path: root/2025.08.22_bylaws-wine_upstream-arm64ec_hack_v2.patch

From 5e385cd354cd4f1fc69ff4193b0b13b841402f59 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Mon, 11 Aug 2025 16:40:01 +0100
Subject: [PATCH 03/20] ntdll/tests: Test THREAD_CREATE_FLAGS_SKIP_LOADER_INIT
 flag.

---
 dlls/ntdll/tests/thread.c | 76 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 76 insertions(+)

diff --git a/dlls/ntdll/tests/thread.c b/dlls/ntdll/tests/thread.c
index 9fd23809488f..dce77cfc68c6 100644
--- a/dlls/ntdll/tests/thread.c
+++ b/dlls/ntdll/tests/thread.c
@@ -35,12 +35,15 @@ static NTSTATUS (WINAPI *pNtAllocateReserveObject)( HANDLE *, const OBJECT_ATTRI
 static NTSTATUS (WINAPI *pNtCreateThreadEx)( HANDLE *, ACCESS_MASK, OBJECT_ATTRIBUTES *,
                                              HANDLE, PRTL_THREAD_START_ROUTINE, void *,
                                              ULONG, ULONG_PTR, SIZE_T, SIZE_T, PS_ATTRIBUTE_LIST * );
+static NTSTATUS (WINAPI *pNtAllocateVirtualMemoryEx)(HANDLE, PVOID *, SIZE_T *, ULONG, ULONG,
+                                                     MEM_EXTENDED_PARAMETER *, ULONG);
 static NTSTATUS  (WINAPI *pNtSuspendProcess)(HANDLE process);
 static NTSTATUS  (WINAPI *pNtResumeProcess)(HANDLE process);
 static NTSTATUS  (WINAPI *pNtQueueApcThreadEx)(HANDLE handle, HANDLE reserve_handle, PNTAPCFUNC func,
                                                ULONG_PTR arg1, ULONG_PTR arg2, ULONG_PTR arg3);
 static NTSTATUS  (WINAPI *pNtQueueApcThreadEx2)(HANDLE handle, HANDLE reserve_handle, ULONG flags, PNTAPCFUNC func,
                                                 ULONG_PTR arg1, ULONG_PTR arg2, ULONG_PTR arg3);
+static NTSTATUS (WINAPI *pRtlWow64GetProcessMachines)(HANDLE,WORD*,WORD*);

 static int * (CDECL *p_errno)(void);

@@ -53,10 +56,12 @@ static void init_function_pointers(void)
     hdll = GetModuleHandleA( "ntdll.dll" );
     GET_FUNC( NtAllocateReserveObject );
     GET_FUNC( NtCreateThreadEx );
+    GET_FUNC( NtAllocateVirtualMemoryEx );
     GET_FUNC( NtSuspendProcess );
     GET_FUNC( NtQueueApcThreadEx );
     GET_FUNC( NtQueueApcThreadEx2 );
     GET_FUNC( NtResumeProcess );
+    GET_FUNC( RtlWow64GetProcessMachines );
     GET_FUNC( _errno );

     hdll = GetModuleHandleA( "kernel32.dll" );
@@ -441,6 +446,66 @@ static void test_skip_thread_attach(void)
     DeleteFileA(path_dll_local);
 }

+struct test_skip_load_init_args
+{
+    USHORT teb_same_teb_flags;
+};
+
+static ULONG patched_code[] =
+{
+    0xd282fdc2, /* mov x2, #0x17ee */
+    0x78626a41, /* ldrh w1, [x18, x2]   (NtCurrentTeb()->SameTebFlags) */
+    0x79000001, /* strh w1, [x0]        (args->teb_same_teb_flags)     */
+    0xd65f03c0, /* ret */
+};
+
+static void test_arm64_skip_loader_init(void)
+{
+    struct test_skip_load_init_args args;
+    HANDLE thread;
+    NTSTATUS status;
+    void *code_mem = NULL;
+#ifdef __x86_64__
+    MEM_EXTENDED_PARAMETER param = { 0 };
+    SIZE_T code_size = 0x10000;
+
+    param.Type = MemExtendedParameterAttributeFlags;
+    param.ULong64 = MEM_EXTENDED_PARAMETER_EC_CODE;
+    if (!pNtAllocateVirtualMemoryEx ||
+        pNtAllocateVirtualMemoryEx( GetCurrentProcess(), &code_mem, &code_size, MEM_RESERVE | MEM_COMMIT,
+                                    PAGE_EXECUTE_READWRITE, &param, 1 ))
+    {
+        trace("NtAllocateVirtualMemoryEx failed\n");
+        return;
+
+    }
+#else
+    code_mem = VirtualAlloc(NULL, 65536, MEM_RESERVE | MEM_COMMIT, PAGE_EXECUTE_READWRITE);
+    if (!code_mem) {
+        trace("VirtualAlloc failed\n");
+        return;
+    }
+#endif
+    if (!pNtCreateThreadEx)
+    {
+        win_skip( "NtCreateThreadEx is not available.\n" );
+        return;
+    }
+
+    memcpy(code_mem, patched_code, sizeof(patched_code));
+
+    status = pNtCreateThreadEx( &thread, THREAD_ALL_ACCESS, NULL, GetCurrentProcess(), (PRTL_THREAD_START_ROUTINE)code_mem,
+                                &args, THREAD_CREATE_FLAGS_SKIP_THREAD_ATTACH | THREAD_CREATE_FLAGS_SKIP_LOADER_INIT, 0, 0, 0, NULL );
+
+    todo_wine ok( status == STATUS_SUCCESS, "Got unexpected status %#lx.\n", status );
+
+    WaitForSingleObject( thread, INFINITE );
+
+    todo_wine ok( (args.teb_same_teb_flags & 0x4008) == 0x4008, "wrong value %x\n", args.teb_same_teb_flags );
+
+    CloseHandle( thread );
+}
+
 START_TEST(thread)
 {
     init_function_pointers();
@@ -464,4 +529,15 @@ START_TEST(thread)
     test_thread_bypass_process_freeze();
     test_NtQueueApcThreadEx();
     test_skip_thread_attach();
+
+    if (pRtlWow64GetProcessMachines)
+    {
+        USHORT current, native;
+        NTSTATUS status = pRtlWow64GetProcessMachines( GetCurrentProcess(), &current, &native );
+        if (!status && native == IMAGE_FILE_MACHINE_ARM64)
+        {
+            trace( "Running arm64 tests.\n" );
+            test_arm64_skip_loader_init();
+        }
+    }
 }

From 58c942999f94dc9ae91adb840c78fadad4244f51 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Thu, 27 Jul 2023 08:00:43 -0700
Subject: [PATCH 06/20] ntdll: Implement WOW64 thread suspension helpers.

---
 dlls/ntdll/loader.c         |  2 ++
 dlls/ntdll/ntdll.spec       |  1 +
 dlls/ntdll/ntdll_misc.h     |  1 +
 dlls/ntdll/signal_arm64.c   | 17 +++++++++++++++++
 dlls/ntdll/signal_arm64ec.c |  7 +++++++
 dlls/ntdll/signal_x86_64.c  |  7 +++++++
 include/winternl.h          |  1 +
 7 files changed, 36 insertions(+)

diff --git a/dlls/ntdll/loader.c b/dlls/ntdll/loader.c
index c6cc93855205..e34bf9bd0d04 100644
--- a/dlls/ntdll/loader.c
+++ b/dlls/ntdll/loader.c
@@ -4285,6 +4285,7 @@ static void build_wow64_main_module(void)
 static void (WINAPI *pWow64LdrpInitialize)( CONTEXT *ctx );

 void (WINAPI *pWow64PrepareForException)( EXCEPTION_RECORD *rec, CONTEXT *context ) = NULL;
+NTSTATUS (WINAPI *pWow64SuspendLocalThread)( HANDLE thread, ULONG *count ) = NULL;

 static void init_wow64( CONTEXT *context )
 {
@@ -4309,6 +4310,7 @@ static void init_wow64( CONTEXT *context )

         GET_PTR( Wow64LdrpInitialize );
         GET_PTR( Wow64PrepareForException );
+        GET_PTR( Wow64SuspendLocalThread );
 #undef GET_PTR
         imports_fixup_done = TRUE;
     }
diff --git a/dlls/ntdll/ntdll.spec b/dlls/ntdll/ntdll.spec
index 604c342ba350..ffcb7a0549cd 100644
--- a/dlls/ntdll/ntdll.spec
+++ b/dlls/ntdll/ntdll.spec
@@ -1136,6 +1136,7 @@
 @ stdcall -arch=win64 RtlWow64PushCrossProcessWorkOntoWorkList(ptr ptr ptr)
 @ stdcall -arch=win64 RtlWow64RequestCrossProcessHeavyFlush(ptr)
 @ stdcall -arch=win64 RtlWow64SetThreadContext(long ptr)
+@ stdcall -arch=win64 RtlWow64SuspendThread(long ptr)
 @ stub RtlWriteMemoryStream
 @ stdcall RtlWriteRegistryValue(long ptr wstr long ptr long)
 @ stub RtlZeroHeap
diff --git a/dlls/ntdll/ntdll_misc.h b/dlls/ntdll/ntdll_misc.h
index 99e0aceeef30..d8de8fb32e91 100644
--- a/dlls/ntdll/ntdll_misc.h
+++ b/dlls/ntdll/ntdll_misc.h
@@ -76,6 +76,7 @@ extern void WINAPI KiUserApcDispatcher(CONTEXT*,ULONG_PTR,ULONG_PTR,ULONG_PTR,PN
 extern void WINAPI KiUserCallbackDispatcher(ULONG,void*,ULONG);
 extern void WINAPI KiUserCallbackDispatcherReturn(void);
 extern void (WINAPI *pWow64PrepareForException)( EXCEPTION_RECORD *rec, CONTEXT *context );
+extern NTSTATUS (WINAPI *pWow64SuspendLocalThread)( HANDLE thread, ULONG *count );

 /* debug helpers */
 extern LPCSTR debugstr_us( const UNICODE_STRING *str );
diff --git a/dlls/ntdll/signal_arm64.c b/dlls/ntdll/signal_arm64.c
index bc981289ea5a..548f87a3a6ee 100644
--- a/dlls/ntdll/signal_arm64.c
+++ b/dlls/ntdll/signal_arm64.c
@@ -643,6 +643,23 @@ BOOLEAN WINAPI RtlIsProcessorFeaturePresent( UINT feature )
             user_shared_data->ProcessorFeatures[feature]);
 }

+/***********************************************************************
+ *              RtlWow64SuspendThread (NTDLL.@)
+ */
+NTSTATUS WINAPI RtlWow64SuspendThread( HANDLE thread, ULONG *count )
+{
+    THREAD_BASIC_INFORMATION tbi;
+
+    NTSTATUS ret = NtQueryInformationThread( thread, ThreadBasicInformation, &tbi, sizeof(tbi), NULL);
+    if (ret) return ret;
+
+    if (tbi.ClientId.UniqueProcess != NtCurrentTeb()->ClientId.UniqueProcess) {
+        FIXME( "Non-local process thread suspend\n" );
+        return STATUS_SUCCESS;
+    }
+
+    return pWow64SuspendLocalThread( thread, count );
+}

 /*************************************************************************
  *		RtlWalkFrameChain (NTDLL.@)
diff --git a/dlls/ntdll/signal_arm64ec.c b/dlls/ntdll/signal_arm64ec.c
index a93aae895540..8a6f3628097c 100644
--- a/dlls/ntdll/signal_arm64ec.c
+++ b/dlls/ntdll/signal_arm64ec.c
@@ -1782,6 +1782,13 @@ BOOLEAN WINAPI RtlIsProcessorFeaturePresent( UINT feature )
     return emulated_processor_features[feature];
 }

+/***********************************************************************
+ *              RtlWow64SuspendThread (NTDLL.@)
+ */
+NTSTATUS WINAPI RtlWow64SuspendThread( HANDLE thread, ULONG *count )
+{
+    return NtSuspendThread( thread, count );
+}

 /*************************************************************************
  *		RtlWalkFrameChain (NTDLL.@)
diff --git a/dlls/ntdll/signal_x86_64.c b/dlls/ntdll/signal_x86_64.c
index 1f184f4ea586..9f14365a5702 100644
--- a/dlls/ntdll/signal_x86_64.c
+++ b/dlls/ntdll/signal_x86_64.c
@@ -844,6 +844,13 @@ BOOLEAN WINAPI RtlIsProcessorFeaturePresent( UINT feature )
     return feature < PROCESSOR_FEATURE_MAX && user_shared_data->ProcessorFeatures[feature];
 }

+/***********************************************************************
+ *              RtlWow64SuspendThread (NTDLL.@)
+ */
+NTSTATUS WINAPI RtlWow64SuspendThread( HANDLE thread, ULONG *count )
+{
+    return NtSuspendThread( thread, count );
+}

 /*************************************************************************
  *		RtlWalkFrameChain (NTDLL.@)
diff --git a/include/winternl.h b/include/winternl.h
index 52d4703145c4..3613216eb24d 100644
--- a/include/winternl.h
+++ b/include/winternl.h
@@ -5186,6 +5186,7 @@ NTSYSAPI USHORT    WINAPI RtlWow64GetCurrentMachine(void);
 NTSYSAPI NTSTATUS  WINAPI RtlWow64GetProcessMachines(HANDLE,USHORT*,USHORT*);
 NTSYSAPI NTSTATUS  WINAPI RtlWow64GetSharedInfoProcess(HANDLE,BOOLEAN*,WOW64INFO*);
 NTSYSAPI NTSTATUS  WINAPI RtlWow64IsWowGuestMachineSupported(USHORT,BOOLEAN*);
+NTSYSAPI NTSTATUS  WINAPI RtlWow64SuspendThread(HANDLE,ULONG*);
 NTSYSAPI NTSTATUS  WINAPI RtlWriteRegistryValue(ULONG,PCWSTR,PCWSTR,ULONG,PVOID,ULONG);
 NTSYSAPI NTSTATUS  WINAPI RtlZombifyActivationContext(struct _ACTIVATION_CONTEXT *);
 NTSYSAPI NTSTATUS  WINAPI RtlpNtCreateKey(PHANDLE,ACCESS_MASK,const OBJECT_ATTRIBUTES*,ULONG,const UNICODE_STRING*,ULONG,PULONG);

From ce44c044fbd7916bd2c6bd509e5013ee334ce4ab Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Thu, 27 Jul 2023 08:20:59 -0700
Subject: [PATCH 07/20] wow64: Support running a BT module provided suspend
 routine.

Suspension in JITs cannot easily be handled on the unix side, so allow BT
modules to provide their own suspend helper.
---
 dlls/wow64/process.c  |  2 +-
 dlls/wow64/syscall.c  | 10 ++++++++++
 dlls/wow64/wow64.spec |  2 +-
 3 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/dlls/wow64/process.c b/dlls/wow64/process.c
index f82f3f2187ed..9b0c50e98303 100644
--- a/dlls/wow64/process.c
+++ b/dlls/wow64/process.c
@@ -1097,7 +1097,7 @@ NTSTATUS WINAPI wow64_NtSuspendThread( UINT *args )
     HANDLE handle = get_handle( &args );
     ULONG *count = get_ptr( &args );

-    return NtSuspendThread( handle, count );
+    return RtlWow64SuspendThread( handle, count );
 }


diff --git a/dlls/wow64/syscall.c b/dlls/wow64/syscall.c
index db925475bc4f..78b3522cf9a8 100644
--- a/dlls/wow64/syscall.c
+++ b/dlls/wow64/syscall.c
@@ -102,6 +102,7 @@ static void     (WINAPI *pBTCpuProcessInit)(void);
 static NTSTATUS (WINAPI *pBTCpuSetContext)(HANDLE,HANDLE,void *,void *);
 static void     (WINAPI *pBTCpuThreadInit)(void);
 static void     (WINAPI *pBTCpuSimulate)(void) __attribute__((used));
+static NTSTATUS (WINAPI *pBTCpuSuspendLocalThread)(HANDLE,ULONG *);
 static void *   (WINAPI *p__wine_get_unix_opcode)(void);
 static void *   (WINAPI *pKiRaiseUserExceptionDispatcher)(void);
 void     (WINAPI *pBTCpuFlushInstructionCache2)( const void *, SIZE_T ) = NULL;
@@ -846,6 +847,7 @@ static DWORD WINAPI process_init( RTL_RUN_ONCE *once, void *param, void **contex
     GET_PTR( BTCpuProcessInit );
     GET_PTR( BTCpuThreadInit );
     GET_PTR( BTCpuResetToConsistentState );
+    GET_PTR( BTCpuSuspendLocalThread );
     GET_PTR( BTCpuSetContext );
     GET_PTR( BTCpuSimulate );
     GET_PTR( BTCpuFlushInstructionCache2 );
@@ -1532,3 +1534,11 @@ NTSTATUS WINAPI Wow64RaiseException( int code, EXCEPTION_RECORD *rec )

     return STATUS_SUCCESS;
 }
+
+/**********************************************************************
+ *            Wow64SuspendLocalThread (wow64.@)
+ */
+NTSTATUS WINAPI Wow64SuspendLocalThread( HANDLE thread, ULONG *count )
+{
+    return pBTCpuSuspendLocalThread( thread, count );
+}
diff --git a/dlls/wow64/wow64.spec b/dlls/wow64/wow64.spec
index 5b4e565e5936..8f95dfcdbe52 100644
--- a/dlls/wow64/wow64.spec
+++ b/dlls/wow64/wow64.spec
@@ -22,7 +22,7 @@
 @ stub Wow64ShallowThunkAllocSecurityQualityOfService32TO64_FNC
 @ stub Wow64ShallowThunkSIZE_T32TO64
 @ stub Wow64ShallowThunkSIZE_T64TO32
-@ stub Wow64SuspendLocalThread
+@ stdcall Wow64SuspendLocalThread(long ptr)
 @ stdcall -norelay Wow64SystemServiceEx(long ptr)
 @ stub Wow64ValidateUserCallTarget
 @ stub Wow64ValidateUserCallTargetFilter

From eb5721e223b30fcd8c1a1025b3440a17d39d3c46 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Fri, 19 Apr 2024 20:25:30 +0000
Subject: [PATCH 08/20] HACK: Use x86 mono on aarch64.

