On Wed, 2008-04-09 at 20:22 +0400, Oleg Nesterov wrote:
> On 04/09, David Woodhouse wrote:
> > One of the supposed advantages of TIF_RESTORE_SIGMASK in the first
> > place, iirc, was that it allowed us to return a result code other than
> > -EINTR as _well_ as restoring the signal mask.
> Agreed, good point. ERESTART_ is not that flexible.
> Somehow I assumed we will never need something "special" here, this is
> not very clever.
Well, it's not clear that we _will_ need it to be so special. You could
perhaps argue that it's overengineering. It's just that at the time I
did it, I _thought_ I'd need it for ppoll().
It's only in later optimisations that I realised we only ever really
needed to use TIF_RESTORE_SIGMASK in the case where ppoll() or pselect()