2.6.9-rc4-mm1-VP-T7 - horrid death in vortex_init at boot

[Valdis . Kletnieks]

2.6.9-rc4-mm1-VP-T7 plus Ingo's patch to profile.c and 3c59x.c to use
raw_rwlock_t and raw_spinlock_t rather than the non-raw variant.  It croaked
when it found the onboard ethernet controller on a Dell Latitude C840 laptop:

lspci says it's a:
02:00.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 78)

Got the following (admittedly truncated - was handwritten and CTS is a pain,
literally) at very early boot:

3c59x: Donald Becker and others. ...
0000:02:00.0: 3Com PCI 3c905C Tornado at 0xec80. Vers LK1.1.19
kernel BUG at net/core/net-sysfs.c:384
process swapper
... registers skipped
call trace:
	show_stack
	show_registers
	die
	do_invalid_op
	error_code
	class_hotplug
	kobject_hotplug
	kobject_add
	class_device_add
	netdev_register_sysfs
	netdev_run_todo
	register_netdev
	vortex_probe1
	vortex_init_one
	pci_device_probe_static
	__pci_device_probe
	pci_device_probe
	bus_match
	driver_attach
	bus_add_driver
	driver_register
	pci_register_driver
	vortex_init
	do_initcalls

The offending code:

static void netdev_release(struct class_device *cd)
{
        struct net_device *dev
                = container_of(cd, struct net_device, class_dev);

        BUG_ON(dev->reg_state != NETREG_RELEASED);

        kfree((char *)dev - dev->padded);
}

I'm guessing something broken in the bk-driver-core patch in -rc4-mm1, as
that completely overhauled this stuff.

This ring any bells?  If need be, I'll scare up a serial cable and get a
more complete trace - somehow, I don't think netconsole will help here.. ;)

(As an aside, kobject_hotplug calls call_usermodehelper() - which seems
like a Bad Idea if we haven't gotten userspace up and running yet?

Attachment: pgp00000.pgp
Description: PGP signature