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, | |
53 | struct amdgpu_ctx *ctx, | |
54 | uint32_t id) | |
55 | { | |
56 | int i; | |
57 | memset(ctx, 0, sizeof(*ctx)); | |
58 | ctx->adev = adev; | |
59 | kref_init(&ctx->refcount); | |
60 | spin_lock_init(&ctx->ring_lock); | |
61 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) | |
62 | ctx->rings[i].sequence = 1; | |
63 | } | |
64 | ||
0b492a4c AD |
65 | int amdgpu_ctx_alloc(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, |
66 | uint32_t *id) | |
d38ceaf9 | 67 | { |
d38ceaf9 | 68 | struct amdgpu_ctx *ctx; |
9cb7e5a9 | 69 | int i, j, r; |
d38ceaf9 AD |
70 | |
71 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
72 | if (!ctx) | |
73 | return -ENOMEM; | |
23ca0e4e CZ |
74 | if (fpriv) { |
75 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
76 | mutex_lock(&mgr->lock); | |
77 | r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL); | |
78 | if (r < 0) { | |
79 | mutex_unlock(&mgr->lock); | |
80 | kfree(ctx); | |
81 | return r; | |
82 | } | |
83 | *id = (uint32_t)r; | |
84 | amdgpu_ctx_init(adev, fpriv, ctx, *id); | |
0147ee0f | 85 | mutex_unlock(&mgr->lock); |
23ca0e4e CZ |
86 | } else { |
87 | if (adev->kernel_ctx) { | |
88 | DRM_ERROR("kernel cnotext has been created.\n"); | |
89 | kfree(ctx); | |
90 | return 0; | |
91 | } | |
92 | *id = AMD_KERNEL_CONTEXT_ID; | |
93 | amdgpu_ctx_init(adev, fpriv, ctx, *id); | |
94 | ||
95 | adev->kernel_ctx = ctx; | |
d38ceaf9 | 96 | } |
d38ceaf9 | 97 | |
9cb7e5a9 CZ |
98 | if (amdgpu_enable_scheduler) { |
99 | /* create context entity for each ring */ | |
100 | for (i = 0; i < adev->num_rings; i++) { | |
101 | struct amd_run_queue *rq; | |
102 | if (fpriv) | |
103 | rq = &adev->rings[i]->scheduler->sched_rq; | |
104 | else | |
105 | rq = &adev->rings[i]->scheduler->kernel_rq; | |
106 | r = amd_context_entity_init(adev->rings[i]->scheduler, | |
107 | &ctx->rings[i].c_entity, | |
1333f723 JZ |
108 | NULL, rq, *id, |
109 | amdgpu_sched_jobs); | |
9cb7e5a9 CZ |
110 | if (r) |
111 | break; | |
112 | } | |
113 | ||
114 | if (i < adev->num_rings) { | |
115 | for (j = 0; j < i; j++) | |
116 | amd_context_entity_fini(adev->rings[j]->scheduler, | |
117 | &ctx->rings[j].c_entity); | |
118 | kfree(ctx); | |
119 | return -EINVAL; | |
120 | } | |
121 | } | |
d38ceaf9 AD |
122 | |
123 | return 0; | |
124 | } | |
125 | ||
126 | int amdgpu_ctx_free(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, uint32_t id) | |
127 | { | |
d38ceaf9 | 128 | struct amdgpu_ctx *ctx; |
d38ceaf9 | 129 | |
23ca0e4e CZ |
130 | if (fpriv) { |
131 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
132 | mutex_lock(&mgr->lock); | |
133 | ctx = idr_find(&mgr->ctx_handles, id); | |
134 | if (ctx) { | |
135 | idr_remove(&mgr->ctx_handles, id); | |
136 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); | |
137 | mutex_unlock(&mgr->lock); | |
138 | return 0; | |
139 | } | |
0147ee0f | 140 | mutex_unlock(&mgr->lock); |
23ca0e4e CZ |
141 | } else { |
142 | ctx = adev->kernel_ctx; | |
143 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); | |
f11358da | 144 | return 0; |
