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 | ||
20874179 | 28 | static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx) |
d38ceaf9 | 29 | { |
21c16bf6 | 30 | unsigned i, j; |
47f38501 | 31 | int r; |
d38ceaf9 | 32 | |
23ca0e4e CZ |
33 | memset(ctx, 0, sizeof(*ctx)); |
34 | ctx->adev = adev; | |
35 | kref_init(&ctx->refcount); | |
36 | spin_lock_init(&ctx->ring_lock); | |
a750b47e CK |
37 | ctx->fences = kcalloc(amdgpu_sched_jobs * AMDGPU_MAX_RINGS, |
38 | sizeof(struct fence*), GFP_KERNEL); | |
37cd0ca2 CZ |
39 | if (!ctx->fences) |
40 | return -ENOMEM; | |
d38ceaf9 | 41 | |
37cd0ca2 CZ |
42 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
43 | ctx->rings[i].sequence = 1; | |
a750b47e | 44 | ctx->rings[i].fences = &ctx->fences[amdgpu_sched_jobs * i]; |
37cd0ca2 | 45 | } |
ce199ad6 NH |
46 | |
47 | ctx->reset_counter = atomic_read(&adev->gpu_reset_counter); | |
48 | ||
cadf97b1 CZ |
49 | /* create context entity for each ring */ |
50 | for (i = 0; i < adev->num_rings; i++) { | |
20874179 | 51 | struct amdgpu_ring *ring = adev->rings[i]; |
cadf97b1 | 52 | struct amd_sched_rq *rq; |
20874179 CK |
53 | |
54 | rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL]; | |
55 | r = amd_sched_entity_init(&ring->sched, &ctx->rings[i].entity, | |
cadf97b1 CZ |
56 | rq, amdgpu_sched_jobs); |
57 | if (r) | |
58 | break; | |
59 | } | |
60 | ||
61 | if (i < adev->num_rings) { | |
62 | for (j = 0; j < i; j++) | |
63 | amd_sched_entity_fini(&adev->rings[j]->sched, | |
64 | &ctx->rings[j].entity); | |
65 | kfree(ctx->fences); | |
54ddf3a6 | 66 | ctx->fences = NULL; |
cadf97b1 | 67 | return r; |
9cb7e5a9 | 68 | } |
d38ceaf9 AD |
69 | return 0; |
70 | } | |
71 | ||
20874179 | 72 | static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx) |
d38ceaf9 | 73 | { |
47f38501 CK |
74 | struct amdgpu_device *adev = ctx->adev; |
75 | unsigned i, j; | |
76 | ||
fe295b27 DA |
77 | if (!adev) |
78 | return; | |
79 | ||
47f38501 | 80 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) |
37cd0ca2 | 81 | for (j = 0; j < amdgpu_sched_jobs; ++j) |
47f38501 | 82 | fence_put(ctx->rings[i].fences[j]); |
37cd0ca2 | 83 | kfree(ctx->fences); |
54ddf3a6 | 84 | ctx->fences = NULL; |
47f38501 | 85 | |
cadf97b1 CZ |
86 | for (i = 0; i < adev->num_rings; i++) |
87 | amd_sched_entity_fini(&adev->rings[i]->sched, | |
88 | &ctx->rings[i].entity); | |
47f38501 CK |
89 | } |
90 | ||
91 | static int amdgpu_ctx_alloc(struct amdgpu_device *adev, | |
92 | struct amdgpu_fpriv *fpriv, | |
93 | uint32_t *id) | |
94 | { | |
95 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
d38ceaf9 | 96 | struct amdgpu_ctx *ctx; |
47f38501 | 97 | int r; |
d38ceaf9 | 98 | |
47f38501 CK |
99 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
100 | if (!ctx) | |
101 | return -ENOMEM; | |
102 | ||
103 | mutex_lock(&mgr->lock); | |
104 | r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL); | |
105 | if (r < 0) { | |
0147ee0f | 106 | mutex_unlock(&mgr->lock); |
47f38501 CK |
107 | kfree(ctx); |
108 | return r; | |
109 | } | |
110 | *id = (uint32_t)r; | |
20874179 | 111 | r = amdgpu_ctx_init(adev, ctx); |
c648ed7c CZ |
112 | if (r) { |
113 | idr_remove(&mgr->ctx_handles, *id); | |
114 | *id = 0; | |
115 | kfree(ctx); | |
116 | } | |
47f38501 | 117 | mutex_unlock(&mgr->lock); |
47f38501 CK |
118 | return r; |
119 | } | |
120 | ||
121 | static void amdgpu_ctx_do_release(struct kref *ref) | |
122 | { | |
123 | struct amdgpu_ctx *ctx; | |
124 | ||
125 | ctx = container_of(ref, struct amdgpu_ctx, refcount); | |
126 | ||
127 | amdgpu_ctx_fini(ctx); | |
128 | ||
129 | kfree(ctx); | |
130 | } | |
131 | ||
132 | static int amdgpu_ctx_free(struct amdgpu_fpriv *fpriv, uint32_t id) | |
133 | { | |
134 | struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr; | |
135 | struct amdgpu_ctx *ctx; | |
136 | ||
137 | mutex_lock(&mgr->lock); | |
138 | ctx = idr_find(&mgr->ctx_handles, id); | |
139 | if (ctx) { | |
140 | idr_remove(&mgr->ctx_handles, id); | |
23ca0e4e | 141 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); |
47f38501 | 142 | mutex_unlock(&mgr->lock); |
f11358da | 143 | return 0; |
d38ceaf9 | 144 | } |
47f38501 | 145 | mutex_unlock(&mgr->lock); |
d38ceaf9 AD |
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 | ||
