yeah - sorry - the '16' was a blast from the past, i just took the old
2.6.24 condition on the 32-bit side which didnt use PF_INSTR (PF_INSTR
came in as a cleanup during the unification), and when the confirmation
came that this fixed the crash i sent the pull request. That's how this
nonsensical mixing happened, and that's why i missed the 'if (error_code
& PF_INSTR)' branch.
The patch below (ontop of the tree) is a first cut at fixing all these
problems - but i'd wait at least 24 hours with applying this to let it
be tested through - it affects both 32-bit and 64-bit. The fix further
cleans up this codepath and removes an #ifdef.
Ingo
------------------------>
Subject: x86: prefetch fix #2
From: Ingo Molnar <mingo@elte.hu>
Date: Thu Mar 27 21:29:09 CET 2008
Linus noticed a second bug and an uncleanliness:
- we'd return on any instruction fetch fault
- we'd use both the value of 16 and the PF_INSTR symbol which are
the same and make no sense
the cleanup nicely unifies this piece of logic.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
arch/x86/mm/fault.c | 12 ++++--------
1 file changed, 4 insertions(+), 8 deletions(-)
Index: linux-x86.q/arch/x86/mm/fault.c
===================================================================
--- linux-x86.q.orig/arch/x86/mm/fault.c
+++ linux-x86.q/arch/x86/mm/fault.c
@@ -103,14 +103,10 @@ static int is_prefetch(struct pt_regs *r
int prefetch = 0;
unsigned char *max_instr;
-#ifdef CONFIG_X86_32
- /* Catch an obscure case of prefetch inside an NX page: */
- if ((__supported_pte_mask & _PAGE_NX) && (error_code & 16))
- return 0;
-#endif
-
- /* If it was a exec fault on NX page, ignore */
- if (error_code & PF_INSTR)
+ /*
+ * Catch an obscure case of prefetch inside an NX page:
+ */
+ if ((__supported_pte_mask & _PAGE_NX) && (error_code & PF_INSTR))
return 0;
instr = (unsigned char *)convert_ip_to_linear(current, regs);
--
