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
|
From a3e2b400ab91822ed1680131b319cd1ee6832cb0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Mon, 25 Jan 2016 14:27:31 +0100
Subject: [PATCH 0252/1110] drm/amdgpu: split VM mappings into smaller
operations (v3)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
If we can't copy entries from the GTT or fill them with one command split
up the mapping operation into multiple ones.
v2: agd: rebase on upstream
v3: squash in Christian's fix
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 104 ++++++++++++++++++++++++---------
1 file changed, 77 insertions(+), 27 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 3469d11..ae1d20a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -703,42 +703,32 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
*
* @adev: amdgpu_device pointer
* @gtt: GART instance to use for mapping
+ * @gtt_flags: flags as they are used for GTT
* @vm: requested vm
- * @mapping: mapped range and flags to use for the update
+ * @start: start of mapped range
+ * @last: last mapped entry
+ * @flags: flags for the entries
* @addr: addr to set the area to
- * @gtt_flags: flags as they are used for GTT
* @fence: optional resulting fence
*
- * Fill in the page table entries for @mapping.
+ * Fill in the page table entries between @start and @last.
* Returns 0 for success, -EINVAL for failure.
- *
- * Object have to be reserved and mutex must be locked!
*/
static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
struct amdgpu_gart *gtt,
uint32_t gtt_flags,
struct amdgpu_vm *vm,
- struct amdgpu_bo_va_mapping *mapping,
- uint64_t addr, struct fence **fence)
+ uint64_t start, uint64_t last,
+ uint32_t flags, uint64_t addr,
+ struct fence **fence)
{
struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
unsigned nptes, ncmds, ndw;
- uint32_t flags = gtt_flags;
struct amdgpu_ib *ib;
struct fence *f = NULL;
int r;
- /* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here
- * but in case of something, we filter the flags in first place
- */
- if (!(mapping->flags & AMDGPU_PTE_READABLE))
- flags &= ~AMDGPU_PTE_READABLE;
- if (!(mapping->flags & AMDGPU_PTE_WRITEABLE))
- flags &= ~AMDGPU_PTE_WRITEABLE;
-
- trace_amdgpu_vm_bo_update(mapping);
-
- nptes = mapping->it.last - mapping->it.start + 1;
+ nptes = last - start + 1;
/*
* reserve space for one command every (1 << BLOCK_SIZE)
@@ -780,10 +770,8 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
ib->length_dw = 0;
- r = amdgpu_vm_update_ptes(adev, gtt, gtt_flags, vm, ib,
- mapping->it.start, mapping->it.last + 1,
- addr + mapping->offset, flags);
-
+ r = amdgpu_vm_update_ptes(adev, gtt, gtt_flags, vm, ib, start,
+ last + 1, addr, flags);
if (r) {
amdgpu_ib_free(adev, ib);
kfree(ib);
@@ -814,6 +802,68 @@ error_free:
}
/**
+ * amdgpu_vm_bo_split_mapping - split a mapping into smaller chunks
+ *
+ * @adev: amdgpu_device pointer
+ * @gtt: GART instance to use for mapping
+ * @vm: requested vm
+ * @mapping: mapped range and flags to use for the update
+ * @addr: addr to set the area to
+ * @gtt_flags: flags as they are used for GTT
+ * @fence: optional resulting fence
+ *
+ * Split the mapping into smaller chunks so that each update fits
+ * into a SDMA IB.
+ * Returns 0 for success, -EINVAL for failure.
+ */
+static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev,
+ struct amdgpu_gart *gtt,
+ uint32_t gtt_flags,
+ struct amdgpu_vm *vm,
+ struct amdgpu_bo_va_mapping *mapping,
+ uint64_t addr, struct fence **fence)
+{
+ const uint64_t max_size = 64ULL * 1024ULL * 1024ULL / AMDGPU_GPU_PAGE_SIZE;
+
+ uint64_t start = mapping->it.start;
+ uint32_t flags = gtt_flags;
+ int r;
+
+ /* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here
+ * but in case of something, we filter the flags in first place
+ */
+ if (!(mapping->flags & AMDGPU_PTE_READABLE))
+ flags &= ~AMDGPU_PTE_READABLE;
+ if (!(mapping->flags & AMDGPU_PTE_WRITEABLE))
+ flags &= ~AMDGPU_PTE_WRITEABLE;
+
+ trace_amdgpu_vm_bo_update(mapping);
+
+ addr += mapping->offset;
+
+ if (!gtt || ((gtt == &adev->gart) && (flags == gtt_flags)))
+ return amdgpu_vm_bo_update_mapping(adev, gtt, gtt_flags, vm,
+ start, mapping->it.last,
+ flags, addr, fence);
+
+ while (start != mapping->it.last + 1) {
+ uint64_t last;
+
+ last = min((uint64_t)mapping->it.last, start + max_size);
+ r = amdgpu_vm_bo_update_mapping(adev, gtt, gtt_flags, vm,
+ start, last, flags, addr,
+ fence);
+ if (r)
+ return r;
+
+ start = last + 1;
+ addr += max_size;
+ }
+
+ return 0;
+}
+
+/**
* amdgpu_vm_bo_update - update all BO mappings in the vm page table
*
* @adev: amdgpu_device pointer
@@ -862,8 +912,8 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
spin_unlock(&vm->status_lock);
list_for_each_entry(mapping, &bo_va->invalids, list) {
- r = amdgpu_vm_bo_update_mapping(adev, gtt, flags, vm, mapping, addr,
- &bo_va->last_pt_update);
+ r = amdgpu_vm_bo_split_mapping(adev, gtt, flags, vm, mapping, addr,
+ &bo_va->last_pt_update);
if (r)
return r;
}
@@ -909,8 +959,8 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
struct amdgpu_bo_va_mapping, list);
list_del(&mapping->list);
spin_unlock(&vm->freed_lock);
- r = amdgpu_vm_bo_update_mapping(adev, NULL, 0, vm, mapping,
- 0, NULL);
+ r = amdgpu_vm_bo_split_mapping(adev, NULL, 0, vm, mapping,
+ 0, NULL);
kfree(mapping);
if (r)
return r;
--
2.7.4
|
