Re: slub-i386-support.patch

[Hugh Dickins <hugh@veritas.com>]

On Fri, 11 May 2007, Andrew Morton wrote:
> On Fri, 11 May 2007 10:29:30 +0200 Andi Kleen <ak@suse.de> wrote:
> 
> > > I'm guessing (haven't rechecked source) that the cpu_idle() call comes
> > > about because the top level pgd of a process gets freed very late in
> > > its exit, and after a great flurry of processes have just exited,
> > > perhaps there was nothing to free up the accumulation.  Though
> > > it still strikes me as an odd place to do it.
> > 
> > I always found it odd and probably the wrong place too.
> 
> so... what's the bottom line here, guys?  Should we change that patch?

The bottom line... I can see why you're asking for that ;)

I'd say delete the change to arch/i386/kernel/smp.c - contrary to
what Christoph says, no other arch buries a check_pgt_cache() call
in flush_tlb_mm(), that just seems to be a thinko: i386 has the usual
call to it from tlb_finish_mmu() - _that_ is the one which he and
David were talking about.

I'm just worried that there might somewhere be a call to flush_tlb_mm()
which would now be surprised to be freeing pages: almost certainly not,
but why raise that concern?  It's just not flush_tlb_mm()'s business.

The cpu_idle() call should stay for now: we're agreed that it's odd,
but there's plenty of precedent for it, and it's easier to imagine it
serves a real purpose, and shouldn't be removed without replacement.

Bill raised a real concern about unnecessary PAE pgd memory usage,
but let's get the patch into -rc1 to enable the wider SLUB testing,
before coming back to fix that up.  His micro-optimizations can wait.

IMHO
Hugh