diff --git a/common/src/x86emitter/LnxCpuDetect.cpp b/common/src/x86emitter/LnxCpuDetect.cpp
index 747b1c784d..de2f913e05 100644
--- a/common/src/x86emitter/LnxCpuDetect.cpp
+++ b/common/src/x86emitter/LnxCpuDetect.cpp
@@ -28,13 +28,6 @@ void x86capabilities::CountLogicalCores()
 	LogicalCores = wxThread::GetCPUCount();
 }
 
-bool CanEmitShit()
-{
-	// In Linux I'm pretty sure TLS always works, none of the funny business that Windows
-	// has involving DLLs. >_<
-	return true;
-}
-
 // Not implemented yet for linux (see cpudetect_internal.h for details)
 SingleCoreAffinity::SingleCoreAffinity() {}
 SingleCoreAffinity::~SingleCoreAffinity() throw() {}
diff --git a/common/src/x86emitter/WinCpuDetect.cpp b/common/src/x86emitter/WinCpuDetect.cpp
index 35957087fe..7de69371e6 100644
--- a/common/src/x86emitter/WinCpuDetect.cpp
+++ b/common/src/x86emitter/WinCpuDetect.cpp
@@ -39,21 +39,6 @@ void x86capabilities::CountLogicalCores()
 	LogicalCores = CPUs;
 }
 
-bool CanEmitShit()
-{
-	// Under Windows, pre 0.9.6 versions of PCSX2 may not initialize the TLS
-	// register (FS register), so plugins (DLLs) using our x86emitter in multithreaded
-	// mode will just crash/fail if it tries to do the instruction set tests.
-
-#if x86EMIT_MULTITHREADED
-	static __threadlocal int tls_failcheck;
-	__try { tls_failcheck = 1; }
-	__except(EXCEPTION_EXECUTE_HANDLER) { return false; }
-#endif
-
-	return true;
-}
-
 SingleCoreAffinity::SingleCoreAffinity()
 {
 	s_threadId	= NULL;
diff --git a/common/src/x86emitter/cpudetect.cpp b/common/src/x86emitter/cpudetect.cpp
index 7b5388be30..0e58986660 100644
--- a/common/src/x86emitter/cpudetect.cpp
+++ b/common/src/x86emitter/cpudetect.cpp
@@ -22,7 +22,6 @@ using namespace x86Emitter;
 __aligned16 x86capabilities x86caps;
 
 // Recompiled code buffer for SSE and MXCSR feature testing.
-static __pagealigned u8 recSSE[__pagesize];
 static __pagealigned u8 targetFXSAVE[512];
 
 static const char* bool_to_char( bool testcond )
@@ -34,6 +33,7 @@ static const char* bool_to_char( bool testcond )
 // MSVC PGO builds.  The problem was fixed when I moved the MXCSR code to this function, and
 // moved the recSSE[] array to a global static (it was local to cpudetectInit).  Commented
 // here in case the nutty crash ever re-surfaces. >_<
+// Note: recSSE was deleted
 void x86capabilities::SIMD_EstablishMXCSRmask()
 {
 	if( !hasStreamingSIMDExtensions ) return;
@@ -48,32 +48,9 @@ void x86capabilities::SIMD_EstablishMXCSRmask()
 
 		MXCSR_Mask.bitmask = 0xFFFF;	// SSE2 features added
 	}
-#ifdef _M_X86_64
-#ifdef _MSC_VER
-	// Use the intrinsic that is provided with MSVC 2012
+
+	// Work for recent enough GCC/CLANG/MSVC 2012
 	_fxsave(&targetFXSAVE);
-#else
-	// GCC path is supported since GCC 4.6.x
-	__asm __volatile ("fxsave %0" : "+m" (targetFXSAVE));
-#endif
-#else
-	// Grab the MXCSR mask the x86_32 way.
-	//
-	// the fxsave buffer must be 16-byte aligned to avoid GPF.  I just save it to an
-	// unused portion of recSSE, since it has plenty of room to spare.
-
-	if( !CanEmitShit() ) return;
-
-	HostSys::MemProtectStatic( recSSE, PageAccess_ReadWrite() );
-
-	xSetPtr( recSSE );
-	xFXSAVE( ptr[&targetFXSAVE] );
-	xRET();
-
-	HostSys::MemProtectStatic( recSSE, PageAccess_ExecOnly() );
-
-	CallAddress( recSSE );
-#endif
 
 	u32 result = (u32&)targetFXSAVE[28];			// bytes 28->32 are the MXCSR_Mask.
 	if( result != 0 )
diff --git a/common/src/x86emitter/cpudetect_internal.h b/common/src/x86emitter/cpudetect_internal.h
index d1e4fb30c2..94be9b1329 100644
--- a/common/src/x86emitter/cpudetect_internal.h
+++ b/common/src/x86emitter/cpudetect_internal.h
@@ -37,18 +37,3 @@ public:
 	SingleCoreAffinity();
 	virtual ~SingleCoreAffinity() throw();
 };
-
-// --------------------------------------------------------------------------------------
-//  SIMD "Manual" Detection, using Invalid Instruction Checks
-// --------------------------------------------------------------------------------------
-//
-// Note: This API doesn't support GCC/Linux.  Looking online it seems the only
-// way to simulate the Microsoft SEH model is to use unix signals, and the 'sigaction'
-// function specifically.  A linux coder could implement this using sigaction at a later
-// date, however its not really a big deal:  CPUID should be 99-100% accurate, as no modern
-// software would work on the CPU if it mis-reported capabilities.  However there are known
-// cases of a CPU failing to report supporting instruction sets it does in fact support.
-// This secondary test fixes such cases (although apparently a CMOS reset does as well).
-//
-
-extern bool CanEmitShit();