Richard Henderson <r...@twiddle.net> writes:
> Add all of cmpxchg, op_fetch, fetch_op, and xchg.
> Handle both endian-ness, and sizes up to 8.
> Handle expanding non-atomically, when emulating in serial.
>
> Signed-off-by: Richard Henderson <r...@twiddle.net>
<snip>
> /* For the benefit of TCG generated code, we want to avoid the complication
> of ABI-specific return type promotion and always return a value extended
> to the register size of the host. This is tcg_target_long, except in the
> @@ -508,11 +507,184 @@ void probe_write(CPUArchState *env, target_ulong addr,
> int mmu_idx,
> }
> }
> #endif
> +
> +#if DATA_SIZE == 1
> +# define HE_SUFFIX _mmu
> +#elif defined(HOST_WORDS_BIGENDIAN)
> +# define HE_SUFFIX _be_mmu
> +# define RE_SUFFIX _le_mmu
> +#else
> +# define HE_SUFFIX _le_mmu
> +# define RE_SUFFIX _be_mmu
> +#endif
> +
> +#define ATOMIC_MMU_BODY \
> + DATA_TYPE *haddr; \
> + do { \
> + unsigned mmu_idx = get_mmuidx(oi); \
> + int index = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1); \
> + CPUTLBEntry *tlbe = &env->tlb_table[mmu_idx][index]; \
> + target_ulong tlb_addr = tlbe->addr_write; \
> + int a_bits = get_alignment_bits(get_memop(oi)); \
> + \
> + /* Adjust the given return address. */ \
> + retaddr -= GETPC_ADJ; \
> + \
> + /* Enforce guest required alignment. */ \
> + if (unlikely(a_bits > 0 && (addr & ((1 << a_bits) - 1)))) { \
> + /* ??? Maybe indicate atomic op to cpu_unaligned_access */ \
> + cpu_unaligned_access(ENV_GET_CPU(env), addr, MMU_DATA_STORE, \
> + mmu_idx, retaddr); \
> + } \
> + /* Enforce qemu required alignment. */ \
> + if (unlikely(addr & ((1 << SHIFT) - 1))) { \
> + /* We get here if guest alignment was not requested, \
> + or was not enforced by cpu_unaligned_access above. \
> + We might widen the access and emulate, but for now \
> + mark an exception and exit the cpu loop. */ \
> + cpu_loop_exit_atomic(ENV_GET_CPU(env), retaddr); \
> + } \
> + \
> + /* Check TLB entry and enforce page permissions. */ \
> + if ((addr & TARGET_PAGE_MASK) \
> + != (tlb_addr & (TARGET_PAGE_MASK | TLB_INVALID_MASK))) { \
> + if (!VICTIM_TLB_HIT(addr_write)) { \
> + tlb_fill(ENV_GET_CPU(env), addr, MMU_DATA_STORE, \
> + mmu_idx, retaddr); \
> + } \
> + tlb_addr = tlbe->addr_write; \
> + } else if (unlikely(tlbe->addr_read != tlb_addr)) { \
> + /* Let the guest notice RMW on a write-only page. */ \
> + tlb_fill(ENV_GET_CPU(env), addr, MMU_DATA_LOAD, \
> + mmu_idx, retaddr); \
> + } \
> + \
> + /* Notice an IO access. */ \
> + if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) { \
> + /* There's really nothing that can be done to \
> + support this apart from stop-the-world. */ \
> + cpu_loop_exit_atomic(ENV_GET_CPU(env), retaddr);
> \
Hmm what are we doing here? I don't think there is a reason to stop the
world because atomicity should be guaranteed by the BQL. Although I
suspect anything doing atomic access to an I/O address (especially with
LL/SC primitives) is going to be disappointed.
