ARM: Correct several inconsistencies with ldrd and strd

* lib/jit_disasm.c: Change thumb or arm disassemble based on
	jit code before disassembly.

	* lib/jit_arm-cpu.c: Correct reversed arguments to LDRD and
	STRD instructions, and correct checking for support of those.

	* lib/jit_arm-swf.c: Correct wrong use of LDRD and STRD and
	only use those if the register is even.

	* check/check.arm.swf.sh, check/check.arm4.swf.sh: New files
	to test LDRD and STRD, as well as the alternate code path
	when those are not available, in the .arm4. test case.

	* check/Makefile.am: Update for the new test cases.

ChangeLog

@@ -1,3 +1,20 @@
+2014-11-20 Paulo Andrade <pcpa@gnu.org>
+
+	* lib/jit_disasm.c: Change thumb or arm disassemble based on
+	jit code before disassembly.
+
+	* lib/jit_arm-cpu.c: Correct reversed arguments to LDRD and
+	STRD instructions, and correct checking for support of those.
+
+	* lib/jit_arm-swf.c: Correct wrong use of LDRD and STRD and
+	only use those if the register is even.
+
+	* check/check.arm.swf.sh, check/check.arm4.swf.sh: New files
+	to test LDRD and STRD, as well as the alternate code path
+	when those are not available, in the .arm4. test case.
+
+	* check/Makefile.am: Update for the new test cases.
+
 2014-11-08 Paulo Andrade <pcpa@gnu.org>
 
 	* include/lightning/jit_private.h, lib/jit_aarch64.c,

check/Makefile.am

@@ -92,6 +92,8 @@ EXTRA_DIST =				\
 	check.sh			\
 	check.x87.sh			\
 	check.arm.sh	check.swf.sh	\
+	check.arm.swf.sh		\
+	check.arm4.swf.sh		\
 	check.nodata.sh			\
 	check.x87.nodata.sh		\
 	run-test	all.tst
@@ -213,6 +215,53 @@ swf_TESTS =					\
 $(swf_TESTS):	check.swf.sh
 	$(LN_S) $(srcdir)/check.swf.sh $@
 TESTS += $(swf_TESTS)
+if test_arm_arm
+#arm_swf_TESTS = $(addsuffix .arm.swf, $(base_TESTS))
+arm_swf_TESTS =							\
+	3to2.arm.swf add.arm.swf allocai.arm.swf		\
+	bp.arm.swf divi.arm.swf fib.arm.swf rpn.arm.swf		\
+	ldstr.arm.swf ldsti.arm.swf				\
+	ldstxr.arm.swf ldstxi.arm.swf				\
+	ldstr-c.arm.swf ldstxr-c.arm.swf ldstxi-c.arm.swf	\
+	cvt.arm.swf branch.arm.swf				\
+	alu_add.arm.swf alux_add.arm.swf			\
+	alu_sub.arm.swf alux_sub.arm.swf alu_rsb.arm.swf	\
+	alu_mul.arm.swf alu_div.arm.swf alu_rem.arm.swf		\
+	alu_and.arm.swf alu_or.arm.swf alu_xor.arm.swf		\
+	alu_lsh.arm.swf alu_rsh.arm.swf				\
+	alu_com.arm.swf alu_neg.arm.swf				\
+	fop_abs.arm.swf fop_sqrt.arm.swf			\
+	varargs.arm.swf stack.arm.swf				\
+	clobber.arm.swf carry.arm.swf call.arm.swf		\
+	float.arm.swf jmpr.arm.swf tramp.arm.swf range.arm.swf
+$(arm_swf_TESTS):	check.arm.swf.sh
+	$(LN_S) $(srcdir)/check.arm.swf.sh $@
+TESTS += $(arm_swf_TESTS)
+endif
+if test_arm_arm
+#arm4_swf_TESTS = $(addsuffix .arm4.swf, $(base_TESTS))
+arm4_swf_TESTS =						\
+	3to2.arm4.swf add.arm4.swf allocai.arm4.swf		\
+	bp.arm4.swf divi.arm4.swf fib.arm4.swf rpn.arm4.swf	\
+	ldstr.arm4.swf ldsti.arm4.swf				\
+	ldstxr.arm4.swf ldstxi.arm4.swf				\
+	ldstr-c.arm4.swf ldstxr-c.arm4.swf ldstxi-c.arm4.swf	\
+	cvt.arm4.swf branch.arm4.swf				\
+	alu_add.arm4.swf alux_add.arm4.swf			\
+	alu_sub.arm4.swf alux_sub.arm4.swf alu_rsb.arm4.swf	\
+	alu_mul.arm4.swf alu_div.arm4.swf alu_rem.arm4.swf	\
+	alu_and.arm4.swf alu_or.arm4.swf alu_xor.arm4.swf	\
+	alu_lsh.arm4.swf alu_rsh.arm4.swf			\
+	alu_com.arm4.swf alu_neg.arm4.swf			\
+	fop_abs.arm4.swf fop_sqrt.arm4.swf			\
+	varargs.arm4.swf stack.arm4.swf				\
+	clobber.arm4.swf carry.arm4.swf call.arm4.swf		\
+	float.arm4.swf jmpr.arm4.swf tramp.arm4.swf		\
+	range.arm4.swf
+$(arm4_swf_TESTS):	check.arm4.swf.sh
+	$(LN_S) $(srcdir)/check.arm4.swf.sh $@
+TESTS += $(arm4_swf_TESTS)
+endif
 endif
 
