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 const int tcg_target_call_oarg_regs[] = {
121 #define TCG_CT_CONST_ZERO 0x100
122 #define TCG_CT_CONST_S12 0x200
123 #define TCG_CT_CONST_N12 0x400
124 #define TCG_CT_CONST_M12 0x800
126 #define ALL_GENERAL_REGS MAKE_64BIT_MASK(0, 32)
128 * For softmmu, we need to avoid conflicts with the first 5
129 * argument registers to call the helper. Some of these are
130 * also used for the tlb lookup.
132 #ifdef CONFIG_SOFTMMU
133 #define SOFTMMU_RESERVE_REGS MAKE_64BIT_MASK(TCG_REG_A0, 5)
135 #define SOFTMMU_RESERVE_REGS 0
139 static inline tcg_target_long sextreg(tcg_target_long val, int pos, int len)
141 if (TCG_TARGET_REG_BITS == 32) {
142 return sextract32(val, pos, len);
144 return sextract64(val, pos, len);
148 /* test if a constant matches the constraint */
149 static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
151 if (ct & TCG_CT_CONST) {
154 if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
158 * Sign extended from 12 bits: [-0x800, 0x7ff].
159 * Used for most arithmetic, as this is the isa field.
161 if ((ct & TCG_CT_CONST_S12) && val >= -0x800 && val <= 0x7ff) {
165 * Sign extended from 12 bits, negated: [-0x7ff, 0x800].
166 * Used for subtraction, where a constant must be handled by ADDI.
168 if ((ct & TCG_CT_CONST_N12) && val >= -0x7ff && val <= 0x800) {
172 * Sign extended from 12 bits, +/- matching: [-0x7ff, 0x7ff].
173 * Used by addsub2, which may need the negative operation,
174 * and requires the modified constant to be representable.
176 if ((ct & TCG_CT_CONST_M12) && val >= -0x7ff && val <= 0x7ff) {
183 * RISC-V Base ISA opcodes (IM)
199 OPC_DIVU = 0x2005033,
211 OPC_MULH = 0x2001033,
212 OPC_MULHSU = 0x2002033,
213 OPC_MULHU = 0x2003033,
217 OPC_REMU = 0x2007033,
227 OPC_SRA = 0x40005033,
228 OPC_SRAI = 0x40005013,
231 OPC_SUB = 0x40000033,
236 #if TCG_TARGET_REG_BITS == 64
239 OPC_DIVUW = 0x200503b,
240 OPC_DIVW = 0x200403b,
241 OPC_MULW = 0x200003b,
242 OPC_REMUW = 0x200703b,
243 OPC_REMW = 0x200603b,
246 OPC_SRAIW = 0x4000501b,
247 OPC_SRAW = 0x4000503b,
250 OPC_SUBW = 0x4000003b,
252 /* Simplify code throughout by defining aliases for RV32. */
253 OPC_ADDIW = OPC_ADDI,
255 OPC_DIVUW = OPC_DIVU,
258 OPC_REMUW = OPC_REMU,
260 OPC_SLLIW = OPC_SLLI,
262 OPC_SRAIW = OPC_SRAI,
264 OPC_SRLIW = OPC_SRLI,
269 OPC_FENCE = 0x0000000f,
270 OPC_NOP = OPC_ADDI, /* nop = addi r0,r0,0 */
274 * RISC-V immediate and instruction encoders (excludes 16-bit RVC)
279 static int32_t encode_r(RISCVInsn opc, TCGReg rd, TCGReg rs1, TCGReg rs2)
281 return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20;
286 static int32_t encode_imm12(uint32_t imm)
288 return (imm & 0xfff) << 20;
291 static int32_t encode_i(RISCVInsn opc, TCGReg rd, TCGReg rs1, uint32_t imm)
293 return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | encode_imm12(imm);
298 static int32_t encode_simm12(uint32_t imm)
302 ret |= (imm & 0xFE0) << 20;
303 ret |= (imm & 0x1F) << 7;
308 static int32_t encode_s(RISCVInsn opc, TCGReg rs1, TCGReg rs2, uint32_t imm)
310 return opc | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20 | encode_simm12(imm);
315 static int32_t encode_sbimm12(uint32_t imm)
319 ret |= (imm & 0x1000) << 19;
320 ret |= (imm & 0x7e0) << 20;
321 ret |= (imm & 0x1e) << 7;
322 ret |= (imm & 0x800) >> 4;
327 static int32_t encode_sb(RISCVInsn opc, TCGReg rs1, TCGReg rs2, uint32_t imm)
329 return opc | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20 | encode_sbimm12(imm);
334 static int32_t encode_uimm20(uint32_t imm)
336 return imm & 0xfffff000;
339 static int32_t encode_u(RISCVInsn opc, TCGReg rd, uint32_t imm)
341 return opc | (rd & 0x1f) << 7 | encode_uimm20(imm);
346 static int32_t encode_ujimm20(uint32_t imm)
350 ret |= (imm & 0x0007fe) << (21 - 1);
351 ret |= (imm & 0x000800) << (20 - 11);
352 ret |= (imm & 0x0ff000) << (12 - 12);
353 ret |= (imm & 0x100000) << (31 - 20);
358 static int32_t encode_uj(RISCVInsn opc, TCGReg rd, uint32_t imm)
360 return opc | (rd & 0x1f) << 7 | encode_ujimm20(imm);
364 * RISC-V instruction emitters
367 static void tcg_out_opc_reg(TCGContext *s, RISCVInsn opc,
368 TCGReg rd, TCGReg rs1, TCGReg rs2)
370 tcg_out32(s, encode_r(opc, rd, rs1, rs2));
373 static void tcg_out_opc_imm(TCGContext *s, RISCVInsn opc,
374 TCGReg rd, TCGReg rs1, TCGArg imm)
376 tcg_out32(s, encode_i(opc, rd, rs1, imm));
379 static void tcg_out_opc_store(TCGContext *s, RISCVInsn opc,
380 TCGReg rs1, TCGReg rs2, uint32_t imm)
382 tcg_out32(s, encode_s(opc, rs1, rs2, imm));
385 static void tcg_out_opc_branch(TCGContext *s, RISCVInsn opc,
386 TCGReg rs1, TCGReg rs2, uint32_t imm)
388 tcg_out32(s, encode_sb(opc, rs1, rs2, imm));
391 static void tcg_out_opc_upper(TCGContext *s, RISCVInsn opc,
392 TCGReg rd, uint32_t imm)
394 tcg_out32(s, encode_u(opc, rd, imm));
397 static void tcg_out_opc_jump(TCGContext *s, RISCVInsn opc,
398 TCGReg rd, uint32_t imm)
400 tcg_out32(s, encode_uj(opc, rd, imm));
403 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
406 for (i = 0; i < count; ++i) {
415 static bool reloc_sbimm12(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
417 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
418 intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
420 tcg_debug_assert((offset & 1) == 0);
421 if (offset == sextreg(offset, 0, 12)) {
422 *src_rw |= encode_sbimm12(offset);
429 static bool reloc_jimm20(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
431 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
432 intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
434 tcg_debug_assert((offset & 1) == 0);
435 if (offset == sextreg(offset, 0, 20)) {
436 *src_rw |= encode_ujimm20(offset);
443 static bool reloc_call(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
445 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
446 intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
447 int32_t lo = sextreg(offset, 0, 12);
