aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c66
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h4
2 files changed, 21 insertions, 49 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 1a8f4e0dd023..f0deedcaf1c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -119,9 +119,7 @@ static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
119 * is currently evicted. add the bo to the evicted list to make sure it 119 * is currently evicted. add the bo to the evicted list to make sure it
120 * is validated on next vm use to avoid fault. 120 * is validated on next vm use to avoid fault.
121 * */ 121 * */
122 spin_lock(&vm->status_lock);
123 list_move_tail(&base->vm_status, &vm->evicted); 122 list_move_tail(&base->vm_status, &vm->evicted);
124 spin_unlock(&vm->status_lock);
125} 123}
126 124
127/** 125/**
@@ -228,7 +226,6 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
228 struct ttm_bo_global *glob = adev->mman.bdev.glob; 226 struct ttm_bo_global *glob = adev->mman.bdev.glob;
229 int r; 227 int r;
230 228
231 spin_lock(&vm->status_lock);
232 while (!list_empty(&vm->evicted)) { 229 while (!list_empty(&vm->evicted)) {
233 struct amdgpu_vm_bo_base *bo_base; 230 struct amdgpu_vm_bo_base *bo_base;
234 struct amdgpu_bo *bo; 231 struct amdgpu_bo *bo;
@@ -236,10 +233,8 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
236 bo_base = list_first_entry(&vm->evicted, 233 bo_base = list_first_entry(&vm->evicted,
237 struct amdgpu_vm_bo_base, 234 struct amdgpu_vm_bo_base,
238 vm_status); 235 vm_status);
239 spin_unlock(&vm->status_lock);
240 236
241 bo = bo_base->bo; 237 bo = bo_base->bo;
242 BUG_ON(!bo);
243 if (bo->parent) { 238 if (bo->parent) {
244 r = validate(param, bo); 239 r = validate(param, bo);
245 if (r) 240 if (r)
@@ -259,13 +254,14 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
259 return r; 254 return r;
260 } 255 }
261 256
262 spin_lock(&vm->status_lock); 257 if (bo->tbo.type != ttm_bo_type_kernel) {
263 if (bo->tbo.type != ttm_bo_type_kernel) 258 spin_lock(&vm->moved_lock);
264 list_move(&bo_base->vm_status, &vm->moved); 259 list_move(&bo_base->vm_status, &vm->moved);
265 else 260 spin_unlock(&vm->moved_lock);
261 } else {
266 list_move(&bo_base->vm_status, &vm->relocated); 262 list_move(&bo_base->vm_status, &vm->relocated);
263 }
267 } 264 }
268 spin_unlock(&vm->status_lock);
269 265
270 return 0; 266 return 0;
271} 267}
@@ -279,13 +275,7 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
279 */ 275 */
280bool amdgpu_vm_ready(struct amdgpu_vm *vm) 276bool amdgpu_vm_ready(struct amdgpu_vm *vm)
281{ 277{
282 bool ready; 278 return list_empty(&vm->evicted);
283
284 spin_lock(&vm->status_lock);
285 ready = list_empty(&vm->evicted);
286 spin_unlock(&vm->status_lock);
287
288 return ready;
289} 279}
290 280
291/** 281/**
@@ -477,9 +467,7 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
477 pt->parent = amdgpu_bo_ref(parent->base.bo); 467 pt->parent = amdgpu_bo_ref(parent->base.bo);
478 468
479 amdgpu_vm_bo_base_init(&entry->base, vm, pt); 469 amdgpu_vm_bo_base_init(&entry->base, vm, pt);
480 spin_lock(&vm->status_lock);
481 list_move(&entry->base.vm_status, &vm->relocated); 470 list_move(&entry->base.vm_status, &vm->relocated);
482 spin_unlock(&vm->status_lock);
