tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/amdgpu: add command submission workflow tracepoint
OGL needs these tracepoints to investigate performance issue.
Change-Id: I5e58187d061253f7d665dfce8e4e163ba91d3e2b
Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
authored by
Chunming Zhou
and committed by
Alex Deucher
7034decf
5f2e816b
+76
-4
+1
-1
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+1
-1
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+2
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+2
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
···
26
26
#include <linux/sched.h>
27
27
#include <drm/drmP.h>
28
28
#include "amdgpu.h"
29
+
#include "amdgpu_trace.h"
29
30
30
31
static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
31
32
{
···
46
45
}
47
46
job = to_amdgpu_job(sched_job);
48
47
mutex_lock(&job->job_lock);
48
+
trace_amdgpu_sched_run_job(job);
49
49
r = amdgpu_ib_schedule(job->adev,
50
50
job->num_ibs,
51
51
job->ibs,
+51
drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
+51
drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
···
48
48
__entry->fences)
49
49
);
50
50
51
+
TRACE_EVENT(amdgpu_cs_ioctl,
52
+
TP_PROTO(struct amdgpu_job *job),
53
+
TP_ARGS(job),
54
+
TP_STRUCT__entry(
55
+
__field(struct amdgpu_device *, adev)
56
+
__field(struct amd_sched_job *, sched_job)
57
+
__field(struct amdgpu_ib *, ib)
58
+
__field(struct fence *, fence)
59
+
__field(char *, ring_name)
60
+
__field(u32, num_ibs)
61
+
),
62
+
63
+
TP_fast_assign(
64
+
__entry->adev = job->adev;
65
+
__entry->sched_job = &job->base;
66
+
__entry->ib = job->ibs;
67
+
__entry->fence = &job->base.s_fence->base;
68
+
__entry->ring_name = job->ibs[0].ring->name;
69
+
__entry->num_ibs = job->num_ibs;
70
+
),
71
+
TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u",
72
+
__entry->adev, __entry->sched_job, __entry->ib,
73
+
__entry->fence, __entry->ring_name, __entry->num_ibs)
74
+
);
75
+
76
+
TRACE_EVENT(amdgpu_sched_run_job,
77
+
TP_PROTO(struct amdgpu_job *job),
78
+
TP_ARGS(job),
79
+
TP_STRUCT__entry(
80
+
__field(struct amdgpu_device *, adev)
81
+
__field(struct amd_sched_job *, sched_job)
82
+
__field(struct amdgpu_ib *, ib)
83
+
__field(struct fence *, fence)
84
+
__field(char *, ring_name)
85
+
__field(u32, num_ibs)
86
+
),
87
+
88
+
TP_fast_assign(
89
+
__entry->adev = job->adev;
90
+
__entry->sched_job = &job->base;
91
+
__entry->ib = job->ibs;
92
+
__entry->fence = &job->base.s_fence->base;
93
+
__entry->ring_name = job->ibs[0].ring->name;
94
+
__entry->num_ibs = job->num_ibs;
95
+
),
96
+
TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u",
97
+
__entry->adev, __entry->sched_job, __entry->ib,
98
+
__entry->fence, __entry->ring_name, __entry->num_ibs)
99
+
);
100
+
101
+
51
102
TRACE_EVENT(amdgpu_vm_grab_id,
52
103
TP_PROTO(unsigned vmid, int ring),
53
104
TP_ARGS(vmid, ring),
+21
-3
drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
+21
-3
drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
···
16
16
TP_ARGS(sched_job),
17
17
TP_STRUCT__entry(
18
18
__field(struct amd_sched_entity *, entity)
19
+
__field(struct amd_sched_job *, sched_job)
20
+
__field(struct fence *, fence)
19
21
__field(const char *, name)
20
22
__field(u32, job_count)
21
23
__field(int, hw_job_count)
···
25
23
26
24
TP_fast_assign(
27
25
__entry->entity = sched_job->s_entity;
26
+
__entry->sched_job = sched_job;
27
+
__entry->fence = &sched_job->s_fence->base;
28
28
__entry->name = sched_job->sched->name;
29
29
__entry->job_count = kfifo_len(
30
30
&sched_job->s_entity->job_queue) / sizeof(sched_job);
31
31
__entry->hw_job_count = atomic_read(
32
32
&sched_job->sched->hw_rq_count);
33
33
),
34
-
TP_printk("entity=%p, ring=%s, job count:%u, hw job count:%d",
35
-
__entry->entity, __entry->name, __entry->job_count,
36
-
__entry->hw_job_count)
34
+
TP_printk("entity=%p, sched job=%p, fence=%p, ring=%s, job count:%u, hw job count:%d",
35
+
__entry->entity, __entry->sched_job, __entry->fence, __entry->name,
36
+
__entry->job_count, __entry->hw_job_count)
37
37
);
38
+
39
+
TRACE_EVENT(amd_sched_process_job,
40
+
TP_PROTO(struct amd_sched_fence *fence),
41
+
TP_ARGS(fence),
42
+
TP_STRUCT__entry(
43
+
__field(struct fence *, fence)
44
+
),
45
+
46
+
TP_fast_assign(
47
+
__entry->fence = &fence->base;
48
+
),
49
+
TP_printk("fence=%p signaled", __entry->fence)
50
+
);
51
+
38
52
#endif
39
53
40
54
/* This part must be outside protection */
+1
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+1
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c