Index: lib/builtins/arm/sync-ops.h
===================================================================
--- lib/builtins/arm/sync-ops.h
+++ lib/builtins/arm/sync-ops.h
@@ -15,6 +15,20 @@
 
 #include "../assembly.h"
 
+#if __ARM_ARCH_PROFILE == 'M'
+#define SYNC_OP_4(op) \
+        .thumb ; \
+        .syntax unified ; \
+        DEFINE_COMPILERRT_THUMB_FUNCTION(__sync_fetch_and_ ## op) \
+        mrs r3, PRIMASK ; \
+        cpsid i ; \
+        mov r2, r0 ; \
+        ldr r0, [r2] ; \
+        op(r1, r1, r0) ; \
+        str r1, [r2] ; \
+        msr PRIMASK, r3 ; \
+        bx lr;
+#else
 #define SYNC_OP_4(op) \
         .p2align 2 ; \
         .thumb ; \
@@ -30,7 +44,25 @@
         bne LOCAL_LABEL(tryatomic_ ## op) ; \
         dmb ; \
         bx lr
+#endif
 
+#if __ARM_ARCH_PROFILE == 'M'
+#define SYNC_OP_8(op) \
+        .thumb ; \
+        .syntax unified ; \
+        DEFINE_COMPILERRT_THUMB_FUNCTION(__sync_fetch_and_ ## op) \
+        push {r4, r5, lr} ; \
+        mov r4, r0 ; \
+        mrs r5, PRIMASK ; \
+        cpsid i ; \
+        ldr r0, [r4, #0] ; \
+        ldr r1, [r4, #4] ; \
+        op(r2, r3, r2, r3, r0, r1) ; \
+        str r2, [r4, #0] ; \
+        str r3, [r4, #4] ; \
+        msr PRIMASK, r5 ; \
+        pop {r4, r5, pc} ;
+#else
 #define SYNC_OP_8(op) \
         .p2align 2 ; \
         .thumb ; \
@@ -47,6 +79,7 @@
         bne LOCAL_LABEL(tryatomic_ ## op) ; \
         dmb ; \
         pop {r4, r5, r6, pc}
+#endif
 
 #define MINMAX_4(rD, rN, rM, cmp_kind) \
         cmp rN, rM ; \
Index: lib/builtins/arm/sync_fetch_and_add_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_add_8.S
+++ lib/builtins/arm/sync_fetch_and_add_8.S
@@ -14,13 +14,11 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define add_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI) \
     adds rD_LO, rN_LO, rM_LO ; \
     adc rD_HI, rN_HI, rM_HI
 
 SYNC_OP_8(add_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE
 
Index: lib/builtins/arm/sync_fetch_and_and_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_and_8.S
+++ lib/builtins/arm/sync_fetch_and_and_8.S
@@ -14,13 +14,11 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define and_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI) \
     and rD_LO, rN_LO, rM_LO ; \
     and rD_HI, rN_HI, rM_HI
 
 SYNC_OP_8(and_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE
 
Index: lib/builtins/arm/sync_fetch_and_max_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_max_8.S
+++ lib/builtins/arm/sync_fetch_and_max_8.S
@@ -14,11 +14,9 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define max_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI)         MINMAX_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI, gt)
 
 SYNC_OP_8(max_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE
 
Index: lib/builtins/arm/sync_fetch_and_nand_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_nand_8.S
+++ lib/builtins/arm/sync_fetch_and_nand_8.S
@@ -14,13 +14,11 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define nand_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI) \
     bic rD_LO, rN_LO, rM_LO ; \
     bic rD_HI, rN_HI, rM_HI
 
 SYNC_OP_8(nand_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE
 
Index: lib/builtins/arm/sync_fetch_and_or_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_or_8.S
+++ lib/builtins/arm/sync_fetch_and_or_8.S
@@ -14,13 +14,11 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define or_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI) \
     orr rD_LO, rN_LO, rM_LO ; \
     orr rD_HI, rN_HI, rM_HI
 
 SYNC_OP_8(or_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE
 
Index: lib/builtins/arm/sync_fetch_and_sub_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_sub_8.S
+++ lib/builtins/arm/sync_fetch_and_sub_8.S
@@ -14,13 +14,11 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define sub_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI) \
     subs rD_LO, rN_LO, rM_LO ; \
     sbc rD_HI, rN_HI, rM_HI
 
 SYNC_OP_8(sub_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE
 
Index: lib/builtins/arm/sync_fetch_and_xor_8.S
===================================================================
--- lib/builtins/arm/sync_fetch_and_xor_8.S
+++ lib/builtins/arm/sync_fetch_and_xor_8.S
@@ -14,13 +14,11 @@
 
 #include "sync-ops.h"
 
-#if __ARM_ARCH_PROFILE != 'M'
 #define xor_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI) \
     eor rD_LO, rN_LO, rM_LO ; \
     eor rD_HI, rN_HI, rM_HI
 
 SYNC_OP_8(xor_8)
-#endif
 
 NO_EXEC_STACK_DIRECTIVE