target/arm: Correct load exclusive pair atomicity

We are not providing the required single-copy atomic semantics for
the 64-bit operation that is the 32-bit paired load.

At the same time, leave the entire 64-bit value in cpu_exclusive_val
and stop writing to cpu_exclusive_high. This means that we do not
have to re-assemble the 64-bit quantity when it comes time to store.

At the same time, drop a redundant temporary and perform all loads
directly into the cpu_exclusive_* globals.

Backports commit 19514cde3b92938df750acaecf2caaa85e1d36a6 from qemu
This commit is contained in:
Richard Henderson 2018-03-04 01:47:37 -05:00 committed by Lioncash
parent 009a52dd13
commit 4a8f556c29
No known key found for this signature in database
GPG key ID: 4E3C3CC1031BA9C7

View file

@ -1897,29 +1897,43 @@ static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
TCGv_i64 addr, int size, bool is_pair)
{
TCGContext *tcg_ctx = s->uc->tcg_ctx;
TCGv_i64 tmp = tcg_temp_new_i64(tcg_ctx);
TCGMemOp memop = s->be_data + size;
int idx = get_mem_index(s);
TCGMemOp memop = s->be_data;
g_assert(size <= 3);
tcg_gen_qemu_ld_i64(s->uc, tmp, addr, get_mem_index(s), memop);
if (is_pair) {
TCGv_i64 addr2 = tcg_temp_new_i64(tcg_ctx);
TCGv_i64 hitmp = tcg_temp_new_i64(tcg_ctx);
g_assert(size >= 2);
tcg_gen_addi_i64(tcg_ctx, addr2, addr, 1 << size);
tcg_gen_qemu_ld_i64(s->uc, hitmp, addr2, get_mem_index(s), memop);
tcg_temp_free_i64(tcg_ctx, addr2);
tcg_gen_mov_i64(tcg_ctx, tcg_ctx->cpu_exclusive_high, hitmp);
tcg_gen_mov_i64(tcg_ctx, cpu_reg(s, rt2), hitmp);
tcg_temp_free_i64(tcg_ctx, hitmp);
if (size == 2) {
/* The pair must be single-copy atomic for the doubleword. */
memop |= MO_64;
tcg_gen_qemu_ld_i64(s->uc, tcg_ctx->cpu_exclusive_val, addr, idx, memop);
if (s->be_data == MO_LE) {
tcg_gen_extract_i64(tcg_ctx, cpu_reg(s, rt), tcg_ctx->cpu_exclusive_val, 0, 32);
tcg_gen_extract_i64(tcg_ctx, cpu_reg(s, rt2), tcg_ctx->cpu_exclusive_val, 32, 32);
} else {
tcg_gen_extract_i64(tcg_ctx, cpu_reg(s, rt), tcg_ctx->cpu_exclusive_val, 32, 32);
tcg_gen_extract_i64(tcg_ctx, cpu_reg(s, rt2), tcg_ctx->cpu_exclusive_val, 0, 32);
}
} else {
/* The pair must be single-copy atomic for *each* doubleword,
but not the entire quadword. */
memop |= MO_64;
tcg_gen_qemu_ld_i64(s->uc, tcg_ctx->cpu_exclusive_val, addr, idx, memop);
TCGv_i64 addr2 = tcg_temp_new_i64(tcg_ctx);
tcg_gen_addi_i64(tcg_ctx, addr2, addr, 8);
tcg_gen_qemu_ld_i64(s->uc, tcg_ctx->cpu_exclusive_high, addr2, idx, memop);
tcg_temp_free_i64(tcg_ctx, addr2);
tcg_gen_mov_i64(tcg_ctx, cpu_reg(s, rt), tcg_ctx->cpu_exclusive_val);
tcg_gen_mov_i64(tcg_ctx, cpu_reg(s, rt2), tcg_ctx->cpu_exclusive_high);
}
} else {
memop |= size;
tcg_gen_qemu_ld_i64(s->uc, tcg_ctx->cpu_exclusive_val, addr, idx, memop);
tcg_gen_mov_i64(tcg_ctx, cpu_reg(s, rt), tcg_ctx->cpu_exclusive_val);
}
tcg_gen_mov_i64(tcg_ctx, tcg_ctx->cpu_exclusive_val, tmp);
tcg_gen_mov_i64(tcg_ctx, cpu_reg(s, rt), tmp);
tcg_temp_free_i64(tcg_ctx, tmp);
tcg_gen_mov_i64(tcg_ctx, tcg_ctx->cpu_exclusive_addr, addr);
}
@ -1954,14 +1968,15 @@ static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
if (is_pair) {
if (size == 2) {
TCGv_i64 val = tcg_temp_new_i64(tcg_ctx);
tcg_gen_concat32_i64(tcg_ctx, tmp, cpu_reg(s, rt), cpu_reg(s, rt2));
tcg_gen_concat32_i64(tcg_ctx, val, tcg_ctx->cpu_exclusive_val, tcg_ctx->cpu_exclusive_high);
tcg_gen_atomic_cmpxchg_i64(tcg_ctx, tmp, addr, val, tmp,
if (s->be_data == MO_LE) {
tcg_gen_concat32_i64(tcg_ctx, tmp, cpu_reg(s, rt), cpu_reg(s, rt2));
} else {
tcg_gen_concat32_i64(tcg_ctx, tmp, cpu_reg(s, rt2), cpu_reg(s, rt));
}
tcg_gen_atomic_cmpxchg_i64(tcg_ctx, tmp, addr, tcg_ctx->cpu_exclusive_val, tmp,
get_mem_index(s),
MO_64 | MO_ALIGN | s->be_data);
tcg_gen_setcond_i64(tcg_ctx, TCG_COND_NE, tmp, tmp, val);
tcg_temp_free_i64(tcg_ctx, val);
tcg_gen_setcond_i64(tcg_ctx, TCG_COND_NE, tmp, tmp, tcg_ctx->cpu_exclusive_val);
} else if (s->be_data == MO_LE) {
gen_helper_paired_cmpxchg64_le(tcg_ctx, tmp, tcg_ctx->cpu_env, addr, cpu_reg(s, rt),
cpu_reg(s, rt2));