RE: 2.6.11, USB: High latency?

[Alan Stern <stern@rowland.harvard.edu>]

On Fri, 1 Apr 2005, kus Kusche Klaus wrote:

> > The biggest advantage would come from using a bottom-half 
> > handler to do 
> > most of the work.  Right now the uhci-hcd driver does 
> > everything in its 
> > interrupt handler.  This would certainly help IRQ latency; it 
> > might not 
> > affect application latency very much.
> 
> Sounds very reasonable, thanks. Also helps application latency,
> because with the RT patches, I can tune the rt prio of softirq
> execution (that's where bottom-half goes, doesn't it?) w.r.t. the
> rt prio of the application threads.

Yes.  Bear in mind, however, that if these application threads are doing 
I/O over USB then they will be forced to wait for the bottom half to 
execute, regardless of its priority.

> However, if I understand things correctly, if you really need 
> to disable all interrupts while doing the USB work, it will not
> make any difference if IRQs are disabled while you are in the
> USB IRQ handler, or if they are disabled for the same amount of 
> work/time in the bottom-half code.

For most of the USB work it will be necessary only to insure that no more
than one copy of the bottom-half handler is running at a time, which I 
think the kernel does automatically for tasklets.  There are a few places 
where all IRQs will have to be disabled, but those places are relatively 
small and short.

Right now, of course, everything runs with IRQs disabled.

> > We'll see what happens with the upcoming changes.  Maybe 
> > you'll be able to 
> > test them for me?
> 
> Basically, yes (as long as our company doesn't decide to stop the
> linux experiments).
> 
> However, I depend on Ingo's RT patch, which is against the -rc series,
> not against the -mm series. So I will probably not be able to apply
> patches created against -mm.

Okay.  It will be a while before the new code is ready and the changes on 
which it depends have gotten into -rc.

Alan Stern

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/