Re: [PATCH] mm: disable preemption in apply_to_pte_range

[Peter Zijlstra <peterz@infradead.org>]

On Fri, 2009-02-13 at 09:24 -0800, Jeremy Fitzhardinge wrote:
> Peter Zijlstra wrote:
> >> The specific rules are that 
> >> arch_enter_lazy_mmu_mode()/arch_leave_lazy_mmu_mode() require you to be 
> >> holding the appropriate pte locks for the ptes you're updating, so 
> >> preemption is naturally disabled in that case.
> >>     
> >
> > Right, except on -rt where the pte lock is a mutex.
> >   
> 
> Hm, that's interesting.  The requirement isn't really "no preemption", 
> its "must not migrate to another cpu".  Is there a better way to express 
> that?

Not really, in the past something like migrate_disable() has been
proposed, however that's problematic in that it can generate latencies
that are _very_ hard to track down, so we've always resisted that and
found other ways.

> >> This all goes a bit strange with init_mm's non-requirement for taking 
> >> pte locks.  The caller has to arrange for some kind of serialization on 
> >> updating the range in question, and that could be a mutex.  Explicitly 
> >> disabling preemption in enter_lazy_mmu_mode would make sense for this 
> >> case, but it would be redundant for the common case of batched updates 
> >> to usermode ptes.
> >>     
> >
> > I really utterly hate how you just plonk preempt_disable() in there
> > unconditionally and without very clear comments on how and why.
> >   
> 
> Well, there's the commit comment.  They're important, right?  That's why 
> we spend time writing good commit comments?  So they get read?  ;)

Andrew taught me that indeed, but still when looking at the code its
good to have some text there explaining things too.