Commit | Line | Data |
---|---|---|
d38ceaf9 AD |
1 | /* |
2 | * Copyright 2015 Advanced Micro Devices, Inc. | |
3 | * | |
4 | * Permission is hereby granted, free of charge, to any person obtaining a | |
5 | * copy of this software and associated documentation files (the "Software"), | |
6 | * to deal in the Software without restriction, including without limitation | |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
8 | * and/or sell copies of the Software, and to permit persons to whom the | |
9 | * Software is furnished to do so, subject to the following conditions: | |
10 | * | |
11 | * The above copyright notice and this permission notice shall be included in | |
12 | * all copies or substantial portions of the Software. | |
13 | * | |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
20 | * OTHER DEALINGS IN THE SOFTWARE. | |
21 | * | |
22 | * Authors: monk liu <monk.liu@amd.com> | |
23 | */ | |
24 | ||
25 | #include <drm/drmP.h> | |
26 | #include "amdgpu.h" | |
27 | ||
28 | static void amdgpu_ctx_do_release(struct kref *ref) | |
29 | { | |
30 | struct amdgpu_ctx *ctx; | |
9cb7e5a9 | 31 | struct amdgpu_device *adev; |
21c16bf6 | 32 | unsigned i, j; |
d38ceaf9 AD |
33 | |
34 | ctx = container_of(ref, struct amdgpu_ctx, refcount); | |
9cb7e5a9 CZ |
35 | adev = ctx->adev; |
36 | ||
21c16bf6 CK |
37 | |
38 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) | |
39 | for (j = 0; j < AMDGPU_CTX_MAX_CS_PENDING; ++j) | |
40 | fence_put(ctx->rings[i].fences[j]); | |
9cb7e5a9 CZ |
41 | |
42 | if (amdgpu_enable_scheduler) { | |
43 | for (i = 0; i < adev->num_rings; i++) | |
44 | amd_context_entity_fini(adev->rings[i]->scheduler, | |
45 | &ctx->rings[i].c_entity); | |
46 | } | |
47 | ||
d38ceaf9 AD |
48 | kfree(ctx); |
49 | } | |
50 | ||
23ca0e4e CZ |
51 | static void amdgpu_ctx_init(struct amdgpu_device *adev, |
52 | struct amdgpu_fpriv *fpriv, | |
0e89d0c1 | 53 | struct amdgpu_ctx *ctx) |
23ca0e4e CZ |
54 | { |
55 | int i; | |
56 | memset(ctx, 0, sizeof(*ctx)); | |
57 | ctx->adev = adev; | |
58 | kref_init(&ctx->refcount); | |
59 | spin_lock_init(&ctx->ring_lock); | |
60 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) | |
61 | ctx->rings[i].sequence = 1; | |
62 | } | |
63 | ||
0b492a4c AD |
64 | int amdgpu_ctx_alloc(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, |
65 | uint32_t *id) | |
d38ceaf9 | 66 | { |
d38ceaf9 | 67 | struct amdgpu_ctx *ctx; |
9cb7e5a9 | 68 | int i, j, r; |
d38ceaf9 AD |
69 | |
70 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
71 | if (!ctx) | |
72 | return -ENOMEM; | |
23ca0e4e CZ |
73 | if (fpriv) { |
74 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
75 | mutex_lock(&mgr->lock); | |
76 | r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL); | |
77 | if (r < 0) { | |
78 | mutex_unlock(&mgr->lock); | |
79 | kfree(ctx); | |
80 | return r; | |
81 | } | |
82 | *id = (uint32_t)r; | |
0e89d0c1 | 83 | amdgpu_ctx_init(adev, fpriv, ctx); |
0147ee0f | 84 | mutex_unlock(&mgr->lock); |
23ca0e4e CZ |
85 | } else { |
86 | if (adev->kernel_ctx) { | |
87 | DRM_ERROR("kernel cnotext has been created.\n"); | |
88 | kfree(ctx); | |
89 | return 0; | |
90 | } | |
0e89d0c1 | 91 | amdgpu_ctx_init(adev, fpriv, ctx); |
23ca0e4e CZ |
92 | |
93 | adev->kernel_ctx = ctx; | |
d38ceaf9 | 94 | } |
d38ceaf9 | 95 | |
9cb7e5a9 CZ |
96 | if (amdgpu_enable_scheduler) { |
97 | /* create context entity for each ring */ | |
98 | for (i = 0; i < adev->num_rings; i++) { | |
99 | struct amd_run_queue *rq; | |
100 | if (fpriv) | |
101 | rq = &adev->rings[i]->scheduler->sched_rq; | |
102 | else | |
103 | rq = &adev->rings[i]->scheduler->kernel_rq; | |
104 | r = amd_context_entity_init(adev->rings[i]->scheduler, | |
105 | &ctx->rings[i].c_entity, | |
