lkml.org 
[lkml]   [2006]   [Feb]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] tmpfs: fix mount mpol nodelist parsing
On Tue, 21 Feb 2006, Andrew Morton wrote:
> Hugh Dickins <hugh@veritas.com> wrote:
> >
> > Move the mpol= parsing to shmem_parse_mpol under CONFIG_NUMA, reject
> > all its options as invalid if not NUMA.
>
> That's a bit irritating, really. It means that userspace needs to be
> different for NUMA kernels (or more different, which is still bad). Boot
> into a non-NUMA kernel and whoops, no tmpfs and quite possibly no boot.

Well spotted.

That was a choice that gave me pause between making it and sending the
patch. But in the end I decided we might as well. Repeating what I
wrote to Robin about it...

I did wonder for a while whether I'd been unhelpful to make mpol= fail
when not CONFIG_NUMA - tiresome for someone switching between NUMA and
non-NUMA kernels. But this is an advanced option, not something for
everybody's /etc/fstab; and once I realized that all but the trivial
nodelist "0" would get rejected anyway if not CONFIG_NUMA, decided it
is best to placate the anti-bloaters with that CONFIG_NUMA after all.

> But last time I whined about this Albert had a list of fairly
> reasonable-sounding reasons why filesystems shouldn't silently accept
> not-understood options.
>
> But in this case, we _do_ understand them. We're just not going to do
> anything about them.
>
> I just wonder if we're being as friendly as we possibly can be to admins
> and distro-makers.

I doubt the distro-makers will want to be putting "mpol=" options into
their tmpfs lines in /etc/fstab. I hope the admins of such systems
that need it can cope.

But perhaps I should expand the mention of CONFIG_NUMA in tmpfs.txt,
to explain the issue, and suggest that "mpol=" be used in remounts
rather than automatic mounts on systems where it might be a problem.
I'll dream up some wording later.

> [ Vaguely suprised that tmpfs isn't using match_token()... ]

I did briefly consider that back in the days when I noticed a host of
fs filesystems got converted. But didn't see any point in messing
with what was already working. Haven't looked recently: would it
actually be a useful change to make?

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

\
 
 \ /
  Last update: 2006-02-22 08:23    [from the cache]
©2003-2011 Jasper Spaans