kernel26-xen/patches-2.6.25-r1/1152-2.6.25-xen-Clear-__START_KERNEL_map-PTEs-on-free_init_page.patch

From 93ee2997ca1b50cd133d3a9e767851205c222fca Mon Sep 17 00:00:00 2001
From: Eduardo Habkost <ehabkost@redhat.com>
Date: Tue, 15 Apr 2008 17:59:31 -0300
Subject: [PATCH] Xen: Clear __START_KERNEL_map PTEs on free_init_pages() (REVIEWME)

We need to kill the __START_KERNEL_map aliases of the freed init pages
otherwise we won't be able to pin those pages as pagetables later, because
there will be a read-write alias even after make_lowpage_readonly().

Probably using set_memory_ro() instead of make_lowpage_readonly()
when pinning pages will be better.

This should fix the pagetable pinning problems reported at
https://bugzilla.redhat.com/show_bug.cgi?id=441543

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
---
 arch/x86/mm/init_64.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)

diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 763914c..d4083c8 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -1089,6 +1089,16 @@ void free_init_pages(char *what, unsigned long begin, unsigned long end)
 			POISON_FREE_INITMEM, PAGE_SIZE);
 		free_page(addr);
 		totalram_pages++;
+		if (begin >= __START_KERNEL_map) {
+			pte_t *kpte;
+			unsigned int l;
+			/* __START_KERNEL_map addresses must be unmapped
+			 * because we map physical memory at __PAGE_OFFSET
+			 * and we don't want to keep aliases.
+			 */
+			kpte = lookup_address(addr, &l);
+			set_pte(kpte, __pte(0));
+		}
 	}
 #endif
 }
-- 
1.5.4.1

1	From 93ee2997ca1b50cd133d3a9e767851205c222fca Mon Sep 17 00:00:00 2001
2	From: Eduardo Habkost <ehabkost@redhat.com>
3	Date: Tue, 15 Apr 2008 17:59:31 -0300
4	Subject: [PATCH] Xen: Clear __START_KERNEL_map PTEs on free_init_pages() (REVIEWME)
5
6	We need to kill the __START_KERNEL_map aliases of the freed init pages
7	otherwise we won't be able to pin those pages as pagetables later, because
8	there will be a read-write alias even after make_lowpage_readonly().
9
10	Probably using set_memory_ro() instead of make_lowpage_readonly()
11	when pinning pages will be better.
12
13	This should fix the pagetable pinning problems reported at
14	https://bugzilla.redhat.com/show_bug.cgi?id=441543
15
16	Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
17	---
18	arch/x86/mm/init_64.c \| 10 ++++++++++
19	1 files changed, 10 insertions(+), 0 deletions(-)
20
21	diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
22	index 763914c..d4083c8 100644
23	--- a/arch/x86/mm/init_64.c
24	+++ b/arch/x86/mm/init_64.c
25	@@ -1089,6 +1089,16 @@ void free_init_pages(char *what, unsigned long begin, unsigned long end)
26	POISON_FREE_INITMEM, PAGE_SIZE);
27	free_page(addr);
28	totalram_pages++;
29	+ if (begin >= __START_KERNEL_map) {
30	+ pte_t *kpte;
31	+ unsigned int l;
32	+ /* __START_KERNEL_map addresses must be unmapped
33	+ * because we map physical memory at __PAGE_OFFSET
34	+ * and we don't want to keep aliases.
35	+ */
36	+ kpte = lookup_address(addr, &l);
37	+ set_pte(kpte, __pte(0));
38	+ }
39	}
40	#endif
41	}
42	--
43	1.5.4.1
44