Re: [RFC,PATCH] use rcu for fasync_lock

[Manfred Spraul]

Linus Torvalds wrote:

> On Sun, 21 Dec 2003, Manfred Spraul wrote:
>  
>
> > Initially I tried to keep the patch as tiny as possible, thus I avoided 
> > adding an inline function. But Stephen Hemminger convinced me to update 
> > the network code, and thus it didn't matter and I've switched to an 
> > inline function.
> > What do you think about the attached patch?
> >    
>
> Please, NO!
>
> Stuff like this
>
> 	-       write_lock_irq(&fasync_lock);
> 	+       if (s)
> 	+               lock_sock(s);
> 	+       else
> 	+               spin_lock(&fasync_lock);
> 	+
>
> should not be allowed. That's especially true since the choice really is a 
> static one depending on the caller.
>
> Just make the caller do the locking.
>  
It's not that simple: the function does
   kmalloc();
   spin_lock();
   use_allocation.
If the caller does the locking, then the kmalloc would have to use 
GFP_ATOMIC, or the caller would have to do the alloc.
But: as far as I can see, these lines usually run under lock_kernel(). 
If this is true, then the spin_lock(&fasync_lock) won't cause any 
scalability regression, and I'll use that lock instead of lock_sock, 
even for network sockets.

--
   Manfred

-
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/