On Mon, Aug 20, 2018 at 3:02 PM Thomas Gleixner tglx@linutronix.de wrote:
On Mon, 20 Aug 2018, Jinpu Wang wrote:
On Sat, Aug 18, 2018 at 2:50 PM Thomas Gleixner tglx@linutronix.de wrote:
On Fri, 17 Aug 2018, Greg Kroah-Hartman wrote:
On Fri, Aug 17, 2018 at 02:39:00PM +0200, Jinpu Wang wrote:
I found the problem, CONFIG_HOTPLUG_SMT is enable due to CONFIG_SMP, but I did explicitly enable CONFIG_HOTPLUG_CPU.
That's why the smt dir is missing, and kernel panic when I boot with 'nosmt' or 'l1tf=full' probably also other combination.
With CONFIG_HOTPLUG_CPU=y, it works fine so far.
So do we need a new configuration dependancy here to resolve this problem? Why is this not an issue in Linus's tree?
That's probably the same problem on Linus tree. Have not tested it yet though. I'll have a look.
Thanks,
tglxYeah, it's the same problem on Linus tree.
Probably makes sense to do the following, I can sent a formal patch if you think it's right thing to do:
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index b0312f8..8a52f0b 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -188,7 +188,7 @@ config X86 select HAVE_SYSCALL_TRACEPOINTS select HAVE_UNSTABLE_SCHED_CLOCK select HAVE_USER_RETURN_NOTIFIER
select HOTPLUG_SMT if SMP
select HOTPLUG_SMT if SMP && HOTPLUG_CPUThat works, but we might as well force 'select HOTPLUG_CPU if SMP' so the L1TF mitigations are functional. Sigh....
Thanks,
tglx
right, sth like? diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index b0312f8..5852948 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -188,7 +188,8 @@ config X86 select HAVE_SYSCALL_TRACEPOINTS select HAVE_UNSTABLE_SCHED_CLOCK select HAVE_USER_RETURN_NOTIFIER - select HOTPLUG_SMT if SMP + select HOTPLUG_CPU if SMP + select HOTPLUG_SMT if SMP && HOTPLUG_CPU select IRQ_FORCED_THREADING select NEED_SG_DMA_LENGTH select PCI_LOCKLESS_CONFIG