From: Stanimir Varbanov <stanimir.varbanov@linaro.org>
To: Alexandre Courbot <acourbot@chromium.org>
Cc: linux-media@vger.kernel.org, linux-arm-msm@vger.kernel.org,
linux-kernel@vger.kernel.org,
Vikash Garodia <vgarodia@codeaurora.org>,
Mansur Alisha Shaik <mansur@codeaurora.org>,
Dikshita Agarwal <dikshita@codeaurora.org>
Subject: Re: [PATCH 1/3] venus: venc: Init the session only once in queue_setup
Date: Fri, 27 Nov 2020 01:50:58 +0200 [thread overview]
Message-ID: <dbd68079-6736-28bf-b9ab-dda51a07aa71@linaro.org> (raw)
In-Reply-To: <CAPBb6MUU2tKUarrdPennSQ0NVnOUp+jwS3DN3Qye-wiq-aT5_Q@mail.gmail.com>
On 11/25/20 5:13 AM, Alexandre Courbot wrote:
> Hi Stan,
>
> On Fri, Nov 20, 2020 at 9:12 AM Stanimir Varbanov
> <stanimir.varbanov@linaro.org> wrote:
>>
>> Init the hfi session only once in queue_setup and also cover that
>> with inst->lock.
>>
>> Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
>> ---
>> drivers/media/platform/qcom/venus/venc.c | 98 ++++++++++++++++++------
>> 1 file changed, 73 insertions(+), 25 deletions(-)
>>
>> diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
>> index 4ecf78e30b59..3a2e449663d8 100644
>> --- a/drivers/media/platform/qcom/venus/venc.c
>> +++ b/drivers/media/platform/qcom/venus/venc.c
>> @@ -725,8 +725,10 @@ static int venc_init_session(struct venus_inst *inst)
>> int ret;
>>
>> ret = hfi_session_init(inst, inst->fmt_cap->pixfmt);
>> - if (ret)
>> - return ret;
>> + if (ret == -EINVAL)
>> + return 0;
>
> Why is it safe to ignore EINVAL here?
The confusion comes from hfi_session_init() return values. Presently
hfi_session_init will return EINVAL when the session is already init.
Maybe EINVAL is not fitting well with the expected behavior of the
function. I thought about EALREADY, EBUSY but it doesn't fit well to me too.
>
>> + else if (ret)
>> + goto deinit;
>>
>> ret = venus_helper_set_input_resolution(inst, inst->width,
>> inst->height);
>> @@ -762,17 +764,13 @@ static int venc_out_num_buffers(struct venus_inst *inst, unsigned int *num)
>> struct hfi_buffer_requirements bufreq;
>> int ret;
>>
>> - ret = venc_init_session(inst);
>> + ret = venus_helper_get_bufreq(inst, HFI_BUFFER_INPUT, &bufreq);
>> if (ret)
>> return ret;
>>
>> - ret = venus_helper_get_bufreq(inst, HFI_BUFFER_INPUT, &bufreq);
>> -
>> *num = bufreq.count_actual;
>>
>> - hfi_session_deinit(inst);
>> -
>> - return ret;
>> + return 0;
>> }
>>
>> static int venc_queue_setup(struct vb2_queue *q,
>> @@ -781,7 +779,7 @@ static int venc_queue_setup(struct vb2_queue *q,
>> {
>> struct venus_inst *inst = vb2_get_drv_priv(q);
>> unsigned int num, min = 4;
>> - int ret = 0;
>> + int ret;
>>
>> if (*num_planes) {
>> if (q->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE &&
>> @@ -803,6 +801,17 @@ static int venc_queue_setup(struct vb2_queue *q,
>> return 0;
>> }
>>
>> + ret = mutex_lock_interruptible(&inst->lock);
I'll keep original mutex_lock here in next version.
