linux-kernel - Re: [PATCH tip/core/rcu 13/22] rcu: Fix grace-period hangs due to race with CPU offline

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:   Thu, 28 Jun 2018 15:06:46 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
Cc:     linux-kernel@...r.kernel.org, mingo@...nel.org,
        jiangshanlai@...il.com, dipankar@...ibm.com,
        akpm@...ux-foundation.org, mathieu.desnoyers@...icios.com,
        josh@...htriplett.org, tglx@...utronix.de, rostedt@...dmis.org,
        dhowells@...hat.com, edumazet@...gle.com, fweisbec@...il.com,
        oleg@...hat.com, joel@...lfernandes.org
Subject: Re: [PATCH tip/core/rcu 13/22] rcu: Fix grace-period hangs due to
 race with CPU offline

On Thu, Jun 28, 2018 at 05:38:33AM -0700, Paul E. McKenney wrote:
> Please let me try again.
> 
> The approach you are suggesting, clever though it is, disables a check

  https://lkml.kernel.org/r/20180627094633.GG2512@hirez.programming.kicks-ass.net

Is the one we're talking about, right?

That does not disable any actual check afaict. It simply does not do a
wakeup when ran on an offline CPU. And ensures we do an unconditional
wakeup soon after from a still running CPU.

> of a type that has proved to be an important diagnostic in the past.
> It is only reasonable to assume that this check would be important
> and helpful in the future, but only if that check remains in the code.

I am confused..

> Yes, agreed, given the current structure of the code, this particular
> instance of the check would not matter, but experience indicates that
> RCU code restructuring is not at all uncommon, with the current effort
> being but one case in point.

Once more confused...

> So, unless I am missing something, the only possible benefit of disabling
> this check is getting rid of an acquisition of an uncontended lock in
> a code path that is miles (sorry, kilometers) away from any fastpath.
> So, again, yes, it is clever.  If it sped up a fastpath, I might be
> sorely tempted to take it.  But the alternative is straightforward and
> isn't anywhere near a fastpath.  So, though I do very much appreciate
> the cleverness and creativity, I am not seeing your change to be a
> good tradeoff from a long-term maintainability viewpoint.

I think you mean guarantee/invariant instead of check. But I see it no
different than any other missed rcu_gp_kthread_wake(). You can similarly
fail to make the call while restructuring.