2 * Tiny Code Generator for QEMU
4 * Copyright (c) 2009 Ulrich Hecht <uli@suse.de>
5 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
6 * Copyright (c) 2010 Richard Henderson <rth@twiddle.net>
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 /* We only support generating code for 64-bit mode. */
28 #if TCG_TARGET_REG_BITS != 64
29 #error "unsupported code generation mode"
32 #include "../tcg-pool.c.inc"
35 /* ??? The translation blocks produced by TCG are generally small enough to
36 be entirely reachable with a 16-bit displacement. Leaving the option for
37 a 32-bit displacement here Just In Case. */
38 #define USE_LONG_BRANCHES 0
40 #define TCG_CT_CONST_S16 0x100
41 #define TCG_CT_CONST_S32 0x200
42 #define TCG_CT_CONST_S33 0x400
43 #define TCG_CT_CONST_ZERO 0x800
45 /* Several places within the instruction set 0 means "no register"
46 rather than TCG_REG_R0. */
47 #define TCG_REG_NONE 0
49 /* A scratch register that may be be used throughout the backend. */
50 #define TCG_TMP0 TCG_REG_R1
52 /* A scratch register that holds a pointer to the beginning of the TB.
53 We don't need this when we have pc-relative loads with the general
54 instructions extension facility. */
55 #define TCG_REG_TB TCG_REG_R12
56 #define USE_REG_TB (!(s390_facilities & FACILITY_GEN_INST_EXT))
58 #ifndef CONFIG_SOFTMMU
59 #define TCG_GUEST_BASE_REG TCG_REG_R13
62 /* All of the following instructions are prefixed with their instruction
63 format, and are defined as 8- or 16-bit quantities, even when the two
64 halves of the 16-bit quantity may appear 32 bits apart in the insn.
65 This makes it easy to copy the values from the tables in Appendix B. */
66 typedef enum S390Opcode {
256 #ifdef CONFIG_DEBUG_TCG
257 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
258 "%r0", "%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7",
259 "%r8", "%r9", "%r10" "%r11" "%r12" "%r13" "%r14" "%r15"
263 /* Since R6 is a potential argument register, choose it last of the
264 call-saved registers. Likewise prefer the call-clobbered registers
265 in reverse order to maximize the chance of avoiding the arguments. */
266 static const int tcg_target_reg_alloc_order[] = {
267 /* Call saved registers. */
276 /* Call clobbered registers. */
280 /* Argument registers, in reverse order of allocation. */
287 static const int tcg_target_call_iarg_regs[] = {
295 static const int tcg_target_call_oarg_regs[] = {
303 #define S390_CC_NE (S390_CC_LT | S390_CC_GT)
304 #define S390_CC_LE (S390_CC_LT | S390_CC_EQ)
305 #define S390_CC_GE (S390_CC_GT | S390_CC_EQ)
306 #define S390_CC_NEVER 0
307 #define S390_CC_ALWAYS 15
309 /* Condition codes that result from a COMPARE and COMPARE LOGICAL. */
310 static const uint8_t tcg_cond_to_s390_cond[] = {
311 [TCG_COND_EQ] = S390_CC_EQ,
312 [TCG_COND_NE] = S390_CC_NE,
313 [TCG_COND_LT] = S390_CC_LT,
314 [TCG_COND_LE] = S390_CC_LE,
315 [TCG_COND_GT] = S390_CC_GT,
316 [TCG_COND_GE] = S390_CC_GE,
317 [TCG_COND_LTU] = S390_CC_LT,
318 [TCG_COND_LEU] = S390_CC_LE,
319 [TCG_COND_GTU] = S390_CC_GT,
320 [TCG_COND_GEU] = S390_CC_GE,
323 /* Condition codes that result from a LOAD AND TEST. Here, we have no
324 unsigned instruction variation, however since the test is vs zero we
325 can re-map the outcomes appropriately. */
326 static const uint8_t tcg_cond_to_ltr_cond[] = {
327 [TCG_COND_EQ] = S390_CC_EQ,
328 [TCG_COND_NE] = S390_CC_NE,
329 [TCG_COND_LT] = S390_CC_LT,
330 [TCG_COND_LE] = S390_CC_LE,
331 [TCG_COND_GT] = S390_CC_GT,
332 [TCG_COND_GE] = S390_CC_GE,
333 [TCG_COND_LTU] = S390_CC_NEVER,
334 [TCG_COND_LEU] = S390_CC_EQ,
335 [TCG_COND_GTU] = S390_CC_NE,
336 [TCG_COND_GEU] = S390_CC_ALWAYS,
339 #ifdef CONFIG_SOFTMMU
340 static void * const qemu_ld_helpers[16] = {
341 [MO_UB] = helper_ret_ldub_mmu,
342 [MO_SB] = helper_ret_ldsb_mmu,
343 [MO_LEUW] = helper_le_lduw_mmu,
344 [MO_LESW] = helper_le_ldsw_mmu,
345 [MO_LEUL] = helper_le_ldul_mmu,
346 [MO_LESL] = helper_le_ldsl_mmu,
347 [MO_LEQ] = helper_le_ldq_mmu,
348 [MO_BEUW] = helper_be_lduw_mmu,
349 [MO_BESW] = helper_be_ldsw_mmu,
350 [MO_BEUL] = helper_be_ldul_mmu,
351 [MO_BESL] = helper_be_ldsl_mmu,
352 [MO_BEQ] = helper_be_ldq_mmu,
355 static void * const qemu_st_helpers[16] = {
356 [MO_UB] = helper_ret_stb_mmu,
357 [MO_LEUW] = helper_le_stw_mmu,
358 [MO_LEUL] = helper_le_stl_mmu,
359 [MO_LEQ] = helper_le_stq_mmu,
360 [MO_BEUW] = helper_be_stw_mmu,
361 [MO_BEUL] = helper_be_stl_mmu,
362 [MO_BEQ] = helper_be_stq_mmu,
366 static tcg_insn_unit *tb_ret_addr;
367 uint64_t s390_facilities;
369 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
370 intptr_t value, intptr_t addend)
376 pcrel2 = (tcg_insn_unit *)value - code_ptr;
380 if (pcrel2 == (int16_t)pcrel2) {
381 tcg_patch16(code_ptr, pcrel2);
386 if (pcrel2 == (int32_t)pcrel2) {
387 tcg_patch32(code_ptr, pcrel2);
392 if (value == sextract64(value, 0, 20)) {
393 old = *(uint32_t *)code_ptr & 0xf00000ff;
394 old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
395 tcg_patch32(code_ptr, old);
400 g_assert_not_reached();
405 /* parse target specific constraints */
406 static const char *target_parse_constraint(TCGArgConstraint *ct,
407 const char *ct_str, TCGType type)
410 case 'r': /* all registers */
413 case 'L': /* qemu_ld/st constraint */
415 tcg_regset_reset_reg(ct->regs, TCG_REG_R2);
416 tcg_regset_reset_reg(ct->regs, TCG_REG_R3);
417 tcg_regset_reset_reg(ct->regs, TCG_REG_R4);
419 case 'a': /* force R2 for division */
421 tcg_regset_set_reg(ct->regs, TCG_REG_R2);
423 case 'b': /* force R3 for division */
425 tcg_regset_set_reg(ct->regs, TCG_REG_R3);
428 ct->ct |= TCG_CT_CONST_S33;
431 ct->ct |= TCG_CT_CONST_S16;
434 ct->ct |= TCG_CT_CONST_S32;
437 ct->ct |= TCG_CT_CONST_ZERO;
445 /* Test if a constant matches the constraint. */
446 static int tcg_target_const_match(tcg_target_long val, TCGType type,
447 const TCGArgConstraint *arg_ct)
451 if (ct & TCG_CT_CONST) {
455 if (type == TCG_TYPE_I32) {
459 /* The following are mutually exclusive. */
460 if (ct & TCG_CT_CONST_S16) {
461 return val == (int16_t)val;
462 } else if (ct & TCG_CT_CONST_S32) {
463 return val == (int32_t)val;
464 } else if (ct & TCG_CT_CONST_S33) {
465 return val >= -0xffffffffll && val <= 0xffffffffll;
466 } else if (ct & TCG_CT_CONST_ZERO) {
473 /* Emit instructions according to the given instruction format. */
475 static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
477 tcg_out16(s, (op << 8) | (r1 << 4) | r2);
480 static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
481 TCGReg r1, TCGReg r2)
483 tcg_out32(s, (op << 16) | (r1 << 4) | r2);
486 static void tcg_out_insn_RRF(TCGContext *s, S390Opcode op,
487 TCGReg r1, TCGReg r2, int m3)
489 tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
492 static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
494 tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
497 static void tcg_out_insn_RIE(TCGContext *s, S390Opcode op, TCGReg r1,
500 tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
501 tcg_out32(s, (i2 << 16) | (op & 0xff));
504 static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
506 tcg_out16(s, op | (r1 << 4));
510 static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
511 TCGReg b2, TCGReg r3, int disp)
513 tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
517 static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
518 TCGReg b2, TCGReg r3, int disp)
520 tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
521 tcg_out32(s, (op & 0xff) | (b2 << 28)
522 | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
525 #define tcg_out_insn_RX tcg_out_insn_RS
526 #define tcg_out_insn_RXY tcg_out_insn_RSY
528 /* Emit an opcode with "type-checking" of the format. */
529 #define tcg_out_insn(S, FMT, OP, ...) \
530 glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
533 /* emit 64-bit shifts */
534 static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
535 TCGReg src, TCGReg sh_reg, int sh_imm)
537 tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
540 /* emit 32-bit shifts */
541 static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
542 TCGReg sh_reg, int sh_imm)
544 tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
547 static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
550 if (type == TCG_TYPE_I32) {
551 tcg_out_insn(s, RR, LR, dst, src);
553 tcg_out_insn(s, RRE, LGR, dst, src);
559 static const S390Opcode lli_insns[4] = {
560 RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
563 static bool maybe_out_small_movi(TCGContext *s, TCGType type,
564 TCGReg ret, tcg_target_long sval)
566 tcg_target_ulong uval = sval;
569 if (type == TCG_TYPE_I32) {
570 uval = (uint32_t)sval;
571 sval = (int32_t)sval;
574 /* Try all 32-bit insns that can load it in one go. */
575 if (sval >= -0x8000 && sval < 0x8000) {
576 tcg_out_insn(s, RI, LGHI, ret, sval);
580 for (i = 0; i < 4; i++) {
581 tcg_target_long mask = 0xffffull << i*16;
582 if ((uval & mask) == uval) {
583 tcg_out_insn_RI(s, lli_insns[i], ret, uval >> i*16);
591 /* load a register with an immediate value */
592 static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
593 tcg_target_long sval, bool in_prologue)
595 tcg_target_ulong uval;
597 /* Try all 32-bit insns that can load it in one go. */
598 if (maybe_out_small_movi(s, type, ret, sval)) {
603 if (type == TCG_TYPE_I32) {
604 uval = (uint32_t)sval;
605 sval = (int32_t)sval;
608 /* Try all 48-bit insns that can load it in one go. */
609 if (s390_facilities & FACILITY_EXT_IMM) {
610 if (sval == (int32_t)sval) {
611 tcg_out_insn(s, RIL, LGFI, ret, sval);
614 if (uval <= 0xffffffff) {
615 tcg_out_insn(s, RIL, LLILF, ret, uval);
618 if ((uval & 0xffffffff) == 0) {
619 tcg_out_insn(s, RIL, LLIHF, ret, uval >> 32);
624 /* Try for PC-relative address load. For odd addresses,
625 attempt to use an offset from the start of the TB. */
626 if ((sval & 1) == 0) {
627 ptrdiff_t off = tcg_pcrel_diff(s, (void *)sval) >> 1;
628 if (off == (int32_t)off) {
629 tcg_out_insn(s, RIL, LARL, ret, off);
632 } else if (USE_REG_TB && !in_prologue) {
633 ptrdiff_t off = sval - (uintptr_t)s->code_gen_ptr;
634 if (off == sextract64(off, 0, 20)) {
635 /* This is certain to be an address within TB, and therefore
636 OFF will be negative; don't try RX_LA. */
637 tcg_out_insn(s, RXY, LAY, ret, TCG_REG_TB, TCG_REG_NONE, off);
642 /* A 32-bit unsigned value can be loaded in 2 insns. And given
643 that LLILL, LLIHL, LLILF above did not succeed, we know that
644 both insns are required. */
645 if (uval <= 0xffffffff) {
646 tcg_out_insn(s, RI, LLILL, ret, uval);
647 tcg_out_insn(s, RI, IILH, ret, uval >> 16);
651 /* Otherwise, stuff it in the constant pool. */
652 if (s390_facilities & FACILITY_GEN_INST_EXT) {
653 tcg_out_insn(s, RIL, LGRL, ret, 0);
654 new_pool_label(s, sval, R_390_PC32DBL, s->code_ptr - 2, 2);
655 } else if (USE_REG_TB && !in_prologue) {
656 tcg_out_insn(s, RXY, LG, ret, TCG_REG_TB, TCG_REG_NONE, 0);
657 new_pool_label(s, sval, R_390_20, s->code_ptr - 2,
658 -(intptr_t)s->code_gen_ptr);
660 TCGReg base = ret ? ret : TCG_TMP0;
661 tcg_out_insn(s, RIL, LARL, base, 0);
662 new_pool_label(s, sval, R_390_PC32DBL, s->code_ptr - 2, 2);
663 tcg_out_insn(s, RXY, LG, ret, base, TCG_REG_NONE, 0);
667 static void tcg_out_movi(TCGContext *s, TCGType type,
668 TCGReg ret, tcg_target_long sval)
670 tcg_out_movi_int(s, type, ret, sval, false);
673 /* Emit a load/store type instruction. Inputs are:
674 DATA: The register to be loaded or stored.
675 BASE+OFS: The effective address.
676 OPC_RX: If the operation has an RX format opcode (e.g. STC), otherwise 0.
677 OPC_RXY: The RXY format opcode for the operation (e.g. STCY). */
679 static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
680 TCGReg data, TCGReg base, TCGReg index,
683 if (ofs < -0x80000 || ofs >= 0x80000) {
684 /* Combine the low 20 bits of the offset with the actual load insn;
685 the high 44 bits must come from an immediate load. */
686 tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
687 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
690 /* If we were already given an index register, add it in. */
691 if (index != TCG_REG_NONE) {
692 tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
697 if (opc_rx && ofs >= 0 && ofs < 0x1000) {
698 tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
700 tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
705 /* load data without address translation or endianness conversion */
706 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
707 TCGReg base, intptr_t ofs)
709 if (type == TCG_TYPE_I32) {
710 tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
712 tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
716 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
717 TCGReg base, intptr_t ofs)
719 if (type == TCG_TYPE_I32) {
720 tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
722 tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
726 static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
727 TCGReg base, intptr_t ofs)
732 /* load data from an absolute host address */
733 static void tcg_out_ld_abs(TCGContext *s, TCGType type, TCGReg dest, void *abs)
735 intptr_t addr = (intptr_t)abs;
737 if ((s390_facilities & FACILITY_GEN_INST_EXT) && !(addr & 1)) {
738 ptrdiff_t disp = tcg_pcrel_diff(s, abs) >> 1;
739 if (disp == (int32_t)disp) {
740 if (type == TCG_TYPE_I32) {
741 tcg_out_insn(s, RIL, LRL, dest, disp);
743 tcg_out_insn(s, RIL, LGRL, dest, disp);
749 ptrdiff_t disp = abs - (void *)s->code_gen_ptr;
750 if (disp == sextract64(disp, 0, 20)) {
751 tcg_out_ld(s, type, dest, TCG_REG_TB, disp);
756 tcg_out_movi(s, TCG_TYPE_PTR, dest, addr & ~0xffff);
757 tcg_out_ld(s, type, dest, dest, addr & 0xffff);
760 static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
761 int msb, int lsb, int ofs, int z)
764 tcg_out16(s, (RIE_RISBG & 0xff00) | (dest << 4) | src);
765 tcg_out16(s, (msb << 8) | (z << 7) | lsb);
766 tcg_out16(s, (ofs << 8) | (RIE_RISBG & 0xff));
769 static void tgen_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
771 if (s390_facilities & FACILITY_EXT_IMM) {
772 tcg_out_insn(s, RRE, LGBR, dest, src);
776 if (type == TCG_TYPE_I32) {
778 tcg_out_sh32(s, RS_SLL, dest, TCG_REG_NONE, 24);
780 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 24);
782 tcg_out_sh32(s, RS_SRA, dest, TCG_REG_NONE, 24);
784 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 56);
785 tcg_out_sh64(s, RSY_SRAG, dest, dest, TCG_REG_NONE, 56);
789 static void tgen_ext8u(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
791 if (s390_facilities & FACILITY_EXT_IMM) {
792 tcg_out_insn(s, RRE, LLGCR, dest, src);
797 tcg_out_movi(s, type, TCG_TMP0, 0xff);
800 tcg_out_movi(s, type, dest, 0xff);
802 if (type == TCG_TYPE_I32) {
803 tcg_out_insn(s, RR, NR, dest, src);
805 tcg_out_insn(s, RRE, NGR, dest, src);
809 static void tgen_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
811 if (s390_facilities & FACILITY_EXT_IMM) {
812 tcg_out_insn(s, RRE, LGHR, dest, src);
816 if (type == TCG_TYPE_I32) {
818 tcg_out_sh32(s, RS_SLL, dest, TCG_REG_NONE, 16);
820 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 16);
822 tcg_out_sh32(s, RS_SRA, dest, TCG_REG_NONE, 16);
824 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 48);
825 tcg_out_sh64(s, RSY_SRAG, dest, dest, TCG_REG_NONE, 48);
829 static void tgen_ext16u(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
831 if (s390_facilities & FACILITY_EXT_IMM) {
832 tcg_out_insn(s, RRE, LLGHR, dest, src);
837 tcg_out_movi(s, type, TCG_TMP0, 0xffff);
840 tcg_out_movi(s, type, dest, 0xffff);
842 if (type == TCG_TYPE_I32) {
843 tcg_out_insn(s, RR, NR, dest, src);
845 tcg_out_insn(s, RRE, NGR, dest, src);
849 static inline void tgen_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
851 tcg_out_insn(s, RRE, LGFR, dest, src);
854 static inline void tgen_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
856 tcg_out_insn(s, RRE, LLGFR, dest, src);
859 /* Accept bit patterns like these:
