Date:   Mon, 28 Mar 2022 14:02:04 +0200
From:   Halil Pasic <pasic@...ux.ibm.com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Toke Høiland-Jørgensen <toke@...e.dk>,
        Robin Murphy <robin.murphy@....com>,
        Christoph Hellwig <hch@....de>,
        Maxime Bizon <mbizon@...ebox.fr>,
        Oleksandr Natalenko <oleksandr@...alenko.name>,
        Marek Szyprowski <m.szyprowski@...sung.com>,
        Kalle Valo <kvalo@...nel.org>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        Olha Cherevyk <olha.cherevyk@...il.com>,
        iommu <iommu@...ts.linux-foundation.org>,
        linux-wireless <linux-wireless@...r.kernel.org>,
        Netdev <netdev@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        stable <stable@...r.kernel.org>,
        Halil Pasic <pasic@...ux.ibm.com>
Subject: Re: [REGRESSION] Recent swiotlb DMA_FROM_DEVICE fixes break
 ath9k-based AP

On Sun, 27 Mar 2022 17:30:01 -0700
Linus Torvalds <torvalds@...ux-foundation.org> wrote:

> On Sun, Mar 27, 2022 at 4:52 PM Halil Pasic <pasic@...ux.ibm.com> wrote:
> >
> > I have no intention of pursuing this.  When fixing the information leak,
> > I happened to realize, that a somewhat similar situation can emerge when
> > mappings are reused. It seemed like an easy fix, so I asked the swiotlb
> > maintainers, and they agreed. It ain't my field of expertise, and the
> > drivers I'm interested in don't need this functionality.  
> 
> Ok.
> 
> That said, I think you are putting yourself down when you said in an
> earlier email that you aren't veryt knowledgeable in this area.
> 
> I think the fact that you *did* think of this other similar situation
> is actually very interesting, and it's something people probably
> _haven't_ been thinking about.

Thank you!

> 
> So I think your first commit fixes the straightforward and common case
> where you do that "map / partial dma / unmap" case.
> 
> And that straightforward case is probably all that the disk IO case
> ever really triggers, which is presumably why those "drivers I'm
> interested in don't need this functionality" don't need anything else?
> 

I agree.

> And yes, your second commit didn't work, but hey, whatever. The whole
> "multiple operations on the same double buffering allocation"
> situation is something I don't think people have necessarily thought
> about enough.
> 
> And by that I don't mean you. I mean very much the whole history of
> our dma mapping code.
> 

I agree. We are in the process of catching up! :) My idea was to aid
a process, as a relatively naive pair of eyes: somebody didn't read any
data sheets describing non-cache-coherent DMA, and never programmed
a DMA. It is a fairly common problem, that for the very knowledgeable
certain things seem obvious, self-explanatory or trivial, but for the
less knowledgeable the are not. And knowledge can create bias.

> I then get opinionated and probably too forceful, but please don't
> take it as being about you - it's about just my frustration with that
> code - and if it comes off too negative then please accept my
> apologies.

I have to admit, I did feel a little uncomfortable, and I did look for
an exit strategy. I do believe, that people in your position do have to
occasionally get forceful, and even abrasive to maintain efficiency. I
try to not ignore the social aspect of things, but I do get carried away
occasionally.

Your last especially paragraph is very encouraging and welcome. Thank
you!

Regards,
Halil

[..]

Powered by blists - more mailing lists