On Thu, 1 Jul 2021, at 13:10, Joel Stanley wrote:

On Fri, 25 Jun 2021 at 06:10, Andrew Jeffery [off-list ref] wrote:

quoted

The degree values were reversed out from the magic tap values of 7 (in)
and 15 + inversion (out) initially suggested by Aspeed.

With the patch tacoma survives several gigabytes of reads and writes
using dd while without it locks up randomly during the boot process.

Signed-off-by: Andrew Jeffery <redacted>

Thanks for the fix. Is this required due to "mmc: sdhci-of-aspeed: Add
AST2600 bus clock support" or "mmc: sdhci-of-aspeed: Expose clock
phase controls"?

Sort of neither, it's really a bug with the devicetrees.

On the topic of those patches, it would be good if we could operate
the devices (with the slower speed?) when the device tree does not
provide the phase values. Think about system bringup, or where you
need the system booting in order to determine the phase calculations.

You can use the maximum-frequency binding to make things go slow enough 
to paper over phase issues. This helped us limp along early on.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/devicetree/bindings/mmc/mmc-controller.yaml?h=v5.13#n90

But really it depends on how bad the issues are at a given speed.

What changes would be required to the host driver for it to work out of the box?

Maybe the above is enough of a crutch?

Andrew