864 Copied from gcc sources. */
865 static inline bool risbg_mask(uint64_t c)
868 /* We don't change the number of transitions by inverting,
869 so make sure we start with the LSB zero. */
873 /* Reject all zeros or all ones. */
877 /* Find the first transition. */
879 /* Invert to look for a second transition. */
881 /* Erase the first transition. */
883 /* Find the second transition, if any. */
885 /* Match if all the bits are 1's, or if c is zero. */
889 static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
892 if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
893 /* Achieve wraparound by swapping msb and lsb. */
894 msb = 64 - ctz64(~val);
895 lsb = clz64(~val) - 1;
898 lsb = 63 - ctz64(val);
900 tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
903 static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
905 static const S390Opcode ni_insns[4] = {
906 RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
908 static const S390Opcode nif_insns[2] = {
911 uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
914 /* Look for the zero-extensions. */
915 if ((val & valid) == 0xffffffff) {
916 tgen_ext32u(s, dest, dest);
919 if (s390_facilities & FACILITY_EXT_IMM) {
920 if ((val & valid) == 0xff) {
921 tgen_ext8u(s, TCG_TYPE_I64, dest, dest);
924 if ((val & valid) == 0xffff) {
925 tgen_ext16u(s, TCG_TYPE_I64, dest, dest);
930 /* Try all 32-bit insns that can perform it in one go. */
931 for (i = 0; i < 4; i++) {
932 tcg_target_ulong mask = ~(0xffffull << i*16);
933 if (((val | ~valid) & mask) == mask) {
934 tcg_out_insn_RI(s, ni_insns[i], dest, val >> i*16);
939 /* Try all 48-bit insns that can perform it in one go. */
940 if (s390_facilities & FACILITY_EXT_IMM) {
941 for (i = 0; i < 2; i++) {
942 tcg_target_ulong mask = ~(0xffffffffull << i*32);
943 if (((val | ~valid) & mask) == mask) {
944 tcg_out_insn_RIL(s, nif_insns[i], dest, val >> i*32);
949 if ((s390_facilities & FACILITY_GEN_INST_EXT) && risbg_mask(val)) {
950 tgen_andi_risbg(s, dest, dest, val);
954 /* Use the constant pool if USE_REG_TB, but not for small constants. */
956 if (!maybe_out_small_movi(s, type, TCG_TMP0, val)) {
957 tcg_out_insn(s, RXY, NG, dest, TCG_REG_TB, TCG_REG_NONE, 0);
958 new_pool_label(s, val & valid, R_390_20, s->code_ptr - 2,
959 -(intptr_t)s->code_gen_ptr);
963 tcg_out_movi(s, type, TCG_TMP0, val);
965 if (type == TCG_TYPE_I32) {
966 tcg_out_insn(s, RR, NR, dest, TCG_TMP0);
968 tcg_out_insn(s, RRE, NGR, dest, TCG_TMP0);
972 static void tgen_ori(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
974 static const S390Opcode oi_insns[4] = {
975 RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
977 static const S390Opcode oif_insns[2] = {
983 /* Look for no-op. */
984 if (unlikely(val == 0)) {
988 /* Try all 32-bit insns that can perform it in one go. */
989 for (i = 0; i < 4; i++) {
990 tcg_target_ulong mask = (0xffffull << i*16);
991 if ((val & mask) != 0 && (val & ~mask) == 0) {
992 tcg_out_insn_RI(s, oi_insns[i], dest, val >> i*16);
997 /* Try all 48-bit insns that can perform it in one go. */
998 if (s390_facilities & FACILITY_EXT_IMM) {
999 for (i = 0; i < 2; i++) {
1000 tcg_target_ulong mask = (0xffffffffull << i*32);
1001 if ((val & mask) != 0 && (val & ~mask) == 0) {
1002 tcg_out_insn_RIL(s, oif_insns[i], dest, val >> i*32);
1008 /* Use the constant pool if USE_REG_TB, but not for small constants. */
1009 if (maybe_out_small_movi(s, type, TCG_TMP0, val)) {
1010 if (type == TCG_TYPE_I32) {
1011 tcg_out_insn(s, RR, OR, dest, TCG_TMP0);
1013 tcg_out_insn(s, RRE, OGR, dest, TCG_TMP0);
1015 } else if (USE_REG_TB) {
1016 tcg_out_insn(s, RXY, OG, dest, TCG_REG_TB, TCG_REG_NONE, 0);
1017 new_pool_label(s, val, R_390_20, s->code_ptr - 2,
1018 -(intptr_t)s->code_gen_ptr);
1020 /* Perform the OR via sequential modifications to the high and
1021 low parts. Do this via recursion to handle 16-bit vs 32-bit
1022 masks in each half. */
1023 tcg_debug_assert(s390_facilities & FACILITY_EXT_IMM);
1024 tgen_ori(s, type, dest, val & 0x00000000ffffffffull);
1025 tgen_ori(s, type, dest, val & 0xffffffff00000000ull);
1029 static void tgen_xori(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1031 /* Try all 48-bit insns that can perform it in one go. */
1032 if (s390_facilities & FACILITY_EXT_IMM) {
1033 if ((val & 0xffffffff00000000ull) == 0) {
1034 tcg_out_insn(s, RIL, XILF, dest, val);
1037 if ((val & 0x00000000ffffffffull) == 0) {
1038 tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1043 /* Use the constant pool if USE_REG_TB, but not for small constants. */
1044 if (maybe_out_small_movi(s, type, TCG_TMP0, val)) {
1045 if (type == TCG_TYPE_I32) {
1046 tcg_out_insn(s, RR, XR, dest, TCG_TMP0);
1048 tcg_out_insn(s, RRE, XGR, dest, TCG_TMP0);
1050 } else if (USE_REG_TB) {
1051 tcg_out_insn(s, RXY, XG, dest, TCG_REG_TB, TCG_REG_NONE, 0);
1052 new_pool_label(s, val, R_390_20, s->code_ptr - 2,
1053 -(intptr_t)s->code_gen_ptr);
1055 /* Perform the xor by parts. */
1056 tcg_debug_assert(s390_facilities & FACILITY_EXT_IMM);
1057 if (val & 0xffffffff) {
1058 tcg_out_insn(s, RIL, XILF, dest, val);
1060 if (val > 0xffffffff) {
1061 tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1066 static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1067 TCGArg c2, bool c2const, bool need_carry)
1069 bool is_unsigned = is_unsigned_cond(c);
1074 if (!(is_unsigned && need_carry)) {
1075 if (type == TCG_TYPE_I32) {
1076 tcg_out_insn(s, RR, LTR, r1, r1);
1078 tcg_out_insn(s, RRE, LTGR, r1, r1);
1080 return tcg_cond_to_ltr_cond[c];
1084 if (!is_unsigned && c2 == (int16_t)c2) {
1085 op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1086 tcg_out_insn_RI(s, op, r1, c2);
1090 if (s390_facilities & FACILITY_EXT_IMM) {
1091 if (type == TCG_TYPE_I32) {
1092 op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1093 tcg_out_insn_RIL(s, op, r1, c2);
1095 } else if (c2 == (is_unsigned ? (uint32_t)c2 : (int32_t)c2)) {
1096 op = (is_unsigned ? RIL_CLGFI : RIL_CGFI);
1097 tcg_out_insn_RIL(s, op, r1, c2);
1102 /* Use the constant pool, but not for small constants. */
1103 if (maybe_out_small_movi(s, type, TCG_TMP0, c2)) {
1105 /* fall through to reg-reg */
1106 } else if (USE_REG_TB) {
1107 if (type == TCG_TYPE_I32) {
1108 op = (is_unsigned ? RXY_CLY : RXY_CY);
1109 tcg_out_insn_RXY(s, op, r1, TCG_REG_TB, TCG_REG_NONE, 0);
1110 new_pool_label(s, (uint32_t)c2, R_390_20, s->code_ptr - 2,
1111 4 - (intptr_t)s->code_gen_ptr);
1113 op = (is_unsigned ? RXY_CLG : RXY_CG);
1114 tcg_out_insn_RXY(s, op, r1, TCG_REG_TB, TCG_REG_NONE, 0);
1115 new_pool_label(s, c2, R_390_20, s->code_ptr - 2,
1116 -(intptr_t)s->code_gen_ptr);
1120 if (type == TCG_TYPE_I32) {
1121 op = (is_unsigned ? RIL_CLRL : RIL_CRL);
1122 tcg_out_insn_RIL(s, op, r1, 0);
1123 new_pool_label(s, (uint32_t)c2, R_390_PC32DBL,
1124 s->code_ptr - 2, 2 + 4);
1126 op = (is_unsigned ? RIL_CLGRL : RIL_CGRL);
1127 tcg_out_insn_RIL(s, op, r1, 0);
1128 new_pool_label(s, c2, R_390_PC32DBL, s->code_ptr - 2, 2);
1134 if (type == TCG_TYPE_I32) {
1135 op = (is_unsigned ? RR_CLR : RR_CR);
1136 tcg_out_insn_RR(s, op, r1, c2);
1138 op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1139 tcg_out_insn_RRE(s, op, r1, c2);
1143 return tcg_cond_to_s390_cond[c];
1146 static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1147 TCGReg dest, TCGReg c1, TCGArg c2, int c2const)
1152 /* With LOC2, we can always emit the minimum 3 insns. */
1153 if (s390_facilities & FACILITY_LOAD_ON_COND2) {
1154 /* Emit: d = 0, d = (cc ? 1 : d). */
1155 cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1156 tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1157 tcg_out_insn(s, RIE, LOCGHI, dest, 1, cc);
1161 have_loc = (s390_facilities & FACILITY_LOAD_ON_COND) != 0;
1163 /* For HAVE_LOC, only the paths through GTU/GT/LEU/LE are smaller. */
1167 /* X != 0 is X > 0. */
1168 if (c2const && c2 == 0) {
1169 cond = TCG_COND_GTU;
1177 /* The result of a compare has CC=2 for GT and CC=3 unused.
