1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
|
From 42327493a6177b62edf449ff1142a1cd9c1d046c Mon Sep 17 00:00:00 2001
From: = Christian Koenig <christian.koenig@amd.com>
Date: Wed, 9 Aug 2017 14:15:46 +0200
Subject: [PATCH 0864/4131] drm/amdgpu: rework page directory filling v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Keep track off relocated PDs/PTs instead of walking and checking all PDs.
v2: fix root PD handling
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Kalyan Alle <kalyan.alle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 89 +++++++++++++++++++++++-----------
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 3 ++
2 files changed, 63 insertions(+), 29 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 05737f2..413dc75 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -196,7 +196,7 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
}
spin_lock(&vm->status_lock);
- list_del_init(&bo_base->vm_status);
+ list_move(&bo_base->vm_status, &vm->relocated);
}
spin_unlock(&vm->status_lock);
@@ -314,8 +314,10 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
entry->base.vm = vm;
entry->base.bo = pt;
list_add_tail(&entry->base.bo_list, &pt->va);
- INIT_LIST_HEAD(&entry->base.vm_status);
- entry->addr = 0;
+ spin_lock(&vm->status_lock);
+ list_add(&entry->base.vm_status, &vm->relocated);
+ spin_unlock(&vm->status_lock);
+ entry->addr = ~0ULL;
}
if (level < adev->vm_manager.num_level) {
@@ -1000,18 +1002,17 @@ static int amdgpu_vm_wait_pd(struct amdgpu_device *adev, struct amdgpu_vm *vm,
*/
static int amdgpu_vm_update_level(struct amdgpu_device *adev,
struct amdgpu_vm *vm,
- struct amdgpu_vm_pt *parent,
- unsigned level)
+ struct amdgpu_vm_pt *parent)
{
struct amdgpu_bo *shadow;
struct amdgpu_ring *ring = NULL;
uint64_t pd_addr, shadow_addr = 0;
- uint32_t incr = amdgpu_vm_bo_size(adev, level + 1);
uint64_t last_pde = ~0, last_pt = ~0, last_shadow = ~0;
unsigned count = 0, pt_idx, ndw = 0;
struct amdgpu_job *job;
struct amdgpu_pte_update_params params;
struct dma_fence *fence = NULL;
+ uint32_t incr;
int r;
@@ -1059,12 +1060,17 @@ static int amdgpu_vm_update_level(struct amdgpu_device *adev,
/* walk over the address space and update the directory */
for (pt_idx = 0; pt_idx <= parent->last_entry_used; ++pt_idx) {
- struct amdgpu_bo *bo = parent->entries[pt_idx].base.bo;
+ struct amdgpu_vm_pt *entry = &parent->entries[pt_idx];
+ struct amdgpu_bo *bo = entry->base.bo;
uint64_t pde, pt;
if (bo == NULL)
continue;
+ spin_lock(&vm->status_lock);
+ list_del_init(&entry->base.vm_status);
+ spin_unlock(&vm->status_lock);
+
pt = amdgpu_bo_gpu_offset(bo);
pt = amdgpu_gart_get_vm_pde(adev, pt);
/* Don't update huge pages here */
@@ -1075,6 +1081,7 @@ static int amdgpu_vm_update_level(struct amdgpu_device *adev,
parent->entries[pt_idx].addr = pt | AMDGPU_PTE_VALID;
pde = pd_addr + pt_idx * 8;
+ incr = amdgpu_bo_size(bo);
if (((last_pde + 8 * count) != pde) ||
((last_pt + incr * count) != pt) ||
(count == AMDGPU_VM_MAX_UPDATE_SIZE)) {
@@ -1135,20 +1142,6 @@ static int amdgpu_vm_update_level(struct amdgpu_device *adev,
dma_fence_put(fence);
}
}
- /*
- * Recurse into the subdirectories. This recursion is harmless because
- * we only have a maximum of 5 layers.