 if test_nodata

check/check.arm.swf.sh

@@ -0,0 +1,15 @@
+#!/bin/sh
+test=`basename $0 | sed -e 's|\.arm\.swf$||'`
+./lightning -mthumb=0 -mvfp=0 $srcdir/$test.tst | tr -d \\r > $test.out
+if test $? != 0; then
+  exit $?
+fi
+
+cmp -s $srcdir/$test.ok $test.out
+result=$?
+if test $result != 0; then
+    diff $srcdir/$test.ok $test.out
+    rm $test.out
+    exit 1
+fi
+rm $test.out

check/check.arm4.swf.sh

@@ -0,0 +1,15 @@
+#!/bin/sh
+test=`basename $0 | sed -e 's|\.arm4\.swf$||'`
+./lightning -mcpu=4 -mthumb=0 -mvfp=0 $srcdir/$test.tst | tr -d \\r > $test.out
+if test $? != 0; then
+  exit $?
+fi
+
+cmp -s $srcdir/$test.ok $test.out
+result=$?
+if test $result != 0; then
+    diff $srcdir/$test.ok $test.out
+    rm $test.out
+    exit 1
+fi
+rm $test.out

lib/jit_arm-cpu.c

@@ -31,7 +31,7 @@
 #  define jit_thumb_p()			jit_cpu.thumb
 #  define jit_no_set_flags()		_jitc->no_set_flags
 #  define jit_armv5_p()			(jit_cpu.version >= 5)
-#  define jit_armv5e_p()		(jit_cpu.version >= 5 && jit_cpu.extend)
+#  define jit_armv5e_p()		(jit_cpu.version > 5 || (jit_cpu.version == 5 && jit_cpu.extend))
 #  define jit_armv6_p()			(jit_cpu.version >= 6)
 #  define jit_armv7r_p()		0
 #  define stack_framesize		48
@@ -709,9 +709,9 @@ static void _torl(jit_state_t*,int,int,int) maybe_unused;
 #  define LDRD(rt,rn,rm)		CC_LDRD(ARM_CC_AL,rt,rn,rm)
 #  define T2_LDRDI(rt,rt2,rn,im)	torrri8(THUMB2_LDRDI|ARM_P,rn,rt,rt2,im)
 #  define CC_LDRDN(cc,rt,rn,rm)		corrr(cc,ARM_LDRD,rn,rt,rm)
-#  define LDRDN(rd,rn,rm)		CC_LDRDN(ARM_CC_AL,rn,rt,rm)
+#  define LDRDN(rd,rn,rm)		CC_LDRDN(ARM_CC_AL,rt,rn,rm)
 #  define CC_LDRDI(cc,rt,rn,im)		corri8(cc,ARM_LDRDI|ARM_P,rn,rt,im)
-#  define LDRDI(rt,rn,im)		CC_LDRDI(ARM_CC_AL,rn,rt,im)
+#  define LDRDI(rt,rn,im)		CC_LDRDI(ARM_CC_AL,rt,rn,im)
 #  define CC_LDRDIN(cc,rt,rn,im)	corri8(cc,ARM_LDRDI,rn,rt,im)
 #  define LDRDIN(rt,rn,im)		CC_LDRDIN(ARM_CC_AL,rt,rn,im)
 #  define T2_LDRDIN(rt,rt2,rn,im)	torrri8(THUMB2_LDRDI,rn,rt,rt2,im)
@@ -758,7 +758,7 @@ static void _torl(jit_state_t*,int,int,int) maybe_unused;
 #  define CC_STRIN(cc,rt,rn,im)		corri(cc,ARM_STRI,rn,rt,im)
 #  define STRIN(rt,rn,im)		CC_STRIN(ARM_CC_AL,rt,rn,im)
 #  define T2_STRIN(rt,rn,im)		torri8(THUMB2_STRI,rn,rt,im)
-#  define CC_STRD(cc,rt,rn,rm)		corrr(cc,ARM_STRD|ARM_P,rt,rn,rm)
+#  define CC_STRD(cc,rt,rn,rm)		corrr(cc,ARM_STRD|ARM_P,rn,rt,rm)
 #  define STRD(rt,rn,rm)		CC_STRD(ARM_CC_AL,rt,rn,rm)
 #  define CC_STRDN(cc,rt,rn,rm)		corrr(cc,ARM_STRD,rn,rt,rm)
 #  define STRDN(rt,rn,rm)		CC_STRDN(ARM_CC_AL,rt,rn,rm)

