[PATCH 0/6] mmc: Set PROBE_PREFER_ASYNCHRONOUS for all host drivers

[Ulf Hansson]

On Mon, 7 Sep 2020 at 06:08, Chen-Yu Tsai <wens at kernel.org> wrote:
>
> (Resent from kernel.org)
>
> On Mon, Sep 7, 2020 at 11:57 AM Anand Moon <linux.amoon at gmail.com> wrote:
> >
> > Hi Martin.
> >
> > On Sat, 5 Sep 2020 at 03:24, Martin Blumenstingl
> > <martin.blumenstingl at googlemail.com> wrote:
> > >
> > > Hi Douglas,
> > >
> > > On Fri, Sep 4, 2020 at 1:25 AM Douglas Anderson <dianders at chromium.org> wrote:
> > > >
> > > > As per discussion [1], it seems like it should be quite safe to turn
> > > > on PROBE_PREFER_ASYNCHRONOUS for all sd/mmc host controllers.  Let's
> > > > give it a shot.  For some discussion about this flag, see the commit
> > > > message for commit 3d3451124f3d ("mmc: sdhci-msm: Prefer asynchronous
> > > > probe").
> > > can this somehow change the order in which the MMC drivers end up loading?
> > > on Amlogic SoCs we have up to three MMC controllers, some SoCs even
> > > use two different MMC controller IPs (and therefore two different
> > > drivers).
> > > so far the MMC controller naming (/dev/mmcblk* etc.) was consistent -
> > > can that change with this patch?
> > >

Consistency has never been guaranteed. Just imagine one of the mmc
host drivers ending up lacking some of its resources during ->probe()
and returning -EPROBE_DEFER.

UUID/PARTID has been the only way.

> >
> > We could resolve this by setting up aliases for mmc nodes in the dts.

Yes, this is now possible due to the recently [1] applied patches for
supporting mmc aliases.

>
> Right now, only the dw_mmc driver family supports aliases for mmc nodes.

That's "mshc" specific (related to the old multiple slot support I
think), but not affecting the mmc aliases, which the mmc core now is
supporting.

So the mmc aliases are common for all mmc hosts. Please have a look at
the new OF parsing in mmc_alloc_host().

[...]

Kind regards
Uffe

[1]
https://patchwork.kernel.org/patch/11747669/
https://patchwork.kernel.org/patch/11747671/