netdev - Re: [RFC PATCH net-next 0/9] Userspace spinning on net-sysfs access

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: <163350719413.4226.2526174755566600987@kwain>
Date:   Wed, 06 Oct 2021 09:59:54 +0200
From:   Antoine Tenart <atenart@...nel.org>
To:     Michal Hocko <mhocko@...e.com>
Cc:     davem@...emloft.net, kuba@...nel.org, pabeni@...hat.com,
        gregkh@...uxfoundation.org, ebiederm@...ssion.com,
        stephen@...workplumber.org, herbert@...dor.apana.org.au,
        juri.lelli@...hat.com, netdev@...r.kernel.org,
        Jiri Bohac <jbohac@...e.cz>
Subject: Re: [RFC PATCH net-next 0/9] Userspace spinning on net-sysfs access

Quoting Michal Hocko (2021-10-06 08:37:47)
> On Tue 28-09-21 14:54:51, Antoine Tenart wrote:
> thanks for posting this. Coincidentally we have come across a similar
> problem as well just recently.
> 
> > What made those syscalls to spin is the following construction (which is
> > found a lot in net sysfs and sysctl code):
> > 
> >   if (!rtnl_trylock())
> >           return restart_syscall();
> 
> One of our customer is using Prometeus (https://github.com/prometheus/prometheus)
> for monitoring and they have noticed that running several instances of
> node-exporter can lead to a high CPU utilization. After some
> investigation it has turned out that most instances are busy looping on
> on of the sysfs files while one instance is processing sysfs speed file
> for mlx driver which performs quite a convoluted set of operations (send
> commands to the lower layers via workqueues) to talk to the device to
> get the information.
> 
> The problem becomes more visible with more instance of node-exporter
> running at parallel. This results in some monitoring alarms at the said
> machine because the high CPU utilization is not expected.
> 
> I would appreciate if you CC me on next versions of this patchset.

Sure, will do!

Nice to see this can help others. Any help on (extensively) testing is
welcomed :-)

Thanks,
Antoine