On Wed, 2018-05-09 at 01:13 +0000, Peng Fan wrote:
+0800, Anson Huang wrote:
Add i.MX7 PSCI system reset support, linux kernel now can use "reboot" command to reset system.
+__secure void imx_system_reset(void) {
- writew(1 << 2, WDOG1_BASE_ADDR); }
This does not work properly on our board.
You could try write 0x14 to WDOG1_BASE_ADDR to see whether it works. Here using (1 << 2) triggers both SRS and WDOG_B which seems wrong.
It works when the Linux driver does that and appears to work identically when uboot does it that way.
Triggering both SRS and WDOG_B does not work as it causes a very short wdog_b assertion.
Yes.
In case it wasn't clear, this code is in the new 2018.05 u-boot release and I think will cause a problem for any imx7d board using the typical reset erratum option of connecting wdog_b to a pfuze3000 pmic's pwrwon pin.
An external watchdog monitor chip, I think as used on sabreD for instance, might still respond to the short pulse and allow it work.
There is a little difficulty here in just changing the code to use WCR_WDE|WCR_SRS, as that would be wrong if the board does not use an external signal. The Linux driver uses the device tree to determine what to do, but this will not work for PSCI.
U-Boot need parse the device tree and configure the WDOG accordingly.
Yes, but consider the difficulties:
There could be multiple Linux device trees, e.g. in a FIT image and chosen based on board revision. Should the PSCI code use the u-boot device tree? Which would then need to also have this board revision support. Or should it use the Linux DT, which it would need to get information from when the kernel DT is prepared before boot?
Of the five external watchdog signals from the iMX7d, which one should be used? I believe each imx watchdog configured in Linux will register a reset handler that are called in turn, so Linux should handle this ok. The uboot PSCI currently assumes wdog1 is the correct one to use.
The Linux driver interfaces with Linux pinctl to iomux the watchdog signal. The PSCI code is assuming the wdog signal is correctly muxed. Default iomux values doesn't have any wdog signal routed to a pin.
Not impossible, but it is not so simple as the original three lines of code might have made it seem at first glance.
If one has the Linux imx watchdog driver working, to use as watchdog, then one gets a wdog system reset for free with all the above taken care of. Which makes me wonder if disabling the PSCI reset might not be a better option in that situration.