> On Tue, Apr 6, 2021 at 12:52 AM Bjorn Andersson > <bjorn.andersson@xxxxxxxxxx> wrote: > > On Mon 05 Apr 15:02 CDT 2021, Petr Vorel wrote: > > > Simple testing with /sys/class/gpio/export showed that 85-88. > > > 3 disables UART. I expect 0-2 are also reserved as on other msm8998. > > Are you saying that once you export these gpios the uart stops working? > That might be a Fixes: but certainly not a regression. > Using GPIO sysfs access is dangerous and if someone enables it into > their kernel (which requires setting CONFIG_EXPERT) they are > certainly aware that they are taking a risk. (Same goes for using > the character device from userspace.) Not sure if we understand each other. You might think I export GPIO from userspace on mainline kernel or use CONFIG_EXPERT. None of these is true. The real problem is that mainline kernel compiled with defconfig resets really early, thus v2 of this patch [2] is needed. I just exported GPIO on *downstream* kernel to observe the behaviour. It was just a simple way for me to find GPIOs which cause reset. (having docs would be great) And there is similar reset much later but still before reaching initramfs in loop_init since a99163e9e708 (during loop_add() on lo->lo_number: 3). And if I disable CONFIG_BLK_DEV_LOOP, it resets right after reaching initramfs. I guess that's another problem with DT, but haven't got much further. But that's a separate problem, gpio-reserved-ranges = <85 4> on tlmm [2] is still some improvement. Kind regards, Petr [2] https://lore.kernel.org/linux-arm-msm/20210406202936.22500-1-petr.vorel > Yours, > Linus Walleij
