Re: [PATCH v2 04/19] dmaengine: dw-edma: Make DMA link list work as a circular buffer

On Thu, Jul 23, 2026 at 11:43:29AM -0500, Frank Li wrote:
> On Thu, Jul 23, 2026 at 09:07:05AM +0000, [email protected] wrote:
> > Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> > - [High] Ring buffer capacity is permanently leaked on error and termination paths because `ll_end` is not advanced.
> > --
> >
> > --- Patch [4]: [PATCH v2 04/19] dmaengine: dw-edma: Make DMA link list work as a circular buffer ---
> > commit 2d5b6554b58830487f70a13eab5e1ac3de304658
> > Author: Frank Li <[email protected]>
> >
> > dmaengine: dw-edma: Make DMA link list work as a circular buffer
> >
> > This commit updates the driver to use the DMA linked list as a circular
> > ring buffer. It appends entries at ll_head with the current cycle bit and
> > reserves the final entry to link back to the start.
> >
> > > diff --git a/drivers/dma/dw-edma/dw-edma-core.c b/drivers/dma/dw-edma/dw-edma-core.c
> > > --- a/drivers/dma/dw-edma/dw-edma-core.c
> > > +++ b/drivers/dma/dw-edma/dw-edma-core.c
> >
> > [ ... ]
> >
> > > @@ -645,6 +687,8 @@ static void dw_edma_done_interrupt(struct dw_edma_chan *chan)
> > >  							    DMA_TRANS_NOERROR);
> > >  				list_del(&vd->node);
> > >  				vchan_cookie_complete(vd);
> > > +				if (!chan->non_ll)
> > > +					chan->ll_end = chan->ll_head;
> > >  			}
> > >
> > >  			if (chan->request == EDMA_REQ_PAUSE) {
> >
> > [Severity: High]
> > Does this leak ring capacity when descriptors are aborted or terminated?
> >
> > This block successfully reclaims ring capacity on successful completion by
> > advancing chan->ll_end. However, if a client driver queues transfers and
> > then aborts them (for example, via dmaengine_terminate_all() or during an
> > abort interrupt), the descriptors are removed from the list but chan->ll_end
> > does not appear to be updated to reflect the aborted descriptors.
> >
> > Since chan->ll_valid remains true after a termination, subsequent transfers
> > will bypass the dw_edma_core_reset_ll() ring reset logic and continue
> > appending to the un-reclaimed ring.
> 
> Make sense. Need change at these abnormal case.

Indeed. This is a bisectability issue. The final tree handles it in patch 14,
but patches 4 through 13 still have this issue. I'll sort it out so each
intermediate patch remains valid.

Best regards,
Koichiro

> 
> Frank
> >
> > Will this eventually cause dw_edma_core_get_free_num() to permanently return
> > 0 and deadlock the channel once the un-reclaimed slots exhaust the ring?
> >
> > --
> > Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4