Re: [PATCH] dmaengine: Add API documentation for slave dma usage

[Per Forlin]

On 25 May 2011 11:34, Koul, Vinod <vinod.koul@xxxxxxxxx> wrote:
> On Wed, 2011-05-25 at 11:47 +0200, Per Forlin wrote:
>> On 25 May 2011 10:26, Koul, Vinod <vinod.koul@xxxxxxxxx> wrote:
>> > I am okay with making this same for the slave devices as well.
>> >
>> > That would also require to change few drivers which start the txn in
>> > submit()
>> Yes,
>> The trivial part is to move enable from submit to issue_pending. I had
>> a quick look and it looks like the following drivers need to be
>> changed.
>> mpc512x_dma.c
>> imx-dma.c
>> imx-sdma.c
>> mxs-dma.c
>>
>> The other part is more complex. To make submit only add descs to a
>> queue that will be pushed down to the DMA only after issue_pending. It
>> is not as trivial to identify which of the drivers support this or
>> not.
>> I still think it make sense to fix the documentation first and then
>> fix the drivers. Keep a list inside the dmaengine.txt of drivers that
>> need to be fixed in order to comply with the API. Over time drivers
>> will be fixed and removed from the list. When we have agreed upon the
>> API (we may not be there yet) I am willingly to make a draft of such a
>> list.
> Yes this sounds right, I will send updated version and now drivers would
> need to call issue_pending.
>
> I know dw_dmac, intel_mid_dma and ste_dma40 would need changes as well.
>
I think ste_dma40 treats issue_pending and submit correctly but it may
be other things not aligned. I need to double check this since it was
a long time since I did any work in this driver. I can update
ste_dma40 where it need changes.

> Was thinking to add TODO in drivers/dma with these and any other changes
> we feel are required moving forward.
> We need dw_dmac to be moved to new slave interface and remove old slave
> pointer mechanism when all are moved.
>
I agree, TODO is a better place for it.

I would like to ask a question on usage of DMA_CTRL_ACK while we're at it.
In ste_dma40 I added support to reuse descriptors if DMA_CTRL_ACK is
clear. The purpose of DMA_CTRL_ACK is to set up dependency but is it
also a valid to reuse it?
I mean submit the same descriptor again and again without having to
call prepare_slave_sg.
Example:
desc = prepare_slave_sg()
submit(desc)
issue_pending(desc)

wait_for_callback()

submit(desc)
issue_pending(desc)

/Per
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/