Re: [PATCH] 2.5.3 remove global semaphore_lock spin lock.

[Jens Axboe <axboe@suse.de>]

On Thu, Jan 31 2002, Jeff Garzik wrote:
> On Thu, Jan 31, 2002 at 03:55:02PM -0800, Andrew Morton wrote:
> > > +       unsigned long flags;
> > > +       wq_write_lock_irqsave(&sem->wait.lock, flags);
> > > -       spin_lock_irq(&semaphore_lock);
> > 
> > I rather dislike spin_lock_irq(), because it's fragile (makes
> 
> It's less flexible for architectures, too.
> 
> spin_lock_irqsave is considered 100% portable AFAIK, and I make it my
> own policy to s/_irq/_irqsave/ when the opportunity strikes in my PCI
> drivers.

spin_lock_irq is cheaper, though, and sometimes you _know_ it's safe to
use. For instance, if the function in question can block (ie never
called with interrupts disabled) then using spin_lock_irq is always
safe.

I've heard this portability argument before, anyone care to outline
_what_ the problem allegedly is?? Major part of the kernel uses
spin_lock_irq and I suspect we would be seeing lots of request list
corruption did it not work.

-- 
Jens Axboe