target/arm: single-binary
Subsystems (2)
v10 cover target/arm: single-binary
This series tackles (most of) the rest of target/arm, especially tcg folder.
We extract tcg memory operation in a new header, with a new type representing virtual addresses.
Initially, I went down the way to have a dynamic definition of TCGv, which required much more boilerplate and runtime checks. After discussing with Richard, I realized it was not needed, and that could simply split 32 and 64 bit properly in different files instead. This approach will not work with other architectures, requiring something else, but it's enough for target/arm.
Files left are:
- target/arm/tcg/mve_helper.c
- target/arm/tcg/m_helper.c
- target/arm/tcg/cpu32.c
v10
---
- extract patch for endianness handling in arm_disas_set_info
- added some comments
v9
--
- introduce aarch64_translate_code *before* adding parameter to translator_loop
- use TCG_TYPE_VA as address type for translator_loop
- remove inline marker on functions merged from arm_ldst.h to translate.c
- refactor arm_ldl_code/arm_lduw_code
v8
--
- make translate.c compilation unit common.
v7
--
- add parameter to translator_loop to indicate address type for current tb.
- remove TCGv from target/arm/tcg/translate.c
v6
--
- meson.build: extract decode files in separate variables and explicitly include them in user and system code. The duplication is now minimized: one instance for all system binaries, and one per user binary.
v5
--
- reimplemented what v0 did with new approach defining new tcg-op-mem header
- reapply all straightforward patches from v0
- only translate.c is left, but first want to validate that changes here are ok for maintainers before doing more.
v0
--
See original thread:
https://lore.kernel.org/qemu-devel/[email protected]/#t
Note about v1 -> v4
-------------------
- After v0 which was trying to do too many changes, v1 to v4 focused on smaller things, and all patches have been merged already.
https://patchew.org/QEMU/[email protected]/
Pierrick Bouvier (21):
include/tcg/tcg-op: extract memory operations to tcg-op-mem.h target/arm/translate.h: remove tcg-op.h include target/arm/tcg/translate.h: remove tcg-op-gvec.h include target/arm/tcg/translate.h: remove TARGET_AARCH64 target/arm/tcg/translate-vfp.c: make compilation unit common target/arm/tcg/translate-neon.c: make compilation unit common target/arm/tcg/translate-mve.c: make compilation unit common target/arm/tcg/translate-m-nocp.c: make compilation unit common target/arm/tcg/op_helper.c: make compilation unit common target/arm/tcg/gengvec.c: make compilation unit common target/arm/tcg/translate.c: remove MO_TE usage target/arm/tcg/translate.c: replace target_ulong with uint32_t target/arm/tcg/translate.c: extract aarch64_translate_code() tcg/translator: add parameter to translator_loop for current addr type target/arm/tcg/translate.c: replace TCGv with TCGv_va target/arm/tcg/translate-a64.c: use translator_ldl_end instead of
arm_ldl_code
target/arm/tcg/arm_ldst.h: merge in translate.c target/arm/tcg/translate.c: replace translator_ldl_swap with
translator_ldl_end
target/arm/cpu.c: simplify endianness handling in arm_disas_set_info target/arm/tcg/translate.c: remove target_ulong target/arm/tcg/translate.c: make compilation unit common
include/exec/translator.h | 4 +- include/tcg/tcg-op-mem.h | 126 +++++++++++++++++++++++++++++++ include/tcg/tcg-op.h | 100 +----------------------- target/arm/internals.h | 2 + target/arm/tcg/arm_ldst.h | 47 ------------ target/arm/tcg/translate.h | 14 +--- accel/tcg/translate-all.c | 1 - accel/tcg/translator.c | 4 +- target/alpha/translate.c | 3 +- target/arm/cpu.c | 6 +- target/arm/tcg/op_helper.c | 1 - target/arm/tcg/stubs32.c | 24 ++++++ target/arm/tcg/translate-a64.c | 17 ++++- target/arm/tcg/translate-sme.c | 1 + target/arm/tcg/translate-sve.c | 1 + target/arm/tcg/translate.c | 73 +++++++++++------- target/avr/translate.c | 3 +- target/hexagon/translate.c | 3 +- target/hppa/translate.c | 3 +- target/i386/tcg/translate.c | 3 +- target/loongarch/tcg/translate.c | 3 +- target/m68k/translate.c | 3 +- target/microblaze/translate.c | 3 +- target/mips/tcg/translate.c | 3 +- target/or1k/translate.c | 3 +- target/ppc/translate.c | 3 +- target/riscv/translate.c | 3 +- target/rx/translate.c | 3 +- target/s390x/tcg/translate.c | 3 +- target/sh4/translate.c | 3 +- target/sparc/translate.c | 3 +- target/tricore/translate.c | 3 +- target/xtensa/translate.c | 3 +- target/arm/tcg/meson.build | 57 ++++++++++---- 34 files changed, 304 insertions(+), 228 deletions(-) create mode 100644 include/tcg/tcg-op-mem.h create mode 100644 target/arm/tcg/stubs32.c
Signature · 1 line
2.47.3
PATCH v10 1/21 include/tcg/tcg-op: extract memory operations to tcg-op-mem.h
This new header defines a new type for target virtual address, independent from TCGv and is parameterized by a new define TCG_ADDRESS_BITS (name was suggested by Paolo instead of TARGET_ADDRESS_BITS).
By default, tcg-op.h include set this define to TARGET_LONG_BITS, but it's also possible to include only tcg-op-common.h and tcg-op-mem.h and set TCG_ADDRESS_BITS manually, which is what next commits will do.
We preserve existing MIT license when extracting this new header.
Implemented from:
https://lore.kernel.org/qemu-devel/[email protected]/
--- include/tcg/tcg-op-mem.h | 126 +++++++++++++++++++++++++++++++++++++++ include/tcg/tcg-op.h | 100 +------------------------------ 2 files changed, 129 insertions(+), 97 deletions(-) create mode 100644 include/tcg/tcg-op-mem.h
include/tcg/tcg-op-mem.h new file +126 -0
@@ -0,0 +1,126 @@ +/* SPDX-License-Identifier: MIT */ +/* +* Target dependent memory related functions. +* +* Copyright (c) 2008 Fabrice Bellard +*/ + +#ifndef TCG_TCG_OP_MEM_H +#define TCG_TCG_OP_MEM_H + +#ifndef TCG_ADDRESS_BITS +#error TCG_ADDRESS_BITS must be defined +#endif + +#if TCG_ADDRESS_BITS ==32 +typedef TCGv_i32 TCGv_va ; +#define TCG_TYPE_VA TCG_TYPE_I32 +#define tcgv_va_temp tcgv_i32_temp +#define tcgv_va_temp_new tcg_temp_new_i32 +#elif TCG_ADDRESS_BITS ==64 +typedef TCGv_i64 TCGv_va ; +#define TCG_TYPE_VA TCG_TYPE_I64 +#define tcgv_va_temp tcgv_i64_temp +#define tcgv_va_temp_new tcg_temp_new_i64 +#else +#error +#endif + +static inline void +tcg_gen_qemu_ld_i32 (TCGv_i32 v,TCGv_va a,TCGArg i,MemOp m) +{ +tcg_gen_qemu_ld_i32_chk (v,tcgv_va_temp (a), i, m,TCG_TYPE_VA ); +} + +static inline void +tcg_gen_qemu_st_i32 (TCGv_i32 v,TCGv_va a,TCGArg i,MemOp m) +{ +tcg_gen_qemu_st_i32_chk (v,tcgv_va_temp (a), i, m,TCG_TYPE_VA ); +} + +static inline void +tcg_gen_qemu_ld_i64 (TCGv_i64 v,TCGv_va a,TCGArg i,MemOp m) +{ +tcg_gen_qemu_ld_i64_chk (v,tcgv_va_temp (a), i, m,TCG_TYPE_VA ); +} + +static inline void +tcg_gen_qemu_st_i64 (TCGv_i64 v,TCGv_va a,TCGArg i,MemOp m) +{ +tcg_gen_qemu_st_i64_chk (v,tcgv_va_temp (a), i, m,TCG_TYPE_VA ); +} + +static inline void +tcg_gen_qemu_ld_i128 (TCGv_i128 v,TCGv_va a,TCGArg i,MemOp m) +{ +tcg_gen_qemu_ld_i128_chk (v,tcgv_va_temp (a), i, m,TCG_TYPE_VA ); +} + +static inline void +tcg_gen_qemu_st_i128 (TCGv_i128 v,TCGv_va a,TCGArg i,MemOp m) +{ +tcg_gen_qemu_st_i128_chk (v,tcgv_va_temp (a), i, m,TCG_TYPE_VA ); +} + +#define DEF_ATOMIC2 (N ,S ) \ + static inline void N##_##S(TCGv_##S r, TCGv_va a, TCGv_##S v, \ + TCGArg i, MemOp m) \ + { N##_##S##_chk(r, tcgv_va_temp(a), v, i, m, TCG_TYPE_VA); } + +#define DEF_ATOMIC3 (N ,S ) \ + static inline void N##_##S(TCGv_##S r, TCGv_va a, TCGv_##S o, \ + TCGv_##S n, TCGArg i, MemOp m) \ + { N##_##S##_chk(r, tcgv_va_temp(a), o, n, i, m, TCG_TYPE_VA); } + +DEF_ATOMIC3(tcg_gen_atomic_cmpxchg, i32) +DEF_ATOMIC3(tcg_gen_atomic_cmpxchg, i64) +DEF_ATOMIC3(tcg_gen_atomic_cmpxchg, i128) + +DEF_ATOMIC3(tcg_gen_nonatomic_cmpxchg, i32) +DEF_ATOMIC3(tcg_gen_nonatomic_cmpxchg, i64) +DEF_ATOMIC3(tcg_gen_nonatomic_cmpxchg, i128) + +DEF_ATOMIC2(tcg_gen_atomic_xchg, i32) +DEF_ATOMIC2(tcg_gen_atomic_xchg, i64) +DEF_ATOMIC2(tcg_gen_atomic_xchg, i128) + +DEF_ATOMIC2(tcg_gen_atomic_fetch_add, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_add, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_and, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_and, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_and, i128) +DEF_ATOMIC2(tcg_gen_atomic_fetch_or, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_or, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_or, i128) +DEF_ATOMIC2(tcg_gen_atomic_fetch_xor, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_xor, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_smin, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_smin, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_umin, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_umin, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_smax, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_smax, i64) +DEF_ATOMIC2(tcg_gen_atomic_fetch_umax, i32) +DEF_ATOMIC2(tcg_gen_atomic_fetch_umax, i64) + +DEF_ATOMIC2(tcg_gen_atomic_add_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_add_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_and_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_and_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_or_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_or_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_xor_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_xor_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_smin_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_smin_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_umin_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_umin_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_smax_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_smax_fetch, i64) +DEF_ATOMIC2(tcg_gen_atomic_umax_fetch, i32) +DEF_ATOMIC2(tcg_gen_atomic_umax_fetch, i64) + +#undef DEF_ATOMIC2 +#undef DEF_ATOMIC3 + +#endif /* TCG_TCG_OP_MEM_H */
include/tcg/tcg-op.h +3 -97
112 lines not on this page · raw message
Signature · 1 line
2.47.3
PATCH v10 2/21 target/arm/translate.h: remove tcg-op.h include
In reply to Pierrick BouvierInclude tcg-op-common.h instead, and include target specific tcg-op.h in files needing it. This intermediate step allows to clean up every file (TCGv, tcg_gen.*_tl.*) in separate commits.
--- target/arm/tcg/translate.h | 2 +- target/arm/tcg/translate-a64.c | 1 + target/arm/tcg/translate-sme.c | 1 + target/arm/tcg/translate-sve.c | 1 + target/arm/tcg/translate.c | 1 + 5 files changed, 5 insertions(+), 1 deletion(-)
target/arm/tcg/translate.h +1 -1
@@ -2,7 +2,7 @@#define TARGET_ARM_TRANSLATE_H #include "cpu.h" -#include "tcg/tcg-op.h" +#include "tcg/tcg-op-common.h" #include "tcg/tcg-op-gvec.h" #include "exec/translator.h" #include "exec/translation-block.h"
target/arm/tcg/translate-a64.c +1 -0
@@ -23,6 +23,7 @@#include "helper-sve.h" #include "translate.h" #include "translate-a64.h" +#include "tcg/tcg-op.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h"
target/arm/tcg/translate-sme.c +1 -0
@@ -23,6 +23,7 @@#include "helper-sve.h" #include "translate.h" #include "translate-a64.h" +#include "tcg/tcg-op.h" /* * Include the generated decoder.
target/arm/tcg/translate-sve.c +1 -0
@@ -23,6 +23,7 @@#include "helper-sve.h" #include "translate.h" #include "translate-a64.h" +#include "tcg/tcg-op.h" #include "fpu/softfloat.h"
target/arm/tcg/translate.c +1 -0
@@ -22,6 +22,7 @@#include "translate.h" #include "translate-a32.h" +#include "tcg/tcg-op.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h"
Signature · 1 line
2.47.3
PATCH v10 3/21 target/arm/tcg/translate.h: remove tcg-op-gvec.h include
In reply to Pierrick BouvierInclude tcg-op-gvec-common.h instead.
No target/arm code rely on target specifics for gvec ops.
--- target/arm/tcg/translate.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
target/arm/tcg/translate.h +1 -1
@@ -3,7 +3,7 @@#include "cpu.h" #include "tcg/tcg-op-common.h" -#include "tcg/tcg-op-gvec.h" +#include "tcg/tcg-op-gvec-common.h" #include "exec/translator.h" #include "exec/translation-block.h" #include "helper.h"
Signature · 1 line
2.47.3
PATCH v10 4/21 target/arm/tcg/translate.h: remove TARGET_AARCH64
In reply to Pierrick BouvierWe need to stub a64_translate_init and gen_a64_update_pc. At this point, we don't need to do anything for aarch64_translator_ops since it's just an external symbol.
We can now include target/arm/tcg/translate.h from common code, since all target specific bits have been removed, or can be specialized with specific defines.
--- target/arm/tcg/translate.h | 10 ---------- target/arm/tcg/stubs32.c | 17 +++++++++++++++++ target/arm/tcg/meson.build | 1 + 3 files changed, 18 insertions(+), 10 deletions(-) create mode 100644 target/arm/tcg/stubs32.c
target/arm/tcg/translate.h +0 -10
@@ -357,19 +357,9 @@ static inline int curr_insn_len(DisasContext *s)/* CPU state was modified dynamically; no need to exit, but do not chain. */ #define DISAS_UPDATE_NOCHAIN DISAS_TARGET_10 -#ifdef TARGET_AARCH64 void a64_translate_init (void );void gen_a64_update_pc (DisasContext *s,int64_t diff);extern const TranslatorOps aarch64_translator_ops; -#else -static inline void a64_translate_init (void ) -{ -} - -static inline void gen_a64_update_pc (DisasContext *s,int64_t diff) -{ -} -#endif void arm_test_cc (DisasCompare *cmp,int cc);void arm_jump_cc (DisasCompare *cmp,TCGLabel *label);
target/arm/tcg/stubs32.c new file +17 -0
@@ -0,0 +1,17 @@ +/* +* SPDX-License-Identifier: GPL-2.0-or-later +*/ + +#include "qemu/osdep.h" +#include "target/arm/tcg/translate.h" + + +void gen_a64_update_pc (DisasContext *s,int64_t diff) +{ +g_assert_not_reached (); +} + +void a64_translate_init (void ) +{ +/* Don't initialize for 32 bits. Call site will be fixed later. */ +}
target/arm/tcg/meson.build +1 -0
@@ -21,6 +21,7 @@ gen_a32 = [ arm_ss.add(gen_a32) arm_ss.add(when: 'TARGET_AARCH64', if_true: gen_a64) +arm_ss.add(when: 'TARGET_AARCH64', if_false: files('stubs32.c')) arm_ss.add(files( 'cpu32.c',
Signature · 1 line
2.47.3
PATCH v10 5/21 target/arm/tcg/translate-vfp.c: make compilation unit common
In reply to Pierrick BouvierGenerated decode files must be duplicated between user and system, as they are generated in private folders per libs, and can't be included otherwise, as meson does not give control on output folder. Indeed, meson generator is a different approach than custom_target, and this is a limitation by design.
They were already duplicated between arch variants anyway, so nothing new here. They will now be compiled once for system binaries, and still per target for user binaries.
