2 * Copyright 2016 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
22 * Authors: Christian König
24 #ifndef __AMDGPU_VM_H__
25 #define __AMDGPU_VM_H__
27 #include <linux/rbtree.h>
29 #include "gpu_scheduler.h"
30 #include "amdgpu_sync.h"
31 #include "amdgpu_ring.h"
35 struct amdgpu_bo_list_entry;
41 /* maximum number of VMIDs */
42 #define AMDGPU_NUM_VM 16
44 /* Maximum number of PTEs the hardware can write with one command */
45 #define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFF
47 /* number of entries in page table */
48 #define AMDGPU_VM_PTE_COUNT(adev) (1 << (adev)->vm_manager.block_size)
50 /* PTBs (Page Table Blocks) need to be aligned to 32K */
51 #define AMDGPU_VM_PTB_ALIGN_SIZE 32768
53 #define AMDGPU_PTE_VALID (1ULL << 0)
54 #define AMDGPU_PTE_SYSTEM (1ULL << 1)
55 #define AMDGPU_PTE_SNOOPED (1ULL << 2)
58 #define AMDGPU_PTE_EXECUTABLE (1ULL << 4)
60 #define AMDGPU_PTE_READABLE (1ULL << 5)
61 #define AMDGPU_PTE_WRITEABLE (1ULL << 6)
63 #define AMDGPU_PTE_FRAG(x) ((x & 0x1fULL) << 7)
65 /* TILED for VEGA10, reserved for older ASICs */
66 #define AMDGPU_PTE_PRT (1ULL << 51)
68 /* PDE is handled as PTE for VEGA10 */
69 #define AMDGPU_PDE_PTE (1ULL << 54)
72 #define AMDGPU_PTE_MTYPE(a) ((uint64_t)a << 57)
73 #define AMDGPU_PTE_MTYPE_MASK AMDGPU_PTE_MTYPE(3ULL)
75 /* How to programm VM fault handling */
76 #define AMDGPU_VM_FAULT_STOP_NEVER 0
77 #define AMDGPU_VM_FAULT_STOP_FIRST 1
78 #define AMDGPU_VM_FAULT_STOP_ALWAYS 2
80 /* max number of VMHUB */
81 #define AMDGPU_MAX_VMHUBS 2
82 #define AMDGPU_GFXHUB 0
83 #define AMDGPU_MMHUB 1
85 /* hardcode that limit for now */
86 #define AMDGPU_VA_RESERVED_SIZE (8 << 20)
87 /* max vmids dedicated for process */
88 #define AMDGPU_VM_MAX_RESERVED_VMID 1
90 #define AMDGPU_VM_CONTEXT_GFX 0
91 #define AMDGPU_VM_CONTEXT_COMPUTE 1
93 /* See vm_update_mode */
94 #define AMDGPU_VM_USE_CPU_FOR_GFX (1 << 0)
95 #define AMDGPU_VM_USE_CPU_FOR_COMPUTE (1 << 1)
97 /* base structure for tracking BO usage in a VM */
98 struct amdgpu_vm_bo_base {
99 /* constant after initialization */
100 struct amdgpu_vm *vm;
101 struct amdgpu_bo *bo;
103 /* protected by bo being reserved */
104 struct list_head bo_list;
106 /* protected by spinlock */
107 struct list_head vm_status;
109 /* protected by the BO being reserved */
113 struct amdgpu_vm_pt {
114 struct amdgpu_vm_bo_base base;
117 /* array of page tables, one for each directory entry */
118 struct amdgpu_vm_pt *entries;
119 unsigned last_entry_used;
123 /* tree of virtual addresses mapped */
126 /* protecting invalidated */
127 spinlock_t status_lock;
129 /* BOs who needs a validation */
130 struct list_head evicted;
132 /* PT BOs which relocated and their parent need an update */
133 struct list_head relocated;
135 /* BOs moved, but not yet updated in the PT */
136 struct list_head moved;
138 /* BO mappings freed, but not yet updated in the PT */
139 struct list_head freed;
141 /* contains the page directory */
142 struct amdgpu_vm_pt root;
143 struct dma_fence *last_update;
145 /* protecting freed */
146 spinlock_t freed_lock;
148 /* Scheduler entity for page table updates */
149 struct amd_sched_entity entity;
153 /* dedicated to vm */
154 struct amdgpu_vm_id *reserved_vmid[AMDGPU_MAX_VMHUBS];
156 /* Flag to indicate if VM tables are updated by CPU or GPU (SDMA) */
157 bool use_cpu_for_update;
159 /* Flag to indicate ATS support from PTE for GFX9 */
160 bool pte_support_ats;
163 struct amdgpu_vm_id {
164 struct list_head list;
165 struct amdgpu_sync active;
166 struct dma_fence *last_flush;
169 uint64_t pd_gpu_addr;
170 /* last flushed PD/PT update */
171 struct dma_fence *flushed_updates;
173 uint32_t current_gpu_reset_count;
183 struct amdgpu_vm_id_manager {
186 struct list_head ids_lru;