0e89d0c1 | 106 | NULL, rq, amdgpu_sched_jobs); |
9cb7e5a9 CZ |
107 | if (r) |
108 | break; | |
109 | } | |
110 | ||
111 | if (i < adev->num_rings) { | |
112 | for (j = 0; j < i; j++) | |
113 | amd_context_entity_fini(adev->rings[j]->scheduler, | |
114 | &ctx->rings[j].c_entity); | |
115 | kfree(ctx); | |
116 | return -EINVAL; | |
117 | } | |
118 | } | |
d38ceaf9 AD |
119 | |
120 | return 0; | |
121 | } | |
122 | ||
123 | int amdgpu_ctx_free(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, uint32_t id) | |
124 | { | |
d38ceaf9 | 125 | struct amdgpu_ctx *ctx; |
d38ceaf9 | 126 | |
23ca0e4e CZ |
127 | if (fpriv) { |
128 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
129 | mutex_lock(&mgr->lock); | |
130 | ctx = idr_find(&mgr->ctx_handles, id); | |
131 | if (ctx) { | |
132 | idr_remove(&mgr->ctx_handles, id); | |
133 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); | |
134 | mutex_unlock(&mgr->lock); | |
135 | return 0; | |
136 | } | |
0147ee0f | 137 | mutex_unlock(&mgr->lock); |
23ca0e4e CZ |
138 | } else { |
139 | ctx = adev->kernel_ctx; | |
140 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); | |
f11358da | 141 | return 0; |
d38ceaf9 AD |
142 | } |
143 | return -EINVAL; | |
144 | } | |
145 | ||
d94aed5a MO |
146 | static int amdgpu_ctx_query(struct amdgpu_device *adev, |
147 | struct amdgpu_fpriv *fpriv, uint32_t id, | |
148 | union drm_amdgpu_ctx_out *out) | |
d38ceaf9 AD |
149 | { |
150 | struct amdgpu_ctx *ctx; | |
23ca0e4e | 151 | struct amdgpu_ctx_mgr *mgr; |
d94aed5a | 152 | unsigned reset_counter; |
d38ceaf9 | 153 | |
23ca0e4e CZ |
154 | if (!fpriv) |
155 | return -EINVAL; | |
156 | ||
157 | mgr = &fpriv->ctx_mgr; | |
0147ee0f | 158 | mutex_lock(&mgr->lock); |
d38ceaf9 | 159 | ctx = idr_find(&mgr->ctx_handles, id); |
d94aed5a | 160 | if (!ctx) { |
0147ee0f | 161 | mutex_unlock(&mgr->lock); |
d94aed5a | 162 | return -EINVAL; |
d38ceaf9 | 163 | } |
d94aed5a MO |
164 | |
165 | /* TODO: these two are always zero */ | |
0b492a4c AD |
166 | out->state.flags = 0x0; |
167 | out->state.hangs = 0x0; | |
d94aed5a MO |
168 | |
169 | /* determine if a GPU reset has occured since the last call */ | |
170 | reset_counter = atomic_read(&adev->gpu_reset_counter); | |
171 | /* TODO: this should ideally return NO, GUILTY, or INNOCENT. */ | |
172 | if (ctx->reset_counter == reset_counter) | |
173 | out->state.reset_status = AMDGPU_CTX_NO_RESET; | |
174 | else | |
175 | out->state.reset_status = AMDGPU_CTX_UNKNOWN_RESET; | |
176 | ctx->reset_counter = reset_counter; | |
177 | ||
0147ee0f | 178 | mutex_unlock(&mgr->lock); |
d94aed5a | 179 | return 0; |
d38ceaf9 AD |
180 | } |
181 | ||
d38ceaf9 | 182 | int amdgpu_ctx_ioctl(struct drm_device *dev, void *data, |
d94aed5a | 183 | struct drm_file *filp) |
d38ceaf9 AD |
184 | { |
185 | int r; | |
186 | uint32_t id; | |
d38ceaf9 AD |
187 | |
188 | union drm_amdgpu_ctx *args = data; | |
189 | struct amdgpu_device *adev = dev->dev_private; | |
190 | struct amdgpu_fpriv *fpriv = filp->driver_priv; | |
191 | ||
192 | r = 0; | |
193 | id = args->in.ctx_id; | |
d38ceaf9 AD |
194 | |
195 | switch (args->in.op) { | |
196 | case AMDGPU_CTX_OP_ALLOC_CTX: | |
0b492a4c | 197 | r = amdgpu_ctx_alloc(adev, fpriv, &id); |
d38ceaf9 AD |
198 | args->out.alloc.ctx_id = id; |
199 | break; | |
200 | case AMDGPU_CTX_OP_FREE_CTX: | |
201 | r = amdgpu_ctx_free(adev, fpriv, id); | |
202 | break; | |
203 | case AMDGPU_CTX_OP_QUERY_STATE: | |
d94aed5a | 204 | r = amdgpu_ctx_query(adev, fpriv, id, &args->out); |
d38ceaf9 AD |
205 | break; |
206 | default: | |
207 | return -EINVAL; | |
208 | } | |
209 | ||
210 | return r; | |
211 | } | |
66b3cf2a JZ |
212 | |
213 | struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id) | |
214 | { | |
215 | struct amdgpu_ctx *ctx; | |
23ca0e4e CZ |
