Re: [RFC][PATCH] HWPOISON: remove the unsafe __set_page_locked()

[Nick Piggin]

On Sat, Sep 26, 2009 at 05:05:55PM +0200, Andi Kleen wrote:
> > However we may well end up to accept the fact that "we just cannot do
> > hwpoison 100% correct", and settle with a simple and 99% correct code.
> 
> I would prefer to avoid any oopses, but if they are unlikely enough
> and too hard to fix that's bearable. The race window here is certainly rather 
> small. 

Well, several places non-atomically modify page flags, including
within preempt-enabled regions... It's nasty to introduce these
oopses in the hwposion code! I'm ashamed I didn't pick up on this
problem seeing as I introduced several of them.

 
> On the other hand if you cannot detect a difference in benchmarks I see
> no reason not to add the additional steps, as long as the code isn't
> complicated or ugly. These changes are neither.

The patch to add atomics back into the fastpaths? I don't think that's
acceptable at all. A config option doesn't go far enough either because
distros will have to turn it on.

--
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/