]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
selftests/x86/mov_ss_trap: Fix the SYSENTER test
authorAndy Lutomirski <luto@kernel.org>
Wed, 20 Nov 2019 20:59:13 +0000 (12:59 -0800)
committerPeter Zijlstra <peterz@infradead.org>
Thu, 21 Nov 2019 20:55:51 +0000 (21:55 +0100)
For reasons that I haven't quite fully diagnosed, running
mov_ss_trap_32 on a 32-bit kernel results in an infinite loop in
userspace.  This appears to be because the hacky SYSENTER test
doesn't segfault as desired; instead it corrupts the program state
such that it infinite loops.

Fix it by explicitly clearing EBP before doing SYSENTER.  This will
give a more reliable segfault.

Fixes: 59c2a7226fc5 ("x86/selftests: Add mov_to_ss test")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
tools/testing/selftests/x86/mov_ss_trap.c

index 3c3a022654f36ee52c31a57f342083607a06fcda..6da0ac3f0135811321c163cf55cf6b3b5aea027b 100644 (file)
@@ -257,7 +257,8 @@ int main()
                        err(1, "sigaltstack");
                sethandler(SIGSEGV, handle_and_longjmp, SA_RESETHAND | SA_ONSTACK);
                nr = SYS_getpid;
-               asm volatile ("mov %[ss], %%ss; SYSENTER" : "+a" (nr)
+               /* Clear EBP first to make sure we segfault cleanly. */
+               asm volatile ("xorl %%ebp, %%ebp; mov %[ss], %%ss; SYSENTER" : "+a" (nr)
                              : [ss] "m" (ss) : "flags", "rcx"
 #ifdef __x86_64__
                                , "r11"