing IB on GFX ring (%d).\n", r); | |
294 | adev->accel_working = false; | |
295 | return r; | |
296 | ||
297 | } else { | |
298 | /* still not good, but we can live with it */ | |
299 | DRM_ERROR("amdgpu: failed testing IB on ring %d (%d).\n", i, r); | |
300 | } | |
301 | } | |
302 | } | |
303 | return 0; | |
304 | } | |
305 | ||
306 | /* | |
307 | * Debugfs info | |
308 | */ | |
309 | #if defined(CONFIG_DEBUG_FS) | |
310 | ||
311 | static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data) | |
312 | { | |
313 | struct drm_info_node *node = (struct drm_info_node *) m->private; | |
314 | struct drm_device *dev = node->minor->dev; | |
315 | struct amdgpu_device *adev = dev->dev_private; | |
316 | ||
317 | amdgpu_sa_bo_dump_debug_info(&adev->ring_tmp_bo, m); | |
318 | ||
319 | return 0; | |
320 | ||
321 | } | |
322 | ||
06ab6832 | 323 | static const struct drm_info_list amdgpu_debugfs_sa_list[] = { |
d38ceaf9 AD |
324 | {"amdgpu_sa_info", &amdgpu_debugfs_sa_info, 0, NULL}, |
325 | }; | |
326 | ||
327 | #endif | |
328 | ||
329 | static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev) | |
330 | { | |
331 | #if defined(CONFIG_DEBUG_FS) | |
332 | return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_sa_list, 1); | |
333 | #else | |
334 | return 0; | |
335 | #endif | |
336 | } |