On 22/05/2019 10:17, Marc Gonzalez wrote: > [ Re-sending to a wider audience, hoping to catch someone's eye ] > > https://patchwork.kernel.org/patch/10953257/ > > On 21/05/2019 14:03, Marc Gonzalez wrote: > >> With CPUIDLE enabled, my system starts to boot, hangs a few seconds, >> then silently reboots (killed by FW) > > Using extremely high-tech debugging tools (i.e. spraying printk left and right) > I traced this crash down to: > > psci_cpu_suspend_enter: 435 > psci_cpu_suspend: 171 > psci_cpu_suspend: __invoke_psci_fn_smc c4000001 > __invoke_psci_fn_smc: id=c4000001 3 0 0 > /*** execution never returns from arm_smccc_smc() ***/ > > > The following dmesg log caught my eye, and might be relevant: > > ARM_SMCCC_ARCH_WORKAROUND_1 missing from firmware > > Is that bad, doctor? I don't think it would explain the crash though... It doesn't explain it. It is bad though: your vendor doesn't tell the kernel about the HW being free of Spectre-v2, and doesn't provide a mitigation either. Hopefully, this is a responsible vendor that will provide you with a firmware update that fixes it. > > > If I revert the patch, psci_cpu_suspend_enter() is never called, > so we don't tickle the arm_smccc_smc() monster. > > Could it be that this FW doesn't support PSCI? Probably not in as well as it could, at least for CPU suspend. M. -- Jazz is not dead. It just smells funny...