> + } \
> + haddr = (DATA_TYPE *)((uintptr_t)addr + tlbe->addend); \
> + } while (0)
> +
> +DATA_TYPE glue(glue(helper_atomic_cmpxchg, SUFFIX), HE_SUFFIX)
> + (CPUArchState *env, target_ulong addr, DATA_TYPE cmpv, DATA_TYPE newv,
> + TCGMemOpIdx oi, uintptr_t retaddr)
> +{
> + ATOMIC_MMU_BODY;
> + return atomic_cmpxchg(haddr, cmpv, newv);
> +}
> +
> +#define GEN_ATOMIC_HELPER(NAME) \
> +DATA_TYPE glue(glue(glue(helper_atomic_, NAME), SUFFIX), HE_SUFFIX) \
> + (CPUArchState *env, target_ulong addr, DATA_TYPE val, \
> + TCGMemOpIdx oi, uintptr_t retaddr) \
> +{ \
> + ATOMIC_MMU_BODY; \
> + return glue(atomic_, NAME)(haddr, val); \
> +}
> +
> +GEN_ATOMIC_HELPER(fetch_add)
> +GEN_ATOMIC_HELPER(fetch_and)
> +GEN_ATOMIC_HELPER(fetch_or)
> +GEN_ATOMIC_HELPER(fetch_xor)
> +
> +GEN_ATOMIC_HELPER(add_fetch)
> +GEN_ATOMIC_HELPER(and_fetch)
> +GEN_ATOMIC_HELPER(or_fetch)
> +GEN_ATOMIC_HELPER(xor_fetch)
> +
> +GEN_ATOMIC_HELPER(xchg)
> +
> +#undef GEN_ATOMIC_HELPER
> +
> +#if DATA_SIZE > 1
> +DATA_TYPE glue(glue(helper_atomic_cmpxchg, SUFFIX), RE_SUFFIX)
> + (CPUArchState *env, target_ulong addr, DATA_TYPE cmpv, DATA_TYPE newv,
> + TCGMemOpIdx oi, uintptr_t retaddr)
> +{
> + DATA_TYPE retv;
> + cmpv = BSWAP(cmpv);
> + newv = BSWAP(newv);
> + retv = (glue(glue(helper_atomic_cmpxchg, SUFFIX), HE_SUFFIX)
> + (env, addr, cmpv, newv, oi, retaddr));
> + return BSWAP(retv);
> +}
> +
> +#define GEN_ATOMIC_HELPER(NAME) \
> +DATA_TYPE glue(glue(glue(helper_atomic_, NAME), SUFFIX), RE_SUFFIX) \
> + (CPUArchState *env, target_ulong addr, DATA_TYPE val, \
> + TCGMemOpIdx oi, uintptr_t retaddr) \
> +{ \
> + DATA_TYPE ret; \
> + val = BSWAP(val); \
> + ret = (glue(glue(glue(helper_atomic_, NAME), SUFFIX), HE_SUFFIX) \
> + (env, addr, val, oi, retaddr)); \
> + return BSWAP(ret); \
> +}
> +
> +GEN_ATOMIC_HELPER(fetch_and)
> +GEN_ATOMIC_HELPER(fetch_or)
> +GEN_ATOMIC_HELPER(fetch_xor)
> +
> +GEN_ATOMIC_HELPER(and_fetch)
> +GEN_ATOMIC_HELPER(or_fetch)
> +GEN_ATOMIC_HELPER(xor_fetch)
> +
> +GEN_ATOMIC_HELPER(xchg)
> +
> +#undef GEN_ATOMIC_HELPER
> +
> +/* Note that for addition, we need to use a separate cmpxchg loop instead
> + of bswaps around the host-endian helpers. */
> +DATA_TYPE glue(glue(helper_atomic_fetch_add, SUFFIX), RE_SUFFIX)
> + (CPUArchState *env, target_ulong addr, DATA_TYPE val,
> + TCGMemOpIdx oi, uintptr_t retaddr)
> +{
> + DATA_TYPE ldo, ldn, ret, sto;
> + ATOMIC_MMU_BODY;
> +
> + ldo = *haddr;
> + while (1) {
> + ret = BSWAP(ldo);
> + sto = BSWAP(ret + val);
> + ldn = atomic_cmpxchg(haddr, ldo, sto);
> + if (ldn == ldo) {
> + return ret;
> + }
> + ldo = ldn;
> + }
> +}
> +
> +DATA_TYPE glue(glue(helper_atomic_add_fetch, SUFFIX), RE_SUFFIX)
> + (CPUArchState *env, target_ulong addr, DATA_TYPE val,
> + TCGMemOpIdx oi, uintptr_t retaddr)
> +{
> + DATA_TYPE ldo, ldn, ret, sto;
> + ATOMIC_MMU_BODY;
> +
> + ldo = *haddr;
> + while (1) {
> + ret = BSWAP(ldo) + val;
> + sto = BSWAP(ret);
> + ldn = atomic_cmpxchg(haddr, ldo, sto);
> + if (ldn == ldo) {
> + return ret;
> + }
> + ldo = ldn;
> + }
> +}
> +#endif /* DATA_SIZE > 1 */
> +
> +#undef ATOMIC_MMU_BODY
> +
> #endif /* !defined(SOFTMMU_CODE_ACCESS) */
>
> #undef READ_ACCESS_TYPE
> #undef SHIFT
> #undef DATA_TYPE
> +#undef DATAX_TYPE
Accidental typo?
> #undef SUFFIX
> #undef LSUFFIX
> #undef DATA_SIZE
> @@ -524,8 +696,6 @@ void probe_write(CPUArchState *env, target_ulong addr,
> int mmu_idx,
> #undef BSWAP
> #undef TGT_BE
> #undef TGT_LE
> -#undef CPU_BE
> -#undef CPU_LE
I suspect these belong in a separate clean-up patch?
