Currently syscall instruction is buggy on user mode X86_64,
the EIP is updated after do_syscall(), that is too late for
clone(). Because clone() will create a thread at the env->EIP
(the address of syscall insn), and then child thread enters
do_syscall() again, that is not expected. Sometimes it is tragic.
User mode syscall insn emulation is not used MSR, so the
action should be same to INT 0x80. INT 0x80 will update EIP in
do_interrupt(), ditto for syscall() for consistency.
Signed-off-by: Jincheng Miao <jm...@redhat.com>
---
linux-user/main.c | 1 -
target-i386/seg_helper.c | 4 ++--
2 files changed, 2 insertions(+), 3 deletions(-)
diff --git a/linux-user/main.c b/linux-user/main.c
index b453a39..472a16d 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -309,7 +309,6 @@ void cpu_loop(CPUX86State *env)
env->regs[8],
env->regs[9],
0, 0);
- env->eip = env->exception_next_eip;
break;
#endif
case EXCP0B_NOSEG:
diff --git a/target-i386/seg_helper.c b/target-i386/seg_helper.c
index 2d970d0..13eefba 100644
--- a/target-i386/seg_helper.c
+++ b/target-i386/seg_helper.c
@@ -1127,8 +1127,8 @@ static void do_interrupt_user(CPUX86State *env, int
intno, int is_int,
/* Since we emulate only user space, we cannot do more than
exiting the emulation with the suitable exception and error
- code */
- if (is_int) {
+ code. So update EIP for INT 0x80 and EXCP_SYSCALL. */
+ if (is_int || intno == EXCP_SYSCALL) {
env->eip = next_eip;
}
}
--
1.7.1
