linux-kernel - Re: use generic DMA mapping code in powerpc V4

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 11 Dec 2018 07:49:21 +1100
From:   Benjamin Herrenschmidt <benh@...nel.crashing.org>
To:     Christoph Hellwig <hch@....de>,
        Rui Salvaterra <rsalvaterra@...il.com>
Cc:     linux-mm@...ck.org, linuxppc-dev@...ts.ozlabs.org,
        linux-kernel@...r.kernel.org
Subject: Re: use generic DMA mapping code in powerpc V4

On Mon, 2018-12-10 at 20:33 +0100, Christoph Hellwig wrote:
> On Mon, Dec 10, 2018 at 05:04:46PM +0000, Rui Salvaterra wrote:
> > Hi, Christoph and Ben,
> > 
> > It just came to my mind (and this is most likely a stupid question,
> > but still)… Is there any possibility of these changes having an
> > (positive) effect on the long-standing problem of Power Mac machines
> > with AGP graphics cards (which have to be limited to PCI transfers,
> > otherwise they'll hang, due to coherence issues)? If so, I have a G4
> > machine where I'd gladly test them.
> 
> These patches themselves are not going to affect that directly.
> But IFF the problem really is that the AGP needs to be treated as not
> cache coherent (I have no idea if that is true) the generic direct
> mapping code has full support for a per-device coherent flag, so
> support for a non-coherent AGP slot could be implemented relatively
> simply.

AGP is a gigantic nightmare :-) It's not just cache coherency issues
(some implementations are coherent, some aren't, Apple's is ... weird).

Apple has all sort of bugs, and Darwin source code only sheds light on
some of them. Some implementation can only read, not write I think, for
example. There are issues with transfers crossing some boundaries I
beleive, but it's all unclear.

Apple makes this work with a combination of hacks in the AGP "driver"
and the closed source GPU driver, which we don't see.

I have given up trying to make that stuff work reliably a decade ago :)

Cheers,
Ben.