483 } 471 }
484 472
485 if (level < AMDGPU_VM_PTB) { 473 if (level < AMDGPU_VM_PTB) {
@@ -926,10 +914,8 @@ static void amdgpu_vm_invalidate_level(struct amdgpu_device *adev,
926 if (!entry->base.bo) 914 if (!entry->base.bo)
927 continue; 915 continue;
928 916
929 spin_lock(&vm->status_lock);
930 if (list_empty(&entry->base.vm_status)) 917 if (list_empty(&entry->base.vm_status))
931 list_add(&entry->base.vm_status, &vm->relocated); 918 list_add(&entry->base.vm_status, &vm->relocated);
932 spin_unlock(&vm->status_lock);
933 amdgpu_vm_invalidate_level(adev, vm, entry, level + 1); 919 amdgpu_vm_invalidate_level(adev, vm, entry, level + 1);
934 } 920 }
935} 921}
@@ -974,7 +960,6 @@ restart:
974 params.func = amdgpu_vm_do_set_ptes; 960 params.func = amdgpu_vm_do_set_ptes;
975 } 961 }
976 962
977 spin_lock(&vm->status_lock);
978 while (!list_empty(&vm->relocated)) { 963 while (!list_empty(&vm->relocated)) {
979 struct amdgpu_vm_bo_base *bo_base, *parent; 964 struct amdgpu_vm_bo_base *bo_base, *parent;
980 struct amdgpu_vm_pt *pt, *entry; 965 struct amdgpu_vm_pt *pt, *entry;
@@ -984,13 +969,10 @@ restart:
984 struct amdgpu_vm_bo_base, 969 struct amdgpu_vm_bo_base,
985 vm_status); 970 vm_status);
986 list_del_init(&bo_base->vm_status); 971 list_del_init(&bo_base->vm_status);
987 spin_unlock(&vm->status_lock);
988 972
989 bo = bo_base->bo->parent; 973 bo = bo_base->bo->parent;
990 if (!bo) { 974 if (!bo)
991 spin_lock(&vm->status_lock);
992 continue; 975 continue;
993 }
994 976
995 parent = list_first_entry(&bo->va, struct amdgpu_vm_bo_base, 977 parent = list_first_entry(&bo->va, struct amdgpu_vm_bo_base,
996 bo_list); 978 bo_list);
@@ -999,12 +981,10 @@ restart:
999 981
1000 amdgpu_vm_update_pde(&params, vm, pt, entry); 982 amdgpu_vm_update_pde(&params, vm, pt, entry);
1001 983
1002 spin_lock(&vm->status_lock);
1003 if (!vm->use_cpu_for_update && 984 if (!vm->use_cpu_for_update &&
1004 (ndw - params.ib->length_dw) < 32) 985 (ndw - params.ib->length_dw) < 32)
1005 break; 986 break;
1006 } 987 }
1007 spin_unlock(&vm->status_lock);
1008 988
1009 if (vm->use_cpu_for_update) { 989 if (vm->use_cpu_for_update) {
1010 /* Flush HDP */ 990 /* Flush HDP */
@@ -1107,9 +1087,7 @@ static void amdgpu_vm_handle_huge_pages(struct amdgpu_pte_update_params *p,
1107 if (entry->huge) { 1087 if (entry->huge) {
1108 /* Add the entry to the relocated list to update it. */ 1088 /* Add the entry to the relocated list to update it. */
1109 entry->huge = false; 1089 entry->huge = false;
1110 spin_lock(&p->vm->status_lock);
1111 list_move(&entry->base.vm_status, &p->vm->relocated); 1090 list_move(&entry->base.vm_status, &p->vm->relocated);
1112 spin_unlock(&p->vm->status_lock);
1113 } 1091 }
1114 return; 1092 return;
1115 } 1093 }
@@ -1588,8 +1566,9 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
1588 amdgpu_asic_flush_hdp(adev, NULL); 1566 amdgpu_asic_flush_hdp(adev, NULL);
1589 } 1567 }
1590 1568
1591 spin_lock(&vm->status_lock); 1569 spin_lock(&vm->moved_lock);
1592 list_del_init(&bo_va->base.vm_status); 1570 list_del_init(&bo_va->base.vm_status);
1571 spin_unlock(&vm->moved_lock);
1593 1572
1594 /* If the BO is not in its preferred location add it back to 1573 /* If the BO is not in its preferred location add it back to