> #undef helper_le_ld_name
> #undef helper_be_ld_name
> #undef helper_le_lds_name
> @@ -534,3 +704,5 @@ void probe_write(CPUArchState *env, target_ulong addr,
> int mmu_idx,
> #undef helper_be_st_name
> #undef helper_te_ld_name
> #undef helper_te_st_name
> +#undef HE_SUFFIX
> +#undef RE_SUFFIX
> diff --git a/tcg-runtime.c b/tcg-runtime.c
> index ea2ad64..63f78fc 100644
> --- a/tcg-runtime.c
> +++ b/tcg-runtime.c
> @@ -23,17 +23,10 @@
> */
> #include "qemu/osdep.h"
> #include "qemu/host-utils.h"
> -
> -/* This file is compiled once, and thus we can't include the standard
> - "exec/helper-proto.h", which has includes that are target specific. */
> -
> -#include "exec/helper-head.h"
> -
> -#define DEF_HELPER_FLAGS_2(name, flags, ret, t1, t2) \
> - dh_ctype(ret) HELPER(name) (dh_ctype(t1), dh_ctype(t2));
> -
> -#include "tcg-runtime.h"
> -
> +#include "cpu.h"
> +#include "exec/helper-proto.h"
> +#include "exec/cpu_ldst.h"
> +#include "exec/exec-all.h"
>
> /* 32-bit helpers */
>
> @@ -107,3 +100,15 @@ int64_t HELPER(mulsh_i64)(int64_t arg1, int64_t arg2)
> muls64(&l, &h, arg1, arg2);
> return h;
> }
> +
> +#define SHIFT 0
> +#include "atomic_template.h"
> +
> +#define SHIFT 1
> +#include "atomic_template.h"
> +
> +#define SHIFT 2
> +#include "atomic_template.h"
> +
> +#define SHIFT 3
> +#include "atomic_template.h"
> diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
> index 293b854..bc72c17 100644
> --- a/tcg/tcg-op.c
> +++ b/tcg/tcg-op.c
> @@ -1958,3 +1958,327 @@ void tcg_gen_qemu_st_i64(TCGv_i64 val, TCGv addr,
> TCGArg idx, TCGMemOp memop)
> addr, trace_mem_get_info(memop, 1));
> gen_ldst_i64(INDEX_op_qemu_st_i64, val, addr, memop, idx);
> }
> +
> +static void tcg_gen_ext_i32(TCGv_i32 ret, TCGv_i32 val, TCGMemOp opc)
> +{
> + switch (opc & MO_SSIZE) {
> + case MO_SB:
> + tcg_gen_ext8s_i32(ret, val);
> + break;
> + case MO_UB:
> + tcg_gen_ext8u_i32(ret, val);
> + break;
> + case MO_SW:
> + tcg_gen_ext16s_i32(ret, val);
> + break;
> + case MO_UW:
> + tcg_gen_ext16u_i32(ret, val);
> + break;
> + default:
> + tcg_gen_mov_i32(ret, val);
> + break;
> + }
> +}
> +
> +static void tcg_gen_ext_i64(TCGv_i64 ret, TCGv_i64 val, TCGMemOp opc)
> +{
> + switch (opc & MO_SSIZE) {
> + case MO_SB:
> + tcg_gen_ext8s_i64(ret, val);
> + break;
> + case MO_UB:
> + tcg_gen_ext8u_i64(ret, val);
> + break;
> + case MO_SW:
> + tcg_gen_ext16s_i64(ret, val);
> + break;
> + case MO_UW:
> + tcg_gen_ext16u_i64(ret, val);
> + break;
> + case MO_SL:
> + tcg_gen_ext32s_i64(ret, val);
> + break;
> + case MO_UL:
> + tcg_gen_ext32u_i64(ret, val);
> + break;
> + default:
> + tcg_gen_mov_i64(ret, val);
> + break;
> + }
> +}
Should these two be separate patches?
> +
> +#ifdef CONFIG_USER_ONLY
> +typedef void (*gen_atomic_cx_i32)(TCGv_i32, TCGv, TCGv_i32, TCGv_i32);
> +typedef void (*gen_atomic_cx_i64)(TCGv_i64, TCGv, TCGv_i64, TCGv_i64);
> +typedef void (*gen_atomic_op_i32)(TCGv_i32, TCGv, TCGv_i32);
> +typedef void (*gen_atomic_op_i64)(TCGv_i64, TCGv, TCGv_i64);
> +#else
> +typedef void (*gen_atomic_cx_i32)(TCGv_i32, TCGv_env, TCGv,
> + TCGv_i32, TCGv_i32, TCGv_i32);
> +typedef void (*gen_atomic_cx_i64)(TCGv_i64, TCGv_env, TCGv,
> + TCGv_i64, TCGv_i64, TCGv_i32);
> +typedef void (*gen_atomic_op_i32)(TCGv_i32, TCGv_env, TCGv, TCGv_i32,
> TCGv_i32);
> +typedef void (*gen_atomic_op_i64)(TCGv_i64, TCGv_env, TCGv, TCGv_i64,
> TCGv_i32);
> +#endif
> +
> +static void * const table_cmpxchg[16] = {
> + [MO_8] = gen_helper_atomic_cmpxchgb,
> + [MO_16 | MO_LE] = gen_helper_atomic_cmpxchgw_le,
> + [MO_16 | MO_BE] = gen_helper_atomic_cmpxchgw_be,
> + [MO_32 | MO_LE] = gen_helper_atomic_cmpxchgl_le,
> + [MO_32 | MO_BE] = gen_helper_atomic_cmpxchgl_be,
> + [MO_64 | MO_LE] = gen_helper_atomic_cmpxchgq_le,
> + [MO_64 | MO_BE] = gen_helper_atomic_cmpxchgq_be,
> +};
> +
> +void tcg_gen_atomic_cmpxchg_i32(TCGv_i32 retv, TCGv addr, TCGv_i32 cmpv,
> + TCGv_i32 newv, TCGArg idx, TCGMemOp memop)
> +{
> + memop = tcg_canonicalize_memop(memop, 0, 0);
> +
> + if (!parallel_cpus) {
> + TCGv_i32 t1 = tcg_temp_new_i32();
> + TCGv_i32 t2 = tcg_temp_new_i32();
> +
> + tcg_gen_ext_i32(t2, cmpv, memop & MO_SIZE);
> +
> + tcg_gen_qemu_ld_i32(t1, addr, idx, memop & ~MO_SIGN);
> + tcg_gen_movcond_i32(TCG_COND_EQ, t2, t1, t2, newv, t1);
> + tcg_gen_qemu_st_i32(t2, addr, idx, memop);
> + tcg_temp_free_i32(t2);
> +
> + if (memop & MO_SIGN) {
> + tcg_gen_ext_i32(retv, t1, memop);
> + } else {
> + tcg_gen_mov_i32(retv, t1);
> + }
> + tcg_temp_free_i32(t1);
> + } else {
> + gen_atomic_cx_i32 gen;
> +
> + gen = table_cmpxchg[memop & (MO_SIZE | MO_BSWAP)];
> + tcg_debug_assert(gen != NULL);
> +
> +#ifdef CONFIG_USER_ONLY
> + gen(retv, addr, cmpv, newv);
> +#else
> + {
> + TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop & ~MO_SIGN,
> idx));
> + gen(retv, tcg_ctx.tcg_env, addr, cmpv, newv, oi);
> + tcg_temp_free_i32(oi);
> + }
> +#endif
> +
> + if (memop & MO_SIGN) {
> + tcg_gen_ext_i32(retv, retv, memop);
> + }
> + }
> +}
> +
> +void tcg_gen_atomic_cmpxchg_i64(TCGv_i64 retv, TCGv addr, TCGv_i64 cmpv,
> + TCGv_i64 newv, TCGArg idx, TCGMemOp memop)
> +{
> + memop = tcg_canonicalize_memop(memop, 1, 0);
> +
> + if (!parallel_cpus) {
> + TCGv_i64 t1 = tcg_temp_new_i64();
> + TCGv_i64 t2 = tcg_temp_new_i64();
> +
> + tcg_gen_ext_i64(t2, cmpv, memop & MO_SIZE);
> +
> + tcg_gen_qemu_ld_i64(t1, addr, idx, memop & ~MO_SIGN);
> + tcg_gen_movcond_i64(TCG_COND_EQ, t2, t1, t2, newv, t1);
> + tcg_gen_qemu_st_i64(t2, addr, idx, memop);
> + tcg_temp_free_i64(t2);
> +
> + if (memop & MO_SIGN) {
> + tcg_gen_ext_i64(retv, t1, memop);
> + } else {
> + tcg_gen_mov_i64(retv, t1);
> + }
> + tcg_temp_free_i64(t1);
> + } else if ((memop & MO_SIZE) == MO_64) {
> + gen_atomic_cx_i64 gen;
> +
> + gen = table_cmpxchg[memop & (MO_SIZE | MO_BSWAP)];
> + tcg_debug_assert(gen != NULL);
> +
> +#ifdef CONFIG_USER_ONLY
> + gen(retv, addr, cmpv, newv);
> +#else
> + TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop, idx));
> + gen(retv, tcg_ctx.tcg_env, addr, cmpv, newv, oi);
> + tcg_temp_free_i32(oi);
> +#endif
> + } else {
> + TCGv_i32 c32 = tcg_temp_new_i32();
> + TCGv_i32 n32 = tcg_temp_new_i32();
> + TCGv_i32 r32 = tcg_temp_new_i32();
> +
> + tcg_gen_extrl_i64_i32(c32, cmpv);
> + tcg_gen_extrl_i64_i32(n32, newv);
> + tcg_gen_atomic_cmpxchg_i32(r32, addr, c32, n32, idx, memop &
> ~MO_SIGN);
> + tcg_temp_free_i32(c32);
> + tcg_temp_free_i32(n32);
> +
> + tcg_gen_extu_i32_i64(retv, r32);
> + tcg_temp_free_i32(r32);
> +
> + if (memop & MO_SIGN) {
> + tcg_gen_ext_i64(retv, retv, memop);
> + }
> + }
> +}
> +
> +static void do_nonatomic_op_i32(TCGv_i32 ret, TCGv addr, TCGv_i32 val,
> + TCGArg idx, TCGMemOp memop, bool new_val,
> + void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32))
> +{
> + TCGv_i32 t1 = tcg_temp_new_i32();
> + TCGv_i32 t2 = tcg_temp_new_i32();
> +
> + memop = tcg_canonicalize_memop(memop, 0, 0);
> +
> + tcg_gen_qemu_ld_i32(t1, addr, idx, memop & ~MO_SIGN);
> + gen(t2, t1, val);
> + tcg_gen_qemu_st_i32(t2, addr, idx, memop);
> +
