linux-kernel - Re: [PATCH 06/13] locking/qspinlock: merge qspinlock

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 for Android: free password hash cracker in your pocket

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

Date:   Wed, 6 Jul 2022 16:16:27 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Waiman Long <longman@...hat.com>
Cc:     Nicholas Piggin <npiggin@...il.com>,
        Ingo Molnar <mingo@...hat.com>, Will Deacon <will@...nel.org>,
        Boqun Feng <boqun.feng@...il.com>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 06/13] locking/qspinlock: merge qspinlock_paravirt.h into
 qspinlock.c

On Wed, Jul 06, 2022 at 09:35:48AM -0400, Waiman Long wrote:
> On 7/5/22 13:20, Peter Zijlstra wrote:
> > On Tue, Jul 05, 2022 at 12:38:13AM +1000, Nicholas Piggin wrote:
> > > There isn't much reason to keep these separate.
> > The reason was so that other paravirt implementations could be added.
> > 
> > The CNA thing was also implemented this way...
> 
> Do you have any plan to take CNA [1] some time in the future?

I do mean to look at it, but somehow I never get around to it :-(
ISTR the O(n) on unlock got fixed.

Also; and I think this is why it never really gets to the top of the
todo list, I do still feel it is fixing the wrong problem. IIRC the
whole reason for CNA is the futex hash lock and I think that a better
futex ABI would be a *much* better solution there.

> Anyway, the main reason the paravirt code is separated into a separated file
> is to leave only the core part in qspinlock.c so that new users are
> overwhelmed with the messy details for the paravirt code. Putting everything
> into a single can make it harder to read for the newbies.
> 
> Also eliminating the preprocessor trick will make it harder to integrate a
> different qspinlock variant, like CNA, into the code base.

Fixed that already: YsR8BIyrSCQ8AlEo@...ktop.programming.kicks-ass.net