[PATCH] rpcrt4: Handle absense of RPC flags in call_stubless_func() thunk.

[Zebediah Figura]

If Oi_HAS_RPCFLAGS is not set, then the stack size is at offset 4, not 8.

Signed-off-by: Zebediah Figura <z.figura12(a)gmail.com>
---
 dlls/rpcrt4/cproxy.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/dlls/rpcrt4/cproxy.c b/dlls/rpcrt4/cproxy.c
index 2f49bb5..39dd34e 100644
--- a/dlls/rpcrt4/cproxy.c
+++ b/dlls/rpcrt4/cproxy.c
@@ -64,7 +64,13 @@ __ASM_GLOBAL_FUNC(call_stubless_func,
                   "movl 8(%ecx),%edx\n\t"         /* info->FormatStringOffset */
                   "movzwl (%edx,%eax,2),%edx\n\t" /* FormatStringOffset[index] */
                   "addl 4(%ecx),%edx\n\t"         /* info->ProcFormatString + offset */
+                  "testb $0x08,1(%edx)\n\t"       /* Oi_HAS_RPCFLAGS */
+                  "jz 1f\n\t"
                   "movzwl 8(%edx),%eax\n\t"       /* arguments size */
+                  "jmp 2f\n\t"
+                  "1:\n\t"
+                  "movzwl 4(%edx),%eax\n\t"
+                  "2:\n\t"
                   "pushl %eax\n\t"
                   __ASM_CFI(".cfi_adjust_cfa_offset 4\n\t")
                   "leal 8(%esp),%eax\n\t"         /* &This */
-- 
2.7.4

[Huw Davies]

Signed-off-by: Huw Davies <huw(a)codeweavers.com>

[Huw Davies]

On Wed, Oct 03, 2018 at 12:08:21PM -0500, Zebediah Figura wrote:
> If Oi_HAS_RPCFLAGS is not set, then the stack size is at offset 4, not 8.
>
> Signed-off-by: Zebediah Figura <z.figura12(a)gmail.com>
> ---
>  dlls/rpcrt4/cproxy.c | 6 ++++++
>  1 file changed, 6 insertions(+)
>
> diff --git a/dlls/rpcrt4/cproxy.c b/dlls/rpcrt4/cproxy.c
> index 2f49bb5..39dd34e 100644
> --- a/dlls/rpcrt4/cproxy.c
> +++ b/dlls/rpcrt4/cproxy.c
> @@ -64,7 +64,13 @@ __ASM_GLOBAL_FUNC(call_stubless_func,
>                    "movl 8(%ecx),%edx\n\t"         /* info->FormatStringOffset */
>                    "movzwl (%edx,%eax,2),%edx\n\t" /* FormatStringOffset[index] */
>                    "addl 4(%ecx),%edx\n\t"         /* info->ProcFormatString + offset */
> +                  "testb $0x08,1(%edx)\n\t"       /* Oi_HAS_RPCFLAGS */
> +                  "jz 1f\n\t"
>                    "movzwl 8(%edx),%eax\n\t"       /* arguments size */
> +                  "jmp 2f\n\t"
> +                  "1:\n\t"
> +                  "movzwl 4(%edx),%eax\n\t"
> +                  "2:\n\t"
>                    "pushl %eax\n\t"
>                    __ASM_CFI(".cfi_adjust_cfa_offset 4\n\t")
>                    "leal 8(%esp),%eax\n\t"         /* &This */

Actually, thinking about this a bit more, it would be nice to avoid
the jumps.  It should be easy enough to mask off the flag bit and
then use that to calculate the correct offset from %edx.

Huw.

[Zebediah Figura]

On 04/10/18 07:51, Huw Davies wrote:
> On Wed, Oct 03, 2018 at 12:08:21PM -0500, Zebediah Figura wrote:
> > If Oi_HAS_RPCFLAGS is not set, then the stack size is at offset 4, not 8.
> >
> > Signed-off-by: Zebediah Figura <z.figura12(a)gmail.com>
> > ---
> >  dlls/rpcrt4/cproxy.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> >
> > diff --git a/dlls/rpcrt4/cproxy.c b/dlls/rpcrt4/cproxy.c
> > index 2f49bb5..39dd34e 100644
> > --- a/dlls/rpcrt4/cproxy.c
> > +++ b/dlls/rpcrt4/cproxy.c
> > @@ -64,7 +64,13 @@ __ASM_GLOBAL_FUNC(call_stubless_func,
> >                    "movl 8(%ecx),%edx\n\t"         /* info->FormatStringOffset */
> >                    "movzwl (%edx,%eax,2),%edx\n\t" /* FormatStringOffset[index] */
> >                    "addl 4(%ecx),%edx\n\t"         /* info->ProcFormatString + offset */
> > +                  "testb $0x08,1(%edx)\n\t"       /* Oi_HAS_RPCFLAGS */
> > +                  "jz 1f\n\t"
> >                    "movzwl 8(%edx),%eax\n\t"       /* arguments size */
> > +                  "jmp 2f\n\t"
> > +                  "1:\n\t"
> > +                  "movzwl 4(%edx),%eax\n\t"
> > +                  "2:\n\t"
> >                    "pushl %eax\n\t"
> >                    __ASM_CFI(".cfi_adjust_cfa_offset 4\n\t")
> >                    "leal 8(%esp),%eax\n\t"         /* &This */
>
> Actually, thinking about this a bit more, it would be nice to avoid
> the jumps.  It should be easy enough to mask off the flag bit and
> then use that to calculate the correct offset from %edx.
>
> Huw.

Sure, can do.