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
|
From 592390152b5e395db896bf6073a32e0f4cbcb1c7 Mon Sep 17 00:00:00 2001
From: Felix Kuehling <Felix.Kuehling@amd.com>
Date: Tue, 13 Mar 2018 16:05:59 -0400
Subject: [PATCH 2721/2940] drm/amdkfd: Fix and simplify sync object handling
for KFD
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
The adev parameter in amdgpu_sync_fence and amdgpu_sync_resv is only
needed for updating sync->last_vm_update. This breaks if different
adevs are passed to calls for the same sync object.
Always pass NULL for calls from KFD because sync objects used for
KFD don't belong to any particular device, and KFD doesn't need the
sync->last_vm_update fence.
This fixes kernel log warnings on multi-GPU systems after recent
changes in amdgpu_amdkfd_gpuvm_restore_process_bos.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 28 ++++---------------
1 file changed, 5 insertions(+), 23 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 190b92f81153..59d6dcf8ba43 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -398,23 +398,6 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
return 0;
}
-static int sync_vm_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
- struct dma_fence *f)
-{
- int ret = amdgpu_sync_fence(adev, sync, f, false);
-
- /* Sync objects can't handle multiple GPUs (contexts) updating
- * sync->last_vm_update. Fortunately we don't need it for
- * KFD's purposes, so we can just drop that fence.
- */
- if (sync->last_vm_update) {
- dma_fence_put(sync->last_vm_update);
- sync->last_vm_update = NULL;
- }
-
- return ret;
-}
-
static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
{
struct amdgpu_bo *pd = vm->root.base.bo;
@@ -425,7 +408,7 @@ static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
if (ret)
return ret;
- return sync_vm_fence(adev, sync, vm->last_update);
+ return amdgpu_sync_fence(NULL, sync, vm->last_update, false);
}
/* add_bo_to_vm - Add a BO to a VM
@@ -823,7 +806,7 @@ static int unmap_bo_from_gpuvm(struct amdgpu_device *adev,
/* Add the eviction fence back */
amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true);
- sync_vm_fence(adev, sync, bo_va->last_pt_update);
+ amdgpu_sync_fence(NULL, sync, bo_va->last_pt_update, false);
return 0;
}
@@ -842,7 +825,7 @@ static int update_gpuvm_pte(struct amdgpu_device *adev,
return ret;
}
- return sync_vm_fence(adev, sync, bo_va->last_pt_update);
+ return amdgpu_sync_fence(NULL, sync, bo_va->last_pt_update, false);
}
static int map_bo_to_gpuvm(struct amdgpu_device *adev,
@@ -902,7 +885,7 @@ static int process_sync_pds_resv(struct amdkfd_process_info *process_info,
vm_list_node) {
struct amdgpu_bo *pd = peer_vm->root.base.bo;
- ret = amdgpu_sync_resv(amdgpu_ttm_adev(pd->tbo.bdev),
+ ret = amdgpu_sync_resv(NULL,
sync, pd->tbo.resv,
AMDGPU_FENCE_OWNER_UNDEFINED, false);
if (ret)
@@ -2075,8 +2058,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
pr_debug("Memory eviction: Validate BOs failed. Try again\n");
goto validate_map_fail;
}
- ret = amdgpu_sync_fence(amdgpu_ttm_adev(bo->tbo.bdev),
- &sync_obj, bo->tbo.moving, false);
+ ret = amdgpu_sync_fence(NULL, &sync_obj, bo->tbo.moving, false);
if (ret) {
pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
goto validate_map_fail;
--
2.17.1
|
