2006-05-19 22:09:09

by Aleksey Gorelov

Subject: [PATCH] fix broken vm86 interrupt/signal handling

Hi,

This patch
http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h
=c3ff8ec31c1249d268cd11390649768a12bec1b9 has broken vm86
interrupt/signal handling in case when vm86 is called from kernel space.
In this scenario, if signal is pending because of vm86 interrupt,
do_notify_resume/do_signal exits immediately due to user_mode() check,
without processing any signals. Thus, resume_userspace handler is
spinning in a tight loop with signal pending and TIF_SIGPENDING is set.
Previously everything worked Ok.

The following patch fixes the issue

Signed-off-by: Aleksey Gorelov <[email protected]>

--- linux-2.6.16/arch/i386/kernel/entry.S-old 2006-05-19
14:31:41.000000000 -0700
+++ linux-2.6.16/arch/i386/kernel/entry.S 2006-05-19
14:33:31.000000000 -0700
@@ -82,6 +82,12 @@
#define resume_kernel restore_nocheck
#endif

+#ifdef CONFIG_VM86
+#define resume_userspace_sig check_userspace
+#else
+#define resume_userspace_sig resume_userspace
+#endif
+
#define SAVE_ALL \
cld; \
pushl %es; \
@@ -143,6 +149,7 @@
preempt_stop
ret_from_intr:
GET_THREAD_INFO(%ebp)
+check_userspace:
movl EFLAGS(%esp), %eax # mix EFLAGS and CS
movb CS(%esp), %al
testl $(VM_MASK | 3), %eax
@@ -319,7 +326,7 @@
# vm86-space
xorl %edx, %edx
call do_notify_resume
- jmp resume_userspace
+ jmp resume_userspace_sig

ALIGN
work_notifysig_v86:
@@ -330,7 +337,7 @@
movl %eax, %esp
xorl %edx, %edx
call do_notify_resume
- jmp resume_userspace
+ jmp resume_userspace_sig
#endif

# perform syscall exit tracing

2006-05-20 02:29:51

by Arjan van de Ven

Subject: Re: [PATCH] fix broken vm86 interrupt/signal handling

On Fri, 2006-05-19 at 15:09 -0700, Aleksey Gorelov wrote:
> Hi,
>
> This patch
> http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h
> =c3ff8ec31c1249d268cd11390649768a12bec1b9 has broken vm86
> interrupt/signal handling in case when vm86 is called from kernel space.

can you point out where vm86 is called from kernel space?

2006-05-22 11:51:54

by Andi Kleen

Subject: Re: [PATCH] fix broken vm86 interrupt/signal handling

Arjan van de Ven <[email protected]> writes:

D> On Fri, 2006-05-19 at 15:09 -0700, Aleksey Gorelov wrote:
> > Hi,
> >
> > This patch
> > http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h
> > =c3ff8ec31c1249d268cd11390649768a12bec1b9 has broken vm86
> > interrupt/signal handling in case when vm86 is called from kernel space.
>
> can you point out where vm86 is called from kernel space?

Programs inside vm86 can run with segments that look like ring 0.
That is why i386 always checks (ring > 0 || vmmask in eflags) when it checks
for user mode.

I guess Aleksey meant that.

-Andi