--- target/arm/tcg/meson.build | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-)
target/arm/tcg/meson.build +13 -5
@@ -5,12 +5,15 @@ gen_a64 = [ decodetree.process('sme-fa64.decode', extra_args: '--static-decode=disas_sme_fa64'), ] +vfp_d = [ + decodetree.process('vfp.decode', extra_args: '--decode=disas_vfp'), + decodetree.process('vfp-uncond.decode', extra_args: '--decode=disas_vfp_uncond'), +] + gen_a32 = [ decodetree.process('neon-shared.decode', extra_args: '--decode=disas_neon_shared'), decodetree.process('neon-dp.decode', extra_args: '--decode=disas_neon_dp'), decodetree.process('neon-ls.decode', extra_args: '--decode=disas_neon_ls'), - decodetree.process('vfp.decode', extra_args: '--decode=disas_vfp'), - decodetree.process('vfp-uncond.decode', extra_args: '--decode=disas_vfp_uncond'), decodetree.process('m-nocp.decode', extra_args: '--decode=disas_m_nocp'), decodetree.process('mve.decode', extra_args: '--decode=disas_mve'), decodetree.process('a32.decode', extra_args: '--static-decode=disas_a32'),
@@ -30,7 +33,6 @@ arm_ss.add(files( 'translate-m-nocp.c', 'translate-mve.c', 'translate-neon.c', - 'translate-vfp.c', 'm_helper.c', 'mve_helper.c', 'op_helper.c',
@@ -60,7 +62,9 @@ arm_common_ss.add(files( 'crypto_helper.c', )) -arm_common_system_ss.add(files( +arm_common_system_ss.add( + vfp_d, + files( 'cpregs-at.c', 'debug.c', 'hflags.c',
@@ -68,14 +72,18 @@ arm_common_system_ss.add(files( 'psci.c', 'tlb_helper.c', 'tlb-insns.c', + 'translate-vfp.c', 'vec_helper.c', 'vfp_helper.c', )) -arm_user_ss.add(files( +arm_user_ss.add( + vfp_d, + files( 'debug.c', 'hflags.c', 'neon_helper.c', 'tlb_helper.c', + 'translate-vfp.c', 'vec_helper.c', 'vfp_helper.c', ))
Signature · 1 line
2.47.3
PATCH v10 6/21 target/arm/tcg/translate-neon.c: make compilation unit common
In reply to Pierrick Bouvier--- target/arm/tcg/meson.build | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
target/arm/tcg/meson.build +8 -2
@@ -10,10 +10,13 @@ vfp_d = [ decodetree.process('vfp-uncond.decode', extra_args: '--decode=disas_vfp_uncond'), ] -gen_a32 = [ +neon_d = [ decodetree.process('neon-shared.decode', extra_args: '--decode=disas_neon_shared'), decodetree.process('neon-dp.decode', extra_args: '--decode=disas_neon_dp'), decodetree.process('neon-ls.decode', extra_args: '--decode=disas_neon_ls'), +] + +gen_a32 = [ decodetree.process('m-nocp.decode', extra_args: '--decode=disas_m_nocp'), decodetree.process('mve.decode', extra_args: '--decode=disas_mve'), decodetree.process('a32.decode', extra_args: '--static-decode=disas_a32'),
@@ -32,7 +35,6 @@ arm_ss.add(files( 'translate.c', 'translate-m-nocp.c', 'translate-mve.c', - 'translate-neon.c', 'm_helper.c', 'mve_helper.c', 'op_helper.c',
@@ -63,6 +65,7 @@ arm_common_ss.add(files( )) arm_common_system_ss.add( + neon_d, vfp_d, files( 'cpregs-at.c',
@@ -72,17 +75,20 @@ arm_common_system_ss.add( 'psci.c', 'tlb_helper.c', 'tlb-insns.c', + 'translate-neon.c', 'translate-vfp.c', 'vec_helper.c', 'vfp_helper.c', )) arm_user_ss.add( + neon_d, vfp_d, files( 'debug.c', 'hflags.c', 'neon_helper.c', 'tlb_helper.c', + 'translate-neon.c', 'translate-vfp.c', 'vec_helper.c', 'vfp_helper.c',
Signature · 1 line
2.47.3
PATCH v10 7/21 target/arm/tcg/translate-mve.c: make compilation unit common
In reply to Pierrick Bouvier--- target/arm/tcg/meson.build | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
target/arm/tcg/meson.build +6 -2
@@ -16,9 +16,10 @@ neon_d = [ decodetree.process('neon-ls.decode', extra_args: '--decode=disas_neon_ls'), ] +mve_d = decodetree.process('mve.decode', extra_args: '--decode=disas_mve') + gen_a32 = [ decodetree.process('m-nocp.decode', extra_args: '--decode=disas_m_nocp'), - decodetree.process('mve.decode', extra_args: '--decode=disas_mve'), decodetree.process('a32.decode', extra_args: '--static-decode=disas_a32'), decodetree.process('a32-uncond.decode', extra_args: '--static-decode=disas_a32_uncond'), decodetree.process('t32.decode', extra_args: '--static-decode=disas_t32'),
@@ -34,7 +35,6 @@ arm_ss.add(files( 'gengvec.c', 'translate.c', 'translate-m-nocp.c', - 'translate-mve.c', 'm_helper.c', 'mve_helper.c', 'op_helper.c',
@@ -65,6 +65,7 @@ arm_common_ss.add(files( )) arm_common_system_ss.add( + mve_d, neon_d, vfp_d, files(
@@ -75,12 +76,14 @@ arm_common_system_ss.add( 'psci.c', 'tlb_helper.c', 'tlb-insns.c', + 'translate-mve.c', 'translate-neon.c', 'translate-vfp.c', 'vec_helper.c', 'vfp_helper.c', )) arm_user_ss.add( + mve_d, neon_d, vfp_d, files(
@@ -88,6 +91,7 @@ arm_user_ss.add( 'hflags.c', 'neon_helper.c', 'tlb_helper.c', + 'translate-mve.c', 'translate-neon.c', 'translate-vfp.c', 'vec_helper.c',
Signature · 1 line
2.47.3
PATCH v10 8/21 target/arm/tcg/translate-m-nocp.c: make compilation unit common
In reply to Pierrick Bouvier--- target/arm/tcg/meson.build | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
target/arm/tcg/meson.build +6 -2
@@ -18,8 +18,9 @@ neon_d = [ mve_d = decodetree.process('mve.decode', extra_args: '--decode=disas_mve') +m_nocp_d = decodetree.process('m-nocp.decode', extra_args: '--decode=disas_m_nocp') + gen_a32 = [ - decodetree.process('m-nocp.decode', extra_args: '--decode=disas_m_nocp'), decodetree.process('a32.decode', extra_args: '--static-decode=disas_a32'), decodetree.process('a32-uncond.decode', extra_args: '--static-decode=disas_a32_uncond'), decodetree.process('t32.decode', extra_args: '--static-decode=disas_t32'),
@@ -34,7 +35,6 @@ arm_ss.add(files( 'cpu32.c', 'gengvec.c', 'translate.c', - 'translate-m-nocp.c', 'm_helper.c', 'mve_helper.c', 'op_helper.c',
@@ -65,6 +65,7 @@ arm_common_ss.add(files( )) arm_common_system_ss.add( + m_nocp_d, mve_d, neon_d, vfp_d,
@@ -76,6 +77,7 @@ arm_common_system_ss.add( 'psci.c', 'tlb_helper.c', 'tlb-insns.c', + 'translate-m-nocp.c', 'translate-mve.c', 'translate-neon.c', 'translate-vfp.c',
@@ -83,6 +85,7 @@ arm_common_system_ss.add( 'vfp_helper.c', )) arm_user_ss.add( + m_nocp_d, mve_d, neon_d, vfp_d,
@@ -91,6 +94,7 @@ arm_user_ss.add( 'hflags.c', 'neon_helper.c', 'tlb_helper.c', + 'translate-m-nocp.c', 'translate-mve.c', 'translate-neon.c', 'translate-vfp.c',
Signature · 1 line
2.47.3
PATCH v10 9/21 target/arm/tcg/op_helper.c: make compilation unit common
In reply to Pierrick BouvierRemove unused header accel/tcg/cpu-ldst.h that has target specifics.
