drm/amdgpu: consistenly use VM moved flag
[linux-2.6-block.git] / drivers / gpu / drm / amd / amdgpu / amdgpu_vm.h
CommitLineData
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
37struct amdgpu_bo_va;
38struct amdgpu_job;
39struct 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
959a2091
YZ
78
79/* For GFX9 */
ca02061c
AD
80#define AMDGPU_PTE_MTYPE(a) ((uint64_t)a << 57)
81#define AMDGPU_PTE_MTYPE_MASK AMDGPU_PTE_MTYPE(3ULL)
82
959a2091 83#define AMDGPU_MTYPE_NC 0
6d16dac8
YZ
84#define AMDGPU_MTYPE_CC 2
85
86#define AMDGPU_PTE_DEFAULT_ATC (AMDGPU_PTE_SYSTEM \
87 | AMDGPU_PTE_SNOOPED \
88 | AMDGPU_PTE_EXECUTABLE \
89 | AMDGPU_PTE_READABLE \
90 | AMDGPU_PTE_WRITEABLE \
91 | AMDGPU_PTE_MTYPE(AMDGPU_MTYPE_CC))
92
073440d2
CK
93/* How to programm VM fault handling */
94#define AMDGPU_VM_FAULT_STOP_NEVER 0
95#define AMDGPU_VM_FAULT_STOP_FIRST 1
96#define AMDGPU_VM_FAULT_STOP_ALWAYS 2
97
eb60ef2b
CK
98/* max number of VMHUB */
99#define AMDGPU_MAX_VMHUBS 2
100#define AMDGPU_GFXHUB 0
101#define AMDGPU_MMHUB 1
102
103/* hardcode that limit for now */
18d09e63 104#define AMDGPU_VA_RESERVED_SIZE (1ULL << 20)
ff4cd389 105
bb7939b2
CK
106/* VA hole for 48bit addresses on Vega10 */
107#define AMDGPU_VA_HOLE_START 0x0000800000000000ULL
108#define AMDGPU_VA_HOLE_END 0xffff800000000000ULL
109
110/*
111 * Hardware is programmed as if the hole doesn't exists with start and end
112 * address values.
113 *
114 * This mask is used to remove the upper 16bits of the VA and so come up with
115 * the linear addr value.
116 */
117#define AMDGPU_VA_HOLE_MASK 0x0000ffffffffffffULL
118
c3505770
CZ
119/* max vmids dedicated for process */
120#define AMDGPU_VM_MAX_RESERVED_VMID 1
eb60ef2b 121
9a4b7d4c
HK
122#define AMDGPU_VM_CONTEXT_GFX 0
123#define AMDGPU_VM_CONTEXT_COMPUTE 1
124
125/* See vm_update_mode */
126#define AMDGPU_VM_USE_CPU_FOR_GFX (1 << 0)
127#define AMDGPU_VM_USE_CPU_FOR_COMPUTE (1 << 1)
128
196f7489
CZ
129/* VMPT level enumerate, and the hiberachy is:
130 * PDB2->PDB1->PDB0->PTB
131 */
132enum amdgpu_vm_level {
133 AMDGPU_VM_PDB2,
134 AMDGPU_VM_PDB1,
135 AMDGPU_VM_PDB0,
136 AMDGPU_VM_PTB
137};
138
ec681545
CK
139/* base structure for tracking BO usage in a VM */
140struct amdgpu_vm_bo_base {
141 /* constant after initialization */
142 struct amdgpu_vm *vm;
143 struct amdgpu_bo *bo;
144
145 /* protected by bo being reserved */
146 struct list_head bo_list;
147
148 /* protected by spinlock */
149 struct list_head vm_status;
3d7d4d3a
CK
150
151 /* protected by the BO being reserved */
152 bool moved;
ec681545 153};
9a4b7d4c 154
073440d2 155struct amdgpu_vm_pt {
3f3333f8 156 struct amdgpu_vm_bo_base base;
78eb2f0c 157 bool huge;
67003a15
CK
158
159 /* array of page tables, one for each directory entry */
3f3333f8 160 struct amdgpu_vm_pt *entries;
073440d2
CK
161};
162
a2f14820
FK
163#define AMDGPU_VM_FAULT(pasid, addr) (((u64)(pasid) << 48) | (addr))
164#define AMDGPU_VM_FAULT_PASID(fault) ((u64)(fault) >> 48)
165#define AMDGPU_VM_FAULT_ADDR(fault) ((u64)(fault) & 0xfffffffff000ULL)
166
073440d2
CK
167struct amdgpu_vm {
168 /* tree of virtual addresses mapped */
f808c13f 169 struct rb_root_cached va;
073440d2 170
3f3333f8
CK
171 /* BOs who needs a validation */
172 struct list_head evicted;
173
ea09729c
CK
174 /* PT BOs which relocated and their parent need an update */
175 struct list_head relocated;
176
073440d2 177 /* BOs moved, but not yet updated in the PT */
27c7b9ae 178 struct list_head moved;
af4c0f65 179 spinlock_t moved_lock;
073440d2 180
073440d2
CK
181 /* BO mappings freed, but not yet updated in the PT */
182 struct list_head freed;
183
184 /* contains the page directory */
67003a15 185 struct amdgpu_vm_pt root;
d5884513 186 struct dma_fence *last_update;
073440d2 187
073440d2 188 /* Scheduler entity for page table updates */
1b1f42d8 189 struct drm_sched_entity entity;
073440d2 190
02208441 191 unsigned int pasid;
36bbf3bf 192 /* dedicated to vm */
620f774f 193 struct amdgpu_vmid *reserved_vmid[AMDGPU_MAX_VMHUBS];
9a4b7d4c
HK
194
195 /* Flag to indicate if VM tables are updated by CPU or GPU (SDMA) */
196 bool use_cpu_for_update;
51ac7eec
YZ
197
198 /* Flag to indicate ATS support from PTE for GFX9 */
199 bool pte_support_ats;
a2f14820 200
c98171cc 201 /* Up to 128 pending retry page faults */
a2f14820 202 DECLARE_KFIFO(faults, u64, 128);
c98171cc
FK
203
204 /* Limit non-retry fault storms */
205 unsigned int fault_credit;
5b21d3e5
FK
206
207 /* Points to the KFD process VM info */
208 struct amdkfd_process_info *process_info;
209
210 /* List node in amdkfd_process_info.vm_list_head */
