Use mcr for dmb instruction on armv6

[raj.khem]

Summary

At present compiler-rt cross compiles for armv6 ( -march=armv6 ) but includes dmb instructions which are only available in armv7+ this causes SIGILL on clang+compiler-rt compiled components on rpi0w platforms.

[raj.khem]

ping,

[MaskRay]

compiler-rt/lib/builtins/assembly.h

195

I see that musl src/thread/arm/atomics.s uses .arch armv6t2; mcr p15,0,r0,c7,c10,5

Is #elif __ARM_ARCH >= 6 accurate?

[raj.khem]

compiler-rt/lib/builtins/assembly.h

195

yes I think it is correct. Musl is using that directive for generating thumb2 code when -mthumb is used along with mcr barrier. mcr is available in both thumb/arm ISAs on ARMv6

[raj.khem]

rebased

[raj.khem]

@MaskRay can you help in moving this forward ?

[efriedma]

I'm not sure this is right... even if dmb barriers do work, strex is only available in v7+ in Thumb mode, so I suspect this still causes a SIGILL on v6, whether or not Thumb2 is available.

compiler-rt/lib/builtins/arm/sync-ops.h

 
 #include "../assembly.h"
 
 #define SYNC_OP_4(op)                                                          \
   .p2align 2;                                                                  \
   .thumb;                                                                      \
   .syntax unified;                                                             \
   DEFINE_COMPILERRT_THUMB_FUNCTION(__sync_fetch_and_##op)                      \
-  dmb;                                                                         \
+  DMB;                                                                         \
   mov r12, r0;                                                                 \
   LOCAL_LABEL(tryatomic_##op) : ldrex r0, [r12];                               \
   op(r2, r0, r1);                                                              \
   strex r3, r2, [r12];                                                         \
   cmp r3, #0;                                                                  \
   bne LOCAL_LABEL(tryatomic_##op);                                             \
-  dmb;                                                                         \
+  DMB;                                                                         \
   bx lr
 
 #define SYNC_OP_8(op)                                                          \
   .p2align 2;                                                                  \
   .thumb;                                                                      \
   .syntax unified;                                                             \
   DEFINE_COMPILERRT_THUMB_FUNCTION(__sync_fetch_and_##op)                      \
   push {r4, r5, r6, lr};                                                       \
-  dmb;                                                                         \
+  DMB;                                                                         \
   mov r12, r0;                                                                 \
   LOCAL_LABEL(tryatomic_##op) : ldrexd r0, r1, [r12];                          \
   op(r4, r5, r0, r1, r2, r3);                                                  \
   strexd r6, r4, r5, [r12];                                                    \
   cmp r6, #0;                                                                  \
   bne LOCAL_LABEL(tryatomic_##op);                                             \
-  dmb;                                                                         \
+  DMB;                                                                         \
   pop { r4, r5, r6, pc }
 
 #define MINMAX_4(rD, rN, rM, cmp_kind)                                         \
   cmp rN, rM;                                                                  \
   mov rD, rM;                                                                  \
   it cmp_kind;                                                                 \
   mov##cmp_kind rD, rN
 
 #define MINMAX_8(rD_LO, rD_HI, rN_LO, rN_HI, rM_LO, rM_HI, cmp_kind)           \
   cmp rN_LO, rM_LO;                                                            \
   sbcs rN_HI, rM_HI;                                                           \
   mov rD_LO, rM_LO;                                                            \
   mov rD_HI, rM_HI;                                                            \
   itt cmp_kind;                                                                \
   mov##cmp_kind rD_LO, rN_LO;                                                  \
   mov##cmp_kind rD_HI, rN_HI

compiler-rt/lib/builtins/assembly.h

 #if __ARM_ARCH >= 5
 #define POP_PC() pop {pc}
 #else
 #define POP_PC()                                                               \
   pop {ip};                                                                    \
   JMP(ip)
 #endif
 
+#if __ARM_ARCH >= 7
+#define DMB dmb
+#elif __ARM_ARCH >= 6
+#define DMB mcr p15, #0, r0, c7, c10, #5
+#else
+#error only supported on ARMv6+
+#endif
+
 #if defined(USE_THUMB_2)
 #define WIDE(op) op.w
 #else
 #define WIDE(op) op
 #endif
 #else // !defined(__arm)
 #define DECLARE_FUNC_ENCODING
 #define DEFINE_CODE_STATE