Re: [PATCH] Replace completions with semaphores

[Ingo Molnar <mingo@elte.hu>]

* Matthew Wilcox <matthew@wil.cx> wrote:

> On Tue, Apr 15, 2008 at 07:05:56PM +0200, Ingo Molnar wrote:
>
> > i very much agree with the "get rid of semaphores" argument - the 
> > reason why i initially supported the "move to generic semaphores" 
> > step was because i saw it basically as the precursor to full 
> > removal: it is the removal of semaphores from all architectures - 
> > with a small generic compatibility wrapper to handle the remaining 
> > few uses of semaphores.
> 
> Hm.  I thought you initially supported it because it deleted so much 
> code. [...]

... sorry, but i always thought of semaphores to be removed completely.

> [...]  I don't want to go and add down_killable() to each architecture 
> again, and you were pretty enthusiastic about adding down_killable().

... the killable sleeps should and are already propagated everywhere - i 
never thought of them as a semaphore-only feature.

killable sleeps are probably the next best thing to true 
interruptability.

btw., has anyone thought about killable sync/fsync syscalls - would that 
surprise too many programs?

> > i got thoroughly surprised by the "increase the scope of semaphores" 
> > angle to the patchset though, and in hindsight i'd rather see 
> > neither of those generalizations and see semaphores die a slow but 
> > sure natural death than to see their prolongation :-/
> 
> I'm fully in favour of reducing the number of semaphore users, and 
> eventually eliminating them.  Arjan and I discussed a way to do that 
> just now ... I'll write some code, see how it looks.

cool!

	Ingo