[PATCH 1/5] ntdll/tests: Add test for xstate in extended context.

[Paul Gofman]

Signed-off-by: Paul Gofman 
---
 dlls/ntdll/tests/exception.c | 160 +++++++++++++++++++++++++++++++++++
 1 file changed, 160 insertions(+)

diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c
index 1effffd421f..460f0e9a03e 100644
--- a/dlls/ntdll/tests/exception.c
+++ b/dlls/ntdll/tests/exception.c
@@ -30,6 +30,7 @@
 #include "winternl.h"
 #include "excpt.h"
 #include "wine/test.h"
+#include "intrin.h"

static void *code_mem;

@@ -43,6 +44,7 @@ static ULONG     (WINAPI *pRtlRemoveVectoredExceptionHandler)(PVOID handler);
 static PVOID     (WINAPI *pRtlAddVectoredContinueHandler)(ULONG first, PVECTORED_EXCEPTION_HANDLER func);
 static ULONG     (WINAPI *pRtlRemoveVectoredContinueHandler)(PVOID handler);
 static void      (WINAPI *pRtlSetUnhandledExceptionFilter)(PRTL_EXCEPTION_FILTER filter);
+static ULONG64   (WINAPI *pRtlGetEnabledExtendedFeatures)(ULONG64);
 static NTSTATUS  (WINAPI *pNtReadVirtualMemory)(HANDLE, const void*, void*, SIZE_T, SIZE_T*);
 static NTSTATUS  (WINAPI *pNtTerminateProcess)(HANDLE handle, LONG exit_code);
 static NTSTATUS  (WINAPI *pNtQueryInformationProcess)(HANDLE, PROCESSINFOCLASS, PVOID, ULONG, PULONG);
@@ -5458,6 +5460,161 @@ static void test_unload_trace(void)
     }
 }

+#if defined(__i386__) || defined(__x86_64__)
+
+static const unsigned int test_extended_context_data[8] = {1, 2, 3, 4, 5, 6, 7, 8};
+static const unsigned test_extended_context_spoil_data1[8] = {0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70, 0x80};
+static const unsigned test_extended_context_spoil_data2[8] = {0x15, 0x25, 0x35, 0x45, 0x55, 0x65, 0x75, 0x85};
+
+static BOOL test_extended_context_modified_state;
+
+static DWORD test_extended_context_handler(EXCEPTION_RECORD *rec, EXCEPTION_REGISTRATION_RECORD *frame,
+        CONTEXT *context, EXCEPTION_REGISTRATION_RECORD **dispatcher)
+{
+    static const ULONG64 expected_compaction_mask = 0x8000000000000004;
+    CONTEXT_EX *xctx = (CONTEXT_EX *)(context + 1);
+    unsigned int *context_ymm_data;
+    DWORD expected_min_offset;
+    BOOL compaction;
+    int regs[4];
+    XSTATE *xs;
+
+    /* Since we got xstates enabled by OS this cpuid level should be supported. */
+    __cpuidex(regs, 0xd, 1);
+    compaction = regs[0] & 2;
+    todo_wine ok((context->ContextFlags & (CONTEXT_FULL | CONTEXT_XSTATE)) == (CONTEXT_FULL | CONTEXT_XSTATE),
+            "Got unexpected ContextFlags %#x.\n", context->ContextFlags);
+
+    if ((context->ContextFlags & (CONTEXT_FULL | CONTEXT_XSTATE)) != (CONTEXT_FULL | CONTEXT_XSTATE))
+        goto done;
+
+#ifdef __x86_64__
+    {
+        /* Unwind contexts do not inherit xstate information. */
+        DISPATCHER_CONTEXT *dispatch = (DISPATCHER_CONTEXT *)dispatcher;
+
+        ok(!(dispatch->ContextRecord->ContextFlags & 0x40), "Got unexpected ContextRecord->ContextFlags %#x.\n",
+                dispatch->ContextRecord->ContextFlags);
+    }
+#endif
+
+    ok(xctx->Legacy.Offset == -(int)(sizeof(CONTEXT)), "Got unexpected Legacy.Offset %d.\n", xctx->Legacy.Offset);
+    ok(xctx->Legacy.Length == sizeof(CONTEXT), "Got unexpected Legacy.Length %d.\n", xctx->Legacy.Length);
+    ok(xctx->All.Offset == -(int)sizeof(CONTEXT), "Got unexpected All.Offset %d.\n", xctx->All.Offset);
+    ok(xctx->All.Length == sizeof(CONTEXT) + xctx->XState.Offset + xctx->XState.Length,
+            "Got unexpected All.Offset %d.\n", xctx->All.Offset);
+    expected_min_offset = sizeof(void *) == 8 ? sizeof(CONTEXT_EX) + sizeof(EXCEPTION_RECORD) : sizeof(CONTEXT_EX);
+    ok(xctx->XState.Offset >= expected_min_offset,
+            "Got unexpected XState.Offset %d.\n", xctx->XState.Offset);
+    ok(xctx->XState.Length >= sizeof(XSTATE), "Got unexpected XState.Length %d.\n", xctx->XState.Length);
+
+    xs = (XSTATE *)((char *)xctx + xctx->XState.Offset);
+    context_ymm_data = (unsigned int *)&xs->YmmContext;
+    ok(!((ULONG_PTR)xs % 64), "Got unexpected xs %p.\n", xs);
+
+    ok((compaction && (xs->CompactionMask & (expected_compaction_mask | 3)) == expected_compaction_mask)
+            || (!compaction && !xs->CompactionMask), "Got unexpected CompactionMask %s, compaction %#x.\n",
+            wine_dbgstr_longlong(xs->CompactionMask), compaction);
+
+    if (test_extended_context_modified_state)
+    {
+        ok((xs->Mask & 7) == 4, "Got unexpected Mask %s.\n", wine_dbgstr_longlong(xs->Mask));
+        ok(!memcmp(context_ymm_data, test_extended_context_data + 4, sizeof(M128A)),
+                "Got unexpected context data.\n");
+    }
+    else
+    {
+        ok(!xs->Mask, "Got unexpected Mask %s.\n", wine_dbgstr_longlong(xs->Mask));
+        /* The save area has garbage in this case, the state should be restored to INIT_STATE
+         * without using these data. */
+        memcpy(context_ymm_data, test_extended_context_spoil_data1 + 4, sizeof(M128A));
+    }
+
+done:
+#ifdef __GNUC__
+    __asm__ volatile("vmovups %0,%%ymm0" : : "m"(test_extended_context_spoil_data2));
+#endif
+#ifdef __x86_64__
+    ++context->Rip;
+#else
+    if (*(BYTE *)context->Eip == 0xcc)
+        ++context->Eip;
+#endif
+    return ExceptionContinueExecution;
+}
+
+static void test_extended_context(void)
+{
+    static BYTE except_code_set_ymm0[] =
+    {
+#ifdef __x86_64__
+        0x48,
+#endif
+        0xb8,                         /* mov imm,%ax */
+        0x00, 0x00, 0x00, 0x00,
+#ifdef __x86_64__
+        0x00, 0x00, 0x00, 0x00,
+#endif
+
+        0xc5, 0xfc, 0x10, 0x00,       /* vmovups (%ax),%ymm0 */
+        0xcc,                         /* int3 */
+        0xc5, 0xfc, 0x11, 0x00,       /* vmovups %ymm0,(%ax) */
+        0xc3,                         /* ret  */
+    };
+    static BYTE except_code_reset_ymm_state[] =
+    {
+#ifdef __x86_64__
+        0x48,
+#endif
+        0xb8,                         /* mov imm,%ax */
+        0x00, 0x00, 0x00, 0x00,
+#ifdef __x86_64__
+        0x00, 0x00, 0x00, 0x00,
+#endif
+        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+
+        0xc5, 0xf8, 0x77,             /* vzeroupper */
+        0x0f, 0x57, 0xc0,             /* xorps  %xmm0,%xmm0 */
+        0xcc,                         /* int3 */
+        0xc5, 0xfc, 0x11, 0x00,       /* vmovups %ymm0,(%ax) */
+        0xc3,                         /* ret  */
+    };
+    unsigned int i, address_offset;
+    unsigned data[8];
+
+    address_offset = sizeof(void *) == 8 ? 2 : 1;
+    *(void **)(except_code_set_ymm0 + address_offset) = data;
+    *(void **)(except_code_reset_ymm_state + address_offset) = data;
+
+    if (!pRtlGetEnabledExtendedFeatures || !(pRtlGetEnabledExtendedFeatures(~(ULONG64)0) & (1 << XSTATE_AVX)))
+    {
+        skip("AVX is not supported.\n");
+        return;
+    }
+
+    memset(data, 0xff, sizeof(data));
+    test_extended_context_modified_state = FALSE;
+    run_exception_test(test_extended_context_handler, NULL, except_code_reset_ymm_state,
+            ARRAY_SIZE(except_code_reset_ymm_state), PAGE_EXECUTE_READ);
+    for (i = 0; i < 8; ++i)
+    {
+        /* Older Windows version do not reset AVX context to INIT_STATE on x86. */
+        todo_wine_if(i >= 4)
+        ok(!data[i] || broken(i >= 4 && sizeof(void *) == 4 && data[i] == test_extended_context_spoil_data2[i]),
+                "Got unexpected data %#x, i %u.\n", data[i], i);
+    }
+
+    memcpy(data, test_extended_context_data, sizeof(data));
+    test_extended_context_modified_state = TRUE;
+    run_exception_test(test_extended_context_handler, NULL, except_code_set_ymm0,
+            ARRAY_SIZE(except_code_set_ymm0), PAGE_EXECUTE_READ);
+
+    for (i = 0; i < 8; ++i)
+        todo_wine_if(i >= 4)
+        ok(data[i] == test_extended_context_data[i], "Got unexpected data %#x, i %u.\n", data[i], i);
+}
+#endif
+
 START_TEST(exception)
 {
     HMODULE hntdll = GetModuleHandleA("ntdll.dll");
@@ -5500,6 +5657,7 @@ START_TEST(exception)
     X(NtResumeProcess);
     X(RtlGetUnloadEventTrace);
     X(RtlGetUnloadEventTraceEx);
+    X(RtlGetEnabledExtendedFeatures);
 #undef X

