[PATCH 1/3] ntdll: Align stack pointer when calling KiUserCallbackDispatcher.

[Torge Matthies]

From: Torge Matthies

Overwatch hooks KiUserCallbackDispatcher and expects the stack pointer to be aligned to a multiple of 16 bytes,
instead of the usual 8-byte misalignment, otherwise it will crash on a misaligned movaps.

Fix this by aligning the stack pointer when calling the dispatcher and again inside the dispatcher.

Signed-off-by: Torge Matthies 
---
 dlls/ntdll/signal_x86_64.c      | 28 ++++++++++++++++++++++------
 dlls/ntdll/unix/signal_x86_64.c |  2 +-
 2 files changed, 23 insertions(+), 7 deletions(-)

diff --git a/dlls/ntdll/signal_x86_64.c b/dlls/ntdll/signal_x86_64.c
index b1ab4933b93..5af82a6a09f 100644
--- a/dlls/ntdll/signal_x86_64.c
+++ b/dlls/ntdll/signal_x86_64.c
@@ -654,12 +654,7 @@ __ASM_GLOBAL_FUNC( KiUserApcDispatcher,
                    "int3")

-/*******************************************************************
- *		KiUserCallbackDispatcher (NTDLL.@)
- *
- * FIXME: not binary compatible
- */
-void WINAPI KiUserCallbackDispatcher( ULONG id, void *args, ULONG len )
+void WINAPI user_callback_dispatcher( ULONG id, void *args, ULONG len )
 {
     NTSTATUS status;

@@ -678,6 +673,27 @@ void WINAPI KiUserCallbackDispatcher( ULONG id, void *args, ULONG len )
     RtlRaiseStatus( status );
 }

+/*******************************************************************
+ *		KiUserCallbackDispatcher (NTDLL.@)
+ *
+ * FIXME: not binary compatible
+ */
+#ifdef __x86_64__
+__ASM_GLOBAL_FUNC( KiUserCallbackDispatcher,
+                  "movq %rsp,%rbp\n\t"
+                  __ASM_SEH(".seh_setframe %rbp, 0\n\t")
+                  __ASM_CFI(".cfi_def_cfa rbp, 8\n\t")
+                  "andq $0xFFFFFFFFFFFFFFF0, %rsp\n\t"
+                  __ASM_SEH(".seh_endprologue\n\t")
+                  "call " __ASM_NAME("user_callback_dispatcher") "\n\t"
+                  "int3")
+#else
+void WINAPI DECLSPEC_HOTPATCH KiUserCallbackDispatcher( ULONG id, void *args, ULONG len )
+{
+    return user_callback_dispatcher( id, args, len );
+}
+#endif
+

static ULONG64 get_int_reg( CONTEXT *context, int reg )
 {
diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c
index 5787f1dc6f9..0526db9d762 100644
--- a/dlls/ntdll/unix/signal_x86_64.c
+++ b/dlls/ntdll/unix/signal_x86_64.c
@@ -2399,7 +2399,7 @@ NTSTATUS WINAPI KeUserModeCallback( ULONG id, const void *args, ULONG len, void
     if (!__wine_setjmpex( &callback_frame.jmpbuf, NULL ))
     {
         struct syscall_frame *frame = amd64_thread_data()->syscall_frame;
-        void *args_data = (void *)((frame->rsp - len) & ~15);
+        void *args_data = (void *)(((frame->rsp - len) & ~15) - 8);

memcpy( args_data, args, len );