187 struct amdgpu_vm_id ids[AMDGPU_NUM_VM];
188 atomic_t reserved_vmid_num;
191 struct amdgpu_vm_manager {
192 /* Handling of VMIDs */
193 struct amdgpu_vm_id_manager id_mgr[AMDGPU_MAX_VMHUBS];
195 /* Handling of VM fences */
197 unsigned seqno[AMDGPU_MAX_RINGS];
203 uint32_t fragment_size;
204 /* vram base address for page table entry */
205 u64 vram_base_offset;
206 /* vm pte handling */
207 const struct amdgpu_vm_pte_funcs *vm_pte_funcs;
208 struct amdgpu_ring *vm_pte_rings[AMDGPU_MAX_RINGS];
209 unsigned vm_pte_num_rings;
210 atomic_t vm_pte_next_ring;
211 /* client id counter */
212 atomic64_t client_counter;
214 /* partial resident texture handling */
216 atomic_t num_prt_users;
218 /* controls how VM page tables are updated for Graphics and Compute.
219 * BIT0[= 0] Graphics updated by SDMA [= 1] by CPU
220 * BIT1[= 0] Compute updated by SDMA [= 1] by CPU
225 void amdgpu_vm_manager_init(struct amdgpu_device *adev);
226 void amdgpu_vm_manager_fini(struct amdgpu_device *adev);
227 int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
229 void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
230 void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
231 struct list_head *validated,
232 struct amdgpu_bo_list_entry *entry);
233 bool amdgpu_vm_ready(struct amdgpu_vm *vm);
234 int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
235 int (*callback)(void *p, struct amdgpu_bo *bo),
237 int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
238 struct amdgpu_vm *vm,
239 uint64_t saddr, uint64_t size);
240 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
241 struct amdgpu_sync *sync, struct dma_fence *fence,
242 struct amdgpu_job *job);
243 int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_pipe_sync);
244 void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vmhub,
246 void amdgpu_vm_reset_all_ids(struct amdgpu_device *adev);
247 int amdgpu_vm_update_directories(struct amdgpu_device *adev,
248 struct amdgpu_vm *vm);
249 int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
250 struct amdgpu_vm *vm,
251 struct dma_fence **fence);
252 int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
253 struct amdgpu_vm *vm);
254 int amdgpu_vm_bo_update(struct amdgpu_device *adev,
255 struct amdgpu_bo_va *bo_va,
257 void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
258 struct amdgpu_bo *bo, bool evicted);
259 struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm,
260 struct amdgpu_bo *bo);
261 struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
262 struct amdgpu_vm *vm,
263 struct amdgpu_bo *bo);
264 int amdgpu_vm_bo_map(struct amdgpu_device *adev,
265 struct amdgpu_bo_va *bo_va,
266 uint64_t addr, uint64_t offset,
267 uint64_t size, uint64_t flags);
268 int amdgpu_vm_bo_replace_map(struct amdgpu_device *adev,
269 struct amdgpu_bo_va *bo_va,
270 uint64_t addr, uint64_t offset,
271 uint64_t size, uint64_t flags);
272 int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
273 struct amdgpu_bo_va *bo_va,
275 int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev,
276 struct amdgpu_vm *vm,
277 uint64_t saddr, uint64_t size);
278 struct amdgpu_bo_va_mapping *amdgpu_vm_bo_lookup_mapping(struct amdgpu_vm *vm,
280 void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
281 struct amdgpu_bo_va *bo_va);
282 void amdgpu_vm_set_fragment_size(struct amdgpu_device *adev,
283 uint32_t fragment_size_default);
284 void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint64_t vm_size,
285 uint32_t fragment_size_default);
286 int amdgpu_vm_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
287 bool amdgpu_vm_need_pipeline_sync(struct amdgpu_ring *ring,
288 struct amdgpu_job *job);
289 void amdgpu_vm_check_compute_bug(struct amdgpu_device *adev);