1178 ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit. */
1179 tgen_cmp(s, type, cond, c1, c2, c2const, true);
1180 tcg_out_movi(s, type, dest, 0);
1181 tcg_out_insn(s, RRE, ALCGR, dest, dest);
1185 /* X == 0 is X <= 0. */
1186 if (c2const && c2 == 0) {
1187 cond = TCG_COND_LEU;
1195 /* As above, but we're looking for borrow, or !carry.
1196 The second insn computes d - d - borrow, or -1 for true
1197 and 0 for false. So we must mask to 1 bit afterward. */
1198 tgen_cmp(s, type, cond, c1, c2, c2const, true);
1199 tcg_out_insn(s, RRE, SLBGR, dest, dest);
1200 tgen_andi(s, type, dest, 1);
1207 /* Swap operands so that we can use LEU/GTU/GT/LE. */
1212 tcg_out_movi(s, type, TCG_TMP0, c2);
1221 cond = tcg_swap_cond(cond);
1225 g_assert_not_reached();
1228 cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1230 /* Emit: d = 0, t = 1, d = (cc ? t : d). */
1231 tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1232 tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 1);
1233 tcg_out_insn(s, RRF, LOCGR, dest, TCG_TMP0, cc);
1235 /* Emit: d = 1; if (cc) goto over; d = 0; over: */
1236 tcg_out_movi(s, type, dest, 1);
1237 tcg_out_insn(s, RI, BRC, cc, (4 + 4) >> 1);
1238 tcg_out_movi(s, type, dest, 0);
1242 static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1243 TCGReg c1, TCGArg c2, int c2const,
1244 TCGArg v3, int v3const)
1247 if (s390_facilities & FACILITY_LOAD_ON_COND) {
1248 cc = tgen_cmp(s, type, c, c1, c2, c2const, false);
1250 tcg_out_insn(s, RIE, LOCGHI, dest, v3, cc);
1252 tcg_out_insn(s, RRF, LOCGR, dest, v3, cc);
1255 c = tcg_invert_cond(c);
1256 cc = tgen_cmp(s, type, c, c1, c2, c2const, false);
1258 /* Emit: if (cc) goto over; dest = r3; over: */
1259 tcg_out_insn(s, RI, BRC, cc, (4 + 4) >> 1);
1260 tcg_out_insn(s, RRE, LGR, dest, v3);
1264 static void tgen_clz(TCGContext *s, TCGReg dest, TCGReg a1,
1265 TCGArg a2, int a2const)
1267 /* Since this sets both R and R+1, we have no choice but to store the
1268 result into R0, allowing R1 == TCG_TMP0 to be clobbered as well. */
1269 QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
1270 tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
1272 if (a2const && a2 == 64) {
1273 tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
1276 tcg_out_movi(s, TCG_TYPE_I64, dest, a2);
1278 tcg_out_mov(s, TCG_TYPE_I64, dest, a2);
1280 if (s390_facilities & FACILITY_LOAD_ON_COND) {
1281 /* Emit: if (one bit found) dest = r0. */
1282 tcg_out_insn(s, RRF, LOCGR, dest, TCG_REG_R0, 2);
1284 /* Emit: if (no one bit found) goto over; dest = r0; over: */
1285 tcg_out_insn(s, RI, BRC, 8, (4 + 4) >> 1);
1286 tcg_out_insn(s, RRE, LGR, dest, TCG_REG_R0);
1291 static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1292 int ofs, int len, int z)
1294 int lsb = (63 - ofs);
1295 int msb = lsb - (len - 1);
1296 tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1299 static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1302 tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1305 static void tgen_gotoi(TCGContext *s, int cc, tcg_insn_unit *dest)
1307 ptrdiff_t off = dest - s->code_ptr;
1308 if (off == (int16_t)off) {
1309 tcg_out_insn(s, RI, BRC, cc, off);
1310 } else if (off == (int32_t)off) {
1311 tcg_out_insn(s, RIL, BRCL, cc, off);
1313 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1314 tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1318 static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1321 tgen_gotoi(s, cc, l->u.value_ptr);
1322 } else if (USE_LONG_BRANCHES) {
1323 tcg_out16(s, RIL_BRCL | (cc << 4));
1324 tcg_out_reloc(s, s->code_ptr, R_390_PC32DBL, l, 2);
1327 tcg_out16(s, RI_BRC | (cc << 4));
1328 tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1333 static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1334 TCGReg r1, TCGReg r2, TCGLabel *l)
1339 off = l->u.value_ptr - s->code_ptr;
1340 tcg_debug_assert(off == (int16_t)off);
1342 tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1345 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1347 tcg_out16(s, cc << 12 | (opc & 0xff));
1350 static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1351 TCGReg r1, int i2, TCGLabel *l)
1353 tcg_target_long off = 0;
1356 off = l->u.value_ptr - s->code_ptr;
1357 tcg_debug_assert(off == (int16_t)off);
1359 tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1362 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1364 tcg_out16(s, (i2 << 8) | (opc & 0xff));
1367 static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1368 TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1372 if (s390_facilities & FACILITY_GEN_INST_EXT) {
1373 bool is_unsigned = is_unsigned_cond(c);
1377 cc = tcg_cond_to_s390_cond[c];
1380 opc = (type == TCG_TYPE_I32
1381 ? (is_unsigned ? RIE_CLRJ : RIE_CRJ)
1382 : (is_unsigned ? RIE_CLGRJ : RIE_CGRJ));
1383 tgen_compare_branch(s, opc, cc, r1, c2, l);
1387 /* COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1388 If the immediate we've been given does not fit that range, we'll
1389 fall back to separate compare and branch instructions using the
1390 larger comparison range afforded by COMPARE IMMEDIATE. */
1391 if (type == TCG_TYPE_I32) {
1394 in_range = (uint32_t)c2 == (uint8_t)c2;
1397 in_range = (int32_t)c2 == (int8_t)c2;
1402 in_range = (uint64_t)c2 == (uint8_t)c2;
1405 in_range = (int64_t)c2 == (int8_t)c2;
1409 tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1414 cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1415 tgen_branch(s, cc, l);
1418 static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest)
1420 ptrdiff_t off = dest - s->code_ptr;
1421 if (off == (int32_t)off) {
1422 tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1424 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1425 tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1429 static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1430 TCGReg base, TCGReg index, int disp)
1432 switch (opc & (MO_SSIZE | MO_BSWAP)) {
1434 tcg_out_insn(s, RXY, LLGC, data, base, index, disp);
1437 tcg_out_insn(s, RXY, LGB, data, base, index, disp);
1440 case MO_UW | MO_BSWAP:
1441 /* swapped unsigned halfword load with upper bits zeroed */
1442 tcg_out_insn(s, RXY, LRVH, data, base, index, disp);
1443 tgen_ext16u(s, TCG_TYPE_I64, data, data);
1446 tcg_out_insn(s, RXY, LLGH, data, base, index, disp);
1449 case MO_SW | MO_BSWAP:
1450 /* swapped sign-extended halfword load */
1451 tcg_out_insn(s, RXY, LRVH, data, base, index, disp);
1452 tgen_ext16s(s, TCG_TYPE_I64, data, data);
1455 tcg_out_insn(s, RXY, LGH, data, base, index, disp);
1458 case MO_UL | MO_BSWAP:
1459 /* swapped unsigned int load with upper bits zeroed */
1460 tcg_out_insn(s, RXY, LRV, data, base, index, disp);