---
 dlls/appwiz.cpl/addons.c | 4 ++--
 dlls/mscoree/metahost.c  | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/dlls/appwiz.cpl/addons.c b/dlls/appwiz.cpl/addons.c
index 5791b782a060..fbd7e0d8fecd 100644
--- a/dlls/appwiz.cpl/addons.c
+++ b/dlls/appwiz.cpl/addons.c
@@ -48,7 +48,7 @@ WINE_DEFAULT_DEBUG_CHANNEL(appwizcpl);
 #ifdef __i386__
 #define GECKO_ARCH "x86"
 #define GECKO_SHA "26cecc47706b091908f7f814bddb074c61beb8063318e9efc5a7f789857793d6"
-#elif defined(__x86_64__)
+#elif defined(__x86_64__) || defined(__aarch64__)
 #define GECKO_ARCH "x86_64"
 #define GECKO_SHA "e590b7d988a32d6aa4cf1d8aa3aa3d33766fdd4cf4c89c2dcc2095ecb28d066f"
 #else
@@ -57,7 +57,7 @@ WINE_DEFAULT_DEBUG_CHANNEL(appwizcpl);
 #endif

 #define MONO_VERSION "10.2.0"
-#if defined(__i386__) || defined(__x86_64__)
+#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__)
 #define MONO_ARCH "x86"
 #define MONO_SHA "4e1ed3f02e92d053133d03ddfbefcf6db4a4dc231a9aed3367b17117a88847d8"
 #else
diff --git a/dlls/mscoree/metahost.c b/dlls/mscoree/metahost.c
index 0e34d7388a0f..a4ff9fd90268 100644
--- a/dlls/mscoree/metahost.c
+++ b/dlls/mscoree/metahost.c
@@ -711,7 +711,7 @@ HRESULT ICLRRuntimeInfo_GetRuntimeHost(ICLRRuntimeInfo *iface, RuntimeHost **res

 #ifdef __i386__
 static const WCHAR libmono2_arch_dll[] = {'\\','b','i','n','\\','l','i','b','m','o','n','o','-','2','.','0','-','x','8','6','.','d','l','l',0};
-#elif defined(__x86_64__)
+#elif defined(__x86_64__) || defined(__aarch64__)
 static const WCHAR libmono2_arch_dll[] = {'\\','b','i','n','\\','l','i','b','m','o','n','o','-','2','.','0','-','x','8','6','_','6','4','.','d','l','l',0};
 #else
 static const WCHAR libmono2_arch_dll[] = {'\\','b','i','n','\\','l','i','b','m','o','n','o','-','2','.','0','.','d','l','l',0};

From e0e23983b170b085e0b3d78f6f03e993cebd3504 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Tue, 23 Apr 2024 12:58:04 +0000
Subject: [PATCH 09/20] configure: Enable vcruntime140_1 and dpnsvcr for arm64x
 builds.

---
 configure.ac                            | 4 ++--
 dlls/vcruntime140_1/vcruntime140_1.spec | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/configure.ac b/configure.ac
index 3647015114ff..9e32170e5ecc 100644
--- a/configure.ac
+++ b/configure.ac
@@ -2371,7 +2371,7 @@ esac

 dnl *** Check for modules to disable by default

-enable_vcruntime140_1=${enable_vcruntime140_1:-x86_64,arm64ec}
+enable_vcruntime140_1=${enable_vcruntime140_1:-x86_64,arm64ec,aarch64}

 if test -n "$PE_ARCHS"
 then
@@ -2401,7 +2401,7 @@ enable_wuauserv=${enable_wuauserv:-$HOST_ARCH}
 dnl Explicitly enable some programs for arm64ec
 enable_cmd=${enable_cmd:-yes}
 enable_dllhost=${enable_dllhost:-yes}
-enable_dpnsvr=${enable_dpnsvr:-i386,x86_64,arm64ec}
+enable_dpnsvr=${enable_dpnsvr:-i386,x86_64,arm64ec,aarch64}
 enable_dxdiag=${enable_dxdiag:-yes}
 enable_msiexec=${enable_msiexec:-yes}
 enable_netsh=${enable_netsh:-yes}
diff --git a/dlls/vcruntime140_1/vcruntime140_1.spec b/dlls/vcruntime140_1/vcruntime140_1.spec
index 0b126fa95d46..af295940a61a 100644
--- a/dlls/vcruntime140_1/vcruntime140_1.spec
+++ b/dlls/vcruntime140_1/vcruntime140_1.spec
@@ -1,3 +1,3 @@
-@ cdecl __CxxFrameHandler4(ptr long ptr ptr)
+@ cdecl -arch=arm64ec,x86_64 __CxxFrameHandler4(ptr long ptr ptr)
 @ stub __NLG_Dispatch2
 @ stub __NLG_Return2

From a85a6e410ad6395f8a43d857ff9b31222080971f Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Mon, 19 Aug 2024 20:29:50 +0000
Subject: [PATCH 10/20] include: Add arm64 SVE xstate headers.

---
 include/winnt.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/include/winnt.h b/include/winnt.h
index e8a47fec963d..80329281ea4b 100644
--- a/include/winnt.h
+++ b/include/winnt.h
@@ -1473,6 +1473,10 @@ typedef struct _KNONVOLATILE_CONTEXT_POINTERS
 #define XSTATE_MASK_LEGACY                  (XSTATE_MASK_LEGACY_FLOATING_POINT | XSTATE_MASK_LEGACY_SSE)
 #define XSTATE_MASK_GSSE                    (1 << XSTATE_GSSE)

+#define XSTATE_ARM64_SVE 2
+
+#define XSTATE_MASK_ARM64_SVE (1 << XSTATE_ARM64_SVE)
+
 typedef struct _XSTATE_FEATURE
 {
     ULONG Offset;
@@ -1513,6 +1517,13 @@ typedef struct _XSAVE_AREA_HEADER
 }
 XSAVE_AREA_HEADER, *PXSAVE_AREA_HEADER;

+typedef struct _XSAVE_ARM64_SVE_HEADER {
+    ULONG VectorLength;
+    ULONG VectorRegisterOffset;
+    ULONG PredicateRegisterOffset;
+    ULONG Reserved[5];
+} XSAVE_ARM64_SVE_HEADER, *PXSAVE_ARM64_SVE_HEADER;
+
 typedef struct _YMMCONTEXT
 {
     M128A Ymm0;
@@ -1693,6 +1704,7 @@ typedef struct _KNONVOLATILE_CONTEXT_POINTERS
 #define CONTEXT_ARM64_FLOATING_POINT  (CONTEXT_ARM64 | 0x00000004)
 #define CONTEXT_ARM64_DEBUG_REGISTERS (CONTEXT_ARM64 | 0x00000008)
 #define CONTEXT_ARM64_X18       (CONTEXT_ARM64 | 0x00000010)
+#define CONTEXT_ARM64_XSTATE    (CONTEXT_ARM64 | 0x00000020)
 #define CONTEXT_ARM64_FULL (CONTEXT_ARM64_CONTROL | CONTEXT_ARM64_INTEGER | CONTEXT_ARM64_FLOATING_POINT)
 #define CONTEXT_ARM64_ALL  (CONTEXT_ARM64_FULL | CONTEXT_ARM64_DEBUG_REGISTERS | CONTEXT_ARM64_X18)

@@ -1972,6 +1984,7 @@ typedef struct DECLSPEC_ALIGN(16) _ARM64EC_NT_CONTEXT
 #define CONTEXT_INTEGER CONTEXT_ARM64_INTEGER
 #define CONTEXT_FLOATING_POINT CONTEXT_ARM64_FLOATING_POINT
 #define CONTEXT_DEBUG_REGISTERS CONTEXT_ARM64_DEBUG_REGISTERS
+#define CONTEXT_XSTATE CONTEXT_ARM64_XSTATE
 #define CONTEXT_FULL CONTEXT_ARM64_FULL
 #define CONTEXT_ALL CONTEXT_ARM64_ALL
 #define CONTEXT_RET_TO_GUEST CONTEXT_ARM64_RET_TO_GUEST

From 61fab5fde12c31560b78190900bc99cdd36a6061 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Wed, 21 Aug 2024 22:26:28 +0000
Subject: [PATCH 11/20] FEX specific AVX support for arm64ec

---
 dlls/ntdll/signal_arm64ec.c | 56 ++++++++++++++++++++++++-------------
 dlls/ntdll/signal_x86_64.c  |  2 +-
 dlls/ntdll/unix/system.c    | 19 +++++++++++++
 dlls/ntdll/unwind.h         | 16 +++++++++++
 include/winnt.h             |  5 ++--
 5 files changed, 75 insertions(+), 23 deletions(-)

diff --git a/dlls/ntdll/signal_arm64ec.c b/dlls/ntdll/signal_arm64ec.c
index 8a6f3628097c..3d53dc91459f 100644
--- a/dlls/ntdll/signal_arm64ec.c
+++ b/dlls/ntdll/signal_arm64ec.c
@@ -36,6 +36,11 @@
 #include "wine/debug.h"
 #include "ntsyscalls.h"

+union ARM64EC_NT_XCONTEXT {
+    ARM64EC_NT_CONTEXT context;
+    BYTE buffer[0x800];
+};
+
 WINE_DEFAULT_DEBUG_CHANNEL(seh);
 WINE_DECLARE_DEBUG_CHANNEL(relay);

@@ -1234,7 +1239,11 @@ NTSTATUS call_seh_handlers( EXCEPTION_RECORD *rec, CONTEXT *orig_context )
  */
 void dispatch_emulation( ARM64_NT_CONTEXT *arm_ctx )
 {
-    context_arm_to_x64( get_arm64ec_cpu_area()->ContextAmd64, arm_ctx );
+    ARM64EC_NT_CONTEXT *context = get_arm64ec_cpu_area()->ContextAmd64;
+    CONTEXT_EX *xctx;
+
+    RtlInitializeExtendedContext( context, ctx_flags_arm_to_x64( arm_ctx->ContextFlags), &xctx );
+    context_arm_to_x64( context, arm_ctx );
     get_arm64ec_cpu_area()->InSimulation = 1;
     pBeginSimulation();
 }
@@ -1264,11 +1273,13 @@ static void dispatch_syscall( ARM64_NT_CONTEXT *context )
 }


