2006-10-24 23:19:41

by Joe Korty

[permalink] [raw]
Subject: [PATCH] strange work_notifysig code since 2.6.16

tree 9b5daef5280800a0006343a17f63072658d91a1d is surely wrong.

[PATCH] Make vm86 support optional
Jan 8, 2006, tree first appears in 2.6.16

If the branch to work_notifysig_86 is taken and CONFIG_VM86=n,
then entry.S drops into unrelated assembly code.

The branch to work_notifysig_v86, a few lines above the patch, says:
"returning to kernel-space or vm86-space"
which implies if vm86-space isn't being supported we still need the
branch & jumped-to code in order to handle the kernel case.

However, I don't understand this area all that well, so take this patch
as more of an indication of a possible problem area than as a true fix.

Signed-off-by: Joe Korty <[email protected]>

Index: 2.6.18.1/arch/i386/kernel/entry.S
===================================================================
--- 2.6.18.1.orig/arch/i386/kernel/entry.S 2006-09-19 23:42:06.000000000 -0400
+++ 2.6.18.1/arch/i386/kernel/entry.S 2006-10-24 19:08:36.000000000 -0400
@@ -457,7 +457,6 @@

ALIGN
work_notifysig_v86:
-#ifdef CONFIG_VM86
pushl %ecx # save ti_flags for do_notify_resume
CFI_ADJUST_CFA_OFFSET 4
call save_v86_state # %eax contains pt_regs pointer
@@ -467,7 +466,6 @@
xorl %edx, %edx
call do_notify_resume
jmp resume_userspace_sig
-#endif

# perform syscall exit tracing
ALIGN


2006-10-25 05:49:34

by Matt Mackall

[permalink] [raw]
Subject: Re: [PATCH] strange work_notifysig code since 2.6.16

On Tue, Oct 24, 2006 at 07:19:21PM -0400, Joe Korty wrote:
> tree 9b5daef5280800a0006343a17f63072658d91a1d is surely wrong.
>
> [PATCH] Make vm86 support optional
> Jan 8, 2006, tree first appears in 2.6.16
>
> If the branch to work_notifysig_86 is taken and CONFIG_VM86=n,
> then entry.S drops into unrelated assembly code.
>
> The branch to work_notifysig_v86, a few lines above the patch, says:
> "returning to kernel-space or vm86-space"
> which implies if vm86-space isn't being supported we still need the
> branch & jumped-to code in order to handle the kernel case.
>
> However, I don't understand this area all that well, so take this patch
> as more of an indication of a possible problem area than as a true fix.
>
> Signed-off-by: Joe Korty <[email protected]>
>
> Index: 2.6.18.1/arch/i386/kernel/entry.S
> ===================================================================
> --- 2.6.18.1.orig/arch/i386/kernel/entry.S 2006-09-19 23:42:06.000000000 -0400
> +++ 2.6.18.1/arch/i386/kernel/entry.S 2006-10-24 19:08:36.000000000 -0400
> @@ -457,7 +457,6 @@
>
> ALIGN
> work_notifysig_v86:
> -#ifdef CONFIG_VM86
> pushl %ecx # save ti_flags for do_notify_resume
> CFI_ADJUST_CFA_OFFSET 4
> call save_v86_state # %eax contains pt_regs pointer

I suspect this won't link with CONFIG_VM86 disabled because
save_v86_state goes away. I think we just need to move the #endif up a
few lines.

> @@ -467,7 +466,6 @@
> xorl %edx, %edx
> call do_notify_resume
> jmp resume_userspace_sig
> -#endif
>
> # perform syscall exit tracing
> ALIGN

--
Mathematics is the supreme nostalgia of our time.

2006-10-25 14:29:37

by Joe Korty

[permalink] [raw]
Subject: Re: [PATCH] strange work_notifysig code since 2.6.16

> I suspect this won't link with CONFIG_VM86 disabled because save_v86_state
> goes away. I think we just need to move the #endif up a few lines.

Hi Matt,
Since that also makes the 'then' and 'else' branches identical, perhaps
this patch would be better .. it eliminates the VM86 test entirely when
CONFIG_VM86=n.

Boot tested with CONFIG_VM86=y.

Regards,
Joe

The entry.S code at work_notifysig is surely wrong. It drops into unrelated
code if the branch to work_notifysig_v86 is taken, and CONFIG_VM86=n.

[PATCH] Make vm86 support optional
tree 9b5daef5280800a0006343a17f63072658d91a1d
pushed to git Jan 8, 2006, and first appears in 2.6.16

The 'fix' here is to also compile out the vm86 test & branch when
CONFIG_VM86=n.

Signed-off-by: Joe Korty <[email protected]>

Index: 2.6.18.1/arch/i386/kernel/entry.S
===================================================================
--- 2.6.18.1.orig/arch/i386/kernel/entry.S 2006-10-25 10:06:25.000000000 -0400
+++ 2.6.18.1/arch/i386/kernel/entry.S 2006-10-25 10:10:38.000000000 -0400
@@ -447,6 +447,7 @@

work_notifysig: # deal with pending signals and
# notify-resume requests
+#ifdef CONFIG_VM86
testl $VM_MASK, EFLAGS(%esp)
movl %esp, %eax
jne work_notifysig_v86 # returning to kernel-space or
@@ -457,17 +458,18 @@

ALIGN
work_notifysig_v86:
-#ifdef CONFIG_VM86
pushl %ecx # save ti_flags for do_notify_resume
CFI_ADJUST_CFA_OFFSET 4
call save_v86_state # %eax contains pt_regs pointer
popl %ecx
CFI_ADJUST_CFA_OFFSET -4
movl %eax, %esp
+#else
+ movl %esp, %eax
+#endif
xorl %edx, %edx
call do_notify_resume
jmp resume_userspace_sig
-#endif

# perform syscall exit tracing
ALIGN

2006-10-25 19:03:01

by Andi Kleen

[permalink] [raw]
Subject: Re: [PATCH] strange work_notifysig code since 2.6.16

Joe Korty <[email protected]> writes:

> > I suspect this won't link with CONFIG_VM86 disabled because save_v86_state
> > goes away. I think we just need to move the #endif up a few lines.
>
> Hi Matt,
> Since that also makes the 'then' and 'else' branches identical, perhaps
> this patch would be better .. it eliminates the VM86 test entirely when
> CONFIG_VM86=n.
>
> Boot tested with CONFIG_VM86=y.

Added.

-Andi