Re: [RFCv1 0/8] RK3399 clean shutdown issue

From: Tobias Schramm
Date: Mon Dec 09 2019 - 09:52:10 EST


On Mon, Dec 9, 2019 at 8:29 AM Robin Murphy <robin.murphy@xxxxxxx> wrote:
On 06/12/2019 6:45 pm, Anand Moon wrote:
Most of the RK3399 SBC boards do not perform clean
shutdown and clean reboot.
FWIW reboot problems on RK3399 have been tracked down to issues in
upstream ATF, and are unrelated to the PMIC.

These patches try to help resolve the issue with proper
shutdown by turning off the PMIC.
As mentioned elsewhere[1], although this is what the BSP kernel seems to
do, and in practice it's unlikely to matter for the majority of devboard
users like you and me, I still feel a bit uncomfortable with this
solution for systems using ATF as in principle the secure world might
want to know about orderly shutdowns, and this effectively makes every
shutdown an unexpected power loss from secure software's point of view.

Since ATF is operating completely in volatile memory, and shouldn't be
touching hardware once it passes off control to the kernel anyways,
what is the harm of pulling the rug out from under it?
If this idea is to prevent issues in the future, such as if ATF does
gain the ability to preempt hardware control, then at that time ATF
will need to be able to handle actually powering off devices using the
same functionality.

As far as I know ATF implements PSCI, doesn't it? Thus I would assume that it should most definitely handle power off for all platforms as indicated by the presence of platform handlers in [1].

But as we discussed previously, ATF doesn't have this capability, so
in this case any board without a dedicated power-off gpio will be
unable to power off at all.
Also it seems that giving ATF this functionality, with the current
state of ATF, would be cost prohibitive.

I personally feel that allowing the kernel to do this is a solution to
the problem we have now.

Maybe I'm missing something here but I'd suggest that implementing an i2c driver in the rockchip platform part of ATF using libfdt to find the PMIC from the devicetree would be the way to go.