linux-kernel - Re: [PATCH] block: elevator.c: Check elevator kernel argument again

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: <75c5681a-389f-cfaf-7f3f-31a2daec9da4@kernel.dk>
Date:   Sat, 13 Jul 2019 15:53:28 -0600
From:   Jens Axboe <axboe@...nel.dk>
To:     Marcos Paulo de Souza <marcos.souza.org@...il.com>,
        linux-kernel@...r.kernel.org
Cc:     "open list:BLOCK LAYER" <linux-block@...r.kernel.org>
Subject: Re: [PATCH] block: elevator.c: Check elevator kernel argument again

On 7/12/19 9:52 PM, Marcos Paulo de Souza wrote:
> Since the inclusion of blk-mq, elevator= kernel argument was not being
> considered anymore, making it impossible to specify a specific elevator
> at boot time as it was used before.
> 
> This is done by checking chosen_elevator global variable, which is
> populated once elevator= kernel argument is passed. Without this patch,
> mq-deadline is the only elevator that is can be used at boot time.
> 
> Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@...il.com>
> ---
> 
>   I found this issue while inspecting why noop scheduler was gone, and
>   so I found that was now impossible to use a scheduler different from
>   mq-deadeline.
> 
>   Am I missing something? Is this a desirable behavior?

Just google, I'm sure 2-3 discussions on this topic will come up.

tldr is that the original parameter was a mistake and doesn't work at
all for multiple devices. Today it's even worse, as we have device
types that won't even work properly with any scheduler, liked the
zoned devices. The parameter was never enabled for blk-mq because of
that, and hence died when the legacy IO path was scrapped. It's not
coming back.

-- 
Jens Axboe