From: "Christian König" <ckoenig.leichtzumerken@gmail.com>
To: "Zhu, Jiadong" <Jiadong.Zhu@amd.com>,
"amd-gfx@lists.freedesktop.org" <amd-gfx@lists.freedesktop.org>,
Alex Deucher <Alexander.Deucher@amd.com>
Subject: Re: [PATCH v2] drm/amdgpu: set start timestamp of fence in the right place
Date: Wed, 10 Jul 2024 11:27:22 +0200 [thread overview]
Message-ID: <3f6f779d-33ff-4cbb-9eaf-035888c200e5@gmail.com> (raw)
In-Reply-To: <DS7PR12MB633361A48187FDC04193B93CF4A42@DS7PR12MB6333.namprd12.prod.outlook.com>
Am 10.07.24 um 09:54 schrieb Zhu, Jiadong:
> [AMD Official Use Only - AMD Internal Distribution Only]
>
>> -----Original Message-----
>> From: Christian König <ckoenig.leichtzumerken@gmail.com>
>> Sent: Wednesday, July 10, 2024 3:17 PM
>> To: Zhu, Jiadong <Jiadong.Zhu@amd.com>; amd-gfx@lists.freedesktop.org
>> Subject: Re: [PATCH v2] drm/amdgpu: set start timestamp of fence in the
>> right place
>>
>> Am 10.07.24 um 02:31 schrieb jiadong.zhu@amd.com:
>>> From: Jiadong Zhu <Jiadong.Zhu@amd.com>
>>>
>>> The job's embedded fence is dma_fence which shall not be conversed to
>>> amdgpu_fence.
>> Good catch.
>>
>>> The start timestamp shall be saved on job for hw_fence.
>> But NAK to that approach. Why do we need the start time here in the first
>> place?
>>
>> Regards,
>> Christian.
>>
> The start timestamp is used for ring mux to check if the fences are unsignaled for a period of time under mcbp scenarios (by calling amdgpu_fence_last_unsignaled_time_us).
I can't find a reason for doing that in the first place. What is the
background of this?
Regards,
Christian.
>
> Thanks,
> Jiadong
>>> v2: optimize get_fence_start_time.
>>> Signed-off-by: Jiadong Zhu <Jiadong.Zhu@amd.com>
>>> ---
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 31
>> ++++++++++++++++++++---
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_job.h | 3 +++
>>> 2 files changed, 31 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>> index 2f24a6aa13bf..72bb007e48c8 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>> @@ -88,6 +88,31 @@ static inline struct amdgpu_fence
>> *to_amdgpu_fence(struct dma_fence *f)
>>> return NULL;
>>> }
>>>
>>> +static inline void set_fence_start_time(struct dma_fence *f, ktime_t
>>> +start_timestamp) {
>>> + if (f->ops == &amdgpu_fence_ops) {
>>> + struct amdgpu_fence *__f = container_of(f, struct
>> amdgpu_fence,
>>> +base);
>>> +
>>> + __f->start_timestamp = start_timestamp;
>>> + } else if (f->ops == &amdgpu_job_fence_ops) {
>>> + struct amdgpu_job *job = container_of(f, struct amdgpu_job,
>>> +hw_fence);
>>> +
>>> + job->start_timestamp = start_timestamp;
>>> + }
>>> +}
>>> +
>>> +static inline ktime_t get_fence_start_time(struct dma_fence *f) {
>>> + if (unlikely(f->ops == &amdgpu_fence_ops)) {
>>> + struct amdgpu_fence *__f = container_of(f, struct
>> amdgpu_fence,
>>> +base);
>>> +
>>> + return __f->start_timestamp;
>>> + }
>>> + struct amdgpu_job *job = container_of(f, struct amdgpu_job,
>>> +hw_fence);
>>> +
>>> + return job->start_timestamp;
>>> +}
>>> +
>>> /**
>>> * amdgpu_fence_write - write a fence value
>>> *
>>> @@ -197,7 +222,7 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring,
>> struct dma_fence **f, struct amd
>>> }
>>> }
>>>
>>> - to_amdgpu_fence(fence)->start_timestamp = ktime_get();
>>> + set_fence_start_time(fence, ktime_get());
>>>
>>> /* This function can't be called concurrently anyway, otherwise
>>> * emitting the fence would mess up the hardware ring buffer.
>>> @@ -428,7 +453,7 @@ u64
>> amdgpu_fence_last_unsignaled_time_us(struct amdgpu_ring *ring)
>>> return 0;
>>>
>>> return ktime_us_delta(ktime_get(),
>>> - to_amdgpu_fence(fence)->start_timestamp);
>>> + get_fence_start_time(fence));
>>> }
>>>
>>> /**
>>> @@ -451,7 +476,7 @@ void
>> amdgpu_fence_update_start_timestamp(struct amdgpu_ring *ring,
>> uint32_t seq,
>>> if (!fence)
>>> return;
>>>
>>> - to_amdgpu_fence(fence)->start_timestamp = timestamp;
>>> + set_fence_start_time(fence, timestamp);
>>> }
>>>
>>> /**
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
>>> index a963a25ddd62..3a73fe11a1ce 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
>>> @@ -73,6 +73,9 @@ struct amdgpu_job {
>>> uint64_t gds_va;
>>> bool init_shadow;
>>>
>>> + /* start timestamp for hw_fence*/
>>> + ktime_t start_timestamp;
>>> +
>>> /* job_run_counter >= 1 means a resubmit job */
>>> uint32_t job_run_counter;
>>>
next prev parent reply other threads:[~2024-07-10 9:27 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-07-10 0:31 [PATCH v2] drm/amdgpu: set start timestamp of fence in the right place jiadong.zhu
2024-07-10 7:17 ` Christian König
2024-07-10 7:42 ` Zhu, Jiadong
2024-07-10 7:54 ` Zhu, Jiadong
2024-07-10 9:27 ` Christian König [this message]
2024-07-10 10:15 ` Zhu, Jiadong
2024-07-10 12:45 ` Christian König
2024-07-11 1:31 ` Zhu, Jiadong
2024-07-11 11:56 ` Christian König
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3f6f779d-33ff-4cbb-9eaf-035888c200e5@gmail.com \
--to=ckoenig.leichtzumerken@gmail.com \
--cc=Alexander.Deucher@amd.com \
--cc=Jiadong.Zhu@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox