2 * Tiny Code Generator for QEMU
4 * Copyright (c) 2018 SiFive, Inc
5 * Copyright (c) 2008-2009 Arnaud Patard <arnaud.patard@rtp-net.org>
6 * Copyright (c) 2009 Aurelien Jarno <aurelien@aurel32.net>
7 * Copyright (c) 2008 Fabrice Bellard
9 * Based on i386/tcg-target.c and mips/tcg-target.c
11 * Permission is hereby granted, free of charge, to any person obtaining a copy
12 * of this software and associated documentation files (the "Software"), to deal
13 * in the Software without restriction, including without limitation the rights
14 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
15 * copies of the Software, and to permit persons to whom the Software is
16 * furnished to do so, subject to the following conditions:
18 * The above copyright notice and this permission notice shall be included in
19 * all copies or substantial portions of the Software.
21 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
22 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
23 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
24 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
25 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
26 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30 #include "../tcg-ldst.c.inc"
31 #include "../tcg-pool.c.inc"
33 #ifdef CONFIG_DEBUG_TCG
34 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
70 static const int tcg_target_reg_alloc_order[] = {
71 /* Call saved registers */
72 /* TCG_REG_S0 reservered for TCG_AREG0 */
85 /* Call clobbered registers */
94 /* Argument registers */
105 static const int tcg_target_call_iarg_regs[] = {
116 static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
118 tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
119 tcg_debug_assert(slot >= 0 && slot <= 1);
120 return TCG_REG_A0 + slot;
123 #define TCG_CT_CONST_ZERO 0x100
124 #define TCG_CT_CONST_S12 0x200
125 #define TCG_CT_CONST_N12 0x400
126 #define TCG_CT_CONST_M12 0x800
128 #define ALL_GENERAL_REGS MAKE_64BIT_MASK(0, 32)
130 * For softmmu, we need to avoid conflicts with the first 5
131 * argument registers to call the helper. Some of these are
132 * also used for the tlb lookup.
134 #ifdef CONFIG_SOFTMMU
135 #define SOFTMMU_RESERVE_REGS MAKE_64BIT_MASK(TCG_REG_A0, 5)
137 #define SOFTMMU_RESERVE_REGS 0
141 static inline tcg_target_long sextreg(tcg_target_long val, int pos, int len)
143 if (TCG_TARGET_REG_BITS == 32) {
144 return sextract32(val, pos, len);
146 return sextract64(val, pos, len);
150 /* test if a constant matches the constraint */
151 static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
153 if (ct & TCG_CT_CONST) {
156 if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
160 * Sign extended from 12 bits: [-0x800, 0x7ff].
161 * Used for most arithmetic, as this is the isa field.
163 if ((ct & TCG_CT_CONST_S12) && val >= -0x800 && val <= 0x7ff) {
167 * Sign extended from 12 bits, negated: [-0x7ff, 0x800].
168 * Used for subtraction, where a constant must be handled by ADDI.
170 if ((ct & TCG_CT_CONST_N12) && val >= -0x7ff && val <= 0x800) {
174 * Sign extended from 12 bits, +/- matching: [-0x7ff, 0x7ff].
175 * Used by addsub2, which may need the negative operation,
176 * and requires the modified constant to be representable.
178 if ((ct & TCG_CT_CONST_M12) && val >= -0x7ff && val <= 0x7ff) {
185 * RISC-V Base ISA opcodes (IM)
201 OPC_DIVU = 0x2005033,
213 OPC_MULH = 0x2001033,
214 OPC_MULHSU = 0x2002033,
215 OPC_MULHU = 0x2003033,
219 OPC_REMU = 0x2007033,
229 OPC_SRA = 0x40005033,
230 OPC_SRAI = 0x40005013,
233 OPC_SUB = 0x40000033,
238 #if TCG_TARGET_REG_BITS == 64
241 OPC_DIVUW = 0x200503b,
242 OPC_DIVW = 0x200403b,
243 OPC_MULW = 0x200003b,
244 OPC_REMUW = 0x200703b,
245 OPC_REMW = 0x200603b,
248 OPC_SRAIW = 0x4000501b,
249 OPC_SRAW = 0x4000503b,
252 OPC_SUBW = 0x4000003b,
254 /* Simplify code throughout by defining aliases for RV32. */
255 OPC_ADDIW = OPC_ADDI,
257 OPC_DIVUW = OPC_DIVU,
260 OPC_REMUW = OPC_REMU,
262 OPC_SLLIW = OPC_SLLI,
264 OPC_SRAIW = OPC_SRAI,
266 OPC_SRLIW = OPC_SRLI,
271 OPC_FENCE = 0x0000000f,
272 OPC_NOP = OPC_ADDI, /* nop = addi r0,r0,0 */
276 * RISC-V immediate and instruction encoders (excludes 16-bit RVC)
281 static int32_t encode_r(RISCVInsn opc, TCGReg rd, TCGReg rs1, TCGReg rs2)
283 return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20;
288 static int32_t encode_imm12(uint32_t imm)
290 return (imm & 0xfff) << 20;
293 static int32_t encode_i(RISCVInsn opc, TCGReg rd, TCGReg rs1, uint32_t imm)
295 return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | encode_imm12(imm);
300 static int32_t encode_simm12(uint32_t imm)
304 ret |= (imm & 0xFE0) << 20;
305 ret |= (imm & 0x1F) << 7;
310 static int32_t encode_s(RISCVInsn opc, TCGReg rs1, TCGReg rs2, uint32_t imm)
312 return opc | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20 | encode_simm12(imm);
317 static int32_t encode_sbimm12(uint32_t imm)
321 ret |= (imm & 0x1000) << 19;
322 ret |= (imm & 0x7e0) << 20;
323 ret |= (imm & 0x1e) << 7;
324 ret |= (imm & 0x800) >> 4;
329 static int32_t encode_sb(RISCVInsn opc, TCGReg rs1, TCGReg rs2, uint32_t imm)
331 return opc | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20 | encode_sbimm12(imm);
336 static int32_t encode_uimm20(uint32_t imm)
338 return imm & 0xfffff000;
341 static int32_t encode_u(RISCVInsn opc, TCGReg rd, uint32_t imm)
343 return opc | (rd & 0x1f) << 7 | encode_uimm20(imm);
348 static int32_t encode_ujimm20(uint32_t imm)
352 ret |= (imm & 0x0007fe) << (21 - 1);
353 ret |= (imm & 0x000800) << (20 - 11);
354 ret |= (imm & 0x0ff000) << (12 - 12);
355 ret |= (imm & 0x100000) << (31 - 20);
360 static int32_t encode_uj(RISCVInsn opc, TCGReg rd, uint32_t imm)
362 return opc | (rd & 0x1f) << 7 | encode_ujimm20(imm);
366 * RISC-V instruction emitters
369 static void tcg_out_opc_reg(TCGContext *s, RISCVInsn opc,
370 TCGReg rd, TCGReg rs1, TCGReg rs2)
372 tcg_out32(s, encode_r(opc, rd, rs1, rs2));
375 static void tcg_out_opc_imm(TCGContext *s, RISCVInsn opc,
376 TCGReg rd, TCGReg rs1, TCGArg imm)
378 tcg_out32(s, encode_i(opc, rd, rs1, imm));
381 static void tcg_out_opc_store(TCGContext *s, RISCVInsn opc,
382 TCGReg rs1, TCGReg rs2, uint32_t imm)
384 tcg_out32(s, encode_s(opc, rs1, rs2, imm));
387 static void tcg_out_opc_branch(TCGContext *s, RISCVInsn opc,
388 TCGReg rs1, TCGReg rs2, uint32_t imm)
390 tcg_out32(s, encode_sb(opc, rs1, rs2, imm));
393 static void tcg_out_opc_upper(TCGContext *s, RISCVInsn opc,
394 TCGReg rd, uint32_t imm)
396 tcg_out32(s, encode_u(opc, rd, imm));
399 static void tcg_out_opc_jump(TCGContext *s, RISCVInsn opc,
400 TCGReg rd, uint32_t imm)
402 tcg_out32(s, encode_uj(opc, rd, imm));
405 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
408 for (i = 0; i < count; ++i) {
417 static bool reloc_sbimm12(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
419 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
420 intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
422 tcg_debug_assert((offset & 1) == 0);
423 if (offset == sextreg(offset, 0, 12)) {
424 *src_rw |= encode_sbimm12(offset);
431 static bool reloc_jimm20(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
433 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
434 intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
436 tcg_debug_assert((offset & 1) == 0);
437 if (offset == sextreg(offset, 0, 20)) {
438 *src_rw |= encode_ujimm20(offset);
445 static bool reloc_call(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
447 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
448 intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
449 int32_t lo = sextreg(offset, 0, 12);
450 int32_t hi = offset - lo;
452 if (offset == hi + lo) {
453 src_rw[0] |= encode_uimm20(hi);
454 src_rw[1] |= encode_imm12(lo);
461 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
462 intptr_t value, intptr_t addend)
464 tcg_debug_assert(addend == 0);
467 return reloc_sbimm12(code_ptr, (tcg_insn_unit *)value);
469 return reloc_jimm20(code_ptr, (tcg_insn_unit *)value);
471 return reloc_call(code_ptr, (tcg_insn_unit *)value);
473 g_assert_not_reached();
481 static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
489 tcg_out_opc_imm(s, OPC_ADDI, ret, arg, 0);
492 g_assert_not_reached();
497 static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
500 tcg_target_long lo, hi, tmp;
503 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
507 lo = sextreg(val, 0, 12);
509 tcg_out_opc_imm(s, OPC_ADDI, rd, TCG_REG_ZERO, lo);
514 if (TCG_TARGET_REG_BITS == 32 || val == (int32_t)val) {
515 tcg_out_opc_upper(s, OPC_LUI, rd, hi);
517 tcg_out_opc_imm(s, OPC_ADDIW, rd, rd, lo);
522 /* We can only be here if TCG_TARGET_REG_BITS != 32 */
523 tmp = tcg_pcrel_diff(s, (void *)val);
524 if (tmp == (int32_t)tmp) {
525 tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
526 tcg_out_opc_imm(s, OPC_ADDI, rd, rd, 0);
527 ret = reloc_call(s->code_ptr - 2, (const tcg_insn_unit *)val);
528 tcg_debug_assert(ret == true);
532 /* Look for a single 20-bit section. */
535 if (tmp == sextreg(tmp, 0, 20)) {
536 tcg_out_opc_upper(s, OPC_LUI, rd, tmp << 12);
538 tcg_out_opc_imm(s, OPC_SLLI, rd, rd, shift - 12);
540 tcg_out_opc_imm(s, OPC_SRAI, rd, rd, 12 - shift);
545 /* Look for a few high zero bits, with lots of bits set in the middle. */
548 if (tmp == sextreg(tmp, 12, 20) << 12) {
549 tcg_out_opc_upper(s, OPC_LUI, rd, tmp);
550 tcg_out_opc_imm(s, OPC_SRLI, rd, rd, shift);
552 } else if (tmp == sextreg(tmp, 0, 12)) {
553 tcg_out_opc_imm(s, OPC_ADDI, rd, TCG_REG_ZERO, tmp);
554 tcg_out_opc_imm(s, OPC_SRLI, rd, rd, shift);
558 /* Drop into the constant pool. */
559 new_pool_label(s, val, R_RISCV_CALL, s->code_ptr, 0);
560 tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
561 tcg_out_opc_imm(s, OPC_LD, rd, rd, 0);
564 static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
567 /* This function is only used for passing structs by reference. */
568 g_assert_not_reached();
571 static void tcg_out_ext8u(TCGContext *s, TCGReg ret, TCGReg arg)
573 tcg_out_opc_imm(s, OPC_ANDI, ret, arg, 0xff);
576 static void tcg_out_ext16u(TCGContext *s, TCGReg ret, TCGReg arg)
578 tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 16);
579 tcg_out_opc_imm(s, OPC_SRLIW, ret, ret, 16);
582 static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg)
584 tcg_out_opc_imm(s, OPC_SLLI, ret, arg, 32);
585 tcg_out_opc_imm(s, OPC_SRLI, ret, ret, 32);
588 static void tcg_out_ext8s(TCGContext *s, TCGReg ret, TCGReg arg)
590 tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 24);
591 tcg_out_opc_imm(s, OPC_SRAIW, ret, ret, 24);
594 static void tcg_out_ext16s(TCGContext *s, TCGReg ret, TCGReg arg)
596 tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 16);
597 tcg_out_opc_imm(s, OPC_SRAIW, ret, ret, 16);
600 static void tcg_out_ext32s(TCGContext *s, TCGReg ret, TCGReg arg)
602 tcg_out_opc_imm(s, OPC_ADDIW, ret, arg, 0);
605 static void tcg_out_ldst(TCGContext *s, RISCVInsn opc, TCGReg data,
606 TCGReg addr, intptr_t offset)
608 intptr_t imm12 = sextreg(offset, 0, 12);
610 if (offset != imm12) {
611 intptr_t diff = tcg_pcrel_diff(s, (void *)offset);
613 if (addr == TCG_REG_ZERO && diff == (int32_t)diff) {
614 imm12 = sextreg(diff, 0, 12);
615 tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP2, diff - imm12);
617 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP2, offset - imm12);
618 if (addr != TCG_REG_ZERO) {
619 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, addr);
630 tcg_out_opc_store(s, opc, addr, data, imm12);
639 tcg_out_opc_imm(s, opc, data, addr, imm12);
642 g_assert_not_reached();
646 static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
647 TCGReg arg1, intptr_t arg2)
649 bool is32bit = (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I32);
650 tcg_out_ldst(s, is32bit ? OPC_LW : OPC_LD, arg, arg1, arg2);
653 static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
654 TCGReg arg1, intptr_t arg2)
656 bool is32bit = (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I32);
657 tcg_out_ldst(s, is32bit ? OPC_SW : OPC_SD, arg, arg1, arg2);
660 static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
661 TCGReg base, intptr_t ofs)
664 tcg_out_st(s, type, TCG_REG_ZERO, base, ofs);
670 static void tcg_out_addsub2(TCGContext *s,
671 TCGReg rl, TCGReg rh,
672 TCGReg al, TCGReg ah,
673 TCGArg bl, TCGArg bh,
674 bool cbl, bool cbh, bool is_sub, bool is32bit)
676 const RISCVInsn opc_add = is32bit ? OPC_ADDW : OPC_ADD;
677 const RISCVInsn opc_addi = is32bit ? OPC_ADDIW : OPC_ADDI;
678 const RISCVInsn opc_sub = is32bit ? OPC_SUBW : OPC_SUB;
679 TCGReg th = TCG_REG_TMP1;
681 /* If we have a negative constant such that negating it would
682 make the high part zero, we can (usually) eliminate one insn. */
683 if (cbl && cbh && bh == -1 && bl != 0) {
689 /* By operating on the high part first, we get to use the final
690 carry operation to move back from the temporary. */
692 tcg_out_opc_reg(s, (is_sub ? opc_sub : opc_add), th, ah, bh);
693 } else if (bh != 0 || ah == rl) {
694 tcg_out_opc_imm(s, opc_addi, th, ah, (is_sub ? -bh : bh));
699 /* Note that tcg optimization should eliminate the bl == 0 case. */
702 tcg_out_opc_imm(s, OPC_SLTIU, TCG_REG_TMP0, al, bl);
703 tcg_out_opc_imm(s, opc_addi, rl, al, -bl);
705 tcg_out_opc_reg(s, OPC_SLTU, TCG_REG_TMP0, al, bl);
706 tcg_out_opc_reg(s, opc_sub, rl, al, bl);
708 tcg_out_opc_reg(s, opc_sub, rh, th, TCG_REG_TMP0);
711 tcg_out_opc_imm(s, opc_addi, rl, al, bl);
712 tcg_out_opc_imm(s, OPC_SLTIU, TCG_REG_TMP0, rl, bl);
713 } else if (al == bl) {
715 * If the input regs overlap, this is a simple doubling
716 * and carry-out is the input msb. This special case is
717 * required when the output reg overlaps the input,
718 * but we might as well use it always.
720 tcg_out_opc_imm(s, OPC_SLTI, TCG_REG_TMP0, al, 0);
721 tcg_out_opc_reg(s, opc_add, rl, al, al);
723 tcg_out_opc_reg(s, opc_add, rl, al, bl);
724 tcg_out_opc_reg(s, OPC_SLTU, TCG_REG_TMP0,
725 rl, (rl == bl ? al : bl));
727 tcg_out_opc_reg(s, opc_add, rh, th, TCG_REG_TMP0);
731 static const struct {
734 } tcg_brcond_to_riscv[] = {
735 [TCG_COND_EQ] = { OPC_BEQ, false },
736 [TCG_COND_NE] = { OPC_BNE, false },
737 [TCG_COND_LT] = { OPC_BLT, false },
738 [TCG_COND_GE] = { OPC_BGE, false },
739 [TCG_COND_LE] = { OPC_BGE, true },
740 [TCG_COND_GT] = { OPC_BLT, true },
741 [TCG_COND_LTU] = { OPC_BLTU, false },
742 [TCG_COND_GEU] = { OPC_BGEU, false },
743 [TCG_COND_LEU] = { OPC_BGEU, true },
744 [TCG_COND_GTU] = { OPC_BLTU, true }
747 static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
748 TCGReg arg2, TCGLabel *l)
750 RISCVInsn op = tcg_brcond_to_riscv[cond].op;
752 tcg_debug_assert(op != 0);
754 if (tcg_brcond_to_riscv[cond].swap) {
760 tcg_out_reloc(s, s->code_ptr, R_RISCV_BRANCH, l, 0);
761 tcg_out_opc_branch(s, op, arg1, arg2, 0);
764 static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret,
765 TCGReg arg1, TCGReg arg2)
769 tcg_out_opc_reg(s, OPC_SUB, ret, arg1, arg2);
770 tcg_out_opc_imm(s, OPC_SLTIU, ret, ret, 1);
773 tcg_out_opc_reg(s, OPC_SUB, ret, arg1, arg2);
774 tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, ret);
777 tcg_out_opc_reg(s, OPC_SLT, ret, arg1, arg2);
780 tcg_out_opc_reg(s, OPC_SLT, ret, arg1, arg2);
781 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
784 tcg_out_opc_reg(s, OPC_SLT, ret, arg2, arg1);
785 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
788 tcg_out_opc_reg(s, OPC_SLT, ret, arg2, arg1);
791 tcg_out_opc_reg(s, OPC_SLTU, ret, arg1, arg2);
794 tcg_out_opc_reg(s, OPC_SLTU, ret, arg1, arg2);
795 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
798 tcg_out_opc_reg(s, OPC_SLTU, ret, arg2, arg1);
799 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
802 tcg_out_opc_reg(s, OPC_SLTU, ret, arg2, arg1);
805 g_assert_not_reached();
810 static void tcg_out_brcond2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
811 TCGReg bl, TCGReg bh, TCGLabel *l)
814 g_assert_not_reached();
817 static void tcg_out_setcond2(TCGContext *s, TCGCond cond, TCGReg ret,
818 TCGReg al, TCGReg ah, TCGReg bl, TCGReg bh)
821 g_assert_not_reached();
824 static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
826 TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
827 ptrdiff_t offset = tcg_pcrel_diff(s, arg);
830 tcg_debug_assert((offset & 1) == 0);
831 if (offset == sextreg(offset, 0, 20)) {
832 /* short jump: -2097150 to 2097152 */
833 tcg_out_opc_jump(s, OPC_JAL, link, offset);
834 } else if (TCG_TARGET_REG_BITS == 32 || offset == (int32_t)offset) {
835 /* long jump: -2147483646 to 2147483648 */
836 tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
837 tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
838 ret = reloc_call(s->code_ptr - 2, arg);
839 tcg_debug_assert(ret == true);
840 } else if (TCG_TARGET_REG_BITS == 64) {
841 /* far jump: 64-bit */
842 tcg_target_long imm = sextreg((tcg_target_long)arg, 0, 12);
843 tcg_target_long base = (tcg_target_long)arg - imm;
844 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, base);
845 tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, imm);
847 g_assert_not_reached();
851 static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg,
852 const TCGHelperInfo *info)
854 tcg_out_call_int(s, arg, false);
857 static void tcg_out_mb(TCGContext *s, TCGArg a0)
859 tcg_insn_unit insn = OPC_FENCE;
861 if (a0 & TCG_MO_LD_LD) {
864 if (a0 & TCG_MO_ST_LD) {
867 if (a0 & TCG_MO_LD_ST) {
870 if (a0 & TCG_MO_ST_ST) {
880 #if defined(CONFIG_SOFTMMU)
881 /* helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
882 * MemOpIdx oi, uintptr_t ra)
884 static void * const qemu_ld_helpers[MO_SSIZE + 1] = {
885 [MO_UB] = helper_ret_ldub_mmu,
886 [MO_SB] = helper_ret_ldsb_mmu,
888 [MO_UW] = helper_be_lduw_mmu,
889 [MO_SW] = helper_be_ldsw_mmu,
890 [MO_UL] = helper_be_ldul_mmu,
891 #if TCG_TARGET_REG_BITS == 64
892 [MO_SL] = helper_be_ldsl_mmu,
894 [MO_UQ] = helper_be_ldq_mmu,
896 [MO_UW] = helper_le_lduw_mmu,
897 [MO_SW] = helper_le_ldsw_mmu,
898 [MO_UL] = helper_le_ldul_mmu,
899 #if TCG_TARGET_REG_BITS == 64
900 [MO_SL] = helper_le_ldsl_mmu,
902 [MO_UQ] = helper_le_ldq_mmu,
906 /* helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
907 * uintxx_t val, MemOpIdx oi,
910 static void * const qemu_st_helpers[MO_SIZE + 1] = {
911 [MO_8] = helper_ret_stb_mmu,
913 [MO_16] = helper_be_stw_mmu,
914 [MO_32] = helper_be_stl_mmu,
915 [MO_64] = helper_be_stq_mmu,
917 [MO_16] = helper_le_stw_mmu,
918 [MO_32] = helper_le_stl_mmu,
919 [MO_64] = helper_le_stq_mmu,
923 /* We don't support oversize guests */
924 QEMU_BUILD_BUG_ON(TCG_TARGET_REG_BITS < TARGET_LONG_BITS);
926 /* We expect to use a 12-bit negative offset from ENV. */
927 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
928 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 11));
930 static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
932 tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
933 bool ok = reloc_jimm20(s->code_ptr - 1, target);
934 tcg_debug_assert(ok);
937 static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addrl,
938 TCGReg addrh, MemOpIdx oi,
939 tcg_insn_unit **label_ptr, bool is_load)
941 MemOp opc = get_memop(oi);
942 unsigned s_bits = opc & MO_SIZE;
943 unsigned a_bits = get_alignment_bits(opc);
944 tcg_target_long compare_mask;
945 int mem_index = get_mmuidx(oi);
946 int fast_ofs = TLB_MASK_TABLE_OFS(mem_index);
947 int mask_ofs = fast_ofs + offsetof(CPUTLBDescFast, mask);
948 int table_ofs = fast_ofs + offsetof(CPUTLBDescFast, table);
949 TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
951 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, mask_base, mask_ofs);
952 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, table_base, table_ofs);
954 tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP2, addrl,
955 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
956 tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP0);
957 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP1);
959 /* Load the tlb comparator and the addend. */
960 tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP0, TCG_REG_TMP2,
961 is_load ? offsetof(CPUTLBEntry, addr_read)
962 : offsetof(CPUTLBEntry, addr_write));
963 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP2,
964 offsetof(CPUTLBEntry, addend));
966 /* We don't support unaligned accesses. */
967 if (a_bits < s_bits) {
970 /* Clear the non-page, non-alignment bits from the address. */
971 compare_mask = (tcg_target_long)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
972 if (compare_mask == sextreg(compare_mask, 0, 12)) {
973 tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addrl, compare_mask);
975 tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_TMP1, compare_mask);
976 tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP1, TCG_REG_TMP1, addrl);
979 /* Compare masked address with the TLB entry. */
980 label_ptr[0] = s->code_ptr;
981 tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP0, TCG_REG_TMP1, 0);
983 /* TLB Hit - translate address using addend. */
984 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
985 tcg_out_ext32u(s, TCG_REG_TMP0, addrl);
986 addrl = TCG_REG_TMP0;
988 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_REG_TMP2, addrl);
992 static void add_qemu_ldst_label(TCGContext *s, int is_ld, MemOpIdx oi,
994 TCGReg datalo, TCGReg datahi,
995 TCGReg addrlo, TCGReg addrhi,
996 void *raddr, tcg_insn_unit **label_ptr)
998 TCGLabelQemuLdst *label = new_ldst_label(s);
1000 label->is_ld = is_ld;
1003 label->datalo_reg = datalo;
1004 label->datahi_reg = datahi;
1005 label->addrlo_reg = addrlo;
1006 label->addrhi_reg = addrhi;
1007 label->raddr = tcg_splitwx_to_rx(raddr);
1008 label->label_ptr[0] = label_ptr[0];
1011 static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1013 MemOpIdx oi = l->oi;
1014 MemOp opc = get_memop(oi);
1015 TCGReg a0 = tcg_target_call_iarg_regs[0];
1016 TCGReg a1 = tcg_target_call_iarg_regs[1];
1017 TCGReg a2 = tcg_target_call_iarg_regs[2];
1018 TCGReg a3 = tcg_target_call_iarg_regs[3];
1020 /* We don't support oversize guests */
1021 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1022 g_assert_not_reached();
1025 /* resolve label address */
1026 if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1030 /* call load helper */
1031 tcg_out_mov(s, TCG_TYPE_PTR, a0, TCG_AREG0);
1032 tcg_out_mov(s, TCG_TYPE_PTR, a1, l->addrlo_reg);
1033 tcg_out_movi(s, TCG_TYPE_PTR, a2, oi);
1034 tcg_out_movi(s, TCG_TYPE_PTR, a3, (tcg_target_long)l->raddr);
1036 tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SSIZE], false);
1037 tcg_out_mov(s, (opc & MO_SIZE) == MO_64, l->datalo_reg, a0);
1039 tcg_out_goto(s, l->raddr);