d38ceaf9 | 185 | int amdgpu_ctx_ioctl(struct drm_device *dev, void *data, |
d94aed5a | 186 | struct drm_file *filp) |
d38ceaf9 AD |
187 | { |
188 | int r; | |
189 | uint32_t id; | |
d38ceaf9 AD |
190 | |
191 | union drm_amdgpu_ctx *args = data; | |
192 | struct amdgpu_device *adev = dev->dev_private; | |
193 | struct amdgpu_fpriv *fpriv = filp->driver_priv; | |
194 | ||
195 | r = 0; | |
196 | id = args->in.ctx_id; | |
d38ceaf9 AD |
197 | |
198 | switch (args->in.op) { | |
a750b47e CK |
199 | case AMDGPU_CTX_OP_ALLOC_CTX: |
200 | r = amdgpu_ctx_alloc(adev, fpriv, &id); | |
201 | args->out.alloc.ctx_id = id; | |
202 | break; | |
203 | case AMDGPU_CTX_OP_FREE_CTX: | |
204 | r = amdgpu_ctx_free(fpriv, id); | |
205 | break; | |
206 | case AMDGPU_CTX_OP_QUERY_STATE: | |
207 | r = amdgpu_ctx_query(adev, fpriv, id, &args->out); | |
208 | break; | |
209 | default: | |
210 | return -EINVAL; | |
d38ceaf9 AD |
211 | } |
212 | ||
213 | return r; | |
214 | } | |
66b3cf2a JZ |
215 | |
216 | struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id) | |
217 | { | |
218 | struct amdgpu_ctx *ctx; | |
23ca0e4e CZ |
219 | struct amdgpu_ctx_mgr *mgr; |
220 | ||
221 | if (!fpriv) | |
222 | return NULL; | |
223 | ||
224 | mgr = &fpriv->ctx_mgr; | |
66b3cf2a JZ |
225 | |
226 | mutex_lock(&mgr->lock); | |
227 | ctx = idr_find(&mgr->ctx_handles, id); | |
228 | if (ctx) | |
229 | kref_get(&ctx->refcount); | |
230 | mutex_unlock(&mgr->lock); | |
231 | return ctx; | |
232 | } | |
233 | ||
234 | int amdgpu_ctx_put(struct amdgpu_ctx *ctx) | |
235 | { | |
66b3cf2a JZ |
236 | if (ctx == NULL) |
237 | return -EINVAL; | |
238 | ||
66b3cf2a | 239 | kref_put(&ctx->refcount, amdgpu_ctx_do_release); |
66b3cf2a JZ |
240 | return 0; |
241 | } | |
21c16bf6 CK |
242 | |
243 | uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, | |
ce882e6d | 244 | struct fence *fence) |
21c16bf6 CK |
245 | { |
246 | struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx]; | |
ce882e6d | 247 | uint64_t seq = cring->sequence; |
b43a9a7e CZ |
248 | unsigned idx = 0; |
249 | struct fence *other = NULL; | |
21c16bf6 | 250 | |
5b011235 | 251 | idx = seq & (amdgpu_sched_jobs - 1); |
b43a9a7e | 252 | other = cring->fences[idx]; |
21c16bf6 CK |
253 | if (other) { |
254 | signed long r; | |
255 | r = fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT); | |
256 | if (r < 0) | |
257 | DRM_ERROR("Error (%ld) waiting for fence!\n", r); | |
258 | } | |
259 | ||
260 | fence_get(fence); | |
261 | ||
262 | spin_lock(&ctx->ring_lock); | |
263 | cring->fences[idx] = fence; | |
ce882e6d | 264 | cring->sequence++; |
21c16bf6 CK |
265 | spin_unlock(&ctx->ring_lock); |
266 | ||
267 | fence_put(other); | |
268 | ||
269 | return seq; | |
270 | } | |
271 | ||
272 | struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, | |
273 | struct amdgpu_ring *ring, uint64_t seq) | |
274 | { | |
275 | struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx]; | |
276 | struct fence *fence; | |
277 | ||
278 | spin_lock(&ctx->ring_lock); | |
b43a9a7e | 279 | |
ce882e6d | 280 | if (seq >= cring->sequence) { |
21c16bf6 CK |
281 | spin_unlock(&ctx->ring_lock); |
282 | return ERR_PTR(-EINVAL); | |
283 | } | |
284 | ||
b43a9a7e | 285 | |
37cd0ca2 | 286 | if (seq + amdgpu_sched_jobs < cring->sequence) { |
21c16bf6 CK |
287 | spin_unlock(&ctx->ring_lock); |
288 | return NULL; | |
289 | } | |
290 | ||
5b011235 | 291 | fence = fence_get(cring->fences[seq & (amdgpu_sched_jobs - 1)]); |
21c16bf6 CK |
292 | spin_unlock(&ctx->ring_lock); |
293 | ||
294 | return fence; | |
295 | } | |
efd4ccb5 CK |
296 | |
297 | void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr) | |
298 | { | |
299 | mutex_init(&mgr->lock); | |
300 | idr_init(&mgr->ctx_handles); | |
301 | } | |
302 | ||
303 | void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr) | |
304 | { | |
305 | struct amdgpu_ctx *ctx; | |
306 | struct idr *idp; | |
307 | uint32_t id; | |
308 | ||
309 | idp = &mgr->ctx_handles; | |
310 | ||
311 | idr_for_each_entry(idp, ctx, id) { | |
312 | if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1) | |
313 | DRM_ERROR("ctx %p is still alive\n", ctx); | |
314 | } | |
315 | ||
316 | idr_destroy(&mgr->ctx_handles); | |
317 | mutex_destroy(&mgr->lock); | |
318 | } |