pIsWow64Process = (void *)GetProcAddress(GetModuleHandleA("kernel32.dll"), "IsWow64Process");
@@ -5582,6 +5740,7 @@ START_TEST(exception)
     test_dpe_exceptions();
     test_prot_fault();
     test_kiuserexceptiondispatcher();
+    test_extended_context();

#elif defined(__x86_64__)

@@ -5620,6 +5779,7 @@ START_TEST(exception)
       test_dynamic_unwind();
     else
       skip( "Dynamic unwind functions not found\n" );
+    test_extended_context();
 #endif

test_debugger();

[PATCH 2/5] ntdll: Support AVX context in fault exceptions on Linux x64.

[Paul Gofman]

Signed-off-by: Paul Gofman 
---
 dlls/ntdll/signal_x86_64.c      |   2 +
 dlls/ntdll/tests/exception.c    |   3 +-
 dlls/ntdll/unix/signal_x86_64.c | 114 ++++++++++++++++++++++++++------
 dlls/ntdll/unix/unix_private.h  |   6 ++
 dlls/ntdll/unixlib.h            |  30 +++++++++
 5 files changed, 134 insertions(+), 21 deletions(-)

diff --git a/dlls/ntdll/signal_x86_64.c b/dlls/ntdll/signal_x86_64.c
index e1cf7f3e013..564b74b7be6 100644
--- a/dlls/ntdll/signal_x86_64.c
+++ b/dlls/ntdll/signal_x86_64.c
@@ -428,6 +428,8 @@ static NTSTATUS call_stack_handlers( EXCEPTION_RECORD *rec, CONTEXT *orig_contex
     NTSTATUS status;

context = *orig_context;
+    context.ContextFlags &= ~0x40; /* Clear xstate flag. */
+
     dispatch.TargetIp      = 0;
     dispatch.ContextRecord = &context;
     dispatch.HistoryTable  = &table;
diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c
index 460f0e9a03e..cf704093fce 100644
--- a/dlls/ntdll/tests/exception.c
+++ b/dlls/ntdll/tests/exception.c
@@ -5482,7 +5482,8 @@ static DWORD test_extended_context_handler(EXCEPTION_RECORD *rec, EXCEPTION_REGI
     /* Since we got xstates enabled by OS this cpuid level should be supported. */
     __cpuidex(regs, 0xd, 1);
     compaction = regs[0] & 2;
-    todo_wine ok((context->ContextFlags & (CONTEXT_FULL | CONTEXT_XSTATE)) == (CONTEXT_FULL | CONTEXT_XSTATE),
+    todo_wine_if(sizeof(void *) == 4)
+    ok((context->ContextFlags & (CONTEXT_FULL | CONTEXT_XSTATE)) == (CONTEXT_FULL | CONTEXT_XSTATE),
             "Got unexpected ContextFlags %#x.\n", context->ContextFlags);

if ((context->ContextFlags & (CONTEXT_FULL | CONTEXT_XSTATE)) != (CONTEXT_FULL | CONTEXT_XSTATE))
diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c
index 638ad5cab76..370f1f36fd7 100644
--- a/dlls/ntdll/unix/signal_x86_64.c
+++ b/dlls/ntdll/unix/signal_x86_64.c
@@ -78,6 +78,7 @@
 #define WIN32_NO_STATUS
 #include "windef.h"
 #include "winternl.h"
+#include "ddk/wdm.h"
 #include "wine/exception.h"
 #include "wine/list.h"
 #include "wine/asm.h"
@@ -94,6 +95,10 @@ WINE_DEFAULT_DEBUG_CHANNEL(seh);
 #include <asm/prctl.h>
 static inline int arch_prctl( int func, void *ptr ) { return syscall( __NR_arch_prctl, func, ptr ); }

+#ifndef FP_XSTATE_MAGIC1
+#define FP_XSTATE_MAGIC1 0x46505853
+#endif
+
 #define RAX_sig(context)     ((context)->uc_mcontext.gregs[REG_RAX])
 #define RBX_sig(context)     ((context)->uc_mcontext.gregs[REG_RBX])
 #define RCX_sig(context)     ((context)->uc_mcontext.gregs[REG_RCX])
@@ -118,6 +123,7 @@ static inline int arch_prctl( int func, void *ptr ) { return syscall( __NR_arch_
 #define TRAP_sig(context)    ((context)->uc_mcontext.gregs[REG_TRAPNO])
 #define ERROR_sig(context)   ((context)->uc_mcontext.gregs[REG_ERR])
 #define FPU_sig(context)     ((XMM_SAVE_AREA32 *)((context)->uc_mcontext.fpregs))
+#define XState_sig(fpu)      (((unsigned int *)fpu->Reserved4)[12] == FP_XSTATE_MAGIC1 ? (XSTATE *)(fpu + 1) : NULL)

#elif defined(__FreeBSD__) || defined (__FreeBSD_kernel__)

@@ -148,6 +154,7 @@ static inline int arch_prctl( int func, void *ptr ) { return syscall( __NR_arch_
 #define TRAP_sig(context)    ((context)->uc_mcontext.mc_trapno)
 #define ERROR_sig(context)   ((context)->uc_mcontext.mc_err)
 #define FPU_sig(context)     ((XMM_SAVE_AREA32 *)((context)->uc_mcontext.mc_fpstate))
+#define XState_sig(context)  NULL

#elif defined(__NetBSD__)