1043 static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1045 MemOpIdx oi = l->oi;
1046 MemOp opc = get_memop(oi);
1047 MemOp s_bits = opc & MO_SIZE;
1048 TCGReg a0 = tcg_target_call_iarg_regs[0];
1049 TCGReg a1 = tcg_target_call_iarg_regs[1];
1050 TCGReg a2 = tcg_target_call_iarg_regs[2];
1051 TCGReg a3 = tcg_target_call_iarg_regs[3];
1052 TCGReg a4 = tcg_target_call_iarg_regs[4];
1054 /* We don't support oversize guests */
1055 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1056 g_assert_not_reached();
1059 /* resolve label address */
1060 if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1064 /* call store helper */
1065 tcg_out_mov(s, TCG_TYPE_PTR, a0, TCG_AREG0);
1066 tcg_out_mov(s, TCG_TYPE_PTR, a1, l->addrlo_reg);
1067 tcg_out_mov(s, TCG_TYPE_PTR, a2, l->datalo_reg);
1070 tcg_out_ext8u(s, a2, a2);
1073 tcg_out_ext16u(s, a2, a2);
1078 tcg_out_movi(s, TCG_TYPE_PTR, a3, oi);
1079 tcg_out_movi(s, TCG_TYPE_PTR, a4, (tcg_target_long)l->raddr);
1081 tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE], false);
1083 tcg_out_goto(s, l->raddr);
1088 static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addr_reg,
1091 unsigned a_mask = (1 << a_bits) - 1;
1092 TCGLabelQemuLdst *l = new_ldst_label(s);
1095 l->addrlo_reg = addr_reg;
1097 /* We are expecting a_bits to max out at 7, so we can always use andi. */
1098 tcg_debug_assert(a_bits < 12);
1099 tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addr_reg, a_mask);
1101 l->label_ptr[0] = s->code_ptr;
1102 tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP1, TCG_REG_ZERO, 0);
1104 l->raddr = tcg_splitwx_to_rx(s->code_ptr);
1107 static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l)
1109 /* resolve label address */
1110 if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1114 tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_A1, l->addrlo_reg);
1115 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A0, TCG_AREG0);
1117 /* tail call, with the return address back inline. */
1118 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_RA, (uintptr_t)l->raddr);
1119 tcg_out_call_int(s, (const void *)(l->is_ld ? helper_unaligned_ld
1120 : helper_unaligned_st), true);
1124 static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1126 return tcg_out_fail_alignment(s, l);
1129 static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1131 return tcg_out_fail_alignment(s, l);
1134 #endif /* CONFIG_SOFTMMU */
1136 static void tcg_out_qemu_ld_direct(TCGContext *s, TCGReg lo, TCGReg hi,
1137 TCGReg base, MemOp opc, bool is_64)
1139 /* Byte swapping is left to middle-end expansion. */
1140 tcg_debug_assert((opc & MO_BSWAP) == 0);
1142 switch (opc & (MO_SSIZE)) {
1144 tcg_out_opc_imm(s, OPC_LBU, lo, base, 0);
1147 tcg_out_opc_imm(s, OPC_LB, lo, base, 0);
1150 tcg_out_opc_imm(s, OPC_LHU, lo, base, 0);
1153 tcg_out_opc_imm(s, OPC_LH, lo, base, 0);
1156 if (TCG_TARGET_REG_BITS == 64 && is_64) {
1157 tcg_out_opc_imm(s, OPC_LWU, lo, base, 0);
1162 tcg_out_opc_imm(s, OPC_LW, lo, base, 0);
1165 /* Prefer to load from offset 0 first, but allow for overlap. */
1166 if (TCG_TARGET_REG_BITS == 64) {
1167 tcg_out_opc_imm(s, OPC_LD, lo, base, 0);
1168 } else if (lo != base) {
1169 tcg_out_opc_imm(s, OPC_LW, lo, base, 0);
1170 tcg_out_opc_imm(s, OPC_LW, hi, base, 4);
1172 tcg_out_opc_imm(s, OPC_LW, hi, base, 4);
1173 tcg_out_opc_imm(s, OPC_LW, lo, base, 0);
1177 g_assert_not_reached();
1181 static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
1183 TCGReg addr_regl, addr_regh __attribute__((unused));
1184 TCGReg data_regl, data_regh;
1187 #if defined(CONFIG_SOFTMMU)
1188 tcg_insn_unit *label_ptr[1];
1194 data_regl = *args++;
1195 data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1196 addr_regl = *args++;
1197 addr_regh = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1199 opc = get_memop(oi);
1201 #if defined(CONFIG_SOFTMMU)
1202 base = tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 1);
1203 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64);
1204 add_qemu_ldst_label(s, 1, oi,
1205 (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32),
1206 data_regl, data_regh, addr_regl, addr_regh,
1207 s->code_ptr, label_ptr);
1209 a_bits = get_alignment_bits(opc);
1211 tcg_out_test_alignment(s, true, addr_regl, a_bits);
1214 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1215 tcg_out_ext32u(s, TCG_REG_TMP0, base);
1216 base = TCG_REG_TMP0;
1218 if (guest_base != 0) {
1219 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_GUEST_BASE_REG, base);
1220 base = TCG_REG_TMP0;
1222 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64);
1226 static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg lo, TCGReg hi,
1227 TCGReg base, MemOp opc)
1229 /* Byte swapping is left to middle-end expansion. */
1230 tcg_debug_assert((opc & MO_BSWAP) == 0);
1232 switch (opc & (MO_SSIZE)) {
1234 tcg_out_opc_store(s, OPC_SB, base, lo, 0);
1237 tcg_out_opc_store(s, OPC_SH, base, lo, 0);
1240 tcg_out_opc_store(s, OPC_SW, base, lo, 0);
1243 if (TCG_TARGET_REG_BITS == 64) {
1244 tcg_out_opc_store(s, OPC_SD, base, lo, 0);
1246 tcg_out_opc_store(s, OPC_SW, base, lo, 0);
1247 tcg_out_opc_store(s, OPC_SW, base, hi, 4);
1251 g_assert_not_reached();
1255 static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
1257 TCGReg addr_regl, addr_regh __attribute__((unused));
1258 TCGReg data_regl, data_regh;
1261 #if defined(CONFIG_SOFTMMU)
1262 tcg_insn_unit *label_ptr[1];
1268 data_regl = *args++;
1269 data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1270 addr_regl = *args++;
1271 addr_regh = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1273 opc = get_memop(oi);
1275 #if defined(CONFIG_SOFTMMU)
1276 base = tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 0);
1277 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc);
1278 add_qemu_ldst_label(s, 0, oi,
1279 (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32),
1280 data_regl, data_regh, addr_regl, addr_regh,
1281 s->code_ptr, label_ptr);
1283 a_bits = get_alignment_bits(opc);
1285 tcg_out_test_alignment(s, false, addr_regl, a_bits);
1288 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1289 tcg_out_ext32u(s, TCG_REG_TMP0, base);
1290 base = TCG_REG_TMP0;
1292 if (guest_base != 0) {
1293 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_GUEST_BASE_REG, base);
1294 base = TCG_REG_TMP0;
1296 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc);
1300 static const tcg_insn_unit *tb_ret_addr;
1302 static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1304 /* Reuse the zeroing that exists for goto_ptr. */
1306 tcg_out_call_int(s, tcg_code_gen_epilogue, true);
1308 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A0, a0);
1309 tcg_out_call_int(s, tb_ret_addr, true);
1313 static void tcg_out_goto_tb(TCGContext *s, int which)
1315 /* Direct branch will be patched by tb_target_set_jmp_target. */
1316 set_jmp_insn_offset(s, which);
1317 tcg_out32(s, OPC_JAL);
1319 /* When branch is out of range, fall through to indirect. */
1320 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_REG_ZERO,
1321 get_jmp_target_addr(s, which));
1322 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_TMP0, 0);
1323 set_jmp_reset_offset(s, which);
1326 void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1327 uintptr_t jmp_rx, uintptr_t jmp_rw)
1329 uintptr_t addr = tb->jmp_target_addr[n];
1330 ptrdiff_t offset = addr - jmp_rx;
1333 /* Either directly branch, or fall through to indirect branch. */
1334 if (offset == sextreg(offset, 0, 20)) {
1335 insn = encode_uj(OPC_JAL, TCG_REG_ZERO, offset);
1339 qatomic_set((uint32_t *)jmp_rw, insn);
1340 flush_idcache_range(jmp_rx, jmp_rw, 4);
1343 static void tcg_out_op(TCGContext *s, TCGOpcode opc,
1344 const TCGArg args[TCG_MAX_OP_ARGS],
1345 const int const_args[TCG_MAX_OP_ARGS])
1347 TCGArg a0 = args[0];
1348 TCGArg a1 = args[1];
1349 TCGArg a2 = args[2];
1350 int c2 = const_args[2];
1353 case INDEX_op_goto_ptr:
1354 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, a0, 0);
1358 tcg_out_reloc(s, s->code_ptr, R_RISCV_JAL, arg_label(a0), 0);
1359 tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
1362 case INDEX_op_ld8u_i32:
1363 case INDEX_op_ld8u_i64:
1364 tcg_out_ldst(s, OPC_LBU, a0, a1, a2);
1366 case INDEX_op_ld8s_i32:
1367 case INDEX_op_ld8s_i64:
1368 tcg_out_ldst(s, OPC_LB, a0, a1, a2);
1370 case INDEX_op_ld16u_i32:
1371 case INDEX_op_ld16u_i64:
1372 tcg_out_ldst(s, OPC_LHU, a0, a1, a2);
1374 case INDEX_op_ld16s_i32:
1375 case INDEX_op_ld16s_i64:
1376 tcg_out_ldst(s, OPC_LH, a0, a1, a2);
1378 case INDEX_op_ld32u_i64:
1379 tcg_out_ldst(s, OPC_LWU, a0, a1, a2);
1381 case INDEX_op_ld_i32:
1382 case INDEX_op_ld32s_i64:
1383 tcg_out_ldst(s, OPC_LW, a0, a1, a2);
1385 case INDEX_op_ld_i64:
1386 tcg_out_ldst(s, OPC_LD, a0, a1, a2);
1389 case INDEX_op_st8_i32:
1390 case INDEX_op_st8_i64:
1391 tcg_out_ldst(s, OPC_SB, a0, a1, a2);
1393 case INDEX_op_st16_i32:
1394 case INDEX_op_st16_i64:
1395 tcg_out_ldst(s, OPC_SH, a0, a1, a2);
1397 case INDEX_op_st_i32:
1398 case INDEX_op_st32_i64:
1399 tcg_out_ldst(s, OPC_SW, a0, a1, a2);
1401 case INDEX_op_st_i64:
1402 tcg_out_ldst(s, OPC_SD, a0, a1, a2);
1405 case INDEX_op_add_i32:
1407 tcg_out_opc_imm(s, OPC_ADDIW, a0, a1, a2);
1409 tcg_out_opc_reg(s, OPC_ADDW, a0, a1, a2);
1412 case INDEX_op_add_i64:
1414 tcg_out_opc_imm(s, OPC_ADDI, a0, a1, a2);
1416 tcg_out_opc_reg(s, OPC_ADD, a0, a1, a2);
1420 case INDEX_op_sub_i32:
1422 tcg_out_opc_imm(s, OPC_ADDIW, a0, a1, -a2);
1424 tcg_out_opc_reg(s, OPC_SUBW, a0, a1, a2);
1427 case INDEX_op_sub_i64:
1429 tcg_out_opc_imm(s, OPC_ADDI, a0, a1, -a2);
1431 tcg_out_opc_reg(s, OPC_SUB, a0, a1, a2);
1435 case INDEX_op_and_i32:
1436 case INDEX_op_and_i64:
1438 tcg_out_opc_imm(s, OPC_ANDI, a0, a1, a2);
1440 tcg_out_opc_reg(s, OPC_AND, a0, a1, a2);
1444 case INDEX_op_or_i32:
1445 case INDEX_op_or_i64:
1447 tcg_out_opc_imm(s, OPC_ORI, a0, a1, a2);
1449 tcg_out_opc_reg(s, OPC_OR, a0, a1, a2);
1453 case INDEX_op_xor_i32:
1454 case INDEX_op_xor_i64:
1456 tcg_out_opc_imm(s, OPC_XORI, a0, a1, a2);
1458 tcg_out_opc_reg(s, OPC_XOR, a0, a1, a2);
1462 case INDEX_op_not_i32:
1463 case INDEX_op_not_i64:
1464 tcg_out_opc_imm(s, OPC_XORI, a0, a1, -1);
1467 case INDEX_op_neg_i32:
1468 tcg_out_opc_reg(s, OPC_SUBW, a0, TCG_REG_ZERO, a1);
1470 case INDEX_op_neg_i64:
1471 tcg_out_opc_reg(s, OPC_SUB, a0, TCG_REG_ZERO, a1);
1474 case INDEX_op_mul_i32:
1475 tcg_out_opc_reg(s, OPC_MULW, a0, a1, a2);
1477 case INDEX_op_mul_i64:
1478 tcg_out_opc_reg(s, OPC_MUL, a0, a1, a2);
1481 case INDEX_op_div_i32:
1482 tcg_out_opc_reg(s, OPC_DIVW, a0, a1, a2);
1484 case INDEX_op_div_i64:
1485 tcg_out_opc_reg(s, OPC_DIV, a0, a1, a2);
1488 case INDEX_op_divu_i32:
1489 tcg_out_opc_reg(s, OPC_DIVUW, a0, a1, a2);
1491 case INDEX_op_divu_i64:
1492 tcg_out_opc_reg(s, OPC_DIVU, a0, a1, a2);
1495 case INDEX_op_rem_i32:
1496 tcg_out_opc_reg(s, OPC_REMW, a0, a1, a2);
1498 case INDEX_op_rem_i64:
1499 tcg_out_opc_reg(s, OPC_REM, a0, a1, a2);
1502 case INDEX_op_remu_i32:
1503 tcg_out_opc_reg(s, OPC_REMUW, a0, a1, a2);
1505 case INDEX_op_remu_i64:
1506 tcg_out_opc_reg(s, OPC_REMU, a0, a1, a2);
1509 case INDEX_op_shl_i32:
1511 tcg_out_opc_imm(s, OPC_SLLIW, a0, a1, a2 & 0x1f);
1513 tcg_out_opc_reg(s, OPC_SLLW, a0, a1, a2);
1516 case INDEX_op_shl_i64:
1518 tcg_out_opc_imm(s, OPC_SLLI, a0, a1, a2 & 0x3f);