--- target/arm/tcg/op_helper.c | 1 - target/arm/tcg/meson.build | 3 ++- 2 files changed, 2 insertions(+), 2 deletions(-)
target/arm/tcg/op_helper.c +0 -1
@@ -23,7 +23,6 @@#include "helper.h" #include "internals.h" #include "cpu-features.h" -#include "accel/tcg/cpu-ldst.h" #include "accel/tcg/probe.h" #include "cpregs.h"
target/arm/tcg/meson.build +2 -1
@@ -37,7 +37,6 @@ arm_ss.add(files( 'translate.c', 'm_helper.c', 'mve_helper.c', - 'op_helper.c', )) arm_ss.add(when: 'TARGET_AARCH64', if_true: files(
@@ -74,6 +73,7 @@ arm_common_system_ss.add( 'debug.c', 'hflags.c', 'neon_helper.c', + 'op_helper.c', 'psci.c', 'tlb_helper.c', 'tlb-insns.c',
@@ -93,6 +93,7 @@ arm_user_ss.add( 'debug.c', 'hflags.c', 'neon_helper.c', + 'op_helper.c', 'tlb_helper.c', 'translate-m-nocp.c', 'translate-mve.c',
Signature · 1 line
2.47.3
PATCH v10 10/21 target/arm/tcg/gengvec.c: make compilation unit common
In reply to Pierrick Bouvier--- target/arm/tcg/meson.build | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
target/arm/tcg/meson.build +2 -1
@@ -33,7 +33,6 @@ arm_ss.add(when: 'TARGET_AARCH64', if_false: files('stubs32.c')) arm_ss.add(files( 'cpu32.c', - 'gengvec.c', 'translate.c', 'm_helper.c', 'mve_helper.c',
@@ -72,6 +71,7 @@ arm_common_system_ss.add( 'cpregs-at.c', 'debug.c', 'hflags.c', + 'gengvec.c', 'neon_helper.c', 'op_helper.c', 'psci.c',
@@ -91,6 +91,7 @@ arm_user_ss.add( vfp_d, files( 'debug.c', + 'gengvec.c', 'hflags.c', 'neon_helper.c', 'op_helper.c',
Signature · 1 line
2.47.3
PATCH v10 11/21 target/arm/tcg/translate.c: remove MO_TE usage
In reply to Pierrick Bouvierdc->be_data is already set just above in the same function:
```
dc->be_data = EX_TBFLAG_ANY(tb_flags, BE_DATA) ? MO_BE : MO_LE;
```--- target/arm/tcg/translate.c | 1 - 1 file changed, 1 deletion(-)
target/arm/tcg/translate.c +0 -1
@@ -6344,7 +6344,6 @@ static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)if (arm_feature (env,ARM_FEATURE_M )) { dc->vfp_enabled =1 ; - dc->be_data =MO_TE ; dc->v7m_handler_mode =EX_TBFLAG_M32 (tb_flags,HANDLER ); dc->v8m_secure =EX_TBFLAG_M32 (tb_flags,SECURE ); dc->v8m_stackcheck =EX_TBFLAG_M32 (tb_flags,STACKCHECK );
Signature · 1 line
2.47.3
Re: [PATCH v10 11/21] target/arm/tcg/translate.c: remove MO_TE usage
Quoted 6 lines · Pierrick Bouvier
dc->be_datais already set just above in the same function:``` dc->be_data = EX_TBFLAG_ANY(tb_flags, BE_DATA) ? MO_BE : MO_LE; ```
Why Cc stable? This is just removing pointless code, right?
Quoted 1 line
Fixes a729a46b05a ("target/arm: Add wrapper macros for accessing tbflags")
This commit doesn't seem to be related.
Do you mean commit 79cabf1f473 ("target/arm: Recover 4 bits from TBFLAGs")?
We had completely run out of TBFLAG bits.
Split A- and M-profile bits into two overlapping buckets. This results in 4 free bits.
We used to initialize all of the a32 and m32 fields in DisasContext by assignment, in arm_tr_init_disas_context. Now we only initialize either the a32 or m32 by assignment, because the bits overlap in tbflags. So zero the entire structure in gen_intermediate_code.
Since BE_DATA is in the TBFLAG_ANY set, I don't get why it is set to MO_TE for M-profile.
At any rate this patch description deserves a better explanation IMHO.
Quoted 18 lines
Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/translate.c | 1 - 1 file changed, 1 deletion(-)target/arm/tcg/translate.c +0
-1@@ -6344,7 +6344,6 @@ static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)if (arm_feature (env,ARM_FEATURE_M )) { dc->vfp_enabled =1 ; - dc->be_data =MO_TE ; dc->v7m_handler_mode =EX_TBFLAG_M32 (tb_flags,HANDLER ); dc->v8m_secure =EX_TBFLAG_M32 (tb_flags,SECURE ); dc->v8m_stackcheck =EX_TBFLAG_M32 (tb_flags,STACKCHECK );
Re: [PATCH v10 11/21] target/arm/tcg/translate.c: remove MO_TE usage
Quoted 43 lines · Philippe Mathieu-Daudé
Quoted 6 lines · Pierrick Bouvier
dc->be_datais already set just above in the same function:``` dc->be_data = EX_TBFLAG_ANY(tb_flags, BE_DATA) ? MO_BE : MO_LE; ```Why Cc stable? This is just removing pointless code, right?
Quoted 1 line
Fixes a729a46b05a ("target/arm: Add wrapper macros for accessing tbflags")This commit doesn't seem to be related.
Do you mean commit
79cabf1f473("target/arm: Recover 4 bits from TBFLAGs")?We had completely run out of TBFLAG bits.
Split A- and M-profile bits into two overlapping buckets. This results in 4 free bits.We used to initialize all of the a32 and m32 fields in DisasContext by assignment, in
arm_tr_init_disas_context. Now we only initialize either the a32 or m32 by assignment, because the bits overlap in tbflags. So zero the entire structure ingen_intermediate_code.Since
BE_DATAis in theTBFLAG_ANYset, I don't get why it is set toMO_TEfor M-profile.At any rate this patch description deserves a better explanation IMHO.
Quoted 12 lines
Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/translate.c | 1 - 1 file changed, 1 deletion(-)target/arm/tcg/translate.c +0
-0@@ -6344,7 +6344,6 @@ static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
Quoted 7 lines
target/arm/tcg/translate.c static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs) · line 6349
Quoted 6 lines
target/arm/tcg/translate.c +0
-1if (arm_feature (env,ARM_FEATURE_M )) { dc->vfp_enabled =1 ; - dc->be_data =MO_TE ; dc->v7m_handler_mode =EX_TBFLAG_M32 (tb_flags,HANDLER ); dc->v8m_secure =EX_TBFLAG_M32 (tb_flags,SECURE ); dc->v8m_stackcheck =EX_TBFLAG_M32 (tb_flags,STACKCHECK );
PATCH v10 12/21 target/arm/tcg/translate.c: replace target_ulong with uint32_t
In reply to Pierrick Bouvier--- target/arm/tcg/translate.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
target/arm/tcg/translate.c +2 -2
@@ -6450,7 +6450,7 @@ static void arm_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)* fields here. */ uint32_t condexec_bits; -target_ulong pc_arg = dc->base.pc_next; +uint32_t pc_arg = dc->base.pc_next;if (tb_cflags (dcbase->tb) &CF_PCREL ) { pc_arg &= ~TARGET_PAGE_MASK ;
@@ -6612,7 +6612,7 @@ static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)bool is_16bit;/* TCG op to rewind to if this turns out to be an invalid ECI state */ TCGOp *insn_eci_rewind =NULL ; -target_ulong insn_eci_pc_save =-1 ; +uint32_t insn_eci_pc_save =-1 ;/* Misaligned thumb PC is architecturally impossible. */ assert ((dc->base.pc_next &1 ) ==0 );
Signature · 1 line
2.47.3
PATCH v10 13/21 target/arm/tcg/translate.c: extract aarch64_translate_code()
In reply to Pierrick BouvierThis allows to get rid of TARGET_AARCH64, and helps with next patch which will define at runtime tcg address type, by adding a second entry point in a different source file.