lib/jit_arm-swf.c

@@ -1870,7 +1870,7 @@ _swf_movr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 	    }
 	}
 	else if (jit_fpr_p(r0)) {
-	    if (!jit_thumb_p() && jit_armv5e_p())
+	    if (!jit_thumb_p() && jit_armv5e_p() && !(r1 & 1))
 		STRDIN(r1, _FP_REGNO, swf_off(r0) + 8);
 	    else {
 		swf_strin(r1, _FP_REGNO, swf_off(r0) + 8);
@@ -1967,9 +1967,9 @@ _swf_absr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
     if (jit_fpr_p(r1)) {
 	if (jit_fpr_p(r0) && !jit_thumb_p() && jit_armv5e_p() &&
 	    r0 != r1 && (reg = jit_get_reg_pair()) != JIT_NOREG) {
-	    LDRDIN(rn(reg), _FP_REGNO, swf_off(r1) + 4);
+	    LDRDIN(rn(reg), _FP_REGNO, swf_off(r1) + 8);
-	    swf_bici(rn(reg), rn(reg), 0x80000000);
+	    swf_bici(rn(reg) + 1, rn(reg) + 1, 0x80000000);
-	    STRDIN(rn(reg), _FP_REGNO, swf_off(r0) + 4);
+	    STRDIN(rn(reg), _FP_REGNO, swf_off(r0) + 8);
 	    jit_unget_reg_pair(reg);
 	}
 	else {
@@ -2039,9 +2039,9 @@ _swf_negr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
     if (jit_fpr_p(r1)) {
 	if (jit_fpr_p(r0) && !jit_thumb_p() && jit_armv5e_p() &&
 	    r0 != r1 && (reg = jit_get_reg_pair()) != JIT_NOREG) {
-	    LDRDIN(rn(reg), _FP_REGNO, swf_off(r1) + 4);
+	    LDRDIN(rn(reg), _FP_REGNO, swf_off(r1) + 8);
-	    EORI(rn(reg), rn(reg), encode_arm_immediate(0x80000000));
+	    EORI(rn(reg) + 1, rn(reg) + 1, encode_arm_immediate(0x80000000));
-	    STRDIN(rn(reg), _FP_REGNO, swf_off(r0) + 4);
+	    STRDIN(rn(reg), _FP_REGNO, swf_off(r0) + 8);
 	    jit_unget_reg_pair(reg);
 	}
 	else {
@@ -2197,7 +2197,7 @@ _swf_ldr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 	    jit_unget_reg(reg);
 	}
     }
-    else if (!jit_thumb_p() && jit_armv5e_p())
+    else if (!jit_thumb_p() && jit_armv5e_p() && !(r0 & 1))
 	LDRDI(r0, r1, 0);
     else {
 	ldxi_i(r0, r1, 0);
@@ -2241,7 +2241,7 @@ _swf_ldi_d(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
 	    swf_strin(rn(rg0), _FP_REGNO, swf_off(r0) + 4);
 	    jit_unget_reg(rg0);
 	}
-	else if (!jit_thumb_p() && jit_armv5e_p())
+	else if (!jit_thumb_p() && jit_armv5e_p() && !(r0 & 1))
 	    LDRDI(r0, rn(rg1), 0);
 	else {
 	    ldxi_i(r0, rn(rg1), 0);
@@ -2289,7 +2289,7 @@ _swf_ldxr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2)
 	}
     }
     else {
-	if (!jit_thumb_p() && jit_armv5e_p())
+	if (!jit_thumb_p() && jit_armv5e_p() && !(r0 & 1))
 	    LDRD(r0, r1, r2);
 	else {
 	    rg1 = jit_get_reg(jit_class_gpr);
@@ -2360,9 +2360,11 @@ _swf_ldxi_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
 	}
     }
     else {
-	if (!jit_thumb_p() && jit_armv5e_p() && i0 >= 0 && i0 <= 255)
+	if (!jit_thumb_p() && jit_armv5e_p() &&
+	    i0 >= 0 && i0 <= 255 && !(r0 & 1))
 	    LDRDI(r0, r1, i0);
-	else if (!jit_thumb_p() && jit_armv5e_p() && i0 < 0 && i0 >= -255)
+	else if (!jit_thumb_p() && jit_armv5e_p() &&
+		 i0 < 0 && i0 >= -255 && !(r0 & 1))
 	    LDRDIN(r0, r1, -i0);
 	else if (i0 >= 0 && i0 + 4 <= 4095) {
 	    ldxi_i(r0, r1, i0);
@@ -2417,7 +2419,7 @@ _swf_str_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 	}
     }
     else {
-	if (!jit_thumb_p() && jit_armv5e_p())
+	if (!jit_thumb_p() && jit_armv5e_p() && !(r1 & 1))
 	    STRDI(r1, r0, 0);
 	else {
 	    stxi_i(0, r0, r1);
@@ -2469,7 +2471,7 @@ _swf_sti_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)
     else {
 	rg1 = jit_get_reg(jit_class_gpr);
 	movi(rn(rg1), i0);
-	if (!jit_thumb_p() && jit_armv5e_p())
+	if (!jit_thumb_p() && jit_armv5e_p() && !(r0 & 1))
 	    STRDI(r0, rn(rg1), 0);
 	else {
 	    stxi_i(0, rn(rg1), r0);
@@ -2517,7 +2519,7 @@ _swf_stxr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2)
 	}
     }
     else {
-	if (!jit_thumb_p() && jit_armv5e_p())
+	if (!jit_thumb_p() && jit_armv5e_p() && !(r2 & 1))
 	    STRD(r0, r1, r2);
 	else {
 	    rg1 = jit_get_reg(jit_class_gpr);
@@ -2556,6 +2558,7 @@ _swf_stxi_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
 		STRDI(rn(rg0), r0, i0);
 	    else
 		STRDIN(rn(rg0), r0, -i0);
+	    jit_unget_reg_pair(rg0);
 	}
 	else if (i0 >= 0 && i0 + 4 <= 4095) {
 	    rg0 = jit_get_reg(jit_class_gpr);
@@ -2587,9 +2590,11 @@ _swf_stxi_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
 	}
     }
     else {
-	if (!jit_thumb_p() && jit_armv5e_p() && i0 >= 0 && i0 <= 255)
+	if (!jit_thumb_p() && jit_armv5e_p() &&
+	    i0 >= 0 && i0 <= 255 && !(r1 & 1))
 	    STRDI(r1, r0, i0);
-	else if (!jit_thumb_p() && jit_armv5e_p() && i0 < 0 && i0 >= -255)
+	else if (!jit_thumb_p() && jit_armv5e_p() &&
+		 i0 < 0 && i0 >= -255 && !(r1 & 1))
 	    STRDIN(r1, r0, -i0);
 	else if (i0 >= 0 && i0 + 4 <= 4095) {
 	    stxi_i(i0, r0, r1);