-static void * __attribute__((used)) prepare_exception_arm64ec( EXCEPTION_RECORD *rec, ARM64EC_NT_CONTEXT *context, ARM64_NT_CONTEXT *arm_ctx )
+static void * __attribute__((used)) prepare_exception_arm64ec( EXCEPTION_RECORD *rec, union ARM64EC_NT_XCONTEXT *context, ARM64_NT_CONTEXT *arm_ctx )
 {
+    CONTEXT_EX *xctx;
     if (rec->ExceptionCode == STATUS_EMULATION_SYSCALL) dispatch_syscall( arm_ctx );
-    context_arm_to_x64( context, arm_ctx );
-    if (pResetToConsistentState) pResetToConsistentState( rec, &context->AMD64_Context, arm_ctx );
+    RtlInitializeExtendedContext( context, ctx_flags_arm_to_x64( arm_ctx->ContextFlags ), &xctx );
+    context_arm_to_x64( &context->context, arm_ctx );
+    if (pResetToConsistentState) pResetToConsistentState( rec, &context->context.AMD64_Context, arm_ctx );
     /* call x64 dispatcher if the thunk or the function pointer was modified */
     if (pWow64PrepareForException || memcmp( KiUserExceptionDispatcher_thunk, KiUserExceptionDispatcher_orig,
                                              sizeof(KiUserExceptionDispatcher_orig) ))
@@ -1283,12 +1294,13 @@ void __attribute__((naked)) KiUserExceptionDispatcher( EXCEPTION_RECORD *rec, CO
 {
     asm( ".seh_proc \"#KiUserExceptionDispatcher\"\n\t"
          ".seh_context\n\t"
-         "sub sp, sp, #0x4d0\n\t"       /* sizeof(ARM64EC_NT_CONTEXT) */
-         ".seh_stackalloc 0x4d0\n\t"
+         "sub sp, sp, #0xcd0\n\t"       /* sizeof(union ARM64EC_NT_XCONTEXT) */
+         ".seh_stackalloc 0xcd0\n\t"
          ".seh_endprologue\n\t"
-         "add x0, sp, #0x3b0+0x4d0\n\t" /* rec */
+         "add x0, sp, #0xcd0\n\t"
+         "add x0, x0, #0x3b0\n\t"       /* rec */
          "mov x1, sp\n\t"               /* context */
-         "add x2, sp, #0x4d0\n\t"       /* arm_ctx (context + 1) */
+         "add x2, sp, #0xcd0\n\t"       /* arm_ctx (context + 1) */
          "bl \"#prepare_exception_arm64ec\"\n\t"
          "cbz x0, 1f\n\t"
          /* bypass exit thunk to avoid messing up the stack */
@@ -1296,8 +1308,9 @@ void __attribute__((naked)) KiUserExceptionDispatcher( EXCEPTION_RECORD *rec, CO
          "ldr x16, [x16, #:lo12:__os_arm64x_dispatch_call_no_redirect]\n\t"
          "mov x9, x0\n\t"
          "blr x16\n"
-         "1:\tadd x0, sp, #0x3b0+0x4d0\n\t" /* rec */
-         "mov x1, sp\n\t"                   /* context */
+         "1:\tadd x0, sp, #0xcd0\n\t"
+         "add x0, x0, #0x3b0\n\t"       /* rec */
+         "mov x1, sp\n\t"               /* context */
          "bl #dispatch_exception\n\t"
          "brk #1\n\t"
          ".seh_endproc" );
@@ -1311,11 +1324,12 @@ static void __attribute__((used)) dispatch_apc( void (CALLBACK *func)(ULONG_PTR,
                                                 ULONG_PTR arg1, ULONG_PTR arg2, ULONG_PTR arg3,
                                                 BOOLEAN alertable, ARM64_NT_CONTEXT *arm_ctx )
 {
-    ARM64EC_NT_CONTEXT context;
-
-    context_arm_to_x64( &context, arm_ctx );
-    func( arg1, arg2, arg3, &context.AMD64_Context );
-    NtContinue( &context.AMD64_Context, alertable );
+    union ARM64EC_NT_XCONTEXT context;
+    CONTEXT_EX *xctx;
+    RtlInitializeExtendedContext( &context, ctx_flags_arm_to_x64( arm_ctx->ContextFlags), &xctx );
+    context_arm_to_x64( &context.context, arm_ctx );
+    func( arg1, arg2, arg3, &context.context.AMD64_Context );
+    NtContinue( &context.context.AMD64_Context, alertable );
 }
 __ASM_GLOBAL_FUNC( "#KiUserApcDispatcher",
                    ".seh_context\n\t"
@@ -2070,7 +2084,9 @@ void __attribute__((naked)) RtlUserThreadStart( PRTL_THREAD_START_ROUTINE entry,
  */
 void WINAPI LdrInitializeThunk( CONTEXT *arm_context, ULONG_PTR unk2, ULONG_PTR unk3, ULONG_PTR unk4 )
 {
-    ARM64EC_NT_CONTEXT context;
+    union ARM64EC_NT_XCONTEXT context;
+    CONTEXT_EX *xctx;
+    RtlInitializeExtendedContext( &context, ctx_flags_arm_to_x64( arm_context->ContextFlags), &xctx );

     if (!__os_arm64x_check_call)
     {
@@ -2081,10 +2097,10 @@ void WINAPI LdrInitializeThunk( CONTEXT *arm_context, ULONG_PTR unk2, ULONG_PTR
         __os_arm64x_set_x64_information = LdrpSetX64Information;
     }

-    context_arm_to_x64( &context, (ARM64_NT_CONTEXT *)arm_context );
-    loader_init( &context.AMD64_Context, (void **)&context.X0 );
-    TRACE_(relay)( "\1Starting thread proc %p (arg=%p)\n", (void *)context.X0, (void *)context.X1 );
-    NtContinue( &context.AMD64_Context, TRUE );
+    context_arm_to_x64( &context.context, (ARM64_NT_CONTEXT *)arm_context );
+    loader_init( &context.context.AMD64_Context, (void **)&context.context.X0 );
+    TRACE_(relay)( "\1Starting thread proc %p (arg=%p)\n", (void *)context.context.X0, (void *)context.context.X1 );
+    NtContinue( &context.context.AMD64_Context, TRUE );
 }


diff --git a/dlls/ntdll/signal_x86_64.c b/dlls/ntdll/signal_x86_64.c
index 9f14365a5702..ab95aa87c2c8 100644
--- a/dlls/ntdll/signal_x86_64.c
+++ b/dlls/ntdll/signal_x86_64.c
@@ -83,7 +83,7 @@ __ASM_GLOBAL_FUNC( "EXP+#KiUserExceptionDispatcher",
                    "test %rax,%rax\n\t"
                    "jz 1f\n\t"
                    "subq $0x28,%rsp\n\t"
-                   "leaq 0x30+0x3b0+0x4d0(%rsp),%rcx\n\t" /* rec */
+                   "leaq 0x30+0x3b0+0xcd0(%rsp),%rcx\n\t" /* rec */
                    "leaq 0x30(%rsp),%rdx\n\t"             /* context */
                    "call *%rax\n"
                    "addq $0x28,%rsp\n"
diff --git a/dlls/ntdll/unix/system.c b/dlls/ntdll/unix/system.c
index 342e443f5e7b..24d8c10f0fa1 100644
--- a/dlls/ntdll/unix/system.c
+++ b/dlls/ntdll/unix/system.c
@@ -597,6 +597,23 @@ static ULONGLONG get_cpu_features(void)
     return 0;  /* FIXME */
 }

+static void init_xstate_features( XSTATE_CONFIGURATION *xstate )
+{
+    xstate->EnabledFeatures = (1 << XSTATE_LEGACY_FLOATING_POINT) | (1 << XSTATE_LEGACY_SSE) | (1 << XSTATE_AVX);
+    xstate->EnabledVolatileFeatures = xstate->EnabledFeatures;
+    xstate->AllFeatureSize = 0x340;
+
+    xstate->OptimizedSave = 0;
+    xstate->CompactionEnabled = 0;
+
+    xstate->Features[0].Size = xstate->AllFeatures[0] = offsetof(XSAVE_FORMAT, XmmRegisters);
+    xstate->Features[1].Size = xstate->AllFeatures[1] = sizeof(M128A) * 16;
+    xstate->Features[1].Offset = xstate->Features[0].Size;
+    xstate->Features[2].Offset = 0x240;
+    xstate->Features[2].Size = 0x100;
+    xstate->Size = 0x340;
+}
+
 void init_shared_data_cpuinfo( KUSER_SHARED_DATA *data )
 {
     BOOLEAN *features = data->ProcessorFeatures;
@@ -679,6 +696,8 @@ void init_shared_data_cpuinfo( KUSER_SHARED_DATA *data )
             break;
         }
     }
+
+    init_xstate_features( &data->XState );
 }

 #endif /* End architecture specific feature detection for CPUs */
diff --git a/dlls/ntdll/unwind.h b/dlls/ntdll/unwind.h
index faca8a259fb6..71a4d23da043 100644
--- a/dlls/ntdll/unwind.h
+++ b/dlls/ntdll/unwind.h
@@ -36,6 +36,7 @@ static inline ULONG ctx_flags_x64_to_arm( ULONG flags )
     if (flags & CONTEXT_AMD64_CONTROL) ret |= CONTEXT_ARM64_CONTROL;
     if (flags & CONTEXT_AMD64_INTEGER) ret |= CONTEXT_ARM64_INTEGER;
     if (flags & CONTEXT_AMD64_FLOATING_POINT) ret |= CONTEXT_ARM64_FLOATING_POINT;
+    if (flags & CONTEXT_AMD64_XSTATE) ret |= CONTEXT_ARM64_FEX_YMMSTATE;
     return ret;
 }

@@ -47,6 +48,7 @@ static inline ULONG ctx_flags_arm_to_x64( ULONG flags )
     if (flags & CONTEXT_ARM64_CONTROL) ret |= CONTEXT_AMD64_CONTROL;
     if (flags & CONTEXT_ARM64_INTEGER) ret |= CONTEXT_AMD64_INTEGER;
     if (flags & CONTEXT_ARM64_FLOATING_POINT) ret |= CONTEXT_AMD64_FLOATING_POINT;
+    if (flags & CONTEXT_ARM64_FEX_YMMSTATE) ret |= CONTEXT_AMD64_XSTATE;
     return ret;
 }

