Re: [RFC PATCH 0/3] fastboot: sunxi: Determine MMC device at runtime

From: Maxime Ripard <maxime@cerno.tech>
To: Sean Anderson <sean.anderson@seco.com>
Cc: Andre Przywara <andre.przywara@arm.com>,
	Simon Glass <sjg@chromium.org>,
	Jagan Teki <jagan@amarulasolutions.com>,
	u-boot@lists.denx.de, linux-sunxi@lists.linux.dev,
	Patrick Delaunay <patrick.delaunay@foss.st.com>,
	Heiko Schocher <hs@denx.de>,
	Kever Yang <kever.yang@rock-chips.com>,
	Philipp Tomsich <philipp.tomsich@vrull.eu>
Subject: Re: [RFC PATCH 0/3] fastboot: sunxi: Determine MMC device at runtime
Date: Mon, 24 May 2021 17:28:31 +0200	[thread overview]
Message-ID: <20210524152831.unrg7zexsdbz6uki@gilmour> (raw)
In-Reply-To: <f8fa6aca-7525-355f-ec77-26819e4c785d@seco.com>

[-- Attachment #1: Type: text/plain, Size: 2302 bytes --]

On Mon, May 24, 2021 at 10:37:31AM -0400, Sean Anderson wrote:
> 
> 
> On 5/23/21 8:36 PM, Andre Przywara wrote:
> > At the moment the fastboot code relies on the Kconfig variable
> > CONFIG_FASTBOOT_FLASH_MMC_DEV to point to the MMC device to use for the
> > flash command. This value needs to be the *U-Boot device number*, which
> > is picked by the U-Boot device model at runtime. This makes it quite
> > tricky and fragile to fix this variable at compile time, as other DT
> > nodes and aliases influence the enumeration process.
> >
> > To make this process more robust, allow the device number to be picked at
> > runtime, which sounds like a better fit to find this device number. Patch
> > 1/3 introduces a weak function for that purpose.
> > Patch 2/3 then implements this function for the Allwinner platform. The
> > code follows the idea behind the existing Kconfig defaults: Use the eMMC
> > device, if that exists, or the SD card otherwise. This patch is actually
> > not sunxi specific, so might be adopted by other platforms as well.
> > Patch 3/3 then drops the existing Kconfig defaults for sunxi, to clean
> > this up and remove the implicit assumption that the eMMC device is always
> > device 1 (as least for the fastboot code).
> >
> > I would be curious if others think this is the right way forward.
> > The fact that the U-Boot device numbers are determined at runtime
> > seems to conflict with the idea of a Kconfig variable in the first place,
> > hence this series. This brings us one step closer to the end goal of
> > removing the "eMMC is device 1" assumption.
> 
> I would actually favor removing CONFIG_FASTBOOT_FLASH_MMC_DEV
> altogether, and just specifying the device explicitly in fastboot
> commands. If you need to dynamically change the device, you can create
> some aliases. E.g. you could have something like
> 
> "fastboot_aliases=setenv fastboot_partition_alias_user ${mmcdev}.0:0"
> 
> and then run this variable just before calling `fastboot 0` (or whatever
> your usb device is).

If we're going that way, maybe we can just pass the interface on the
command line like dfu does? That way the new requirement would be very
obvious instead of introducing a new environment variable no one really
expects?

Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]