--- target/arm/internals.h | 2 ++ target/arm/tcg/stubs32.c | 7 +++++++ target/arm/tcg/translate-a64.c | 9 +++++++++ target/arm/tcg/translate.c | 19 +++++++++---------- 4 files changed, 27 insertions(+), 10 deletions(-)
target/arm/internals.h +2 -0
@@ -381,6 +381,8 @@ void arm_init_cpreg_list(ARMCPU *cpu);void arm_cpu_register_gdb_regs_for_features (ARMCPU *cpu);void arm_translate_init (void ); +void aarch64_translate_code (CPUState *cs,TranslationBlock *tb, +int *max_insns,vaddr pc,void *host_pc);void arm_translate_code (CPUState *cs,TranslationBlock *tb,int *max_insns,vaddr pc,void *host_pc);
target/arm/tcg/stubs32.c +7 -0
@@ -3,6 +3,7 @@*/ #include "qemu/osdep.h" +#include "target/arm/internals.h" #include "target/arm/tcg/translate.h"
@@ -15,3 +16,9 @@ void a64_translate_init(void) {/* Don't initialize for 32 bits. Call site will be fixed later. */ } + +void aarch64_translate_code (CPUState *cs,TranslationBlock *tb, +int *max_insns,vaddr pc,void *host_pc) +{ +g_assert_not_reached (); +}
target/arm/tcg/translate-a64.c +9 -0
@@ -18,6 +18,7 @@*/ #include "qemu/osdep.h" #include "exec/target_page.h" +#include "exec/translator.h" #include "helper-a64.h" #include "helper-sme.h" #include "helper-sve.h"
@@ -10949,3 +10950,11 @@ const TranslatorOps aarch64_translator_ops = { .translate_insn = aarch64_tr_translate_insn, .tb_stop = aarch64_tr_tb_stop, }; + +void aarch64_translate_code (CPUState *cpu,TranslationBlock *tb, +int *max_insns,vaddr pc,void *host_pc) +{ +DisasContext dc = {}; +translator_loop (cpu, tb, max_insns, pc, host_pc, + &aarch64_translator_ops, &dc.base); +}
target/arm/tcg/translate.c +9 -10
@@ -28,6 +28,7 @@#include "semihosting/semihost.h" #include "cpregs.h" #include "exec/target_page.h" +#include "exec/translator.h" #include "helper.h" #include "helper-mve.h"
@@ -6878,18 +6879,16 @@ static const TranslatorOps thumb_translator_ops = {void arm_translate_code (CPUState *cpu,TranslationBlock *tb,int *max_insns,vaddr pc,void *host_pc) { -DisasContext dc = { }; -const TranslatorOps *ops = &arm_translator_ops;CPUARMTBFlags tb_flags =arm_tbflags_from_tb (tb); -if (EX_TBFLAG_AM32 (tb_flags,THUMB )) { - ops = &thumb_translator_ops; - } -#ifdef TARGET_AARCH64 if (EX_TBFLAG_ANY (tb_flags,AARCH64_STATE )) { - ops = &aarch64_translator_ops; +aarch64_translate_code (cpu, tb, max_insns, pc, host_pc); + }else { +DisasContext dc = { }; +translator_loop (cpu, tb, max_insns, pc, host_pc, + (EX_TBFLAG_AM32 (tb_flags,THUMB ) + ? &thumb_translator_ops + : &arm_translator_ops), + &dc.base); } -#endif - -translator_loop (cpu, tb, max_insns, pc, host_pc, ops, &dc.base); }
Signature · 1 line
2.47.3
Re: [PATCH v10 13/21] target/arm/tcg/translate.c: extract aarch64_translate_code()
Quoted 7 lines · Pierrick Bouvier
This allows to get rid of
TARGET_AARCH64, and helps with next patch which will define at runtime tcg address type, by adding a second entry point in a different source file.Suggested-by Richard Henderson <[email protected]>Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>
Quoted 6 lines
--- target/arm/internals.h | 2 ++ target/arm/tcg/stubs32.c | 7 +++++++ target/arm/tcg/translate-a64.c | 9 +++++++++ target/arm/tcg/translate.c | 19 +++++++++---------- 4 files changed, 27 insertions(+), 10 deletions(-)
PATCH v10 14/21 tcg/translator: add parameter to translator_loop for current addr type
In reply to Pierrick BouvierWith TCG_ADDRESS_BITS mechanism, it's now possible to specify which variant every source file is written for. Compared to before, it means that addr_type will now vary per tb translation, where it was constant for a given target previously.
Thus, we add new a parameter to translator_loop(). This will allow us to convert targets one by one.
--- include/exec/translator.h | 4 +++- accel/tcg/translate-all.c | 1 - accel/tcg/translator.c | 4 +++- target/alpha/translate.c | 3 ++- target/arm/tcg/translate-a64.c | 3 ++- target/arm/tcg/translate.c | 2 +- target/avr/translate.c | 3 ++- target/hexagon/translate.c | 3 ++- target/hppa/translate.c | 3 ++- target/i386/tcg/translate.c | 3 ++- target/loongarch/tcg/translate.c | 3 ++- target/m68k/translate.c | 3 ++- target/microblaze/translate.c | 3 ++- target/mips/tcg/translate.c | 3 ++- target/or1k/translate.c | 3 ++- target/ppc/translate.c | 3 ++- target/riscv/translate.c | 3 ++- target/rx/translate.c | 3 ++- target/s390x/tcg/translate.c | 3 ++- target/sh4/translate.c | 3 ++- target/sparc/translate.c | 3 ++- target/tricore/translate.c | 3 ++- target/xtensa/translate.c | 3 ++- 23 files changed, 45 insertions(+), 23 deletions(-)
include/exec/translator.h +3 -1
@@ -20,6 +20,7 @@#include "exec/memop.h" #include "exec/vaddr.h" +#include "tcg/tcg.h" /** * DisasJumpType:
@@ -132,6 +133,7 @@ typedef struct TranslatorOps {* @host_pc: host physical program counter address * @ops: Target-specific operations. * @db: Disassembly context. +* @addr_type: TCG Type for addresses (TCGv_va). * * Generic translator loop. *
@@ -147,7 +149,7 @@ typedef struct TranslatorOps {*/ void translator_loop (CPUState *cpu,TranslationBlock *tb,int *max_insns,vaddr pc,void *host_pc,const TranslatorOps *ops, -DisasContextBase *db); +DisasContextBase *db,TCGType addr_type);/** * translator_use_goto_tb
accel/tcg/translate-all.c +0 -1
@@ -316,7 +316,6 @@ TranslationBlock *tb_gen_code(CPUState *cpu, TCGTBCPUState s) } tcg_ctx->gen_tb = tb; - tcg_ctx->addr_type =target_long_bits () ==32 ?TCG_TYPE_I32 :TCG_TYPE_I64 ; tcg_ctx->guest_mo = cpu->cc->tcg_ops->guest_default_memory_order; restart_translate:
accel/tcg/translator.c +3 -1
@@ -121,13 +121,15 @@ bool translator_use_goto_tb(DisasContextBase *db, vaddr dest)void translator_loop (CPUState *cpu,TranslationBlock *tb,int *max_insns,vaddr pc,void *host_pc,const TranslatorOps *ops, -DisasContextBase *db) +DisasContextBase *db,TCGType addr_type) {uint32_t cflags =tb_cflags (tb);TCGOp *icount_start_insn;TCGOp *first_insn_start =NULL ;bool plugin_enabled; + tcg_ctx->addr_type = addr_type; +/* Initialize DisasContext */ db->tb = tb; db->pc_first = pc;
target/alpha/translate.c +2 -1
@@ -2953,5 +2953,6 @@ void alpha_translate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; -translator_loop (cpu, tb, max_insns, pc, host_pc, &alpha_tr_ops, &dc.base); +translator_loop (cpu, tb, max_insns, pc, host_pc, &alpha_tr_ops, &dc.base, +TCG_TYPE_VA ); }
target/arm/tcg/translate-a64.c +2 -1
@@ -10956,5 +10956,6 @@ void aarch64_translate_code(CPUState *cpu, TranslationBlock *tb, {DisasContext dc = {};translator_loop (cpu, tb, max_insns, pc, host_pc, - &aarch64_translator_ops, &dc.base); + &aarch64_translator_ops, &dc.base, +TCG_TYPE_VA ); }
target/arm/tcg/translate.c +1 -1
@@ -6889,6 +6889,6 @@ void arm_translate_code(CPUState *cpu, TranslationBlock *tb, (EX_TBFLAG_AM32 (tb_flags,THUMB ) ? &thumb_translator_ops : &arm_translator_ops), - &dc.base); + &dc.base, TCG_TYPE_VA); } }
target/avr/translate.c +2 -1
@@ -2802,5 +2802,6 @@ void avr_cpu_translate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, vaddr pc, void *host_pc) { DisasContext dc = { }; -translator_loop (cs, tb, max_insns, pc, host_pc, &avr_tr_ops, &dc.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &avr_tr_ops, &dc.base, +TCG_TYPE_VA ); }
target/hexagon/translate.c +2 -1
@@ -1077,7 +1077,8 @@ void hexagon_translate_code(CPUState *cs, TranslationBlock *tb,DisasContext ctx;translator_loop (cs, tb, max_insns, pc, host_pc, - &hexagon_tr_ops, &ctx.base); + &hexagon_tr_ops, &ctx.base, +TCG_TYPE_VA ); }#define NAME_LEN 64
target/hppa/translate.c +2 -1
@@ -4899,5 +4899,6 @@ void hppa_translate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx = { }; -translator_loop (cs, tb, max_insns, pc, host_pc, &hppa_tr_ops, &ctx.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &hppa_tr_ops, &ctx.base, +TCG_TYPE_VA ); }
target/i386/tcg/translate.c +2 -1
@@ -3615,5 +3615,6 @@ void x86_translate_code(CPUState *cpu, TranslationBlock *tb, {DisasContext dc; -translator_loop (cpu, tb, max_insns, pc, host_pc, &i386_tr_ops, &dc.base); +translator_loop (cpu, tb, max_insns, pc, host_pc, &i386_tr_ops, &dc.base, +TCG_TYPE_VA ); }
target/loongarch/tcg/translate.c +2 -1
@@ -342,7 +342,8 @@ void loongarch_translate_code(CPUState *cs, TranslationBlock *tb,DisasContext ctx;translator_loop (cs, tb, max_insns, pc, host_pc, - &loongarch_tr_ops, &ctx.base); + &loongarch_tr_ops, &ctx.base, +TCG_TYPE_VA ); }void loongarch_translate_init (void )
target/m68k/translate.c +2 -1
@@ -6126,7 +6126,8 @@ void m68k_translate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; -translator_loop (cpu, tb, max_insns, pc, host_pc, &m68k_tr_ops, &dc.base); +translator_loop (cpu, tb, max_insns, pc, host_pc, &m68k_tr_ops, &dc.base, +TCG_TYPE_VA ); }static double floatx80_to_double (CPUM68KState *env,uint16_t high,uint64_t low)
target/microblaze/translate.c +2 -1
@@ -1788,7 +1788,8 @@ void mb_translate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; -translator_loop (cpu, tb, max_insns, pc, host_pc, &mb_tr_ops, &dc.base); +translator_loop (cpu, tb, max_insns, pc, host_pc, &mb_tr_ops, &dc.base, +TCG_TYPE_VA ); }void mb_cpu_dump_state (CPUState *cs,FILE *f,int flags)
target/mips/tcg/translate.c +2 -1
@@ -15242,7 +15242,8 @@ void mips_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext ctx; -translator_loop (cs, tb, max_insns, pc, host_pc, &mips_tr_ops, &ctx.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &mips_tr_ops, &ctx.base, +TCG_TYPE_VA ); }void mips_tcg_init (void )
target/or1k/translate.c +2 -1
@@ -1647,7 +1647,8 @@ void openrisc_translate_code(CPUState *cs, TranslationBlock *tb,DisasContext ctx;translator_loop (cs, tb, max_insns, pc, host_pc, - &openrisc_tr_ops, &ctx.base); + &openrisc_tr_ops, &ctx.base, +TCG_TYPE_VA ); }void openrisc_cpu_dump_state (CPUState *cs,FILE *f,int flags)
target/ppc/translate.c +2 -1
@@ -6719,5 +6719,6 @@ void ppc_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext ctx; -translator_loop (cs, tb, max_insns, pc, host_pc, &ppc_tr_ops, &ctx.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &ppc_tr_ops, &ctx.base, +TCG_TYPE_VA ); }
target/riscv/translate.c +2 -1
@@ -1440,7 +1440,8 @@ void riscv_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext ctx; -translator_loop (cs, tb, max_insns, pc, host_pc, &riscv_tr_ops, &ctx.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &riscv_tr_ops, &ctx.base, +TCG_TYPE_VA ); }void riscv_translate_init (void )
target/rx/translate.c +2 -1
@@ -2270,7 +2270,8 @@ void rx_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext dc; -translator_loop (cs, tb, max_insns, pc, host_pc, &rx_tr_ops, &dc.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &rx_tr_ops, &dc.base, +TCG_TYPE_VA ); }#define ALLOC_REGISTER (sym, name) \
target/s390x/tcg/translate.c +2 -1
@@ -6509,7 +6509,8 @@ void s390x_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext dc; -translator_loop (cs, tb, max_insns, pc, host_pc, &s390x_tr_ops, &dc.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &s390x_tr_ops, &dc.base, +TCG_TYPE_VA ); } void s390x_restore_state_to_opc(CPUState *cs,
target/sh4/translate.c +2 -1
@@ -2316,5 +2316,6 @@ void sh4_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext ctx; -translator_loop (cs, tb, max_insns, pc, host_pc, &sh4_tr_ops, &ctx.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &sh4_tr_ops, &ctx.base, +TCG_TYPE_VA ); }
target/sparc/translate.c +2 -1
@@ -5853,7 +5853,8 @@ void sparc_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext dc = {}; -translator_loop (cs, tb, max_insns, pc, host_pc, &sparc_tr_ops, &dc.base); +translator_loop (cs, tb, max_insns, pc, host_pc, &sparc_tr_ops, &dc.base, +TCG_TYPE_VA ); }void sparc_tcg_init (void )
target/tricore/translate.c +2 -1
@@ -8500,7 +8500,8 @@ void tricore_translate_code(CPUState *cs, TranslationBlock *tb, {DisasContext ctx;translator_loop (cs, tb, max_insns, pc, host_pc, - &tricore_tr_ops, &ctx.base); + &tricore_tr_ops, &ctx.base, +TCG_TYPE_VA ); }/*
target/xtensa/translate.c +2 -1
@@ -1233,7 +1233,8 @@ void xtensa_translate_code(CPUState *cpu, TranslationBlock *tb, {DisasContext dc = {};translator_loop (cpu, tb, max_insns, pc, host_pc, - &xtensa_translator_ops, &dc.base); + &xtensa_translator_ops, &dc.base, +TCG_TYPE_VA ); }void xtensa_cpu_dump_state (CPUState *cs,FILE *f,int flags)
Signature · 1 line
2.47.3
Re: [PATCH v10 14/21] tcg/translator: add parameter to translator_loop for current addr type
Quoted 49 lines · Pierrick Bouvier
With
TCG_ADDRESS_BITSmechanism, it's now possible to specify which variant every source file is written for. Compared to before, it means thataddr_typewill now vary per tb translation, where it was constant for a given target previously.Thus, we add new a parameter to
translator_loop(). This will allow us to convert targets one by one.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- include/exec/translator.h | 4 +++- accel/tcg/translate-all.c | 1 - accel/tcg/translator.c | 4 +++- target/alpha/translate.c | 3 ++- target/arm/tcg/translate-a64.c | 3 ++- target/arm/tcg/translate.c | 2 +- target/avr/translate.c | 3 ++- target/hexagon/translate.c | 3 ++- target/hppa/translate.c | 3 ++- target/i386/tcg/translate.c | 3 ++- target/loongarch/tcg/translate.c | 3 ++- target/m68k/translate.c | 3 ++- target/microblaze/translate.c | 3 ++- target/mips/tcg/translate.c | 3 ++- target/or1k/translate.c | 3 ++- target/ppc/translate.c | 3 ++- target/riscv/translate.c | 3 ++- target/rx/translate.c | 3 ++- target/s390x/tcg/translate.c | 3 ++- target/sh4/translate.c | 3 ++- target/sparc/translate.c | 3 ++- target/tricore/translate.c | 3 ++- target/xtensa/translate.c | 3 ++- 23 files changed, 45 insertions(+), 23 deletions(-)include/exec/translator.h +1
-0@@ -20,6 +20,7 @@#include "exec/memop.h" #include "exec/vaddr.h" +#include "tcg/tcg.h" /** * DisasJumpType:
include/exec/translator.h typedef struct TranslatorOps { · line 136
Quoted 4 lines
include/exec/translator.h +1
-0@@ -132,6 +133,7 @@ typedef struct TranslatorOps {* @host_pc: host physical program counter address * @ops: Target-specific operations. * @db: Disassembly context. +* @addr_type: TCG Type for addresses (TCGv_va).
"TCGv_va" confuses me here, do you mean TCG_TYPE_VA?
What about:
"@addr_type: #TCG_TYPE_VA for TCGv_va address expansion."
Otherwise,
Re: [PATCH v10 14/21] tcg/translator: add parameter to translator_loop for current addr type
Quoted 57 lines · Philippe Mathieu-Daudé
Quoted 49 lines · Pierrick Bouvier
With
TCG_ADDRESS_BITSmechanism, it's now possible to specify which variant every source file is written for. Compared to before, it means thataddr_typewill now vary per tb translation, where it was constant for a given target previously.Thus, we add new a parameter to
translator_loop(). This will allow us to convert targets one by one.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- include/exec/translator.h | 4 +++- accel/tcg/translate-all.c | 1 - accel/tcg/translator.c | 4 +++- target/alpha/translate.c | 3 ++- target/arm/tcg/translate-a64.c | 3 ++- target/arm/tcg/translate.c | 2 +- target/avr/translate.c | 3 ++- target/hexagon/translate.c | 3 ++- target/hppa/translate.c | 3 ++- target/i386/tcg/translate.c | 3 ++- target/loongarch/tcg/translate.c | 3 ++- target/m68k/translate.c | 3 ++- target/microblaze/translate.c | 3 ++- target/mips/tcg/translate.c | 3 ++- target/or1k/translate.c | 3 ++- target/ppc/translate.c | 3 ++- target/riscv/translate.c | 3 ++- target/rx/translate.c | 3 ++- target/s390x/tcg/translate.c | 3 ++- target/sh4/translate.c | 3 ++- target/sparc/translate.c | 3 ++- target/tricore/translate.c | 3 ++- target/xtensa/translate.c | 3 ++- 23 files changed, 45 insertions(+), 23 deletions(-)include/exec/translator.h +1
-0@@ -20,6 +20,7 @@#include "exec/memop.h" #include "exec/vaddr.h" +#include "tcg/tcg.h" /** * DisasJumpType: include/exec/translator.h typedef struct TranslatorOps { · line 136
Quoted 4 lines
include/exec/translator.h +1
-0@@ -132,6 +133,7 @@ typedef struct TranslatorOps {* @host_pc: host physical program counter address * @ops: Target-specific operations. * @db: Disassembly context. +* @addr_type: TCG Type for addresses (TCGv_va). "
TCGv_va" confuses me here, do you meanTCG_TYPE_VA?