@@ -166,6 +168,13 @@ static inline void context_x64_to_arm( ARM64_NT_CONTEXT *arm_ctx, const ARM64EC_
     fpcsr = mxcsr_to_fpcsr( ec_ctx->AMD64_MxCsr );
     arm_ctx->Fpcr = fpcsr;
     arm_ctx->Fpsr = fpcsr >> 32;
+
+    if ((ec_ctx->ContextFlags & CONTEXT_XSTATE) == CONTEXT_XSTATE)
+    {
+        CONTEXT_EX *ec_xctx = (CONTEXT_EX *)(ec_ctx + 1);
+        YMMCONTEXT *ec_ymm = RtlLocateExtendedFeature( ec_xctx, XSTATE_AVX, NULL );
+        memcpy( arm_ctx->V + 16, ec_ymm, sizeof(*ec_ymm) );
+    }
 }

 static inline void context_arm_to_x64( ARM64EC_NT_CONTEXT *ec_ctx, const ARM64_NT_CONTEXT *arm_ctx )
@@ -216,6 +225,13 @@ static inline void context_arm_to_x64( ARM64EC_NT_CONTEXT *ec_ctx, const ARM64_N
     ec_ctx->X17_3 = arm_ctx->X17 >> 48;

     memcpy( ec_ctx->V, arm_ctx->V, sizeof(ec_ctx->V) );
+
+    if ((arm_ctx->ContextFlags & CONTEXT_ARM64_FEX_YMMSTATE) == CONTEXT_ARM64_FEX_YMMSTATE)
+    {
+        CONTEXT_EX *ec_xctx = (CONTEXT_EX *)(ec_ctx + 1);
+        YMMCONTEXT *ec_ymm = RtlLocateExtendedFeature( ec_xctx, XSTATE_AVX, NULL );
+        memcpy( ec_ymm, arm_ctx->V + 16, sizeof(*ec_ymm) );
+    }
 }

 #endif /* __aarch64__ || __arm64ec__ */
diff --git a/include/winnt.h b/include/winnt.h
index 80329281ea4b..a914996eb61f 100644
--- a/include/winnt.h
+++ b/include/winnt.h
@@ -1703,8 +1703,9 @@ typedef struct _KNONVOLATILE_CONTEXT_POINTERS
 #define CONTEXT_ARM64_INTEGER         (CONTEXT_ARM64 | 0x00000002)
 #define CONTEXT_ARM64_FLOATING_POINT  (CONTEXT_ARM64 | 0x00000004)
 #define CONTEXT_ARM64_DEBUG_REGISTERS (CONTEXT_ARM64 | 0x00000008)
-#define CONTEXT_ARM64_X18       (CONTEXT_ARM64 | 0x00000010)
-#define CONTEXT_ARM64_XSTATE    (CONTEXT_ARM64 | 0x00000020)
+#define CONTEXT_ARM64_X18             (CONTEXT_ARM64 | 0x00000010)
+#define CONTEXT_ARM64_XSTATE          (CONTEXT_ARM64 | 0x00000020)
+#define CONTEXT_ARM64_FEX_YMMSTATE   (CONTEXT_ARM64 | 0x00000040)
 #define CONTEXT_ARM64_FULL (CONTEXT_ARM64_CONTROL | CONTEXT_ARM64_INTEGER | CONTEXT_ARM64_FLOATING_POINT)
 #define CONTEXT_ARM64_ALL  (CONTEXT_ARM64_FULL | CONTEXT_ARM64_DEBUG_REGISTERS | CONTEXT_ARM64_X18)


From 510598df5bf0ad91ce6e2c2b25c6a469041ce13f Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Wed, 17 Apr 2024 23:44:33 +0000
Subject: [PATCH 12/20] Use FEX wow64/ARM64EC dll names

---
 dlls/ntdll/loader.c  | 3 +--
 dlls/wow64/syscall.c | 2 +-
 loader/wine.inf.in   | 4 ++--
 3 files changed, 4 insertions(+), 5 deletions(-)

diff --git a/dlls/ntdll/loader.c b/dlls/ntdll/loader.c
index e34bf9bd0d04..476b6bf740b4 100644
--- a/dlls/ntdll/loader.c
+++ b/dlls/ntdll/loader.c
@@ -4235,7 +4235,7 @@ static void load_arm64ec_module(void)
     ULONG buffer[16];
     KEY_VALUE_PARTIAL_INFORMATION *info = (KEY_VALUE_PARTIAL_INFORMATION *)buffer;
     UNICODE_STRING nameW = RTL_CONSTANT_STRING( L"\\Registry\\Machine\\Software\\Microsoft\\Wow64\\amd64" );
-    WCHAR module[64] = L"C:\\windows\\system32\\xtajit64.dll";
+    WCHAR module[64] = L"C:\\windows\\system32\\libarm64ecfex.dll";
     OBJECT_ATTRIBUTES attr;
     WINE_MODREF *wm;
     NTSTATUS status;
@@ -4319,7 +4319,6 @@ static void init_wow64( CONTEXT *context )
     pWow64LdrpInitialize( context );
 }

-
 #else

 void *Wow64Transition = NULL;
diff --git a/dlls/wow64/syscall.c b/dlls/wow64/syscall.c
index 78b3522cf9a8..264b3bf58cc1 100644
--- a/dlls/wow64/syscall.c
+++ b/dlls/wow64/syscall.c
@@ -752,7 +752,7 @@ static const WCHAR *get_cpu_dll_name(void)
     {
     case IMAGE_FILE_MACHINE_I386:
         RtlInitUnicodeString( &nameW, L"\\Registry\\Machine\\Software\\Microsoft\\Wow64\\x86" );
-        ret = (native_machine == IMAGE_FILE_MACHINE_ARM64 ? L"xtajit.dll" : L"wow64cpu.dll");
+        ret = (native_machine == IMAGE_FILE_MACHINE_ARM64 ? L"libwow64fex.dll" : L"wow64cpu.dll");
         break;
     case IMAGE_FILE_MACHINE_ARMNT:
         RtlInitUnicodeString( &nameW, L"\\Registry\\Machine\\Software\\Microsoft\\Wow64\\arm" );
diff --git a/loader/wine.inf.in b/loader/wine.inf.in
index 000ba36bf369..ab29d3a55fee 100644
--- a/loader/wine.inf.in
+++ b/loader/wine.inf.in
@@ -433,9 +433,9 @@ HKLM,%CurrentVersion%,"ProgramFilesDir (x86)",,"%16426%"
 HKLM,%CurrentVersion%,"CommonFilesDir (x86)",,"%16428%"
 HKLM,%CurrentVersion%,"ProgramFilesDir (Arm)",,"C:\Program Files (Arm)"
 HKLM,%CurrentVersion%,"CommonFilesDir (Arm)",,"C:\Program Files (Arm)\Common Files"
-HKLM,Software\Microsoft\Wow64\amd64,,2,"xtajit64.dll"
+HKLM,Software\Microsoft\Wow64\amd64,,2,"libarm64ecfex.dll"
 HKLM,Software\Microsoft\Wow64\arm,,2,"wowarmhw.dll"
-HKLM,Software\Microsoft\Wow64\x86,,2,"xtajit.dll"
+HKLM,Software\Microsoft\Wow64\x86,,2,"libwow64fex.dll"

 [CurrentVersionWow64.ntx86]
 HKLM,%CurrentVersion%,"ProgramFilesDir (x86)",,"%16422%"

From b379b4caf4f3be5faab4addcb937f9854dab4401 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Thu, 23 Jan 2025 01:04:14 +0000
Subject: [PATCH 13/20] ntdll: Add an interface to map the FEX stats shm.

---
 dlls/ntdll/unix/virtual.c | 53 +++++++++++++++++++++++++++++++++++++++
 include/winternl.h        |  7 ++++++
 2 files changed, 60 insertions(+)

diff --git a/dlls/ntdll/unix/virtual.c b/dlls/ntdll/unix/virtual.c
index 4b4d9d6cd7f6..e43b50180fc2 100644
--- a/dlls/ntdll/unix/virtual.c
+++ b/dlls/ntdll/unix/virtual.c
@@ -241,6 +241,9 @@ static void *preload_reserve_end;
 static BOOL force_exec_prot;  /* whether to force PROT_EXEC on all PROT_READ mmaps */
 static BOOL enable_write_exceptions;  /* raise exception on writes to executable memory */

+#define FEX_STATS_SHM_MAX_SIZE 0x400000
+static void *fex_stats_shm;
+
 struct range_entry
 {
     void *base;
@@ -5941,6 +5944,49 @@ static unsigned int get_memory_image_info( HANDLE process, LPCVOID addr, MEMORY_
     return status;
 }

+#if defined(linux) && defined(__aarch64__)
+NTSTATUS get_memory_fex_stats_shm( HANDLE process, LPCVOID addr, MEMORY_FEX_STATS_SHM_INFORMATION *info,
+                                   SIZE_T len, SIZE_T *res_len)
+{
+    char buf[0x20];
+    int fd;
+    int oflag = O_RDWR;
+
+    if (len != sizeof(*info)) return STATUS_INFO_LENGTH_MISMATCH;
+    if (process != GetCurrentProcess()) return STATUS_INVALID_HANDLE;
+
+    sprintf( buf, "fex-%d-stats", getpid() );
+
+    if (!fex_stats_shm) {
+        fex_stats_shm = mmap( NULL, FEX_STATS_SHM_MAX_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS |
+                              MAP_NORESERVE, -1, 0 );
+        if (fex_stats_shm == MAP_FAILED) {
+            fex_stats_shm = NULL;
+            return STATUS_INTERNAL_ERROR;
+        }
+
+        oflag |= O_CREAT | O_TRUNC;
+    }
+
+    fd = shm_open( buf, oflag, S_IRWXU | S_IRWXG | S_IRWXO );
+    if (fd == -1) return STATUS_INTERNAL_ERROR;
+
+    if (ftruncate( fd, info->map_size )) goto err;
+
+    if (mmap( fex_stats_shm, info->map_size, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, fd,
+              0 ) == MAP_FAILED) goto err;
+
+    close( fd );
+
+    info->shm_base = fex_stats_shm;
+    *res_len = len;
+    return STATUS_SUCCESS;
+
+err:
+    close( fd );
+    return STATUS_INTERNAL_ERROR;
+}
+#endif

