Please review the patch "mmc: core: try to use an id from the devicetree " again

Please review the patch "mmc: core: try to use an id from the devicetree " again

[Johnson CH Chen (陳昭勳)]

Dear Ulf,

We have a requirement to set a specific mmc's disk slot for a root path when booting. I think a specific "slot" is different from a specific "device". For example, When a platform has four mmc/sd slots, and we have ten mmc/sd cards. We need to test some cards for booting in one slot, and other slots can be filled with one card each for another use. Actually the order of these four slots is not constant because it depends on the order of registering mmc/sd drivers, and different platforms may have different situations. Slot id can be related with block id, so it can be related with mmc id for current linux kernel, too. We need to set a specific slot id to satisfy our requirement.

I find the following patch: https://lkml.org/lkml/2019/6/20/638
This is Lubomir provides has a function we need. Actually I have a patch with similar function without mmc alias and just set "mmc-id" in device tree, but I think Lubomir's is better than myself.

Someone think why not use "LABEL" or "UUID" for initramfs, or use "PARTLABEL" or "PARTUUID" for u-boot? I think they assign a specific numbers/characters for mmc/sd cards, not for mmc/sd slots. Besides, if I set the same "LABEL" or "PARTLABEL" in some mmc/sd card, I think kernel will be confused how to find a desired mmc/sd card.

So kindly review the patch "mmc: core: try to use an id from the devicetree " if you are free, thanks!

Best regards,
Johnson

Moxa Inc.
Fl. 4, No. 135, Lane 235, Baoqiao Rd.
Xindian Dist., New Taipei City, Taiwan, R.O.C.
Tel : +886-2-8919-1230 ext.1075
JohnsonCH.Chen@moxa.com 

Re: Please review the patch "mmc: core: try to use an id from the devicetree " again

[Ulf Hansson]

On Thu, 19 Dec 2019 at 11:57, Johnson CH Chen (陳昭勳)
<JohnsonCH.Chen@moxa.com> wrote:
>
> Dear Ulf,
>
> We have a requirement to set a specific mmc's disk slot for a root path when booting. I think a specific "slot" is different from a specific "device". For example, When a platform has four mmc/sd slots, and we have ten mmc/sd cards. We need to test some cards for booting in one slot, and other slots can be filled with one card each for another use. Actually the order of these four slots is not constant because it depends on the order of registering mmc/sd drivers, and different platforms may have different situations. Slot id can be related with block id, so it can be related with mmc id for current linux kernel, too. We need to set a specific slot id to satisfy our requirement.
>
> I find the following patch: https://lkml.org/lkml/2019/6/20/638
> This is Lubomir provides has a function we need. Actually I have a patch with similar function without mmc alias and just set "mmc-id" in device tree, but I think Lubomir's is better than myself.
>
> Someone think why not use "LABEL" or "UUID" for initramfs, or use "PARTLABEL" or "PARTUUID" for u-boot? I think they assign a specific numbers/characters for mmc/sd cards, not for mmc/sd slots. Besides, if I set the same "LABEL" or "PARTLABEL" in some mmc/sd card, I think kernel will be confused how to find a desired mmc/sd card.
>
> So kindly review the patch "mmc: core: try to use an id from the devicetree " if you are free, thanks!

There have been several attempts to fix this and me personally don't
have a problem with using aliases/labels in DT. I said that before as
well.

However, what is missing today, is that someone collects the proper
arguments for *why* this is needed (and especially why UUID/LABEL
doesn't work). I think all that information is available in the
discussion here: https://lore.kernel.org/patchwork/cover/674381/

When I see a new patch posted, that contains these arguments in the
commit message of the patch, I am ready to review it again.

Kind regards
Uffe