Re: [PATCH 5/5] uio: Implement hotunplug support, using libunload

[Hans J. Koch]

On Mon, Oct 04, 2010 at 11:19:34AM -0700, Eric W. Biederman wrote:
> "Hans J. Koch" <hjk@xxxxxxxxxxxxx> writes:
> 
> > On Sun, Sep 26, 2010 at 03:49:03PM -0700, Eric W. Biederman wrote:
> >> 
> >> With this change it is possible to remove a module that implements
> >> a uio device, or to remove the underlying hardware device of a uio
> >> device withot crashing the kernel, or causing user space more problems
> >> than just an I/O error.
> >
> > Well, that I/O error can also be a segfault if userspace accesses
> > memory previously mmap'ed. So a userspace program needs to properly
> > handle -EIO from read(), and has to handle SIG_SEGV. 
> 
> It is SIG_BUS (just like truncate), but yes.

All I saw was "Segmentation fault" from my test app when unloading
my uio_dummy test driver. I didn't check which signal it actually was.
Doesn't matter too much as long as it can be handled.

> If userspace doesn't
> handle there error it can go down.  All of which is a better state
> then the current version where the kernel crash if you hotunplug
> a uio device.
> 
> > This should also
> > be mentioned in Documentation/DocBook/uio-howto.tmpl.
> > Or do you have a better solution?
> 
> A better solution for?

For preventing the segfault situation. But forget that, it's OK as it
is. But it should be mentioned in the documentation.

Thanks,
Hans

> 
> Eric
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/