[PATCH 06/13] DMAENGINE: driver for the ARM PL080/PL081 PrimeCells

[Dan Williams]

On Wed, Dec 22, 2010 at 4:10 PM, Russell King - ARM Linux
<linux at arm.linux.org.uk> wrote:
> On Wed, Dec 22, 2010 at 03:45:39PM -0800, Dan Williams wrote:
>> 3.6 Constraints:
>> 1/ Calls to async_<operation> are not permitted in IRQ context.  Other
>>    contexts are permitted provided constraint #2 is not violated.
>
> BTW, this is misleading.  Have the functions been renamed dma_async_xxx(),
> eg dma_async_memcpy_buf_to_buf etc, or are you referring just to:
>
>        async_dmaengine_get
>        async_dmaengine_put
>        async_dma_find_channel
>        async_dma_find_channel
>        async_tx_ack
>        async_tx_clear_ack
>        async_tx_test_ack
>
> Beware of just renaming it to dma_async_<operation> as there's other
> functions in that namespace which may not be appropriate.
>
> Eg, is it really illegal to issue call dma_async_issue_pending() from
> IRQ context?  That'd make it exceedingly difficult to use the DMA
> engine with the slave API in a lot of device drivers.

This is generic offload (async_{memcpy|xor|etc...}) versus the slave
usage confusion .  In the async_<operation> case the client (md/raid5)
has no idea if a dmaengine is offloading the operation behind the
scenes and should not make any assumptions about submission context
beyond what is allowed in the document.  In the slave case the client
driver knows that it is talking to a specific dma driver.  The
contract between the client driver and dma driver is undocumented.
The slave usage model really is a "I want to use dmaengine to find my
dma device driver / manage channels, and I want a common prep / submit
mechanism, but otherwise stay out of my way".  Drivers that do not
want to meet the constraints expected by the opportunistic offload
clients should do what ste_dma40 does and add "depends on !(NET_DMA ||
ASYNC_TX_DMA)"

--
Dan