> + tcg_gen_ext_i32(ret, (new_val ? t2 : t1), memop);
> + tcg_temp_free_i32(t1);
> + tcg_temp_free_i32(t2);
> +}
> +
> +static void do_atomic_op_i32(TCGv_i32 ret, TCGv addr, TCGv_i32 val,
> + TCGArg idx, TCGMemOp memop, void * const
> table[])
> +{
> + gen_atomic_op_i32 gen;
> +
> + memop = tcg_canonicalize_memop(memop, 0, 0);
> +
> + gen = table[memop & (MO_SIZE | MO_BSWAP)];
> + tcg_debug_assert(gen != NULL);
> +
> +#ifdef CONFIG_USER_ONLY
> + gen(ret, addr, val);
> +#else
> + {
> + TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop & ~MO_SIGN, idx));
> + gen(ret, tcg_ctx.tcg_env, addr, val, oi);
> + tcg_temp_free_i32(oi);
> + }
> +#endif
> +
> + if (memop & MO_SIGN) {
> + tcg_gen_ext_i32(ret, ret, memop);
> + }
> +}
> +
> +static void do_nonatomic_op_i64(TCGv_i64 ret, TCGv addr, TCGv_i64 val,
> + TCGArg idx, TCGMemOp memop, bool new_val,
> + void (*gen)(TCGv_i64, TCGv_i64, TCGv_i64))
> +{
> + TCGv_i64 t1 = tcg_temp_new_i64();
> + TCGv_i64 t2 = tcg_temp_new_i64();
> +
> + memop = tcg_canonicalize_memop(memop, 1, 0);
> +
> + tcg_gen_qemu_ld_i64(t1, addr, idx, memop & ~MO_SIGN);
> + gen(t2, t1, val);
> + tcg_gen_qemu_st_i64(t2, addr, idx, memop);
> +
> + tcg_gen_ext_i64(ret, (new_val ? t2 : t1), memop);
> + tcg_temp_free_i64(t1);
> + tcg_temp_free_i64(t2);
> +}
> +
> +static void do_atomic_op_i64(TCGv_i64 ret, TCGv addr, TCGv_i64 val,
> + TCGArg idx, TCGMemOp memop, void * const
> table[])
> +{
> + memop = tcg_canonicalize_memop(memop, 1, 0);
> +
> + if ((memop & MO_SIZE) == MO_64) {
> + gen_atomic_op_i64 gen;
> +
> + gen = table[memop & (MO_SIZE | MO_BSWAP)];
> + tcg_debug_assert(gen != NULL);
> +
> +#ifdef CONFIG_USER_ONLY
> + gen(ret, addr, val);
> +#else
> + {
> + TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop & ~MO_SIGN,
> idx));
> + gen(ret, tcg_ctx.tcg_env, addr, val, oi);
> + tcg_temp_free_i32(oi);
> + }
> +#endif
> + } else {
> + TCGv_i32 v32 = tcg_temp_new_i32();
> + TCGv_i32 r32 = tcg_temp_new_i32();
> +
> + tcg_gen_extrl_i64_i32(v32, val);
> + do_atomic_op_i32(r32, addr, v32, idx, memop & ~MO_SIGN, table);
> + tcg_temp_free_i32(v32);
> +
> + tcg_gen_extu_i32_i64(ret, r32);
> + tcg_temp_free_i32(r32);
> +
> + if (memop & MO_SIGN) {
> + tcg_gen_ext_i64(ret, ret, memop);
> + }
> + }
> +}
> +
> +#define GEN_ATOMIC_HELPER(NAME, OP, NEW) \
> +static void * const table_##NAME[16] = { \
> + [MO_8] = gen_helper_atomic_##NAME##b, \
> + [MO_16 | MO_LE] = gen_helper_atomic_##NAME##w_le, \
> + [MO_16 | MO_BE] = gen_helper_atomic_##NAME##w_be, \
> + [MO_32 | MO_LE] = gen_helper_atomic_##NAME##l_le, \
> + [MO_32 | MO_BE] = gen_helper_atomic_##NAME##l_be, \
> + [MO_64 | MO_LE] = gen_helper_atomic_##NAME##q_le, \
> + [MO_64 | MO_BE] = gen_helper_atomic_##NAME##q_be, \
> +}; \
> +void tcg_gen_atomic_##NAME##_i32 \
> + (TCGv_i32 ret, TCGv addr, TCGv_i32 val, TCGArg idx, TCGMemOp memop) \
> +{ \
> + if (parallel_cpus) { \
> + do_atomic_op_i32(ret, addr, val, idx, memop, table_##NAME); \
> + } else { \
> + do_nonatomic_op_i32(ret, addr, val, idx, memop, NEW, \
> + tcg_gen_##OP##_i32); \
> + } \
> +} \
> +void tcg_gen_atomic_##NAME##_i64 \
> + (TCGv_i64 ret, TCGv addr, TCGv_i64 val, TCGArg idx, TCGMemOp memop) \
> +{ \
> + if (parallel_cpus) { \
> + do_atomic_op_i64(ret, addr, val, idx, memop, table_##NAME); \
> + } else { \
> + do_nonatomic_op_i64(ret, addr, val, idx, memop, NEW, \
> + tcg_gen_##OP##_i64); \
> + } \
> +}
> +
> +GEN_ATOMIC_HELPER(fetch_add, add, 0)
> +GEN_ATOMIC_HELPER(fetch_and, and, 0)
> +GEN_ATOMIC_HELPER(fetch_or, or, 0)
