[PATCH v2] arm64: dts: allwinner: h6: add eMMC voltage property for Beelink GS1

[Maxime Ripard]

On Thu, Oct 08, 2020 at 10:00:06PM +0200, Clément Péron wrote:
> Hi Maxime,
> 
> Adding linux-sunxi and Jernej Skrabec to this discussion.
> 
> On Thu, 8 Oct 2020 at 17:10, Maxime Ripard <maxime at cerno.tech> wrote:
> >
> > Hi Clément,
> >
> > On Mon, Oct 05, 2020 at 08:47:19PM +0200, Clément Péron wrote:
> > > On Mon, 5 Oct 2020 at 11:21, Maxime Ripard <maxime at cerno.tech> wrote:
> > > >
> > > > Hi Clément,
> > > >
> > > > On Sat, Oct 03, 2020 at 11:20:01AM +0200, Clément Péron wrote:
> > > > > Sunxi MMC driver can't distinguish at runtime what's the I/O voltage
> > > > > for HS200 mode.
> > > >
> > > > Unfortunately, that's not true (or at least, that's not related to your patch).
> > > >
> > > > > Add a property in the device-tree to notify MMC core about this
> > > > > configuration.
> > > > >
> > > > > Fixes: 089bee8dd119 ("arm64: dts: allwinner: h6: Introduce Beelink GS1 board")
> > > > > Signed-off-by: Clément Péron <peron.clem at gmail.com>
> > > > > ---
> > > > >  arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts | 1 +
> > > > >  1 file changed, 1 insertion(+)
> > > > >
> > > > > diff --git a/arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts b/arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts
> > > > > index 049c21718846..3f20d2c9bbbb 100644
> > > > > --- a/arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts
> > > > > +++ b/arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts
> > > > > @@ -145,6 +145,7 @@ &mmc2 {
> > > > >       vqmmc-supply = <&reg_bldo2>;
> > > > >       non-removable;
> > > > >       cap-mmc-hw-reset;
> > > > > +     mmc-hs200-1_8v;
> > > > >       bus-width = <8>;
> > > > >       status = "okay";
> > > > >  };
> > > >
> > > > I'm not really sure what you're trying to fix here, but as far as MMC
> > > > goes, eMMC's can support io voltage of 3.3, 1.8 and 1.2V. Modes up until
> > > > HS DDR (50MHz in DDR) will use an IO voltage of 3.3V, higher speed modes
> > > > (HS200 and HS400) supporting 1.8V and 1.2V.
> > >
> > > Some users report that the eMMC is not working properly on their
> > > Beelink GS1 boards.
> > >
> > > > The mmc-hs200-1_8v property states that the MMC controller supports the
> > > > HS200 mode at 1.8V. Now, I can only assume that since BLDO2 is set up at
> > > > 1.8V then otherwise, the MMC core will rightfully decide to use the
> > > > highest supported mode. In this case, since the driver sets it, it would
> > > > be HS-DDR at 3.3V, which won't work with that fixed regulator.
> > > >
> > > > I can only assume that enabling HS200 at 1.8V only fixes the issue you
> > > > have because otherwise it would use HS-DDR at 3.3V, ie not actually
> > > > fixing the issue but sweeping it under the rug.
> > > >
> > > > Trying to add mmc-ddr-1_8v would be a good idea
> > >
> > > Thanks for the explanation, this is indeed the correct one.
> > > So It looks like the SDIO controller has an issue on some boards when
> > > using HS-DDR mode.
> > >
> > > Is this patch acceptable with the proper commit log?
> >
> > If HS-DDR works, yes, but I assume it doesn't?
> 
> After discussing with Jernej about this issue, I understood that:
> - Automatic delay calibration is not implemented
> - We also miss some handling of DDR related bits in control register
> 
> So none of H5/H6 boards should actually work.
> (Some 'lucky' boards seem to work enough to switch to HS200 mode...)
> 
> To "fix" this the H5 disable the HS-DDR mode in sunxi mmc driver :
> https://github.com/torvalds/linux/blob/master/drivers/mmc/host/sunxi-mmc.c#L1409

I find it suspicious that some boards would have traces not good enough
for HS-DDR (50MHz in DDR) but would work fine in HS200 (200MHz in SDR).
If there's some mismatch on the traces, it will only be worse in HS200.

And for the delay calibration, iirc, that's only necessary for HS400
that we don't support?

> I'm not sure about A64 but it looks like the property "mmc-hs200-1_8v"
> for the PineBook shows the same issue.
> 
> The proper way would of course be to implement the missing feature
> mentioned above.
> But this could take some time and as the eMMC driver is actually
> broken wouldn't it be better to disable the HS-DDR for H6 in the mmc
> driver like it's done for H5 ?

Have you tested with only the mmc-ddr-1_8v property?

Maxime
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20201009/5540f74e/attachment.sig>