[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAGSQo03R+HYcX2JJDSm7LA8V370s_3hrbTBc2s51Pp9nxWTz5w@mail.gmail.com>
Date: Tue, 9 Jul 2024 10:06:44 -0700
From: Matthew Maurer <mmaurer@...gle.com>
To: Sami Tolvanen <samitolvanen@...gle.com>
Cc: Luis Chamberlain <mcgrof@...nel.org>, Petr Mladek <pmladek@...e.com>, Gary Guo <gary@...yguo.net>,
Masahiro Yamada <masahiroy@...nel.org>, Michal Suchánek <msuchanek@...e.de>,
Lucas De Marchi <lucas.demarchi@...el.com>, Andreas Hindborg <nmi@...aspace.dk>,
Josh Poimboeuf <jpoimboe@...nel.org>, Miroslav Benes <mbenes@...e.cz>, Song Liu <song@...nel.org>,
live-patching@...r.kernel.org, linux-kernel@...r.kernel.org, jikos@...nel.org,
joe.lawrence@...hat.com, nathan@...nel.org, morbo@...gle.com,
justinstitt@...gle.com, thunder.leizhen@...wei.com, kees@...nel.org,
kernel-team@...a.com
Subject: Re: [PATCH] kallsyms, livepatch: Fix livepatch with CONFIG_LTO_CLANG
On Mon, Jul 8, 2024 at 5:07 PM Sami Tolvanen <samitolvanen@...gle.com> wrote:
>
> On Mon, Jul 8, 2024 at 2:33 PM Luis Chamberlain <mcgrof@...nel.org> wrote:
> >
> > Looking at this again its not to me why Masahiro Yamada's suggestion on
> > that old patch series to just increase the length and put long symbols
> > names into its own section [0] could not be embraced with a new kconfig
> > option, so new kernels and new userspace could support it:
> >
> > https://lore.kernel.org/lkml/CAK7LNATsuszFR7JB5ZkqVS1W=hWr9=E7bTf+MvgJ+NXT3aZNwg@mail.gmail.com/
>
> Matt, was there a reason we didn't move forward with Masahiro's
> proposal? It sounds reasonable to me, but I might be missing some
> background here.
I liked Masahiro's proposal, and implemented it [1], but in the first
version of the patch, folks complained [2] that I was touching code
that was hard to reason about, and that I should clean it up. In the
second version of the patch, I cleaned it up, and was told the cleanup
patch was too big and needed to be broken down [3].
If we want to revive that, it appears there's one or two PPC userspace
tool issues I need to address first, but I could do so, and attempt to
break the cleanup patch into smaller pieces to be more reviewable. I
had abandoned it because even if completed, without the new work Sami
did on extracting a hash from debug symbols, it would not have been
accepted as a way to enable MODVERSIONS. When we generated this new
approach, I suggested we do hashes to avoid modifying the module
loading code given the previous reaction.
[1]: https://lore.kernel.org/all/20231118025748.2778044-1-mmaurer@google.com/
[2]: https://lore.kernel.org/all/ZVZNh%2FPA5HiVRkeb@bombadil.infradead.org/
[3]: https://lore.kernel.org/all/2023111818-agent-verdict-99a5@gregkh/
Powered by blists - more mailing lists