speedtch usbatm.h,1.4,1.5 usbatm2.c,1.13,1.14

Roman Kagan rkagan at mail.ru
Mon Feb 14 17:28:32 EST 2005


  Hi Duncan,

On Sat, Feb 12, 2005 at 07:55:24PM +0300, Roman Kagan wrote:
> On Sat, Feb 12, 2005 at 03:35:00PM +0100, Duncan Sands wrote:
> > For me the main problem with getting these changes into the kernel tree
> > is that they are not backwards compatible:
> > [...]
> > (2) the ATM device is only created once the USB device is fully
> > initialized, not straight away like now.  This may cause trouble for
> > scripts.
> 
> If the device is created, but the firmware isn't loaded yet, any attempt
> to open the atm device will fail.  So scripts have to work around this
> delay anyway, even with 2.6.10 version.

More precisely, any attempt to open a socket on the existing but still
uninitialized ATM device will fail, for the reason unclear to the
userspace tool.  Some ATM tools don't handle it right.  br2684ctl, for
one, does nasty things in this case: it quits without cleaning up the
Ethernet device it created.

So IMO the current behavior is much saner even with those scripts:
they'll see the ATM device only when it's ready (well, there's still a
small window we discussed on the ATM list), and the ATM tools would fail
with a proper diagnostic if run too early.

> > > Besides, I think we need to make up a usbatm subtree buildable
> > > within the kernel, so that users of 2.6.1[01] could fetch it and
> > > replace the original contents of drivers/usb/atm.  I'll add
> > > Kconfig and Makefile.kernel to CVS, if you don't mind.

I've noticed your updates to that, thanks!  (I wish I knew about
Kbuild.)  The mail notification via usbatm-commits still doesn't work,
sigh...

Cheers,
  Roman.



More information about the Usbatm mailing list