1520 tcg_out_opc_reg(s, OPC_SLL, a0, a1, a2);
1524 case INDEX_op_shr_i32:
1526 tcg_out_opc_imm(s, OPC_SRLIW, a0, a1, a2 & 0x1f);
1528 tcg_out_opc_reg(s, OPC_SRLW, a0, a1, a2);
1531 case INDEX_op_shr_i64:
1533 tcg_out_opc_imm(s, OPC_SRLI, a0, a1, a2 & 0x3f);
1535 tcg_out_opc_reg(s, OPC_SRL, a0, a1, a2);
1539 case INDEX_op_sar_i32:
1541 tcg_out_opc_imm(s, OPC_SRAIW, a0, a1, a2 & 0x1f);
1543 tcg_out_opc_reg(s, OPC_SRAW, a0, a1, a2);
1546 case INDEX_op_sar_i64:
1548 tcg_out_opc_imm(s, OPC_SRAI, a0, a1, a2 & 0x3f);
1550 tcg_out_opc_reg(s, OPC_SRA, a0, a1, a2);
1554 case INDEX_op_add2_i32:
1555 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1556 const_args[4], const_args[5], false, true);
1558 case INDEX_op_add2_i64:
1559 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1560 const_args[4], const_args[5], false, false);
1562 case INDEX_op_sub2_i32:
1563 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1564 const_args[4], const_args[5], true, true);
1566 case INDEX_op_sub2_i64:
1567 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1568 const_args[4], const_args[5], true, false);
1571 case INDEX_op_brcond_i32:
1572 case INDEX_op_brcond_i64:
1573 tcg_out_brcond(s, a2, a0, a1, arg_label(args[3]));
1575 case INDEX_op_brcond2_i32:
1576 tcg_out_brcond2(s, args[4], a0, a1, a2, args[3], arg_label(args[5]));
1579 case INDEX_op_setcond_i32:
1580 case INDEX_op_setcond_i64:
1581 tcg_out_setcond(s, args[3], a0, a1, a2);
1583 case INDEX_op_setcond2_i32:
1584 tcg_out_setcond2(s, args[5], a0, a1, a2, args[3], args[4]);
1587 case INDEX_op_qemu_ld_i32:
1588 tcg_out_qemu_ld(s, args, false);
1590 case INDEX_op_qemu_ld_i64:
1591 tcg_out_qemu_ld(s, args, true);
1593 case INDEX_op_qemu_st_i32:
1594 tcg_out_qemu_st(s, args, false);
1596 case INDEX_op_qemu_st_i64:
1597 tcg_out_qemu_st(s, args, true);
1600 case INDEX_op_ext8u_i32:
1601 case INDEX_op_ext8u_i64:
1602 tcg_out_ext8u(s, a0, a1);
1605 case INDEX_op_ext16u_i32:
1606 case INDEX_op_ext16u_i64:
1607 tcg_out_ext16u(s, a0, a1);
1610 case INDEX_op_ext32u_i64:
1611 case INDEX_op_extu_i32_i64:
1612 tcg_out_ext32u(s, a0, a1);
1615 case INDEX_op_ext8s_i32:
1616 case INDEX_op_ext8s_i64:
1617 tcg_out_ext8s(s, a0, a1);
1620 case INDEX_op_ext16s_i32:
1621 case INDEX_op_ext16s_i64:
1622 tcg_out_ext16s(s, a0, a1);
1625 case INDEX_op_ext32s_i64:
1626 case INDEX_op_extrl_i64_i32:
1627 case INDEX_op_ext_i32_i64:
1628 tcg_out_ext32s(s, a0, a1);
1631 case INDEX_op_extrh_i64_i32:
1632 tcg_out_opc_imm(s, OPC_SRAI, a0, a1, 32);
1635 case INDEX_op_mulsh_i32:
1636 case INDEX_op_mulsh_i64:
1637 tcg_out_opc_reg(s, OPC_MULH, a0, a1, a2);
1640 case INDEX_op_muluh_i32:
1641 case INDEX_op_muluh_i64:
1642 tcg_out_opc_reg(s, OPC_MULHU, a0, a1, a2);
1649 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
1650 case INDEX_op_mov_i64:
1651 case INDEX_op_call: /* Always emitted via tcg_out_call. */
1652 case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */
1653 case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */
1655 g_assert_not_reached();
1659 static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
1662 case INDEX_op_goto_ptr:
1665 case INDEX_op_ld8u_i32:
1666 case INDEX_op_ld8s_i32:
1667 case INDEX_op_ld16u_i32:
1668 case INDEX_op_ld16s_i32:
1669 case INDEX_op_ld_i32:
1670 case INDEX_op_not_i32:
1671 case INDEX_op_neg_i32:
1672 case INDEX_op_ld8u_i64:
1673 case INDEX_op_ld8s_i64:
1674 case INDEX_op_ld16u_i64:
1675 case INDEX_op_ld16s_i64:
1676 case INDEX_op_ld32s_i64:
1677 case INDEX_op_ld32u_i64:
1678 case INDEX_op_ld_i64:
1679 case INDEX_op_not_i64:
1680 case INDEX_op_neg_i64:
1681 case INDEX_op_ext8u_i32:
1682 case INDEX_op_ext8u_i64:
1683 case INDEX_op_ext16u_i32:
1684 case INDEX_op_ext16u_i64:
1685 case INDEX_op_ext32u_i64:
1686 case INDEX_op_extu_i32_i64:
1687 case INDEX_op_ext8s_i32:
1688 case INDEX_op_ext8s_i64:
1689 case INDEX_op_ext16s_i32:
1690 case INDEX_op_ext16s_i64:
1691 case INDEX_op_ext32s_i64:
1692 case INDEX_op_extrl_i64_i32:
1693 case INDEX_op_extrh_i64_i32:
1694 case INDEX_op_ext_i32_i64:
1695 return C_O1_I1(r, r);
1697 case INDEX_op_st8_i32:
1698 case INDEX_op_st16_i32:
1699 case INDEX_op_st_i32:
1700 case INDEX_op_st8_i64:
1701 case INDEX_op_st16_i64:
1702 case INDEX_op_st32_i64:
1703 case INDEX_op_st_i64:
1704 return C_O0_I2(rZ, r);
1706 case INDEX_op_add_i32:
1707 case INDEX_op_and_i32:
1708 case INDEX_op_or_i32:
1709 case INDEX_op_xor_i32:
1710 case INDEX_op_add_i64:
1711 case INDEX_op_and_i64:
1712 case INDEX_op_or_i64:
1713 case INDEX_op_xor_i64:
1714 return C_O1_I2(r, r, rI);
1716 case INDEX_op_sub_i32:
1717 case INDEX_op_sub_i64:
1718 return C_O1_I2(r, rZ, rN);
1720 case INDEX_op_mul_i32:
1721 case INDEX_op_mulsh_i32:
1722 case INDEX_op_muluh_i32:
1723 case INDEX_op_div_i32:
1724 case INDEX_op_divu_i32:
1725 case INDEX_op_rem_i32:
1726 case INDEX_op_remu_i32:
1727 case INDEX_op_setcond_i32:
1728 case INDEX_op_mul_i64:
1729 case INDEX_op_mulsh_i64:
1730 case INDEX_op_muluh_i64:
1731 case INDEX_op_div_i64:
1732 case INDEX_op_divu_i64:
1733 case INDEX_op_rem_i64:
1734 case INDEX_op_remu_i64:
1735 case INDEX_op_setcond_i64:
1736 return C_O1_I2(r, rZ, rZ);
1738 case INDEX_op_shl_i32:
1739 case INDEX_op_shr_i32:
1740 case INDEX_op_sar_i32:
1741 case INDEX_op_shl_i64:
1742 case INDEX_op_shr_i64:
1743 case INDEX_op_sar_i64:
1744 return C_O1_I2(r, r, ri);
1746 case INDEX_op_brcond_i32:
1747 case INDEX_op_brcond_i64:
1748 return C_O0_I2(rZ, rZ);
1750 case INDEX_op_add2_i32:
1751 case INDEX_op_add2_i64:
1752 case INDEX_op_sub2_i32:
1753 case INDEX_op_sub2_i64:
1754 return C_O2_I4(r, r, rZ, rZ, rM, rM);
1756 case INDEX_op_brcond2_i32:
1757 return C_O0_I4(rZ, rZ, rZ, rZ);
1759 case INDEX_op_setcond2_i32:
