diff options
author | Richard Henderson <rth@twiddle.net> | 2016-08-31 10:48:03 -0700 |
---|---|---|
committer | Richard Henderson <rth@twiddle.net> | 2016-10-31 14:46:47 -0600 |
commit | cb21b4da6cca1bb4e3f5fefb698fb9e4d00c8f66 (patch) | |
tree | 95a723b84dc504bacfe90ac377d2668f2632744a /target-sparc | |
parent | f939ffe5a022a8798824e2720ed5a14186fca6b6 (diff) |
target-sparc: Allow 4-byte alignment on fp mem ops
The cpu is allowed to require stricter alignment on these 8- and 16-byte
operations, and the OS is required to fix up the accesses as necessary,
so the previous code was not wrong.
However, we can easily handle this misalignment for all direct 8-byte
operations and for direct 16-byte loads.
We must retain 16-byte alignment for 16-byte stores, so that we don't have
to probe for writability of a second page before performing the first of
two 8-byte stores. We also retain 8-byte alignment for no-fault loads,
since they are rare and it's not worth extending the helpers for this.
Tested-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Diffstat (limited to 'target-sparc')
-rw-r--r-- | target-sparc/translate.c | 86 |
1 files changed, 44 insertions, 42 deletions
diff --git a/target-sparc/translate.c b/target-sparc/translate.c index a869d7d6dc..4920cf28e7 100644 --- a/target-sparc/translate.c +++ b/target-sparc/translate.c @@ -2459,6 +2459,7 @@ static void gen_ldf_asi(DisasContext *dc, TCGv addr, { DisasASI da = get_asi(dc, insn, (size == 4 ? MO_TEUL : MO_TEQ)); TCGv_i32 d32; + TCGv_i64 d64; switch (da.type) { case GET_ASI_EXCP: @@ -2473,12 +2474,17 @@ static void gen_ldf_asi(DisasContext *dc, TCGv addr, gen_store_fpr_F(dc, rd, d32); break; case 8: - tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da.mem_idx, da.memop); + tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da.mem_idx, + da.memop | MO_ALIGN_4); break; case 16: - tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da.mem_idx, da.memop); + d64 = tcg_temp_new_i64(); + tcg_gen_qemu_ld_i64(d64, addr, da.mem_idx, da.memop | MO_ALIGN_4); tcg_gen_addi_tl(addr, addr, 8); - tcg_gen_qemu_ld_i64(cpu_fpr[rd/2+1], addr, da.mem_idx, da.memop); + tcg_gen_qemu_ld_i64(cpu_fpr[rd/2+1], addr, da.mem_idx, + da.memop | MO_ALIGN_4); + tcg_gen_mov_i64(cpu_fpr[rd / 2], d64); + tcg_temp_free_i64(d64); break; default: g_assert_not_reached(); @@ -2534,22 +2540,23 @@ static void gen_ldf_asi(DisasContext *dc, TCGv addr, but we can just use the integer asi helper for them. */ switch (size) { case 4: - { - TCGv d64 = tcg_temp_new_i64(); - gen_helper_ld_asi(d64, cpu_env, addr, r_asi, r_mop); - d32 = gen_dest_fpr_F(dc); - tcg_gen_extrl_i64_i32(d32, d64); - tcg_temp_free_i64(d64); - gen_store_fpr_F(dc, rd, d32); - } + d64 = tcg_temp_new_i64(); + gen_helper_ld_asi(d64, cpu_env, addr, r_asi, r_mop); + d32 = gen_dest_fpr_F(dc); + tcg_gen_extrl_i64_i32(d32, d64); + tcg_temp_free_i64(d64); + gen_store_fpr_F(dc, rd, d32); break; case 8: gen_helper_ld_asi(cpu_fpr[rd / 2], cpu_env, addr, r_asi, r_mop); break; case 16: - gen_helper_ld_asi(cpu_fpr[rd / 2], cpu_env, addr, r_asi, r_mop); + d64 = tcg_temp_new_i64(); + gen_helper_ld_asi(d64, cpu_env, addr, r_asi, r_mop); tcg_gen_addi_tl(addr, addr, 8); gen_helper_ld_asi(cpu_fpr[rd/2+1], cpu_env, addr, r_asi, r_mop); + tcg_gen_mov_i64(cpu_fpr[rd / 2], d64); + tcg_temp_free_i64(d64); break; default: g_assert_not_reached(); @@ -2579,15 +2586,15 @@ static void gen_stf_asi(DisasContext *dc, TCGv addr, tcg_gen_qemu_st_i32(d32, addr, da.mem_idx, da.memop); break; case 8: - /* ??? Only 4-byte alignment required. However, it is legal - for the cpu to signal the alignment fault, and the OS trap - handler is required to fix it up. */ - tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx, da.memop); + tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx, + da.memop | MO_ALIGN_4); break; case 16: - /* Only 4-byte alignment required. See above. Requiring - 16-byte alignment here avoids having to probe the second - page before performing the first write. */ + /* Only 4-byte alignment required. However, it is legal for the + cpu to signal the alignment fault, and the OS trap handler is + required to fix it up. Requiring 16-byte alignment here avoids + having to probe the second page before performing the first + write. */ tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx, da.memop | MO_ALIGN_16); tcg_gen_addi_tl(addr, addr, 8); @@ -5423,18 +5430,15 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn) skip_move: ; #endif } else if (xop >= 0x20 && xop < 0x24) { - TCGv t0; - if (gen_trap_ifnofpu(dc)) { goto jmp_insn; } switch (xop) { case 0x20: /* ldf, load fpreg */ gen_address_mask(dc, cpu_addr); - t0 = get_temp_tl(dc); - tcg_gen_qemu_ld32u(t0, cpu_addr, dc->mem_idx); cpu_dst_32 = gen_dest_fpr_F(dc); - tcg_gen_trunc_tl_i32(cpu_dst_32, t0); + tcg_gen_qemu_ld_i32(cpu_dst_32, cpu_addr, + dc->mem_idx, MO_TEUL); gen_store_fpr_F(dc, rd, cpu_dst_32); break; case 0x21: /* ldfsr, V9 ldxfsr */ @@ -5442,26 +5446,28 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn) gen_address_mask(dc, cpu_addr); if (rd == 1) { TCGv_i64 t64 = tcg_temp_new_i64(); - tcg_gen_qemu_ld64(t64, cpu_addr, dc->mem_idx); + tcg_gen_qemu_ld_i64(t64, cpu_addr, + dc->mem_idx, MO_TEQ); gen_helper_ldxfsr(cpu_fsr, cpu_env, cpu_fsr, t64); tcg_temp_free_i64(t64); break; } #endif cpu_dst_32 = get_temp_i32(dc); - t0 = get_temp_tl(dc); - tcg_gen_qemu_ld32u(t0, cpu_addr, dc->mem_idx); - tcg_gen_trunc_tl_i32(cpu_dst_32, t0); + tcg_gen_qemu_ld_i32(cpu_dst_32, cpu_addr, + dc->mem_idx, MO_TEUL); gen_helper_ldfsr(cpu_fsr, cpu_env, cpu_fsr, cpu_dst_32); break; case 0x22: /* ldqf, load quad fpreg */ CHECK_FPU_FEATURE(dc, FLOAT128); gen_address_mask(dc, cpu_addr); cpu_src1_64 = tcg_temp_new_i64(); - tcg_gen_qemu_ld64(cpu_src1_64, cpu_addr, dc->mem_idx); + tcg_gen_qemu_ld_i64(cpu_src1_64, cpu_addr, dc->mem_idx, + MO_TEQ | MO_ALIGN_4); tcg_gen_addi_tl(cpu_addr, cpu_addr, 8); cpu_src2_64 = tcg_temp_new_i64(); - tcg_gen_qemu_ld64(cpu_src2_64, cpu_addr, dc->mem_idx); + tcg_gen_qemu_ld_i64(cpu_src2_64, cpu_addr, dc->mem_idx, + MO_TEQ | MO_ALIGN_4); gen_store_fpr_Q(dc, rd, cpu_src1_64, cpu_src2_64); tcg_temp_free_i64(cpu_src1_64); tcg_temp_free_i64(cpu_src2_64); @@ -5469,7 +5475,8 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn) case 0x23: /* lddf, load double fpreg */ gen_address_mask(dc, cpu_addr); cpu_dst_64 = gen_dest_fpr_D(dc, rd); - tcg_gen_qemu_ld64(cpu_dst_64, cpu_addr, dc->mem_idx); + tcg_gen_qemu_ld_i64(cpu_dst_64, cpu_addr, dc->mem_idx, + MO_TEQ | MO_ALIGN_4); gen_store_fpr_D(dc, rd, cpu_dst_64); break; default: @@ -5542,13 +5549,10 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn) } switch (xop) { case 0x24: /* stf, store fpreg */ - { - TCGv t = get_temp_tl(dc); - gen_address_mask(dc, cpu_addr); - cpu_src1_32 = gen_load_fpr_F(dc, rd); - tcg_gen_ext_i32_tl(t, cpu_src1_32); - tcg_gen_qemu_st32(t, cpu_addr, dc->mem_idx); - } + gen_address_mask(dc, cpu_addr); + cpu_src1_32 = gen_load_fpr_F(dc, rd); + tcg_gen_qemu_st_i32(cpu_src1_32, cpu_addr, + dc->mem_idx, MO_TEUL); break; case 0x25: /* stfsr, V9 stxfsr */ { @@ -5594,12 +5598,10 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn) #endif #endif case 0x27: /* stdf, store double fpreg */ - /* ??? Only 4-byte alignment required. However, it is - legal for the cpu to signal the alignment fault, and - the OS trap handler is required to fix it up. */ gen_address_mask(dc, cpu_addr); cpu_src1_64 = gen_load_fpr_D(dc, rd); - tcg_gen_qemu_st64(cpu_src1_64, cpu_addr, dc->mem_idx); + tcg_gen_qemu_st_i64(cpu_src1_64, cpu_addr, dc->mem_idx, + MO_TEQ | MO_ALIGN_4); break; default: goto illegal_insn; |