- */
- for (pt_idx = 0; pt_idx <= parent->last_entry_used; ++pt_idx) {
- struct amdgpu_vm_pt *entry = &parent->entries[pt_idx];
-
- if (!entry->base.bo)
- continue;
-
- r = amdgpu_vm_update_level(adev, vm, entry, level + 1);
- if (r)
- return r;
- }
return 0;
@@ -1164,7 +1157,8 @@ static int amdgpu_vm_update_level(struct amdgpu_device *adev,
*
* Mark all PD level as invalid after an error.
*/
-static void amdgpu_vm_invalidate_level(struct amdgpu_vm_pt *parent)
+static void amdgpu_vm_invalidate_level(struct amdgpu_vm *vm,
+ struct amdgpu_vm_pt *parent)
{
unsigned pt_idx;
@@ -1179,7 +1173,10 @@ static void amdgpu_vm_invalidate_level(struct amdgpu_vm_pt *parent)
continue;
entry->addr = ~0ULL;
- amdgpu_vm_invalidate_level(entry);
+ spin_lock(&vm->status_lock);
+ list_move(&entry->base.vm_status, &vm->relocated);
+ spin_unlock(&vm->status_lock);
+ amdgpu_vm_invalidate_level(vm, entry);
}
}
@@ -1197,9 +1194,38 @@ int amdgpu_vm_update_directories(struct amdgpu_device *adev,
{
int r = 0;
- r = amdgpu_vm_update_level(adev, vm, &vm->root, 0);
- if (r)
- amdgpu_vm_invalidate_level(&vm->root);
+ spin_lock(&vm->status_lock);
+ while (!list_empty(&vm->relocated)) {
+ struct amdgpu_vm_bo_base *bo_base;
+ struct amdgpu_bo *bo;
+
+ bo_base = list_first_entry(&vm->relocated,
+ struct amdgpu_vm_bo_base,
+ vm_status);
+ spin_unlock(&vm->status_lock);
+
+ bo = bo_base->bo->parent;
+ if (bo) {
+ struct amdgpu_vm_bo_base *parent;
+ struct amdgpu_vm_pt *pt;
+
+ parent = list_first_entry(&bo->va,
+ struct amdgpu_vm_bo_base,
+ bo_list);
+ pt = container_of(parent, struct amdgpu_vm_pt, base);
+
+ r = amdgpu_vm_update_level(adev, vm, pt);
+ if (r) {
+ amdgpu_vm_invalidate_level(vm, &vm->root);
+ return r;
+ }
+ spin_lock(&vm->status_lock);
+ } else {
+ spin_lock(&vm->status_lock);
+ list_del_init(&bo_base->vm_status);
+ }
+ }
+ spin_unlock(&vm->status_lock);
if (vm->use_cpu_for_update) {
/* Flush HDP */
@@ -1601,7 +1627,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
error_free:
amdgpu_job_free(job);
- amdgpu_vm_invalidate_level(&vm->root);
+ amdgpu_vm_invalidate_level(vm, &vm->root);
return r;
}
@@ -2408,9 +2434,13 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
continue;
}
- /* Don't add page tables to the moved state */
- if (bo->tbo.type == ttm_bo_type_kernel)
+ if (bo->tbo.type == ttm_bo_type_kernel) {
+ spin_lock(&bo_base->vm->status_lock);
+ if (list_empty(&bo_base->vm_status))
+ list_add(&bo_base->vm_status, &vm->relocated);
+ spin_unlock(&bo_base->vm->status_lock);
continue;
+ }
spin_lock(&bo_base->vm->status_lock);
list_move(&bo_base->vm_status, &bo_base->vm->moved);
@@ -2500,6 +2530,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
vm->reserved_vmid[i] = NULL;
spin_lock_init(&vm->status_lock);
INIT_LIST_HEAD(&vm->evicted);
+ INIT_LIST_HEAD(&vm->relocated);
INIT_LIST_HEAD(&vm->moved);
INIT_LIST_HEAD(&vm->freed);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index c01a09c..427a8c7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -129,6 +129,9 @@ struct amdgpu_vm {
/* BOs who needs a validation */
struct list_head evicted;
+ /* PT BOs which relocated and their parent need an update */
+ struct list_head relocated;
+
/* BOs moved, but not yet updated in the PT */
struct list_head moved;
--
2.7.4
|