1595 * the evicted list so that it gets validated again on the 1574 * the evicted list so that it gets validated again on the
@@ -1599,7 +1578,6 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
1599 !(bo->preferred_domains & 1578 !(bo->preferred_domains &
1600 amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type))) 1579 amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type)))
1601 list_add_tail(&bo_va->base.vm_status, &vm->evicted); 1580 list_add_tail(&bo_va->base.vm_status, &vm->evicted);
1602 spin_unlock(&vm->status_lock);
1603 1581
1604 list_splice_init(&bo_va->invalids, &bo_va->valids); 1582 list_splice_init(&bo_va->invalids, &bo_va->valids);
1605 bo_va->cleared = clear; 1583 bo_va->cleared = clear;
@@ -1811,14 +1789,14 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
1811 bool clear; 1789 bool clear;
1812 int r = 0; 1790 int r = 0;
1813 1791
1814 spin_lock(&vm->status_lock); 1792 spin_lock(&vm->moved_lock);
1815 while (!list_empty(&vm->moved)) { 1793 while (!list_empty(&vm->moved)) {
1816 struct amdgpu_bo_va *bo_va; 1794 struct amdgpu_bo_va *bo_va;
1817 struct reservation_object *resv; 1795 struct reservation_object *resv;
1818 1796
1819 bo_va = list_first_entry(&vm->moved, 1797 bo_va = list_first_entry(&vm->moved,
1820 struct amdgpu_bo_va, base.vm_status); 1798 struct amdgpu_bo_va, base.vm_status);
1821 spin_unlock(&vm->status_lock); 1799 spin_unlock(&vm->moved_lock);
1822 1800
1823 resv = bo_va->base.bo->tbo.resv; 1801 resv = bo_va->base.bo->tbo.resv;
1824 1802
@@ -1839,9 +1817,9 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
1839 if (!clear && resv != vm->root.base.bo->tbo.resv) 1817 if (!clear && resv != vm->root.base.bo->tbo.resv)
1840 reservation_object_unlock(resv); 1818 reservation_object_unlock(resv);
1841 1819
1842 spin_lock(&vm->status_lock); 1820 spin_lock(&vm->moved_lock);
1843 } 1821 }
1844 spin_unlock(&vm->status_lock); 1822 spin_unlock(&vm->moved_lock);
1845 1823
1846 return r; 1824 return r;
1847} 1825}
@@ -1903,10 +1881,10 @@ static void amdgpu_vm_bo_insert_map(struct amdgpu_device *adev,
1903 amdgpu_vm_prt_get(adev); 1881 amdgpu_vm_prt_get(adev);
1904 1882
1905 if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) { 1883 if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
1906 spin_lock(&vm->status_lock); 1884 spin_lock(&vm->moved_lock);
1907 if (list_empty(&bo_va->base.vm_status)) 1885 if (list_empty(&bo_va->base.vm_status))
1908 list_add(&bo_va->base.vm_status, &vm->moved); 1886 list_add(&bo_va->base.vm_status, &vm->moved);
1909 spin_unlock(&vm->status_lock); 1887 spin_unlock(&vm->moved_lock);
1910 } 1888 }
1911 trace_amdgpu_vm_bo_map(bo_va, mapping); 1889 trace_amdgpu_vm_bo_map(bo_va, mapping);
1912} 1890}
@@ -2216,9 +2194,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
2216 2194
2217 list_del(&bo_va->base.bo_list); 2195 list_del(&bo_va->base.bo_list);
2218 2196
2219 spin_lock(&vm->status_lock); 2197 spin_lock(&vm->moved_lock);
2220 list_del(&bo_va->base.vm_status); 2198 list_del(&bo_va->base.vm_status);
2221 spin_unlock(&vm->status_lock); 2199 spin_unlock(&vm->moved_lock);
2222 2200
2223 list_for_each_entry_safe(mapping, next, &bo_va->valids, list) { 2201 list_for_each_entry_safe(mapping, next, &bo_va->valids, list) {
