diff --git a/dll/win32/kernel32/client/utils.c b/dll/win32/kernel32/client/utils.c
index c10d2d45004..54a6f05fa5c 100644
--- a/dll/win32/kernel32/client/utils.c
+++ b/dll/win32/kernel32/client/utils.c
@@ -600,6 +600,9 @@ BaseInitializeContext(IN PCONTEXT Context,
     /* Set the EFLAGS */
     Context->EFlags = 0x3000 | EFLAGS_INTERRUPT_MASK; /* IOPL 3 */
 
+    /* Set MXCSR */
+    Context->MxCsr = INITIAL_MXCSR;
+
     if (ContextType == 1)      /* For Threads */
     {
         Context->Rip = (ULONG_PTR)BaseThreadStartup;
diff --git a/ntoskrnl/ke/amd64/context.c b/ntoskrnl/ke/amd64/context.c
index 0decbde88f9..1e9900af11f 100644
--- a/ntoskrnl/ke/amd64/context.c
+++ b/ntoskrnl/ke/amd64/context.c
@@ -56,7 +56,7 @@ KeContextToTrapFrame(IN PCONTEXT Context,
     }
 
     /* Handle floating point registers */
-    if ((ContextFlags & CONTEXT_FLOATING_POINT))
+    if (ContextFlags & CONTEXT_FLOATING_POINT)
     {
         TrapFrame->MxCsr = Context->MxCsr;
         TrapFrame->Xmm0 = Context->Xmm0;
@@ -207,9 +207,9 @@ KeTrapFrameToContext(IN PKTRAP_FRAME TrapFrame,
     }
 
     /* Handle floating point registers */
-    if ((ContextFlags & CONTEXT_FLOATING_POINT) &&
-        ((TrapFrame->SegCs & MODE_MASK) != KernelMode))
+    if (ContextFlags & CONTEXT_FLOATING_POINT)
     {
+        Context->MxCsr = TrapFrame->MxCsr;
         Context->Xmm0 = TrapFrame->Xmm0;
         Context->Xmm1 = TrapFrame->Xmm1;
         Context->Xmm2 = TrapFrame->Xmm2;
diff --git a/ntoskrnl/ke/amd64/cpu.c b/ntoskrnl/ke/amd64/cpu.c
index d4a8888c664..0967411e644 100644
--- a/ntoskrnl/ke/amd64/cpu.c
+++ b/ntoskrnl/ke/amd64/cpu.c
@@ -375,7 +375,7 @@ KiRestoreProcessorControlState(PKPROCESSOR_STATE ProcessorState)
 //    __ltr(&ProcessorState->SpecialRegisters.Tr);
     __lidt(&ProcessorState->SpecialRegisters.Idtr.Limit);
 
-//    __ldmxcsr(&ProcessorState->SpecialRegisters.MxCsr); // FIXME
+    _mm_setcsr(ProcessorState->SpecialRegisters.MxCsr);
 //    ProcessorState->SpecialRegisters.DebugControl
 //    ProcessorState->SpecialRegisters.LastBranchToRip
 //    ProcessorState->SpecialRegisters.LastBranchFromRip
@@ -417,7 +417,7 @@ KiSaveProcessorControlState(OUT PKPROCESSOR_STATE ProcessorState)
     __str(&ProcessorState->SpecialRegisters.Tr);
     __sidt(&ProcessorState->SpecialRegisters.Idtr.Limit);
 
-//    __stmxcsr(&ProcessorState->SpecialRegisters.MxCsr);
+    ProcessorState->SpecialRegisters.MxCsr = _mm_getcsr();
 //    ProcessorState->SpecialRegisters.DebugControl =
 //    ProcessorState->SpecialRegisters.LastBranchToRip =
 //    ProcessorState->SpecialRegisters.LastBranchFromRip =
diff --git a/ntoskrnl/ke/amd64/kiinit.c b/ntoskrnl/ke/amd64/kiinit.c
index d9c8807bf34..310810475ea 100644
--- a/ntoskrnl/ke/amd64/kiinit.c
+++ b/ntoskrnl/ke/amd64/kiinit.c
@@ -152,6 +152,9 @@ KiInitializePcr(IN PKIPCR Pcr,
     Pcr->Prcb.ProcessorState.SpecialRegisters.KernelDr6 = 0;
     Pcr->Prcb.ProcessorState.SpecialRegisters.KernelDr7 = 0;
 
+    /* Initialize MXCSR (all exceptions masked) */
+    Pcr->Prcb.MxCsr = INITIAL_MXCSR;
+
     /* Set the Current Thread */
     Pcr->Prcb.CurrentThread = IdleThread;
 
@@ -231,6 +234,9 @@ KiInitializeCpu(PKIPCR Pcr)
     Pat = (PAT_WB << 0)  | (PAT_WC << 8) | (PAT_UCM << 16) | (PAT_UC << 24) |
           (PAT_WB << 32) | (PAT_WC << 40) | (PAT_UCM << 48) | (PAT_UC << 56);
     __writemsr(MSR_PAT, Pat);
+
+    /* Initialize MXCSR */
+    _mm_setcsr(INITIAL_MXCSR);
 }
 
 VOID
diff --git a/sdk/lib/rtl/amd64/stubs.c b/sdk/lib/rtl/amd64/stubs.c
index 7ffc4fb9658..0f02e60a41a 100644
--- a/sdk/lib/rtl/amd64/stubs.c
+++ b/sdk/lib/rtl/amd64/stubs.c
@@ -23,7 +23,7 @@ NTAPI
 RtlInitializeContext(
     _Reserved_ HANDLE ProcessHandle,
     _Out_ PCONTEXT ThreadContext,
-    _In_ PVOID ThreadStartParam  OPTIONAL,
+    _In_opt_ PVOID ThreadStartParam,
     _In_ PTHREAD_START_ROUTINE ThreadStartAddress,
     _In_ PINITIAL_TEB StackBase)
 {
@@ -66,10 +66,13 @@ RtlInitializeContext(
         ThreadContext->SegSs = KGDT64_R3_DATA |  RPL_MASK;
     }
 
+    ThreadContext->MxCsr = INITIAL_MXCSR;
+
     /* Only the basic Context is initialized */
     ThreadContext->ContextFlags = CONTEXT_CONTROL |
                                   CONTEXT_INTEGER |
-                                  CONTEXT_SEGMENTS;
+                                  CONTEXT_SEGMENTS |
+                                  CONTEXT_FLOATING_POINT;
 
     return;
 }