Re: [patch] x86: Fix /dev/mem mmap breakage when PAT is disabled

[Ravikiran G Thirumalai <kiran@scalex86.org>]

On Thu, Oct 30, 2008 at 02:29:49PM -0700, Arjan van de Ven wrote:
>On Thu, 30 Oct 2008 14:21:21 -0700
>Ravikiran G Thirumalai <kiran@scalex86.org> wrote:
>
>> On Thu, Oct 30, 2008 at 01:21:03PM -0700, Arjan van de Ven wrote:
>> >On Wed, 29 Oct 2008 19:02:03 -0700
>> >Ravikiran G Thirumalai <kiran@scalex86.org> wrote:
>> >
>> >well... technically the aliases are bad without PAT as well...
>> >
>> 
>> Hmm! Well, you mean if mtrr is used to change attributes?
>
>you can also set uncached in the pagetables, like via ioremap_uncached
>and other ways.
>
>and /dev/mem may or may not imply cached/uncached, depending on the
>open flags.
>
>On memory... what would that mean?
>

Well, I am not denying that an incorrectly written userspace program could cause
cache aliasing/crash the system  I am not arguing for or against this
argument at all.

However, there is what seems like an unintended change in behavior from
2.6.25 to 2.6.26 when CONFIG_STRICT_DEVMEM is disabled and PAT is disabled,
which is a bug, and this patch fixes it that's all. If this was intentional,
then I don't see the reason for having CONFIG_STRICT_DEVMEM!!