From: Chris Lalancette <clalance@redhat.com>
Date: Thu, 3 Jul 2008 17:19:07 +0200
Subject: [x86_64] xen: fix syscall return when tracing
Message-id: 486CEDEB.8060802@redhat.com
O-Subject: [RHEL5 PATCH]: Fix syscall return when tracing
Bugzilla: 453394
RH-Acked-by: Bill Burns <bburns@redhat.com>
RH-Acked-by: Markus Armbruster <armbru@redhat.com>
RH-Acked-by: Jarod Wilson <jwilson@redhat.com>

All,
    Attached is a patch to fix BZ 453394, which was spun out of bz 442352.  The
reported problem was that on 64-bit, system calls above valid system calls
(specifically, NR_max_syscalls) were returning the number of the syscall, not
-ENOSYS.  After looking at entry-xen.S for a couple of minutes, the problem was
obvious; we were moving %rax (the syscall number) into the return area right
before returning, so the -ENOSYS was getting lost.  The attached patch fixes
this, and was committed in upstream linux-2.6.18-xen.hg today as c/s 583.

Note that the upstream Linux kernel proper is not affected by this bug, and
neither is the RHEL-4 Xen kernel, and neither is 32-bit RHEL-4, RHEL-5, or upstream.

This patch was tested by me with the following test program:

#include <stdio.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <sys/types.h>

int main()
{
        int foo;

        foo = syscall(289);

        fprintf(stderr, "Error is %d\n",foo);

        return 0;
}

Before this patch, compiling and running the above program would return 289;
after the patch, compiling and running the above program returns the proper -1
with an errno of ENOSYS.

Brew build is here: http://brewweb.devel.redhat.com/brew/taskinfo?taskID=1370049

Please review and ACK.

Chris Lalancette

 arch/x86_64/kernel/entry-xen.S |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86_64/kernel/entry-xen.S b/arch/x86_64/kernel/entry-xen.S
index f177e65..c6f2a23 100644
--- a/arch/x86_64/kernel/entry-xen.S
+++ b/arch/x86_64/kernel/entry-xen.S
@@ -336,10 +336,10 @@ tracesys:
 	LOAD_ARGS ARGOFFSET  /* reload args from stack in case ptrace changed it */
 	RESTORE_REST
 	cmpq $__NR_syscall_max,%rax
-	ja  1f
+	ja  int_ret_from_sys_call
 	movq %r10,%rcx	/* fixup for C */
 	call *sys_call_table(,%rax,8)
-1:	movq %rax,RAX-ARGOFFSET(%rsp)
+	movq %rax,RAX-ARGOFFSET(%rsp)
 	/* Use IRET because user could have changed frame */
 	jmp int_ret_from_sys_call
 	CFI_ENDPROC