Re: [RFC PATCH] watchdog: Add hook for kicking in kdump path

[Don Zickus <dzickus@redhat.com>]

On Fri, Apr 12, 2013 at 02:30:24PM -0700, Guenter Roeck wrote:
> On Fri, Apr 12, 2013 at 05:16:27PM -0400, Don Zickus wrote:
> > On Wed, Apr 10, 2013 at 08:10:41AM -0700, Guenter Roeck wrote:
> > > > have no idea how to even find out if multiple watchdogs are open on the
> > > > system.  Is there a list I could walk?  And with regard to 'watchdog is
> > > 
> > > /* the dev_t structure to store the dynamically allocated watchdog devices */
> > > static dev_t watchdog_devt;
> > > 
> > > One way to look up the allocated watchdogs might be to loop through all kobj
> > > instances for the major device using kobj_lookup. Don't know if there is a
> > > better way.
> > 
> > Hmm, I got around to poking at this today and I am not sure kobj_lookup
> > will work.  Besides being surrounded with another mutex, I don't have
> > access to the character device domain to pass to kobj_lookup.
> > 
> > Perhaps I am not reading the code right, but I can't find a good way
> > forward.
> > 
> > The only other hack I can think of, is to embed a list object in the
> > watchdog structure and list_add each new register'd watchdog.  Then it
> > would be trivial to walk the watchdog list.
> > 
> After looking into it again, I agree. Maybe you can give it a try. At least
> other options look even more complicated (eg creating a watchdog class ?).

I looked at the watchdog class in watchdog_core.c.  Even implemented
class_for_each_device.  But got stuck trying to figure out how to go from
a struct device *dev to a struct watchdog_device.  Then I realized in the
bowels of class_for_each_device were spinlocks. :-(

So I implemented an RCU list.  It isn't the prettiest solution but it
seems to work.

I posted a V2 and realized I forgot to cc you on it.  I apologize for
that. :-(  I hope you can find it, otherwise I can post a pointer to it.

Cheers,
Don