From: Jerome Marchand <jmarchan@redhat.com>
Date: Tue, 24 Jun 2008 14:56:12 +0200
Subject: [x86_64] ptrace: sign-extend orig_rax to 64 bits
Message-id: 4860EEEC.8030301@redhat.com
O-Subject: [RHEL5.3 Patch] x86_64, ptrace: sign-extend orig_rax to 64 bits
Bugzilla: 437882
RH-Acked-by: David S. Miller <davem@redhat.com>
RH-Acked-by: Chuck Ebbert <cebbert@redhat.com>
RH-Acked-by: Roland McGrath <roland@redhat.com>

Bugzilla:
https://bugzilla.redhat.com/show_bug.cgi?id=437882

Description:
GDB testsuite failure for x86_64 debugger running i386 debuggee.
GDB sets orig_rax to 0x00000000ffffffff which is not recognized by
kernel as -1.
That bug is revealed by the fix of 434998. It could not happen
before.

Solution:
Make ptrace always sign-extend orig_rax to 64 bits

Upstream status:
commit 84c6f6046c5a2189160a8f0dca8b90427bf690ea

Test status:
Built on all arch, tested on x86_64 using the reproducer provided on
bugzilla.

Regards,
Jerome

diff --git a/arch/x86_64/kernel/ptrace.c b/arch/x86_64/kernel/ptrace.c
index 2ad504e..053d9d9 100644
--- a/arch/x86_64/kernel/ptrace.c
+++ b/arch/x86_64/kernel/ptrace.c
@@ -279,6 +279,16 @@ static int putreg(struct task_struct *child,
 			return -EIO;
 		child->thread.gs = value;
 		return 0;
+	case offsetof(struct user_regs_struct, orig_rax):
+		/*
+		 * Orig_rax is really just a flag with small positive
+		 * and negative values, so make sure to always
+		 * sign-extend it from 32 bits so that it works
+		 * correctly regardless of whether we come from a
+		 * 32-bit environment or not.
+		 */
+		value = (long) (s32) value;
+		break;
 	case offsetof(struct user_regs_struct, eflags):
 		value &= FLAG_MASK;
 		tmp = get_stack_long(child, EFL_OFFSET);