 /***********************************************************************
  *             NtQueryVirtualMemory   (NTDLL.@)
@@ -5986,6 +6032,13 @@ NTSTATUS WINAPI NtQueryVirtualMemory( HANDLE process, LPCVOID addr,
             }
             return STATUS_INVALID_HANDLE;

+        case MemoryFexStatsShm:
+#if defined(linux) && defined(__aarch64__)
+            return get_memory_fex_stats_shm( process, addr, buffer, len, res_len );
+#else
+            FIXME("MemoryFexStatsShm unsupported\n");
+            break;
+#endif
         default:
             FIXME("(%p,%p,info_class=%d,%p,%ld,%p) Unknown information class\n",
                   process, addr, info_class, buffer, len, res_len);
diff --git a/include/winternl.h b/include/winternl.h
index 3613216eb24d..19c7c80d4cd2 100644
--- a/include/winternl.h
+++ b/include/winternl.h
@@ -2385,6 +2385,7 @@ typedef enum _MEMORY_INFORMATION_CLASS {
     MemoryWineUnixFuncs = 1000,
     MemoryWineUnixWow64Funcs,
 #endif
+    MemoryFexStatsShm = 2000,
 } MEMORY_INFORMATION_CLASS;

 typedef struct _MEMORY_SECTION_NAME
@@ -2451,6 +2452,12 @@ typedef struct _MEMORY_IMAGE_INFORMATION
     };
 } MEMORY_IMAGE_INFORMATION, *PMEMORY_IMAGE_INFORMATION;

+typedef struct _MEMORY_FEX_STATS_SHM_INFORMATION
+{
+    void *shm_base;
+    SIZE_T map_size;
+} MEMORY_FEX_STATS_SHM_INFORMATION, *PMEMORY_FEX_STATS_SHM_INFORMATION;
+
 typedef enum _MUTANT_INFORMATION_CLASS
 {
     MutantBasicInformation

From be5bc3cf442fda67bc8200a0230f1a4bdd292807 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Thu, 10 Jul 2025 16:09:08 +0100
Subject: [PATCH 14/20] ntdll: Add FEX specific Asahi TSO support

---
 dlls/ntdll/unix/process.c | 36 ++++++++++++++++++++++++++++++++++++
 include/winternl.h        |  1 +
 2 files changed, 37 insertions(+)

diff --git a/dlls/ntdll/unix/process.c b/dlls/ntdll/unix/process.c
index fdb81845a7e9..8802a640ab6c 100644
--- a/dlls/ntdll/unix/process.c
+++ b/dlls/ntdll/unix/process.c
@@ -46,6 +46,9 @@
 #ifdef HAVE_SYS_PARAM_H
 # include <sys/param.h>
 #endif
+#ifdef HAVE_SYS_PRCTL_H
+# include <sys/prctl.h>
+#endif
 #ifdef HAVE_SYS_QUEUE_H
 # include <sys/queue.h>
 #endif
@@ -1063,6 +1066,35 @@ void fill_vm_counters( VM_COUNTERS_EX *pvmi, int unix_pid )

 #endif

+static BOOL set_hardware_tso( BOOL enable ) {
+#ifdef HAVE_PRCTL
+#ifndef PR_GET_MEM_MODEL
+#define PR_GET_MEM_MODEL 0x6d4d444c
+#endif
+#ifndef PR_SET_MEM_MODEL
+#define PR_SET_MEM_MODEL 0x4d4d444c
+#endif
+#ifndef PR_SET_MEM_MODEL_DEFAULT
+#define PR_SET_MEM_MODEL_DEFAULT 0
+#endif
+#ifndef PR_SET_MEM_MODEL_TSO
+#define PR_SET_MEM_MODEL_TSO 1
+#endif
+    if (enable)
+    {
+        int ret = prctl( PR_GET_MEM_MODEL, 0, 0, 0, 0 );
+        if (ret == PR_SET_MEM_MODEL_DEFAULT)
+            return !prctl( PR_SET_MEM_MODEL, PR_SET_MEM_MODEL_TSO, 0, 0, 0 );
+        return ret == PR_SET_MEM_MODEL_TSO;
+    }
+
+    prctl( PR_SET_MEM_MODEL, PR_SET_MEM_MODEL_DEFAULT, 0, 0, 0 );
+    return TRUE;
+#else
+    return FALSE;
+#endif
+}
+
 #define UNIMPLEMENTED_INFO_CLASS(c) \
     case c: \
         FIXME( "(process=%p) Unimplemented information class: " #c "\n", handle); \
@@ -1830,6 +1862,10 @@ NTSTATUS WINAPI NtSetInformationProcess( HANDLE handle, PROCESSINFOCLASS class,
         SERVER_END_REQ;
         break;

+    case ProcessFexHardwareTso:
+        if (size != sizeof(BOOL)) return STATUS_INFO_LENGTH_MISMATCH;
+        return set_hardware_tso( *(BOOL *)info ) ? STATUS_SUCCESS : STATUS_NOT_SUPPORTED;
+
     case ProcessPowerThrottlingState:
         FIXME( "ProcessPowerThrottlingState - stub\n" );
         return STATUS_SUCCESS;
diff --git a/include/winternl.h b/include/winternl.h
index 19c7c80d4cd2..c37f2829d3f3 100644
--- a/include/winternl.h
+++ b/include/winternl.h
@@ -1982,6 +1982,7 @@ typedef enum _PROCESSINFOCLASS {
     ProcessWineLdtCopy,
     ProcessWineGrantAdminToken,
 #endif
+    ProcessFexHardwareTso = 2000,
 } PROCESSINFOCLASS;

 #define MEM_EXECUTE_OPTION_DISABLE   0x01

From d27ac683b2157caadcdd4441973c7840196b23b7 Mon Sep 17 00:00:00 2001
From: Jacek Caban <jacek@codeweavers.com>
Date: Thu, 10 Jul 2025 18:55:42 +0100
Subject: [PATCH 15/20] HACK: ntdll: Manually define exit thunks for
 mem{cpy/move/set}

---
 dlls/winecrt0/arm64ec.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)

diff --git a/dlls/winecrt0/arm64ec.c b/dlls/winecrt0/arm64ec.c
index a3af06b380d4..a8c82facbdab 100644
--- a/dlls/winecrt0/arm64ec.c
+++ b/dlls/winecrt0/arm64ec.c
@@ -103,4 +103,45 @@ asm( "\t.section .rdata,\"dr\"\n"
      "\t.rva __os_arm64x_helper7\n"
      "\t.rva __os_arm64x_helper8\n" );

+/*
+ * HACK: Provide an exit thunk for memcpy/memmove/memset to work around
+ * https://github.com/llvm/llvm-project/issues/101355
+ */
+asm( "\t.def $iexit_thunk$cdecl$i8$i8i8i8; .scl 2; .type 32; .endef\n"
+     "\t.section .wowthk$aa,\"xr\",discard,$iexit_thunk$cdecl$i8$i8i8i8\n"
+     "\t.globl  $iexit_thunk$cdecl$i8$i8i8i8\n"
+     "\t.p2align 2\n"
+     "$iexit_thunk$cdecl$i8$i8i8i8:\n"
+     "\t.seh_proc $iexit_thunk$cdecl$i8$i8i8i8\n"
+     "\tsub     sp, sp, #48\n"
+     "\t.seh_stackalloc 48\n"
+     "\tstp     x29, x30, [sp, #32]\n"
+     "\t.seh_save_fplr 32\n"
+     "\tadd     x29, sp, #32\n"
+     "\t.seh_add_fp 32\n"
+     "\t.seh_endprologue\n"
+     "\tadrp    x8, __os_arm64x_dispatch_call_no_redirect\n"
+     "\tldr     x16, [x8, :lo12:__os_arm64x_dispatch_call_no_redirect]\n"
+     "\tblr     x16\n"
+     "\tmov     x0, x8\n"
+     "\t.seh_startepilogue\n"
+     "\tldp     x29, x30, [sp, #32]\n"
+     "\t.seh_save_fplr  32\n"
+     "\tadd     sp, sp, #48\n"
+     "\t.seh_stackalloc 48\n"
+     "\t.seh_endepilogue\n"
+     "\tret\n"
+     "\t.seh_endproc\n"
+
+     "\t.section .hybmp$x,\"yi\"\n"
+     "\t.symidx memcpy\n"
+     "\t.symidx $iexit_thunk$cdecl$i8$i8i8i8\n"
+     "\t.word 4\n"
+     "\t.symidx memmove\n"
+     "\t.symidx $iexit_thunk$cdecl$i8$i8i8i8\n"
+     "\t.word 4\n"
+     "\t.symidx memset\n"
+     "\t.symidx $iexit_thunk$cdecl$i8$i8i8i8\n"
+     "\t.word 4\n" );
+
 #endif /* __arm64ec__ */

From 5b993192c90aa390b3d470d7b39ea4aa39e164cf Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Tue, 15 Jul 2025 17:09:15 +0100
Subject: [PATCH 16/20] HACK: Avoid delay importing rpcrt4

Works around broken variadic exit thunks.
---
 dlls/advapi32/Makefile.in | 4 ++--
 dlls/sechost/Makefile.in  | 3 +--
 2 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/dlls/advapi32/Makefile.in b/dlls/advapi32/Makefile.in
index 57bb25feb9c6..671cb02fcd0c 100644
--- a/dlls/advapi32/Makefile.in
+++ b/dlls/advapi32/Makefile.in
@@ -1,8 +1,8 @@
 EXTRADEFS = -D_ADVAPI32_
 MODULE    = advapi32.dll
 IMPORTLIB = advapi32
-IMPORTS   = kernelbase sechost msvcrt
-DELAYIMPORTS = cryptsp rpcrt4 user32 userenv
+IMPORTS   = kernelbase sechost msvcrt rpcrt4
+DELAYIMPORTS = cryptsp user32 userenv

 SOURCES = \
 	advapi.c \
diff --git a/dlls/sechost/Makefile.in b/dlls/sechost/Makefile.in
index ec1eb100683c..25439e94090f 100644
--- a/dlls/sechost/Makefile.in
+++ b/dlls/sechost/Makefile.in
@@ -1,7 +1,6 @@
 MODULE    = sechost.dll
 IMPORTLIB = sechost
-IMPORTS   = kernelbase
-DELAYIMPORTS = rpcrt4
+IMPORTS   = kernelbase rpcrt4

 SOURCES = \
 	plugplay.idl \

From d9005c80b84708e8a2f32340be367013b7902472 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Wed, 6 Aug 2025 21:56:24 +0100
Subject: [PATCH 17/20] ntdll: Implement FEX-specific kernel-side unaligned
 atomic handling.

