Card detection very unreliable due to physical dimensions

[marcowidmer]

In our projects, we use the DM3AT microSD Card socket. It has a simple physical switch to detect the card:

Unfortunately, on the USBSDMux (both "classic" and "fast" variants), the PCB corner which triggers the card detection switch is cut off:

This results in very unreliable card detection, as it depends on the exact insertion angle. If not inserted perfectly, U-Boot reports MMC: no card present and fails to boot the kernel.

[SmithChart]

Hey @marcowidmer ,
thanks for reaching out!

The chamfer on the USB-SD-Mux outline (Both Classic and FAST) were added in an early prototype to ease the insertion into some Micro-SD sockets.
Thanks for the input that this feature is causing these issues for you. We will consider reduce or remove the chamfer in a future version.

For now I can only image two workarounds:

• We have published the STEP-File of the device. I know of customers that build custom jigs to hold the USB-SD-Mux firmly in place.
• At least on Linux you can disable the card detect in the devicetree using the broken-cd attribute.

(Interestingly: We are using the same DM3AT-Socket on our USB-SD-Muxes. But on the SD-Mux the card detect is not connected to anything. So we never noticed this issue.)

[a3f]

At least on Linux you can disable the card detect in the devicetree using the broken-cd attribute.

I think you'll have to drop the cd-gpios in the same node as well.

But interesting to know this is by design (so to speak). I also wondered why we had flaky card detection on some boards and I settled on adding a broken_cd device parameter in barebox that just applies the described device tree fixup to the kernel device tree automatically on boot.

And then we just store it into the environment once as needed.