1 /* 2 * Copyright 2014 Advanced Micro Devices, Inc. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sub license, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 19 * USE OR OTHER DEALINGS IN THE SOFTWARE. 20 * 21 * The above copyright notice and this permission notice (including the 22 * next paragraph) shall be included in all copies or substantial portions 23 * of the Software. 24 * 25 */ 26 /* 27 * Authors: 28 * Christian König <christian.koenig@amd.com> 29 */ 30 31 #include <drm/drmP.h> 32 #include "amdgpu.h" 33 #include "amdgpu_trace.h" 34 35 struct amdgpu_sync_entry { 36 struct hlist_node node; 37 struct fence *fence; 38 }; 39 40 /** 41 * amdgpu_sync_create - zero init sync object 42 * 43 * @sync: sync object to initialize 44 * 45 * Just clear the sync object for now. 46 */ 47 void amdgpu_sync_create(struct amdgpu_sync *sync) 48 { 49 hash_init(sync->fences); 50 sync->last_vm_update = NULL; 51 } 52 53 static bool amdgpu_sync_same_dev(struct amdgpu_device *adev, struct fence *f) 54 { 55 struct amdgpu_fence *a_fence = to_amdgpu_fence(f); 56 struct amd_sched_fence *s_fence = to_amd_sched_fence(f); 57 58 if (a_fence) 59 return a_fence->ring->adev == adev; 60 61 if (s_fence) { 62 struct amdgpu_ring *ring; 63 64 ring = container_of(s_fence->sched, struct amdgpu_ring, sched); 65 return ring->adev == adev; 66 } 67 68 return false; 69 } 70 71 static bool amdgpu_sync_test_owner(struct fence *f, void *owner) 72 { 73 struct amdgpu_fence *a_fence = to_amdgpu_fence(f); 74 struct amd_sched_fence *s_fence = to_amd_sched_fence(f); 75 if (s_fence) 76 return s_fence->owner == owner; 77 if (a_fence) 78 return a_fence->owner == owner; 79 return false; 80 } 81 82 static void amdgpu_sync_keep_later(struct fence **keep, struct fence *fence) 83 { 84 if (*keep && fence_is_later(*keep, fence)) 85 return; 86 87 fence_put(*keep); 88 *keep = fence_get(fence); 89 } 90 91 /** 92 * amdgpu_sync_fence - remember to sync to this fence 93 * 94 * @sync: sync object to add fence to 95 * @fence: fence to sync to 96 * 97 */ 98 int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync, 99 struct fence *f) 100 { 101 struct amdgpu_sync_entry *e; 102 103 if (!f) 104 return 0; 105 106 if (amdgpu_sync_same_dev(adev, f) && 107 amdgpu_sync_test_owner(f, AMDGPU_FENCE_OWNER_VM)) 108 amdgpu_sync_keep_later(&sync->last_vm_update, f); 109 110 hash_for_each_possible(sync->fences, e, node, f->context) { 111 if (unlikely(e->fence->context != f->context)) 112 continue; 113 114 amdgpu_sync_keep_later(&e->fence, f); 115 return 0; 116 } 117 118 e = kmalloc(sizeof(struct amdgpu_sync_entry), GFP_KERNEL); 119 if (!e) 120 return -ENOMEM; 121 122 hash_add(sync->fences, &e->node, f->context); 123 e->fence = fence_get(f); 124 return 0; 125 } 126 127 static void *amdgpu_sync_get_owner(struct fence *f) 128 { 129 struct amdgpu_fence *a_fence = to_amdgpu_fence(f); 130 struct amd_sched_fence *s_fence = to_amd_sched_fence(f); 131 132 if (s_fence) 133 return s_fence->owner; 134 else if (a_fence) 135 return a_fence->owner; 136 return AMDGPU_FENCE_OWNER_UNDEFINED; 137 } 138 139 /** 140 * amdgpu_sync_resv - sync to a reservation object 141 * 142 * @sync: sync object to add fences from reservation object to 143 * @resv: reservation object with embedded fence 144 * @shared: true if we should only sync to the exclusive fence 145 * 146 * Sync to the fence 147 */ 148 int amdgpu_sync_resv(struct amdgpu_device *adev, 149 struct amdgpu_sync *sync, 150 struct reservation_object *resv, 151 void *owner) 152 { 153 struct reservation_object_list *flist; 154 struct fence *f; 155 void *fence_owner; 156 unsigned i; 157 int r = 0; 158 159 if (resv == NULL) 160 return -EINVAL; 161 162 /* always sync to the exclusive fence */ 163 f = reservation_object_get_excl(resv); 164 r = amdgpu_sync_fence(adev, sync, f); 165 166 flist = reservation_object_get_list(resv); 167 if (!flist || r) 168 return r; 169 170 for (i = 0; i < flist->shared_count; ++i) { 171 f = rcu_dereference_protected(flist->shared[i], 172 reservation_object_held(resv)); 173 if (amdgpu_sync_same_dev(adev, f)) { 174 /* VM updates are only interesting 175 * for other VM updates and moves. 176 */ 177 fence_owner = amdgpu_sync_get_owner(f); 178 if ((owner != AMDGPU_FENCE_OWNER_UNDEFINED) && 179 (fence_owner != AMDGPU_FENCE_OWNER_UNDEFINED) && 180 ((owner == AMDGPU_FENCE_OWNER_VM) != 181 (fence_owner == AMDGPU_FENCE_OWNER_VM))) 182 continue; 183 184 /* Ignore fence from the same owner as 185 * long as it isn't undefined. 186 */ 187 if (owner != AMDGPU_FENCE_OWNER_UNDEFINED && 188 fence_owner == owner) 189 continue; 190 } 191 192 r = amdgpu_sync_fence(adev, sync, f); 193 if (r) 194 break; 195 } 196 return r; 197 } 198 199 struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync) 200 { 201 struct amdgpu_sync_entry *e; 202 struct hlist_node *tmp; 203 struct fence *f; 204 int i; 205 206 hash_for_each_safe(sync->fences, i, tmp, e, node) { 207 208 f = e->fence; 209 210 hash_del(&e->node); 211 kfree(e); 212 213 if (!fence_is_signaled(f)) 214 return f; 215 216 fence_put(f); 217 } 218 return NULL; 219 } 220 221 int amdgpu_sync_wait(struct amdgpu_sync *sync) 222 { 223 struct amdgpu_sync_entry *e; 224 struct hlist_node *tmp; 225 int i, r; 226 227 hash_for_each_safe(sync->fences, i, tmp, e, node) { 228 r = fence_wait(e->fence, false); 229 if (r) 230 return r; 231 232 hash_del(&e->node); 233 fence_put(e->fence); 234 kfree(e); 235 } 236 237 return 0; 238 } 239 240 /** 241 * amdgpu_sync_free - free the sync object 242 * 243 * @sync: sync object to use 244 * 245 * Free the sync object. 246 */ 247 void amdgpu_sync_free(struct amdgpu_sync *sync) 248 { 249 struct amdgpu_sync_entry *e; 250 struct hlist_node *tmp; 251 unsigned i; 252 253 hash_for_each_safe(sync->fences, i, tmp, e, node) { 254 hash_del(&e->node); 255 fence_put(e->fence); 256 kfree(e); 257 } 258 259 fence_put(sync->last_vm_update); 260 } 261