[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250610064316.JbrCJTuL@linutronix.de>
Date: Tue, 10 Jun 2025 08:43:16 +0200
From: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To: Gal Pressman <gal@...dia.com>
Cc: "David S. Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>,
Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>,
Andrew Lunn <andrew+netdev@...n.ch>, netdev@...r.kernel.org,
Aaron Conole <aconole@...hat.com>,
Eelco Chaudron <echaudro@...hat.com>,
Ilya Maximets <i.maximets@....org>, Simon Horman <horms@...nel.org>,
Clark Williams <clrkwllms@...nel.org>,
Steven Rostedt <rostedt@...dmis.org>, dev@...nvswitch.org,
linux-rt-devel@...ts.linux.dev
Subject: Re: [PATCH net 0/3] Revert openvswitch per-CPU storage
On 2025-06-10 09:26:28 [+0300], Gal Pressman wrote:
> This patch series reverts a set of changes that consolidated per-CPU
> storage structures in the openvswitch module.
>
> The original changes were intended to improve performance and reduce
> complexity by merging three separate per-CPU structures into one, but
> they have changed openvswitch to use static percpu allocations, and
> exhausted the reserved chunk on module init.
> This results in allocation of struct ovs_pcpu_storage (6488 bytes)
> failure on ARM.
>
> The reverts are applied in reverse order of the original commits.
Is the limited per-CPU storage the only problem? If so I would towards a
different solution rather than reverting everything.
I assume a defconfig on arm is everything that is needed to reproduce
this.
Sebastian
Powered by blists - more mailing lists