On Thu, Aug 11, 2005 at 06:14:51PM +0100, Christoph Hellwig wrote:
> On Wed, Aug 10, 2005 at 10:11:45AM -0700, Paul E. McKenney wrote:
> > Hello!
> >
> > This patch is an experiment in use of RCU for individual code paths that
> > read-acquire the tasklist lock, in this case, unicast signal delivery.
> > It passes five kernbenches on 4-CPU x86, but obviously needs much more
> > testing before it is considered for serious use, let alone inclusion.
>
> I think we should switch over tasklist_lock to RCU completely instead of
> adding suck hacks. I've started lots of preparation work to get rid of
> tasklist_lock users outside of kernel/, especialy getting rid of any
> use in modules.
I agree fully with your end goal -- I would certainly look funny
disagreeing, right? ;-) And Oleg pointed out an area I missed in my
patch, am fixing, but in the meantime it most certainly does qualify as
a suck hack. :-/ Working on it...
But one of the cool things about RCU is that we can make this change
incrementally, with a series of small patches. We can have some of
the read paths protected by RCU and others continuing to be protected
by read_lock(), so that we can avoid the need for a "flag day" that
hits all 300+ uses of tasklist_lock.
FWIW, the approach that I am taking to fix the bug Oleg and Christoph
spotted is roughly as follows:
o Add "struct rcu_head rcu", "struct sighand_struct *successor",
and "int deleted" to struct sighand_struct. This allows
reliable signal delivery in face of sighand_struct replacements.
If an RCU reader finds (deleted && !successor), that reader
is trying to signal a dying process, so gives up. If an
RCU reader instead finds (deleted && successor), the reader
traverses the successor pointer to find the current version.
o Apply call_rcu() to deletion of struct sighand_struct, with
the exception of a couple of failure paths in exec.c, where
the sighand_struct was never exposed to RCU readers.
Thoughts?
Thanx, Paul
-
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]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
| |
 |