netdev - Re: [PATCH net-next v4] net: psample: Introduce stubs to remove NIC driver dependency

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: <20210208170735.GA207830@shredder.lan>
Date:   Mon, 8 Feb 2021 19:07:35 +0200
From:   Ido Schimmel <idosch@...sch.org>
To:     Leon Romanovsky <leon@...nel.org>
Cc:     Chris Mi <cmi@...dia.com>, Jakub Kicinski <kuba@...nel.org>,
        Cong Wang <xiyou.wangcong@...il.com>,
        Linux Kernel Network Developers <netdev@...r.kernel.org>,
        jiri@...dia.com, Saeed Mahameed <saeedm@...dia.com>,
        kernel test robot <lkp@...el.com>
Subject: Re: [PATCH net-next v4] net: psample: Introduce stubs to remove NIC
 driver dependency

On Mon, Feb 08, 2021 at 11:07:02AM +0200, Leon Romanovsky wrote:
> On Mon, Feb 08, 2021 at 10:57:46AM +0200, Ido Schimmel wrote:
> > On Mon, Feb 08, 2021 at 09:03:50AM +0200, Leon Romanovsky wrote:
> > > On Mon, Feb 01, 2021 at 08:08:37PM +0200, Ido Schimmel wrote:
> > > > On Mon, Feb 01, 2021 at 09:37:11AM +0800, Chris Mi wrote:
> > > > > Hi Ido,
> > > > >
> > > > > On 1/30/2021 10:42 PM, Ido Schimmel wrote:
> > > > > > On Fri, Jan 29, 2021 at 12:30:09PM -0800, Jakub Kicinski wrote:
> > > > > > > On Fri, 29 Jan 2021 14:08:39 +0800 Chris Mi wrote:
> > > > > > > > Instead of discussing it several days, maybe it's better to review
> > > > > > > > current patch, so that we can move forward :)
> > > > > > > It took you 4 revisions to post a patch which builds cleanly and now
> > > > > > > you want to hasten the review? My favorite kind of submission.
> > > > > > >
> > > > > > > The mlxsw core + spectrum drivers are 65 times the size of psample
> > > > > > > on my system. Why is the dependency a problem?
> > > > > > mlxsw has been using psample for ~4 years and I don't remember seeing a
> > > > > > single complaint about the dependency. I don't understand why this patch
> > > > > > is needed.
> > > > > Please see Saeed's comment in previous email:
> > > > >
> > > > > "
> > > > >
> > > > > The issue is with distros who ship modules independently.. having a
> > > > > hard dependency will make it impossible for basic mlx5_core.ko users to
> > > > > load the driver when psample is not installed/loaded.
> > > > >
> > > > > I prefer to have 0 dependency on external modules in a HW driver.
> > > > > "
> > > >
> > > > I saw it, but it basically comes down to personal preferences.
> > >
> > > It is more than personal preferences. In opposite to the mlxsw which is
> > > used for netdev only, the mlx5_core is used by other subsystems, e.g. RDMA,
> > > so Saeed's request to avoid extra dependencies makes sense.
> > >
> > > We don't need psample dependency to run RDMA traffic.
> >
> > Right, you don't need it. The dependency is "PSAMPLE || PSAMPLE=n". You
> > can compile out psample and RDMA will work.
> 
> So do you suggest to all our HPC users recompile their distribution kernel
> just to make sure that psample is not called?

I don't know. What are they complaining about? That psample needs to be
installed for mlx5_core to be loaded? How come the rest of the
dependencies are installed?

Or are they complaining about the size / memory footprint of psample?
Because then they should first check mlx5_core when all of its options
are blindly enabled as part of a distribution config.

AFAICS, mlx5 still does not have any code that uses psample. You can
wrap it in a config option and keep the weak dependency on psample.
Something like:

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
index ad45d20f9d44..d17d03d8cc8b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
+++ b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
@@ -104,6 +104,15 @@ config MLX5_TC_CT
 
          If unsure, set to Y
 
+config MLX5_TC_SAMPLE
+       bool "MLX5 TC sample offload support"
+       depends on MLX5_CLS_ACT
+       depends on PSAMPLE || PSAMPLE=n
+       default n
+       help
+         Say Y here if you want to support offloading tc rules that use sample
+          action.
+
 config MLX5_CORE_EN_DCB
        bool "Data Center Bridging (DCB) Support"
        default y

> 
> >
> > >
> > > >
> > > > >
> > > > > We are working on a tc sample offload feature for mlx5_core. The distros
> > > > > are likely to request us to do this. So we address it before submitting
> > > > > the driver changes.
> > > >
> > > > Which distros? Can they comment here? mlxsw is in RHEL and I don't
> > > > remember queries from them about the psample module.
> > >
> > > There is a huge difference between being in RHEL and actively work with
> > > partners as mlx5 does.
> > >
> > > The open mailing list is not the right place to discuss our partnership
> > > relations.
> >
> > I did not ask about "partnership relations". I asked for someone more
> > familiar with the problem that can explain the distro issue. But if such
> > a basic question can't be asked, then the distro argument should not
> > have been made in the first place.
> 
> It is not what you wrote, but if you don't want to take distro argument
> into account, please don't bring mlxsw either.
> 
> Thanks