> +GEN_ATOMIC_HELPER(fetch_xor, xor, 0)
> +
> +GEN_ATOMIC_HELPER(add_fetch, add, 1)
> +GEN_ATOMIC_HELPER(and_fetch, and, 1)
> +GEN_ATOMIC_HELPER(or_fetch, or, 1)
> +GEN_ATOMIC_HELPER(xor_fetch, xor, 1)
> +
> +static void tcg_gen_mov2_i32(TCGv_i32 r, TCGv_i32 a, TCGv_i32 b)
> +{
> + tcg_gen_mov_i32(r, b);
> +}
> +
> +static void tcg_gen_mov2_i64(TCGv_i64 r, TCGv_i64 a, TCGv_i64 b)
> +{
> + tcg_gen_mov_i64(r, b);
> +}
> +
> +GEN_ATOMIC_HELPER(xchg, mov2, 0)
> +
> +#undef GEN_ATOMIC_HELPER
> diff --git a/tcg/tcg-op.h b/tcg/tcg-op.h
> index f217e80..2a845ae 100644
> --- a/tcg/tcg-op.h
> +++ b/tcg/tcg-op.h
> @@ -852,6 +852,30 @@ static inline void tcg_gen_qemu_st64(TCGv_i64 arg, TCGv
> addr, int mem_index)
> tcg_gen_qemu_st_i64(arg, addr, mem_index, MO_TEQ);
> }
>
> +void tcg_gen_atomic_cmpxchg_i32(TCGv_i32, TCGv, TCGv_i32, TCGv_i32,
> + TCGArg, TCGMemOp);
> +void tcg_gen_atomic_cmpxchg_i64(TCGv_i64, TCGv, TCGv_i64, TCGv_i64,
> + TCGArg, TCGMemOp);
> +
> +void tcg_gen_atomic_xchg_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg, TCGMemOp);
> +void tcg_gen_atomic_xchg_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg, TCGMemOp);
> +void tcg_gen_atomic_fetch_add_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_fetch_add_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_fetch_and_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_fetch_and_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_fetch_or_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg, TCGMemOp);
> +void tcg_gen_atomic_fetch_or_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg, TCGMemOp);
> +void tcg_gen_atomic_fetch_xor_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_fetch_xor_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_add_fetch_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_add_fetch_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_and_fetch_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_and_fetch_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_or_fetch_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg, TCGMemOp);
> +void tcg_gen_atomic_or_fetch_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg, TCGMemOp);
> +void tcg_gen_atomic_xor_fetch_i32(TCGv_i32, TCGv, TCGv_i32, TCGArg,
> TCGMemOp);
> +void tcg_gen_atomic_xor_fetch_i64(TCGv_i64, TCGv, TCGv_i64, TCGArg,
> TCGMemOp);
> +
> #if TARGET_LONG_BITS == 64
> #define tcg_gen_movi_tl tcg_gen_movi_i64
> #define tcg_gen_mov_tl tcg_gen_mov_i64
> @@ -930,6 +954,16 @@ static inline void tcg_gen_qemu_st64(TCGv_i64 arg, TCGv
> addr, int mem_index)
> #define tcg_gen_sub2_tl tcg_gen_sub2_i64
> #define tcg_gen_mulu2_tl tcg_gen_mulu2_i64
> #define tcg_gen_muls2_tl tcg_gen_muls2_i64
> +#define tcg_gen_atomic_cmpxchg_tl tcg_gen_atomic_cmpxchg_i64
> +#define tcg_gen_atomic_xchg_tl tcg_gen_atomic_xchg_i64
> +#define tcg_gen_atomic_fetch_add_tl tcg_gen_atomic_fetch_add_i64
> +#define tcg_gen_atomic_fetch_and_tl tcg_gen_atomic_fetch_and_i64
> +#define tcg_gen_atomic_fetch_or_tl tcg_gen_atomic_fetch_or_i64
> +#define tcg_gen_atomic_fetch_xor_tl tcg_gen_atomic_fetch_xor_i64
> +#define tcg_gen_atomic_add_fetch_tl tcg_gen_atomic_add_fetch_i64
> +#define tcg_gen_atomic_and_fetch_tl tcg_gen_atomic_and_fetch_i64
> +#define tcg_gen_atomic_or_fetch_tl tcg_gen_atomic_or_fetch_i64
> +#define tcg_gen_atomic_xor_fetch_tl tcg_gen_atomic_xor_fetch_i64
> #else
> #define tcg_gen_movi_tl tcg_gen_movi_i32
> #define tcg_gen_mov_tl tcg_gen_mov_i32
> @@ -1007,6 +1041,16 @@ static inline void tcg_gen_qemu_st64(TCGv_i64 arg,
> TCGv addr, int mem_index)
> #define tcg_gen_sub2_tl tcg_gen_sub2_i32
> #define tcg_gen_mulu2_tl tcg_gen_mulu2_i32
> #define tcg_gen_muls2_tl tcg_gen_muls2_i32
> +#define tcg_gen_atomic_cmpxchg_tl tcg_gen_atomic_cmpxchg_i32
> +#define tcg_gen_atomic_xchg_tl tcg_gen_atomic_xchg_i32
> +#define tcg_gen_atomic_fetch_add_tl tcg_gen_atomic_fetch_add_i32
> +#define tcg_gen_atomic_fetch_and_tl tcg_gen_atomic_fetch_and_i32
> +#define tcg_gen_atomic_fetch_or_tl tcg_gen_atomic_fetch_or_i32
> +#define tcg_gen_atomic_fetch_xor_tl tcg_gen_atomic_fetch_xor_i32
> +#define tcg_gen_atomic_add_fetch_tl tcg_gen_atomic_add_fetch_i32
> +#define tcg_gen_atomic_and_fetch_tl tcg_gen_atomic_and_fetch_i32
> +#define tcg_gen_atomic_or_fetch_tl tcg_gen_atomic_or_fetch_i32
> +#define tcg_gen_atomic_xor_fetch_tl tcg_gen_atomic_xor_fetch_i32
> #endif
>
> #if UINTPTR_MAX == UINT32_MAX
> diff --git a/tcg/tcg-runtime.h b/tcg/tcg-runtime.h
> index 23a0c37..b3accf4 100644
> --- a/tcg/tcg-runtime.h
> +++ b/tcg/tcg-runtime.h
> @@ -14,3 +14,78 @@ DEF_HELPER_FLAGS_2(sar_i64, TCG_CALL_NO_RWG_SE, s64, s64,
> s64)
>
> DEF_HELPER_FLAGS_2(mulsh_i64, TCG_CALL_NO_RWG_SE, s64, s64, s64)
> DEF_HELPER_FLAGS_2(muluh_i64, TCG_CALL_NO_RWG_SE, i64, i64, i64)
> +
> +#ifdef CONFIG_USER_ONLY
> +
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgb, TCG_CALL_NO_WG, i32, tl, i32, i32)
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgw_be, TCG_CALL_NO_WG, i32, tl, i32, i32)
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgl_be, TCG_CALL_NO_WG, i32, tl, i32, i32)
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgq_be, TCG_CALL_NO_WG, i64, tl, i64, i64)
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgw_le, TCG_CALL_NO_WG, i32, tl, i32, i32)
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgl_le, TCG_CALL_NO_WG, i32, tl, i32, i32)
> +DEF_HELPER_FLAGS_3(atomic_cmpxchgq_le, TCG_CALL_NO_WG, i64, tl, i64, i64)
> +
> +#define GEN_ATOMIC_HELPERS(NAME) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), b), \
> + TCG_CALL_NO_WG, i32, tl, i32) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), w_le), \
> + TCG_CALL_NO_WG, i32, tl, i32) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), w_be), \
> + TCG_CALL_NO_WG, i32, tl, i32) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), l_le), \
> + TCG_CALL_NO_WG, i32, tl, i32) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), l_be), \
> + TCG_CALL_NO_WG, i32, tl, i32) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), q_le), \
> + TCG_CALL_NO_WG, i64, tl, i64) \
> + DEF_HELPER_FLAGS_2(glue(glue(atomic_, NAME), q_be), \
> + TCG_CALL_NO_WG, i64, tl, i64)
> +
> +#else
> +
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgb, TCG_CALL_NO_WG, i32, env,
> + tl, i32, i32, i32)
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgw_be, TCG_CALL_NO_WG, i32, env,
> + tl, i32, i32, i32)
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgl_be, TCG_CALL_NO_WG, i32, env,
> + tl, i32, i32, i32)
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgq_be, TCG_CALL_NO_WG, i64, env,
> + tl, i64, i64, i32)
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgw_le, TCG_CALL_NO_WG, i32, env,
> + tl, i32, i32, i32)
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgl_le, TCG_CALL_NO_WG, i32, env,
> + tl, i32, i32, i32)
> +DEF_HELPER_FLAGS_5(atomic_cmpxchgq_le, TCG_CALL_NO_WG, i64, env,
> + tl, i64, i64, i32)
> +
> +#define GEN_ATOMIC_HELPERS(NAME) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), b), \
