cvs commit: src/sys/sys proc.h src/sys/kern kern_thread.c

[Jeff Roberson]

On Tue, 12 Jun 2007, Sam Leffler wrote:

> Jeff Roberson wrote:
>> jeff        2007-06-12 07:24:47 UTC
>>
>>   FreeBSD src repository
>>
>>   Modified files:
>>     sys/sys              proc.h     sys/kern             kern_thread.c 
>> Log:
>>   Solve a complex exit race introduced with thread_lock:
>>    - Add a count of exiting threads, p_exitthreads, to struct proc.
>>    - Increment p_exithreads when we set the deadthread in thread_exit().
>>    - When we thread_stash() a deadthread use an atomic to drop the count.
>>    - Spin until the p_exithreads count reaches 0 in thread_wait().
>>    - Lock the last exiting thread momentarily to be certain that it has
>>      exited cpu_throw().
>>    - Restructure thread_wait().  It does not need a loop as there will only
>>      ever be one thread.
>
> What are the symptoms of this race?  Since the sched lock changes I've 
> repeatedly hit user-after-free crashes and other data corruption-like assert 
> failures that appear to trigger through signals and exit.  These are all on 
> UP (laptops).

I hadn't heard of these problems.  This race is not possible on UP.  On MP 
you could have had a NULL deref in cpu_throw().  UP should have been 
relatively unchanged as a result of threadlock and was tested under heavy 
load.  Can you describe the problems you're seeing in any further detail?

Thanks,
Jeff

>
> 	Sam
>