[PATCH] Convert CONFIG_SYS_MMC_ENV_DEV et al to Kconfig

[David Woodhouse]

On 5 August 2020 14:14:58 BST, Tom Rini <trini at konsulko.com> wrote:
>On Wed, Aug 05, 2020 at 02:04:02PM +0100, David Woodhouse wrote:
>> 
>> 
>> On 5 August 2020 13:51:43 BST, Tom Rini <trini at konsulko.com> wrote:
>> >On Wed, Aug 05, 2020 at 11:07:15AM +0200, Stefan Roese wrote:
>> >> Hi Rasmus,
>> >> 
>> >> On 05.08.20 10:47, Rasmus Villemoes wrote:
>> >> > On 24/07/2020 23.14, Tom Rini wrote:
>> >> > > This converts the following to Kconfig:
>> >> > >     CONFIG_SYS_MMC_ENV_DEV
>> >> > >     CONFIG_SYS_MMC_ENV_PART
>> >> > 
>> >> > Yes, please! That's the only thing preventing us from running a
>> >vanilla
>> >> > upstream U-Boot on a number of our boards where we have the
>> >defconfig
>> >> > out of tree - we currently have a single patch touching a few
>> >config
>> >> > headers.
>> >> > 
>> >> > > Cc: Priyanka Jain <priyanka.jain at nxp.com>
>> >> > > Cc: Madalin Bucur <madalin.bucur at oss.nxp.com>
>> >> > > Signed-off-by: Tom Rini <trini at konsulko.com>
>> >> > > ---
>> >> > > Due to some non-environment usage of the variables, cc'ing a
>few
>> >folks.
>> >> > 
>> >> > Yes, and those are the reason I didn't already send a patch.
>> >> > 
>> >> > > diff --git a/env/Kconfig b/env/Kconfig
>> >> > > index 38e7fadbb93a..2eb2fe48da3d 100644
>> >> > > --- a/env/Kconfig
>> >> > > +++ b/env/Kconfig
>> >> > > @@ -567,6 +567,25 @@ config SYS_RELOC_GD_ENV_ADDR
>> >> > >   	  Relocate the early env_addr pointer so we know it is not
>> >inside
>> >> > >   	  the binary. Some systems need this and for the rest, it
>> >doesn't hurt.
>> >> > > +config SYS_MMC_ENV_DEV
>> >> > > +	int "mmc device number"
>> >> > > +	depends on ENV_IS_IN_MMC || ENV_IS_IN_FAT ||
>> >SYS_LS_PPA_FW_IN_MMC || \
>> >> > > +		CMD_MVEBU_BUBT || FMAN_ENET || QE
>> >> > 
>> >> > Slight yuck. But I suppose that those (ab)uses of
>> >CONFIG_SYS_MMC_ENV_DEV
>> >> > can grow their own proper config symbols piecemeal and then get
>> >removed
>> >> > from that list, rather than having to wean off all of them
>before
>> >> > SYS_MMC_ENV_* can get converted.
>> >> > 
>> >> > Why is ENV_IS_IN_FAT in that list? That has its own
>> >already-existing
>> >> > ENV_FAT_DEVICE_AND_PART.
>> >
>> >ENV_IS_IN_FAT is in the list because with 6731bef6966e ("env/fat.c:
>> >allow loading from a FAT partition on the MMC boot device") we can
>pick
>> >up the device dynamically and hard-code the partition.  But that in
>> >turn
>> >means needing a default device.  This even gets expanded to cover
>ext4
>> >here:
>>
>>http://patchwork.ozlabs.org/project/uboot/patch/20200804090547.2175076-2-dwmw2@infradead.org/
>> 
>> 
>> FAT and ext4 don't need to grow their own config symbols because they
>already *have* them. The only reason they're involved here is because
>of the case where they explicitly want to *abdicate* responsibility and
>let platform code determine the device to use, not hard-coded config.
>Like the MT7623 platforms where the preloader tells U-Boot if it was
>actually loaded from the internal eMMC or external SD card.
>> 
>> If someone is setting up the ENV_FAT_DEVICE_AND_PART to start with a
>colon and thus abdicate the device part, but then that decision is only
>coming from a hard-coded SYS_MMC_ENV_DEV option, then they're probably
>somewhat confused.
>
>Right, but the default mmc_get_env_dev() you added uses
>CONFIG_SYS_MMC_ENV_DEV.

Perhaps I could revamp the FAT/ext4 patches so that they don't use the default weak function, and if you use the "get it at runtime" behaviour on any build which doesn't actually have a real platform mmc_get_env_dev() function, the build fails.  That would be sane, but might involve a separate explicit config option instead of just "the DEVICE_AND_PART option starts with a colon".

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.