1461 tgen_ext32u(s, data, data);
1464 tcg_out_insn(s, RXY, LLGF, data, base, index, disp);
1467 case MO_SL | MO_BSWAP:
1468 /* swapped sign-extended int load */
1469 tcg_out_insn(s, RXY, LRV, data, base, index, disp);
1470 tgen_ext32s(s, data, data);
1473 tcg_out_insn(s, RXY, LGF, data, base, index, disp);
1476 case MO_Q | MO_BSWAP:
1477 tcg_out_insn(s, RXY, LRVG, data, base, index, disp);
1480 tcg_out_insn(s, RXY, LG, data, base, index, disp);
1488 static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1489 TCGReg base, TCGReg index, int disp)
1491 switch (opc & (MO_SIZE | MO_BSWAP)) {
1493 if (disp >= 0 && disp < 0x1000) {
1494 tcg_out_insn(s, RX, STC, data, base, index, disp);
1496 tcg_out_insn(s, RXY, STCY, data, base, index, disp);
1500 case MO_UW | MO_BSWAP:
1501 tcg_out_insn(s, RXY, STRVH, data, base, index, disp);
1504 if (disp >= 0 && disp < 0x1000) {
1505 tcg_out_insn(s, RX, STH, data, base, index, disp);
1507 tcg_out_insn(s, RXY, STHY, data, base, index, disp);
1511 case MO_UL | MO_BSWAP:
1512 tcg_out_insn(s, RXY, STRV, data, base, index, disp);
1515 if (disp >= 0 && disp < 0x1000) {
1516 tcg_out_insn(s, RX, ST, data, base, index, disp);
1518 tcg_out_insn(s, RXY, STY, data, base, index, disp);
1522 case MO_Q | MO_BSWAP:
1523 tcg_out_insn(s, RXY, STRVG, data, base, index, disp);
1526 tcg_out_insn(s, RXY, STG, data, base, index, disp);
1534 #if defined(CONFIG_SOFTMMU)
1535 #include "../tcg-ldst.c.inc"
1537 /* We're expecting to use a 20-bit negative offset on the tlb memory ops. */
1538 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1539 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 19));
1541 /* Load and compare a TLB entry, leaving the flags set. Loads the TLB
1542 addend into R2. Returns a register with the santitized guest address. */
1543 static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, MemOp opc,
1544 int mem_index, bool is_ld)
1546 unsigned s_bits = opc & MO_SIZE;
1547 unsigned a_bits = get_alignment_bits(opc);
1548 unsigned s_mask = (1 << s_bits) - 1;
1549 unsigned a_mask = (1 << a_bits) - 1;
1550 int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1551 int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1552 int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1556 tcg_out_sh64(s, RSY_SRLG, TCG_REG_R2, addr_reg, TCG_REG_NONE,
1557 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1558 tcg_out_insn(s, RXY, NG, TCG_REG_R2, TCG_AREG0, TCG_REG_NONE, mask_off);
1559 tcg_out_insn(s, RXY, AG, TCG_REG_R2, TCG_AREG0, TCG_REG_NONE, table_off);
1561 /* For aligned accesses, we check the first byte and include the alignment
1562 bits within the address. For unaligned access, we check that we don't
1563 cross pages using the address of the last byte of the access. */
1564 a_off = (a_bits >= s_bits ? 0 : s_mask - a_mask);
1565 tlb_mask = (uint64_t)TARGET_PAGE_MASK | a_mask;
1566 if ((s390_facilities & FACILITY_GEN_INST_EXT) && a_off == 0) {
1567 tgen_andi_risbg(s, TCG_REG_R3, addr_reg, tlb_mask);
1569 tcg_out_insn(s, RX, LA, TCG_REG_R3, addr_reg, TCG_REG_NONE, a_off);
1570 tgen_andi(s, TCG_TYPE_TL, TCG_REG_R3, tlb_mask);
1574 ofs = offsetof(CPUTLBEntry, addr_read);
1576 ofs = offsetof(CPUTLBEntry, addr_write);
1578 if (TARGET_LONG_BITS == 32) {
1579 tcg_out_insn(s, RX, C, TCG_REG_R3, TCG_REG_R2, TCG_REG_NONE, ofs);
1581 tcg_out_insn(s, RXY, CG, TCG_REG_R3, TCG_REG_R2, TCG_REG_NONE, ofs);
1584 tcg_out_insn(s, RXY, LG, TCG_REG_R2, TCG_REG_R2, TCG_REG_NONE,
1585 offsetof(CPUTLBEntry, addend));
1587 if (TARGET_LONG_BITS == 32) {
1588 tgen_ext32u(s, TCG_REG_R3, addr_reg);
1594 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1595 TCGReg data, TCGReg addr,
1596 tcg_insn_unit *raddr, tcg_insn_unit *label_ptr)
1598 TCGLabelQemuLdst *label = new_ldst_label(s);
1600 label->is_ld = is_ld;
1602 label->datalo_reg = data;
1603 label->addrlo_reg = addr;
1604 label->raddr = raddr;
1605 label->label_ptr[0] = label_ptr;
1608 static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1610 TCGReg addr_reg = lb->addrlo_reg;
1611 TCGReg data_reg = lb->datalo_reg;
1612 TCGMemOpIdx oi = lb->oi;
1613 MemOp opc = get_memop(oi);
1615 if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1616 (intptr_t)s->code_ptr, 2)) {
1620 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_AREG0);
1621 if (TARGET_LONG_BITS == 64) {
1622 tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_R3, addr_reg);
1624 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R4, oi);
1625 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R5, (uintptr_t)lb->raddr);
1626 tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SSIZE)]);
1627 tcg_out_mov(s, TCG_TYPE_I64, data_reg, TCG_REG_R2);
1629 tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1633 static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1635 TCGReg addr_reg = lb->addrlo_reg;
1636 TCGReg data_reg = lb->datalo_reg;
1637 TCGMemOpIdx oi = lb->oi;
1638 MemOp opc = get_memop(oi);
1640 if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1641 (intptr_t)s->code_ptr, 2)) {
1645 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_AREG0);
1646 if (TARGET_LONG_BITS == 64) {
1647 tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_R3, addr_reg);
1649 switch (opc & MO_SIZE) {
1651 tgen_ext8u(s, TCG_TYPE_I64, TCG_REG_R4, data_reg);
1654 tgen_ext16u(s, TCG_TYPE_I64, TCG_REG_R4, data_reg);
1657 tgen_ext32u(s, TCG_REG_R4, data_reg);
1660 tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_R4, data_reg);
1665 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R5, oi);
1666 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R6, (uintptr_t)lb->raddr);
1667 tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1669 tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1673 static void tcg_prepare_user_ldst(TCGContext *s, TCGReg *addr_reg,
1674 TCGReg *index_reg, tcg_target_long *disp)
1676 if (TARGET_LONG_BITS == 32) {
1677 tgen_ext32u(s, TCG_TMP0, *addr_reg);
1678 *addr_reg = TCG_TMP0;
1680 if (guest_base < 0x80000) {
1681 *index_reg = TCG_REG_NONE;
1684 *index_reg = TCG_GUEST_BASE_REG;
1688 #endif /* CONFIG_SOFTMMU */
1690 static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1693 MemOp opc = get_memop(oi);
1694 #ifdef CONFIG_SOFTMMU
1695 unsigned mem_index = get_mmuidx(oi);
1696 tcg_insn_unit *label_ptr;
1699 base_reg = tcg_out_tlb_read(s, addr_reg, opc, mem_index, 1);
1701 tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1702 label_ptr = s->code_ptr;
1705 tcg_out_qemu_ld_direct(s, opc, data_reg, base_reg, TCG_REG_R2, 0);
1707 add_qemu_ldst_label(s, 1, oi, data_reg, addr_reg, s->code_ptr, label_ptr);
1710 tcg_target_long disp;
1712 tcg_prepare_user_ldst(s, &addr_reg, &index_reg, &disp);
