Re: [Linux-fbdev-devel] Re: [ACPI] inappropriate use of in_atomic()

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Fri, 2005-03-11 at 16:13 -0800, Andrew Morton wrote:
> (where'd my cc go?)
> 
> Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
> >
> > On Fri, 2005-03-11 at 01:46 -0800, Andrew Morton wrote:
> > > Jan Kasprzak <kas@fi.muni.cz> wrote:
> > > >
> > > > This may be the cause of 
> > > > 
> > > >  http://bugme.osdl.org/show_bug.cgi?id=4150
> > > 
> > > Looks that way, yes.
> > 
> > Note that it would be interesting to fix that (I mean the reliability of
> > is_atomic() or an alternative). I agree it's quite bad to rely on that
> > in practice, but there are a few corner cases where it's useful (like
> > oops handling in fbdev's etc...)
> > 
> 
> That would require that we increment current->something on every
> spin/read/write_lock and decrement it in unlock, even with !CONFIG_PREEMPT.
> 
> iirc, Anton added an option to do that to the ppc64 build, decoupled from
> CONFIG_PREEMPT (which ppc64 doesn't support).

ppc64 _does_ support PREEMPT nowadays :)

> But it's an appreciable amount of overhead.
-- 
Benjamin Herrenschmidt <benh@kernel.crashing.org>