xen/x86/pvh: Use HVM's flush_tlb_others op
authorBoris Ostrovsky <boris.ostrovsky@oracle.com>
Sun, 13 Dec 2015 00:25:55 +0000 (19:25 -0500)
committerDavid Vrabel <david.vrabel@citrix.com>
Mon, 14 Dec 2015 13:53:40 +0000 (13:53 +0000)
Using MMUEXT_TLB_FLUSH_MULTI doesn't buy us much since the hypervisor
will likely perform same IPIs as would have the guest.

More importantly, using MMUEXT_INVLPG_MULTI may not to invalidate the
guest's address on remote CPU (when, for example, VCPU from another guest
is running there).

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Suggested-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
arch/x86/xen/mmu.c

index ac161db633885872fdebdce976d05b978e1235da..cb5e266a8bf752297eaf6fc0d822c9398239fadf 100644 (file)
@@ -2495,14 +2495,9 @@ void __init xen_init_mmu_ops(void)
 {
        x86_init.paging.pagetable_init = xen_pagetable_init;
 
-       /* Optimization - we can use the HVM one but it has no idea which
-        * VCPUs are descheduled - which means that it will needlessly IPI
-        * them. Xen knows so let it do the job.
-        */
-       if (xen_feature(XENFEAT_auto_translated_physmap)) {
-               pv_mmu_ops.flush_tlb_others = xen_flush_tlb_others;
+       if (xen_feature(XENFEAT_auto_translated_physmap))
                return;
-       }
+
        pv_mmu_ops = xen_mmu_ops;
 
        memset(dummy_mapping, 0xff, PAGE_SIZE);