@@ -178,6 +185,7 @@ static inline int arch_prctl( int func, void *ptr ) { return syscall( __NR_arch_
 #define TRAP_sig(context)   ((context)->uc_mcontext.__gregs[_REG_TRAPNO])
 #define ERROR_sig(context)  ((context)->uc_mcontext.__gregs[_REG_ERR])
 #define FPU_sig(context)    ((XMM_SAVE_AREA32 *)((context)->uc_mcontext.__fpregs))
+#define XState_sig(context) NULL

#elif defined (__APPLE__)

@@ -205,6 +213,7 @@ static inline int arch_prctl( int func, void *ptr ) { return syscall( __NR_arch_
 #define TRAP_sig(context)    ((context)->uc_mcontext->__es.__trapno)
 #define ERROR_sig(context)   ((context)->uc_mcontext->__es.__err)
 #define FPU_sig(context)     ((XMM_SAVE_AREA32 *)&(context)->uc_mcontext->__fs.__fpu_fcw)
+#define XState_sig(context)  NULL

#else
 #error You must define the signal context functions for your platform
@@ -237,16 +246,21 @@ enum i386_trap_code
 struct stack_layout
 {
     CONTEXT           context;
-    ULONG64           unknown[4];
+    CONTEXT_EX        context_ex;
     EXCEPTION_RECORD  rec;
     ULONG64           rsi;
     ULONG64           rdi;
     ULONG64           rbp;
     ULONG64           rip;
-    ULONG64           red_zone[16];
+    ULONG64           align;
+    char              xstate[0]; /* If xstate is present it is allocated
+                                  * dynamically to provide 64 byte alignment. */
 };

-C_ASSERT( sizeof(struct stack_layout) == 0x630 ); /* Should match the size in call_user_exception_dispatcher(). */
+C_ASSERT((offsetof(struct stack_layout, xstate) == sizeof(struct stack_layout)));
+
+C_ASSERT( sizeof(XSTATE) == 0x140 );
+C_ASSERT( sizeof(struct stack_layout) == 0x5b0 ); /* Should match the size in call_user_exception_dispatcher(). */

struct syscall_frame
 {
@@ -1408,8 +1422,10 @@ static inline void set_sigcontext( const CONTEXT *context, ucontext_t *sigcontex
  *
  * Set the register values from a sigcontext.
  */
-static void save_context( CONTEXT *context, const ucontext_t *sigcontext )
+static void save_context( struct xcontext *xcontext, const ucontext_t *sigcontext )
 {
+    CONTEXT *context = &xcontext->c;
+
     context->ContextFlags = CONTEXT_CONTROL | CONTEXT_INTEGER | CONTEXT_SEGMENTS | CONTEXT_DEBUG_REGISTERS;
     context->Rax    = RAX_sig(sigcontext);
     context->Rcx    = RCX_sig(sigcontext);
@@ -1458,6 +1474,11 @@ static void save_context( CONTEXT *context, const ucontext_t *sigcontext )
         context->ContextFlags |= CONTEXT_FLOATING_POINT;
         context->u.FltSave = *FPU_sig(sigcontext);
         context->MxCsr = context->u.FltSave.MxCsr;
+        xcontext->xstate = XState_sig(FPU_sig(sigcontext));
+    }
+    else
+    {
+        xcontext->xstate = NULL;
     }
 }

@@ -1467,8 +1488,10 @@ static void save_context( CONTEXT *context, const ucontext_t *sigcontext )
  *
  * Build a sigcontext from the register values.
  */
-static void restore_context( const CONTEXT *context, ucontext_t *sigcontext )
+static void restore_context( const struct xcontext *xcontext, ucontext_t *sigcontext )
 {
+    const CONTEXT *context = &xcontext->c;
+
     amd64_thread_data()->dr0 = context->Dr0;
     amd64_thread_data()->dr1 = context->Dr1;
     amd64_thread_data()->dr2 = context->Dr2;
@@ -1813,7 +1836,6 @@ NTSTATUS WINAPI NtGetContextThread( HANDLE handle, CONTEXT *context )
     return STATUS_SUCCESS;
 }

-
 extern void CDECL raise_func_trampoline( void *dispatcher );

__ASM_GLOBAL_FUNC( raise_func_trampoline,
@@ -1822,10 +1844,12 @@ __ASM_GLOBAL_FUNC( raise_func_trampoline,
 /***********************************************************************
  *           setup_raise_exception
  */
-static void setup_raise_exception( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, CONTEXT *context )
+static void setup_raise_exception( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, struct xcontext *xcontext )
 {
     void *stack_ptr = (void *)(RSP_sig(sigcontext) & ~15);
+    CONTEXT *context = &xcontext->c;
     struct stack_layout *stack;
+    size_t stack_size;
     NTSTATUS status;

if (rec->ExceptionCode == EXCEPTION_SINGLE_STEP)
@@ -1848,16 +1872,37 @@ static void setup_raise_exception( ucontext_t *sigcontext, EXCEPTION_RECORD *rec
     status = send_debug_event( rec, context, TRUE );
     if (status == DBG_CONTINUE || status == DBG_EXCEPTION_HANDLED)
     {
-        restore_context( context, sigcontext );
+        restore_context( xcontext, sigcontext );
         return;
     }

/* fix up instruction pointer in context for EXCEPTION_BREAKPOINT */
     if (rec->ExceptionCode == EXCEPTION_BREAKPOINT) context->Rip--;

-    stack = virtual_setup_exception( stack_ptr, sizeof(*stack), rec );
+    stack_size = sizeof(*stack);
+    if (xcontext->xstate)
+    {
+        stack_size += (ULONG_PTR)stack_ptr - (((ULONG_PTR)stack_ptr
+                - sizeof(XSTATE)) & ~(ULONG_PTR)63);
+    }
+
+    stack = virtual_setup_exception( stack_ptr, stack_size, rec );
     stack->rec          = *rec;
     stack->context      = *context;
+    if (xcontext->xstate)
+    {
+        XSTATE *dst_xs = (XSTATE *)stack->xstate;
+
+        assert( !((ULONG_PTR)dst_xs & 63) );
+        context_init_xstate( &stack->context, stack->xstate );
+        dst_xs->CompactionMask = user_shared_data->XState.CompactionEnabled ? 0x8000000000000004 : 0;
+        if (xcontext->xstate->Mask & 4)
+        {
+            dst_xs->Mask = 4;
+            memcpy( &dst_xs->YmmContext, &xcontext->xstate->YmmContext, sizeof(dst_xs->YmmContext) );
+        }
+    }
+
     RIP_sig(sigcontext) = (ULONG_PTR)raise_func_trampoline;
     R8_sig(sigcontext)  = (ULONG_PTR)pKiUserExceptionDispatcher;
     RSP_sig(sigcontext) = (ULONG_PTR)stack;
@@ -1875,7 +1920,7 @@ static void setup_raise_exception( ucontext_t *sigcontext, EXCEPTION_RECORD *rec
  */
 static void setup_exception( ucontext_t *sigcontext, EXCEPTION_RECORD *rec )
 {
-    CONTEXT context;
+    struct xcontext context;

rec->ExceptionAddress = (void *)RIP_sig(sigcontext);
     save_context( &context, sigcontext );
@@ -1972,7 +2017,31 @@ void WINAPI do_call_user_exception_dispatcher( EXCEPTION_RECORD *rec, CONTEXT *c
 {
     struct syscall_frame *frame = amd64_thread_data()->syscall_frame;

