netdev - rt hash table / rt hash locks question

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 16 Jun 2010 20:46:33 +1000
From:	Nick Piggin <npiggin@...e.de>
To:	netdev@...r.kernel.org
Subject: rt hash table / rt hash locks question

I'm just converting this scalable dentry/inode hash table to a more
compact form. I was previously using a dumb spinlock per bucket,
but this doubles the size of the tables so isn't production quality.

What I've done at the moment is to use a bit_spinlock in bit 0 of each
list pointer of the table. Bit spinlocks are now pretty nice because
we can do __bit_spin_unlock() which gives non-atomic store with release
ordering, so it should be almost as fast as spinlock.

But I look at rt hash and it seems you use a small hash on the side
for spinlocks. So I wonder, pros for each:

- bitlocks have effectively zero storage
- bitlocks hit the same cacheline that the hash walk hits.
- in RCU list, locked hash walks usually followed by hash modification,
  bitlock should have brought in the line for exclusive.
- bitlock number of locks scales with hash size
- spinlocks may be slightly better at the cacheline level (bitops
  sometimes require explicit load which may not acquire exclusive
  line on some archs). On x86 ll/sc architectures, this shouldn't
  be a problem.
- spinlocks better debugging (could be overcome with a LOCKDEP
  option to revert to spinlocks, but a bit ugly).
- in practice, contention due to aliasing in buckets to lock mapping
  is probably fairly minor.

Net code is obviously tested and tuned well, but instinctively I would
have tought bitlocks are the better way to go. Any comments on this?

Thanks,
Nick

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html