d38ceaf9 AD |
145 | } |
146 | return -EINVAL; | |
147 | } | |
148 | ||
d94aed5a MO |
149 | static int amdgpu_ctx_query(struct amdgpu_device *adev, |
150 | struct amdgpu_fpriv *fpriv, uint32_t id, | |
151 | union drm_amdgpu_ctx_out *out) | |
d38ceaf9 AD |
152 | { |
153 | struct amdgpu_ctx *ctx; | |
23ca0e4e | 154 | struct amdgpu_ctx_mgr *mgr; |
d94aed5a | 155 | unsigned reset_counter; |
d38ceaf9 | 156 | |
23ca0e4e CZ |
157 | if (!fpriv) |
158 | return -EINVAL; | |
159 | ||
160 | mgr = &fpriv->ctx_mgr; | |
0147ee0f | 161 | mutex_lock(&mgr->lock); |
d38ceaf9 | 162 | ctx = idr_find(&mgr->ctx_handles, id); |
d94aed5a | 163 | if (!ctx) { |
0147ee0f | 164 | mutex_unlock(&mgr->lock); |
d94aed5a | 165 | return -EINVAL; |
d38ceaf9 | 166 | } |
d94aed5a MO |
167 | |
168 | /* TODO: these two are always zero */ | |
0b492a4c AD |
169 | out->state.flags = 0x0; |
170 | out->state.hangs = 0x0; | |
d94aed5a MO |
171 | |
172 | /* determine if a GPU reset has occured since the last call */ | |
173 | reset_counter = atomic_read(&adev->gpu_reset_counter); | |
174 | /* TODO: this should ideally return NO, GUILTY, or INNOCENT. */ | |
175 | if (ctx->reset_counter == reset_counter) | |
176 | out->state.reset_status = AMDGPU_CTX_NO_RESET; | |
177 | else | |
178 | out->state.reset_status = AMDGPU_CTX_UNKNOWN_RESET; | |
179 | ctx->reset_counter = reset_counter; | |
180 | ||
0147ee0f | 181 | mutex_unlock(&mgr->lock); |
d94aed5a | 182 | return 0; |
d38ceaf9 AD |
183 | } |
184 | ||
185 | void amdgpu_ctx_fini(struct amdgpu_fpriv *fpriv) | |
186 | { | |
187 | struct idr *idp; | |
188 | struct amdgpu_ctx *ctx; | |
189 | uint32_t id; | |
190 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
191 | idp = &mgr->ctx_handles; | |
192 | ||
193 | idr_for_each_entry(idp,ctx,id) { | |
194 | if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1) | |
0b492a4c | 195 | DRM_ERROR("ctx %p is still alive\n", ctx); |
d38ceaf9 AD |
196 | } |
197 | ||
cdecb65b | 198 | idr_destroy(&mgr->ctx_handles); |
0147ee0f | 199 | mutex_destroy(&mgr->lock); |
d38ceaf9 AD |
200 | } |
201 | ||
202 | int amdgpu_ctx_ioctl(struct drm_device *dev, void *data, | |
d94aed5a | 203 | struct drm_file *filp) |
d38ceaf9 AD |
204 | { |
205 | int r; | |
206 | uint32_t id; | |
d38ceaf9 AD |
207 | |
208 | union drm_amdgpu_ctx *args = data; | |
209 | struct amdgpu_device *adev = dev->dev_private; | |
210 | struct amdgpu_fpriv *fpriv = filp->driver_priv; | |
211 | ||
212 | r = 0; | |
213 | id = args->in.ctx_id; | |
d38ceaf9 AD |
214 | |
215 | switch (args->in.op) { | |
216 | case AMDGPU_CTX_OP_ALLOC_CTX: | |
0b492a4c | 217 | r = amdgpu_ctx_alloc(adev, fpriv, &id); |
d38ceaf9 AD |
218 | args->out.alloc.ctx_id = id; |
219 | break; | |
220 | case AMDGPU_CTX_OP_FREE_CTX: | |
221 | r = amdgpu_ctx_free(adev, fpriv, id); | |
222 | break; | |
223 | case AMDGPU_CTX_OP_QUERY_STATE: | |
d94aed5a | 224 | r = amdgpu_ctx_query(adev, fpriv, id, &args->out); |
d38ceaf9 AD |
225 | break; |
226 | default: | |
227 | return -EINVAL; | |
228 | } | |
229 | ||
230 | return r; | |
231 | } | |
66b3cf2a JZ |
232 | |
233 | struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id) | |
