linux-kernel - Re: [PATCH 2/2] locking/pvqspinlock: Optionally store lock holder cpu into lock

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:   Sun, 12 Jul 2020 19:34:52 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Waiman Long <longman@...hat.com>
Cc:     Ingo Molnar <mingo@...hat.com>, Will Deacon <will.deacon@....com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Borislav Petkov <bp@...en8.de>, Arnd Bergmann <arnd@...db.de>,
        linux-kernel@...r.kernel.org, x86@...nel.org,
        linux-arch@...r.kernel.org, Nicholas Piggin <npiggin@...il.com>,
        Davidlohr Bueso <dave@...olabs.net>
Subject: Re: [PATCH 2/2] locking/pvqspinlock: Optionally store lock holder
 cpu into lock

On Sat, Jul 11, 2020 at 02:21:28PM -0400, Waiman Long wrote:
> The previous patch enables native qspinlock to store lock holder cpu
> number into the lock word when the lock is acquired via the slowpath.
> Since PV qspinlock uses atomic unlock, allowing the fastpath and
> slowpath to put different values into the lock word will further slow
> down the performance. This is certainly undesirable.
> 
> The only way we can do that without too much performance impact is to
> make fastpath and slowpath put in the same value. Still there is a slight
> performance overhead in the additional access to a percpu variable in the
> fastpath as well as the less optimized x86-64 PV qspinlock unlock path.
> 
> A new config option QUEUED_SPINLOCKS_CPUINFO is now added to enable
> distros to decide if they want to enable lock holder cpu information in
> the lock itself for both native and PV qspinlocks across both fastpath
> and slowpath. If this option is not configureed, only native qspinlocks
> in the slowpath will put the lock holder cpu information in the lock
> word.

And this kills it,.. if it doesn't make unconditional sense, we're not
going to do this. It's just too ugly.