RE: 2.6.11, USB: High latency?

["kus Kusche Klaus" <kus@keba.com>]

> * kus Kusche Klaus <kus@keba.com> wrote:
> >    IRQ 7-724   0d..1    1us : end_8259A_irq (do_hardirq)
> >    IRQ 7-724   0d..1    1us!: enable_8259A_irq (do_hardirq)
> >    IRQ 7-724   0d...  832us : do_hardirq (do_irqd)
> >    IRQ 7-724   0d...  833us : trace_irqs_on (do_hardirq)
> 
> >     mmap-1000  0d.h1   21us : end_8259A_irq (__do_IRQ)
> >     mmap-1000  0d.h1   22us!: enable_8259A_irq (__do_IRQ)
> >     mmap-1000  0d.h.  662us : irq_exit (do_IRQ)
> >     mmap-1000  0d..1  662us : do_softirq (irq_exit)
> 
> >     mmap-1000  0d.h.    0us : do_IRQ (c012d6d5 7 0)
> >     mmap-1000  0d.h1    2us!: mask_and_ack_8259A (__do_IRQ)
> >     mmap-1000  0d.h1  938us : redirect_hardirq (__do_IRQ)
> >     mmap-1000  0d.h1  939us : wake_up_process (redirect_hardirq)
> 
> such 'freezes' almost certainly signal some sort of hardware 
> latency - 
> some device holding the system bus up during DMA. There is no 
> algorithmic reason for any of those steps above to take 
> several hundreds 
> of microseconds.
> 
> 	Ingo

I asked our hardware team. The hardware has two devices which are
in use and capable of busmaster/DMA transfers: 
The intel e100 ethernet controller and the intel PIIX4 USB 
controller. 
The IDE interface is also a busmaster, but there are only PIO IDE
devices.

I suspect the latter, as USB reads were running in parallel...
How many bytes are transferred at most by the USB controller 
for a single request? How long may this take?

Any experiences / opinitions / advices?

Moreover, we know from experience that the "WBINDV" instruction
(Write back and invalidate CPU cache) can cause such latencies. 

Does this instruction occur anywhere in Linux?

-- 
Klaus Kusche                 (Software Development - Control Systems)
KEBA AG             Gewerbepark Urfahr, A-4041 Linz, Austria (Europe)
Tel: +43 / 732 / 7090-3120                 Fax: +43 / 732 / 7090-8919
E-Mail: kus@keba.com                                WWW: www.keba.com
-
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/