[patch] Upstream updated patch

1700_define-pvclock-pvti-cpu0-va-setter-for-X86-32.patch (text/plain), 1.76 KB, created by Mike Pagano on 2018-06-21 22:42:02 UTC

pvti_cpu0_va is the address of shared kvmclock data structure.

pvti_cpu0_va is currently kept unset (1) on 32 bit systems, (2) when
kvmclock vsyscall is disabled, and (3) if kvmclock is not stable.
This poses a problem, because kvm_ptp needs pvti_cpu0_va, but (1) can
work on 32 bit, (2) has little relation to the vsyscall, and (3) does
not need stable kvmclock (although kvmclock won't be used for system
clock if it's not stable, so kvm_ptp is pointless in that case).

Expose pvti_cpu0_va whenever kvmclock is enabled to allow all users to
work with it.

This fixes a regression found on Gentoo: https://bugs.gentoo.org/658544.

Fixes: 9f08890ab906 ("x86/pvclock: add setter for pvclock_pvti_cpu0_va")
Reported-by: Andreas Steinmetz <[e-mail]>
Signed-off-by: Radim Krčmář <[e-mail]>
---
 arch/x86/kernel/kvmclock.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kernel/kvmclock.c b/arch/x86/kernel/kvmclock.c
index bf8d1eb7fca3..46ffa8327563 100644
--- a/arch/x86/kernel/kvmclock.c
+++ b/arch/x86/kernel/kvmclock.c
@@ -319,6 +319,8 @@ void __init kvmclock_init(void)
 	printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 		msr_kvm_system_time, msr_kvm_wall_clock);
 
+	pvclock_set_pvti_cpu0_va(hv_clock);
+
 	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 
@@ -366,14 +368,11 @@ int __init kvm_setup_vsyscall_timeinfo(void)
 	vcpu_time = &hv_clock[cpu].pvti;
 	flags = pvclock_read_flags(vcpu_time);
 
-	if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
-		put_cpu();
-		return 1;
-	}
-
-	pvclock_set_pvti_cpu0_va(hv_clock);
 	put_cpu();
 
+	if (!(flags & PVCLOCK_TSC_STABLE_BIT))
+		return 1;
+
 	kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 #endif
 	return 0;
-- 
2.18.0.rc2