Re: [2.5 PATCH] bug if cpufreq driver initialization fails

[Russell King <rmk@arm.linux.org.uk>]

On Sun, Jun 15, 2003 at 02:25:07PM -0400, Samuel Thibault wrote:
> Le dim 15 jun 2003 19:16:50 GMT, Russell King a tapoté sur son clavier :
> > On Sun, Jun 15, 2003 at 02:04:36PM -0400, Samuel Thibault wrote:
> > > I hence modified drivers/base/sys.c to have sysdev_driver_register()
> > > fail as well, and then I also had to modify kernel/cpufreq.c, because
> > > this failure did not imply a setting cpufreq_driver to NULL (preventing
> > > me from reinsmoding speedstep-ich: EBUSY)
> > 
> > Unfortunately, you created a by by doing so.  Eg:
> > 
> > - you have 3 devices on kset.list.
> > - you successfully register 2 of them with a driver.
> > - you fail one.
> > - sysdev_driver_register returns failure.
> > - module is unloaded while other parts of the kernel have references into
> >   the driver.
> > - the kernel oopses.
> Ok, that's what I was wondering, might an exit loop removing the already
> added driver be done?
> 
> (the second half of the patch might still be applied as soon as now)

Also, I think people need to consider whether "all devices failed"
or "any device failed" counts as overall failure.

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html