1713 tcg_out_qemu_ld_direct(s, opc, data_reg, addr_reg, index_reg, disp);
1717 static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1720 MemOp opc = get_memop(oi);
1721 #ifdef CONFIG_SOFTMMU
1722 unsigned mem_index = get_mmuidx(oi);
1723 tcg_insn_unit *label_ptr;
1726 base_reg = tcg_out_tlb_read(s, addr_reg, opc, mem_index, 0);
1728 tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1729 label_ptr = s->code_ptr;
1732 tcg_out_qemu_st_direct(s, opc, data_reg, base_reg, TCG_REG_R2, 0);
1734 add_qemu_ldst_label(s, 0, oi, data_reg, addr_reg, s->code_ptr, label_ptr);
1737 tcg_target_long disp;
1739 tcg_prepare_user_ldst(s, &addr_reg, &index_reg, &disp);
1740 tcg_out_qemu_st_direct(s, opc, data_reg, addr_reg, index_reg, disp);
1744 # define OP_32_64(x) \
1745 case glue(glue(INDEX_op_,x),_i32): \
1746 case glue(glue(INDEX_op_,x),_i64)
1748 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1749 const TCGArg *args, const int *const_args)
1755 case INDEX_op_exit_tb:
1756 /* Reuse the zeroing that exists for goto_ptr. */
1759 tgen_gotoi(s, S390_CC_ALWAYS, s->code_gen_epilogue);
1761 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
1762 tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
1766 case INDEX_op_goto_tb:
1768 if (s->tb_jmp_insn_offset) {
1769 /* branch displacement must be aligned for atomic patching;
1770 * see if we need to add extra nop before branch
1772 if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
1775 tcg_debug_assert(!USE_REG_TB);
1776 tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
1777 s->tb_jmp_insn_offset[a0] = tcg_current_code_size(s);
1780 /* load address stored at s->tb_jmp_target_addr + a0 */
1781 tcg_out_ld_abs(s, TCG_TYPE_PTR, TCG_REG_TB,
1782 s->tb_jmp_target_addr + a0);
1784 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_TB);
1786 set_jmp_reset_offset(s, a0);
1788 /* For the unlinked path of goto_tb, we need to reset
1789 TCG_REG_TB to the beginning of this TB. */
1791 int ofs = -tcg_current_code_size(s);
1792 assert(ofs == (int16_t)ofs);
1793 tcg_out_insn(s, RI, AGHI, TCG_REG_TB, ofs);
1797 case INDEX_op_goto_ptr:
1800 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, a0);
1802 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
1806 /* ??? LLC (RXY format) is only present with the extended-immediate
1807 facility, whereas LLGC is always present. */
1808 tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
1812 /* ??? LB is no smaller than LGB, so no point to using it. */
1813 tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
1817 /* ??? LLH (RXY format) is only present with the extended-immediate
1818 facility, whereas LLGH is always present. */
1819 tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
1822 case INDEX_op_ld16s_i32:
1823 tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
1826 case INDEX_op_ld_i32:
1827 tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1831 tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
1832 TCG_REG_NONE, args[2]);
1836 tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
1837 TCG_REG_NONE, args[2]);
1840 case INDEX_op_st_i32:
1841 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1844 case INDEX_op_add_i32:
1845 a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1846 if (const_args[2]) {
1849 if (a2 == (int16_t)a2) {
1850 tcg_out_insn(s, RI, AHI, a0, a2);
1853 if (s390_facilities & FACILITY_EXT_IMM) {
1854 tcg_out_insn(s, RIL, AFI, a0, a2);
1858 tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
1859 } else if (a0 == a1) {
1860 tcg_out_insn(s, RR, AR, a0, a2);
1862 tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
1865 case INDEX_op_sub_i32:
1866 a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1867 if (const_args[2]) {
1870 } else if (a0 == a1) {
1871 tcg_out_insn(s, RR, SR, a0, a2);
1873 tcg_out_insn(s, RRF, SRK, a0, a1, a2);
1877 case INDEX_op_and_i32:
1878 a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
1879 if (const_args[2]) {
1880 tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
1881 tgen_andi(s, TCG_TYPE_I32, a0, a2);
1882 } else if (a0 == a1) {
1883 tcg_out_insn(s, RR, NR, a0, a2);
1885 tcg_out_insn(s, RRF, NRK, a0, a1, a2);
1888 case INDEX_op_or_i32:
1889 a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
1890 if (const_args[2]) {
1891 tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
1892 tgen_ori(s, TCG_TYPE_I32, a0, a2);
1893 } else if (a0 == a1) {
1894 tcg_out_insn(s, RR, OR, a0, a2);
1896 tcg_out_insn(s, RRF, ORK, a0, a1, a2);
1899 case INDEX_op_xor_i32:
1900 a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
1901 if (const_args[2]) {
1902 tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
1903 tgen_xori(s, TCG_TYPE_I32, a0, a2);
1904 } else if (a0 == a1) {
1905 tcg_out_insn(s, RR, XR, args[0], args[2]);
1907 tcg_out_insn(s, RRF, XRK, a0, a1, a2);
1911 case INDEX_op_neg_i32:
1912 tcg_out_insn(s, RR, LCR, args[0], args[1]);
1915 case INDEX_op_mul_i32:
1916 if (const_args[2]) {
1917 if ((int32_t)args[2] == (int16_t)args[2]) {
1918 tcg_out_insn(s, RI, MHI, args[0], args[2]);
1920 tcg_out_insn(s, RIL, MSFI, args[0], args[2]);
1923 tcg_out_insn(s, RRE, MSR, args[0], args[2]);
1927 case INDEX_op_div2_i32:
1928 tcg_out_insn(s, RR, DR, TCG_REG_R2, args[4]);
1930 case INDEX_op_divu2_i32:
1931 tcg_out_insn(s, RRE, DLR, TCG_REG_R2, args[4]);
1934 case INDEX_op_shl_i32:
1938 a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1940 if (const_args[2]) {
1941 tcg_out_sh32(s, op, a0, TCG_REG_NONE, a2);
1943 tcg_out_sh32(s, op, a0, a2, 0);
1946 /* Using tcg_out_sh64 here for the format; it is a 32-bit shift. */
1947 if (const_args[2]) {
1948 tcg_out_sh64(s, op2, a0, a1, TCG_REG_NONE, a2);
1950 tcg_out_sh64(s, op2, a0, a1, a2, 0);
1954 case INDEX_op_shr_i32:
1958 case INDEX_op_sar_i32:
1963 case INDEX_op_rotl_i32:
1964 /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol. */
1965 if (const_args[2]) {
1966 tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
1968 tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
1971 case INDEX_op_rotr_i32:
1972 if (const_args[2]) {
1973 tcg_out_sh64(s, RSY_RLL, args[0], args[1],
1974 TCG_REG_NONE, (32 - args[2]) & 31);
1976 tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
1977 tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
1981 case INDEX_op_ext8s_i32:
1982 tgen_ext8s(s, TCG_TYPE_I32, args[0], args[1]);
1984 case INDEX_op_ext16s_i32:
1985 tgen_ext16s(s, TCG_TYPE_I32, args[0], args[1]);
1987 case INDEX_op_ext8u_i32:
1988 tgen_ext8u(s, TCG_TYPE_I32, args[0], args[1]);
1990 case INDEX_op_ext16u_i32:
1991 tgen_ext16u(s, TCG_TYPE_I32, args[0], args[1]);
1995 /* The TCG bswap definition requires bits 0-47 already be zero.