234 | { | |
235 | struct amdgpu_ctx *ctx; | |
23ca0e4e CZ |
236 | struct amdgpu_ctx_mgr *mgr; |
237 | ||
238 | if (!fpriv) | |
239 | return NULL; | |
240 | ||
241 | mgr = &fpriv->ctx_mgr; | |
66b3cf2a JZ |
242 | |
243 | mutex_lock(&mgr->lock); | |
244 | ctx = idr_find(&mgr->ctx_handles, id); | |
245 | if (ctx) | |
246 | kref_get(&ctx->refcount); | |
247 | mutex_unlock(&mgr->lock); | |
248 | return ctx; | |
249 | } | |
250 | ||
251 | int amdgpu_ctx_put(struct amdgpu_ctx *ctx) | |
252 | { | |
66b3cf2a JZ |
253 | if (ctx == NULL) |
254 | return -EINVAL; | |
255 | ||
66b3cf2a | 256 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); |
66b3cf2a JZ |
257 | return 0; |
258 | } | |
21c16bf6 CK |
259 | |
260 | uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, | |
261 | struct fence *fence) | |
262 | { | |
263 | struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx]; | |
b43a9a7e CZ |
264 | uint64_t seq = 0; |
265 | unsigned idx = 0; | |
266 | struct fence *other = NULL; | |
21c16bf6 | 267 | |
b43a9a7e CZ |
268 | if (amdgpu_enable_scheduler) |
269 | seq = atomic64_read(&cring->c_entity.last_queued_v_seq); | |
270 | else | |
271 | seq = cring->sequence; | |
272 | idx = seq % AMDGPU_CTX_MAX_CS_PENDING; | |
273 | other = cring->fences[idx]; | |
21c16bf6 CK |
274 | if (other) { |
275 | signed long r; | |
276 | r = fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT); | |
277 | if (r < 0) | |
278 | DRM_ERROR("Error (%ld) waiting for fence!\n", r); | |
279 | } | |
280 | ||
281 | fence_get(fence); | |
282 | ||
283 | spin_lock(&ctx->ring_lock); | |
284 | cring->fences[idx] = fence; | |
b43a9a7e CZ |
285 | if (!amdgpu_enable_scheduler) |
286 | cring->sequence++; | |
21c16bf6 CK |
287 | spin_unlock(&ctx->ring_lock); |
288 | ||
289 | fence_put(other); | |
290 | ||
291 | return seq; | |
292 | } | |
293 | ||
294 | struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, | |
295 | struct amdgpu_ring *ring, uint64_t seq) | |
296 | { | |
297 | struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx]; | |
298 | struct fence *fence; | |
b43a9a7e | 299 | uint64_t queued_seq; |
4b559c90 CZ |
300 | int r; |
301 | ||
302 | if (amdgpu_enable_scheduler) { | |
303 | r = amd_sched_wait_emit(&cring->c_entity, | |
304 | seq, | |
51b9db27 CZ |
305 | false, |
306 | -1); | |
4b559c90 CZ |
307 | if (r) |
308 | return NULL; | |
309 | } | |
21c16bf6 CK |
310 | |
311 | spin_lock(&ctx->ring_lock); | |
b43a9a7e CZ |
312 | if (amdgpu_enable_scheduler) |
313 | queued_seq = atomic64_read(&cring->c_entity.last_queued_v_seq) + 1; | |
314 | else | |
315 | queued_seq = cring->sequence; | |
316 | ||
317 | if (seq >= queued_seq) { | |
21c16bf6 CK |
318 | spin_unlock(&ctx->ring_lock); |
319 | return ERR_PTR(-EINVAL); | |
320 | } | |
321 | ||
b43a9a7e CZ |
322 | |
323 | if (seq + AMDGPU_CTX_MAX_CS_PENDING < queued_seq) { | |
21c16bf6 CK |
324 | spin_unlock(&ctx->ring_lock); |
325 | return NULL; | |
326 | } | |
327 | ||
328 | fence = fence_get(cring->fences[seq % AMDGPU_CTX_MAX_CS_PENDING]); | |
329 | spin_unlock(&ctx->ring_lock); | |
330 | ||
331 | return fence; | |
332 | } |