Re: [git head] X86_PAT & mprotect

[Dave Airlie]

On Sat, May 10, 2008 at 6:05 AM, Venki Pallipadi
<venkatesh.pallipadi@xxxxxxxxx> wrote:
> On Fri, May 09, 2008 at 12:08:18PM +0200, Ingo Molnar wrote:
>  >
>  > * Venki Pallipadi <venkatesh.pallipadi@xxxxxxxxx> wrote:
>  >
>  > > > I've tried doing it slightly differently below, don't know whether
>  > > > you'll consider it an improvement or not.
>  > >
>  > > Hugh: Thanks for looking into this. Yes. I like your modified patch.
>  > > Simpler and smaller.
>  >
>  > i have stuck your original patch into testing and nothing blew up so
>  > far. Due to the mm/ bits this is not for the scope of x86.git, but
>  > obviously it all looks good and is .26-worthy to me:
>  >
>  >  Acked-by: Ingo Molnar <mingo@xxxxxxx>
>  >  Tested-by: Ingo Molnar <mingo@xxxxxxx>
>  >
>  > Venki, could you please send a full patch against -git that has
>  > everything from Hugh included, with an updated changelog, for
>  > Linus/Andrew to ack/apply?
>  >
>
>  Ingo,
>
>  Split up the patch into two parts as the pci part was unrelated to mprotect
>  problem in a sense.
>
>  Here is the first patch.
>
>  Thanks,
>  Venki
>
>
>  Some versions of X used the mprotect workaround to change caching type from
>
> UC to WB, so that it can then use mtrr to program WC for that region [1].
>  Change the mmap of pci space through /sys or /proc interfaces from UC to
>  UC_MINUS. With this change, X will not need to use mprotect
>  workaround to get WC type.
>  Also the bug with mprotect which lets caller to change PAT bits is fixed in
>  the follow on patch. So, this X workaround will stop working as well.
>

Wow this kinda puts X in a nasty position, we have 2.6.25 and previous kernels
where we use the original /sys interfaces and nasty hack to
workaround, but on 2.6.26 we magically need to
switch to the /sys _uc interfaces or the users X will slow down.

Granted I think only F9 is shipping libpciaccess so far, but now we
need to fix it up and make sure a new one exists before
2.6.26 hits users. Build it yourself users are going to be noticing
the slowdown I suspect.

Dave.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/