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
|
From 26605d32bf6727a47cda37191eeecaa7f92138ec Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Tue, 3 Nov 2015 20:58:50 +0100
Subject: [PATCH 0237/1110] drm/amdgpu: grab VMID before submitting job v5
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
This allows the scheduler to handle the dependencies on ID contention as well.
v2: grab id only once
v3: use a separate lock for the VMIDs
v4: cleanup after semaphore removal
v5: minor coding style change
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 ++++
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 17 +++++------------
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 26 +++++++++++++++++++++++++-
4 files changed, 35 insertions(+), 13 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index a1da7e2..f970afa 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -737,6 +737,7 @@ struct amdgpu_ib {
struct amdgpu_ring *ring;
struct amdgpu_fence *fence;
struct amdgpu_user_fence *user;
+ bool grabbed_vmid;
struct amdgpu_vm *vm;
struct amdgpu_ctx *ctx;
struct amdgpu_sync sync;
@@ -886,6 +887,9 @@ struct amdgpu_vm {
};
struct amdgpu_vm_manager {
+ /* protecting IDs */
+ struct mutex lock;
+
struct {
struct fence *active;
atomic_long_t owner;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index c96437c..9bab3dc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -1456,6 +1456,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
/* mutex initialization are all done here so we
* can recall function without having locking issues */
mutex_init(&adev->ring_lock);
+ mutex_init(&adev->vm_manager.lock);
atomic_set(&adev->irq.ih.lock, 0);
mutex_init(&adev->gem.mutex);
mutex_init(&adev->pm.mutex);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 54cede3..56ae9a5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -142,21 +142,17 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
return -EINVAL;
}
+ if (vm && !ibs->grabbed_vmid) {
+ dev_err(adev->dev, "VM IB without ID\n");
+ return -EINVAL;
+ }
+
r = amdgpu_ring_lock(ring, (256 + AMDGPU_NUM_SYNCS * 8) * num_ibs);
if (r) {
dev_err(adev->dev, "scheduling IB failed (%d).\n", r);
return r;
}
- if (vm) {
- /* grab a vm id if necessary */
- r = amdgpu_vm_grab_id(ibs->vm, ibs->ring, &ibs->sync);
- if (r) {
- amdgpu_ring_unlock_undo(ring);
- return r;
- }
- }
-
r = amdgpu_sync_wait(&ibs->sync);
if (r) {
amdgpu_ring_unlock_undo(ring);
@@ -207,9 +203,6 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
AMDGPU_FENCE_FLAG_64BIT);
}
- if (ib->vm)
- amdgpu_vm_fence(adev, ib->vm, &ib->fence->base);
-
amdgpu_ring_unlock_commit(ring);
return 0;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index dd9fac3..b22a95f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -31,7 +31,31 @@
static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
{
struct amdgpu_job *job = to_amdgpu_job(sched_job);
- return amdgpu_sync_get_fence(&job->ibs->sync);
+ struct amdgpu_sync *sync = &job->ibs->sync;
+ struct amdgpu_vm *vm = job->ibs->vm;
+
+ struct fence *fence = amdgpu_sync_get_fence(sync);
+
+ if (fence == NULL && vm && !job->ibs->grabbed_vmid) {
+ struct amdgpu_ring *ring = job->ibs->ring;
+ struct amdgpu_device *adev = ring->adev;
+ int r;
+
+ mutex_lock(&adev->vm_manager.lock);
+ r = amdgpu_vm_grab_id(vm, ring, sync);
+ if (r) {
+ DRM_ERROR("Error getting VM ID (%d)\n", r);
+ } else {
+ fence = &job->base.s_fence->base;
+ amdgpu_vm_fence(ring->adev, vm, fence);
+ job->ibs->grabbed_vmid = true;
+ }
+ mutex_unlock(&adev->vm_manager.lock);
+
+ fence = amdgpu_sync_get_fence(sync);
+ }
+
+ return fence;
}
static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job)
--
2.7.4
|
