This looks good to me, I'm still somewhat puzzled, though, as to why the problem has only be encountered with md(4) devices. Was that simply due to the fact that we happened to test with disks that DISKFLAG_CANDELETE doesn't get set for?

Well, if you look at mmc_discover_cards(), you'll see that mmc(4) in fact will enumerate all MMC devices on a bus and assign different addresses to them:

while (1) {

<...>

ivar->rca = rca++;
err = mmc_set_relative_addr(sc, ivar->rca);

<...>

Well, to get en par with the eMMC partition functionality in mmc(4), there still needs to be code added which actually does something with the extended user data area information and a way to create partitions. As for the later, adding such support to camcontrol(8) probably would be the more natural thing to do with CAM. But given that partitioning eMMC devices is a one-time-operation, I'd strongly suggest to just also add a Linux-compatible MMC IOCTL interface so mmc-utils can be continued to be used unless you want to risk to brick a couple of eMMC chips (besides, there are tutorials how to partition with mmc-utils). I'm okay with this patch getting committed as it currently is, though.
Btw., at least currently I don't seem to be able to access your list of issues; even after multiple attempts to reload that page and besides waiting for several minutes, all Firefox is displaying are the spinning cats waiting/"progress" indicators where the lists should go ...

Again better; at this point, I'm mainly concerned about:

• Handling of the timeout provided in mmc_switch_fill_mmcio(); partly this needs to happen in sdhci(4), i. e. SDHCI_TIMEOUT_CONTROL needs to be set accordingly, but a counterpart to mmc_switch() of mmc_subr.c should live in CAM.
• Handling of suspend/resume in sdda(4) but also already starting in sdhci(4) in the MMCCAM case.
• Usage of cpi.maxio for disk->d_maxsize with the former set to an incorrect value by sdhci(4) (which obviously is an sdhci(4) bug, though).

In D13063#271753, @kibab wrote:

RCA = 2 is just because I need to send _some_ RCA to the card. So I chose to send 2, could also choose any other number :-)

Generally better, but still quite a few TODOs.
My CAM knowledge is rusty and I've also mainly worked on SIM drivers before so someone vivid in CAM needs to review the CAM-specific parts. The way the commands for switching partitions are inserted/handled may be correct, but at least seems somewhat fishy to me.

Well, IMO committing the partition identification code only with all the glory details yet to be done is premature, even for head.
Generally: Please adhere to style(9).