-    memmove(&stack->context, context, sizeof(*context));
+    if ((context->ContextFlags & CONTEXT_XSTATE) == CONTEXT_XSTATE)
+    {
+        CONTEXT_EX *xctx = (CONTEXT_EX *)context + 1;
+        XSTATE *xs, *src_xs, xs_buf;
+
+        src_xs = xstate_from_context(context);
+        if ((CONTEXT *)src_xs >= &stack->context + 1 || src_xs + 1 <= (XSTATE *)&stack->context)
+        {
+            xs = src_xs;
+        }
+        else
+        {
+            xs = &xs_buf;
+            memcpy(xs, src_xs, sizeof(*xs));
+        }
+
+        memmove(&stack->context, context, sizeof(*context) + sizeof(*xctx));
+        assert(!((ULONG_PTR)stack->xstate & 63));
+        context_init_xstate(&stack->context, stack->xstate);
+        memcpy(stack->xstate, xs, sizeof(*xs));
+    }
+    else
+    {
+        memmove(&stack->context, context, sizeof(*context));
+    }
     memcpy(&stack->rec, rec, sizeof(*rec));

/* fix up instruction pointer in context for EXCEPTION_BREAKPOINT */
@@ -1985,7 +2054,11 @@ void WINAPI do_call_user_exception_dispatcher( EXCEPTION_RECORD *rec, CONTEXT *c
 __ASM_GLOBAL_FUNC( call_user_exception_dispatcher,
                    "movq 0x98(%rdx),%r9\n\t" /* context->Rsp */
                    "andq $~0xf,%r9\n\t"
-                   "subq $0x630,%r9\n\t" /* sizeof(struct stack_layout) */
+                   "bt $6,0x30(%rdx)\n\t" /* context->ContextFlags, CONTEXT_XSTATE bit. */
+                   "jnc 1f\n\t"
+                   "subq $0x140,%r9\n\t" /* sizeof(XSTATE) */
+                   "andq $~63,%r9\n\t"
+                   "1:subq $0x5b0,%r9\n\t" /* sizeof(struct stack_layout) */
                    "cmpq %rsp,%r9\n\t"
                    "cmovbq %r9,%rsp\n\t"
                    "jmp " __ASM_NAME("do_call_user_exception_dispatcher") "\n\t")
@@ -2187,8 +2260,10 @@ static void install_bpf(struct sigaction *sig_act)
  *
  * Handle an interrupt.
  */
-static inline BOOL handle_interrupt( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, CONTEXT *context )
+static inline BOOL handle_interrupt( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, struct xcontext *xcontext )
 {
+    CONTEXT *context = &xcontext->c;
+
     switch (ERROR_sig(sigcontext) >> 3)
     {
     case 0x2c:
@@ -2213,11 +2288,10 @@ static inline BOOL handle_interrupt( ucontext_t *sigcontext, EXCEPTION_RECORD *r
     default:
         return FALSE;
     }
-    setup_raise_exception( sigcontext, rec, context );
+    setup_raise_exception( sigcontext, rec, xcontext );
     return TRUE;
 }

-
 /**********************************************************************
  *		segv_handler
  *
@@ -2226,7 +2300,7 @@ static inline BOOL handle_interrupt( ucontext_t *sigcontext, EXCEPTION_RECORD *r
 static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 {
     EXCEPTION_RECORD rec = { 0 };
-    CONTEXT context;
+    struct xcontext context;
     ucontext_t *ucontext = sigcontext;

rec.ExceptionAddress = (void *)RIP_sig(ucontext);
@@ -2250,7 +2324,7 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
     case TRAP_x86_PROTFLT:   /* General protection fault */
         {
             WORD err = ERROR_sig(ucontext);
-            if (!err && (rec.ExceptionCode = is_privileged_instr( &context ))) break;
+            if (!err && (rec.ExceptionCode = is_privileged_instr( &context.c ))) break;
             if ((err & 7) == 2 && handle_interrupt( ucontext, &rec, &context )) return;
             rec.ExceptionCode = EXCEPTION_ACCESS_VIOLATION;
             rec.NumberParameters = 2;
@@ -2293,7 +2367,7 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 static void trap_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 {
     EXCEPTION_RECORD rec = { 0 };
-    CONTEXT context;
+    struct xcontext context;
     ucontext_t *ucontext = sigcontext;

rec.ExceptionAddress = (void *)RIP_sig(ucontext);
@@ -2412,10 +2486,10 @@ static void quit_handler( int signal, siginfo_t *siginfo, void *ucontext )
  */
 static void usr1_handler( int signal, siginfo_t *siginfo, void *ucontext )
 {
-    CONTEXT context;
+    struct xcontext context;

save_context( &context, ucontext );
-    wait_suspend( &context );
+    wait_suspend( &context.c );
     restore_context( &context, ucontext );
 }

diff --git a/dlls/ntdll/unix/unix_private.h b/dlls/ntdll/unix/unix_private.h
index c9f8871fd26..68c60f78b7f 100644
--- a/dlls/ntdll/unix/unix_private.h
+++ b/dlls/ntdll/unix/unix_private.h
@@ -400,4 +400,10 @@ static inline int ntdll_wcsnicmp( const WCHAR *str1, const WCHAR *str2, int n )
 #define towupper(c)        ntdll_towupper(c)
 #define towlower(c)        ntdll_towlower(c)

+struct xcontext
+{
+    CONTEXT c;
+    XSTATE *xstate; /* Points to xstate in sigcontext. */
+};
+
 #endif /* __NTDLL_UNIX_PRIVATE_H */
diff --git a/dlls/ntdll/unixlib.h b/dlls/ntdll/unixlib.h
index 2f57873671c..70b6def530a 100644
--- a/dlls/ntdll/unixlib.h
+++ b/dlls/ntdll/unixlib.h
@@ -129,4 +129,34 @@ struct unix_funcs
                                        const char *function );
 };

+static inline XSTATE *xstate_from_context( const CONTEXT *context )
+{
+    CONTEXT_EX *xctx = (CONTEXT_EX *)(context + 1);
+
+    if ((context->ContextFlags & CONTEXT_XSTATE) != CONTEXT_XSTATE)
+        return NULL;
+
+    return (XSTATE *)((char *)(context + 1) + xctx->XState.Offset);
+}
+
+static inline void context_init_xstate( CONTEXT *context, void *xstate_buffer )
+{
+    CONTEXT_EX *xctx;
+    XSTATE *xs;
+
+    xctx = (CONTEXT_EX *)(context + 1);
+    xctx->Legacy.Length = sizeof(CONTEXT);
+    xctx->Legacy.Offset = -(LONG)sizeof(CONTEXT);
+
+    xctx->XState.Length = sizeof(XSTATE);
+    xctx->XState.Offset = xstate_buffer ? (((ULONG_PTR)xstate_buffer + 63) & ~63) - (ULONG_PTR)xctx
+            : (((ULONG_PTR)context + sizeof(CONTEXT) + sizeof(CONTEXT_EX) + 63) & ~63) - (ULONG_PTR)xctx;
+    xctx->All.Length = sizeof(CONTEXT) + xctx->XState.Offset + xctx->XState.Length;
+    xctx->All.Offset = -(LONG)sizeof(CONTEXT);
+    context->ContextFlags |= 0x40;
+
+    xs = xstate_from_context(context);
+    memset( xs, 0, offsetof(XSTATE, YmmContext) );
+}
+
 #endif /* __NTDLL_UNIXLIB_H */

[PATCH 2/5] ntdll: Support AVX context in fault exceptions on Linux x64.

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=77412

Your paranoid android.

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3968 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_2scr (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_ar (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_he (testbot log) ===

An error occurred while waiting for the test to complete: the 5384 process does not exist or is not a child process

=== w1064v1809_ja (testbot log) ===

An error occurred while waiting for the test to complete: the 1152 process does not exist or is not a child process

=== w1064v1809_zh_CN (testbot log) ===

An error occurred while waiting for the test to complete: the 1436 process does not exist or is not a child process

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3952 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: the 6156 process does not exist or is not a child process

=== debiant (32 bit WoW report) ===

ntdll:
change.c:241: Test failed: should be ready

[PATCH 3/5] ntdll: Restore AVX registers in NtSetContextThread() on x64.

