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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
|
From 279b862e391d8593de1970fc68d4e96c812c455d Mon Sep 17 00:00:00 2001
From: Andres Rodriguez <andresx7@gmail.com>
Date: Mon, 20 Feb 2017 17:53:19 -0500
Subject: [PATCH 1973/4131] drm/amdgpu: add framework for HW specific priority
settings v9
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Add an initial framework for changing the HW priorities of rings. The
framework allows requesting priority changes for the lifetime of an
amdgpu_job. After the job completes the priority will decay to the next
lowest priority for which a request is still valid.
A new ring function set_priority() can now be populated to take care of
the HW specific programming sequence for priority changes.
v2: set priority before emitting IB, and take a ref on amdgpu_job
v3: use AMD_SCHED_PRIORITY_* instead of AMDGPU_CTX_PRIORITY_*
v4: plug amdgpu_ring_restore_priority_cb into amdgpu_job_free_cb
v5: use atomic for tracking job priorities instead of last_job
v6: rename amdgpu_ring_priority_[get/put]() and align parameters
v7: replace spinlocks with mutexes for KIQ compatibility
v8: raise ring priority during cs_ioctl, instead of job_run
v9: priority_get() before push_job()
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 4 ++
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 76 ++++++++++++++++++++++++++-
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 15 ++++++
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 7 +++
5 files changed, 103 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 11ad7f1..7c6f574 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -1226,6 +1226,8 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
job->uf_sequence = seq;
amdgpu_job_free_resources(job);
+ amdgpu_ring_priority_get(job->ring,
+ amd_sched_get_job_priority(&job->base));
trace_amdgpu_cs_ioctl(job);
amd_sched_entity_push_job(&job->base);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 4510627..83d1343 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -103,6 +103,7 @@ static void amdgpu_job_free_cb(struct amd_sched_job *s_job)
{
struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
+ amdgpu_ring_priority_put(job->ring, amd_sched_get_job_priority(s_job));
dma_fence_put(job->fence);
amdgpu_sync_free(&job->sync);
amdgpu_sync_free(&job->dep_sync);
@@ -139,6 +140,8 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
job->fence_ctx = entity->fence_context;
*f = dma_fence_get(&job->base.s_fence->finished);
amdgpu_job_free_resources(job);
+ amdgpu_ring_priority_get(job->ring,
+ amd_sched_get_job_priority(&job->base));
amd_sched_entity_push_job(&job->base);
return 0;
@@ -203,6 +206,7 @@ static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
/* if gpu reset, hw fence will be replaced here */
dma_fence_put(job->fence);
job->fence = dma_fence_get(fence);
+
amdgpu_job_free_resources(job);
return fence;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 019932a..e5ece1f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -155,6 +155,75 @@ void amdgpu_ring_undo(struct amdgpu_ring *ring)
}
/**
+ * amdgpu_ring_priority_put - restore a ring's priority
+ *
+ * @ring: amdgpu_ring structure holding the information
+ * @priority: target priority
+ *
+ * Release a request for executing at @priority
+ */
+void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
+ enum amd_sched_priority priority)
+{
+ int i;
+
+ if (!ring->funcs->set_priority)
+ return;
+
+ if (atomic_dec_return(&ring->num_jobs[priority]) > 0)
+ return;
+
+ /* no need to restore if the job is already at the lowest priority */
+ if (priority == AMD_SCHED_PRIORITY_NORMAL)
+ return;
+
+ mutex_lock(&ring->priority_mutex);
+ /* something higher prio is executing, no need to decay */
+ if (ring->priority > priority)
+ goto out_unlock;
+
+ /* decay priority to the next level with a job available */
+ for (i = priority; i >= AMD_SCHED_PRIORITY_MIN; i--) {
+ if (i == AMD_SCHED_PRIORITY_NORMAL
+ || atomic_read(&ring->num_jobs[i])) {
+ ring->priority = i;
+ ring->funcs->set_priority(ring, i);
+ break;
+ }
+ }
+
+out_unlock:
+ mutex_unlock(&ring->priority_mutex);
+}
+
+/**
+ * amdgpu_ring_priority_get - change the ring's priority
+ *
+ * @ring: amdgpu_ring structure holding the information
+ * @priority: target priority
+ *
+ * Request a ring's priority to be raised to @priority (refcounted).
