ARM: Support avoiding pc as destination register.

3 files changed, 60 insertions, 2 deletions

diff --git a/ports/sysdeps/arm/arm-features.h b/ports/sysdeps/arm/arm-features.h
index 31801cf408..139a40368b 100644
--- a/ports/sysdeps/arm/arm-features.h
+++ b/ports/sysdeps/arm/arm-features.h
@@ -36,4 +36,8 @@
    at runtime (or that we never care about its state) and so need not
    be checked for.  */
 
+/* A more-specific arm-features.h file may define ARM_ALWAYS_BX to indicate
+   that instructions using pc as a destination register must never be used,
+   so a "bx" (or "blx") instruction is always required.  */
+
 #endif  /* arm-features.h */
diff --git a/ports/sysdeps/arm/memcpy.S b/ports/sysdeps/arm/memcpy.S
index eb9699ddf6..779f403576 100644
--- a/ports/sysdeps/arm/memcpy.S
+++ b/ports/sysdeps/arm/memcpy.S
@@ -20,6 +20,7 @@
 /* Thumb requires excessive IT insns here.  */
 #define NO_THUMB
 #include <sysdep.h>
+#include <arm-features.h>
 
 /*
  * Data preload for architectures that support it (ARM V5TE and above)
@@ -89,7 +90,12 @@ ENTRY(memcpy)
 	CALGN(	bcs	2f			)
 	CALGN(	adr	r4, 6f			)
 	CALGN(	subs	r2, r2, r3		)  @ C gets set
+#ifndef ARM_ALWAYS_BX
 	CALGN(	add	pc, r4, ip		)
+#else
+	CALGN(	add	r4, r4, ip		)
+	CALGN(	bx	r4			)
+#endif
 
 	PLD(	pld	[r1, #0]		)
 2:	PLD(	subs	r2, r2, #96		)
@@ -108,8 +114,17 @@ ENTRY(memcpy)
 
 5:		ands	ip, r2, #28
 		rsb	ip, ip, #32
+#ifndef ARM_ALWAYS_BX
 		addne	pc, pc, ip		@ C is always clear here
 		b	7f
+#else
+		beq	7f
+		push	{r10}
+		cfi_adjust_cfa_offset (4)
+		cfi_rel_offset (r10, 0)
+		add	r10, pc, ip
+		bx	r10
+#endif
 6:		nop
 		ldr	r3, [r1], #4
 		ldr	r4, [r1], #4
@@ -119,8 +134,13 @@ ENTRY(memcpy)
 		ldr	r8, [r1], #4
 		ldr	lr, [r1], #4
 
+#ifndef ARM_ALWAYS_BX
 		add	pc, pc, ip
 		nop
+#else
+		add	r10, pc, ip
+		bx	r10
+#endif
 		nop
 		str	r3, [r0], #4
 		str	r4, [r0], #4
@@ -130,6 +150,12 @@ ENTRY(memcpy)
 		str	r8, [r0], #4
 		str	lr, [r0], #4
 
+#ifdef ARM_ALWAYS_BX
+		pop	{r10}
+		cfi_adjust_cfa_offset (-4)
+		cfi_restore (r10)
+#endif
+
 	CALGN(	bcs	2b			)
 
 7:		pop	{r5 - r8}
@@ -147,7 +173,8 @@ ENTRY(memcpy)
 		strbcs	r4, [r0], #1
 		strbcs	ip, [r0]
 
-#if defined (__ARM_ARCH_4T__) && defined(__THUMB_INTERWORK__)
+#if ((defined (__ARM_ARCH_4T__) && defined(__THUMB_INTERWORK__)) \
+     || defined (ARM_ALWAYS_BX))
 		pop	{r0, r4, lr}
 		cfi_adjust_cfa_offset (-12)
 		cfi_restore (r4)
diff --git a/ports/sysdeps/arm/memmove.S b/ports/sysdeps/arm/memmove.S
index 9e8ad65f44..4a2cb92ea4 100644
--- a/ports/sysdeps/arm/memmove.S
+++ b/ports/sysdeps/arm/memmove.S
@@ -20,6 +20,7 @@
 /* Thumb requires excessive IT insns here.  */
 #define NO_THUMB
 #include <sysdep.h>
+#include <arm-features.h>
 
 /*
  * Data preload for architectures that support it (ARM V5TE and above)
@@ -105,7 +106,12 @@ ENTRY(memmove)
 	CALGN(	bcs	2f			)
 	CALGN(	adr	r4, 6f			)
 	CALGN(	subs	r2, r2, ip		)  @ C is set here
+#ifndef ARM_ALWAYS_BX
 	CALGN(	add	pc, r4, ip		)
+#else
+	CALGN(	add	r4, r4, ip		)
+	CALGN(	bx	r4			)
+#endif
 
 	PLD(	pld	[r1, #-4]		)
 2:	PLD(	subs	r2, r2, #96		)
@@ -124,8 +130,17 @@ ENTRY(memmove)
 
 5:		ands	ip, r2, #28
 		rsb	ip, ip, #32
+#ifndef ARM_ALWAYS_BX
 		addne	pc, pc, ip		@ C is always clear here
 		b	7f
+#else
+		beq	7f
+		push	{r10}
+		cfi_adjust_cfa_offset (4)
+		cfi_rel_offset (r10, 0)
+		add	r10, pc, ip
+		bx	r10
+#endif
 6:		nop
 		ldr	r3, [r1, #-4]!
 		ldr	r4, [r1, #-4]!
@@ -135,8 +150,13 @@ ENTRY(memmove)
 		ldr	r8, [r1, #-4]!
 		ldr	lr, [r1, #-4]!
 
+#ifndef ARM_ALWAYS_BX
 		add	pc, pc, ip
 		nop
+#else
+		add	r10, pc, ip
+		bx	r10
+#endif
 		nop
 		str	r3, [r0, #-4]!
 		str	r4, [r0, #-4]!
@@ -146,6 +166,12 @@ ENTRY(memmove)
 		str	r8, [r0, #-4]!
 		str	lr, [r0, #-4]!
 
+#ifdef ARM_ALWAYS_BX
+		pop	{r10}
+		cfi_adjust_cfa_offset (-4)
+		cfi_restore (r10)
+#endif
+
 	CALGN(	bcs	2b			)
 
 7:		pop	{r5 - r8}
@@ -163,7 +189,8 @@ ENTRY(memmove)
 		strbcs	r4, [r0, #-1]!
 		strbcs	ip, [r0, #-1]
 
-#if defined (__ARM_ARCH_4T__) && defined (__THUMB_INTERWORK__)
+#if ((defined (__ARM_ARCH_4T__) && defined (__THUMB_INTERWORK__)) \
+     || defined (ARM_ALWAYS_BX))
 		pop	{r0, r4, lr}
 		cfi_adjust_cfa_offset (-12)
 		cfi_restore (r4)