2004-10-11 22:10:46

by Andrea Arcangeli

[permalink] [raw]
Subject: ptep_establish smp race x86 PAE >4G

This avoid userspace mm corruption during COWs with threads (i.e.
malloc;fork;clone) on x86 PAE with >4G of ram

Signed-Off-By: Andrea Arcangeli <[email protected]>

Index: linux-2.5/include/asm-i386/pgtable-3level.h
===================================================================
RCS file: /home/andrea/crypto/cvs/linux-2.5/include/asm-i386/pgtable-3level.h,v
retrieving revision 1.20
diff -u -p -r1.20 pgtable-3level.h
--- linux-2.5/include/asm-i386/pgtable-3level.h 24 Aug 2004 18:28:07 -0000 1.20
+++ linux-2.5/include/asm-i386/pgtable-3level.h 11 Oct 2004 21:17:56 -0000
@@ -54,6 +54,7 @@ static inline void set_pte(pte_t *ptep,
smp_wmb();
ptep->pte_low = pte.pte_low;
}
+#define __HAVE_ARCH_SET_PTE_ATOMIC
#define set_pte_atomic(pteptr,pteval) \
set_64bit((unsigned long long *)(pteptr),pte_val(pteval))
#define set_pmd(pmdptr,pmdval) \
Index: linux-2.5/include/asm-generic/pgtable.h
===================================================================
RCS file: /home/andrea/crypto/cvs/linux-2.5/include/asm-generic/pgtable.h,v
retrieving revision 1.8
diff -u -p -r1.8 pgtable.h
--- linux-2.5/include/asm-generic/pgtable.h 29 Jul 2004 06:01:30 -0000 1.8
+++ linux-2.5/include/asm-generic/pgtable.h 11 Oct 2004 22:07:31 -0000
@@ -13,11 +13,19 @@
* Note: the old pte is known to not be writable, so we don't need to
* worry about dirty bits etc getting lost.
*/
+#ifndef __HAVE_ARCH_SET_PTE_ATOMIC
#define ptep_establish(__vma, __address, __ptep, __entry) \
do { \
set_pte(__ptep, __entry); \
flush_tlb_page(__vma, __address); \
} while (0)
+#else /* __HAVE_ARCH_SET_PTE_ATOMIC */
+#define ptep_establish(__vma, __address, __ptep, __entry) \
+do { \
+ set_pte_atomic(__ptep, __entry); \
+ flush_tlb_page(__vma, __address); \
+} while (0)
+#endif /* __HAVE_ARCH_SET_PTE_ATOMIC */
#endif

#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS