Re: [S390] cio: kernel stack overflow.

[Martin Schwidefsky <schwidefsky@de.ibm.com>]

On Wed, 2006-08-30 at 21:19 +0200, Heiko Carstens wrote:
> > >-            cdev->id = (struct ccw_device_id) {
> > >-                    .cu_type   = cdev->private->senseid.cu_type,
> > >-                    .cu_model  = cdev->private->senseid.cu_model,
> > >-                    .dev_type  = cdev->private->senseid.dev_type,
> > >-                    .dev_model = cdev->private->senseid.dev_model,
> > >-            };
> > >+            cdev->id.cu_type   = cdev->private->senseid.cu_type;
> > >+            cdev->id.cu_model  = cdev->private->senseid.cu_model;
> > >+            cdev->id.dev_type  = cdev->private->senseid.dev_type;
> > >+            cdev->id.dev_model = cdev->private->senseid.dev_model;
> > 
> > I don't see any obvious place that zeroes out cdev->id.
> > In particular, it looks like cdev->id.match_flags and .driver_info
> > are never cleared (i.e., they retain whatever old garbage they had
> > before).  More importantly, if anyone ever adds any more fields to
> > struct ccw_device_id, then they will also be retain old garbage values,
> > which is a maintenance pitfall.  Is this right, or did I miss something
> > again?
> 
> You're right. Thanks for pointing this out! I will take care of it.

The ccw_device_id structure contains two more fields in addition to the
field that are set up in ccw_device_recog_done, namely match_flags and
driver_info. driver_info is set later in ccw_bus_match, so that is fine.
match_flags of the device is never used, only the match_flags of the
drivers version of the ccw_device_id is used. So the code is correct
even without the memset. But your point about the maintenance pitfall is
valid, we will add a memset after 2.6.18. I don't want to push yet
another patch.

-- 
blue skies,
  Martin.

Martin Schwidefsky
Linux for zSeries Development & Services
IBM Deutschland Entwicklung GmbH

"Reality continues to ruin my life." - Calvin.