I hadn't heard of UEFI memory disks before, so you're ahead of me there. If the memory is physically contiguous, and there's a way to get the base address and length through UEFI apis, the md driver changes should work, and we'd need UEFI loader changes to detect the memory, and hint it to the driver.

Thanks for the quick feedback. I've adjusted things a bit. Let me know what you think of the changes to the printfs in bios_memdisk.c; maybe they should just go away as they probably aren't very useful... and they're hard to see on a normal config anyway (they get output before the loader menu, the loader doesn't lets you scroll back to see them, and release images don't setup loader on the serial console).

address some of the comments

In case my comment on the bugzilla issue wasn't seen, I tested this patch (manually applied to 10.3), and it worked as expected. Thanks!

I've been running this patch (manually applied to 9.3-RELEASE) on one of our machines since Thursday, and all looks well. Prior to this issue, we were seeing retransmit events about once a day, so 4 days without an event looks successful.