linux-kernel - Re: [PATCH 00/13] crypto: x86 - yield FPU context during long loops

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, 20 Dec 2022 20:02:32 +0000
From:   Eric Biggers <ebiggers@...nel.org>
To:     Robert Elliott <elliott@....com>
Cc:     herbert@...dor.apana.org.au, davem@...emloft.net, Jason@...c4.com,
        ardb@...nel.org, ap420073@...il.com, David.Laight@...lab.com,
        tim.c.chen@...ux.intel.com, peter@...jl.ca, tglx@...utronix.de,
        mingo@...hat.com, bp@...en8.de, dave.hansen@...ux.intel.com,
        linux-crypto@...r.kernel.org, x86@...nel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 00/13] crypto: x86 - yield FPU context during long loops

On Mon, Dec 19, 2022 at 04:02:10PM -0600, Robert Elliott wrote:
> This is an offshoot of the previous patch series at:
>   https://lore.kernel.org/linux-crypto/20221219202910.3063036-1-elliott@hpe.com
> 
> Add a kernel_fpu_yield() function for x86 crypto drivers to call
> periodically during long loops.
> 
> Test results
> ============
> I created 28 tcrypt modules so modprobe can run concurrent tests,
> added 1 MiB functional and speed tests to tcrypt, and ran three processes
> spawning 28 subprocesses (one per physical CPU core) each looping forever
> through all the tcrypt test modes. This keeps the system quite busy,
> generating RCU stalls and soft lockups during both generic and x86
> crypto function processing.
> 
> In conjunction with these patch series:
> * [PATCH 0/8] crypto: kernel-doc for assembly language
>   https://lore.kernel.org/linux-crypto/20221219185555.433233-1-elliott@hpe.com
> * [PATCH 0/3] crypto/rcu: suppress unnecessary CPU stall warnings
>   https://lore.kernel.org/linux-crypto/20221219202910.3063036-1-elliott@hpe.com
> * [PATCH 0/3] crypto: yield at end of operations
>   https://lore.kernel.org/linux-crypto/20221219203733.3063192-1-elliott@hpe.com
> 
> while using the default RCU values (60 s stalls, 21 s expedited stalls),
> several nights of testing did not result in any RCU stall warnings or soft
> lockups in any of these preemption modes:
>    preempt=none
>    preempt=voluntary
>    preempt=full
> 
> Setting the shortest possible RCU timeouts (3 s, 20 ms) did still result
> in RCU stalls, but only about one every 2 hours, and not occurring
> on particular modules like sha512_ssse3 and sm4-generic.
> 
> systemd usually crashes and restarts when its journal becomes full from
> all the tcrypt printk messages. Without the patches, that triggered more
> RCU stall reports and soft lockups; with the patches, only userspace
> seems perturbed.
> 

Where does this patch series apply to?

- Eric