448 int32_t hi = offset - lo;
450 if (offset == hi + lo) {
451 src_rw[0] |= encode_uimm20(hi);
452 src_rw[1] |= encode_imm12(lo);
459 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
460 intptr_t value, intptr_t addend)
462 tcg_debug_assert(addend == 0);
465 return reloc_sbimm12(code_ptr, (tcg_insn_unit *)value);
467 return reloc_jimm20(code_ptr, (tcg_insn_unit *)value);
469 return reloc_call(code_ptr, (tcg_insn_unit *)value);
471 g_assert_not_reached();
479 static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
487 tcg_out_opc_imm(s, OPC_ADDI, ret, arg, 0);
490 g_assert_not_reached();
495 static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
498 tcg_target_long lo, hi, tmp;
501 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
505 lo = sextreg(val, 0, 12);
507 tcg_out_opc_imm(s, OPC_ADDI, rd, TCG_REG_ZERO, lo);
512 if (TCG_TARGET_REG_BITS == 32 || val == (int32_t)val) {
513 tcg_out_opc_upper(s, OPC_LUI, rd, hi);
515 tcg_out_opc_imm(s, OPC_ADDIW, rd, rd, lo);
520 /* We can only be here if TCG_TARGET_REG_BITS != 32 */
521 tmp = tcg_pcrel_diff(s, (void *)val);
522 if (tmp == (int32_t)tmp) {
523 tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
524 tcg_out_opc_imm(s, OPC_ADDI, rd, rd, 0);
525 ret = reloc_call(s->code_ptr - 2, (const tcg_insn_unit *)val);
526 tcg_debug_assert(ret == true);
530 /* Look for a single 20-bit section. */
533 if (tmp == sextreg(tmp, 0, 20)) {
534 tcg_out_opc_upper(s, OPC_LUI, rd, tmp << 12);
536 tcg_out_opc_imm(s, OPC_SLLI, rd, rd, shift - 12);
538 tcg_out_opc_imm(s, OPC_SRAI, rd, rd, 12 - shift);
543 /* Look for a few high zero bits, with lots of bits set in the middle. */
546 if (tmp == sextreg(tmp, 12, 20) << 12) {
547 tcg_out_opc_upper(s, OPC_LUI, rd, tmp);
548 tcg_out_opc_imm(s, OPC_SRLI, rd, rd, shift);
550 } else if (tmp == sextreg(tmp, 0, 12)) {
551 tcg_out_opc_imm(s, OPC_ADDI, rd, TCG_REG_ZERO, tmp);
552 tcg_out_opc_imm(s, OPC_SRLI, rd, rd, shift);
556 /* Drop into the constant pool. */
557 new_pool_label(s, val, R_RISCV_CALL, s->code_ptr, 0);
558 tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
559 tcg_out_opc_imm(s, OPC_LD, rd, rd, 0);
562 static void tcg_out_ext8u(TCGContext *s, TCGReg ret, TCGReg arg)
564 tcg_out_opc_imm(s, OPC_ANDI, ret, arg, 0xff);
567 static void tcg_out_ext16u(TCGContext *s, TCGReg ret, TCGReg arg)
569 tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 16);
570 tcg_out_opc_imm(s, OPC_SRLIW, ret, ret, 16);
573 static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg)
575 tcg_out_opc_imm(s, OPC_SLLI, ret, arg, 32);
576 tcg_out_opc_imm(s, OPC_SRLI, ret, ret, 32);
579 static void tcg_out_ext8s(TCGContext *s, TCGReg ret, TCGReg arg)
581 tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 24);
582 tcg_out_opc_imm(s, OPC_SRAIW, ret, ret, 24);
585 static void tcg_out_ext16s(TCGContext *s, TCGReg ret, TCGReg arg)
587 tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 16);
588 tcg_out_opc_imm(s, OPC_SRAIW, ret, ret, 16);
591 static void tcg_out_ext32s(TCGContext *s, TCGReg ret, TCGReg arg)
593 tcg_out_opc_imm(s, OPC_ADDIW, ret, arg, 0);
596 static void tcg_out_ldst(TCGContext *s, RISCVInsn opc, TCGReg data,
597 TCGReg addr, intptr_t offset)
599 intptr_t imm12 = sextreg(offset, 0, 12);
601 if (offset != imm12) {
602 intptr_t diff = tcg_pcrel_diff(s, (void *)offset);
604 if (addr == TCG_REG_ZERO && diff == (int32_t)diff) {
605 imm12 = sextreg(diff, 0, 12);
606 tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP2, diff - imm12);
608 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP2, offset - imm12);
609 if (addr != TCG_REG_ZERO) {
610 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, addr);
621 tcg_out_opc_store(s, opc, addr, data, imm12);
630 tcg_out_opc_imm(s, opc, data, addr, imm12);
633 g_assert_not_reached();
637 static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
638 TCGReg arg1, intptr_t arg2)
640 bool is32bit = (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I32);
641 tcg_out_ldst(s, is32bit ? OPC_LW : OPC_LD, arg, arg1, arg2);
644 static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
645 TCGReg arg1, intptr_t arg2)
647 bool is32bit = (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I32);
648 tcg_out_ldst(s, is32bit ? OPC_SW : OPC_SD, arg, arg1, arg2);
651 static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
652 TCGReg base, intptr_t ofs)
655 tcg_out_st(s, type, TCG_REG_ZERO, base, ofs);
661 static void tcg_out_addsub2(TCGContext *s,
662 TCGReg rl, TCGReg rh,
663 TCGReg al, TCGReg ah,
664 TCGArg bl, TCGArg bh,
665 bool cbl, bool cbh, bool is_sub, bool is32bit)
667 const RISCVInsn opc_add = is32bit ? OPC_ADDW : OPC_ADD;
668 const RISCVInsn opc_addi = is32bit ? OPC_ADDIW : OPC_ADDI;
669 const RISCVInsn opc_sub = is32bit ? OPC_SUBW : OPC_SUB;
670 TCGReg th = TCG_REG_TMP1;
672 /* If we have a negative constant such that negating it would
673 make the high part zero, we can (usually) eliminate one insn. */
674 if (cbl && cbh && bh == -1 && bl != 0) {
680 /* By operating on the high part first, we get to use the final
681 carry operation to move back from the temporary. */
683 tcg_out_opc_reg(s, (is_sub ? opc_sub : opc_add), th, ah, bh);
684 } else if (bh != 0 || ah == rl) {
685 tcg_out_opc_imm(s, opc_addi, th, ah, (is_sub ? -bh : bh));
690 /* Note that tcg optimization should eliminate the bl == 0 case. */
693 tcg_out_opc_imm(s, OPC_SLTIU, TCG_REG_TMP0, al, bl);
694 tcg_out_opc_imm(s, opc_addi, rl, al, -bl);
696 tcg_out_opc_reg(s, OPC_SLTU, TCG_REG_TMP0, al, bl);
697 tcg_out_opc_reg(s, opc_sub, rl, al, bl);
699 tcg_out_opc_reg(s, opc_sub, rh, th, TCG_REG_TMP0);
702 tcg_out_opc_imm(s, opc_addi, rl, al, bl);
703 tcg_out_opc_imm(s, OPC_SLTIU, TCG_REG_TMP0, rl, bl);
704 } else if (al == bl) {
706 * If the input regs overlap, this is a simple doubling
707 * and carry-out is the input msb. This special case is
708 * required when the output reg overlaps the input,
709 * but we might as well use it always.
711 tcg_out_opc_imm(s, OPC_SLTI, TCG_REG_TMP0, al, 0);
712 tcg_out_opc_reg(s, opc_add, rl, al, al);
714 tcg_out_opc_reg(s, opc_add, rl, al, bl);
715 tcg_out_opc_reg(s, OPC_SLTU, TCG_REG_TMP0,
716 rl, (rl == bl ? al : bl));
718 tcg_out_opc_reg(s, opc_add, rh, th, TCG_REG_TMP0);
722 static const struct {
725 } tcg_brcond_to_riscv[] = {
726 [TCG_COND_EQ] = { OPC_BEQ, false },
727 [TCG_COND_NE] = { OPC_BNE, false },
728 [TCG_COND_LT] = { OPC_BLT, false },
729 [TCG_COND_GE] = { OPC_BGE, false },
730 [TCG_COND_LE] = { OPC_BGE, true },
731 [TCG_COND_GT] = { OPC_BLT, true },
732 [TCG_COND_LTU] = { OPC_BLTU, false },
733 [TCG_COND_GEU] = { OPC_BGEU, false },
734 [TCG_COND_LEU] = { OPC_BGEU, true },
735 [TCG_COND_GTU] = { OPC_BLTU, true }
738 static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
739 TCGReg arg2, TCGLabel *l)
741 RISCVInsn op = tcg_brcond_to_riscv[cond].op;
743 tcg_debug_assert(op != 0);
745 if (tcg_brcond_to_riscv[cond].swap) {
751 tcg_out_reloc(s, s->code_ptr, R_RISCV_BRANCH, l, 0);
752 tcg_out_opc_branch(s, op, arg1, arg2, 0);
755 static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret,
756 TCGReg arg1, TCGReg arg2)
760 tcg_out_opc_reg(s, OPC_SUB, ret, arg1, arg2);
761 tcg_out_opc_imm(s, OPC_SLTIU, ret, ret, 1);
764 tcg_out_opc_reg(s, OPC_SUB, ret, arg1, arg2);
765 tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, ret);
768 tcg_out_opc_reg(s, OPC_SLT, ret, arg1, arg2);
771 tcg_out_opc_reg(s, OPC_SLT, ret, arg1, arg2);
772 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
775 tcg_out_opc_reg(s, OPC_SLT, ret, arg2, arg1);
776 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
779 tcg_out_opc_reg(s, OPC_SLT, ret, arg2, arg1);
782 tcg_out_opc_reg(s, OPC_SLTU, ret, arg1, arg2);
785 tcg_out_opc_reg(s, OPC_SLTU, ret, arg1, arg2);
786 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
789 tcg_out_opc_reg(s, OPC_SLTU, ret, arg2, arg1);
790 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
793 tcg_out_opc_reg(s, OPC_SLTU, ret, arg2, arg1);
796 g_assert_not_reached();
801 static void tcg_out_brcond2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
802 TCGReg bl, TCGReg bh, TCGLabel *l)
805 g_assert_not_reached();
808 static void tcg_out_setcond2(TCGContext *s, TCGCond cond, TCGReg ret,
809 TCGReg al, TCGReg ah, TCGReg bl, TCGReg bh)
812 g_assert_not_reached();
815 static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
817 TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
818 ptrdiff_t offset = tcg_pcrel_diff(s, arg);
821 tcg_debug_assert((offset & 1) == 0);
822 if (offset == sextreg(offset, 0, 20)) {
823 /* short jump: -2097150 to 2097152 */
824 tcg_out_opc_jump(s, OPC_JAL, link, offset);
825 } else if (TCG_TARGET_REG_BITS == 32 || offset == (int32_t)offset) {
826 /* long jump: -2147483646 to 2147483648 */
827 tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
828 tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
829 ret = reloc_call(s->code_ptr - 2, arg);
830 tcg_debug_assert(ret == true);
831 } else if (TCG_TARGET_REG_BITS == 64) {
832 /* far jump: 64-bit */
833 tcg_target_long imm = sextreg((tcg_target_long)arg, 0, 12);
834 tcg_target_long base = (tcg_target_long)arg - imm;
835 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, base);
836 tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, imm);
838 g_assert_not_reached();
842 static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg,
843 const TCGHelperInfo *info)
845 tcg_out_call_int(s, arg, false);
848 static void tcg_out_mb(TCGContext *s, TCGArg a0)
850 tcg_insn_unit insn = OPC_FENCE;
852 if (a0 & TCG_MO_LD_LD) {
855 if (a0 & TCG_MO_ST_LD) {
858 if (a0 & TCG_MO_LD_ST) {
861 if (a0 & TCG_MO_ST_ST) {
871 #if defined(CONFIG_SOFTMMU)
872 /* helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
873 * MemOpIdx oi, uintptr_t ra)
875 static void * const qemu_ld_helpers[MO_SSIZE + 1] = {
876 [MO_UB] = helper_ret_ldub_mmu,
877 [MO_SB] = helper_ret_ldsb_mmu,
879 [MO_UW] = helper_be_lduw_mmu,
880 [MO_SW] = helper_be_ldsw_mmu,
881 [MO_UL] = helper_be_ldul_mmu,
882 #if TCG_TARGET_REG_BITS == 64
883 [MO_SL] = helper_be_ldsl_mmu,
885 [MO_UQ] = helper_be_ldq_mmu,
887 [MO_UW] = helper_le_lduw_mmu,
888 [MO_SW] = helper_le_ldsw_mmu,
889 [MO_UL] = helper_le_ldul_mmu,
890 #if TCG_TARGET_REG_BITS == 64
891 [MO_SL] = helper_le_ldsl_mmu,
893 [MO_UQ] = helper_le_ldq_mmu,
897 /* helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
898 * uintxx_t val, MemOpIdx oi,
901 static void * const qemu_st_helpers[MO_SIZE + 1] = {
902 [MO_8] = helper_ret_stb_mmu,
904 [MO_16] = helper_be_stw_mmu,
905 [MO_32] = helper_be_stl_mmu,
906 [MO_64] = helper_be_stq_mmu,
908 [MO_16] = helper_le_stw_mmu,
909 [MO_32] = helper_le_stl_mmu,
910 [MO_64] = helper_le_stq_mmu,
914 /* We don't support oversize guests */
915 QEMU_BUILD_BUG_ON(TCG_TARGET_REG_BITS < TARGET_LONG_BITS);
917 /* We expect to use a 12-bit negative offset from ENV. */
918 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
919 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 11));
921 static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
923 tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
924 bool ok = reloc_jimm20(s->code_ptr - 1, target);
925 tcg_debug_assert(ok);
928 static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addrl,
929 TCGReg addrh, MemOpIdx oi,
930 tcg_insn_unit **label_ptr, bool is_load)
932 MemOp opc = get_memop(oi);
933 unsigned s_bits = opc & MO_SIZE;
934 unsigned a_bits = get_alignment_bits(opc);
935 tcg_target_long compare_mask;
936 int mem_index = get_mmuidx(oi);
937 int fast_ofs = TLB_MASK_TABLE_OFS(mem_index);
938 int mask_ofs = fast_ofs + offsetof(CPUTLBDescFast, mask);
939 int table_ofs = fast_ofs + offsetof(CPUTLBDescFast, table);
940 TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
942 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, mask_base, mask_ofs);
943 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, table_base, table_ofs);
945 tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP2, addrl,
946 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
947 tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP0);
948 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP1);
950 /* Load the tlb comparator and the addend. */
951 tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP0, TCG_REG_TMP2,
952 is_load ? offsetof(CPUTLBEntry, addr_read)
953 : offsetof(CPUTLBEntry, addr_write));
954 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP2,
955 offsetof(CPUTLBEntry, addend));
957 /* We don't support unaligned accesses. */
958 if (a_bits < s_bits) {
961 /* Clear the non-page, non-alignment bits from the address. */
962 compare_mask = (tcg_target_long)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
963 if (compare_mask == sextreg(compare_mask, 0, 12)) {
964 tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addrl, compare_mask);
966 tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_TMP1, compare_mask);
967 tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP1, TCG_REG_TMP1, addrl);
970 /* Compare masked address with the TLB entry. */
971 label_ptr[0] = s->code_ptr;
972 tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP0, TCG_REG_TMP1, 0);
974 /* TLB Hit - translate address using addend. */
975 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
976 tcg_out_ext32u(s, TCG_REG_TMP0, addrl);
977 addrl = TCG_REG_TMP0;
979 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_REG_TMP2, addrl);
983 static void add_qemu_ldst_label(TCGContext *s, int is_ld, MemOpIdx oi,
985 TCGReg datalo, TCGReg datahi,
986 TCGReg addrlo, TCGReg addrhi,
987 void *raddr, tcg_insn_unit **label_ptr)
989 TCGLabelQemuLdst *label = new_ldst_label(s);
991 label->is_ld = is_ld;
994 label->datalo_reg = datalo;
995 label->datahi_reg = datahi;
996 label->addrlo_reg = addrlo;
997 label->addrhi_reg = addrhi;
998 label->raddr = tcg_splitwx_to_rx(raddr);
999 label->label_ptr[0] = label_ptr[0];
1002 static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1004 MemOpIdx oi = l->oi;
1005 MemOp opc = get_memop(oi);
1006 TCGReg a0 = tcg_target_call_iarg_regs[0];
1007 TCGReg a1 = tcg_target_call_iarg_regs[1];
1008 TCGReg a2 = tcg_target_call_iarg_regs[2];
1009 TCGReg a3 = tcg_target_call_iarg_regs[3];
1011 /* We don't support oversize guests */
1012 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1013 g_assert_not_reached();
1016 /* resolve label address */
1017 if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1021 /* call load helper */
1022 tcg_out_mov(s, TCG_TYPE_PTR, a0, TCG_AREG0);
1023 tcg_out_mov(s, TCG_TYPE_PTR, a1, l->addrlo_reg);
1024 tcg_out_movi(s, TCG_TYPE_PTR, a2, oi);
1025 tcg_out_movi(s, TCG_TYPE_PTR, a3, (tcg_target_long)l->raddr);
1027 tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SSIZE], false);
1028 tcg_out_mov(s, (opc & MO_SIZE) == MO_64, l->datalo_reg, a0);
1030 tcg_out_goto(s, l->raddr);
1034 static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1036 MemOpIdx oi = l->oi;
1037 MemOp opc = get_memop(oi);
1038 MemOp s_bits = opc & MO_SIZE;
1039 TCGReg a0 = tcg_target_call_iarg_regs[0];
1040 TCGReg a1 = tcg_target_call_iarg_regs[1];
1041 TCGReg a2 = tcg_target_call_iarg_regs[2];
1042 TCGReg a3 = tcg_target_call_iarg_regs[3];
1043 TCGReg a4 = tcg_target_call_iarg_regs[4];
1045 /* We don't support oversize guests */
1046 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1047 g_assert_not_reached();
1050 /* resolve label address */
1051 if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1055 /* call store helper */
1056 tcg_out_mov(s, TCG_TYPE_PTR, a0, TCG_AREG0);
1057 tcg_out_mov(s, TCG_TYPE_PTR, a1, l->addrlo_reg);
1058 tcg_out_mov(s, TCG_TYPE_PTR, a2, l->datalo_reg);
1061 tcg_out_ext8u(s, a2, a2);
1064 tcg_out_ext16u(s, a2, a2);
1069 tcg_out_movi(s, TCG_TYPE_PTR, a3, oi);
1070 tcg_out_movi(s, TCG_TYPE_PTR, a4, (tcg_target_long)l->raddr);
1072 tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE], false);
1074 tcg_out_goto(s, l->raddr);
1079 static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addr_reg,
1082 unsigned a_mask = (1 << a_bits) - 1;
1083 TCGLabelQemuLdst *l = new_ldst_label(s);
1086 l->addrlo_reg = addr_reg;
1088 /* We are expecting a_bits to max out at 7, so we can always use andi. */
1089 tcg_debug_assert(a_bits < 12);
1090 tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addr_reg, a_mask);
1092 l->label_ptr[0] = s->code_ptr;
1093 tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP1, TCG_REG_ZERO, 0);
1095 l->raddr = tcg_splitwx_to_rx(s->code_ptr);
1098 static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l)
1100 /* resolve label address */
1101 if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1105 tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_A1, l->addrlo_reg);
1106 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A0, TCG_AREG0);
1108 /* tail call, with the return address back inline. */
1109 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_RA, (uintptr_t)l->raddr);
1110 tcg_out_call_int(s, (const void *)(l->is_ld ? helper_unaligned_ld
1111 : helper_unaligned_st), true);
1115 static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1117 return tcg_out_fail_alignment(s, l);
1120 static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1122 return tcg_out_fail_alignment(s, l);
1125 #endif /* CONFIG_SOFTMMU */
1127 static void tcg_out_qemu_ld_direct(TCGContext *s, TCGReg lo, TCGReg hi,
1128 TCGReg base, MemOp opc, bool is_64)
1130 /* Byte swapping is left to middle-end expansion. */
1131 tcg_debug_assert((opc & MO_BSWAP) == 0);
1133 switch (opc & (MO_SSIZE)) {
1135 tcg_out_opc_imm(s, OPC_LBU, lo, base, 0);
1138 tcg_out_opc_imm(s, OPC_LB, lo, base, 0);
1141 tcg_out_opc_imm(s, OPC_LHU, lo, base, 0);
1144 tcg_out_opc_imm(s, OPC_LH, lo, base, 0);
1147 if (TCG_TARGET_REG_BITS == 64 && is_64) {
1148 tcg_out_opc_imm(s, OPC_LWU, lo, base, 0);
1153 tcg_out_opc_imm(s, OPC_LW, lo, base, 0);
1156 /* Prefer to load from offset 0 first, but allow for overlap. */
1157 if (TCG_TARGET_REG_BITS == 64) {
1158 tcg_out_opc_imm(s, OPC_LD, lo, base, 0);
1159 } else if (lo != base) {
1160 tcg_out_opc_imm(s, OPC_LW, lo, base, 0);
1161 tcg_out_opc_imm(s, OPC_LW, hi, base, 4);
1163 tcg_out_opc_imm(s, OPC_LW, hi, base, 4);
1164 tcg_out_opc_imm(s, OPC_LW, lo, base, 0);
1168 g_assert_not_reached();
1172 static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
1174 TCGReg addr_regl, addr_regh __attribute__((unused));
1175 TCGReg data_regl, data_regh;
1178 #if defined(CONFIG_SOFTMMU)
1179 tcg_insn_unit *label_ptr[1];
1185 data_regl = *args++;
1186 data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1187 addr_regl = *args++;
1188 addr_regh = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1190 opc = get_memop(oi);
1192 #if defined(CONFIG_SOFTMMU)
1193 base = tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 1);
1194 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64);
1195 add_qemu_ldst_label(s, 1, oi,
1196 (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32),
1197 data_regl, data_regh, addr_regl, addr_regh,
1198 s->code_ptr, label_ptr);
1200 a_bits = get_alignment_bits(opc);
1202 tcg_out_test_alignment(s, true, addr_regl, a_bits);
1205 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1206 tcg_out_ext32u(s, TCG_REG_TMP0, base);
1207 base = TCG_REG_TMP0;
1209 if (guest_base != 0) {
1210 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_GUEST_BASE_REG, base);
1211 base = TCG_REG_TMP0;
1213 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64);
1217 static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg lo, TCGReg hi,
1218 TCGReg base, MemOp opc)
1220 /* Byte swapping is left to middle-end expansion. */
1221 tcg_debug_assert((opc & MO_BSWAP) == 0);
1223 switch (opc & (MO_SSIZE)) {
1225 tcg_out_opc_store(s, OPC_SB, base, lo, 0);
1228 tcg_out_opc_store(s, OPC_SH, base, lo, 0);
1231 tcg_out_opc_store(s, OPC_SW, base, lo, 0);
1234 if (TCG_TARGET_REG_BITS == 64) {
1235 tcg_out_opc_store(s, OPC_SD, base, lo, 0);
1237 tcg_out_opc_store(s, OPC_SW, base, lo, 0);
1238 tcg_out_opc_store(s, OPC_SW, base, hi, 4);
1242 g_assert_not_reached();
1246 static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
1248 TCGReg addr_regl, addr_regh __attribute__((unused));
1249 TCGReg data_regl, data_regh;
1252 #if defined(CONFIG_SOFTMMU)
1253 tcg_insn_unit *label_ptr[1];
1259 data_regl = *args++;
1260 data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1261 addr_regl = *args++;
1262 addr_regh = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1264 opc = get_memop(oi);
1266 #if defined(CONFIG_SOFTMMU)
1267 base = tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 0);
1268 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc);
1269 add_qemu_ldst_label(s, 0, oi,
1270 (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32),
1271 data_regl, data_regh, addr_regl, addr_regh,
1272 s->code_ptr, label_ptr);
1274 a_bits = get_alignment_bits(opc);
1276 tcg_out_test_alignment(s, false, addr_regl, a_bits);
1279 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1280 tcg_out_ext32u(s, TCG_REG_TMP0, base);
1281 base = TCG_REG_TMP0;
1283 if (guest_base != 0) {
1284 tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_GUEST_BASE_REG, base);
1285 base = TCG_REG_TMP0;
1287 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc);
1291 static const tcg_insn_unit *tb_ret_addr;
1293 static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1295 /* Reuse the zeroing that exists for goto_ptr. */
1297 tcg_out_call_int(s, tcg_code_gen_epilogue, true);
1299 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A0, a0);
1300 tcg_out_call_int(s, tb_ret_addr, true);
1304 static void tcg_out_goto_tb(TCGContext *s, int which)
1306 /* Direct branch will be patched by tb_target_set_jmp_target. */
1307 set_jmp_insn_offset(s, which);
1308 tcg_out32(s, OPC_JAL);
1310 /* When branch is out of range, fall through to indirect. */
1311 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_REG_ZERO,
1312 get_jmp_target_addr(s, which));
1313 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_TMP0, 0);
1314 set_jmp_reset_offset(s, which);
1317 void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1318 uintptr_t jmp_rx, uintptr_t jmp_rw)
1320 uintptr_t addr = tb->jmp_target_addr[n];
1321 ptrdiff_t offset = addr - jmp_rx;
1324 /* Either directly branch, or fall through to indirect branch. */
1325 if (offset == sextreg(offset, 0, 20)) {
1326 insn = encode_uj(OPC_JAL, TCG_REG_ZERO, offset);
1330 qatomic_set((uint32_t *)jmp_rw, insn);
1331 flush_idcache_range(jmp_rx, jmp_rw, 4);
1334 static void tcg_out_op(TCGContext *s, TCGOpcode opc,
1335 const TCGArg args[TCG_MAX_OP_ARGS],
1336 const int const_args[TCG_MAX_OP_ARGS])
1338 TCGArg a0 = args[0];
1339 TCGArg a1 = args[1];
1340 TCGArg a2 = args[2];
1341 int c2 = const_args[2];
1344 case INDEX_op_goto_ptr:
1345 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, a0, 0);
1349 tcg_out_reloc(s, s->code_ptr, R_RISCV_JAL, arg_label(a0), 0);
1350 tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
1353 case INDEX_op_ld8u_i32:
1354 case INDEX_op_ld8u_i64:
1355 tcg_out_ldst(s, OPC_LBU, a0, a1, a2);
1357 case INDEX_op_ld8s_i32:
1358 case INDEX_op_ld8s_i64:
1359 tcg_out_ldst(s, OPC_LB, a0, a1, a2);
1361 case INDEX_op_ld16u_i32:
1362 case INDEX_op_ld16u_i64:
1363 tcg_out_ldst(s, OPC_LHU, a0, a1, a2);
1365 case INDEX_op_ld16s_i32:
1366 case INDEX_op_ld16s_i64:
1367 tcg_out_ldst(s, OPC_LH, a0, a1, a2);
1369 case INDEX_op_ld32u_i64:
1370 tcg_out_ldst(s, OPC_LWU, a0, a1, a2);
1372 case INDEX_op_ld_i32:
1373 case INDEX_op_ld32s_i64:
1374 tcg_out_ldst(s, OPC_LW, a0, a1, a2);
1376 case INDEX_op_ld_i64:
1377 tcg_out_ldst(s, OPC_LD, a0, a1, a2);
1380 case INDEX_op_st8_i32:
1381 case INDEX_op_st8_i64:
1382 tcg_out_ldst(s, OPC_SB, a0, a1, a2);
1384 case INDEX_op_st16_i32:
1385 case INDEX_op_st16_i64:
1386 tcg_out_ldst(s, OPC_SH, a0, a1, a2);
1388 case INDEX_op_st_i32:
1389 case INDEX_op_st32_i64:
1390 tcg_out_ldst(s, OPC_SW, a0, a1, a2);
1392 case INDEX_op_st_i64:
1393 tcg_out_ldst(s, OPC_SD, a0, a1, a2);
1396 case INDEX_op_add_i32:
1398 tcg_out_opc_imm(s, OPC_ADDIW, a0, a1, a2);
1400 tcg_out_opc_reg(s, OPC_ADDW, a0, a1, a2);
1403 case INDEX_op_add_i64:
1405 tcg_out_opc_imm(s, OPC_ADDI, a0, a1, a2);
1407 tcg_out_opc_reg(s, OPC_ADD, a0, a1, a2);
1411 case INDEX_op_sub_i32:
1413 tcg_out_opc_imm(s, OPC_ADDIW, a0, a1, -a2);
1415 tcg_out_opc_reg(s, OPC_SUBW, a0, a1, a2);
1418 case INDEX_op_sub_i64:
1420 tcg_out_opc_imm(s, OPC_ADDI, a0, a1, -a2);
1422 tcg_out_opc_reg(s, OPC_SUB, a0, a1, a2);
1426 case INDEX_op_and_i32:
1427 case INDEX_op_and_i64:
1429 tcg_out_opc_imm(s, OPC_ANDI, a0, a1, a2);
1431 tcg_out_opc_reg(s, OPC_AND, a0, a1, a2);
1435 case INDEX_op_or_i32:
1436 case INDEX_op_or_i64:
1438 tcg_out_opc_imm(s, OPC_ORI, a0, a1, a2);
1440 tcg_out_opc_reg(s, OPC_OR, a0, a1, a2);
1444 case INDEX_op_xor_i32:
1445 case INDEX_op_xor_i64:
1447 tcg_out_opc_imm(s, OPC_XORI, a0, a1, a2);
1449 tcg_out_opc_reg(s, OPC_XOR, a0, a1, a2);
1453 case INDEX_op_not_i32:
1454 case INDEX_op_not_i64:
1455 tcg_out_opc_imm(s, OPC_XORI, a0, a1, -1);
1458 case INDEX_op_neg_i32:
1459 tcg_out_opc_reg(s, OPC_SUBW, a0, TCG_REG_ZERO, a1);
1461 case INDEX_op_neg_i64:
1462 tcg_out_opc_reg(s, OPC_SUB, a0, TCG_REG_ZERO, a1);
1465 case INDEX_op_mul_i32:
1466 tcg_out_opc_reg(s, OPC_MULW, a0, a1, a2);
1468 case INDEX_op_mul_i64:
1469 tcg_out_opc_reg(s, OPC_MUL, a0, a1, a2);
1472 case INDEX_op_div_i32:
1473 tcg_out_opc_reg(s, OPC_DIVW, a0, a1, a2);
1475 case INDEX_op_div_i64:
1476 tcg_out_opc_reg(s, OPC_DIV, a0, a1, a2);
1479 case INDEX_op_divu_i32:
1480 tcg_out_opc_reg(s, OPC_DIVUW, a0, a1, a2);
1482 case INDEX_op_divu_i64:
1483 tcg_out_opc_reg(s, OPC_DIVU, a0, a1, a2);
1486 case INDEX_op_rem_i32:
1487 tcg_out_opc_reg(s, OPC_REMW, a0, a1, a2);
1489 case INDEX_op_rem_i64:
1490 tcg_out_opc_reg(s, OPC_REM, a0, a1, a2);
1493 case INDEX_op_remu_i32:
1494 tcg_out_opc_reg(s, OPC_REMUW, a0, a1, a2);
1496 case INDEX_op_remu_i64:
1497 tcg_out_opc_reg(s, OPC_REMU, a0, a1, a2);
1500 case INDEX_op_shl_i32:
1502 tcg_out_opc_imm(s, OPC_SLLIW, a0, a1, a2 & 0x1f);
1504 tcg_out_opc_reg(s, OPC_SLLW, a0, a1, a2);
1507 case INDEX_op_shl_i64:
1509 tcg_out_opc_imm(s, OPC_SLLI, a0, a1, a2 & 0x3f);
1511 tcg_out_opc_reg(s, OPC_SLL, a0, a1, a2);
1515 case INDEX_op_shr_i32:
1517 tcg_out_opc_imm(s, OPC_SRLIW, a0, a1, a2 & 0x1f);
1519 tcg_out_opc_reg(s, OPC_SRLW, a0, a1, a2);
1522 case INDEX_op_shr_i64:
1524 tcg_out_opc_imm(s, OPC_SRLI, a0, a1, a2 & 0x3f);
1526 tcg_out_opc_reg(s, OPC_SRL, a0, a1, a2);
1530 case INDEX_op_sar_i32:
1532 tcg_out_opc_imm(s, OPC_SRAIW, a0, a1, a2 & 0x1f);
1534 tcg_out_opc_reg(s, OPC_SRAW, a0, a1, a2);
1537 case INDEX_op_sar_i64:
1539 tcg_out_opc_imm(s, OPC_SRAI, a0, a1, a2 & 0x3f);
1541 tcg_out_opc_reg(s, OPC_SRA, a0, a1, a2);
1545 case INDEX_op_add2_i32:
1546 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1547 const_args[4], const_args[5], false, true);
1549 case INDEX_op_add2_i64:
1550 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1551 const_args[4], const_args[5], false, false);
1553 case INDEX_op_sub2_i32:
1554 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1555 const_args[4], const_args[5], true, true);
1557 case INDEX_op_sub2_i64:
1558 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
1559 const_args[4], const_args[5], true, false);
1562 case INDEX_op_brcond_i32:
1563 case INDEX_op_brcond_i64:
1564 tcg_out_brcond(s, a2, a0, a1, arg_label(args[3]));
1566 case INDEX_op_brcond2_i32:
1567 tcg_out_brcond2(s, args[4], a0, a1, a2, args[3], arg_label(args[5]));
1570 case INDEX_op_setcond_i32:
1571 case INDEX_op_setcond_i64:
1572 tcg_out_setcond(s, args[3], a0, a1, a2);
1574 case INDEX_op_setcond2_i32:
1575 tcg_out_setcond2(s, args[5], a0, a1, a2, args[3], args[4]);
1578 case INDEX_op_qemu_ld_i32:
1579 tcg_out_qemu_ld(s, args, false);
1581 case INDEX_op_qemu_ld_i64:
1582 tcg_out_qemu_ld(s, args, true);
1584 case INDEX_op_qemu_st_i32:
1585 tcg_out_qemu_st(s, args, false);
1587 case INDEX_op_qemu_st_i64:
1588 tcg_out_qemu_st(s, args, true);
1591 case INDEX_op_ext8u_i32:
1592 case INDEX_op_ext8u_i64:
1593 tcg_out_ext8u(s, a0, a1);
1596 case INDEX_op_ext16u_i32:
1597 case INDEX_op_ext16u_i64:
1598 tcg_out_ext16u(s, a0, a1);
1601 case INDEX_op_ext32u_i64:
1602 case INDEX_op_extu_i32_i64:
1603 tcg_out_ext32u(s, a0, a1);
1606 case INDEX_op_ext8s_i32:
1607 case INDEX_op_ext8s_i64:
1608 tcg_out_ext8s(s, a0, a1);
1611 case INDEX_op_ext16s_i32:
1612 case INDEX_op_ext16s_i64:
1613 tcg_out_ext16s(s, a0, a1);
1616 case INDEX_op_ext32s_i64:
1617 case INDEX_op_extrl_i64_i32:
1618 case INDEX_op_ext_i32_i64:
1619 tcg_out_ext32s(s, a0, a1);
1622 case INDEX_op_extrh_i64_i32:
1623 tcg_out_opc_imm(s, OPC_SRAI, a0, a1, 32);
1626 case INDEX_op_mulsh_i32:
1627 case INDEX_op_mulsh_i64:
1628 tcg_out_opc_reg(s, OPC_MULH, a0, a1, a2);
1631 case INDEX_op_muluh_i32:
1632 case INDEX_op_muluh_i64:
1633 tcg_out_opc_reg(s, OPC_MULHU, a0, a1, a2);
1640 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
1641 case INDEX_op_mov_i64:
1642 case INDEX_op_call: /* Always emitted via tcg_out_call. */
1643 case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */
1644 case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */
1646 g_assert_not_reached();
1650 static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
1653 case INDEX_op_goto_ptr:
1656 case INDEX_op_ld8u_i32:
1657 case INDEX_op_ld8s_i32:
1658 case INDEX_op_ld16u_i32:
1659 case INDEX_op_ld16s_i32:
1660 case INDEX_op_ld_i32:
1661 case INDEX_op_not_i32:
1662 case INDEX_op_neg_i32:
1663 case INDEX_op_ld8u_i64:
1664 case INDEX_op_ld8s_i64:
1665 case INDEX_op_ld16u_i64:
1666 case INDEX_op_ld16s_i64:
1667 case INDEX_op_ld32s_i64:
1668 case INDEX_op_ld32u_i64:
1669 case INDEX_op_ld_i64:
1670 case INDEX_op_not_i64:
1671 case INDEX_op_neg_i64:
1672 case INDEX_op_ext8u_i32:
1673 case INDEX_op_ext8u_i64:
1674 case INDEX_op_ext16u_i32:
1675 case INDEX_op_ext16u_i64:
1676 case INDEX_op_ext32u_i64:
1677 case INDEX_op_extu_i32_i64:
1678 case INDEX_op_ext8s_i32:
1679 case INDEX_op_ext8s_i64:
1680 case INDEX_op_ext16s_i32:
1681 case INDEX_op_ext16s_i64:
1682 case INDEX_op_ext32s_i64:
1683 case INDEX_op_extrl_i64_i32:
1684 case INDEX_op_extrh_i64_i32:
1685 case INDEX_op_ext_i32_i64:
1686 return C_O1_I1(r, r);
1688 case INDEX_op_st8_i32:
1689 case INDEX_op_st16_i32:
1690 case INDEX_op_st_i32:
1691 case INDEX_op_st8_i64:
1692 case INDEX_op_st16_i64:
1693 case INDEX_op_st32_i64:
1694 case INDEX_op_st_i64:
1695 return C_O0_I2(rZ, r);
1697 case INDEX_op_add_i32:
1698 case INDEX_op_and_i32:
1699 case INDEX_op_or_i32:
1700 case INDEX_op_xor_i32:
1701 case INDEX_op_add_i64:
1702 case INDEX_op_and_i64:
1703 case INDEX_op_or_i64:
1704 case INDEX_op_xor_i64:
1705 return C_O1_I2(r, r, rI);
1707 case INDEX_op_sub_i32:
1708 case INDEX_op_sub_i64:
1709 return C_O1_I2(r, rZ, rN);
1711 case INDEX_op_mul_i32:
1712 case INDEX_op_mulsh_i32:
1713 case INDEX_op_muluh_i32:
1714 case INDEX_op_div_i32:
1715 case INDEX_op_divu_i32:
1716 case INDEX_op_rem_i32:
1717 case INDEX_op_remu_i32:
1718 case INDEX_op_setcond_i32:
1719 case INDEX_op_mul_i64:
1720 case INDEX_op_mulsh_i64:
1721 case INDEX_op_muluh_i64:
1722 case INDEX_op_div_i64:
1723 case INDEX_op_divu_i64:
1724 case INDEX_op_rem_i64:
1725 case INDEX_op_remu_i64:
1726 case INDEX_op_setcond_i64:
1727 return C_O1_I2(r, rZ, rZ);
1729 case INDEX_op_shl_i32:
1730 case INDEX_op_shr_i32:
1731 case INDEX_op_sar_i32:
1732 case INDEX_op_shl_i64:
1733 case INDEX_op_shr_i64:
1734 case INDEX_op_sar_i64:
1735 return C_O1_I2(r, r, ri);
1737 case INDEX_op_brcond_i32:
1738 case INDEX_op_brcond_i64:
1739 return C_O0_I2(rZ, rZ);
1741 case INDEX_op_add2_i32:
1742 case INDEX_op_add2_i64:
1743 case INDEX_op_sub2_i32:
1744 case INDEX_op_sub2_i64:
1745 return C_O2_I4(r, r, rZ, rZ, rM, rM);
1747 case INDEX_op_brcond2_i32:
1748 return C_O0_I4(rZ, rZ, rZ, rZ);
1750 case INDEX_op_setcond2_i32:
1751 return C_O1_I4(r, rZ, rZ, rZ, rZ);
1753 case INDEX_op_qemu_ld_i32:
1754 return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
1755 ? C_O1_I1(r, L) : C_O1_I2(r, L, L));
1756 case INDEX_op_qemu_st_i32:
1757 return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
1758 ? C_O0_I2(LZ, L) : C_O0_I3(LZ, L, L));
1759 case INDEX_op_qemu_ld_i64:
1760 return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L)
