Re: [PATCH] dmaengine: omap-dma: Do not call omap_dma_callback() from tx_status()

[Peter Ujfalusi]

On 2016-02-26 14:45, Peter Ujfalusi wrote:
> On 2016-02-26 13:25, Russell King - ARM Linux wrote:
>>>> Your original commit adding the original hack that you're now removing
>>>> above says that this is to support polled operation: I'm not aware of
>>>> DMA engine supporting such a mode.  DMA_PREP_INTERRUPT is a mechanism
>>>> where requests can be queued without an interrupt to allow batching.
>>>
>>> Also it is used to suppress DMA interrupts during audio playback for example.
>>> In this case we will run w/o interrupts and the position is polled.
>>>
>>>> See the raid5/async_tx code, which queues a set of operations without
>>>> DMA_PREP_INTERRUPT, with the final operation with DMA_PREP_INTERRUPT
>>>> set.
>>>
>>> We only allow the interrupts to be disabled in cyclic or memcpy mode. With
>>> slave_sg we have interrupts as it is needed to move to the next SG.
>>>
>>>> As the driver is reliant on interrupts to move to the next transfer,
>>>> the patch which causes DMA_PREP_INTERRUPT to influence whether
>>>> interrupts are sent is actually buggy, and will prevent several
>>>> queued DMA operations to fail.
>>>
>>> Yes, the omap-dma only allows the interrupts to be actually disabled when it
>>> is save to do so. slave_sg can not work w/o interrupts so there we don't
>>> disable them.
>>
>> I get the impression that you haven't taken in what I've said, because
>> each fragment of your reply is just repeating what the previous fragment
>> said.
> 
> Sorry about that. I got it.
> 
>> Let me state that I don't believe you need any hacks here, and this patch
>> is not necessary: the final operation in a set of chained memcpy()s should
>> have DMA_PREP_INTERRUPT set.
> 
> calling the omap_dma_callback() from the tx_status() was a bad idea, this is
> why I'm fixing it.

The reason that we need to be able to determine if the channel is done with
the copy by polling the tx_status is: we need to read and write tiler
registers with DMA on dra7 as per one ERRATA document. This read/write can be
executed in interrupt context so we can not rely on DMA callback to be
notified about the completion of the read/write.

> As for the DMA_PREP_INTERRUPT not enabling interrupt was introduced by:
> 4ce98c0a20bef (dmaengine: omap-dma: Add support for memcpy)
> 
> While we do not have users submitting multiple descriptors I agree that this
> is a possibility and the driver should not fail in such a case.
> I can send a followup patch to fix the omap_dma_prep_dma_memcpy()

With the updated patch based on your comments I can fix the polling in the
tx_status callback and with a separate patch I can address your concern
regarding to multiple transfers queued with only the last one having
DMA_PREP_INTERRUPT (and callback provided).

-- 
Péter