---
 dlls/ntdll/unix/process.c | 15 +++++++++++++++
 include/winternl.h        |  6 ++++++
 2 files changed, 21 insertions(+)

diff --git a/dlls/ntdll/unix/process.c b/dlls/ntdll/unix/process.c
index 8802a640ab6c..4b9af67058e5 100644
--- a/dlls/ntdll/unix/process.c
+++ b/dlls/ntdll/unix/process.c
@@ -1095,6 +1095,17 @@ static BOOL set_hardware_tso( BOOL enable ) {
 #endif
 }

+static BOOL set_unalign_atomic_mode( ULONG64 flags ) {
+#ifdef HAVE_PRCTL
+#ifndef PR_ARM64_SET_UNALIGN_ATOMIC
+#define PR_ARM64_SET_UNALIGN_ATOMIC 0x46455849
+#endif
+    return !prctl( PR_ARM64_SET_UNALIGN_ATOMIC, flags, 0, 0, 0 );;
+#else
+    return FALSE;
+#endif
+}
+
 #define UNIMPLEMENTED_INFO_CLASS(c) \
     case c: \
         FIXME( "(process=%p) Unimplemented information class: " #c "\n", handle); \
@@ -1866,6 +1877,10 @@ NTSTATUS WINAPI NtSetInformationProcess( HANDLE handle, PROCESSINFOCLASS class,
         if (size != sizeof(BOOL)) return STATUS_INFO_LENGTH_MISMATCH;
         return set_hardware_tso( *(BOOL *)info ) ? STATUS_SUCCESS : STATUS_NOT_SUPPORTED;

+    case ProcessFexUnalignAtomic:
+        if (size != sizeof(ULONG64)) return STATUS_INFO_LENGTH_MISMATCH;
+        return set_unalign_atomic_mode( *(ULONG64 *)info ) ? STATUS_SUCCESS : STATUS_NOT_SUPPORTED;
+
     case ProcessPowerThrottlingState:
         FIXME( "ProcessPowerThrottlingState - stub\n" );
         return STATUS_SUCCESS;
diff --git a/include/winternl.h b/include/winternl.h
index c37f2829d3f3..144b94a12f58 100644
--- a/include/winternl.h
+++ b/include/winternl.h
@@ -1983,8 +1983,14 @@ typedef enum _PROCESSINFOCLASS {
     ProcessWineGrantAdminToken,
 #endif
     ProcessFexHardwareTso = 2000,
+    ProcessFexUnalignAtomic,
 } PROCESSINFOCLASS;

+// These match the prctl flag values
+#define FEX_UNALIGN_ATOMIC_EMULATE       (1ULL << 0)
+#define FEX_UNALIGN_ATOMIC_BACKPATCH     (1ULL << 1)
+#define FEX_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS    (1ULL << 2)
+
 #define MEM_EXECUTE_OPTION_DISABLE   0x01
 #define MEM_EXECUTE_OPTION_ENABLE    0x02
 #define MEM_EXECUTE_OPTION_DISABLE_THUNK_EMULATION 0x04

From b5a04cd8ee5158bc21eb84842cf4d5db266c34fa Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Thu, 7 Aug 2025 20:57:03 +0100
Subject: [PATCH 18/20] WIP: ntdll: ARM64EC suspend support

---
 dlls/ntdll/ntdll_misc.h        |  1 +
 dlls/ntdll/signal_arm64ec.c    |  1 +
 dlls/ntdll/signal_x86_64.c     |  8 +++++
 dlls/ntdll/unix/signal_arm64.c | 57 ++++++++++++++++++++++++++++++----
 4 files changed, 61 insertions(+), 6 deletions(-)

diff --git a/dlls/ntdll/ntdll_misc.h b/dlls/ntdll/ntdll_misc.h
index d8de8fb32e91..ab34a4b1c53e 100644
--- a/dlls/ntdll/ntdll_misc.h
+++ b/dlls/ntdll/ntdll_misc.h
@@ -171,6 +171,7 @@ extern IMAGE_ARM64EC_METADATA *arm64ec_get_module_metadata( HMODULE module );
 extern void arm64ec_update_hybrid_metadata( void *module, IMAGE_NT_HEADERS *nt,
                                             const IMAGE_ARM64EC_METADATA *metadata );
 extern void invoke_arm64ec_syscall(void);
+extern void arm64ec_suspend_point(void);

 extern void *__os_arm64x_check_call;
 extern void *__os_arm64x_check_icall;
diff --git a/dlls/ntdll/signal_arm64ec.c b/dlls/ntdll/signal_arm64ec.c
index 3d53dc91459f..0657ae41fb8a 100644
--- a/dlls/ntdll/signal_arm64ec.c
+++ b/dlls/ntdll/signal_arm64ec.c
@@ -90,6 +90,7 @@ static inline BOOL enter_syscall_callback(void)
 static inline void leave_syscall_callback(void)
 {
     get_arm64ec_cpu_area()->InSyscallCallback = 0;
+    if (get_arm64ec_cpu_area()->SuspendDoorbell && *get_arm64ec_cpu_area()->SuspendDoorbell) arm64ec_suspend_point();
 }

 /**********************************************************************
diff --git a/dlls/ntdll/signal_x86_64.c b/dlls/ntdll/signal_x86_64.c
index ab95aa87c2c8..13a868434358 100644
--- a/dlls/ntdll/signal_x86_64.c
+++ b/dlls/ntdll/signal_x86_64.c
@@ -74,6 +74,14 @@ __ASM_GLOBAL_FUNC( invoke_arm64ec_syscall,
                    "pushq %r10\n\t"         /* and return to syscall thunk */
                    "ret" )

+/**************************************************************************
+ *		arm64ec_suspend_point
+ *
+ * x64 stub to support cooperative suspend when leaving a syscall callack.
+ */
+__ASM_GLOBAL_FUNC( arm64ec_suspend_point,
+                   "ret" )
+
 /*******************************************************************
  *		KiUserExceptionDispatcher (NTDLL.@)
  */
diff --git a/dlls/ntdll/unix/signal_arm64.c b/dlls/ntdll/unix/signal_arm64.c
index f5bd47d597ab..00b62b415c5e 100644
--- a/dlls/ntdll/unix/signal_arm64.c
+++ b/dlls/ntdll/unix/signal_arm64.c
@@ -65,6 +65,18 @@ WINE_DEFAULT_DEBUG_CHANNEL(seh);
 #define NTDLL_DWARF_H_NO_UNWINDER
 #include "dwarf.h"

+struct arm64_thread_data
+{
+    BOOL suspend_pending;
+};
+
+C_ASSERT( sizeof(struct arm64_thread_data) <= sizeof(((struct ntdll_thread_data *)0)->cpu_data) );
+
+static inline struct arm64_thread_data *arm64_thread_data(void)
+{
+    return (struct arm64_thread_data *)ntdll_get_thread_data()->cpu_data;
+}
+
 /***********************************************************************
  * signal context platform-specific definitions
  */
@@ -186,6 +198,10 @@ struct syscall_frame

 C_ASSERT( sizeof( struct syscall_frame ) == 0x330 );

+static BOOL is_arm64ec_suspend_doorbell_valid(void)
+{
+    return is_arm64ec() && NtCurrentTeb()->ChpeV2CpuAreaInfo && NtCurrentTeb()->ChpeV2CpuAreaInfo->SuspendDoorbell;
+}

 /***********************************************************************
  *           context_init_empty_xstate
@@ -210,7 +226,6 @@ void set_process_instrumentation_callback( void *callback )
     if (callback) FIXME( "Not supported.\n" );
 }

-
 /***********************************************************************
  *           syscall_frame_fixup_for_fastpath
  *
@@ -319,7 +334,18 @@ NTSTATUS signal_set_full_context( CONTEXT *context )
     struct syscall_frame *frame = get_syscall_frame();
     NTSTATUS status = NtSetContextThread( GetCurrentThread(), context );

