Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
In-Reply-To: <CAODwPW-9UPokepL_VVGuCLaWTDT7hmN9R9Z2yXABrP4g3_N4uQ@mail.gmail.com>
References: <20220726235115.1184532-1-jrosenth@chromium.org> <CAODwPW-9UPokepL_VVGuCLaWTDT7hmN9R9Z2yXABrP4g3_N4uQ@mail.gmail.com>
From:   Stephen Boyd <swboyd@chromium.org>
User-Agent: alot/0.10
Date:   Tue, 26 Jul 2022 20:40:56 -0500
Message-ID: <CAE-0n51RsN3P5hT+kApw-kPqjm_K7=7bvySWLNdQuQPnrQYZRw@mail.gmail.com>
Subject: Re: [PATCH] firmware: google: Implement vboot workbuf in sysfs
To:     Jack Rosenthal <jrosenth@chromium.org>,
        Julius Werner <jwerner@chromium.org>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        chrome-platform@lists.linux.dev,
        Guenter Roeck <groeck@chromium.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Quoting Julius Werner (2022-07-26 17:26:41)
> Sorry, this wasn't quite what I meant. I think we should definitely
> not hardcode any details about the vboot data structure layout here.
> It's already enough of a pain to keep crossystem in sync with
> different data structure versions, we absolutely don't want to have to
> keep updating that in the kernel as well.
>
> I assume that the reason you went that route seems to have been mostly
> that the lb_cbmem_ref coreboot table entry has no size field, so you
> had to infer the size from within the data structure. Thankfully, we
> don't need to use lb_cbmem_ref for this, that's somewhat of a legacy
> entry type that we're trying to phase out where it's no longer needed
> for backwards-compatibility anyway (and in fact I think we should be
> okay to remove the vboot entry there nowadays). We also have
> lb_cbmem_entry (see
> https://review.coreboot.org/plugins/gitiles/coreboot/+/refs/heads/master/src/lib/imd_cbmem.c#222)
> that exports info about every area in CBMEM, with address, size and
> CBMEM ID. The vboot workbuffer is CBMEM ID 0x78007343.
>
> I think we should just add a general driver for lb_cbmem_entry tags
> here, that uses the "id" as (part of) the device name and contains
> node to read/write the raw bytes of the buffer. Then crossystem can
> easily find the right one for vboot, and the infrastructure may also
> come in handy for other uses (or debugging) in the future.

If there's nothing to really "drive" then a driver is overkill. Would
exposing some raw bytes in /sys/firmware/coreboot be sufficient here,
possibly under some sort of /sys/firmware/coreboot/cbmem_entries/ path?

I honestly have no idea what vboot workbuffer is though so maybe I'm
just talking nonsense. If this ends up in sysfs please document the
files in Documentation/ABI/ and include it in the patch so we know what
is being exposed in the kernel ABI.