[Paul Gofman]

Signed-off-by: Paul Gofman 
---
 dlls/ntdll/tests/exception.c    |  4 ++--
 dlls/ntdll/unix/signal_x86_64.c | 30 ++++++++++++++++++++++++++++++
 2 files changed, 32 insertions(+), 2 deletions(-)

diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c
index cf704093fce..1bbf9f4f912 100644
--- a/dlls/ntdll/tests/exception.c
+++ b/dlls/ntdll/tests/exception.c
@@ -5600,7 +5600,7 @@ static void test_extended_context(void)
     for (i = 0; i < 8; ++i)
     {
         /* Older Windows version do not reset AVX context to INIT_STATE on x86. */
-        todo_wine_if(i >= 4)
+        todo_wine_if(i >= 4 && sizeof(void *) == 4)
         ok(!data[i] || broken(i >= 4 && sizeof(void *) == 4 && data[i] == test_extended_context_spoil_data2[i]),
                 "Got unexpected data %#x, i %u.\n", data[i], i);
     }
@@ -5611,7 +5611,7 @@ static void test_extended_context(void)
             ARRAY_SIZE(except_code_set_ymm0), PAGE_EXECUTE_READ);

for (i = 0; i < 8; ++i)
-        todo_wine_if(i >= 4)
+        todo_wine_if(i >= 4 && sizeof(void *) == 4)
         ok(data[i] == test_extended_context_data[i], "Got unexpected data %#x, i %u.\n", data[i], i);
 }
 #endif
diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c
index 370f1f36fd7..53b9d98e80c 100644
--- a/dlls/ntdll/unix/signal_x86_64.c
+++ b/dlls/ntdll/unix/signal_x86_64.c
@@ -1543,6 +1543,34 @@ __ASM_GLOBAL_FUNC( set_full_cpu_context,
                    "iretq" );

+/***********************************************************************
+ *           restore_xstate
+ *
+ * Restore the XState context.
+ */
+static void restore_xstate( const CONTEXT *context )
+{
+    XSAVE_FORMAT *xrstor_base;
+    XSTATE *xs;
+
+    if (!(xs = xstate_from_context( context )))
+        return;
+
+    xrstor_base = (XSAVE_FORMAT *)xs - 1;
+
+    if (!(xs->CompactionMask & ((ULONG64)1 << 63)))
+    {
+        /* Non-compacted xrstor will load Mxcsr regardless of the specified mask. Loading garbage there
+         * may lead to fault. We have only padding, no more used EXCEPTION_RECORD or unused context fields
+         * at the MxCsr restore location, so just put it there. */
+        assert( (void *)&xrstor_base->MxCsr > (void *)context->VectorRegister );
+        xrstor_base->MxCsr = context->u.FltSave.MxCsr;
+        xrstor_base->MxCsr_Mask = context->u.FltSave.MxCsr_Mask;
+    }
+
+    __asm__ volatile("xor %%edx,%%edx; mov $4,%%eax; xrstor64 %0" : : "m"(*xrstor_base) : "eax", "edx");
+}
+
 /***********************************************************************
  *           get_server_context_flags
  *
@@ -1733,6 +1761,8 @@ NTSTATUS WINAPI NtSetContextThread( HANDLE handle, const CONTEXT *context )
         }
     }

+    restore_xstate( context );
+
     if (flags & CONTEXT_FULL)
     {
         if (!(flags & CONTEXT_CONTROL))

[PATCH 3/5] ntdll: Restore AVX registers in NtSetContextThread() on x64.

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=77413

Your paranoid android.

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3944 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: the 6172 process does not exist or is not a child process

=== w1064v1809_2scr (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_ar (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_he (testbot log) ===

An error occurred while waiting for the test to complete: the 2140 process does not exist or is not a child process

=== w1064v1809_ja (testbot log) ===

An error occurred while waiting for the test to complete: the 5428 process does not exist or is not a child process

=== w1064v1809_zh_CN (testbot log) ===

An error occurred while waiting for the test to complete: the 1408 process does not exist or is not a child process

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3952 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

[PATCH 4/5] ntdll: Support AVX context in fault exceptions on Linux x86.

[Paul Gofman]

Signed-off-by: Paul Gofman 
---
 dlls/ntdll/tests/exception.c  |   2 +-
 dlls/ntdll/unix/signal_i386.c | 146 ++++++++++++++++++++++++----------
 2 files changed, 106 insertions(+), 42 deletions(-)

diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c
index 1bbf9f4f912..860221366ce 100644
--- a/dlls/ntdll/tests/exception.c
+++ b/dlls/ntdll/tests/exception.c
@@ -5482,7 +5482,7 @@ static DWORD test_extended_context_handler(EXCEPTION_RECORD *rec, EXCEPTION_REGI
     /* Since we got xstates enabled by OS this cpuid level should be supported. */
     __cpuidex(regs, 0xd, 1);
     compaction = regs[0] & 2;
-    todo_wine_if(sizeof(void *) == 4)
+
     ok((context->ContextFlags & (CONTEXT_FULL | CONTEXT_XSTATE)) == (CONTEXT_FULL | CONTEXT_XSTATE),
             "Got unexpected ContextFlags %#x.\n", context->ContextFlags);

diff --git a/dlls/ntdll/unix/signal_i386.c b/dlls/ntdll/unix/signal_i386.c
index bc30dd34df8..791ffe90940 100644
--- a/dlls/ntdll/unix/signal_i386.c
+++ b/dlls/ntdll/unix/signal_i386.c
@@ -33,6 +33,7 @@
 #include <stdarg.h>
 #include <stdio.h>
 #include <sys/types.h>
+#include <assert.h>
 #ifdef HAVE_UNISTD_H
 # include <unistd.h>
 #endif
@@ -56,6 +57,8 @@
 #include "ntstatus.h"
 #define WIN32_NO_STATUS
 #include "windef.h"
+#include "winternl.h"
+#include "ddk/wdm.h"
 #include "wine/asm.h"
 #include "wine/exception.h"
 #include "unix_private.h"
@@ -122,6 +125,10 @@ typedef struct ucontext
 } ucontext_t;
 #endif /* HAVE_SYS_UCONTEXT_H */

+#ifndef FP_XSTATE_MAGIC1
+#define FP_XSTATE_MAGIC1 0x46505853
+#endif
+
 #define EAX_sig(context)     ((context)->uc_mcontext.gregs[REG_EAX])
 #define EBX_sig(context)     ((context)->uc_mcontext.gregs[REG_EBX])
 #define ECX_sig(context)     ((context)->uc_mcontext.gregs[REG_ECX])
@@ -145,6 +152,7 @@ typedef struct ucontext

#define FPU_sig(context)     ((FLOATING_SAVE_AREA*)((context)->uc_mcontext.fpregs))
 #define FPUX_sig(context)    (FPU_sig(context) && !((context)->uc_mcontext.fpregs->status >> 16) ? (XSAVE_FORMAT *)(FPU_sig(context) + 1) : NULL)
+#define XState_sig(fpu)      (((unsigned int *)fpu->Reserved4)[12] == FP_XSTATE_MAGIC1 ? (XSTATE *)(fpu + 1) : NULL)

#ifdef __ANDROID__
 /* custom signal restorer since we may have unmapped the one in vdso, and bionic doesn't check for that */
@@ -208,6 +216,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )

#define FPU_sig(context)     NULL  /* FIXME */
 #define FPUX_sig(context)    NULL  /* FIXME */
+#define XState_sig(context)  NULL  /* FIXME */

#elif defined (__OpenBSD__)

@@ -238,6 +247,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )

#define FPU_sig(context)     NULL  /* FIXME */
 #define FPUX_sig(context)    NULL  /* FIXME */
+#define XState_sig(context)  NULL  /* FIXME */

#define T_MCHK T_MACHK
 #define T_XMMFLT T_XFTRAP
@@ -283,6 +293,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )

#define FPU_sig(context)     NULL  /* FIXME */
 #define FPUX_sig(context)    NULL  /* FIXME */
+#define XState_sig(context)  NULL  /* FIXME */