-    if (!status && (context->ContextFlags & CONTEXT_INTEGER) == CONTEXT_INTEGER)
+    if (is_arm64ec_suspend_doorbell_valid() && arm64_thread_data()->suspend_pending)
+    {
+        CONTEXT suspend_context;
+        *NtCurrentTeb()->ChpeV2CpuAreaInfo->SuspendDoorbell = 0;
+        arm64_thread_data()->suspend_pending = FALSE;
+        suspend_context.ContextFlags = CONTEXT_FULL | CONTEXT_EXCEPTION_REPORTING; /* TODO: check */
+        NtGetContextThread( GetCurrentThread(), &suspend_context );
+        wait_suspend( &suspend_context );
+        NtSetContextThread( GetCurrentThread(), &suspend_context );
+    }
+
+    if (!status && (context->ContextFlags & CONTEXT_INTEGER) == CONTEXT_INTEGER) /* TODO: also check with susp */
         frame->restore_flags |= CONTEXT_INTEGER;

     if (is_arm64ec() && !is_ec_code( frame->pc ))
@@ -410,12 +436,20 @@ NTSTATUS WINAPI NtGetContextThread( HANDLE handle, CONTEXT *context )
 {
     struct syscall_frame *frame = get_syscall_frame();
     DWORD needed_flags = context->ContextFlags & ~CONTEXT_ARM64;
-    BOOL self = (handle == GetCurrentThread());
+    THREAD_BASIC_INFORMATION info;
+    NTSTATUS ret;
+    BOOL self;
+
+    NtQueryInformationThread( handle, ThreadBasicInformation, &info, sizeof(info), NULL );
+    self = HandleToULong( info.ClientId.UniqueThread ) == GetCurrentThreadId();

     if (!self)
     {
-        NTSTATUS ret = get_thread_context( handle, context, &self, IMAGE_FILE_MACHINE_ARM64 );
-        if (ret || !self) return ret;
+        /* Avoid exposing JIT code pointers to other processes on ARM64EC */
+        if (is_arm64ec()) NtSuspendThread( handle, NULL );
+        ret = get_thread_context( handle, context, &self, IMAGE_FILE_MACHINE_ARM64 );
+        if (is_arm64ec()) NtResumeThread( handle, NULL );
+        return ret;
     }

     if (needed_flags & CONTEXT_INTEGER)
@@ -1294,6 +1328,14 @@ static void usr1_handler( int signal, siginfo_t *siginfo, void *sigcontext )
     ucontext_t *ucontext = sigcontext;
     CONTEXT context;

+    if (is_arm64ec_suspend_doorbell_valid() &&
+        (NtCurrentTeb()->ChpeV2CpuAreaInfo->InSimulation || NtCurrentTeb()->ChpeV2CpuAreaInfo->InSyscallCallback))
+    {
+        *NtCurrentTeb()->ChpeV2CpuAreaInfo->SuspendDoorbell = 1;
+        arm64_thread_data()->suspend_pending = TRUE;
+        return;
+    }
+
     if (is_inside_syscall( SP_sig(ucontext) ))
     {
         context.ContextFlags = CONTEXT_FULL | CONTEXT_EXCEPTION_REQUEST;
@@ -1448,11 +1490,14 @@ void syscall_dispatcher_return_slowpath(void)
  */
 void init_syscall_frame( LPTHREAD_START_ROUTINE entry, void *arg, BOOL suspend, TEB *teb )
 {
-    struct syscall_frame *frame = ((struct ntdll_thread_data *)&teb->GdiTebBatch)->syscall_frame;
+    struct ntdll_thread_data *thread_data = (struct ntdll_thread_data *)&teb->GdiTebBatch;
+    struct syscall_frame *frame = thread_data->syscall_frame;
     CONTEXT *ctx, context = { CONTEXT_ALL };
     I386_CONTEXT *i386_context;
     ARM_CONTEXT *arm_context;

+    ((struct arm64_thread_data *)(thread_data->cpu_data))->suspend_pending = FALSE;
+
     context.X0  = (DWORD64)entry;
     context.X1  = (DWORD64)arg;
     context.X18 = (DWORD64)teb;

From d5fbbfc4f29bf2688b45069371cd5b7e1baa1bf7 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Mon, 11 Aug 2025 21:56:19 +0100
Subject: [PATCH 19/20] ntdll: Support cross-process wow64 suspend on ARM64.

---
 dlls/ntdll/signal_arm64.c | 62 +++++++++++++++++++++++++++++++++++----
 1 file changed, 56 insertions(+), 6 deletions(-)

diff --git a/dlls/ntdll/signal_arm64.c b/dlls/ntdll/signal_arm64.c
index 548f87a3a6ee..ac861b6dc0ce 100644
--- a/dlls/ntdll/signal_arm64.c
+++ b/dlls/ntdll/signal_arm64.c
@@ -643,19 +643,69 @@ BOOLEAN WINAPI RtlIsProcessorFeaturePresent( UINT feature )
             user_shared_data->ProcessorFeatures[feature]);
 }

+static void suspend_remote_breakin( HANDLE thread )
+{
+    ULONG count;
+    NTSTATUS status = pWow64SuspendLocalThread( thread, &count );
+    if (status >= 0) status = count;
+    NtTerminateThread( GetCurrentThread(),  status );
+}
+
 /***********************************************************************
  *              RtlWow64SuspendThread (NTDLL.@)
  */
 NTSTATUS WINAPI RtlWow64SuspendThread( HANDLE thread, ULONG *count )
 {
+    HANDLE thread_dup;
     THREAD_BASIC_INFORMATION tbi;
+    NTSTATUS status = NtDuplicateObject( NtCurrentProcess(), thread, NtCurrentProcess(), &thread_dup,
+                                         THREAD_QUERY_INFORMATION | THREAD_SUSPEND_RESUME, 0, 0 );
+    if (status) return status;
+    status = NtQueryInformationThread( thread_dup, ThreadBasicInformation, &tbi, sizeof(tbi), NULL);
+    NtClose( thread_dup );
+    if (status) return status;
+
+    if (tbi.ClientId.UniqueProcess != NtCurrentTeb()->ClientId.UniqueProcess)
+    {
+        HANDLE process;
+        HANDLE suspender_thread;
+        HANDLE remote_suspendee_thread;
+        OBJECT_ATTRIBUTES attr = { .Length = sizeof(attr) };
+
+        status = NtOpenProcess( &process, PROCESS_CREATE_THREAD | PROCESS_DUP_HANDLE, &attr, &tbi.ClientId );
+        if (status) return status;
+
+        status = NtDuplicateObject( NtCurrentProcess(), thread, process, &remote_suspendee_thread, 0, 0,
+                                    DUPLICATE_SAME_ACCESS );
+        if (status) goto err_close_proc;
+
+        status = NtCreateThreadEx( &suspender_thread, SYNCHRONIZE | THREAD_QUERY_INFORMATION, NULL, process,
+                                   suspend_remote_breakin, remote_suspendee_thread,
+                                   THREAD_CREATE_FLAGS_SKIP_THREAD_ATTACH | THREAD_CREATE_FLAGS_SKIP_LOADER_INIT |
+                                   THREAD_CREATE_FLAGS_HIDE_FROM_DEBUGGER | THREAD_CREATE_FLAGS_BYPASS_PROCESS_FREEZE,
+                                   0, 0, 0, NULL );
+        if (status) goto err_close_remote_hnd;
+
+        NtWaitForSingleObject( suspender_thread, FALSE, NULL );
+        status = NtQueryInformationThread( suspender_thread, ThreadBasicInformation, &tbi, sizeof(tbi), NULL );
+        if (!status)
+        {
+            if (tbi.ExitStatus < 0)
+            {
+                status = tbi.ExitStatus;
+            }
+            else if (count)
+            {
+                *count = (ULONG)tbi.ExitStatus;
+            }
+        }

-    NTSTATUS ret = NtQueryInformationThread( thread, ThreadBasicInformation, &tbi, sizeof(tbi), NULL);
-    if (ret) return ret;
-
-    if (tbi.ClientId.UniqueProcess != NtCurrentTeb()->ClientId.UniqueProcess) {
-        FIXME( "Non-local process thread suspend\n" );
-        return STATUS_SUCCESS;
+        NtClose( suspender_thread );
+err_close_remote_hnd:
+        NtDuplicateObject( process, remote_suspendee_thread, NULL, NULL, 0, 0, DUPLICATE_CLOSE_SOURCE );
+err_close_proc:
+        NtClose( process );
+        return status;
     }

     return pWow64SuspendLocalThread( thread, count );

From a6844d10622fc1a973ec1f22fc4f78a0fcd6cb29 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Fri, 12 Apr 2024 20:54:16 +0000
Subject: [PATCH 20/20] server: Ensure allocated space for mappings is in
 accessible memory.

---
 server/mapping.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/server/mapping.c b/server/mapping.c
index c3f57b6394c1..2e878251a1c3 100644
--- a/server/mapping.c
+++ b/server/mapping.c
@@ -21,6 +21,7 @@
 #include "config.h"

 #include <assert.h>
+#include <errno.h>
 #include <fcntl.h>
 #include <stdarg.h>
 #include <stdio.h>
@@ -265,11 +266,33 @@ static inline mem_size_t round_size( mem_size_t size, mem_size_t mask )
     return (size + mask) & ~mask;
 }

+static void free_available_high_map_addr( client_ptr_t base, mem_size_t size )
+{
+    unsigned int flags = MAP_PRIVATE | MAP_ANON;
+
+#ifdef MAP_FIXED_NOREPLACE
+    flags |= MAP_FIXED_NOREPLACE;
+#endif
+
+    while (base >> 32)
+    {
+        void *ret = mmap( (void *)base, host_page_mask + 1, PROT_NONE, flags, -1, 0 );
+        if (ret != MAP_FAILED) munmap( ret, host_page_mask + 1 );
+        if ((ret != MAP_FAILED && ret >= (void *)base) || errno == EEXIST)
+        {
+            free_map_addr( base, size );
+            return;
+        }
+        base >>= 1;
+        size >>= 1;
+    }
+}
+
 void init_memory(void)
 {
     host_page_mask = sysconf( _SC_PAGESIZE ) - 1;
     free_map_addr( 0x60000000, 0x1c000000 );
-    free_map_addr( 0x600000000000, 0x100000000000 );
+    free_available_high_map_addr( 0x600000000000, 0x100000000000 );
 }

 static void ranges_dump( struct object *obj, int verbose )