211 struct list_head vm_list_node;
212
213 /* Valid while the PD is reserved or fenced */
214 uint64_t pd_phys_addr;
073440d2
CK
215};
216
073440d2
CK
217struct amdgpu_vm_manager {
218 /* Handling of VMIDs */
620f774f 219 struct amdgpu_vmid_mgr id_mgr[AMDGPU_MAX_VMHUBS];
073440d2
CK
220
221 /* Handling of VM fences */
222 u64 fence_context;
223 unsigned seqno[AMDGPU_MAX_RINGS];
224
22770e5a 225 uint64_t max_pfn;
8437a097 226 uint32_t num_level;
36b32a68 227 uint32_t block_size;
e618d306 228 uint32_t fragment_size;
196f7489 229 enum amdgpu_vm_level root_level;
073440d2
CK
230 /* vram base address for page table entry */
231 u64 vram_base_offset;
073440d2
CK
232 /* vm pte handling */
233 const struct amdgpu_vm_pte_funcs *vm_pte_funcs;
234 struct amdgpu_ring *vm_pte_rings[AMDGPU_MAX_RINGS];
235 unsigned vm_pte_num_rings;
236 atomic_t vm_pte_next_ring;
284710fa
CK
237
238 /* partial resident texture handling */
239 spinlock_t prt_lock;
451bc8eb 240 atomic_t num_prt_users;
9a4b7d4c
HK
241
242 /* controls how VM page tables are updated for Graphics and Compute.
243 * BIT0[= 0] Graphics updated by SDMA [= 1] by CPU
244 * BIT1[= 0] Compute updated by SDMA [= 1] by CPU
245 */
246 int vm_update_mode;
02208441
FK
247
248 /* PASID to VM mapping, will be used in interrupt context to
249 * look up VM of a page fault
250 */
251 struct idr pasid_idr;
252 spinlock_t pasid_lock;
073440d2
CK
253};
254
255void amdgpu_vm_manager_init(struct amdgpu_device *adev);
256void amdgpu_vm_manager_fini(struct amdgpu_device *adev);
9a4b7d4c 257int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
02208441 258 int vm_context, unsigned int pasid);
b236fa1d 259int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm);
073440d2 260void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
c98171cc
FK
261bool amdgpu_vm_pasid_fault_credit(struct amdgpu_device *adev,
262 unsigned int pasid);
073440d2
CK
263void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
264 struct list_head *validated,
265 struct amdgpu_bo_list_entry *entry);
3f3333f8 266bool amdgpu_vm_ready(struct amdgpu_vm *vm);
073440d2
CK
267int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
268 int (*callback)(void *p, struct amdgpu_bo *bo),
269 void *param);
663e4577
CK
270int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
271 struct amdgpu_vm *vm,
272 uint64_t saddr, uint64_t size);
8fdf074f 273int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_pipe_sync);
194d2161
CK
274int amdgpu_vm_update_directories(struct amdgpu_device *adev,
275 struct amdgpu_vm *vm);
073440d2 276int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
f3467818
NH
277 struct amdgpu_vm *vm,
278 struct dma_fence **fence);
73fb16e7 279int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
4e55eb38 280 struct amdgpu_vm *vm);
073440d2
CK
281int amdgpu_vm_bo_update(struct amdgpu_device *adev,
282 struct amdgpu_bo_va *bo_va,
283 bool clear);
284void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
3f3333f8 285 struct amdgpu_bo *bo, bool evicted);
073440d2
CK
286struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm,
287 struct amdgpu_bo *bo);
288struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
289 struct amdgpu_vm *vm,
290 struct amdgpu_bo *bo);
291int amdgpu_vm_bo_map(struct amdgpu_device *adev,
292 struct amdgpu_bo_va *bo_va,
293 uint64_t addr, uint64_t offset,
268c3001 294 uint64_t size, uint64_t flags);
80f95c57
CK
295int amdgpu_vm_bo_replace_map(struct amdgpu_device *adev,
296 struct amdgpu_bo_va *bo_va,
297 uint64_t addr, uint64_t offset,
298 uint64_t size, uint64_t flags);
073440d2
CK
299int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
300 struct amdgpu_bo_va *bo_va,
301 uint64_t addr);
dc54d3d1
CK
302int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev,
303 struct amdgpu_vm *vm,
304 uint64_t saddr, uint64_t size);
aebc5e6f
CK
305struct amdgpu_bo_va_mapping *amdgpu_vm_bo_lookup_mapping(struct amdgpu_vm *vm,
306 uint64_t addr);
073440d2
CK
307void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
308 struct amdgpu_bo_va *bo_va);
fdd5faaa 309void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint32_t vm_size,
f3368128
CK
310 uint32_t fragment_size_default, unsigned max_level,
311 unsigned max_bits);
cfbcacf4 312int amdgpu_vm_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
b9bf33d5
CZ
313bool amdgpu_vm_need_pipeline_sync(struct amdgpu_ring *ring,
314 struct amdgpu_job *job);
e59c0205 315void amdgpu_vm_check_compute_bug(struct amdgpu_device *adev);
073440d2
CK
316
317#endif