> + TCG_CALL_NO_WG, i32, env, tl, i32, i32) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), w_le), \
> + TCG_CALL_NO_WG, i32, env, tl, i32, i32) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), w_be), \
> + TCG_CALL_NO_WG, i32, env, tl, i32, i32) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), l_le), \
> + TCG_CALL_NO_WG, i32, env, tl, i32, i32) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), l_be), \
> + TCG_CALL_NO_WG, i32, env, tl, i32, i32) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), q_le), \
> + TCG_CALL_NO_WG, i64, env, tl, i64, i32) \
> + DEF_HELPER_FLAGS_4(glue(glue(atomic_, NAME), q_be), \
> + TCG_CALL_NO_WG, i64, env, tl, i64, i32)
> +
> +#endif
> +
> +GEN_ATOMIC_HELPERS(fetch_add)
> +GEN_ATOMIC_HELPERS(fetch_and)
> +GEN_ATOMIC_HELPERS(fetch_or)
> +GEN_ATOMIC_HELPERS(fetch_xor)
> +
> +GEN_ATOMIC_HELPERS(add_fetch)
> +GEN_ATOMIC_HELPERS(and_fetch)
> +GEN_ATOMIC_HELPERS(or_fetch)
> +GEN_ATOMIC_HELPERS(xor_fetch)
> +
> +GEN_ATOMIC_HELPERS(xchg)
> +
> +#undef GEN_ATOMIC_HELPERS
> diff --git a/tcg/tcg.h b/tcg/tcg.h
> index ab67537..4e60498 100644
> --- a/tcg/tcg.h
> +++ b/tcg/tcg.h
> @@ -1163,6 +1163,59 @@ uint64_t helper_be_ldq_cmmu(CPUArchState *env,
> target_ulong addr,
> # define helper_ret_ldq_cmmu helper_le_ldq_cmmu
> #endif
>
> +uint8_t helper_atomic_cmpxchgb_mmu(CPUArchState *env, target_ulong addr,
> + uint8_t cmpv, uint8_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +uint16_t helper_atomic_cmpxchgw_le_mmu(CPUArchState *env, target_ulong addr,
> + uint16_t cmpv, uint16_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +uint32_t helper_atomic_cmpxchgl_le_mmu(CPUArchState *env, target_ulong addr,
> + uint32_t cmpv, uint32_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +uint64_t helper_atomic_cmpxchgq_le_mmu(CPUArchState *env, target_ulong addr,
> + uint64_t cmpv, uint64_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +uint16_t helper_atomic_cmpxchgw_be_mmu(CPUArchState *env, target_ulong addr,
> + uint16_t cmpv, uint16_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +uint32_t helper_atomic_cmpxchgl_be_mmu(CPUArchState *env, target_ulong addr,
> + uint32_t cmpv, uint32_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +uint64_t helper_atomic_cmpxchgq_be_mmu(CPUArchState *env, target_ulong addr,
> + uint64_t cmpv, uint64_t newv,
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +
> +#define GEN_ATOMIC_HELPER(NAME, TYPE, SUFFIX) \
> +TYPE helper_atomic_ ## NAME ## SUFFIX ## _mmu \
> + (CPUArchState *env, target_ulong addr, TYPE val, \
> + TCGMemOpIdx oi, uintptr_t retaddr);
> +
> +#define GEN_ATOMIC_HELPER_ALL(NAME) \
> + GEN_ATOMIC_HELPER(NAME, uint8_t, b) \
> + GEN_ATOMIC_HELPER(NAME, uint16_t, w_le) \
> + GEN_ATOMIC_HELPER(NAME, uint32_t, l_le) \
> + GEN_ATOMIC_HELPER(NAME, uint64_t, q_le) \
> + GEN_ATOMIC_HELPER(NAME, uint16_t, w_be) \
> + GEN_ATOMIC_HELPER(NAME, uint32_t, l_be) \
> + GEN_ATOMIC_HELPER(NAME, uint64_t, q_be)
> +
> +GEN_ATOMIC_HELPER_ALL(fetch_add)
> +GEN_ATOMIC_HELPER_ALL(fetch_sub)
> +GEN_ATOMIC_HELPER_ALL(fetch_and)
> +GEN_ATOMIC_HELPER_ALL(fetch_or)
> +GEN_ATOMIC_HELPER_ALL(fetch_xor)
> +
> +GEN_ATOMIC_HELPER_ALL(add_fetch)
> +GEN_ATOMIC_HELPER_ALL(sub_fetch)
> +GEN_ATOMIC_HELPER_ALL(and_fetch)
> +GEN_ATOMIC_HELPER_ALL(or_fetch)
> +GEN_ATOMIC_HELPER_ALL(xor_fetch)
> +
> +GEN_ATOMIC_HELPER_ALL(xchg)
> +
> +#undef GEN_ATOMIC_HELPER_ALL
> +#undef GEN_ATOMIC_HELPER
> +
> #endif /* CONFIG_SOFTMMU */
>
> #endif /* TCG_H */
--
Alex Bennée
