Re: [PATCH v2 03/23] kcsan: Avoid checking scoped accesses from nested contexts

From
Boqun Feng <hidden>
Date
2021-11-29 09:01:36
Also in
linux-arch, linux-kbuild, linux-mm, lkml
Reply to
Marco Elver, 2021-11-18
Hi Marco,

On Thu, Nov 18, 2021 at 09:10:07AM +0100, Marco Elver wrote:
Avoid checking scoped accesses from nested contexts (such as nested
interrupts or in scheduler code) which share the same kcsan_ctx.

This is to avoid detecting false positive races of accesses in the same
Could you provide an example for a false positive?

I think we do want to detect the following race:

	static int v = SOME_VALUE; // a percpu variable.
	static int other_v = ... ;

	void foo(..)
	{
		int tmp;
		int other_tmp;

		preempt_disable();
		{
			ASSERT_EXCLUSIVE_ACCESSS_SCOPED(v);
			tmp = v;
			
			other_tmp = other_v; // int_handler() may run here
			
			v = tmp + 2;
		}
		preempt_enabled();
	}

	void int_handler() // an interrupt handler
	{
		v++;
	}

, if I understand correctly, we can detect this currently, but with this
patch, we cannot detect this if the interrupt happens while we're doing
the check for "other_tmp = other_v;", right? Of course, running tests
multiple times may eventually catch this, but I just want to understand
what's this patch for, thanks!

Regards,
Boqun
quoted hunk ↗ jump to hunk
thread with currently scoped accesses: consider setting up a watchpoint
for a non-scoped (normal) access that also "conflicts" with a current
scoped access. In a nested interrupt (or in the scheduler), which shares
the same kcsan_ctx, we cannot check scoped accesses set up in the parent
context -- simply ignore them in this case.

With the introduction of kcsan_ctx::disable_scoped, we can also clean up
kcsan_check_scoped_accesses()'s recursion guard, and do not need to
modify the list's prev pointer.

Signed-off-by: Marco Elver <elver@google.com>
---
 include/linux/kcsan.h |  1 +
 kernel/kcsan/core.c   | 18 +++++++++++++++---
 2 files changed, 16 insertions(+), 3 deletions(-)
diff --git a/include/linux/kcsan.h b/include/linux/kcsan.h
index fc266ecb2a4d..13cef3458fed 100644
--- a/include/linux/kcsan.h
+++ b/include/linux/kcsan.h
@@ -21,6 +21,7 @@
  */
 struct kcsan_ctx {
 	int disable_count; /* disable counter */
+	int disable_scoped; /* disable scoped access counter */
 	int atomic_next; /* number of following atomic ops */
 
 	/*
diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
index e34a1710b7bc..bd359f8ee63a 100644
--- a/kernel/kcsan/core.c
+++ b/kernel/kcsan/core.c
@@ -204,15 +204,17 @@ check_access(const volatile void *ptr, size_t size, int type, unsigned long ip);
 static noinline void kcsan_check_scoped_accesses(void)
 {
 	struct kcsan_ctx *ctx = get_ctx();
-	struct list_head *prev_save = ctx->scoped_accesses.prev;
 	struct kcsan_scoped_access *scoped_access;
 
-	ctx->scoped_accesses.prev = NULL;  /* Avoid recursion. */
+	if (ctx->disable_scoped)
+		return;
+
+	ctx->disable_scoped++;
 	list_for_each_entry(scoped_access, &ctx->scoped_accesses, list) {
 		check_access(scoped_access->ptr, scoped_access->size,
 			     scoped_access->type, scoped_access->ip);
 	}
-	ctx->scoped_accesses.prev = prev_save;
+	ctx->disable_scoped--;
 }
 
 /* Rules for generic atomic accesses. Called from fast-path. */
@@ -465,6 +467,15 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type, unsigned
 		goto out;
 	}
 
+	/*
+	 * Avoid races of scoped accesses from nested interrupts (or scheduler).
+	 * Assume setting up a watchpoint for a non-scoped (normal) access that
+	 * also conflicts with a current scoped access. In a nested interrupt,
+	 * which shares the context, it would check a conflicting scoped access.
+	 * To avoid, disable scoped access checking.
+	 */
+	ctx->disable_scoped++;
+
 	/*
 	 * Save and restore the IRQ state trace touched by KCSAN, since KCSAN's
 	 * runtime is entered for every memory access, and potentially useful
@@ -578,6 +589,7 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type, unsigned
 	if (!kcsan_interrupt_watcher)
 		local_irq_restore(irq_flags);
 	kcsan_restore_irqtrace(current);
+	ctx->disable_scoped--;
 out:
 	user_access_restore(ua_flags);
 }
-- 
2.34.0.rc2.393.gf8c9666880-goog
Thread (34 messages) 34 messages, 4 authors, last reply 2021-11-29
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help