Received-SPF: pass (google.com: domain of linux-kernel+bounces-699-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Date: Fri, 15 Dec 2023 12:45:57 +0300
From: Alexey Dobriyan <adobriyan@gmail.com>
To: Thomas Gleixner <tglx@linutronix.de>, Ingo Molnar <mingo@redhat.com>,
	Borislav Petkov <bp@alien8.de>,
	Dave Hansen <dave.hansen@linux.intel.com>
Cc: linux-kernel@vger.kernel.org, "H. Peter Anvin" <hpa@zytor.com>
Subject: [PATCH] selftests/x86/mm: fixup nx_stack test stability with
 SA_ONSTACK
Message-ID: <8299b17a-a730-46e5-a258-fac78ff0420b@p183>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline

I forgot that using sigaltstack(2) requires opt-in with SA_ONSTACK.

If userspace stack is NX, then the test continues to work and
reports PASS.

If there is kernel bug and some pages of userspace stack are executable,
then test can be derailed because signal stack frame contents will pass
as random instruction with unpredictable consequences.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
---

 tools/testing/selftests/x86/nx_stack.c |   34 ++++++++++++++++-----------------
 1 file changed, 17 insertions(+), 17 deletions(-)

--- a/tools/testing/selftests/x86/nx_stack.c
+++ b/tools/testing/selftests/x86/nx_stack.c
@@ -160,10 +160,25 @@ static void sigtrap(int _, siginfo_t *__, void *uc_)
 
 int main(void)
 {
+	{
+		/*
+		 * We don't know now much stack SIGSEGV handler uses.
+		 * Bump this by 1 page every time someone complains,
+		 * or rewrite it in assembly.
+		 */
+		const size_t len = SIGSTKSZ;
+		void *p = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+		assert(p != MAP_FAILED);
+		stack_t ss = {};
+		ss.ss_sp = p;
+		ss.ss_size = len;
+		int rv = sigaltstack(&ss, NULL);
+		assert(rv == 0);
+	}
 	{
 		struct sigaction act = {};
 		sigemptyset(&act.sa_mask);
-		act.sa_flags = SA_SIGINFO;
+		act.sa_flags = SA_SIGINFO|SA_ONSTACK;
 		act.sa_sigaction = &sigsegv;
 		int rv = sigaction(SIGSEGV, &act, NULL);
 		assert(rv == 0);
@@ -171,7 +186,7 @@ int main(void)
 	{
 		struct sigaction act = {};
 		sigemptyset(&act.sa_mask);
-		act.sa_flags = SA_SIGINFO;
+		act.sa_flags = SA_SIGINFO|SA_ONSTACK;
 		act.sa_sigaction = &sigtrap;
 		int rv = sigaction(SIGTRAP, &act, NULL);
 		assert(rv == 0);
@@ -188,21 +203,6 @@ int main(void)
 		rv = setrlimit(RLIMIT_STACK, &rlim);
 		assert(rv == 0);
 	}
-	{
-		/*
-		 * We don't know now much stack SIGSEGV handler uses.
-		 * Bump this by 1 page every time someone complains,
-		 * or rewrite it in assembly.
-		 */
-		const size_t len = SIGSTKSZ;
-		void *p = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
-		assert(p != MAP_FAILED);
-		stack_t ss = {};
-		ss.ss_sp = p;
-		ss.ss_size = len;
-		int rv = sigaltstack(&ss, NULL);
-		assert(rv == 0);
-	}
 	make_stack1();
 	/*
 	 * Unreachable, but if _this_ INT3 is ever reached, it's a bug somewhere.