Commit | Line | Data |
---|---|---|
073440d2 CK |
1 | /* |
2 | * Copyright 2016 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: Christian König | |
23 | */ | |
24 | #ifndef __AMDGPU_VM_H__ | |
25 | #define __AMDGPU_VM_H__ | |
26 | ||
02208441 | 27 | #include <linux/idr.h> |
1b1f42d8 LS |
28 | #include <linux/kfifo.h> |
29 | #include <linux/rbtree.h> | |
30 | #include <drm/gpu_scheduler.h> | |
61b100e9 | 31 | #include <drm/drm_file.h> |
073440d2 | 32 | |
073440d2 CK |
33 | #include "amdgpu_sync.h" |
34 | #include "amdgpu_ring.h" | |
620f774f | 35 | #include "amdgpu_ids.h" |
073440d2 CK |
36 | |
37 | struct amdgpu_bo_va; | |
38 | struct amdgpu_job; | |
39 | struct amdgpu_bo_list_entry; | |
40 | ||
41 | /* | |
42 | * GPUVM handling | |
43 | */ | |
44 | ||
073440d2 CK |
45 | /* Maximum number of PTEs the hardware can write with one command */ |
46 | #define AMDGPU_VM_MAX_UPDATE_SIZE 0x3FFFF | |
47 | ||
48 | /* number of entries in page table */ | |
36b32a68 | 49 | #define AMDGPU_VM_PTE_COUNT(adev) (1 << (adev)->vm_manager.block_size) |
073440d2 CK |
50 | |
51 | /* PTBs (Page Table Blocks) need to be aligned to 32K */ | |
52 | #define AMDGPU_VM_PTB_ALIGN_SIZE 32768 | |
53 | ||
35ba15f0 CK |
54 | #define AMDGPU_PTE_VALID (1ULL << 0) |
55 | #define AMDGPU_PTE_SYSTEM (1ULL << 1) | |
56 | #define AMDGPU_PTE_SNOOPED (1ULL << 2) | |
073440d2 CK |
57 | |
58 | /* VI only */ | |
35ba15f0 | 59 | #define AMDGPU_PTE_EXECUTABLE (1ULL << 4) |
073440d2 | 60 | |
35ba15f0 CK |
61 | #define AMDGPU_PTE_READABLE (1ULL << 5) |
62 | #define AMDGPU_PTE_WRITEABLE (1ULL << 6) | |
073440d2 | 63 | |
982a1348 | 64 | #define AMDGPU_PTE_FRAG(x) ((x & 0x1fULL) << 7) |
073440d2 | 65 | |
d0766e98 ZJ |
66 | /* TILED for VEGA10, reserved for older ASICs */ |
67 | #define AMDGPU_PTE_PRT (1ULL << 51) | |
284710fa | 68 | |
cf2f0a37 AD |
69 | /* PDE is handled as PTE for VEGA10 */ |
70 | #define AMDGPU_PDE_PTE (1ULL << 54) | |
71 | ||
6a42fd6f CK |
72 | /* PTE is handled as PDE for VEGA10 (Translate Further) */ |
73 | #define AMDGPU_PTE_TF (1ULL << 56) | |
74 | ||
75 | /* PDE Block Fragment Size for VEGA10 */ | |
76 | #define AMDGPU_PDE_BFS(a) ((uint64_t)a << 59) | |
77 | ||
ca02061c AD |
78 | /* VEGA10 only */ |
79 | #define AMDGPU_PTE_MTYPE(a) ((uint64_t)a << 57) | |
80 | #define AMDGPU_PTE_MTYPE_MASK AMDGPU_PTE_MTYPE(3ULL) | |
81 | ||
6d16dac8 YZ |
82 | /* For Raven */ |
83 | #define AMDGPU_MTYPE_CC 2 | |
84 | ||
85 | #define AMDGPU_PTE_DEFAULT_ATC (AMDGPU_PTE_SYSTEM \ | |
86 | | AMDGPU_PTE_SNOOPED \ | |
87 | | AMDGPU_PTE_EXECUTABLE \ | |
88 | | AMDGPU_PTE_READABLE \ | |
89 | | AMDGPU_PTE_WRITEABLE \ | |
90 | | AMDGPU_PTE_MTYPE(AMDGPU_MTYPE_CC)) | |
91 | ||
073440d2 CK |
92 | /* How to programm VM fault handling */ |
93 | #define AMDGPU_VM_FAULT_STOP_NEVER 0 | |
94 | #define AMDGPU_VM_FAULT_STOP_FIRST 1 | |
95 | #define AMDGPU_VM_FAULT_STOP_ALWAYS 2 | |
96 | ||
eb60ef2b CK |
97 | /* max number of VMHUB */ |
98 | #define AMDGPU_MAX_VMHUBS 2 | |
99 | #define AMDGPU_GFXHUB 0 | |
100 | #define AMDGPU_MMHUB 1 | |
101 | ||
102 | /* hardcode that limit for now */ | |
18d09e63 | 103 | #define AMDGPU_VA_RESERVED_SIZE (1ULL << 20) |
