On Sunday 14 April 2002 13:08, Vojtech Pavlik wrote:
> On Thu, Apr 11, 2002 at 10:08:12PM +0200, Oliver Neukum wrote:
> > there's a problem in most usb network drivers.
> > In these drivers the rx path is completely in interrupt.
> > Thus if resubmitting the urb fails no new rx urb will
> > ever be submitted. Therefore any exhaustion of the
> > atomic pool at an inconvinient moment will kill
> > these drivers and possibly if you are using nfs, the
> > whole box.
>
> This also is not a problem with catc.c I think. There the rx urb is
> sumbitted from an irq urb, which doesn't need to be resumbitted.
>
> So, I don't think this is 'most usb drivers' anymore.
Yes, things are improving. But I am not sure about the solutions
in usbnet and rtl8150. IMHO they can put the computer
into a state in which it executes only tasklets and irq handlers.
Regards
Oliver
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
