Re: + uml-sigwinch-handling-cleanup.patch added to -mm tree

[Blaisorblade <blaisorblade@yahoo.it>]

On Thursday 05 January 2006 23:27, Jeff Dike wrote:
> On Thu, Jan 05, 2006 at 08:54:37PM +0100, Blaisorblade wrote:
> > Meanwhile, the whole content of the new free_winch(), including some
> > syscalls on the host, and various other stuff, is brought back under the
> > winch_handler_lock.
>
> And?  There's no particular problem with host system calls being under
> a lock.  And the various other stuff is a kfree and a free_irq, which
> I don't think have a problem being called under a spinlock.

Indeed, right. Ok, no real objections on the patch.

> > I had carefully brought that stuff out keeping only the list access under
> > the lock, probably while fixing some "scheduling while atomic" warnings -
> > once the element is out of the list it's unreachable thus (IMHO) safely
> > accessible.
>
> Probably?  What in there is sensitive to being called under a lock?

Ok, sorry,  wrong here. I remembered doing the thing but it was for other 
reasons - reducing spinlock hold times. Doing syscalls under spinlocks is 
just (possibly) slow, not wrong. But ok, the commendment was "thou shalt not 
optimize".

> > So, list_del should be brought out from free_winch, which would then
> > become callable without the spinlock held.

> That would increase the amount of code, with no gain that I can see.
> The list_del would be duplicated, and the loop in winch_cleanup would
> have to drop and reacquire the lock around each call to free_winch.

I thought mainly to unregister_winch(); the lock in winch_cleanup() has been 
added now, I didn't see it.

> 				Jeff

-- 
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade

	

	
		
___________________________________ 
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB 
http://mail.yahoo.it