Re: Kernel boot hangs on commit "switch fput to task_work_add"

[Fengguang Wu <wfg@linux.intel.com>]

On Sat, Jul 14, 2012 at 04:19:03PM +0100, Al Viro wrote:
> On Sat, Jul 14, 2012 at 10:34:59PM +0800, Fengguang Wu wrote:
> > On Sat, Jul 14, 2012 at 03:14:15PM +0100, Al Viro wrote:
> > > On Sat, Jul 14, 2012 at 09:58:49PM +0800, Fengguang Wu wrote:
> > > > > Hi Al, here I got the output for
> > > > > 
> > > > > (qemu) sendkey alt-sysrq-l
> > > > > (qemu) sendkey alt-sysrq-t
> > > > 
> > > > I repeated that several times and here are the results.
> > > > 
> > > > (qemu) sendkey alt-sysrq-l
> > > > (qemu) sendkey alt-sysrq-l
> > > > (qemu) sendkey alt-sysrq-l
> > > > (qemu) sendkey alt-sysrq-t
> > > > (qemu) sendkey alt-sysrq-t
> > > > (qemu) quit
> > > > 
> > > > The user space shutdown is polling on something, which prevents the
> > > > system from reboot..
> > > 
> > > Which userland it is?
> > 
> > It's a customized ubuntu core. The original tgz image is downloaded
> > here, however it's not directly usable as initrd.. Should I send my
> > hacked one?
> 
> Sigh...  FWIW, with your config neither the mainline nor the tree you'd
> been testing manage to boot with .deb produced by squeeze make-kpkg.
> Both in the same way - they get to busybox, mount(8) in there keeps
> failing wiht -ENODEV all the time and the damn thing ends up in
> busybox shell.  Fsck knows what's going on - apparently the damn thing
> also redirects all printks someplace invisible as soon as it gets to
> userland, so the obvious ways to see what's going on do not work.
> RTFS(busybox) time, I guess...
> 
> With reasonably sane config both trees work fine.  I'll grab your image
> and see what happens with it.

Err, sorry for that! This is some weird randconfig setup, after all...
I've sent you my hacked initrd in a big email, hope you can receive it.
As I already have the reproduce environment, I can help debug it, too.

Thanks,
Fengguang