[PATCH v1 11/11] locks: give the blocked_hash its own spinlock

Stefan (metze) Metzmacher metze at samba.org
Tue Jun 4 10:19:53 EDT 2013


Hi Jeff,

> There's no reason we have to protect the blocked_hash and file_lock_list
> with the same spinlock. With the tests I have, breaking it in two gives
> a barely measurable performance benefit, but it seems reasonable to make
> this locking as granular as possible.

as file_lock_{list,lock} is only used for debugging (/proc/locks) after this
change, I guess it would be possible to use RCU instead of a spinlock.

@others: this was the related discussion on IRC
(http://irclog.samba.org/) about this:

16:02 < metze> jlayton: do you have time to discuss your file_lock_lock
changes?
16:02 < jlayton> metze: sure, what's up?
16:03 < jlayton> metze: note that it won't help vl's thundering herd
problems...
16:03 < metze> is it correct that after your last patch file_lock_lock
is only used for /proc/locks?
16:03 < jlayton> well, it's only used to protect the list that is used
for /proc/locks
16:04 < jlayton> it still gets taken whenever a lock is acquired or
released in order to manipulate that list
16:04 < metze> would it be a good idea to use rcu instead of a spin lock?
16:04 < jlayton> I tried using RCU, but it turned out to slow everything
down
16:04 < jlayton> this is not a read-mostly workload unfortunately
16:04 < jlayton> so doing it with mutual exclusion turns out to be faster
16:04 < metze> ok
16:05 < jlayton> I might play around with it again sometime, but I don't
think it really helps. What we need to ensure is
                 that we optimize the code that manipulates that list,
and RCU list manipulations have larger overhead
16:06 < jlayton> metze: that's a good question though so if you want to
ask it on the list, please do
16:06 < jlayton> others will probably be wondering the same thing
16:08 < metze> maybe it's worth a comment in commit message and the code
16:08 < metze> btw, why don't you remove the ' /* Protects the
file_lock_list and the blocked_hash */' comment?

metze

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 261 bytes
Desc: OpenPGP digital signature
URL: <http://lists.infradead.org/pipermail/linux-afs/attachments/20130604/5ec71c69/attachment-0001.sig>


More information about the linux-afs mailing list