Re: [PATCH] dmaengine: dw: fix cyclic transfers
From: Måns Rullgård
Date: Mon Jan 11 2016 - 07:20:06 EST
Viresh Kumar <viresh.kumar@xxxxxxxxxx> writes:
> On 10-01-16, 20:54, Mans Rullgard wrote:
>> Commit 61e183f83069 ("dmaengine/dw_dmac: Reconfigure interrupt and
>> chan_cfg register on resume") moved some channel initialisation to
>> a new function which must be called before starting a transfer.
>>
>> This adds the necessary dwc_initialize() call to dw_dma_cyclic_start()
>> which was missed in the original commit.
>>
>> Fixes: 61e183f83069 ("dmaengine/dw_dmac: Reconfigure interrupt and chan_cfg register on resume")
>> Signed-off-by: Mans Rullgard <mans@xxxxxxxxx>
>> ---
>> Andy, do you want to collect this with your other patches? It's trivial
>> enough that it can go in separately without conflicts too.
>> ---
>> drivers/dma/dw/core.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/dma/dw/core.c b/drivers/dma/dw/core.c
>> index 7067b6ddc1db..2917f75b67b4 100644
>> --- a/drivers/dma/dw/core.c
>> +++ b/drivers/dma/dw/core.c
>> @@ -1265,6 +1265,8 @@ int dw_dma_cyclic_start(struct dma_chan *chan)
>> return -EBUSY;
>> }
>>
>> + dwc_initialize(dwc);
>> +
>> dma_writel(dw, CLEAR.ERROR, dwc->mask);
>> dma_writel(dw, CLEAR.XFER, dwc->mask);
>
> Ahh, that's a very very old bug :(
It's not the only one, I'm afraid.
> Okay, things got cleaned up by putting the dwc_initialize() call
> within dwc_dostart() and so it worked for all other DMA APIs.
>
> What about calling dwc_dostart() from within dw_dma_cyclic_start()?
> That will simplify it and remove code duplication issue as well.
Good idea. New patch coming up.
--
Måns Rullgård