Hi, On Fri, Feb 22, 2019 at 10:48 AM David Summers <beagleboard@xxxxxxxxxxxxxxxxxxx> wrote: > > The Problem: > > On ASUS Tinker Board S, when booting from the eMMC, and there is no > card sd slot, then there are constant errors. > > Cause: > > Thanks must go to Robin Murphy @ ARM for idenifying the problem. The > rk808 on the Tinker Board and Tinker Board S has many regulators, one > vccio_sd powers the IO for the sd card. Unfortunatly this is also used > in the card detect. Now when no card is install, the regulator is > powered down. This means that the card detect floats, and this means > random card detection. Yeah, this is broken on a lot of SoCs that use dw_mmc. :( Really the card detect line needs to be on a different rail and this is why all boards I've worked on recently have a the card detect going to a GPIO instead of the dw_mmc CD. IIRC Rockchip moved the Card Detect to a different rail on newer SoCs (like rk3399) but we still used a GPIO even there since we didn't like the default/automatic muxing of JTAG and SD signals. The one board I was involved in that did it wrong (where we discovered this issue) was exynos5250-snow. You can see some discussion about the issue at: http://lists.infradead.org/pipermail/linux-arm-kernel/2014-August/282474.html In that discussion I'm pretty sure that Ulf suggested that a better way to go was to use something like "broken-cd" which I think was supposed to switch us to use polling. AKA periodically the SD card would be powered on and we'd check for a card, then we'd power everything off. ...but that was never implemented for snow at least so there may be something more than just adding the property. You can read through the whole thread for more details. IIRC leaving the IO rail always on like you're proposing can also work OK but there may be some corner cases, especially if you are trying to reach UHS speeds and/or if the bootloader ever tries to use UHS speeds. It's almost certainly busted if the bootloader did UHS since it will leave the line at ~1.8 V and the kernel will expect it to be at ~3.3 V. ...but maybe you rely on the bootloader not doing UHS and maybe things are generally OK if not? There may also be cases where you can't properly power down / reset a card because the card may be drawing power through the IO lines when you power off its main lines. That's not good for the card and can also put it in a bad state. I haven't done all the research here so this may be a bit of FUD--it's just a vague recollection from many years ago. ...so to make a long story short, a better solution is to allow the IO lines to be powered off but then poll for the card periodically. > The Solution: > > Make sure that the sd IO is always powered, this means card detection > is always active, which is what should be done on a board with an sd > slot, which both the Tinker Board and Tinker Board S are. Hence change > is made to the .dtsi which takes effect on all Tinker Boards as > required. > > The change also adds "regulator-boot-on" the Tinker Board boot from > uboot, and the sd card is always one option. Hence the IO must be > powered in uboot, and so setting this flag. > > Also removed is "disable-wp" the micro sd card which are used have no > write protection, so the concept doesn't mean anything, and the > Tinker Boards work without this. Hence it is removed to simply. As others have said, please leave disable-wp. There's no way for the kernel to know if you have a SD or uSD slot and the only difference between the two (electrically) is that there's no write protect for micro SD. Also: please CC dw_mmc people on future patches in this area. $ ./scripts/get_maintainer.pl -f drivers/mmc/host/dw_mmc.c Jaehoon Chung <jh80.chung@xxxxxxxxxxx> (maintainer:SYNOPSYS DESIGNWARE MMC/SD/SDIO DRIVER) Ulf Hansson <ulf.hansson@xxxxxxxxxx> (maintainer:MULTIMEDIA CARD (MMC), SECURE DIGITAL (SD) AND...) linux-mmc@xxxxxxxxxxxxxxx (open list:SYNOPSYS DESIGNWARE MMC/SD/SDIO DRIVER) -Doug
