linux-kernel - Re: [RFC][PATCH 0/5] Optimize percpu-rwsem

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

Date:	Tue, 26 May 2015 14:35:20 -0400
From:	Tejun Heo <tj@...nel.org>
To:	Linus Torvalds <torvalds@...ux-foundation.org>
Cc:	Peter Zijlstra <peterz@...radead.org>,
	Oleg Nesterov <oleg@...hat.com>,
	Paul McKenney <paulmck@...ux.vnet.ibm.com>,
	Ingo Molnar <mingo@...hat.com>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	der.herr@...r.at, Davidlohr Bueso <dave@...olabs.net>
Subject: Re: [RFC][PATCH 0/5] Optimize percpu-rwsem

Hello, Linus.

On Tue, May 26, 2015 at 11:12:04AM -0700, Linus Torvalds wrote:
...
> Is there some new use that I don't know about? Have people really
> looked at that uprobes code deeply? OF COURSE global locks will have
> problems, I'm not at all convinced that "let's make that global lock
> really complicated and clever" is the proper solution.

I've posted a patchset to make threadgroup rwsem cgroup-specific (it
is used only by cgroups) and replace the per-threadgroup rwsems with a
global percpu_rwsem.  This is primarily to allow either succeeding or
failing migrations of multiple tasks atomically.  We already try to do
multiple task migrations in some corner cases and the unified
hierarchy will make wider use of it.  Currently, the implementation is
broken as there's no easy way of reverting changes if the operation
fails in the middle.

Given that task migrations aren't particularly high frequency
operations, a global percpu_rwsem is likely to be workable while being
less tedious and slightly less costly for the usual fork/exit paths.

That said, from cgroup's pov, write-locking per-threadgroup rwsems of
all target tasks works too although we'd need an outer mutex to
serialize down-write attempts to avoid locking order issues.

Thanks.

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