linux-kernel - Re: [RFC 1/1] ARM: Drop !ARCH_MULTIPLATFORM dependency from Cortex A8/9 erratas

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] [day] [month] [year] [list]

Message-ID: <20221210025548.ssinevgkyjzf3622@mercury.elektranox.org>
Date:   Sat, 10 Dec 2022 03:55:48 +0100
From:   Sebastian Reichel <sebastian.reichel@...labora.com>
To:     Arnd Bergmann <arnd@...db.de>
Cc:     Russell King <linux@...linux.org.uk>,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        kernel@...labora.com, Tony Lindgren <tony@...mide.com>,
        Rob Herring <robh@...nel.org>
Subject: Re: [RFC 1/1] ARM: Drop !ARCH_MULTIPLATFORM dependency from Cortex
 A8/9 erratas

Hi,

On Fri, Dec 09, 2022 at 10:36:17PM +0100, Arnd Bergmann wrote:
> On Fri, Dec 9, 2022, at 19:45, Russell King (Oracle) wrote:
> > On Fri, Dec 09, 2022 at 07:20:00PM +0100, Sebastian Reichel wrote:
> >> All the erratas for more recent CPUs (Cortex A15/A12/A17) do not
> >> require !ARCH_MULTIPLATFORM, since there is runtime detection of
> >> effected CPUs in arch/arm/mm/proc-v7.S.
> >> 
> >> Errata config options for the older Cortex A8/A9 CPUs have the
> >> !ARCH_MULTIPLATFORM dependency, but the CPUs are also runtime
> >> detected. Since there is runtime detection, it should be fine
> >> to enable the erratas for a multi platform kernel.
> >> 
> >> Signed-off-by: Sebastian Reichel <sebastian.reichel@...labora.com>
> >> ---
> >> For me the most likely explanation that this has not yet happened is
> >> me missing an important detail, so I sent it as RFC. If you point it
> >> out I will try to document the reason.
> >
> > IIRC, the reason these are marked with !ARCH_MULTIPLATFORM is for when
> > the kernel is running in the non-secure world - the accessed registers
> > may prevent platforms booting. However, it was a long time ago and I
> > don't remember clearly now.
> 
> The git log confirms your memory, this is the description of
> commit 62e4d357aae0 ("ARM: 7609/1: disable errata work-arounds
> which access secure registers"):
> 
>     In order to support secure and non-secure platforms in multi-platform
>     kernels, errata work-arounds that access secure only registers need to
>     be disabled. Make all the errata options that fit in this category
>     depend on !CONFIG_ARCH_MULTIPLATFORM.
>     
>     This will effectively remove the errata options as platforms are
>     converted over to multi-platform.
>     
>     Signed-off-by: Rob Herring <rob.herring@...xeda.com>
>     Acked-by: Tony Lindgren <tony@...mide.com>
>     Signed-off-by: Russell King <rmk+kernel@....linux.org.uk>
> 
> As far as I understand, these are all boot time chicken bits that
> should already be set up by the boot loader before entering the
> kernel. 

Ah, I should have spent more time in the git history instead of
trying to search the mailing list :( I will propose a patch adding
a sentence to the config option helptexts, so that one does not need
to go through the git history to understand the requirement and
correct solution:

This errata workaround requires access to secure registers and thus
is not available on multiplatform kernels. The workaround should be
applied by the bootloader instead.

> Sebastian, are you aware of any particular machines that ship
> without the workaround code in the bootloader or in secure world?

No. I'm investigating sporadic GPU hungs on i.MX6QP on a customer
device and just noticed a couple of erratas missing in the .config.
They might already be enabled in the bootloader (and otherwise
definitely can be enabled there).

Thanks for the fast feedback!

-- Sebastian

Download attachment "signature.asc" of type "application/pgp-signature" (834 bytes)