Re: [PATCH] x86: vdso: fix pvclock races with task migration

[Andy Lutomirski]

On Thu, Apr 2, 2015 at 11:44 AM, Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx> wrote:
> If we were migrated right after __getcpu, but before reading the
> migration_count, we wouldn't notice that we read TSC of a different
> VCPU, nor that KVM's bug made pvti invalid, as only migration_count
> on source VCPU is increased.
>
> Change vdso instead of updating migration_count on destination.

Looks good to me.

--Andy

>
> Fixes: 0a4e6be9ca17 ("x86: kvm: Revert "remove sched notifier for cross-cpu migrations"")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>
> ---
>  Because it we'll get a complete rewrite, this series does not
>  - remove the outdated 'TODO: We can put [...]' comment
>  - use a proper encapsulation for the inner do-while loop
>  - optimize the outer do-while loop
>    (no need to re-read cpu id on version mismatch)
>
>  arch/x86/vdso/vclock_gettime.c | 20 ++++++++++++--------
>  1 file changed, 12 insertions(+), 8 deletions(-)
>
> diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
> index 30933760ee5f..40d2473836c9 100644
> --- a/arch/x86/vdso/vclock_gettime.c
> +++ b/arch/x86/vdso/vclock_gettime.c
> @@ -99,21 +99,25 @@ static notrace cycle_t vread_pvclock(int *mode)
>                  * __getcpu() calls (Gleb).
>                  */
>
> -               pvti = get_pvti(cpu);
> +               /* Make sure migrate_count will change if we leave the VCPU. */
> +               do {
> +                       pvti = get_pvti(cpu);
> +                       migrate_count = pvti->migrate_count;
>
> -               migrate_count = pvti->migrate_count;
> +                       cpu1 = cpu;
> +                       cpu = __getcpu() & VGETCPU_CPU_MASK;
> +               } while (unlikely(cpu != cpu1));
>
>                 version = __pvclock_read_cycles(&pvti->pvti, &ret, &flags);
>
>                 /*
>                  * Test we're still on the cpu as well as the version.
> -                * We could have been migrated just after the first
> -                * vgetcpu but before fetching the version, so we
> -                * wouldn't notice a version change.
> +                * - We must read TSC of pvti's VCPU.
> +                * - KVM doesn't follow the versioning protocol, so data could
> +                *   change before version if we left the VCPU.
>                  */
> -               cpu1 = __getcpu() & VGETCPU_CPU_MASK;
> -       } while (unlikely(cpu != cpu1 ||
> -                         (pvti->pvti.version & 1) ||
> +               smp_rmb();
> +       } while (unlikely((pvti->pvti.version & 1) ||
>                           pvti->pvti.version != version ||
>                           pvti->migrate_count != migrate_count));
>
> --
> 2.3.4
>



-- 
Andy Lutomirski
AMA Capital Management, LLC
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/