Fei has reported that KASAN triggers during apply_alternatives() on
a 5-level paging machine:
BUG: KASAN: out-of-bounds in rcu_is_watching()
Read of size 4 at addr ff110003ee6419a0 by task swapper/0/0
...
__asan_load4()
rcu_is_watching()
trace_hardirqs_on()
text_poke_early()
apply_alternatives()
...
On machines with 5-level paging, cpu_feature_enabled(X86_FEATURE_LA57)
gets patched. It includes KASAN code, where KASAN_SHADOW_START depends on
__VIRTUAL_MASK_SHIFT, which is defined with cpu_feature_enabled().
KASAN gets confused when apply_alternatives() patches the
KASAN_SHADOW_START users. A test patch that makes KASAN_SHADOW_START
static, by replacing __VIRTUAL_MASK_SHIFT with 56, works around the issue.
Fix it for real by disabling KASAN while the kernel is patching alternatives.
[ mingo: updated the changelog ]
Fixes: 6657fca06e3f ("x86/mm: Allow to boot without LA57 if CONFIG_X86_5LEVEL=y")
Reported-by: Fei Yang <fei.yang@xxxxxxxxx>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
Acked-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
Link: https://lore.kernel.org/r/20231012100424.1456-1-kirill.shutemov@xxxxxxxxxxxxxxx
(cherry picked from commit d35652a5fc9944784f6f50a5c979518ff8dacf61)
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
---
arch/x86/kernel/alternative.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index d1d92897ed6b..46b7ee0ab01a 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -270,6 +270,17 @@ void __init_or_module noinline apply_alternatives(struct alt_instr *start,
u8 insn_buff[MAX_PATCH_LEN];
DPRINTK("alt table %px, -> %px", start, end);
+
+ /*
+ * In the case CONFIG_X86_5LEVEL=y, KASAN_SHADOW_START is defined using
+ * cpu_feature_enabled(X86_FEATURE_LA57) and is therefore patched here.
+ * During the process, KASAN becomes confused seeing partial LA57
+ * conversion and triggers a false-positive out-of-bound report.
+ *
+ * Disable KASAN until the patching is complete.
+ */
+ kasan_disable_current();
+
/*
* The scan order should be from start to end. A later scanned
* alternative code can overwrite previously scanned alternative code.
@@ -337,6 +348,8 @@ void __init_or_module noinline apply_alternatives(struct alt_instr *start,
next:
optimize_nops(instr, a->instrlen);
}
+
+ kasan_enable_current();
}
static inline bool is_jcc32(struct insn *insn)
--
2.41.0
