x86: Fix code patching for paravirt-alternatives on 486
PATCH 1/1 x86: Fix code patching for paravirt-alternatives on 486
As reported in <http://bugs.debian.org/511703> and
<http://bugs.debian.org/515982>, kernels with paravirt-alternatives enabled crash in text_poke_early() on at least some 486-class processors.
The problem is that text_poke_early() itself contains paravirt- alternatives and therefore will modify instructions that have already been prefetched. Pentium and later processors will invalidate the prefetched instructions in this case, but 486-class processors do not. We must use a jmp instruction to limit prefetching.
There is then a further problem in that sync_core() uses "cpuid" which isn't implemented by most 486-class processors. Since they also do not perform speculative execution, we can make this conditional on the processor family.
---
This has been tested as a change to 2.6.26 by Richard Kettlewell. This code doesn't appear to have changed significantly since then, so hopefully the change is still correct.
Possible the call to sync_core() should be moved above the local_irq_restore() and should incorporate the dummy jmp?
Ben.
arch/x86/include/asm/processor.h | 6 ++++++ arch/x86/kernel/alternative.c | 3 +++ 2 files changed, 9 insertions(+), 0 deletions(-)
arch/x86/include/asm/processor.h +6 -0
@@ -708,6 +708,12 @@ static inline void sync_core(void) {int tmp; +#if defined(CONFIG_M386 ) || defined(CONFIG_M486 ) +/* This is unnecessary on 386- and 486-class processors, most of +which don't even implement CPUID. */ +if (boot_cpu_data.x86 <5 ) +return ; +#endif asmvolatile ("cpuid" :"=a" (tmp) :"0" (1 ) :"ebx" ,"ecx" ,"edx" ,"memory" ); }
arch/x86/kernel/alternative.c +3 -0
@@ -498,6 +498,9 @@ static void *__init_or_module text_poke_early(void *addr, const void *opcode,unsigned long flags;local_irq_save (flags);memcpy (addr, opcode, len); +/* Force 486-class processors to flush prefetched instructions, +since we may have just patched local_irq_restore(). */ + asmvolatile ("jmp 1f\n1:\n" :::"memory" );local_irq_restore (flags);sync_core ();/* Could also do a CLFLUSH here to speed up CPU recovery; but
Signature · 1 line
1.6.3.3
H. Peter Anvin
reviewerQuoted 23 lines · Ben Hutchings
arch/x86/include/asm/processor.h +6
-0@@ -708,6 +708,12 @@ static inline void sync_core(void) {int tmp; +#if defined(CONFIG_M386 ) || defined(CONFIG_M486 ) +/* This is unnecessary on 386- and 486-class processors, most of +which don't even implement CPUID. */ +if (boot_cpu_data.x86 <5 ) +return ; +#endif asmvolatile ("cpuid" :"=a" (tmp) :"0" (1 ) :"ebx" ,"ecx" ,"edx" ,"memory" ); }arch/x86/kernel/alternative.c +0
-0
arch/x86/kernel/alternative.c static void *__init_or_module text_poke_early(void *addr, const void *opcode, · line 506
Quoted 9 lines
arch/x86/kernel/alternative.c +3
-0@@ -498,6 +498,9 @@ static void *__init_or_module text_poke_early(void *addr, const void *opcode,unsigned long flags;local_irq_save (flags);memcpy (addr, opcode, len); +/* Force 486-class processors to flush prefetched instructions, +since we may have just patched local_irq_restore(). */ + asmvolatile ("jmp 1f\n1:\n" :::"memory" );local_irq_restore (flags);sync_core ();/* Could also do a CLFLUSH here to speed up CPU recovery; but
I'm wondering if it wouldn't be cleaner to fold the jump into sync_core() and moving the sync_core() up before local_irq_restore().
-hpa
On Tue, 2009-09-08 at 17:31 -0700, H. Peter Anvin wrote:
[...]
Quoted 2 lines
I'm wondering if it wouldn't be cleaner to fold the jump into
sync_core()and moving thesync_core()up beforelocal_irq_restore().
Exactly as I suggested below the dashes. :-)
The only reason I didn't do that initially was that I don't know whether or not there's a good reason for the current placement of sync_core() after local_irq_restore().
Ben.
Signature · 2 lines
Ben Hutchings Who are all these weirdos? - David Bowie, about L-Space IRC channel #afp
H. Peter Anvin
reviewerQuoted 7 lines · Ben Hutchings
On Tue, 2009-09-08 at 17:31 -0700, H. Peter Anvin wrote:
[...]Quoted 2 lines
I'm wondering if it wouldn't be cleaner to fold the jump into
sync_core()and moving thesync_core()up beforelocal_irq_restore().Exactly as I suggested below the dashes. :-)
Ah, yes, so you did indeed.
Quoted 3 lines
The only reason I didn't do that initially was that I don't know whether or not there's a good reason for the current placement of
sync_core()afterlocal_irq_restore().
I can personally not fathom how moving the sync_core() before local_irq_restore() could cause problems -- in fact, the current code seems logically wrong, although probably correct in practice, as the interrupt itself should perform the necessary synchronization (although it is not architecturally required to do so.) Not that I haven't been wrong before.
-hpa