Re: PI BUG with -rt13

[david singleton]

On Nov 18, 2005, at 5:27 AM, Ingo Molnar wrote:

> * Dinakar Guniguntala <dino@xxxxxxxxxx> wrote:
>
> > Ingo, Thanks for providing the fix. However I still hit the same bug
> > even with your changes
>
> even with my patch the robust-futex code is still quite broken. E.g. in
> down_futex(), it accesses rt-mutex internals without any locking (!):
>
>         if (rt_mutex_free(lock)) {
>                 __down_mutex(lock __EIP__);
>                 rt_mutex_set_owner(lock, owner_task->thread_info);
>         }
>
> both rt_mutex_free() and rt_mutex_set_owner() must be called with the
> proper locking. David?

        Yes, the lock  needs to be protected by the robust semaphore.
        The locking order is:

                mmap_sem to protect the vma that holds the pthread_mutex

                robust_sem to protect the futex_mutex chain.

                futex_mutex - the rt_mutex associated with the 
pthread_mutex.


        This section of code performs the locking of the first waiter
        on the pthread_mutex, which was locked by another thread in the
        fast path in userspace.  The fast path locking does not enter 
the
        kernel.

        The first waiter locks the futex_mutex and then changes the
        owner to the 'owning' thread.  The waiter then calls 
down_interruptible
        to block on the futex_mutex.

        I'll provide a patch to protect this piece of code.

David

> 	Ingo

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/