1760 return C_O1_I4(r, rZ, rZ, rZ, rZ);
1762 case INDEX_op_qemu_ld_i32:
1763 return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
1764 ? C_O1_I1(r, L) : C_O1_I2(r, L, L));
1765 case INDEX_op_qemu_st_i32:
1766 return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
1767 ? C_O0_I2(LZ, L) : C_O0_I3(LZ, L, L));
1768 case INDEX_op_qemu_ld_i64:
1769 return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L)
1770 : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O2_I1(r, r, L)
1771 : C_O2_I2(r, r, L, L));
1772 case INDEX_op_qemu_st_i64:
1773 return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(LZ, L)
1774 : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O0_I3(LZ, LZ, L)
1775 : C_O0_I4(LZ, LZ, L, L));
1778 g_assert_not_reached();
1782 static const int tcg_target_callee_save_regs[] = {
1783 TCG_REG_S0, /* used for the global env (TCG_AREG0) */
1795 TCG_REG_RA, /* should be last for ABI compliance */
1798 /* Stack frame parameters. */
1799 #define REG_SIZE (TCG_TARGET_REG_BITS / 8)
1800 #define SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * REG_SIZE)
1801 #define TEMP_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1802 #define FRAME_SIZE ((TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE + SAVE_SIZE \
1803 + TCG_TARGET_STACK_ALIGN - 1) \
1804 & -TCG_TARGET_STACK_ALIGN)
1805 #define SAVE_OFS (TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE)
1807 /* We're expecting to be able to use an immediate for frame allocation. */
1808 QEMU_BUILD_BUG_ON(FRAME_SIZE > 0x7ff);
1810 /* Generate global QEMU prologue and epilogue code */
1811 static void tcg_target_qemu_prologue(TCGContext *s)
1815 tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE, TEMP_SIZE);
1818 tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, -FRAME_SIZE);
1819 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
1820 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1821 TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
1824 #if !defined(CONFIG_SOFTMMU)
1825 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
1826 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1829 /* Call generated code */
1830 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1831 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, tcg_target_call_iarg_regs[1], 0);
1833 /* Return path for goto_ptr. Set return value to 0 */
1834 tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
1835 tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_A0, TCG_REG_ZERO);
1838 tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
1839 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
1840 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1841 TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
1844 tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, FRAME_SIZE);
1845 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_RA, 0);
1848 static void tcg_target_init(TCGContext *s)
1850 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
1851 if (TCG_TARGET_REG_BITS == 64) {
1852 tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
1855 tcg_target_call_clobber_regs = -1u;
1856 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S0);
1857 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S1);
1858 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S2);
1859 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S3);
1860 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S4);
1861 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S5);
1862 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S6);
1863 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S7);
1864 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S8);
1865 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S9);
1866 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S10);
1867 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S11);
1869 s->reserved_regs = 0;
1870 tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO);
1871 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
1872 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
1873 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
1874 tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
1875 tcg_regset_set_reg(s->reserved_regs, TCG_REG_GP);
1876 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TP);
1881 uint8_t fde_def_cfa[4];
1882 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2];
1885 #define ELF_HOST_MACHINE EM_RISCV
1887 static const DebugFrame debug_frame = {
1888 .h.cie.len = sizeof(DebugFrameCIE) - 4, /* length after .len member */
1891 .h.cie.code_align = 1,
1892 .h.cie.data_align = -(TCG_TARGET_REG_BITS / 8) & 0x7f, /* sleb128 */
1893 .h.cie.return_column = TCG_REG_RA,
1895 /* Total FDE size does not include the "len" member. */
1896 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
1899 12, TCG_REG_SP, /* DW_CFA_def_cfa sp, ... */
1900 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
1904 0x80 + 9, 12, /* DW_CFA_offset, s1, -96 */
1905 0x80 + 18, 11, /* DW_CFA_offset, s2, -88 */
1906 0x80 + 19, 10, /* DW_CFA_offset, s3, -80 */
1907 0x80 + 20, 9, /* DW_CFA_offset, s4, -72 */
1908 0x80 + 21, 8, /* DW_CFA_offset, s5, -64 */
1909 0x80 + 22, 7, /* DW_CFA_offset, s6, -56 */
1910 0x80 + 23, 6, /* DW_CFA_offset, s7, -48 */
1911 0x80 + 24, 5, /* DW_CFA_offset, s8, -40 */
1912 0x80 + 25, 4, /* DW_CFA_offset, s9, -32 */
1913 0x80 + 26, 3, /* DW_CFA_offset, s10, -24 */
1914 0x80 + 27, 2, /* DW_CFA_offset, s11, -16 */
1915 0x80 + 1 , 1, /* DW_CFA_offset, ra, -8 */
1919 void tcg_register_jit(const void *buf, size_t buf_size)
1921 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));