linux-kernel - Re: [PATCH] irqchip/qcom-irq-combiner: Rename driver struct to end in

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: <53ugz6qg3rfgpcokrre6sqfr73tdjwiz67dvgvlnbiu6kkyc5u@bapjxdheqrzt>
Date: Tue, 1 Jul 2025 15:58:35 +0200
From: Uwe Kleine-König <u.kleine-koenig@...libre.com>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: Agustin Vega-Frias <agustinv@...eaurora.org>, 
	Marc Zyngier <maz@...nel.org>, linux-arm-msm@...r.kernel.org, linux-kernel@...r.kernel.org, 
	Markus Schneider-Pargmann <msp@...libre.com>, Masahiro Yamada <masahiroy@...nel.org>, 
	Nathan Chancellor <nathan@...nel.org>, Nicolas Schier <nicolas.schier@...ux.dev>, 
	linux-kbuild@...r.kernel.org
Subject: Re: [PATCH] irqchip/qcom-irq-combiner: Rename driver struct to end
 in _driver

On Tue, Jul 01, 2025 at 11:01:18AM +0200, Thomas Gleixner wrote:
> On Mon, Jun 30 2025 at 21:40, Uwe Kleine-König wrote:
> > On Mon, Jun 30, 2025 at 08:01:53PM +0200, Thomas Gleixner wrote:
> >> On Mon, Jun 30 2025 at 19:23, Uwe Kleine-König wrote:
> >> > The modpost section mismatch checks are more lax for objects that have a
> >> > name that ends in "_probe". This is not justified here though, so rename
> >> 
> >> That's a truly bad design or lack of such.
> >> 
> >> Why can't this muck use foo_driver(name) foo_probe(name) annotations to
> >> make it entirely clear what is tested for instead of oracling it out of
> >> the name itself. That would make it too easy to understand and analyse.
> >
> > I don't understand what you're suggesting here. Either I got it wrong or
> > it is insufficient because every object is checked, not only the driver
> > structs. That would result in more exceptions/special cases than we have
> > now.
> >
> > Anyhow, I agree that depending on the name is unfortunate, maybe we can
> > come up with something more clever?
> 
> That's what I was referring to. Doing checks based on struct names is a
> bad idea. Having distinct '...driver_probe(name)' and ...driver(name)'
> macros to distinguish the functionality is the proper thing to do and
> way simpler to analyse than names.

A driver struct should have no reference to .init.text (i.e. no callback
to a function marked with __init) no matter if it is registered using
module_platform_driver_probe() or module_platform_driver(). But even if
the requirements for those were different, how do you signal in the
binary if the driver was registered using the normal (i.e
platform_driver_register()) or the platform_driver_probe() way? Or do
you want to check the source file?

And note that if you have that, you covered only platform drivers, with
a bit of luck mostly all drivers (note that console drivers are special
and are allowed to have an __init callback). modpost checks all objects,
not only driver structs.

Best regards
Uwe

Download attachment "signature.asc" of type "application/pgp-signature" (489 bytes)