Quoted 4 lines
What about:
"@addr_type: #TCG_TYPE_VA for
TCGv_vaaddress expansion."
Right, I'll update comment.
Quoted 4 lines
Otherwise,
Reviewed-by Philippe Mathieu-Daudé <[email protected]>
PATCH v10 15/21 target/arm/tcg/translate.c: replace TCGv with TCGv_va
In reply to Pierrick BouvierWe know this file is for 32-bit runtime target, so we can set TCG_ADDRESS_BITS. TCG_TYPE_VA is derived accordingly and is already passed to translator_loop.
--- target/arm/tcg/translate.c | 29 +++++++++++++++-------------- 1 file changed, 15 insertions(+), 14 deletions(-)
target/arm/tcg/translate.c +15 -14
@@ -22,7 +22,8 @@#include "translate.h" #include "translate-a32.h" -#include "tcg/tcg-op.h" +#define TCG_ADDRESS_BITS 32 +#include "tcg/tcg-op-mem.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h"
@@ -910,14 +911,14 @@ MemOp pow2_align(unsigned i)* that the address argument is TCGv_i32 rather than TCGv. */ -static TCGv gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) +static TCGv_va gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) { -TCGv addr =tcg_temp_new (); -tcg_gen_extu_i32_tl (addr, a32); +TCGv_va addr =tcgv_va_temp_new (); +tcg_gen_mov_i32 (addr, a32);/* Not needed for user-mode BE32, where we use MO_BE instead. */ if (!IS_USER_ONLY && s->sctlr_b && (op &MO_SIZE ) <MO_32 ) { -tcg_gen_xori_tl (addr, addr,4 - (1 << (op &MO_SIZE ))); +tcg_gen_xori_i32 (addr, addr,4 - (1 << (op &MO_SIZE ))); }return addr; }
@@ -929,21 +930,21 @@ static TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)void gen_aa32_ld_internal_i32 (DisasContext *s,TCGv_i32 val,TCGv_i32 a32,int index,MemOp opc) { -TCGv addr =gen_aa32_addr (s, a32, opc); +TCGv_va addr =gen_aa32_addr (s, a32, opc);tcg_gen_qemu_ld_i32 (val, addr, index, opc); }void gen_aa32_st_internal_i32 (DisasContext *s,TCGv_i32 val,TCGv_i32 a32,int index,MemOp opc) { -TCGv addr =gen_aa32_addr (s, a32, opc); +TCGv_va addr =gen_aa32_addr (s, a32, opc);tcg_gen_qemu_st_i32 (val, addr, index, opc); }void gen_aa32_ld_internal_i64 (DisasContext *s,TCGv_i64 val,TCGv_i32 a32,int index,MemOp opc) { -TCGv addr =gen_aa32_addr (s, a32, opc); +TCGv_va addr =gen_aa32_addr (s, a32, opc);tcg_gen_qemu_ld_i64 (val, addr, index, opc);
@@ -956,7 +957,7 @@ void gen_aa32_ld_internal_i64(DisasContext *s, TCGv_i64 val,void gen_aa32_st_internal_i64 (DisasContext *s,TCGv_i64 val,TCGv_i32 a32,int index,MemOp opc) { -TCGv addr =gen_aa32_addr (s, a32, opc); +TCGv_va addr =gen_aa32_addr (s, a32, opc);/* Not needed for user-mode BE32, where we use MO_BE instead. */ if (!IS_USER_ONLY && s->sctlr_b && (opc &MO_SIZE ) ==MO_64 ) {
@@ -2036,7 +2037,7 @@ static void gen_load_exclusive(DisasContext *s, int rt, int rt2,* architecturally 64-bit access, but instead do a 64-bit access * using MO_BE if appropriate and then split the two halves. */ -TCGv taddr =gen_aa32_addr (s, addr, opc); +TCGv_va taddr =gen_aa32_addr (s, addr, opc);tcg_gen_qemu_ld_i64 (t64, taddr,get_mem_index (s), opc);tcg_gen_mov_i64 (cpu_exclusive_val, t64);
@@ -2065,7 +2066,7 @@ static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2, {TCGv_i32 t0, t1, t2;TCGv_i64 extaddr; -TCGv taddr; +TCGv_va taddr;TCGLabel *done_label;TCGLabel *fail_label;MemOp opc = size |MO_ALIGN | s->be_data;
@@ -3792,7 +3793,7 @@ static void do_ldrd_load(DisasContext *s, TCGv_i32 addr, int rt, int rt2)*/ int mem_idx =get_mem_index (s);MemOp opc =MO_64 |MO_ALIGN_4 |MO_ATOM_SUBALIGN | s->be_data; -TCGv taddr =gen_aa32_addr (s, addr, opc); +TCGv_va taddr =gen_aa32_addr (s, addr, opc);TCGv_i64 t64 =tcg_temp_new_i64 ();TCGv_i32 tmp =tcg_temp_new_i32 ();TCGv_i32 tmp2 =tcg_temp_new_i32 ();
@@ -3847,7 +3848,7 @@ static void do_strd_store(DisasContext *s, TCGv_i32 addr, int rt, int rt2)*/ int mem_idx =get_mem_index (s);MemOp opc =MO_64 |MO_ALIGN_4 |MO_ATOM_SUBALIGN | s->be_data; -TCGv taddr =gen_aa32_addr (s, addr, opc); +TCGv_va taddr =gen_aa32_addr (s, addr, opc);TCGv_i32 t1 =load_reg (s, rt);TCGv_i32 t2 =load_reg (s, rt2);TCGv_i64 t64 =tcg_temp_new_i64 ();
@@ -4068,7 +4069,7 @@ DO_LDST(STRH, store, MO_UW)static bool op_swp (DisasContext *s,arg_SWP *a,MemOp opc) {TCGv_i32 addr, tmp; -TCGv taddr; +TCGv_va taddr; opc |= s->be_data; addr =load_reg (s, a->rn);
Signature · 1 line
2.47.3
Re: [PATCH v10 15/21] target/arm/tcg/translate.c: replace TCGv with TCGv_va
Quoted 25 lines · Pierrick Bouvier
We know this file is for 32-bit runtime target, so we can set
TCG_ADDRESS_BITS.TCG_TYPE_VAis derived accordingly and is already passed totranslator_loop.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/translate.c | 29 +++++++++++++++-------------- 1 file changed, 15 insertions(+), 14 deletions(-)target/arm/tcg/translate.c +2
-1@@ -22,7 +22,8 @@#include "translate.h" #include "translate-a32.h" -#include "tcg/tcg-op.h" +#define TCG_ADDRESS_BITS 32 +#include "tcg/tcg-op-mem.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h"
target/arm/tcg/translate.c MemOp pow2_align(unsigned i) · line 917
Quoted 10 lines
target/arm/tcg/translate.c +3
-3@@ -910,14 +911,14 @@ MemOp pow2_align(unsigned i)* that the address argument is TCGv_i32 rather than TCGv. */ -static TCGv gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) +static TCGv_va gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) { -TCGv addr =tcg_temp_new (); -tcg_gen_extu_i32_tl (addr, a32); +TCGv_va addr =tcgv_va_temp_new (); +tcg_gen_mov_i32 (addr, a32);
The implicit TCGv_va -> TCGv_i32 bugs me here, but the overall change will make my sleep better :) So:
Quoted 16 lines
target/arm/tcg/translate.c +2
-2/* Not needed for user-mode BE32, where we use MO_BE instead. */ if (!IS_USER_ONLY && s->sctlr_b && (op &MO_SIZE ) <MO_32 ) { -tcg_gen_xori_tl (addr, addr,4 - (1 << (op &MO_SIZE ))); +tcg_gen_xori_i32 (addr, addr,4 - (1 << (op &MO_SIZE ))); }return addr; }@@ -929,21 +930,21 @@ static TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)void gen_aa32_ld_internal_i32 (DisasContext *s,TCGv_i32 val,TCGv_i32 a32,int index,MemOp opc) { -TCGv addr =gen_aa32_addr (s, a32, opc); +TCGv_va addr =gen_aa32_addr (s, a32, opc);tcg_gen_qemu_ld_i32 (val, addr, index, opc); }
Re: [PATCH v10 15/21] target/arm/tcg/translate.c: replace TCGv with TCGv_va
Quoted 29 lines · Philippe Mathieu-Daudé
Quoted 28 lines · Pierrick Bouvier
We know this file is for 32-bit runtime target, so we can set
TCG_ADDRESS_BITS.TCG_TYPE_VAis derived accordingly and is already passed totranslator_loop.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/translate.c | 29 +++++++++++++++-------------- 1 file changed, 15 insertions(+), 14 deletions(-)target/arm/tcg/translate.c +2
-1@@ -22,7 +22,8 @@#include "translate.h" #include "translate-a32.h" -#include "tcg/tcg-op.h" +#define TCG_ADDRESS_BITS 32 +#include "tcg/tcg-op-mem.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h" @@ -910,14 +911,14 @@ MemOp pow2_align(unsigned i)* that the address argument is TCGv_i32 rather than TCGv. */
Quoted 11 lines
target/arm/tcg/translate.c MemOp pow2_align(unsigned i) · line 917
Quoted 7 lines
target/arm/tcg/translate.c +3
-3-static TCGv gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) +static TCGv_va gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) { -TCGv addr =tcg_temp_new (); -tcg_gen_extu_i32_tl (addr, a32); +TCGv_va addr =tcgv_va_temp_new (); +tcg_gen_mov_i32 (addr, a32);The implicit
TCGv_va->TCGv_i32bugs me here, but the overall change will make my sleep better :) So:
We could use TCGv_i32 directly, TCGv_va is just useful here to document the intention. I can update this if Richard and you prefer this.
