tcg: Reduce serial context atomicity earlier

Reduce atomicity while emitting opcodes, instead of later
during code generation.  This ensures that any helper called
also sees the reduced atomicity requirement.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2034
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20231212193542.149117-1-richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2023-12-12 11:35:42 -08:00
parent b5e0d5d22f
commit cbb145567c
2 changed files with 25 additions and 12 deletions

View File

@ -77,6 +77,13 @@ static MemOp tcg_canonicalize_memop(MemOp op, bool is64, bool st)
if (st) { if (st) {
op &= ~MO_SIGN; op &= ~MO_SIGN;
} }
/* In serial mode, reduce atomicity. */
if (!(tcg_ctx->gen_tb->cflags & CF_PARALLEL)) {
op &= ~MO_ATOM_MASK;
op |= MO_ATOM_NONE;
}
return op; return op;
} }
@ -428,8 +435,7 @@ static bool use_two_i64_for_i128(MemOp mop)
case MO_ATOM_SUBALIGN: case MO_ATOM_SUBALIGN:
case MO_ATOM_WITHIN16: case MO_ATOM_WITHIN16:
case MO_ATOM_WITHIN16_PAIR: case MO_ATOM_WITHIN16_PAIR:
/* In a serialized context, no atomicity is required. */ return false;
return !(tcg_ctx->gen_tb->cflags & CF_PARALLEL);
default: default:
g_assert_not_reached(); g_assert_not_reached();
} }
@ -499,13 +505,20 @@ static void maybe_free_addr64(TCGv_i64 a64)
static void tcg_gen_qemu_ld_i128_int(TCGv_i128 val, TCGTemp *addr, static void tcg_gen_qemu_ld_i128_int(TCGv_i128 val, TCGTemp *addr,
TCGArg idx, MemOp memop) TCGArg idx, MemOp memop)
{ {
const MemOpIdx orig_oi = make_memop_idx(memop, idx); MemOpIdx orig_oi;
TCGv_i64 ext_addr = NULL; TCGv_i64 ext_addr = NULL;
TCGOpcode opc; TCGOpcode opc;
check_max_alignment(get_alignment_bits(memop)); check_max_alignment(get_alignment_bits(memop));
tcg_gen_req_mo(TCG_MO_LD_LD | TCG_MO_ST_LD); tcg_gen_req_mo(TCG_MO_LD_LD | TCG_MO_ST_LD);
/* In serial mode, reduce atomicity. */
if (!(tcg_ctx->gen_tb->cflags & CF_PARALLEL)) {
memop &= ~MO_ATOM_MASK;
memop |= MO_ATOM_NONE;
}
orig_oi = make_memop_idx(memop, idx);
/* TODO: For now, force 32-bit hosts to use the helper. */ /* TODO: For now, force 32-bit hosts to use the helper. */
if (TCG_TARGET_HAS_qemu_ldst_i128 && TCG_TARGET_REG_BITS == 64) { if (TCG_TARGET_HAS_qemu_ldst_i128 && TCG_TARGET_REG_BITS == 64) {
TCGv_i64 lo, hi; TCGv_i64 lo, hi;
@ -608,13 +621,20 @@ void tcg_gen_qemu_ld_i128_chk(TCGv_i128 val, TCGTemp *addr, TCGArg idx,
static void tcg_gen_qemu_st_i128_int(TCGv_i128 val, TCGTemp *addr, static void tcg_gen_qemu_st_i128_int(TCGv_i128 val, TCGTemp *addr,
TCGArg idx, MemOp memop) TCGArg idx, MemOp memop)
{ {
const MemOpIdx orig_oi = make_memop_idx(memop, idx); MemOpIdx orig_oi;
TCGv_i64 ext_addr = NULL; TCGv_i64 ext_addr = NULL;
TCGOpcode opc; TCGOpcode opc;
check_max_alignment(get_alignment_bits(memop)); check_max_alignment(get_alignment_bits(memop));
tcg_gen_req_mo(TCG_MO_ST_LD | TCG_MO_ST_ST); tcg_gen_req_mo(TCG_MO_ST_LD | TCG_MO_ST_ST);
/* In serial mode, reduce atomicity. */
if (!(tcg_ctx->gen_tb->cflags & CF_PARALLEL)) {
memop &= ~MO_ATOM_MASK;
memop |= MO_ATOM_NONE;
}
orig_oi = make_memop_idx(memop, idx);
/* TODO: For now, force 32-bit hosts to use the helper. */ /* TODO: For now, force 32-bit hosts to use the helper. */
if (TCG_TARGET_HAS_qemu_ldst_i128 && TCG_TARGET_REG_BITS == 64) { if (TCG_TARGET_HAS_qemu_ldst_i128 && TCG_TARGET_REG_BITS == 64) {

View File

@ -5440,15 +5440,8 @@ static TCGAtomAlign atom_and_align_for_opc(TCGContext *s, MemOp opc,
MemOp align = get_alignment_bits(opc); MemOp align = get_alignment_bits(opc);
MemOp size = opc & MO_SIZE; MemOp size = opc & MO_SIZE;
MemOp half = size ? size - 1 : 0; MemOp half = size ? size - 1 : 0;
MemOp atom = opc & MO_ATOM_MASK;
MemOp atmax; MemOp atmax;
MemOp atom;
/* When serialized, no further atomicity required. */
if (s->gen_tb->cflags & CF_PARALLEL) {
atom = opc & MO_ATOM_MASK;
} else {
atom = MO_ATOM_NONE;
}
switch (atom) { switch (atom) {
case MO_ATOM_NONE: case MO_ATOM_NONE: