Re: shm updates broke UML

[Marc Zyngier <marc.zyngier@arm.com>]

On 01/08/11 18:32, Vasiliy Kulikov wrote:
> Marc,
> 
> On Mon, Aug 01, 2011 at 21:24 +0400, Vasiliy Kulikov wrote:
>> On Mon, Aug 01, 2011 at 19:19 +0200, Marc Zyngier wrote:
>>> Here you go (sorry about the line wrapping, damned webmail...):
>>
>> Thank you!  Now I see that the problem is rw_mutex is not initialized:
>>
>> 	down_write(&shm_ids(ns).rw_mutex);
>>
>> void __sched __down_write_nested(struct rw_semaphore *sem, int subclass)
>> {
>>     ...
>> 	if (sem->activity == 0 && list_empty(&sem->wait_list)) {
>>                                          ^^^^^^^^^^^^^^^ is NULL
>>
>> Wtf - ipc namespace should be fully initialized for the moment of
>> threads run...
> 
> Does ARM try to run _any_ threads before do_initcalls()?  IPC
> initialization is initcall, so any thread before do_initcalls() is a
> dependency bug.

I don't think it does anything different from other architectures.
The crash I'm observing seem to be from kworker, which is created from
an early_initcall (kernel/workqueue.c).

This looks very much like a generic problem to me.

	M.
-- 
Jazz is not dead. It just smells funny...