Re: [PATCH] fill in scsi_host_template module owner

[Christoph Hellwig <hch@infradead.org>]

[please don't remove me from the To: or Cc: line, please - I tend
to read personal email much faster than list traffic, and it's the
normal policy for most kernel development lists these days]

On Wed, May 02, 2007 at 10:07:43AM +0200, Olaf Kirch wrote:
> On Monday 30 April 2007 18:02, Christoph Hellwig wrote:
> >  - a driver should handle hot unplugs proper, even if THIS_MODULE
> >    is set this can happen if the hardware just goes away.
> 
> In the case of the iscsi bug I looked at, it wasn't so much the
> hardware going away, it was the rmmod that was kind of hurtful ;)

In the iscsi case this is true as it's not a traditional driver
with a ->remove method that is called on a hot unplug.  Every pci
or usb driver needs to handle hot removal, though.  And yes, I expect
a lot of lingering bug in those areas outside of the few drivers used
on modern highend systems.

> I also thought about adding something like this to
> scsi_add_host:
> 
> +       if (sht->module == NULL && sht->queuecommand &&
> +           !core_kernel_text((unsigned long) sht->queuecommand))
> +               printk(KERN_ERR "%s: didn't set module owner field\n", sht->name);
> 
> But maybe that's overdoing things a little.

->queuecommand is also mandatory, so you could remove the check for it
beeing present, but I think this is really overding things a little.