1996 Thus we don't need the G-type insns to implement bswap16_i64. */
1997 tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
1998 tcg_out_sh32(s, RS_SRL, args[0], TCG_REG_NONE, 16);
2001 tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2004 case INDEX_op_add2_i32:
2005 if (const_args[4]) {
2006 tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2008 tcg_out_insn(s, RR, ALR, args[0], args[4]);
2010 tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2012 case INDEX_op_sub2_i32:
2013 if (const_args[4]) {
2014 tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2016 tcg_out_insn(s, RR, SLR, args[0], args[4]);
2018 tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2022 tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2025 case INDEX_op_brcond_i32:
2026 tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2027 args[1], const_args[1], arg_label(args[3]));
2029 case INDEX_op_setcond_i32:
2030 tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2031 args[2], const_args[2]);
2033 case INDEX_op_movcond_i32:
2034 tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2035 args[2], const_args[2], args[3], const_args[3]);
2038 case INDEX_op_qemu_ld_i32:
2039 /* ??? Technically we can use a non-extending instruction. */
2040 case INDEX_op_qemu_ld_i64:
2041 tcg_out_qemu_ld(s, args[0], args[1], args[2]);
2043 case INDEX_op_qemu_st_i32:
2044 case INDEX_op_qemu_st_i64:
2045 tcg_out_qemu_st(s, args[0], args[1], args[2]);
2048 case INDEX_op_ld16s_i64:
2049 tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2051 case INDEX_op_ld32u_i64:
2052 tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2054 case INDEX_op_ld32s_i64:
2055 tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2057 case INDEX_op_ld_i64:
2058 tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2061 case INDEX_op_st32_i64:
2062 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2064 case INDEX_op_st_i64:
2065 tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2068 case INDEX_op_add_i64:
2069 a0 = args[0], a1 = args[1], a2 = args[2];
2070 if (const_args[2]) {
2073 if (a2 == (int16_t)a2) {
2074 tcg_out_insn(s, RI, AGHI, a0, a2);
2077 if (s390_facilities & FACILITY_EXT_IMM) {
2078 if (a2 == (int32_t)a2) {
2079 tcg_out_insn(s, RIL, AGFI, a0, a2);
2081 } else if (a2 == (uint32_t)a2) {
2082 tcg_out_insn(s, RIL, ALGFI, a0, a2);
2084 } else if (-a2 == (uint32_t)-a2) {
2085 tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2090 tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2091 } else if (a0 == a1) {
2092 tcg_out_insn(s, RRE, AGR, a0, a2);
2094 tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2097 case INDEX_op_sub_i64:
2098 a0 = args[0], a1 = args[1], a2 = args[2];
2099 if (const_args[2]) {
2102 } else if (a0 == a1) {
2103 tcg_out_insn(s, RRE, SGR, a0, a2);
2105 tcg_out_insn(s, RRF, SGRK, a0, a1, a2);
2109 case INDEX_op_and_i64:
2110 a0 = args[0], a1 = args[1], a2 = args[2];
2111 if (const_args[2]) {
2112 tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2113 tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
2114 } else if (a0 == a1) {
2115 tcg_out_insn(s, RRE, NGR, args[0], args[2]);
2117 tcg_out_insn(s, RRF, NGRK, a0, a1, a2);
2120 case INDEX_op_or_i64:
2121 a0 = args[0], a1 = args[1], a2 = args[2];
2122 if (const_args[2]) {
2123 tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2124 tgen_ori(s, TCG_TYPE_I64, a0, a2);
2125 } else if (a0 == a1) {
2126 tcg_out_insn(s, RRE, OGR, a0, a2);
2128 tcg_out_insn(s, RRF, OGRK, a0, a1, a2);
2131 case INDEX_op_xor_i64:
2132 a0 = args[0], a1 = args[1], a2 = args[2];
2133 if (const_args[2]) {
2134 tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2135 tgen_xori(s, TCG_TYPE_I64, a0, a2);
2136 } else if (a0 == a1) {
2137 tcg_out_insn(s, RRE, XGR, a0, a2);
2139 tcg_out_insn(s, RRF, XGRK, a0, a1, a2);
2143 case INDEX_op_neg_i64:
2144 tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2146 case INDEX_op_bswap64_i64:
2147 tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2150 case INDEX_op_mul_i64:
2151 if (const_args[2]) {
2152 if (args[2] == (int16_t)args[2]) {
2153 tcg_out_insn(s, RI, MGHI, args[0], args[2]);
2155 tcg_out_insn(s, RIL, MSGFI, args[0], args[2]);
2158 tcg_out_insn(s, RRE, MSGR, args[0], args[2]);
2162 case INDEX_op_div2_i64:
2163 /* ??? We get an unnecessary sign-extension of the dividend
2164 into R3 with this definition, but as we do in fact always
2165 produce both quotient and remainder using INDEX_op_div_i64
2166 instead requires jumping through even more hoops. */
2167 tcg_out_insn(s, RRE, DSGR, TCG_REG_R2, args[4]);
2169 case INDEX_op_divu2_i64:
2170 tcg_out_insn(s, RRE, DLGR, TCG_REG_R2, args[4]);
2172 case INDEX_op_mulu2_i64:
2173 tcg_out_insn(s, RRE, MLGR, TCG_REG_R2, args[3]);
2176 case INDEX_op_shl_i64:
2179 if (const_args[2]) {
2180 tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2182 tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2185 case INDEX_op_shr_i64:
2188 case INDEX_op_sar_i64:
2192 case INDEX_op_rotl_i64:
2193 if (const_args[2]) {
2194 tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2195 TCG_REG_NONE, args[2]);
2197 tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2200 case INDEX_op_rotr_i64:
2201 if (const_args[2]) {
2202 tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2203 TCG_REG_NONE, (64 - args[2]) & 63);
2205 /* We can use the smaller 32-bit negate because only the
2206 low 6 bits are examined for the rotate. */
2207 tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2208 tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2212 case INDEX_op_ext8s_i64:
2213 tgen_ext8s(s, TCG_TYPE_I64, args[0], args[1]);
2215 case INDEX_op_ext16s_i64:
2216 tgen_ext16s(s, TCG_TYPE_I64, args[0], args[1]);
2218 case INDEX_op_ext_i32_i64:
2219 case INDEX_op_ext32s_i64:
2220 tgen_ext32s(s, args[0], args[1]);
2222 case INDEX_op_ext8u_i64:
2223 tgen_ext8u(s, TCG_TYPE_I64, args[0], args[1]);
2225 case INDEX_op_ext16u_i64:
2226 tgen_ext16u(s, TCG_TYPE_I64, args[0], args[1]);
2228 case INDEX_op_extu_i32_i64:
2229 case INDEX_op_ext32u_i64:
2230 tgen_ext32u(s, args[0], args[1]);
2233 case INDEX_op_add2_i64:
2234 if (const_args[4]) {
2235 if ((int64_t)args[4] >= 0) {
2236 tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2238 tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2241 tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2243 tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2245 case INDEX_op_sub2_i64:
2246 if (const_args[4]) {
2247 if ((int64_t)args[4] >= 0) {
2248 tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2250 tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2253 tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2255 tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2258 case INDEX_op_brcond_i64:
2259 tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2260 args[1], const_args[1], arg_label(args[3]));
2262 case INDEX_op_setcond_i64:
2263 tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2264 args[2], const_args[2]);
2266 case INDEX_op_movcond_i64:
2267 tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2268 args[2], const_args[2], args[3], const_args[3]);
2272 a0 = args[0], a1 = args[1], a2 = args[2];
2273 if (const_args[1]) {
2274 tgen_deposit(s, a0, a2, args[3], args[4], 1);
2276 /* Since we can't support "0Z" as a constraint, we allow a1 in
2277 any register. Fix things up as if a matching constraint. */
2279 TCGType type = (opc == INDEX_op_deposit_i64);
2281 tcg_out_mov(s, type, TCG_TMP0, a2);
2284 tcg_out_mov(s, type, a0, a1);
2286 tgen_deposit(s, a0, a2, args[3], args[4], 0);
2291 tgen_extract(s, args[0], args[1], args[2], args[3]);
2294 case INDEX_op_clz_i64:
2295 tgen_clz(s, args[0], args[1], args[2], const_args[2]);
2299 /* The host memory model is quite strong, we simply need to
2300 serialize the instruction stream. */
2301 if (args[0] & TCG_MO_ST_LD) {
2302 tcg_out_insn(s, RR, BCR,
2303 s390_facilities & FACILITY_FAST_BCR_SER ? 14 : 15, 0);
2307 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
2308 case INDEX_op_mov_i64:
2309 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
2310 case INDEX_op_movi_i64:
2311 case INDEX_op_call: /* Always emitted via tcg_out_call. */
2317 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
2319 static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
2320 static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
2321 static const TCGTargetOpDef r_L = { .args_ct_str = { "r", "L" } };
2322 static const TCGTargetOpDef L_L = { .args_ct_str = { "L", "L" } };
2323 static const TCGTargetOpDef r_ri = { .args_ct_str = { "r", "ri" } };
2324 static const TCGTargetOpDef r_r_ri = { .args_ct_str = { "r", "r", "ri" } };
