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
|
From f227567c996e8347ebc2720c68526da69cee6534 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Thu, 5 Nov 2015 12:57:10 +0100
Subject: [PATCH 0766/1050] drm/amdgpu: optimize scheduler fence handling
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
We only need to wait for jobs to be scheduled when
the dependency is from the same scheduler.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
---
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 51 ++++++++++++++++++++-------
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 5 ++-
drivers/gpu/drm/amd/scheduler/sched_fence.c | 13 +++++++
3 files changed, 55 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index ea30d6a..b7cd108 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -211,6 +211,41 @@ static void amd_sched_entity_wakeup(struct fence *f, struct fence_cb *cb)
amd_sched_wakeup(entity->sched);
}
+static bool amd_sched_entity_add_dependency_cb(struct amd_sched_entity *entity)
+{
+ struct amd_gpu_scheduler *sched = entity->sched;
+ struct fence * fence = entity->dependency;
+ struct amd_sched_fence *s_fence;
+
+ if (fence->context == entity->fence_context) {
+ /* We can ignore fences from ourself */
+ fence_put(entity->dependency);
+ return false;
+ }
+
+ s_fence = to_amd_sched_fence(fence);
+ if (s_fence && s_fence->sched == sched) {
+ /* Fence is from the same scheduler */
+ if (test_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &fence->flags)) {
+ /* Ignore it when it is already scheduled */
+ fence_put(entity->dependency);
+ return false;
+ }
+
+ /* Wait for fence to be scheduled */
+ entity->cb.func = amd_sched_entity_wakeup;
+ list_add_tail(&entity->cb.node, &s_fence->scheduled_cb);
+ return true;
+ }
+
+ if (!fence_add_callback(entity->dependency, &entity->cb,
+ amd_sched_entity_wakeup))
+ return true;
+
+ fence_put(entity->dependency);
+ return false;
+}
+
static struct amd_sched_job *
amd_sched_entity_pop_job(struct amd_sched_entity *entity)
{
@@ -223,20 +258,9 @@ amd_sched_entity_pop_job(struct amd_sched_entity *entity)
if (!kfifo_out_peek(&entity->job_queue, &sched_job, sizeof(sched_job)))
return NULL;
- while ((entity->dependency = sched->ops->dependency(sched_job))) {
-
- if (entity->dependency->context == entity->fence_context) {
- /* We can ignore fences from ourself */
- fence_put(entity->dependency);
- continue;
- }
-
- if (fence_add_callback(entity->dependency, &entity->cb,
- amd_sched_entity_wakeup))
- fence_put(entity->dependency);
- else
+ while ((entity->dependency = sched->ops->dependency(sched_job)))
+ if (amd_sched_entity_add_dependency_cb(entity))
return NULL;
- }
return sched_job;
}
@@ -400,6 +424,7 @@ static int amd_sched_main(void *param)
atomic_inc(&sched->hw_rq_count);
fence = sched->ops->run_job(sched_job);
+ amd_sched_fence_scheduled(s_fence);
if (fence) {
r = fence_add_callback(fence, &s_fence->cb,
amd_sched_process_job);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 939692b..a0f0ae5 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -27,6 +27,8 @@
#include <linux/kfifo.h>
#include <linux/fence.h>
+#define AMD_SCHED_FENCE_SCHEDULED_BIT FENCE_FLAG_USER_BITS
+
struct amd_gpu_scheduler;
struct amd_sched_rq;
@@ -68,6 +70,7 @@ struct amd_sched_rq {
struct amd_sched_fence {
struct fence base;
struct fence_cb cb;
+ struct list_head scheduled_cb;
struct amd_gpu_scheduler *sched;
spinlock_t lock;
void *owner;
@@ -134,7 +137,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
struct amd_sched_fence *amd_sched_fence_create(
struct amd_sched_entity *s_entity, void *owner);
+void amd_sched_fence_scheduled(struct amd_sched_fence *fence);
void amd_sched_fence_signal(struct amd_sched_fence *fence);
-
#endif
diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c
index 8d2130b..87c78ee 100644
--- a/drivers/gpu/drm/amd/scheduler/sched_fence.c
+++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c
@@ -35,6 +35,8 @@ struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *s_entity
fence = kmem_cache_zalloc(sched_fence_slab, GFP_KERNEL);
if (fence == NULL)
return NULL;
+
+ INIT_LIST_HEAD(&fence->scheduled_cb);
fence->owner = owner;
fence->sched = s_entity->sched;
spin_lock_init(&fence->lock);
@@ -55,6 +57,17 @@ void amd_sched_fence_signal(struct amd_sched_fence *fence)
FENCE_TRACE(&fence->base, "was already signaled\n");
}
+void amd_sched_fence_scheduled(struct amd_sched_fence *s_fence)
+{
+ struct fence_cb *cur, *tmp;
+
+ set_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &s_fence->base.flags);
+ list_for_each_entry_safe(cur, tmp, &s_fence->scheduled_cb, node) {
+ list_del_init(&cur->node);
+ cur->func(&s_fence->base, cur);
+ }
+}
+
static const char *amd_sched_fence_get_driver_name(struct fence *fence)
{
return "amd_sched";
--
1.9.1
|
