linux-kernel - Re: Reasoning about memory ordering

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]

Message-ID: <c7e3e457-fac3-7e9f-2045-f4b9a7d54702@suse.com>
Date:   Fri, 23 Feb 2018 17:59:53 +0200
From:   Nikolay Borisov <nborisov@...e.com>
To:     Alan Cox <gnomes@...rguk.ukuu.org.uk>
Cc:     LKML <linux-kernel@...r.kernel.org>,
        "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>,
        mathieu.desnoyers@...icios.com,
        Peter Zijlstra <peterz@...radead.org>, parri.andrea@...il.com
Subject: Re: Reasoning about memory ordering

On 23.02.2018 17:38, Alan Cox wrote:
>> Given this is the current state of the code (it's part of btrfs) I believe
>> the following could/should be done: 
> 
> Is there benchmarking data to show that a custom lock is justified
> (especiaally given it's going to mean btrfs and rtlinux don't play
> together nicely since it won't be able to see the mutex lock and do
> priority boosting ?)

No, unfortunately this is not code that was written by me. I'm just
arguing it's not entirely correct. Generally in btrfs we'd like to allow
concurrent DIO read/write accesses to unrelated portionsin a file. Hence
I can't just use inode_Lock (DIO writes are synchronized with
inode_lock). Also I cannot use inode_shared_Lock to synchronize multiple
DIO reader against writers (including truncate) since this tanks
perfromance and I already tried[0] this approach and received a NAK...

[0] https://patchwork.kernel.org/patch/10232015/

> 
> Alan
> 
> 
>