linux-kernel - [PATCH AUTOSEL 4.9 29/48] selftests/x86/fsgsbase: Fix GS == 1, 2, and 3 tests

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20201223022417.2794032-29-sashal@kernel.org>
Date:   Tue, 22 Dec 2020 21:23:57 -0500
From:   Sasha Levin <sashal@...nel.org>
To:     linux-kernel@...r.kernel.org, stable@...r.kernel.org
Cc:     Andy Lutomirski <luto@...nel.org>, Borislav Petkov <bp@...e.de>,
        Sasha Levin <sashal@...nel.org>,
        linux-kselftest@...r.kernel.org
Subject: [PATCH AUTOSEL 4.9 29/48] selftests/x86/fsgsbase: Fix GS == 1, 2, and 3 tests

From: Andy Lutomirski <luto@...nel.org>

[ Upstream commit 716572b0003ef67a4889bd7d85baf5099c5a0248 ]

Setting GS to 1, 2, or 3 causes a nonsensical part of the IRET microcode
to change GS back to zero on a return from kernel mode to user mode. The
result is that these tests fail randomly depending on when interrupts
happen. Detect when this happens and let the test pass.

Signed-off-by: Andy Lutomirski <luto@...nel.org>
Signed-off-by: Borislav Petkov <bp@...e.de>
Link: https://lkml.kernel.org/r/7567fd44a1d60a9424f25b19a998f12149993b0d.1604346596.git.luto@kernel.org
Signed-off-by: Sasha Levin <sashal@...nel.org>
---
 tools/testing/selftests/x86/fsgsbase.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/x86/fsgsbase.c b/tools/testing/selftests/x86/fsgsbase.c
index f249e042b3b51..026cd644360f6 100644
--- a/tools/testing/selftests/x86/fsgsbase.c
+++ b/tools/testing/selftests/x86/fsgsbase.c
@@ -318,8 +318,8 @@ static void set_gs_and_switch_to(unsigned long local,
 		local = read_base(GS);
 
 		/*
-		 * Signal delivery seems to mess up weird selectors.  Put it
-		 * back.
+		 * Signal delivery is quite likely to change a selector
+		 * of 1, 2, or 3 back to 0 due to IRET being defective.
 		 */
 		asm volatile ("mov %0, %%gs" : : "rm" (force_sel));
 	} else {
@@ -337,6 +337,14 @@ static void set_gs_and_switch_to(unsigned long local,
 	if (base == local && sel_pre_sched == sel_post_sched) {
 		printf("[OK]\tGS/BASE remained 0x%hx/0x%lx\n",
 		       sel_pre_sched, local);
+	} else if (base == local && sel_pre_sched >= 1 && sel_pre_sched <= 3 &&
+		   sel_post_sched == 0) {
+		/*
+		 * IRET is misdesigned and will squash selectors 1, 2, or 3
+		 * to zero.  Don't fail the test just because this happened.
+		 */
+		printf("[OK]\tGS/BASE changed from 0x%hx/0x%lx to 0x%hx/0x%lx because IRET is defective\n",
+		       sel_pre_sched, local, sel_post_sched, base);
 	} else {
 		nerrs++;
 		printf("[FAIL]\tGS/BASE changed from 0x%hx/0x%lx to 0x%hx/0x%lx\n",
-- 
2.27.0