From b5e0d5d22fbffc3d8f7d3e86d7a2d05a1a974e27 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Tue, 12 Dec 2023 09:01:38 -0800 Subject: [PATCH 1/2] target/i386: Fix 32-bit wrapping of pc/eip computation In 32-bit mode, pc = eip + cs_base is also 32-bit, and must wrap. Failure to do so results in incorrect memory exceptions to the guest. Before 732d548732ed, this was implicitly done via truncation to target_ulong but only in qemu-system-i386, not qemu-system-x86_64. To fix this, we must add conditional zero-extensions. Since we have to test for 32 vs 64-bit anyway, note that cs_base is always zero in 64-bit mode. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2022 Signed-off-by: Richard Henderson Reviewed-by: Paolo Bonzini Message-Id: <20231212172510.103305-1-richard.henderson@linaro.org> --- target/i386/cpu.h | 9 +++++++-- target/i386/tcg/tcg-cpu.c | 11 +++++++++-- target/i386/tcg/translate.c | 23 +++++++++++++++++------ 3 files changed, 33 insertions(+), 10 deletions(-) diff --git a/target/i386/cpu.h b/target/i386/cpu.h index cd2e295bd6..ef987f344c 100644 --- a/target/i386/cpu.h +++ b/target/i386/cpu.h @@ -2324,10 +2324,15 @@ static inline int cpu_mmu_index_kernel(CPUX86State *env) static inline void cpu_get_tb_cpu_state(CPUX86State *env, vaddr *pc, uint64_t *cs_base, uint32_t *flags) { - *cs_base = env->segs[R_CS].base; - *pc = *cs_base + env->eip; *flags = env->hflags | (env->eflags & (IOPL_MASK | TF_MASK | RF_MASK | VM_MASK | AC_MASK)); + if (env->hflags & HF_CS64_MASK) { + *cs_base = 0; + *pc = env->eip; + } else { + *cs_base = env->segs[R_CS].base; + *pc = (uint32_t)(*cs_base + env->eip); + } } void do_cpu_init(X86CPU *cpu); diff --git a/target/i386/tcg/tcg-cpu.c b/target/i386/tcg/tcg-cpu.c index 2c6a12c835..6e881e9e27 100644 --- a/target/i386/tcg/tcg-cpu.c +++ b/target/i386/tcg/tcg-cpu.c @@ -52,7 +52,12 @@ static void x86_cpu_synchronize_from_tb(CPUState *cs, /* The instruction pointer is always up to date with CF_PCREL. */ if (!(tb_cflags(tb) & CF_PCREL)) { CPUX86State *env = cpu_env(cs); - env->eip = tb->pc - tb->cs_base; + + if (tb->flags & HF_CS64_MASK) { + env->eip = tb->pc; + } else { + env->eip = (uint32_t)(tb->pc - tb->cs_base); + } } } @@ -66,8 +71,10 @@ static void x86_restore_state_to_opc(CPUState *cs, if (tb_cflags(tb) & CF_PCREL) { env->eip = (env->eip & TARGET_PAGE_MASK) | data[0]; + } else if (tb->flags & HF_CS64_MASK) { + env->eip = data[0]; } else { - env->eip = data[0] - tb->cs_base; + env->eip = (uint32_t)(data[0] - tb->cs_base); } if (cc_op != CC_OP_DYNAMIC) { env->cc_op = cc_op; diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c index 587d88692a..037bc47e7c 100644 --- a/target/i386/tcg/translate.c +++ b/target/i386/tcg/translate.c @@ -552,8 +552,10 @@ static void gen_update_eip_cur(DisasContext *s) assert(s->pc_save != -1); if (tb_cflags(s->base.tb) & CF_PCREL) { tcg_gen_addi_tl(cpu_eip, cpu_eip, s->base.pc_next - s->pc_save); + } else if (CODE64(s)) { + tcg_gen_movi_tl(cpu_eip, s->base.pc_next); } else { - tcg_gen_movi_tl(cpu_eip, s->base.pc_next - s->cs_base); + tcg_gen_movi_tl(cpu_eip, (uint32_t)(s->base.pc_next - s->cs_base)); } s->pc_save = s->base.pc_next; } @@ -563,8 +565,10 @@ static void gen_update_eip_next(DisasContext *s) assert(s->pc_save != -1); if (tb_cflags(s->base.tb) & CF_PCREL) { tcg_gen_addi_tl(cpu_eip, cpu_eip, s->pc - s->pc_save); + } else if (CODE64(s)) { + tcg_gen_movi_tl(cpu_eip, s->base.pc_next); } else { - tcg_gen_movi_tl(cpu_eip, s->pc - s->cs_base); + tcg_gen_movi_tl(cpu_eip, (uint32_t)(s->base.pc_next - s->cs_base)); } s->pc_save = s->pc; } @@ -610,8 +614,10 @@ static TCGv eip_next_tl(DisasContext *s) TCGv ret = tcg_temp_new(); tcg_gen_addi_tl(ret, cpu_eip, s->pc - s->pc_save); return ret; + } else if (CODE64(s)) { + return tcg_constant_tl(s->pc); } else { - return tcg_constant_tl(s->pc - s->cs_base); + return tcg_constant_tl((uint32_t)(s->pc - s->cs_base)); } } @@ -622,8 +628,10 @@ static TCGv eip_cur_tl(DisasContext *s) TCGv ret = tcg_temp_new(); tcg_gen_addi_tl(ret, cpu_eip, s->base.pc_next - s->pc_save); return ret; + } else if (CODE64(s)) { + return tcg_constant_tl(s->base.pc_next); } else { - return tcg_constant_tl(s->base.pc_next - s->cs_base); + return tcg_constant_tl((uint32_t)(s->base.pc_next - s->cs_base)); } } @@ -2837,6 +2845,10 @@ static void gen_jmp_rel(DisasContext *s, MemOp ot, int diff, int tb_num) } } new_eip &= mask; + new_pc = new_eip + s->cs_base; + if (!CODE64(s)) { + new_pc = (uint32_t)new_pc; + } gen_update_cc_op(s); set_cc_op(s, CC_OP_DYNAMIC); @@ -2854,8 +2866,7 @@ static void gen_jmp_rel(DisasContext *s, MemOp ot, int diff, int tb_num) } } - if (use_goto_tb && - translator_use_goto_tb(&s->base, new_eip + s->cs_base)) { + if (use_goto_tb && translator_use_goto_tb(&s->base, new_pc)) { /* jump to same page: we can use a direct jump */ tcg_gen_goto_tb(tb_num); if (!(tb_cflags(s->base.tb) & CF_PCREL)) { From cbb145567c21d2bc41e8197a243c7d5480c6fca0 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Tue, 12 Dec 2023 11:35:42 -0800 Subject: [PATCH 2/2] tcg: Reduce serial context atomicity earlier MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 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 Tested-by: Alex Bennée Message-Id: <20231212193542.149117-1-richard.henderson@linaro.org> --- tcg/tcg-op-ldst.c | 28 ++++++++++++++++++++++++---- tcg/tcg.c | 9 +-------- 2 files changed, 25 insertions(+), 12 deletions(-) diff --git a/tcg/tcg-op-ldst.c b/tcg/tcg-op-ldst.c index e2c55df217..f11043b449 100644 --- a/tcg/tcg-op-ldst.c +++ b/tcg/tcg-op-ldst.c @@ -77,6 +77,13 @@ static MemOp tcg_canonicalize_memop(MemOp op, bool is64, bool st) if (st) { 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; } @@ -428,8 +435,7 @@ static bool use_two_i64_for_i128(MemOp mop) case MO_ATOM_SUBALIGN: case MO_ATOM_WITHIN16: case MO_ATOM_WITHIN16_PAIR: - /* In a serialized context, no atomicity is required. */ - return !(tcg_ctx->gen_tb->cflags & CF_PARALLEL); + return false; default: 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, TCGArg idx, MemOp memop) { - const MemOpIdx orig_oi = make_memop_idx(memop, idx); + MemOpIdx orig_oi; TCGv_i64 ext_addr = NULL; TCGOpcode opc; check_max_alignment(get_alignment_bits(memop)); 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. */ if (TCG_TARGET_HAS_qemu_ldst_i128 && TCG_TARGET_REG_BITS == 64) { 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, TCGArg idx, MemOp memop) { - const MemOpIdx orig_oi = make_memop_idx(memop, idx); + MemOpIdx orig_oi; TCGv_i64 ext_addr = NULL; TCGOpcode opc; check_max_alignment(get_alignment_bits(memop)); 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. */ if (TCG_TARGET_HAS_qemu_ldst_i128 && TCG_TARGET_REG_BITS == 64) { diff --git a/tcg/tcg.c b/tcg/tcg.c index d2ea22b397..896a36caeb 100644 --- a/tcg/tcg.c +++ b/tcg/tcg.c @@ -5440,15 +5440,8 @@ static TCGAtomAlign atom_and_align_for_opc(TCGContext *s, MemOp opc, MemOp align = get_alignment_bits(opc); MemOp size = opc & MO_SIZE; MemOp half = size ? size - 1 : 0; + MemOp atom = opc & MO_ATOM_MASK; 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) { case MO_ATOM_NONE: