[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH] Make pte writes in map_pages_to_xen() atomic



With debug=y, Xen calls map_pages_to_xen() to remap its own page tables. On pae builds, the pte writes are nonatomic. This causes Xen to crash under kvm (and presumably under other shadow page table caching VMMs), because after the first half of the pte update, the code under eip is no longer accessible. It's also possible under bare metal if an SMI arrives exactly at the wrong moment, and in addition that SMI flushes the tlb.

Fix by using atomic pte writes.  With this, xen pae debug=y boots under kvm.

Reported by Jeremy Fitzhardinge <jeremy@xxxxxxxx>.

Signed-off-by: Avi Kivity <avi@xxxxxxxxxxxx>

--
error compiling committee.c: too many arguments to function

diff -r 039daabebad5 xen/arch/x86/mm.c
--- a/xen/arch/x86/mm.c Fri Apr 13 16:07:48 2007 +0100
+++ b/xen/arch/x86/mm.c Sun Apr 15 18:36:59 2007 +0300
@@ -3454,14 +3454,14 @@ int map_pages_to_xen(
                     l1e_write(&pl1e[i],
                               l1e_from_pfn(l2e_get_pfn(*pl2e) + i,
                                            l2e_get_flags(*pl2e) & ~_PAGE_PSE));
-                l2e_write(pl2e, l2e_from_pfn(virt_to_mfn(pl1e),
+                l2e_write_atomic(pl2e, l2e_from_pfn(virt_to_mfn(pl1e),
                                              __PAGE_HYPERVISOR));
                 local_flush_tlb_pge();
             }
 
             pl1e  = l2e_to_l1e(*pl2e) + l1_table_offset(virt);
             ol1e  = *pl1e;
-            l1e_write(pl1e, l1e_from_pfn(mfn, flags));
+            l1e_write_atomic(pl1e, l1e_from_pfn(mfn, flags));
             if ( (l1e_get_flags(ol1e) & _PAGE_PRESENT) )
                 local_flush_tlb_one(virt);
 
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.