Re: sashiko.dev review (Re: [PATCH v4 13/17] dmaengine: sh: rz-dmac: Add cyclic DMA support)

[Claudiu Beznea]

On 4/20/26 15:51, Claudiu Beznea wrote:
>  > Also, because the lock is dropped when exiting the scoped_guard before
>  > vchan_tx_prep() is called, a concurrent terminate_all() could potentially
>  > clear the flag before the descriptor is queued.
>  > Could the driver rely on the desc->type field instead of a global state
>  > flag to identify cyclic transfers?
>
> I'm going to explore that but I think this will complicates the code further.

As the channel can only accept a single cyclic transfer enqueued and no other 
types of transfers (since the HW cannot handle mixed cyclic and non-cyclic 
transfers), the exact scenario described above can only happen if the user of 
this channel terminates the cyclic transfer before it even prepared it. I don't 
think that is a valid case.

Thank you,
Claudiu