[Xen-changelog] [xen-unstable] [IA64] Avoid iterative VHPT purges

[Xen patchbot-unstable <patchbot-unstable@xxxxxxxxxxxxxxxxxxx>]

# HG changeset patch
# User awilliam@xxxxxxxxxxx
# Node ID a34659228c2466f4443f22543ea2bebd3859b077
# Parent  36a3c92cdf8f4a2c06f32b9b3ab28432288f1562
[IA64] Avoid iterative VHPT purges

Workaround to avoid iteratively purging 16M itr.d using
16k Xen VHPT page size.

Signed-off-by: Anthony Xu <anthony.xu@xxxxxxxxx>
---
 xen/arch/ia64/vmx/vmmu.c |   10 +++++++++-
 1 files changed, 9 insertions(+), 1 deletion(-)

diff -r 36a3c92cdf8f -r a34659228c24 xen/arch/ia64/vmx/vmmu.c
--- a/xen/arch/ia64/vmx/vmmu.c  Thu Sep 21 15:33:50 2006 -0600
+++ b/xen/arch/ia64/vmx/vmmu.c  Thu Sep 21 15:34:10 2006 -0600
@@ -456,7 +456,15 @@ IA64FAULT vmx_vcpu_itr_d(VCPU *vcpu, u64
     }
 #endif   
     pte &= ~PAGE_FLAGS_RV_MASK;
-    thash_purge_entries(vcpu, va, ps);
+
+    /* This is a bad workaround
+       In Linux, region 7 use 16M pagesize and is identity mapped.
+       VHPT page size is 16K in XEN.  If purge VHPT while guest insert 16M,
+       it will iteratively purge VHPT 1024 times, which makes XEN/IPF very
+       slow.  XEN doesn't purge VHPT
+    */   
+    if (ps != _PAGE_SIZE_16M)
+        thash_purge_entries(vcpu, va, ps);
     gpfn = (pte & _PAGE_PPN_MASK)>> PAGE_SHIFT;
     if (VMX_DOMAIN(vcpu) && __gpfn_is_io(vcpu->domain, gpfn))
         pte |= VTLB_PTE_IO;

_______________________________________________
Xen-changelog mailing list
Xen-changelog@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-changelog