#elif defined (__APPLE__)

@@ -310,6 +321,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )
 #define ERROR_sig(context)   ((context)->uc_mcontext->__es.__err)
 #define FPU_sig(context)     NULL
 #define FPUX_sig(context)    ((XSAVE_FORMAT *)&(context)->uc_mcontext->__fs.__fpu_fcw)
+#define XState_sig(context)  NULL  /* FIXME */
 #else
 #define EAX_sig(context)     ((context)->uc_mcontext->ss.eax)
 #define EBX_sig(context)     ((context)->uc_mcontext->ss.ebx)
@@ -331,6 +343,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )
 #define ERROR_sig(context)   ((context)->uc_mcontext->es.err)
 #define FPU_sig(context)     NULL
 #define FPUX_sig(context)    ((XSAVE_FORMAT *)&(context)->uc_mcontext->fs.fpu_fcw)
+#define XState_sig(context)  NULL  /* FIXME */
 #endif

#elif defined(__NetBSD__)
@@ -361,6 +374,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )

#define FPU_sig(context)     NULL
 #define FPUX_sig(context)    ((XSAVE_FORMAT *)&((context)->uc_mcontext.__fpregs))
+#define XState_sig(context)  NULL  /* FIXME */

#define T_MCHK T_MCA
 #define T_XMMFLT T_XMM
@@ -393,6 +407,7 @@ static inline int set_thread_area( struct modify_ldt_s *ptr )

#define FPU_sig(context)     ((FLOATING_SAVE_AREA *)&(context)->uc_mcontext.fpregs.fp_reg_set.fpchip_state)
 #define FPUX_sig(context)    NULL
+#define XState_sig(context)  NULL  /* FIXME */

#else
 #error You must define the signal context functions for your platform
@@ -752,10 +767,11 @@ static void fpux_to_fpu( FLOATING_SAVE_AREA *fpu, const XSAVE_FORMAT *fpux )
  *
  * Build a context structure from the signal info.
  */
-static inline void save_context( CONTEXT *context, const ucontext_t *sigcontext )
+static inline void save_context( struct xcontext *xcontext, const ucontext_t *sigcontext )
 {
     FLOATING_SAVE_AREA *fpu = FPU_sig(sigcontext);
     XSAVE_FORMAT *fpux = FPUX_sig(sigcontext);
+    CONTEXT *context = &xcontext->c;

memset(context, 0, sizeof(*context));
     context->ContextFlags = CONTEXT_FULL | CONTEXT_DEBUG_REGISTERS;
@@ -792,6 +808,11 @@ static inline void save_context( CONTEXT *context, const ucontext_t *sigcontext
         context->ContextFlags |= CONTEXT_FLOATING_POINT | CONTEXT_EXTENDED_REGISTERS;
         memcpy( context->ExtendedRegisters, fpux, sizeof(*fpux) );
         if (!fpu) fpux_to_fpu( &context->FloatSave, fpux );
+        xcontext->xstate = XState_sig(fpux);
+    }
+    else
+    {
+        xcontext->xstate = NULL;
     }
     if (!fpu && !fpux) save_fpu( context );
 }
@@ -802,10 +823,11 @@ static inline void save_context( CONTEXT *context, const ucontext_t *sigcontext
  *
  * Restore the signal info from the context.
  */
-static inline void restore_context( const CONTEXT *context, ucontext_t *sigcontext )
+static inline void restore_context( const struct xcontext *xcontext, ucontext_t *sigcontext )
 {
     FLOATING_SAVE_AREA *fpu = FPU_sig(sigcontext);
     XSAVE_FORMAT *fpux = FPUX_sig(sigcontext);
+    const CONTEXT *context = &xcontext->c;

x86_thread_data()->dr0 = context->Dr0;
     x86_thread_data()->dr1 = context->Dr1;
@@ -831,7 +853,18 @@ static inline void restore_context( const CONTEXT *context, ucontext_t *sigconte
     SS_sig(sigcontext)  = context->SegSs;

if (fpu) *fpu = context->FloatSave;
-    if (fpux) memcpy( fpux, context->ExtendedRegisters, sizeof(*fpux) );
+    if (fpux)
+    {
+        XSTATE *src_xs, *dst_xs;
+
+        memcpy( fpux, context->ExtendedRegisters, sizeof(*fpux) );
+
+        if ((dst_xs = XState_sig(fpux)) && (src_xs = xstate_from_context( context )))
+        {
+            memcpy( &dst_xs->YmmContext, &src_xs->YmmContext, sizeof(dst_xs->YmmContext) );
+            dst_xs->Mask |= src_xs->Mask;
+        }
+    }
     if (!fpu && !fpux) restore_fpu( context );
 }

@@ -1442,48 +1475,76 @@ static BOOL check_atl_thunk( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, CONT
  *
  * Setup the exception record and context on the thread stack.
  */
-static void *setup_exception_record( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, CONTEXT *context )
+static void *setup_exception_record( ucontext_t *sigcontext, EXCEPTION_RECORD *rec, struct xcontext *xcontext )
 {
     void *stack = init_handler( sigcontext );

rec->ExceptionAddress = (void *)EIP_sig( sigcontext );
-    save_context( context, sigcontext );
+    save_context( xcontext, sigcontext );
     return stack;
 }

-
 /***********************************************************************
  *           setup_raise_exception
  *
  * Change context to setup a call to a raise exception function.
  */
 static void setup_raise_exception( ucontext_t *sigcontext, void *stack_ptr,
-                                   EXCEPTION_RECORD *rec, CONTEXT *context )
+                                   EXCEPTION_RECORD *rec, struct xcontext *xcontext )
 {
-    struct
+    CONTEXT *context = &xcontext->c;
+    size_t stack_size;
+
+    struct stack_layout
     {
         EXCEPTION_RECORD *rec_ptr;       /* first arg for KiUserExceptionDispatcher */
         CONTEXT          *context_ptr;   /* second arg for KiUserExceptionDispatcher */
         CONTEXT           context;
+        CONTEXT_EX        context_ex;
         EXCEPTION_RECORD  rec;
         DWORD             ebp;
         DWORD             eip;
+        char              xstate[0];
     } *stack;

+C_ASSERT( (offsetof(struct stack_layout, xstate) == sizeof(struct stack_layout)) );
+
     NTSTATUS status = send_debug_event( rec, context, TRUE );

if (status == DBG_CONTINUE || status == DBG_EXCEPTION_HANDLED)
     {
-        restore_context( context, sigcontext );
+        restore_context( xcontext, sigcontext );
         return;
     }

/* fix up instruction pointer in context for EXCEPTION_BREAKPOINT */
     if (rec->ExceptionCode == EXCEPTION_BREAKPOINT) context->Eip--;

-    stack = virtual_setup_exception( stack_ptr, sizeof(*stack), rec );
+    stack_size = sizeof(*stack);
+    if (xcontext->xstate)
+    {
+        stack_size += (ULONG_PTR)stack_ptr - (((ULONG_PTR)stack_ptr
+                - sizeof(XSTATE)) & ~(ULONG_PTR)63);
+    }
+
+    stack = virtual_setup_exception( stack_ptr, stack_size, rec );
     stack->rec          = *rec;
     stack->context      = *context;
+
+    if (xcontext->xstate)
+    {
+        XSTATE *dst_xs = (XSTATE *)stack->xstate;
+
+        assert(!((ULONG_PTR)dst_xs & 63));
+        context_init_xstate( &stack->context, stack->xstate );
+        dst_xs->CompactionMask = user_shared_data->XState.CompactionEnabled ? 0x8000000000000004 : 0;
+        if (xcontext->xstate->Mask & 4)
+        {
+            dst_xs->Mask = 4;
+            memcpy( &dst_xs->YmmContext, &xcontext->xstate->YmmContext, sizeof(dst_xs->YmmContext) );
+        }
+    }
+
     stack->rec_ptr      = &stack->rec;
     stack->context_ptr  = &stack->context;
     ESP_sig(sigcontext) = (DWORD)stack;
@@ -1506,9 +1567,10 @@ static void setup_raise_exception( ucontext_t *sigcontext, void *stack_ptr,
  */
 static void setup_exception( ucontext_t *sigcontext, EXCEPTION_RECORD *rec )
 {
-    CONTEXT context;
-    void *stack = setup_exception_record( sigcontext, rec, &context );
-    setup_raise_exception( sigcontext, stack, rec, &context );
+    struct xcontext xcontext;
+    void *stack = setup_exception_record( sigcontext, rec, &xcontext );
+
+    setup_raise_exception( sigcontext, stack, rec, &xcontext );
 }

