[PATCH] DMA: extend documentation to provide more API details

[dan.j.williams@intel.com (Dan Williams)]

On Thu, Oct 17, 2013 at 7:55 AM, Stephen Warren <swarren@wwwdotorg.org> wrote:
> On 10/16/2013 11:16 PM, Vinod Koul wrote:
>> On Wed, Oct 16, 2013 at 01:33:44PM -0600, Stephen Warren wrote:
>>> On 10/08/2013 07:34 PM, Dan Williams wrote:
>>>> On Mon, Oct 7, 2013 at 3:39 AM, Vinod Koul <vinod.koul@intel.com> wrote:
>>>>> On Mon, Oct 07, 2013 at 12:17:28PM +0100, Russell King - ARM Linux wrote:
>>> ...
>>>>>> What may be better is to change the wording here: not DMA_SUCCESS but
>>>>>> DMA_COMPLETED.  That doesn't imply that it has been successful, merely
>>>>>> that the DMA engine has finished with the transaction.
>>>>>
>>>>> Agreed that its not indication of success but of DMA completetion. I have seen
>>>>> cases where slave perhiphral got stuck while sending last FIFO but since DMA
>>>>> finished transferiing to FIFO it says complete.
>>>
>>> In that case, the DMA *has* completed. DMA is the transfer into the
>>> FIFO, not the handling of the FIFO content by the peripheral.
>>>
>>>>> Dan do you agree?
>>>>
>>>> Yes, it's an indication of completion, not necessarily success.
>>>
>>> Surely by definition, a DMA can't *complete* without being successful.
>>> If the DMA failed, then it didn't complete, but rather must have been
>>> aborted or error'd out, without completing the whole transfer.
>>
>> DMA means transferring of data, and that is what DMA compeletion would mean.
>> Once the data has left the DMA FIFO, we don't know if it filled up the memory
>> propery or get stuck in periphral FIFO, the dmaengine would have no knowledge or
>> control of it. So it can't claim the transfer was success from a data point of
>> view, hence the rename now!
>
> Well, I obviously don't agreee, but I guess I won't bother pursuing this.
>

What name would you prefer,
DMA_THAT_COOKIE_IS_NOT_PART_OF_MY_ACTIVE_SET_SO_SOMETHING_MUST_HAVE_HAPPENED
:-)?

But I think you are pointing out that we don't have any rules around
what to expect when a channel encounters an error.  Does the chain
stop advancing and cookie stays valid until the client can clean up,
or does the engine mark an error and advance its last_completed_cookie
past the error?

--
Dan