Re: [RFC] logdev debugging memory device for tough to debug areas

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, 2005-03-29 at 13:56 +0200, Ingo Molnar wrote:

> > To have a task take back the ownership, I had the stealer call 
> > task_blocks_on_lock on the task that it stole it from. To get this to 
> > work, when a task is given the pending ownership, it doesn't NULL the 
> > blocked_on at that point (although the waiter->task is set to NULL).  
> > But this gives the race condition in pi_setprio where it checks for 
> > p->blocked_on still exists. Reason is that I don't want the waking up 
> > of a process to call any more locks. To solve this, I had to (and this 
> > is what I don't like right now) add another flag for the process 
> > called PF_BLOCKED. So that this can tell the pi_setprio when to stop.  
> > This flag is set in task_blocks_on_lock and cleared in pick_new_owner 
> > where the setting of blocked_on to NULL use to be.
> 
> which locks are affected? I'd prefer the simplest solution. If there's 
> more overhead with deadlock detection (which is a debugging feature), 
> that doesnt matter much.

I've already covered the deadlock detection problems, and that didn't
add anymore overhead. That was just what kept breaking every time I
changed something ;-)

The overhead is caused by the pi_setprio knowing when to stop following
the chain of blocked processes. Since it can't check for p->blocked_on
== NULL anymore. I first had it check that or p->flags & PF_PENDOWNER,
but since the process waking up can happen at anytime without grabbing
any lock, this flag can be cleared as well as the blocked_on at the time
of this test.  So I added another flag to p->flags to tell when the
process is locked and not pending ownership.  I don't know yet if this
works, I have to run to an appointment now and I'll find out when I get
back.

Oh, and I disabled the deadlock check on the stolen case. So when a
pending owner goes back to blocked, right now it doesn't check the
deadlock, since the code uses current as the test, and that no longer
applies. So for now I have a test in task_blocks_on_lock to see if task
== current otherwise, don't check for deadlocks.

-- Steve


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

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux