[PATCH v3 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      | 18 ++++++++++++------
 dlls/ntdll/unix/signal_x86_64.c |  2 +-
 2 files changed, 13 insertions(+), 7 deletions(-)

diff --git a/dlls/ntdll/signal_x86_64.c b/dlls/ntdll/signal_x86_64.c
index b1ab4933b93..59dc20ed693 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 +673,17 @@ void WINAPI KiUserCallbackDispatcher( ULONG id, void *args, ULONG len )
     RtlRaiseStatus( status );
 }

+/*******************************************************************
+ *		KiUserCallbackDispatcher (NTDLL.@)
+ *
+ * FIXME: not binary compatible
+ */
+__ASM_GLOBAL_FUNC( KiUserCallbackDispatcher,
+                  "andq $0xFFFFFFFFFFFFFFF0, %rsp\n\t"
+                  __ASM_SEH(".seh_endprologue\n\t")
+                  "call " __ASM_NAME("user_callback_dispatcher") "\n\t"
+                  "int3")
+

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..db01bc3a736 100644
--- a/dlls/ntdll/unix/signal_x86_64.c
+++ b/dlls/ntdll/unix/signal_x86_64.c
@@ -2410,7 +2410,7 @@ NTSTATUS WINAPI KeUserModeCallback( ULONG id, const void *args, ULONG len, void
         callback_frame.frame.fs            = amd64_thread_data()->fs;
         callback_frame.frame.gs            = ds64_sel;
         callback_frame.frame.ss            = ds64_sel;
-        callback_frame.frame.rsp           = (ULONG_PTR)args_data - 0x28;
+        callback_frame.frame.rsp           = (ULONG_PTR)args_data - 0x30;
         callback_frame.frame.rip           = (ULONG_PTR)pKiUserCallbackDispatcher;
         callback_frame.frame.eflags        = 0x200;
         callback_frame.frame.restore_flags = CONTEXT_CONTROL | CONTEXT_INTEGER;