I've been able to get an unmodified Xen guest to OOPS once after a
lot of test iterations without this patch. I think this patch fixes
the problem. I'm a bit surprised that we don't see much more severe
LDT problems on Xen without this fix.
Once the synchronous modify_ldt code causes modify_ldt to more
aggressively reallocate the LDT, the OOPSes become much more common
without this fix.
Cc: sta...@vger.kernel.org
Signed-off-by: Andy Lutomirski <l...@kernel.org>
---
arch/x86/xen/enlighten.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index 0b95c9b8283f..e417d08c56c4 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -32,6 +32,7 @@
#include <linux/gfp.h>
#include <linux/memblock.h>
#include <linux/edd.h>
+#include <linux/vmalloc.h>
#include <xen/xen.h>
#include <xen/events.h>
@@ -512,6 +513,10 @@ static void xen_alloc_ldt(struct desc_struct *ldt,
unsigned entries)
for(i = 0; i < entries; i += entries_per_page)
set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
+
+ /* If there are stray aliases, the LDT won't work. */
+ if (is_vmalloc_addr(ldt))
+ vm_unmap_aliases();
}
static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
@@ -519,6 +524,13 @@ static void xen_free_ldt(struct desc_struct *ldt, unsigned
entries)
const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
int i;
+ /*
+ * The set_aliased_prot call may OOPS due to a hypercall failure
+ * if there are any lazy vmap aliases of the page. We don't
+ * need to call vm_unmap_aliases() here, though, because we
+ * already eliminated any aliases in xen_alloc_ldt.
+ */
+
for(i = 0; i < entries; i += entries_per_page)
set_aliased_prot(ldt + i, PAGE_KERNEL);
}
--
2.4.3
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel
