[Qemu-devel] [AMD64] Possible problem with the DF flag during the SYSCALL instruction

[Qemu-devel] [AMD64] Possible problem with the DF flag during the SYSCALL instruction

[Jakub Jermar]

Hi,

yesterday I tried to mask off the DF flag from the RFLAGS register 
during the SYSCALL instruction by specifying this bit in the SFMASK MSR 
register while running HelenOS.

With QEMU, this didn't work (but it worked in Simics, for instance), so 
I suspect that there is something wrong with handling RFLAGS.DF during 
SYSCALL. Anyone wants to have a look?

Regards,
Jakub

[PATCH]Re: [Qemu-devel] [AMD64] Possible problem with the DF flag during the SYSCALL instruction

[Jakub Jermar]

[-- Attachment #1: Type: text/plain, Size: 798 bytes --]

Jakub Jermar wrote:
> yesterday I tried to mask off the DF flag from the RFLAGS register 
> during the SYSCALL instruction by specifying this bit in the SFMASK MSR 
> register while running HelenOS.
> 
> With QEMU, this didn't work (but it worked in Simics, for instance), so 
> I suspect that there is something wrong with handling RFLAGS.DF during 
> SYSCALL. Anyone wants to have a look?

I have looked into the problem a little more and found out that the
culprit here is probably a missing update of the DF variable in the
helper_syscall() function, in target-i386/helper.c.  See the attached
patch.

The patch fixes the problem for me, but is probably not complete (i.e.
there might be other places that need a similar treatment), but I can't
really tell since I am not a qemu expert.

Jakub

[-- Attachment #2: patch --]
[-- Type: text/plain, Size: 382 bytes --]

--- helper.c	2008-01-06 20:38:45.000000000 +0100
+++ helper.c.new	2008-03-20 22:39:17.000000000 +0100
@@ -1008,6 +1008,7 @@
                                DESC_S_MASK |
                                DESC_W_MASK | DESC_A_MASK);
         env->eflags &= ~env->fmask;
+	DF = 1 - (2 * ((env->eflags >> 10) & 1));
         if (code64)
             env->eip = env->lstar;
         else