MMC order inconsistency

[hansemro]

Problem: eMMC starts up sometimes as mmcblk0, mmclbk1, or even mmcblk2. Because the kernel boot images assume a target location, the kernel can sometimes fail to find root.

The issue is made worse or caused by enabling MMC5 for WLAN.

I have recently read a patch and the discussion of this inconsistent behavior on OMAP mailing list: https://www.spinics.net/lists/linux-omap/msg159229.html

Patch TLDR: set aliases for each mmc device in the device tree to fix their order.

• [x] Confirm consistent behavior with patch

[mj-sakellaropoulos]

Seems like aliases are already present in omap4.dtsi :

...
/ {
    compatible = "ti,omap4430", "ti,omap4";
    interrupt-parent = <&wakeupgen>;
    #address-cells = <1>;
    #size-cells = <1>;
    chosen { };

    aliases {
        i2c0 = &i2c1;
        i2c1 = &i2c2;
        i2c2 = &i2c3;
        i2c3 = &i2c4;
        mmc0 = &mmc1;
        mmc1 = &mmc2;
        mmc2 = &mmc3;
        mmc3 = &mmc4;
        mmc4 = &mmc5;
        serial0 = &uart1;
        serial1 = &uart2;
        serial2 = &uart3;
        serial3 = &uart4;
        rproc0 = &dsp;
        rproc1 = &ipu;
    };
        ...

[hansemro]

They were not present in <5.11.x, but it is good to know that it has been added to omap4

[mj-sakellaropoulos]

I should clarify I experienced the issue with those aliases

[hansemro]

Well, the issue is that &mmc2 should be aliased to mmc0 for it to be consistent with how things were. So with the new defaults, you would have to adjust root parameter to point to dev/mmcblk1p9 instead of /dev/mmcblk0p9

[mj-sakellaropoulos]

Oh, ok that makes sense.

[hansemro]

The aliases seem to fix the mapping of mmc block devices. Closing the issue.