[RTL/x64] Fix RtlCaptureContext/RtlpRestoreContextInternal in kernel mode

According to tests, legacy fp state is not saved in kernel mode.
Also add an int 2c to the path that changes cs, as it should not be used and probably never will be.

sdk/lib/rtl/amd64/except_asm.S

@@ -94,12 +94,20 @@ PUBLIC RtlCaptureContext
     movaps [rcx + CxXmm14], xmm14
     movaps [rcx + CxXmm15], xmm15
 
-    /* Store legacy floating point registers */
-    fxsave [rcx + CxFltSave]
+    /* Store eflags */
+    mov [rcx + CxEFlags], eax
+
+    /* Store mxcsr */
     stmxcsr [rcx + CxMxCsr]
 
-    /* Store rflags */
-    mov [rcx + CxEFlags], eax
+    /* Check if we are in user mode */
+    test byte ptr [rcx + CxSegCs], 3
+    jz RtlCaptureContextExit
+
+    /* Store legacy floating point registers */
+    fxsave [rcx + CxFltSave]
+
+RtlCaptureContextExit:
 
     /* Cleanup stack and return */
     add rsp, 8
@@ -119,10 +127,6 @@ PUBLIC RtlpRestoreContextInternal
     .ALLOCSTACK 8
     .ENDPROLOG
 
-    /* Restore legacy floating point registers (It is slow, so do it first) */
-    ldmxcsr [rcx + CxMxCsr]
-    fxrstor [rcx + CxFltSave]
-
     /* Load the target stack pointer into rdx */
     mov rdx, [rcx + CxRsp]
 
@@ -185,6 +189,18 @@ PUBLIC RtlpRestoreContextInternal
     mov r8,  [rcx + CxR8]
     mov r9,  [rcx + CxR9]
 
+    /* Restore MXCSR */
+    ldmxcsr [rcx + CxMxCsr]
+
+    /* Check if we go to user mode */
+    test byte ptr [rcx + CxSegCs], 3
+    jz Exit
+
+    /* Restore legacy floating point registers */
+    fxrstor [rcx + CxFltSave]
+
+Exit:
+
     /* Check if we go to a different cs */
     mov ax, cs
     cmp [rcx + CxSegCs], ax
@@ -208,6 +224,9 @@ PUBLIC RtlpRestoreContextInternal
 
 ReturnFar:
 
+    // We should never need this path
+    int HEX(2c)
+
     /* Put cs on the stack for the far return */
     mov ax, [rcx + CxSegCs]
     mov [rdx - 1 * 8], ax