2325 static const TCGTargetOpDef r_0_ri = { .args_ct_str = { "r", "0", "ri" } };
2326 static const TCGTargetOpDef r_0_rI = { .args_ct_str = { "r", "0", "rI" } };
2327 static const TCGTargetOpDef r_0_rJ = { .args_ct_str = { "r", "0", "rJ" } };
2328 static const TCGTargetOpDef a2_r
2329 = { .args_ct_str = { "r", "r", "0", "1", "r", "r" } };
2330 static const TCGTargetOpDef a2_ri
2331 = { .args_ct_str = { "r", "r", "0", "1", "ri", "r" } };
2332 static const TCGTargetOpDef a2_rA
2333 = { .args_ct_str = { "r", "r", "0", "1", "rA", "r" } };
2336 case INDEX_op_goto_ptr:
2339 case INDEX_op_ld8u_i32:
2340 case INDEX_op_ld8u_i64:
2341 case INDEX_op_ld8s_i32:
2342 case INDEX_op_ld8s_i64:
2343 case INDEX_op_ld16u_i32:
2344 case INDEX_op_ld16u_i64:
2345 case INDEX_op_ld16s_i32:
2346 case INDEX_op_ld16s_i64:
2347 case INDEX_op_ld_i32:
2348 case INDEX_op_ld32u_i64:
2349 case INDEX_op_ld32s_i64:
2350 case INDEX_op_ld_i64:
2351 case INDEX_op_st8_i32:
2352 case INDEX_op_st8_i64:
2353 case INDEX_op_st16_i32:
2354 case INDEX_op_st16_i64:
2355 case INDEX_op_st_i32:
2356 case INDEX_op_st32_i64:
2357 case INDEX_op_st_i64:
2360 case INDEX_op_add_i32:
2361 case INDEX_op_add_i64:
2363 case INDEX_op_sub_i32:
2364 case INDEX_op_sub_i64:
2365 case INDEX_op_and_i32:
2366 case INDEX_op_and_i64:
2367 case INDEX_op_or_i32:
2368 case INDEX_op_or_i64:
2369 case INDEX_op_xor_i32:
2370 case INDEX_op_xor_i64:
2371 return (s390_facilities & FACILITY_DISTINCT_OPS ? &r_r_ri : &r_0_ri);
2373 case INDEX_op_mul_i32:
2374 /* If we have the general-instruction-extensions, then we have
2375 MULTIPLY SINGLE IMMEDIATE with a signed 32-bit, otherwise we
2376 have only MULTIPLY HALFWORD IMMEDIATE, with a signed 16-bit. */
2377 return (s390_facilities & FACILITY_GEN_INST_EXT ? &r_0_ri : &r_0_rI);
2378 case INDEX_op_mul_i64:
2379 return (s390_facilities & FACILITY_GEN_INST_EXT ? &r_0_rJ : &r_0_rI);
2381 case INDEX_op_shl_i32:
2382 case INDEX_op_shr_i32:
2383 case INDEX_op_sar_i32:
2384 return (s390_facilities & FACILITY_DISTINCT_OPS ? &r_r_ri : &r_0_ri);
2386 case INDEX_op_shl_i64:
2387 case INDEX_op_shr_i64:
2388 case INDEX_op_sar_i64:
2391 case INDEX_op_rotl_i32:
2392 case INDEX_op_rotl_i64:
2393 case INDEX_op_rotr_i32:
2394 case INDEX_op_rotr_i64:
2397 case INDEX_op_brcond_i32:
2398 case INDEX_op_brcond_i64:
2401 case INDEX_op_bswap16_i32:
2402 case INDEX_op_bswap16_i64:
2403 case INDEX_op_bswap32_i32:
2404 case INDEX_op_bswap32_i64:
2405 case INDEX_op_bswap64_i64:
2406 case INDEX_op_neg_i32:
2407 case INDEX_op_neg_i64:
2408 case INDEX_op_ext8s_i32:
2409 case INDEX_op_ext8s_i64:
2410 case INDEX_op_ext8u_i32:
2411 case INDEX_op_ext8u_i64:
2412 case INDEX_op_ext16s_i32:
2413 case INDEX_op_ext16s_i64:
2414 case INDEX_op_ext16u_i32:
2415 case INDEX_op_ext16u_i64:
2416 case INDEX_op_ext32s_i64:
2417 case INDEX_op_ext32u_i64:
2418 case INDEX_op_ext_i32_i64:
2419 case INDEX_op_extu_i32_i64:
2420 case INDEX_op_extract_i32:
2421 case INDEX_op_extract_i64:
2424 case INDEX_op_clz_i64:
2425 case INDEX_op_setcond_i32:
2426 case INDEX_op_setcond_i64:
2429 case INDEX_op_qemu_ld_i32:
2430 case INDEX_op_qemu_ld_i64:
2432 case INDEX_op_qemu_st_i64:
2433 case INDEX_op_qemu_st_i32:
2436 case INDEX_op_deposit_i32:
2437 case INDEX_op_deposit_i64:
2439 static const TCGTargetOpDef dep
2440 = { .args_ct_str = { "r", "rZ", "r" } };
2443 case INDEX_op_movcond_i32:
2444 case INDEX_op_movcond_i64:
2446 static const TCGTargetOpDef movc
2447 = { .args_ct_str = { "r", "r", "ri", "r", "0" } };
2448 static const TCGTargetOpDef movc_l
2449 = { .args_ct_str = { "r", "r", "ri", "rI", "0" } };
2450 return (s390_facilities & FACILITY_LOAD_ON_COND2 ? &movc_l : &movc);
2452 case INDEX_op_div2_i32:
2453 case INDEX_op_div2_i64:
2454 case INDEX_op_divu2_i32:
2455 case INDEX_op_divu2_i64:
2457 static const TCGTargetOpDef div2
2458 = { .args_ct_str = { "b", "a", "0", "1", "r" } };
2461 case INDEX_op_mulu2_i64:
2463 static const TCGTargetOpDef mul2
2464 = { .args_ct_str = { "b", "a", "0", "r" } };
2468 case INDEX_op_add2_i32:
2469 case INDEX_op_sub2_i32:
2470 return (s390_facilities & FACILITY_EXT_IMM ? &a2_ri : &a2_r);
2471 case INDEX_op_add2_i64:
2472 case INDEX_op_sub2_i64:
2473 return (s390_facilities & FACILITY_EXT_IMM ? &a2_rA : &a2_r);
2481 static void query_s390_facilities(void)
2483 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2485 /* Is STORE FACILITY LIST EXTENDED available? Honestly, I believe this
2486 is present on all 64-bit systems, but let's check for it anyway. */
2487 if (hwcap & HWCAP_S390_STFLE) {
2488 register int r0 __asm__("0");
2489 register void *r1 __asm__("1");
2492 r1 = &s390_facilities;
2493 asm volatile(".word 0xb2b0,0x1000"
2494 : "=r"(r0) : "0"(0), "r"(r1) : "memory", "cc");
2498 static void tcg_target_init(TCGContext *s)
2500 query_s390_facilities();
2502 tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
2503 tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
2505 tcg_target_call_clobber_regs = 0;
2506 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2507 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2508 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2509 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2510 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
2511 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
2512 /* The r6 register is technically call-saved, but it's also a parameter
2513 register, so it can get killed by setup for the qemu_st helper. */
2514 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
2515 /* The return register can be considered call-clobbered. */
2516 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2518 s->reserved_regs = 0;
2519 tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
2520 /* XXX many insns can't be used with R0, so we better avoid it for now */
2521 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
2522 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2524 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
2528 #define FRAME_SIZE ((int)(TCG_TARGET_CALL_STACK_OFFSET \
2529 + TCG_STATIC_CALL_ARGS_SIZE \
2530 + CPU_TEMP_BUF_NLONGS * sizeof(long)))
2532 static void tcg_target_qemu_prologue(TCGContext *s)
2534 /* stmg %r6,%r15,48(%r15) (save registers) */
2535 tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
2537 /* aghi %r15,-frame_size */
2538 tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
2540 tcg_set_frame(s, TCG_REG_CALL_STACK,
2541 TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
2542 CPU_TEMP_BUF_NLONGS * sizeof(long));
2544 #ifndef CONFIG_SOFTMMU
2545 if (guest_base >= 0x80000) {
2546 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2547 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2551 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2553 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB,
2554 tcg_target_call_iarg_regs[1]);
2557 /* br %r3 (go to TB) */
2558 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
2561 * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
2562 * and fall through to the rest of the epilogue.
2564 s->code_gen_epilogue = s->code_ptr;
2565 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
2568 tb_ret_addr = s->code_ptr;
2570 /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
2571 tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
2574 /* br %r14 (return) */
2575 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
2578 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2580 memset(p, 0x07, count * sizeof(tcg_insn_unit));
2585 uint8_t fde_def_cfa[4];
2586 uint8_t fde_reg_ofs[18];
2589 /* We're expecting a 2 byte uleb128 encoded value. */
2590 QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2592 #define ELF_HOST_MACHINE EM_S390
2594 static const DebugFrame debug_frame = {
2595 .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2598 .h.cie.code_align = 1,
2599 .h.cie.data_align = 8, /* sleb128 8 */
2600 .h.cie.return_column = TCG_REG_R14,
2602 /* Total FDE size does not include the "len" member. */
2603 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
2606 12, TCG_REG_CALL_STACK, /* DW_CFA_def_cfa %r15, ... */
2607 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2611 0x86, 6, /* DW_CFA_offset, %r6, 48 */
2612 0x87, 7, /* DW_CFA_offset, %r7, 56 */
2613 0x88, 8, /* DW_CFA_offset, %r8, 64 */
2614 0x89, 9, /* DW_CFA_offset, %r92, 72 */
2615 0x8a, 10, /* DW_CFA_offset, %r10, 80 */
2616 0x8b, 11, /* DW_CFA_offset, %r11, 88 */
2617 0x8c, 12, /* DW_CFA_offset, %r12, 96 */
2618 0x8d, 13, /* DW_CFA_offset, %r13, 104 */
2619 0x8e, 14, /* DW_CFA_offset, %r14, 112 */
2623 void tcg_register_jit(void *buf, size_t buf_size)
2625 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));