On Sun, 2010-04-18 at 18:37 +0200, Gary Jennejohn wrote:
quoted text
> On Sun, 18 Apr 2010 08:24:59 -0700
> Greg KH <greg@kroah.com> wrote:
> 
> > On Sun, Apr 18, 2010 at 11:40:31AM +0200, Gary Jennejohn wrote:
> > > It's a pity that Linux doesn't allowing blocking single interrupts, like
> > > that various BSDs out there.
> > 
> > Um, what do you mean?  You can implicitly block an interrupt, but why
> > would you want to do that?
> > 
> 
> Plenty of reasons.  One can block the interrupt to a certain driver which
> may require that for weird hardware reasons while reenabling and allowing
> other interrupts to be fielded and served.
> 
> That's how the VAX and Motorola 680x0 worked and AT&T Ver. III and IV had
> the splx() macros to block individual interrupts.  I worked on this stuff
> in the 1980's and it was extremely convenient.

Bear in mind that Linux has about two orders of magnitude more drivers
than all of the above put together and that Greg manages a large share
of them, not to mention most of the driver framework. If your question
puzzles Greg, it's a good sign it's not actually needed in practice.

quoted text
> Modern HW may not require it as much, but always blocking all interrupts
> with cli() isn't optimal, although it does force one to come up with
> creative solutions/workarounds :)

cli() has been deprecated for a very long time. It's usefulness to
drivers is just about nil in a multiprocessor environment. The modern
equivalent (local_irq_disable) is used in only a small handful of truly
ancient drivers.

-- 
http://selenic.com : development and support for Mercurial and Linux


_______________________________________________
Kernel-mentors mailing list
Kernel-mentors@selenic.com
http://selenic.com/mailman/listinfo/kernel-mentors