Tearing down DMA transfer setup after DMA client has finished

Måns Rullgård mans at mansr.com
Thu Dec 8 08:36:28 PST 2016


Vinod Koul <vinod.koul at intel.com> writes:

> On Thu, Dec 08, 2016 at 12:20:30PM +0000, Måns Rullgård wrote:
>> >
>> > I'm far from claiming that drivers/tty/serial/sh-sci.c is perfect, but
>> > it does request DMA channels at open time, not at probe time.
>> 
>> In the part quoted above, I said most drivers request dma channels in
>> their probe or open functions.  For the purposes of this discussion,
>> that distinction is irrelevant.  In either case, the channel is held
>> indefinitely.  
>
> And the answer was it is wrong and not _all_ do that!!

Show me one that doesn't.

>> If this wasn't the correct way to use the dmaengine,
>> there would be no need for the virt-dma helpers which are specifically
>> designed for cases the one currently at hand.
>
> That is incorrect.
>
> virt-dma helps to have multiple request from various clients. For many
> controllers which implement a SW mux, they can transfer data for one client
> and then next transfer can be for some other one.
>
> This allows better utilization of dma channels and helps in case where we
> have fewer dma channels than users.

Which is *exactly* the situation we have here.

I have no idea what you're arguing for or against any more.  Perhaps
you're just arguing.

>> The only problem we have is that nobody envisioned hardware where the
>> dma engine indicates completion slightly too soon.  I suspect there's a
>> fifo or such somewhere, and the interrupt is triggered when the last
>> byte has been placed in the fifo rather than when it has been removed
>> which would have been more correct.
>
> That is pretty common hardware optimization but usually hardware shows up
> with flush commands to let in flight transactions be completed.

Well, this hardware seems to lack that particular feature.  Pretending
otherwise isn't helping.

-- 
Måns Rullgård



More information about the linux-arm-kernel mailing list