Re: Regression - Re: 2.6.28-rc2: runaway loop modprobe char-major-5-1

["Kay Sievers" <kay.sievers@vrfy.org>]

On Mon, Nov 24, 2008 at 14:31, Patrick McHardy <kaber@trash.net> wrote:
> Meelis Roos wrote:
>>>>
>>>> request_module: runaway loop modprobe char-major-5-1
>>>>
>>>> and then hangs.
>>>>
>>>> char-major-5-1 seems to mean ttyS0.
>>>
>>> I'm seeing the same problem on a Debian unstable system using
>>> initramfs and dm-crypt, but it seems to be caused by userspace
>>> changes, using an old initramfs image with just the new modules
>>> added manually works fine (well, breaks elsewhere).
>>
>> I have done some eveninfs testing, taking apart and putting together my
>> own initramfs. It looks like this is a kernel bug, not userspace - the hang
>> happens after initramfs is unpacked and after VFS and quotas are installed,
>> and after "msgmni has been set to 627" message but before BSG loading
>> message and before io scheduler initialization. And log before init is
>> executed from initramfs (and no modprobe is executed either, as my wrapper
>> does not run).
>>
>> Rerepeating, same userspace generats fine initramfs for 2.6.27 and the
>> initramfs contents are the same except modules tree that is for different
>> version.
>>
>> It happens on 2 of my computers, both have custom config (serial compiled
>> in, not modular). The hang happens before ttyS driver initialization.
>
> char-major-5-1 is /dev/console here.
>
>> Seems like a kernel regression to me.
>
> It does, I changed some configuration options from =m to =y and
> it started working again. Unfortunately I'm unable to find the
> exact option that fixed it since I don't have the previous
> configuration anymore and random attempts to trigger the problem
> again failed.

Maybe you just miss /dev/console in initramfs? That would trigger the
kernel forked modprobe with that major/minor.

Some device nodes must be created or included in initramfs before any
action is taken.

Kay