@@ -1623,8 +1685,10 @@ static inline DWORD get_fpu_code( const CONTEXT *context )
  * Handle an interrupt.
  */
 static BOOL handle_interrupt( unsigned int interrupt, ucontext_t *sigcontext, void *stack,
-                              EXCEPTION_RECORD *rec, CONTEXT *context )
+                              EXCEPTION_RECORD *rec, struct xcontext *xcontext )
 {
+    CONTEXT *context = &xcontext->c;
+
     switch(interrupt)
     {
     case 0x2d:
@@ -1649,7 +1713,7 @@ static BOOL handle_interrupt( unsigned int interrupt, ucontext_t *sigcontext, vo
         rec->ExceptionInformation[0] = context->Eax;
         rec->ExceptionInformation[1] = context->Ecx;
         rec->ExceptionInformation[2] = context->Edx;
-        setup_raise_exception( sigcontext, stack, rec, context );
+        setup_raise_exception( sigcontext, stack, rec, xcontext );
         return TRUE;
     default:
         return FALSE;
@@ -1665,9 +1729,9 @@ static BOOL handle_interrupt( unsigned int interrupt, ucontext_t *sigcontext, vo
 static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 {
     EXCEPTION_RECORD rec = { 0 };
-    CONTEXT context;
+    struct xcontext xcontext;
     ucontext_t *ucontext = sigcontext;
-    void *stack = setup_exception_record( sigcontext, &rec, &context );
+    void *stack = setup_exception_record( sigcontext, &rec, &xcontext );

switch (TRAP_sig(ucontext))
     {
@@ -1687,8 +1751,8 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
     case TRAP_x86_PROTFLT:   /* General protection fault */
         {
             WORD err = ERROR_sig(ucontext);
-            if (!err && (rec.ExceptionCode = is_privileged_instr( &context ))) break;
-            if ((err & 7) == 2 && handle_interrupt( err >> 3, ucontext, stack, &rec, &context )) return;
+            if (!err && (rec.ExceptionCode = is_privileged_instr( &xcontext.c ))) break;
+            if ((err & 7) == 2 && handle_interrupt( err >> 3, ucontext, stack, &rec, &xcontext )) return;
             rec.ExceptionCode = EXCEPTION_ACCESS_VIOLATION;
             rec.NumberParameters = 2;
             rec.ExceptionInformation[0] = 0;
@@ -1697,7 +1761,7 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
             else
             {
                 rec.ExceptionInformation[1] = 0xffffffff;
-                if (check_invalid_gs( ucontext, &context )) return;
+                if (check_invalid_gs( ucontext, &xcontext.c )) return;
             }
         }
         break;
@@ -1714,7 +1778,7 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
             NtQueryInformationProcess( GetCurrentProcess(), ProcessExecuteFlags,
                                        &flags, sizeof(flags), NULL );
             if (!(flags & MEM_EXECUTE_OPTION_DISABLE_THUNK_EMULATION) &&
-                check_atl_thunk( ucontext, &rec, &context ))
+                check_atl_thunk( ucontext, &rec, &xcontext.c ))
                 return;

/* send EXCEPTION_EXECUTE_FAULT only if data execution prevention is enabled */
@@ -1723,7 +1787,7 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
         break;
     case TRAP_x86_ALIGNFLT:  /* Alignment check exception */
         /* FIXME: pass through exception handler first? */
-        if (context.EFlags & 0x00040000)
+        if (xcontext.c.EFlags & 0x00040000)
         {
             EFL_sig(ucontext) &= ~0x00040000;  /* disable AC flag */
             return;
@@ -1742,7 +1806,7 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
         rec.ExceptionCode = EXCEPTION_ILLEGAL_INSTRUCTION;
         break;
     }
-    setup_raise_exception( ucontext, stack, &rec, &context );
+    setup_raise_exception( ucontext, stack, &rec, &xcontext );
 }

@@ -1754,9 +1818,9 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 static void trap_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 {
     EXCEPTION_RECORD rec = { 0 };
-    CONTEXT context;
+    struct xcontext xcontext;
     ucontext_t *ucontext = sigcontext;
-    void *stack = setup_exception_record( sigcontext, &rec, &context );
+    void *stack = setup_exception_record( sigcontext, &rec, &xcontext );

switch (TRAP_sig(ucontext))
     {
@@ -1765,15 +1829,15 @@ static void trap_handler( int signal, siginfo_t *siginfo, void *sigcontext )
         /* when single stepping can't tell whether this is a hw bp or a
          * single step interrupt. try to avoid as much overhead as possible
          * and only do a server call if there is any hw bp enabled. */
-        if (!(context.EFlags & 0x100) || (context.Dr7 & 0xff))
+        if (!(xcontext.c.EFlags & 0x100) || (xcontext.c.Dr7 & 0xff))
         {
             /* (possible) hardware breakpoint, fetch the debug registers */
-            DWORD saved_flags = context.ContextFlags;
-            context.ContextFlags = CONTEXT_DEBUG_REGISTERS;
-            NtGetContextThread( GetCurrentThread(), &context );
-            context.ContextFlags |= saved_flags;  /* restore flags */
+            DWORD saved_flags = xcontext.c.ContextFlags;
+            xcontext.c.ContextFlags = CONTEXT_DEBUG_REGISTERS;
+            NtGetContextThread( GetCurrentThread(), &xcontext.c );
+            xcontext.c.ContextFlags |= saved_flags;  /* restore flags */
         }
-        context.EFlags &= ~0x100;  /* clear single-step flag */
+        xcontext.c.EFlags &= ~0x100;  /* clear single-step flag */
         break;
     case TRAP_x86_BPTFLT:   /* Breakpoint exception */
         rec.ExceptionAddress = (char *)rec.ExceptionAddress - 1;  /* back up over the int3 instruction */
@@ -1786,7 +1850,7 @@ static void trap_handler( int signal, siginfo_t *siginfo, void *sigcontext )
         rec.ExceptionInformation[2] = 0; /* FIXME */
         break;
     }
-    setup_raise_exception( sigcontext, stack, &rec, &context );
+    setup_raise_exception( sigcontext, stack, &rec, &xcontext );
 }

@@ -1798,9 +1862,9 @@ static void trap_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 static void fpe_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 {
     EXCEPTION_RECORD rec = { 0 };
-    CONTEXT context;
+    struct xcontext xcontext;
     ucontext_t *ucontext = sigcontext;
-    void *stack = setup_exception_record( sigcontext, &rec, &context );
+    void *stack = setup_exception_record( sigcontext, &rec, &xcontext );

switch (TRAP_sig(ucontext))
     {
@@ -1811,8 +1875,8 @@ static void fpe_handler( int signal, siginfo_t *siginfo, void *sigcontext )
         rec.ExceptionCode = EXCEPTION_FLT_INVALID_OPERATION;
         break;
     case TRAP_x86_ARITHTRAP:  /* Floating point exception */
-        rec.ExceptionCode = get_fpu_code( &context );
-        rec.ExceptionAddress = (void *)context.FloatSave.ErrorOffset;
+        rec.ExceptionCode = get_fpu_code( &xcontext.c );
+        rec.ExceptionAddress = (void *)xcontext.c.FloatSave.ErrorOffset;
         break;
     case TRAP_x86_CACHEFLT:  /* SIMD exception */
         /* TODO:
@@ -1832,7 +1896,7 @@ static void fpe_handler( int signal, siginfo_t *siginfo, void *sigcontext )
         rec.ExceptionCode = EXCEPTION_FLT_INVALID_OPERATION;
         break;
     }
-    setup_raise_exception( sigcontext, stack, &rec, &context );
+    setup_raise_exception( sigcontext, stack, &rec, &xcontext );
 }