lib/jit_disasm.c

@@ -84,12 +84,6 @@ jit_init_debug(const char *progname)
     disasm_info.mach = bfd_mach_i386_i386;
 #    endif
 #  endif
-#  if defined(__arm__)
-    /* FIXME add mapping for prolog switching to arm and possible jump
-     * before first prolog also in arm mode */
-    if (jit_cpu.thumb)
-	disasm_info.disassembler_options = "force-thumb";
-#  endif
 #  if defined(__powerpc__)
     disasm_info.arch = bfd_arch_powerpc;
     disasm_info.mach = bfd_mach_ppc64;
@@ -204,8 +198,15 @@ void
 _jit_disassemble(jit_state_t *_jit)
 {
 #if DISASSEMBLER
-    if (disasm_bfd)
+    if (disasm_bfd) {
+#  if defined(__arm__)
+	/* FIXME add mapping for prolog switching to arm and possible jump
+	 * before first prolog also in arm mode */
+	disasm_info.disassembler_options = jit_cpu.thumb ? "force-thumb" : "";
+#  endif
+
 	disassemble(_jit->code.ptr, _jit->pc.uc - _jit->code.ptr);
+    }
 #endif
 }
 
@@ -326,8 +327,7 @@ _disassemble(jit_state_t *_jit, jit_pointer_t code, jit_int32_t length)
     again:
 	if (data_info) {
 	    while (_jitc->data_info.ptr[data_offset].code < pc) {
-		data_offset += 2;
+		if (++data_offset >= _jitc->data_info.length) {
-		if (data_offset >= _jitc->data_info.length) {
 		    data_info = 0;
 		    goto again;
 		}