public inbox for linux-media@vger.kernel.org
 help / color / mirror / Atom feed
From: Hans Verkuil <hverkuil@xs4all.nl>
To: Ran Shalit <ranshalit@gmail.com>
Cc: linux-media@vger.kernel.org
Subject: Re: dma start/stop & vb2 APIs
Date: Tue, 19 Jan 2016 08:33:51 +0100	[thread overview]
Message-ID: <569DE6DF.3060508@xs4all.nl> (raw)
In-Reply-To: <CAJ2oMh+BXUnqehqGzaxqQK07+7HiEOpjRf4+GjqDNRbRNb5NKg@mail.gmail.com>

On 01/19/2016 05:45 AM, Ran Shalit wrote:
> On Mon, Jan 18, 2016 at 7:45 PM, Hans Verkuil <hverkuil@xs4all.nl> wrote:
>> On 01/18/2016 05:26 PM, Ran Shalit wrote:
>>> Hello,
>>>
>>> I am trying to understand how to implement dma transfer correctly
>>> using videobuf2 APIs.
>>>
>>> Normally, application will do semthing like this (video test API):
>>>
>>>                 xioctl(fd, VIDIOC_DQBUF, &buf)
>>>                 process_image(buffers[buf.index].start, buf.bytesused);
>>>                 xioctl(fd, VIDIOC_QBUF, &buf)
>>>
>>> Therefore, in the driver below I will assume that:
>>> 1. VIDIOC_DQBUF -  trigger dma to start
>>
>> No. DMA typically starts when VIDIOC_STREAMON is called, although depending
>> on the hardware the start of the DMA may be delayed if insufficient buffers
>> have been queued. When the start_streaming op is called you know that STREAMON
>> has been called and that at least min_buffers_needed buffers have been queued.
>>
>>> 2. interrupt handler in driver - stop dma
>>
>> ??? No, this just passes the buffer that has been filled by the DMA engine
>> to vb2 via vb2_buffer_done. The DMA will continue filling the next queued buffer.
>>
> 
> Hi,
> Just to be sure I got it all right:
> "The DMA will continue filling the next queued buffer" is usually the
> responsibility of the interrupt handler .
> Interrrupt will get the new buffer from list and trigger next dma
> transaction with that buffer.
>  Is that Right ?

Usually that's true. Again, I can't give absolutes because it depends on
the DMA hardware details. There tend to be two main types of DMA:

1) the DMA engine has pointers to the current and next frame: in that case
the irq handler has to update the pointers once the current frame finishes
the DMA.

2) the DMA engine has a list of descriptors describing each frame and that
links each frame to the next one. In that case the list of descriptors is
updated whenever a new buffer is queued. For engines like this the interrupt
function just returns the DMAed buffer and doesn't have to do anything else.

Regards,

	Hans

  reply	other threads:[~2016-01-19  7:33 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-01-18 16:26 dma start/stop & vb2 APIs Ran Shalit
2016-01-18 17:45 ` Hans Verkuil
2016-01-19  4:45   ` Ran Shalit
2016-01-19  7:33     ` Hans Verkuil [this message]
2016-01-30 11:55       ` Ran Shalit
2016-01-30 12:26         ` Hans Verkuil

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=569DE6DF.3060508@xs4all.nl \
    --to=hverkuil@xs4all.nl \
    --cc=linux-media@vger.kernel.org \
    --cc=ranshalit@gmail.com \
    /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