+ */
+void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
+ enum amd_sched_priority priority)
+{
+ if (!ring->funcs->set_priority)
+ return;
+
+ atomic_inc(&ring->num_jobs[priority]);
+
+ mutex_lock(&ring->priority_mutex);
+ if (priority <= ring->priority)
+ goto out_unlock;
+
+ ring->priority = priority;
+ ring->funcs->set_priority(ring, priority);
+
+out_unlock:
+ mutex_unlock(&ring->priority_mutex);
+}
+
+/**
* amdgpu_ring_init - init driver ring struct.
*
* @adev: amdgpu_device pointer
@@ -169,7 +238,7 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
unsigned max_dw, struct amdgpu_irq_src *irq_src,
unsigned irq_type)
{
- int r;
+ int r, i;
int sched_hw_submission = amdgpu_sched_hw_submission;
/* Set the hw submission limit higher for KIQ because
@@ -247,9 +316,14 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
}
ring->max_dw = max_dw;
+ ring->priority = AMD_SCHED_PRIORITY_NORMAL;
+ mutex_init(&ring->priority_mutex);
INIT_LIST_HEAD(&ring->lru_list);
amdgpu_ring_lru_touch(adev, ring);
+ for (i = 0; i < AMD_SCHED_PRIORITY_MAX; ++i)
+ atomic_set(&ring->num_jobs[i], 0);
+
if (amdgpu_debugfs_ring_init(adev, ring)) {
DRM_ERROR("Failed to register debugfs file for rings !\n");
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index 1a05a4b..494ff48 100755
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -24,6 +24,7 @@
#ifndef __AMDGPU_RING_H__
#define __AMDGPU_RING_H__
+#include <drm/amdgpu_drm.h>
#include "gpu_scheduler.h"
/* max number of rings */
@@ -57,6 +58,7 @@ struct amdgpu_device;
struct amdgpu_ring;
struct amdgpu_ib;
struct amdgpu_cs_parser;
+struct amdgpu_job;
/*
* Fences.
@@ -148,6 +150,9 @@ struct amdgpu_ring_funcs {
void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg);
void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val);
void (*emit_tmz)(struct amdgpu_ring *ring, bool start);
+ /* priority functions */
+ void (*set_priority) (struct amdgpu_ring *ring,
+ enum amd_sched_priority priority);
};
struct amdgpu_ring {
@@ -188,6 +193,12 @@ struct amdgpu_ring {
volatile u32 *cond_exe_cpu_addr;
unsigned vm_inv_eng;
bool has_compute_vm_bug;
+
+ atomic_t num_jobs[AMD_SCHED_PRIORITY_MAX];
+ struct mutex priority_mutex;
+ /* protected by priority_mutex */
+ int priority;
+
#if defined(CONFIG_DEBUG_FS)
struct dentry *ent;
#endif
@@ -198,6 +209,10 @@ void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
void amdgpu_ring_commit(struct amdgpu_ring *ring);
void amdgpu_ring_undo(struct amdgpu_ring *ring);
+void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
+ enum amd_sched_priority priority);
+void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
+ enum amd_sched_priority priority);
int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
unsigned ring_size, struct amdgpu_irq_src *irq_src,
unsigned irq_type);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index e2e13fa..19d5e97 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -169,4 +169,11 @@ void amd_sched_job_recovery(struct amd_gpu_scheduler *sched);
bool amd_sched_dependency_optimized(struct dma_fence* fence,
struct amd_sched_entity *entity);
void amd_sched_job_kickout(struct amd_sched_job *s_job);
+
+static inline enum amd_sched_priority
+amd_sched_get_job_priority(struct amd_sched_job *job)
+{
+ return (job->s_entity->rq - job->sched->sched_rq);
+}
+
#endif
--
2.7.4
|
