The commit 97e3d26b5e5f ("x86/mm: Randomize per-cpu entry area") fixed
an omission of KASLR on CPU entry areas. It doesn't take into account
KASLR switches though, which may result in unintended non-determinism
when a user wants to avoid it (e.g. debugging, benchmarking).
Generate only a single combination of CPU entry areas offsets -- the
linear array that existed prior randomization when KASLR is turned off.
Signed-off-by: Michal Koutný <[email protected]>
---
arch/x86/mm/cpu_entry_area.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/arch/x86/mm/cpu_entry_area.c b/arch/x86/mm/cpu_entry_area.c
index 7316a8224259..f5e93df096fb 100644
--- a/arch/x86/mm/cpu_entry_area.c
+++ b/arch/x86/mm/cpu_entry_area.c
@@ -10,6 +10,7 @@
#include <asm/fixmap.h>
#include <asm/desc.h>
#include <asm/kasan.h>
+#include <asm/setup.h>
static DEFINE_PER_CPU_PAGE_ALIGNED(struct entry_stack_page, entry_stack_storage);
@@ -29,6 +30,12 @@ static __init void init_cea_offsets(void)
unsigned int max_cea;
unsigned int i, j;
+ if (!kaslr_memory_enabled()) {
+ for_each_possible_cpu(i)
+ per_cpu(_cea_offset, i) = i;
+ return;
+ }
+
max_cea = (CPU_ENTRY_AREA_MAP_SIZE - PAGE_SIZE) / CPU_ENTRY_AREA_SIZE;
/* O(sodding terrible) */
--
2.39.2
On 3/3/23 08:06, Michal Koutný wrote:
> @@ -29,6 +30,12 @@ static __init void init_cea_offsets(void)
> unsigned int max_cea;
> unsigned int i, j;
>
> + if (!kaslr_memory_enabled()) {
> + for_each_possible_cpu(i)
> + per_cpu(_cea_offset, i) = i;
> + return;
> + }
Should this be kaslr_memory_enabled() or kaslr_enabled()?
The delta seems to be CONFIG_KASAN, and the cpu entry area randomization
works just fine with KASAN after some recent fixes. I _think_ that
makes cpu entry area randomization more like module randomization which
would point toward kaslr_enabled().
On Fri, Mar 03, 2023 at 01:24:53PM -0800, Dave Hansen <[email protected]> wrote:
> Should this be kaslr_memory_enabled() or kaslr_enabled()?
Originally, I had chosen kaslr_enabled(), seeing the PGD requirement of
KASAN (whole randomization area CPU_ENTRY_AREA_MAP_SIZE would fit in PGD
afterall).
> The delta seems to be CONFIG_KASAN, and the cpu entry area randomization
> works just fine with KASAN after some recent fixes.
But then I found KASAN code trying to be smart and having the fixups,
hence I chickened out to kaslr_memory_enabled().
> I _think_ that makes cpu entry area randomization more like module
> randomization which would point toward kaslr_enabled().
<del>I understood the only difference between kaslr_enabled and
kaslr_memory_enabled is the PGD alignment of the respective regions.
(Although, I don't see where KASAN breaks with unaligned ranges except
for better efficiency of page tables.)</del>
I've just found your [1], wondering the similar.
That being said, I will send v2 with just kaslr_enabled() guard and
updated commit message to beware of KASAN fixups (when backporting).
Thanks,
Michal
[1] https://lore.kernel.org/r/[email protected]/