2224 list_del(&mapping->list); 2202 list_del(&mapping->list);
@@ -2261,28 +2239,24 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
2261 2239
2262 bo_base->moved = true; 2240 bo_base->moved = true;
2263 if (evicted && bo->tbo.resv == vm->root.base.bo->tbo.resv) { 2241 if (evicted && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
2264 spin_lock(&bo_base->vm->status_lock);
2265 if (bo->tbo.type == ttm_bo_type_kernel) 2242 if (bo->tbo.type == ttm_bo_type_kernel)
2266 list_move(&bo_base->vm_status, &vm->evicted); 2243 list_move(&bo_base->vm_status, &vm->evicted);
2267 else 2244 else
2268 list_move_tail(&bo_base->vm_status, 2245 list_move_tail(&bo_base->vm_status,
2269 &vm->evicted); 2246 &vm->evicted);
2270 spin_unlock(&bo_base->vm->status_lock);
2271 continue; 2247 continue;
2272 } 2248 }
2273 2249
2274 if (bo->tbo.type == ttm_bo_type_kernel) { 2250 if (bo->tbo.type == ttm_bo_type_kernel) {
2275 spin_lock(&bo_base->vm->status_lock);
2276 if (list_empty(&bo_base->vm_status)) 2251 if (list_empty(&bo_base->vm_status))
2277 list_add(&bo_base->vm_status, &vm->relocated); 2252 list_add(&bo_base->vm_status, &vm->relocated);
2278 spin_unlock(&bo_base->vm->status_lock);
2279 continue; 2253 continue;
2280 } 2254 }
2281 2255
2282 spin_lock(&bo_base->vm->status_lock); 2256 spin_lock(&bo_base->vm->moved_lock);
2283 if (list_empty(&bo_base->vm_status)) 2257 if (list_empty(&bo_base->vm_status))
2284 list_add(&bo_base->vm_status, &vm->moved); 2258 list_add(&bo_base->vm_status, &vm->moved);
2285 spin_unlock(&bo_base->vm->status_lock); 2259 spin_unlock(&bo_base->vm->moved_lock);
2286 } 2260 }
2287} 2261}
2288 2262
@@ -2391,9 +2365,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
2391 vm->va = RB_ROOT_CACHED; 2365 vm->va = RB_ROOT_CACHED;
2392 for (i = 0; i < AMDGPU_MAX_VMHUBS; i++) 2366 for (i = 0; i < AMDGPU_MAX_VMHUBS; i++)
2393 vm->reserved_vmid[i] = NULL; 2367 vm->reserved_vmid[i] = NULL;
2394 spin_lock_init(&vm->status_lock);
2395 INIT_LIST_HEAD(&vm->evicted); 2368 INIT_LIST_HEAD(&vm->evicted);
2396 INIT_LIST_HEAD(&vm->relocated); 2369 INIT_LIST_HEAD(&vm->relocated);
2370 spin_lock_init(&vm->moved_lock);
2397 INIT_LIST_HEAD(&vm->moved); 2371 INIT_LIST_HEAD(&vm->moved);
2398 INIT_LIST_HEAD(&vm->freed); 2372 INIT_LIST_HEAD(&vm->freed);
2399 2373
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index d6827083572a..0196b9a782f2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -168,9 +168,6 @@ struct amdgpu_vm {
168 /* tree of virtual addresses mapped */ 168 /* tree of virtual addresses mapped */
169 struct rb_root_cached va; 169 struct rb_root_cached va;
170 170
171 /* protecting invalidated */
172 spinlock_t status_lock;
173
174 /* BOs who needs a validation */ 171 /* BOs who needs a validation */
175 struct list_head evicted; 172 struct list_head evicted;
176 173
@@ -179,6 +176,7 @@ struct amdgpu_vm {
179 176
180 /* BOs moved, but not yet updated in the PT */ 177 /* BOs moved, but not yet updated in the PT */
181 struct list_head moved; 178 struct list_head moved;
179 spinlock_t moved_lock;
182 180
183 /* BO mappings freed, but not yet updated in the PT */ 181 /* BO mappings freed, but not yet updated in the PT */
184 struct list_head freed; 182 struct list_head freed;