@@ -1880,12 +1944,12 @@ static void quit_handler( int signal, siginfo_t *siginfo, void *sigcontext )
  */
 static void usr1_handler( int signal, siginfo_t *siginfo, void *sigcontext )
 {
-    CONTEXT context;
+    struct xcontext xcontext;

init_handler( sigcontext );
-    save_context( &context, sigcontext );
-    wait_suspend( &context );
-    restore_context( &context, sigcontext );
+    save_context( &xcontext, sigcontext );
+    wait_suspend( &xcontext.c );
+    restore_context( &xcontext, sigcontext );
 }

[PATCH 4/5] ntdll: Support AVX context in fault exceptions on Linux x86.

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=77414

Your paranoid android.

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3960 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_2scr (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_ar (testbot log) ===

An error occurred while waiting for the test to complete: the 780 process does not exist or is not a child process

=== w1064v1809_he (testbot log) ===

An error occurred while waiting for the test to complete: the 1872 process does not exist or is not a child process

=== w1064v1809_ja (testbot log) ===

An error occurred while waiting for the test to complete: the 2532 process does not exist or is not a child process

=== w1064v1809_zh_CN (testbot log) ===

An error occurred while waiting for the test to complete: the 5136 process does not exist or is not a child process

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3940 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

[PATCH 5/5] ntdll: Restore AVX registers in NtSetContextThread() on x86.

[Paul Gofman]

Signed-off-by: Paul Gofman 
---
 dlls/ntdll/tests/exception.c  |  2 --
 dlls/ntdll/unix/signal_i386.c | 38 +++++++++++++++++++++++++++++++++--
 2 files changed, 36 insertions(+), 4 deletions(-)

diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c
index 860221366ce..9f93212acf6 100644
--- a/dlls/ntdll/tests/exception.c
+++ b/dlls/ntdll/tests/exception.c
@@ -5600,7 +5600,6 @@ static void test_extended_context(void)
     for (i = 0; i < 8; ++i)
     {
         /* Older Windows version do not reset AVX context to INIT_STATE on x86. */
-        todo_wine_if(i >= 4 && sizeof(void *) == 4)
         ok(!data[i] || broken(i >= 4 && sizeof(void *) == 4 && data[i] == test_extended_context_spoil_data2[i]),
                 "Got unexpected data %#x, i %u.\n", data[i], i);
     }
@@ -5611,7 +5610,6 @@ static void test_extended_context(void)
             ARRAY_SIZE(except_code_set_ymm0), PAGE_EXECUTE_READ);

for (i = 0; i < 8; ++i)
-        todo_wine_if(i >= 4 && sizeof(void *) == 4)
         ok(data[i] == test_extended_context_data[i], "Got unexpected data %#x, i %u.\n", data[i], i);
 }
 #endif
diff --git a/dlls/ntdll/unix/signal_i386.c b/dlls/ntdll/unix/signal_i386.c
index 791ffe90940..e03cb4a517b 100644
--- a/dlls/ntdll/unix/signal_i386.c
+++ b/dlls/ntdll/unix/signal_i386.c
@@ -687,7 +687,7 @@ static inline void save_fpux( CONTEXT *context )
 /***********************************************************************
  *           restore_fpu
  *
- * Restore the FPU context to a sigcontext.
+ * Restore the x87 FPU context
  */
 static inline void restore_fpu( const CONTEXT *context )
 {
@@ -701,7 +701,7 @@ static inline void restore_fpu( const CONTEXT *context )
 /***********************************************************************
  *           restore_fpux
  *
- * Restore the FPU extended context to a sigcontext.
+ * Restore the FPU extended context
  */
 static inline void restore_fpux( const CONTEXT *context )
 {
@@ -715,6 +715,38 @@ static inline void restore_fpux( const CONTEXT *context )
     __asm__ __volatile__( "fxrstor %0" : : "m" (*state) );
 }

+/***********************************************************************
+ *           restore_xstate
+ *
+ * Restore the XState context
+ */
+static inline void restore_xstate( const CONTEXT *context )
+{
+    XSAVE_FORMAT *xrstor_base;
+    XSTATE *xs;
+
+    if (!(xs = xstate_from_context( context )))
+        return;
+
+    xrstor_base = (XSAVE_FORMAT *)xs - 1;
+
+    if (!(xs->CompactionMask & ((ULONG64)1 << 63)))
+    {
+        /* Non-compacted xrstor will load Mxcsr regardless of the specified mask. Loading garbage there
+         * may lead to fault. FPUX state should be restored by now, so we can reuse some space in
+         * ExtendedRegisters. */
+        XSAVE_FORMAT *fpux = (XSAVE_FORMAT *)context->ExtendedRegisters;
+        DWORD mxcsr, mxcsr_mask;
+
+        mxcsr = fpux->MxCsr;
+        mxcsr_mask = fpux->MxCsr_Mask;
+
+        assert( (void *)&xrstor_base->MxCsr > (void *)context->ExtendedRegisters );
+        xrstor_base->MxCsr = mxcsr;
+        xrstor_base->MxCsr_Mask = mxcsr_mask;
+    }
+    __asm__ volatile("xor %%edx,%%edx; mov $4,%%eax; xrstor %0" : : "m"(*xrstor_base) : "eax", "edx");
+}

/***********************************************************************
  *           fpux_to_fpu
@@ -1129,6 +1161,8 @@ NTSTATUS WINAPI NtSetContextThread( HANDLE handle, const CONTEXT *context )
     if (flags & CONTEXT_EXTENDED_REGISTERS) restore_fpux( context );
     else if (flags & CONTEXT_FLOATING_POINT) restore_fpu( context );

+    restore_xstate( context );
+
     if (flags & CONTEXT_FULL)
     {
         if (!(flags & CONTEXT_CONTROL))

[PATCH 5/5] ntdll: Restore AVX registers in NtSetContextThread() on x86.

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=77415

Your paranoid android.

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3976 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: the 5356 process does not exist or is not a child process

=== w1064v1809_2scr (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_ar (testbot log) ===

An error occurred while waiting for the test to complete: the 1448 process does not exist or is not a child process

=== w1064v1809_he (testbot log) ===

An error occurred while waiting for the test to complete: the 1560 process does not exist or is not a child process

=== w1064v1809_ja (testbot log) ===

An error occurred while waiting for the test to complete: the 1508 process does not exist or is not a child process

=== w1064v1809_zh_CN (testbot log) ===

An error occurred while waiting for the test to complete: the 5404 process does not exist or is not a child process

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3952 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=77411

Your paranoid android.

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3952 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_2scr (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_ar (testbot log) ===

An error occurred while waiting for the test to complete: network read timed out (wait2/connect:AgentVersion.h:0/9)

=== w1064v1809_he (testbot log) ===

An error occurred while waiting for the test to complete: the 5540 process does not exist or is not a child process

=== w1064v1809_ja (testbot log) ===

An error occurred while waiting for the test to complete: the 4144 process does not exist or is not a child process

=== w1064v1809_zh_CN (testbot log) ===

An error occurred while waiting for the test to complete: the 2792 process does not exist or is not a child process

=== w1064v1507 (testbot log) ===

An error occurred while waiting for the test to complete: the 3948 process does not exist or is not a child process

=== w1064v1809 (testbot log) ===

An error occurred while waiting for the test to complete: network read got a premature EOF (wait2/connect:AgentVersion.h:0/9)