| Message ID | 1340618133-10755-3-git-send-email-stefan.bader@canonical.com |
|---|---|
| State | New |
| Headers | show |
diff --git a/arch/x86/include/mach-xen/asm/spinlock_types.h b/arch/x86/include/mach-xen/asm/spinlock_types.h index 7d01ae1..3fa1833 100644 --- a/arch/x86/include/mach-xen/asm/spinlock_types.h +++ b/arch/x86/include/mach-xen/asm/spinlock_types.h @@ -13,7 +13,7 @@ typedef union { /* * Xen versions prior to 3.2.x have a race condition with HYPERVISOR_poll(). */ -#if CONFIG_XEN_COMPAT >= 0x030200 +#if CONFIG_XEN_COMPAT >= 0x030002 /* * On Xen we support a single level of interrupt re-enabling per lock. Hence * we can have twice as many outstanding tickets. Thus the cut-off for using
The SUSE patchset did limit the ticket lock code to be used only on Xen hypervisors version 3.2 or higher because HYPERVISOR_poll() is supposed to have issues before. However we have been using that call before and the chance to be running on EC2 on such an old hypervisor is getting even less. Not using ticket locks and the hypervisor assisting could increase host cpu usage and unfairness in lock contention situations. BugLink: http://bugs.launchpad.net/bugs/929941 Signed-off-by: Stefan Bader <stefan.bader@canonical.com> --- arch/x86/include/mach-xen/asm/spinlock_types.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)