1761 : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O2_I1(r, r, L)
1762 : C_O2_I2(r, r, L, L));
1763 case INDEX_op_qemu_st_i64:
1764 return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(LZ, L)
1765 : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O0_I3(LZ, LZ, L)
1766 : C_O0_I4(LZ, LZ, L, L));
1769 g_assert_not_reached();
1773 static const int tcg_target_callee_save_regs[] = {
1774 TCG_REG_S0, /* used for the global env (TCG_AREG0) */
1786 TCG_REG_RA, /* should be last for ABI compliance */
1789 /* Stack frame parameters. */
1790 #define REG_SIZE (TCG_TARGET_REG_BITS / 8)
1791 #define SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * REG_SIZE)
1792 #define TEMP_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1793 #define FRAME_SIZE ((TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE + SAVE_SIZE \
1794 + TCG_TARGET_STACK_ALIGN - 1) \
1795 & -TCG_TARGET_STACK_ALIGN)
1796 #define SAVE_OFS (TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE)
1798 /* We're expecting to be able to use an immediate for frame allocation. */
1799 QEMU_BUILD_BUG_ON(FRAME_SIZE > 0x7ff);
1801 /* Generate global QEMU prologue and epilogue code */
1802 static void tcg_target_qemu_prologue(TCGContext *s)
1806 tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE, TEMP_SIZE);
1809 tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, -FRAME_SIZE);
1810 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
1811 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1812 TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
1815 #if !defined(CONFIG_SOFTMMU)
1816 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
1817 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1820 /* Call generated code */
1821 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1822 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, tcg_target_call_iarg_regs[1], 0);
1824 /* Return path for goto_ptr. Set return value to 0 */
1825 tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
1826 tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_A0, TCG_REG_ZERO);
1829 tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
1830 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
1831 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1832 TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
1835 tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, FRAME_SIZE);
1836 tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_RA, 0);
1839 static void tcg_target_init(TCGContext *s)
1841 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
1842 if (TCG_TARGET_REG_BITS == 64) {
1843 tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
1846 tcg_target_call_clobber_regs = -1u;
1847 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S0);
1848 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S1);
1849 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S2);
1850 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S3);
1851 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S4);
1852 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S5);
1853 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S6);
1854 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S7);
1855 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S8);
1856 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S9);
1857 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S10);
1858 tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S11);
1860 s->reserved_regs = 0;
1861 tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO);
1862 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
1863 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
1864 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
1865 tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
1866 tcg_regset_set_reg(s->reserved_regs, TCG_REG_GP);
1867 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TP);
1872 uint8_t fde_def_cfa[4];
1873 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2];
1876 #define ELF_HOST_MACHINE EM_RISCV
1878 static const DebugFrame debug_frame = {
1879 .h.cie.len = sizeof(DebugFrameCIE) - 4, /* length after .len member */
1882 .h.cie.code_align = 1,
1883 .h.cie.data_align = -(TCG_TARGET_REG_BITS / 8) & 0x7f, /* sleb128 */
1884 .h.cie.return_column = TCG_REG_RA,
1886 /* Total FDE size does not include the "len" member. */
1887 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
1890 12, TCG_REG_SP, /* DW_CFA_def_cfa sp, ... */
1891 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
1895 0x80 + 9, 12, /* DW_CFA_offset, s1, -96 */
1896 0x80 + 18, 11, /* DW_CFA_offset, s2, -88 */
1897 0x80 + 19, 10, /* DW_CFA_offset, s3, -80 */
1898 0x80 + 20, 9, /* DW_CFA_offset, s4, -72 */
1899 0x80 + 21, 8, /* DW_CFA_offset, s5, -64 */
1900 0x80 + 22, 7, /* DW_CFA_offset, s6, -56 */
1901 0x80 + 23, 6, /* DW_CFA_offset, s7, -48 */
1902 0x80 + 24, 5, /* DW_CFA_offset, s8, -40 */
1903 0x80 + 25, 4, /* DW_CFA_offset, s9, -32 */
1904 0x80 + 26, 3, /* DW_CFA_offset, s10, -24 */
1905 0x80 + 27, 2, /* DW_CFA_offset, s11, -16 */
1906 0x80 + 1 , 1, /* DW_CFA_offset, ra, -8 */
1910 void tcg_register_jit(const void *buf, size_t buf_size)
1912 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));