linux-kernel - Re: [PATCH v7 2/2] arm64: boot: Support Flat Image Tree

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]

Date:   Sat, 2 Dec 2023 09:36:43 -0700
From:   Simon Glass <sjg@...omium.org>
To:     Ahmad Fatoum <a.fatoum@...gutronix.de>,
        Doug Anderson <dianders@...omium.org>
Cc:     linux-arm-kernel@...ts.infradead.org,
        Masahiro Yamada <masahiroy@...nel.org>,
        Tom Rini <trini@...sulko.com>,
        lkml <linux-kernel@...r.kernel.org>,
        U-Boot Mailing List <u-boot@...ts.denx.de>,
        Catalin Marinas <catalin.marinas@....com>,
        Nathan Chancellor <nathan@...nel.org>,
        Nick Terrell <terrelln@...com>,
        Nicolas Schier <nicolas@...sle.eu>,
        Will Deacon <will@...nel.org>, linux-kbuild@...r.kernel.org,
        Pengutronix Kernel Team <kernel@...gutronix.de>
Subject: Re: [PATCH v7 2/2] arm64: boot: Support Flat Image Tree

Hi Ahmad,

On Thu, 30 Nov 2023 at 19:04, Ahmad Fatoum <a.fatoum@...gutronix.de> wrote:
>
> Hello Simon,
>
> On 30.11.23 21:30, Simon Glass wrote:
> > On Wed, 29 Nov 2023 at 12:54, Ahmad Fatoum <a.fatoum@...gutronix.de> wrote:
> >> On 29.11.23 20:44, Simon Glass wrote:
> >>> On Wed, 29 Nov 2023 at 12:33, Ahmad Fatoum <a.fatoum@...gutronix.de> wrote:
> >>>>
> >>>> On 29.11.23 20:27, Simon Glass wrote:
> >>>>> On Wed, 29 Nov 2023 at 12:15, Ahmad Fatoum <a.fatoum@...gutronix.de> wrote:
> >>>>>> On 29.11.23 20:02, Simon Glass wrote:
> >>>>>>> On Wed, 29 Nov 2023 at 11:59, Ahmad Fatoum <a.fatoum@...gutronix.de> wrote:
> >>>>>>>> The specification says that this is the root U-Boot compatible,
> >>>>>>>> which I presume to mean the top-level compatible, which makes sense to me.
> >>>>>>>>
> >>>>>>>> The code here though adds all compatible strings from the device tree though,
> >>>>>>>> is this intended?
> >>>>>>>
> >>>>>>> Yes, since it saves needing to read in each DT just to get the
> >>>>>>> compatible stringlist.
> >>>>>>
> >>>>>> The spec reads as if only one string (root) is supposed to be in the list.
> >>>>>> The script adds all compatibles though. This is not really useful as a bootloader
> >>>>>> that's compatible with e.g. fsl,imx8mm would just take the first device tree
> >>>>>> with that SoC, which is most likely to be wrong. It would be better to just
> >>>>>> specify the top-level compatible, so the bootloader fails instead of taking
> >>>>>> the first DT it finds.
> >>>>>
> >>>>> We do need to have a list, since we have to support different board revs, etc.
> >>>>
> >>>> Can you give me an example? The way I see it, a bootloader with
> >>>> compatible "vendor,board" and a FIT with configuration with compatibles:
> >>>>
> >>>>   "vendor,board-rev-a", "vendor,board"
> >>>>   "vendor,board-rev-b", "vendor,board"
> >>>>
> >>>> would just result in the bootloader booting the first configuration, even if
> >>>> the device is actually rev-b.
> >>>
> >>> You need to find the best match, not just any match. This is
> >>> documented in the function comment for fit_conf_find_compat().
> >>
> >> In my above example, both configuration are equally good.
> >> Can you give me an example where it makes sense to have multiple
> >> compatibles automatically extracted from the device tree compatible?
> >>
> >> The way I see it having more than one compatible here just has
> >> downsides.
> >
> > I don't have an example to hand, but this is the required mechanism of
> > FIT. This feature has been in place for many years and is used by
> > ChromeOS, at least.
>
> I see the utility of a FIT configuration with
>
>     compatible = "vendor,board-rev-a", "vendor,board-rev-b";
>
> I fail to see a utility for a configuration with
>
>     compatible = "vendor,board", "vendor,SoM", "vendor,SoC";
>
> Any configuration that ends up being booted because "vendor,SoC" was matched is
> most likely doomed to fail. Therefore, I would suggest that only the top level
> configuration is written into the FIT configurations automatically.

Firstly, I am not an expert on this.

Say you have a board with variants. The compatible string in U-Boot
may be something like:

"google,veyron-brain-rev1", "google,veyron-brain", "google,veyron",
"rockchip,rk3288";

If you then have several FIT configurations, they may be something like:

"google,veyron-brain-rev0", "google,veyron-brain", "google,veyron",
"rockchip,rk3288";
"google,veyron-brain-rev1", "google,veyron-brain", "google,veyron",
"rockchip,rk3288";
"google,veyron-brain-rev2", "google,veyron-brain", "google,veyron",
"rockchip,rk3288";

You want to choose the second one, since it is a better match than the others.

+Doug Anderson who knows a lot more about this than me.

Regards,
Simon