Re: [PATCH] scsi_host sysfs updates scsi-misc-2.5 [0/2]

[Mike Anderson <andmike@us.ibm.com>]

James Bottomley [James.Bottomley@steeleye.com] wrote:
> On Mon, 2003-05-12 at 13:41, Mike Anderson wrote:
> > I would like to have feedback from Mochel on the device_initialize
> > call.
> > 
> > The documentation I have read indicates the device_initialize call was to
> > allow the struct device to be usable for reference counting. I think we
> > should determine if that is the intent and can we call kobj_set_kset_s
> > in device_initialize.
> > 
> > Another option may be to set the ktype value on our struct device
> > kobject until device_add overrides this with the call to kobj_set_kset_s.
> 
> I can confirm that moving kobj_set_kset_s fixes my boot up problem (with
> your other patch).  However, now I'm running into an obscure slab
> corruption issue, so I suspect the move has to be more carefully
> orchestrated.

I checked out that setting kobj.ktype also seems to solve the problem
(using a hacked version of scsi_debug and calling scsi_register followed
by a call to scsi_unregister). This would give a more localized work
around, but device_initialize kobj_set_kset_s may be a better choice.

What slab size? (size-512)

-andmike
--
Michael Anderson
andmike@us.ibm.com