On Tue, Jun 19, 2018 at 07:00:13AM -0700, kernelci.org bot wrote:
Today and yesterday's -next is failing to boot on rk3399-firefly in two
configs:
defconfig: rk3399-firefly: lab-baylibre-seattle: failing since 1 day (last pass: next-20180615 - first fail: next-20180618)>> defconfig+CONFIG_RANDOMIZE_BASE=y: rk3399-firefly: lab-baylibre-seattle: failing since 1 day (last pass: next-20180615 - first fail: next-20180618)>
The log ends with:
[ 2.629278] SWITCH_REG2: supplied by vcc3v3_sys [ 2.633030] fan53555-regulator 0-0040: FAN53555 Option[8] Rev[1] Detected! [ 2.633723] fan53555-reg: supplied by vcc_sys [ 2.6353|] Internal error: undefined instruction: 0 [#1] PREEMPT SMP
which is rather entertaining.
Am Dienstag, 19. Juni 2018, 16:41:43 CEST schrieb Mark Brown: On Tue, Jun 19, 2018 at 03:22:47PM +0100, Sudeep Holla wrote:
Based on the logs from successful boots, I see it's CPUFreq that gets probed after fan53555-reg. I also see that it warns about unlisted OPPs in the successful boots. I guess, the undefined instruction could occur if there's issue with CPU DVFS(under voltage or frequency that's not supported). I don't have much knowledge on the platform, I am just guessing on the possible cause here.
Yeah, that was my guess - either an out of spec voltage/frequency was selected or the regulator isn't able to keep up with the voltage transition or a load spike and the power browns out.
I guess this is related to arm64: dts: rockchip: correct voltage selector Firefly-RK3399 [0]
which is in so far interesting as the patch seems to fix a similar issue to the above with the opposite "fcs,suspend-voltage-selector" value. And also claims to be in line with the vendor kernel.
From looking at the schematics, we might have a discrepancy between the
fcs,suspend-voltage-selector and vsel input-pin leading to a gpio [likely depending on what the bootloader set this pin up for].
I'll drop the patch for now and get in contact with the patch submitter.
Heiko