216 | struct amdgpu_ctx_mgr *mgr; |
217 | ||
218 | if (!fpriv) | |
219 | return NULL; | |
220 | ||
221 | mgr = &fpriv->ctx_mgr; | |
66b3cf2a JZ |
222 | |
223 | mutex_lock(&mgr->lock); | |
224 | ctx = idr_find(&mgr->ctx_handles, id); | |
225 | if (ctx) | |
226 | kref_get(&ctx->refcount); | |
227 | mutex_unlock(&mgr->lock); | |
228 | return ctx; | |
229 | } | |
230 | ||
231 | int amdgpu_ctx_put(struct amdgpu_ctx *ctx) | |
232 | { | |
66b3cf2a JZ |
233 | if (ctx == NULL) |
234 | return -EINVAL; | |
235 | ||
66b3cf2a | 236 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); |
66b3cf2a JZ |
237 | return 0; |
238 | } | |
21c16bf6 CK |
239 | |
240 | uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, | |
d1ff9086 | 241 | struct fence *fence, uint64_t queued_seq) |
21c16bf6 CK |
242 | { |
243 | struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx]; | |
b43a9a7e CZ |
244 | uint64_t seq = 0; |
245 | unsigned idx = 0; | |
246 | struct fence *other = NULL; | |
21c16bf6 | 247 | |
b43a9a7e | 248 | if (amdgpu_enable_scheduler) |
d1ff9086 | 249 | seq = queued_seq; |
b43a9a7e CZ |
250 | else |
251 | seq = cring->sequence; | |
252 | idx = seq % AMDGPU_CTX_MAX_CS_PENDING; | |
253 | other = cring->fences[idx]; | |
21c16bf6 CK |
254 | if (other) { |
255 | signed long r; | |
256 | r = fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT); | |
257 | if (r < 0) | |
258 | DRM_ERROR("Error (%ld) waiting for fence!\n", r); | |
259 | } | |
260 | ||
261 | fence_get(fence); | |
262 | ||
263 | spin_lock(&ctx->ring_lock); | |
264 | cring->fences[idx] = fence; | |
b43a9a7e CZ |
265 | if (!amdgpu_enable_scheduler) |
266 | cring->sequence++; | |
21c16bf6 CK |
267 | spin_unlock(&ctx->ring_lock); |
268 | ||
269 | fence_put(other); | |
270 | ||
271 | return seq; | |
272 | } | |
273 | ||
274 | struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, | |
275 | struct amdgpu_ring *ring, uint64_t seq) | |
276 | { | |
277 | struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx]; | |
278 | struct fence *fence; | |
b43a9a7e | 279 | uint64_t queued_seq; |
4b559c90 CZ |
280 | int r; |
281 | ||
282 | if (amdgpu_enable_scheduler) { | |
283 | r = amd_sched_wait_emit(&cring->c_entity, | |
284 | seq, | |
51b9db27 CZ |
285 | false, |
286 | -1); | |
4b559c90 CZ |
287 | if (r) |
288 | return NULL; | |
289 | } | |
21c16bf6 CK |
290 | |
291 | spin_lock(&ctx->ring_lock); | |
b43a9a7e | 292 | if (amdgpu_enable_scheduler) |
27f6642d | 293 | queued_seq = amd_sched_next_queued_seq(&cring->c_entity); |
b43a9a7e CZ |
294 | else |
295 | queued_seq = cring->sequence; | |
296 | ||
297 | if (seq >= queued_seq) { | |
21c16bf6 CK |
298 | spin_unlock(&ctx->ring_lock); |
299 | return ERR_PTR(-EINVAL); | |
300 | } | |
301 | ||
b43a9a7e CZ |
302 | |
303 | if (seq + AMDGPU_CTX_MAX_CS_PENDING < queued_seq) { | |
21c16bf6 CK |
304 | spin_unlock(&ctx->ring_lock); |
305 | return NULL; | |
306 | } | |
307 | ||
308 | fence = fence_get(cring->fences[seq % AMDGPU_CTX_MAX_CS_PENDING]); | |
309 | spin_unlock(&ctx->ring_lock); | |
310 | ||
311 | return fence; | |
312 | } | |
efd4ccb5 CK |
313 | |
314 | void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr) | |
315 | { | |
316 | mutex_init(&mgr->lock); | |
317 | idr_init(&mgr->ctx_handles); | |
318 | } | |
319 | ||
320 | void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr) | |
321 | { | |
322 | struct amdgpu_ctx *ctx; | |
323 | struct idr *idp; | |
324 | uint32_t id; | |
325 | ||
326 | idp = &mgr->ctx_handles; | |
327 | ||
328 | idr_for_each_entry(idp, ctx, id) { | |
329 | if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1) | |
330 | DRM_ERROR("ctx %p is still alive\n", ctx); | |
331 | } | |
332 | ||
333 | idr_destroy(&mgr->ctx_handles); | |
334 | mutex_destroy(&mgr->lock); | |
335 | } |