[sane-devel] linux-2.6.1-smp,usb,oops,sane-backends-1.0.13,epson-1640

[Jaeger, Gerhard]

Hi,

On Samstag, 17. Januar 2004 22:22, Henning Meier-Geinitz wrote:
> Hi,
>
> On Sat, Jan 17, 2004 at 08:04:40PM +0000, Till Kamppeter wrote:
> > Is there a way to block the usage of the scanner.o module by means of
> > the SAME package (compile-time config or so)?
>
> No and that wouldn't help. When the kernel scanner driver is loaded,
> it claims the USB devices it knows. libusb can't access them in this
> case.
>
> > Or is it the best to ask
> > our kernel maintainer to make a kernel package without scanner.o? Is it
> > safe to publish a distro (Mandrake Linux 10.0) without scanner.o in its
> > kernel? Would I loose compatibility for certain scanner models or not?
>
> To my knowledge, every scanner that works with the scanner module also
> works with libusb. But this is only true for SANE, I don't know if
> it's the same for other software like vuescan.

AFAIK vuescan works only with the scanner module.

> If that's possible I would include the scanner driver, but not load it
> by default. So if the user really needs it, he can load it. I have no
> idea how to do that with hotplug, however.

Using hotplug, you should add the scanner module to the "blacklist" file
in /etc/hotplug

> > And does the oops only appear with the SMP kernel 2.6.x or also with the
> > standard kernel 2.6.x.
>
> There seem to be two oopses (maybe related): One that happens on
> unplugging a scanner. That one happens also without SMP and without
> preemtible kernel.
>
> The other one that seems to accour when a scanner is accessed (e.g.
> scanimage -L) only occurs with preemption turned on. I haven't tested
> SMP.

Probably you need to some spinlocks to protect the driver from being scheduled
in some cases!?

If I find some time, I'll check that...

Ciao,
  Gerhard

> I can't reproduce any of the libusb problems but I use different
> scanners.
>
> Bye,
>   Henning