Re: [PATCH 2/3] dmaengine: rcar-dmac: Add missing dma_descriptor_unmap()

[Vinod Koul <vkoul@kernel.org>]

On 29-09-26, 13:13, Koichiro Den wrote:
> On Sun, Sep 27, 2026 at 03:11:26PM +0200, Wolfram Sang wrote:
> > 
> > > I share this thought, and I guess, Laurent expressed this, too. This is
> > > a separate task to tackle, though. I still think your original patch
> > > here makes the situation better by making sure the cache gets
> > > invalidated and should be applied. Or am I missing something?
> > 
> > Okay, I got now that Sashiko's comment was related to cookie-completion
> > and not the callback. For me, its comment also makes sense because it is
> > basically the same argument which the commits I quoted to Frank used:
> > ensure cache completion. This is not only good before the callback but
> > also before cookie completion. Or am I missing something?
> > 
> > This is easy to fix. If we can agree on switching unmapping before
> > cookie completion, I can update this patch and then send a series to fix
> > other drivers, too.
> 
> I agree with changing {cookie -> unmap}** to {unmap -> cookie} in this patch.
> It looks like an improvement, I don't see any obvious downside to doing so.
> 
> Many drivers do {cookie -> unmap}, so it may be worth making the same change
> there too. I haven't found a specific reason for that order in the history I
> checked, but I'd also like to hear Frank's thoughts.

I cna chime in. I think Shashiko is correct here. We should ideally
unmap first and then call complete. This would ensure anyone seeing the
completion would get the right data buffer and chances of stale data are
eliminated.

So Wolfram, can you please reverse the order.
Also, lets document this.

Thanks
-- 
~Vinod