>> + if (ret)
>> + return ret;
>> +
>> + ret = venc_init_session(inst);
>> +
>> + mutex_unlock(&inst->lock);
>> +
>> + if (ret)
>> + return ret;
>> +
>> switch (q->type) {
>> case V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE:
>> *num_planes = inst->fmt_out->num_planes;
>> @@ -838,6 +847,54 @@ static int venc_queue_setup(struct vb2_queue *q,
>> return ret;
>> }
>>
>> +static int venc_buf_init(struct vb2_buffer *vb)
>> +{
>> + struct venus_inst *inst = vb2_get_drv_priv(vb->vb2_queue);
>> +
>> + inst->buf_count++;
>> +
>> + return venus_helper_vb2_buf_init(vb);
>> +}
>> +
>> +static void venc_release_session(struct venus_inst *inst)
>> +{
>> + int ret, abort = 0;
>> +
>> + mutex_lock(&inst->lock);
>> +
>> + ret = hfi_session_deinit(inst);
>> + abort = (ret && ret != -EINVAL) ? 1 : 0;
>
> Here as well, I think a comment is warranted to explain why we can
> ignore EINVAL.
OK, will update that.
>
>> +
>> + if (inst->session_error)
>> + abort = 1;
>> +
>> + if (abort)
>> + hfi_session_abort(inst);
>> +
>> + mutex_unlock(&inst->lock);
>> +
>> + venus_pm_load_scale(inst);
>> + INIT_LIST_HEAD(&inst->registeredbufs);
>> + venus_pm_release_core(inst);
>> +}
>> +
>> +static void venc_buf_cleanup(struct vb2_buffer *vb)
>> +{
>> + struct venus_inst *inst = vb2_get_drv_priv(vb->vb2_queue);
>> + struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
>> + struct venus_buffer *buf = to_venus_buffer(vbuf);
>> +
>> + mutex_lock(&inst->lock);
>> + if (vb->type == V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE)
>> + if (!list_empty(&inst->registeredbufs))
>> + list_del_init(&buf->reg_list);
>> + mutex_unlock(&inst->lock);
>> +
>> + inst->buf_count--;
>> + if (!inst->buf_count)
>> + venc_release_session(inst);
>
> We are calling venc_init_session() during the queue setup but
> venc_release_session() when the last buffer is cleaned up. For
> symmetry, wouldn't it make sense to call venc_init_session() when the
> first buffer is initialized by venc_buf_init()? Otherwise we can
No, the session must be initialized in queue_setup in order to return
the number and sizes of source/destination buffers.
I raised several times the need of symmetrical operation to queue_setup
to cover reqbuf(0) but there is no progress on that. Latest suggestion
was to use .vidioc_reqbufs ioctl op but I fall with some other issues
and at the end I came to this counting buf_init|cleanup solution.
> potentially have a scenario where the queue is set up, but no buffer
> is ever created, leading to the session never being released.
dmabuf import case?
<cut>
--
regards,
Stan
next prev parent reply other threads:[~2020-11-26 23:51 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-20 0:10 [PATCH 0/3] Venus encoder improvements Stanimir Varbanov
2020-11-20 0:10 ` [PATCH 1/3] venus: venc: Init the session only once in queue_setup Stanimir Varbanov
2020-11-21 1:33 ` Fritz Koenig
2020-11-25 3:13 ` Alexandre Courbot
2020-11-26 23:50 ` Stanimir Varbanov [this message]
2020-12-01 12:22 ` dikshita
2020-11-20 0:10 ` [PATCH 2/3] venus: Limit HFI sessions to the maximum supported Stanimir Varbanov
2020-11-21 1:14 ` Fritz Koenig
2020-11-22 14:48 ` Stanimir Varbanov
2020-11-22 21:05 ` Fritz Koenig
2020-11-25 3:46 ` Alexandre Courbot
2020-11-25 13:01 ` Stanimir Varbanov
2020-11-26 6:28 ` Alexandre Courbot
2020-11-26 22:41 ` Stanimir Varbanov
2020-11-27 2:12 ` Alexandre Courbot
2020-11-20 0:10 ` [PATCH 3/3] media: hfi_venus: Request interrupt for sync cmds Stanimir Varbanov
2020-11-21 1:02 ` Fritz Koenig
2020-11-22 14:49 ` Stanimir Varbanov
2020-11-25 8:08 ` Alexandre Courbot
2020-11-25 13:03 ` Stanimir Varbanov
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=dbd68079-6736-28bf-b9ab-dda51a07aa71@linaro.org \
--to=stanimir.varbanov@linaro.org \
--cc=acourbot@chromium.org \
--cc=dikshita@codeaurora.org \
--cc=linux-arm-msm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=mansur@codeaurora.org \
--cc=vgarodia@codeaurora.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;
as well as URLs for NNTP newsgroup(s).