ff4cd389 | 104 | |
bb7939b2 CK |
105 | /* VA hole for 48bit addresses on Vega10 */ |
106 | #define AMDGPU_VA_HOLE_START 0x0000800000000000ULL | |
107 | #define AMDGPU_VA_HOLE_END 0xffff800000000000ULL | |
108 | ||
109 | /* | |
110 | * Hardware is programmed as if the hole doesn't exists with start and end | |
111 | * address values. | |
112 | * | |
113 | * This mask is used to remove the upper 16bits of the VA and so come up with | |
114 | * the linear addr value. | |
115 | */ | |
116 | #define AMDGPU_VA_HOLE_MASK 0x0000ffffffffffffULL | |
117 | ||
c3505770 CZ |
118 | /* max vmids dedicated for process */ |
119 | #define AMDGPU_VM_MAX_RESERVED_VMID 1 | |
eb60ef2b | 120 | |
9a4b7d4c HK |
121 | #define AMDGPU_VM_CONTEXT_GFX 0 |
122 | #define AMDGPU_VM_CONTEXT_COMPUTE 1 | |
123 | ||
124 | /* See vm_update_mode */ | |
125 | #define AMDGPU_VM_USE_CPU_FOR_GFX (1 << 0) | |
126 | #define AMDGPU_VM_USE_CPU_FOR_COMPUTE (1 << 1) | |
127 | ||
196f7489 CZ |
128 | /* VMPT level enumerate, and the hiberachy is: |
129 | * PDB2->PDB1->PDB0->PTB | |
130 | */ | |
131 | enum amdgpu_vm_level { | |
132 | AMDGPU_VM_PDB2, | |
133 | AMDGPU_VM_PDB1, | |
134 | AMDGPU_VM_PDB0, | |
135 | AMDGPU_VM_PTB | |
136 | }; | |
137 | ||
ec681545 CK |
138 | /* base structure for tracking BO usage in a VM */ |
139 | struct amdgpu_vm_bo_base { | |
140 | /* constant after initialization */ | |
141 | struct amdgpu_vm *vm; | |
142 | struct amdgpu_bo *bo; | |
143 | ||
144 | /* protected by bo being reserved */ | |
145 | struct list_head bo_list; | |
146 | ||
147 | /* protected by spinlock */ | |
148 | struct list_head vm_status; | |
3d7d4d3a CK |
149 | |
150 | /* protected by the BO being reserved */ | |
151 | bool moved; | |
ec681545 | 152 | }; |
9a4b7d4c | 153 | |
073440d2 | 154 | struct amdgpu_vm_pt { |
3f3333f8 | 155 | struct amdgpu_vm_bo_base base; |
78eb2f0c | 156 | bool huge; |
67003a15 CK |
157 | |
158 | /* array of page tables, one for each directory entry */ | |
3f3333f8 | 159 | struct amdgpu_vm_pt *entries; |
073440d2 CK |
160 | }; |
161 | ||
a2f14820 FK |
162 | #define AMDGPU_VM_FAULT(pasid, addr) (((u64)(pasid) << 48) | (addr)) |
163 | #define AMDGPU_VM_FAULT_PASID(fault) ((u64)(fault) >> 48) | |
164 | #define AMDGPU_VM_FAULT_ADDR(fault) ((u64)(fault) & 0xfffffffff000ULL) | |
165 | ||
073440d2 CK |
166 | struct amdgpu_vm { |
167 | /* tree of virtual addresses mapped */ | |
f808c13f | 168 | struct rb_root_cached va; |
073440d2 CK |
169 | |
170 | /* protecting invalidated */ | |
171 | spinlock_t status_lock; | |
172 | ||
3f3333f8 CK |
173 | /* BOs who needs a validation */ |
174 | struct list_head evicted; | |
175 | ||
ea09729c CK |
176 | /* PT BOs which relocated and their parent need an update */ |
177 | struct list_head relocated; | |
178 | ||
073440d2 | 179 | /* BOs moved, but not yet updated in the PT */ |
27c7b9ae | 180 | struct list_head moved; |
073440d2 | 181 | |
073440d2 CK |
182 | /* BO mappings freed, but not yet updated in the PT */ |
183 | struct list_head freed; | |
184 | ||
185 | /* contains the page directory */ | |
67003a15 | 186 | struct amdgpu_vm_pt root; |
d5884513 | 187 | struct dma_fence *last_update; |
073440d2 | 188 | |
073440d2 CK |
189 | /* protecting freed */ |
190 | spinlock_t freed_lock; | |
191 | ||
192 | /* Scheduler entity for page table updates */ | |
1b1f42d8 | 193 | struct drm_sched_entity entity; |
073440d2 | 194 | |
02208441 | 195 | unsigned int pasid; |
36bbf3bf | 196 | /* dedicated to vm */ |
620f774f | 197 | struct amdgpu_vmid *reserved_vmid[AMDGPU_MAX_VMHUBS]; |
9a4b7d4c HK |
198 | |
199 | /* Flag to indicate if VM tables are updated by CPU or GPU (SDMA) */ | |
200 | bool use_cpu_for_update; | |
51ac7eec YZ |
201 | |
202 | /* Flag to indicate ATS support from PTE for GFX9 */ | |
203 | bool pte_support_ats; | |
a2f14820 | 204 | |
c98171cc | 205 | /* Up to 128 pending retry page faults */ |
a2f14820 | 206 | DECLARE_KFIFO(faults, u64, 128); |
c98171cc FK |
207 | |
208 | /* Limit non-retry fault storms */ | |
209 | unsigned int fault_credit; | |
5b21d3e5 FK |
210 | |
211 | /* Points to the KFD process VM info */ | |
212 | struct amdkfd_process_info *process_info; | |
213 | ||
214 | /* List node in amdkfd_process_info.vm_list_head */ | |
215 | struct list_head vm_list_node; | |
216 | ||
217 | /* Valid while the PD is reserved or fenced */ | |
218 | uint64_t pd_phys_addr; | |
073440d2 CK |
219 | }; |
220 | ||
073440d2 CK |
221 | struct amdgpu_vm_manager { |
222 | /* Handling of VMIDs */ | |
620f774f | 223 | struct amdgpu_vmid_mgr id_mgr[AMDGPU_MAX_VMHUBS]; |
073440d2 CK |
224 | |
225 | /* Handling of VM fences */ | |
226 | u64 fence_context; | |
227 | unsigned seqno[AMDGPU_MAX_RINGS]; | |
228 | ||
22770e5a | 229 | uint64_t max_pfn; |
8437a097 | 230 | uint32_t num_level; |
36b32a68 | 231 | uint32_t block_size; |
e618d306 | 232 | uint32_t fragment_size; |
196f7489 | 233 | enum amdgpu_vm_level root_level; |
073440d2 CK |
234 | /* vram base address for page table entry */ |
235 | u64 vram_base_offset; | |
073440d2 CK |
236 | /* vm pte handling */ |
237 | const struct amdgpu_vm_pte_funcs *vm_pte_funcs; | |
238 | struct amdgpu_ring *vm_pte_rings[AMDGPU_MAX_RINGS]; | |
239 | unsigned vm_pte_num_rings; | |
240 | atomic_t vm_pte_next_ring; | |
284710fa CK |
241 | |
242 | /* partial resident texture handling */ | |
243 | spinlock_t prt_lock; | |
451bc8eb | 244 | atomic_t num_prt_users; |
9a4b7d4c HK |
245 | |
246 | /* controls how VM page tables are updated for Graphics and Compute. | |
247 | * BIT0[= 0] Graphics updated by SDMA [= 1] by CPU | |
248 | * BIT1[= 0] Compute updated by SDMA [= 1] by CPU | |
249 | */ | |
250 | int vm_update_mode; | |
02208441 FK |
251 | |
252 | /* PASID to VM mapping, will be used in interrupt context to | |
253 | * look up VM of a page fault | |
254 | */ | |
255 | struct idr pasid_idr; | |
256 | spinlock_t pasid_lock; | |
073440d2 CK |
257 | }; |
258 | ||
259 | void amdgpu_vm_manager_init(struct amdgpu_device *adev); | |
260 | void amdgpu_vm_manager_fini(struct amdgpu_device *adev); | |
9a4b7d4c | 261 | int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, |
02208441 | 262 | int vm_context, unsigned int pasid); |
b236fa1d | 263 | int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm); |
073440d2 | 264 | void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); |
