[PATCH v2] irq_work: Make irq_work_queue_on() NMI-safe again

[PATCH v2] irq_work: Make irq_work_queue_on() NMI-safe again

[Jun Miao]

We should not put NMI unsafe code in irq_work_queue_on().

The KASAN of kasan_record_aux_stack_noalloc() is not NMI safe. Because which
will call the spinlock. While the irq_work_queue_on() is also very carefully
crafted to be exactly that.
When unable CONFIG_SMP or local CPU, the irq_work_queue_on() is even same to
irq_work_queue(). So delete KASAN instantly.

Fixes: e2b5bcf9f5ba ("irq_work: record irq_work_queue() call stack")
Suggested by: "Huang, Ying" <ying.huang@intel.com>
Signed-off-by: Jun Miao <jun.miao@intel.com>
Acked-by: Marco Elver <elver@google.com>
---
 kernel/irq_work.c | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/kernel/irq_work.c b/kernel/irq_work.c
index 7afa40fe5cc4..e7f48aa8d8af 100644
--- a/kernel/irq_work.c
+++ b/kernel/irq_work.c
@@ -20,7 +20,6 @@
 #include <linux/smp.h>
 #include <linux/smpboot.h>
 #include <asm/processor.h>
-#include <linux/kasan.h>
 
 static DEFINE_PER_CPU(struct llist_head, raised_list);
 static DEFINE_PER_CPU(struct llist_head, lazy_list);
@@ -137,8 +136,6 @@ bool irq_work_queue_on(struct irq_work *work, int cpu)
 	if (!irq_work_claim(work))
 		return false;
 
-	kasan_record_aux_stack_noalloc(work);
-
 	preempt_disable();
 	if (cpu != smp_processor_id()) {
 		/* Arch remote IPI send/receive backend aren't NMI safe */
-- 
2.32.0

Re: [PATCH v2] irq_work: Make irq_work_queue_on() NMI-safe again

[Peter Zijlstra]

On Wed, Apr 27, 2022 at 09:55:49PM +0800, Jun Miao wrote:
> We should not put NMI unsafe code in irq_work_queue_on().

Why not, irq_work_queue_on() is not NMI safe. Only irq_work_queue() is.