c98171cc FK |
265 | bool amdgpu_vm_pasid_fault_credit(struct amdgpu_device *adev, |
266 | unsigned int pasid); | |
073440d2 CK |
267 | void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm, |
268 | struct list_head *validated, | |
269 | struct amdgpu_bo_list_entry *entry); | |
3f3333f8 | 270 | bool amdgpu_vm_ready(struct amdgpu_vm *vm); |
073440d2 CK |
271 | int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm, |
272 | int (*callback)(void *p, struct amdgpu_bo *bo), | |
273 | void *param); | |
663e4577 CK |
274 | int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, |
275 | struct amdgpu_vm *vm, | |
276 | uint64_t saddr, uint64_t size); | |
8fdf074f | 277 | int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_pipe_sync); |
194d2161 CK |
278 | int amdgpu_vm_update_directories(struct amdgpu_device *adev, |
279 | struct amdgpu_vm *vm); | |
073440d2 | 280 | int amdgpu_vm_clear_freed(struct amdgpu_device *adev, |
f3467818 NH |
281 | struct amdgpu_vm *vm, |
282 | struct dma_fence **fence); | |
73fb16e7 | 283 | int amdgpu_vm_handle_moved(struct amdgpu_device *adev, |
4e55eb38 | 284 | struct amdgpu_vm *vm); |
073440d2 CK |
285 | int amdgpu_vm_bo_update(struct amdgpu_device *adev, |
286 | struct amdgpu_bo_va *bo_va, | |
287 | bool clear); | |
288 | void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, | |
3f3333f8 | 289 | struct amdgpu_bo *bo, bool evicted); |
073440d2 CK |
290 | struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm, |
291 | struct amdgpu_bo *bo); | |
292 | struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, | |
293 | struct amdgpu_vm *vm, | |
294 | struct amdgpu_bo *bo); | |
295 | int amdgpu_vm_bo_map(struct amdgpu_device *adev, | |
296 | struct amdgpu_bo_va *bo_va, | |
297 | uint64_t addr, uint64_t offset, | |
268c3001 | 298 | uint64_t size, uint64_t flags); |
80f95c57 CK |
299 | int amdgpu_vm_bo_replace_map(struct amdgpu_device *adev, |
300 | struct amdgpu_bo_va *bo_va, | |
301 | uint64_t addr, uint64_t offset, | |
302 | uint64_t size, uint64_t flags); | |
073440d2 CK |
303 | int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, |
304 | struct amdgpu_bo_va *bo_va, | |
305 | uint64_t addr); | |
dc54d3d1 CK |
306 | int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev, |
307 | struct amdgpu_vm *vm, | |
308 | uint64_t saddr, uint64_t size); | |
aebc5e6f CK |
309 | struct amdgpu_bo_va_mapping *amdgpu_vm_bo_lookup_mapping(struct amdgpu_vm *vm, |
310 | uint64_t addr); | |
073440d2 CK |
311 | void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, |
312 | struct amdgpu_bo_va *bo_va); | |
fdd5faaa | 313 | void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint32_t vm_size, |
f3368128 CK |
314 | uint32_t fragment_size_default, unsigned max_level, |
315 | unsigned max_bits); | |
cfbcacf4 | 316 | int amdgpu_vm_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); |
b9bf33d5 CZ |
317 | bool amdgpu_vm_need_pipeline_sync(struct amdgpu_ring *ring, |
318 | struct amdgpu_job *job); | |
e59c0205 | 319 | void amdgpu_vm_check_compute_bug(struct amdgpu_device *adev); |
073440d2 CK |
320 | |
321 | #endif |