Quoted 19 lines
Reviewed-by Philippe Mathieu-Daudé <[email protected]>Quoted 16 lines
target/arm/tcg/translate.c +2
-2/* Not needed for user-mode BE32, where we use MO_BE instead. */ if (!IS_USER_ONLY && s->sctlr_b && (op &MO_SIZE ) <MO_32 ) { -tcg_gen_xori_tl (addr, addr,4 - (1 << (op &MO_SIZE ))); +tcg_gen_xori_i32 (addr, addr,4 - (1 << (op &MO_SIZE ))); }return addr; }@@ -929,21 +930,21 @@ static TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)void gen_aa32_ld_internal_i32 (DisasContext *s,TCGv_i32 val,TCGv_i32 a32,int index,MemOp opc) { -TCGv addr =gen_aa32_addr (s, a32, opc); +TCGv_va addr =gen_aa32_addr (s, a32, opc);tcg_gen_qemu_ld_i32 (val, addr, index, opc); }
Re: [PATCH v10 15/21] target/arm/tcg/translate.c: replace TCGv with TCGv_va
Quoted 40 lines · Pierrick Bouvier
Quoted 29 lines · Philippe Mathieu-Daudé
Quoted 28 lines · Pierrick Bouvier
We know this file is for 32-bit runtime target, so we can set
TCG_ADDRESS_BITS.TCG_TYPE_VAis derived accordingly and is already passed totranslator_loop.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>---
target/arm/tcg/translate.c| 29 +++++++++++++++--------------1 file changed, 15 insertions(+), 14 deletions(-)target/arm/tcg/translate.c +3
-2@@ -22,7 +22,8 @@ #include "translate.h" #include "translate-a32.h" -#include "tcg/tcg-op.h" +#define TCG_ADDRESS_BITS 32 +#include "tcg/tcg-op-mem.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h"@@ -910,14 +911,14 @@ MemOp pow2_align(unsigned i)* that the address argument is TCGv_i32 rather than TCGv. */ -static TCGv gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) +static TCGv_va gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op)Quoted 9 lines
target/arm/tcg/translate.c MemOp pow2_align(unsigned i) · line 917
Quoted 5 lines
target/arm/tcg/translate.c +2
-2{ - TCGv addr = tcg_temp_new(); - tcg_gen_extu_i32_tl(addr, a32); + TCGv_va addr = tcgv_va_temp_new(); + tcg_gen_mov_i32(addr, a32);The implicit
TCGv_va->TCGv_i32bugs me here, but the overall change will make my sleep better :) So:
Quoted 1 line
We could use
TCGv_i32directly,TCGv_vais just useful here to document the intention. I can update this if Richard and you prefer this.
I think this is fine, as I'm not keen on introducing a complete set of tcg_gen_foo_va macros. As is, any mismatch will be caught at compile-time.
r~
Re: [PATCH v10 15/21] target/arm/tcg/translate.c: replace TCGv with TCGv_va
Quoted 41 lines · Richard Henderson
Quoted 40 lines · Pierrick Bouvier
Quoted 38 lines · Philippe Mathieu-Daudé
Quoted 24 lines · Pierrick Bouvier
We know this file is for 32-bit runtime target, so we can set
TCG_ADDRESS_BITS.TCG_TYPE_VAis derived accordingly and is already passed totranslator_loop.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/translate.c | 29 +++++++++++++++-------------- 1 file changed, 15 insertions(+), 14 deletions(-)target/arm/tcg/translate.c +2
-1@@ -22,7 +22,8 @@ #include "translate.h" #include "translate-a32.h" -#include "tcg/tcg-op.h" +#define TCG_ADDRESS_BITS 32 +#include "tcg/tcg-op-mem.h" #include "qemu/log.h" #include "arm_ldst.h" #include "semihosting/semihost.h"target/arm/tcg/translate.c MemOp pow2_align(unsigned i) · line 916
Quoted 9 lines
target/arm/tcg/translate.c +3
-3@@ -910,14 +911,14 @@ MemOp pow2_align(unsigned i)* that the address argument is TCGv_i32 rather than TCGv. */ -static TCGv gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) +static TCGv_va gen_aa32_addr (DisasContext *s,TCGv_i32 a32,MemOp op) { - TCGv addr = tcg_temp_new(); - tcg_gen_extu_i32_tl(addr, a32); + TCGv_va addr = tcgv_va_temp_new(); + tcg_gen_mov_i32(addr, a32);The implicit
TCGv_va->TCGv_i32bugs me here, but the overall change will make my sleep better :) So:
Quoted 6 lines
Quoted 2 lines
We could use
TCGv_i32directly,TCGv_vais just useful here to document the intention. I can update this if Richard and you prefer this.I think this is fine, as I'm not keen on introducing a complete set of
tcg_gen_foo_vamacros. As is, any mismatch will be caught at compile-time.
Agreed. Usage should be reduced to address related functions.
Quoted 2 lines
r~
PATCH v10 16/21 target/arm/tcg/translate-a64.c: use translator_ldl_end instead of arm_ldl_code
In reply to Pierrick BouvierAllows to reduce scope of target/arm/tcg/arm_ldst.h to aarch32 only.
--- target/arm/tcg/translate-a64.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
target/arm/tcg/translate-a64.c +3 -3
@@ -26,7 +26,6 @@#include "translate-a64.h" #include "tcg/tcg-op.h" #include "qemu/log.h" -#include "arm_ldst.h" #include "semihosting/semihost.h" #include "cpregs.h"
@@ -10801,7 +10800,7 @@ static void aarch64_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)if (pc &3 ) {/* * PC alignment fault. This has priority over the instruction abort -* that we would receive from a translation fault via arm_ldl_code. +* that we would receive from a translation fault via translator_ldl_end. * This should only be possible after an indirect branch, at the * start of the TB. */
@@ -10813,7 +10812,8 @@ static void aarch64_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu) } s->pc_curr = pc; - insn =arm_ldl_code (env, &s->base, pc, s->sctlr_b); +/* Code is always little-endian on Aarch64 */ + insn =translator_ldl_end (env, &s->base, pc,MO_LE ); s->insn = insn; s->base.pc_next = pc +4 ;
Signature · 1 line
2.47.3
PATCH v10 17/21 target/arm/tcg/arm_ldst.h: merge in translate.c
In reply to Pierrick BouvierOnly translate.c uses those functions, thus move them back to this file.
--- target/arm/tcg/arm_ldst.h | 47 -------------------------------------- target/arm/tcg/translate.c | 22 +++++++++++++++++- 2 files changed, 21 insertions(+), 48 deletions(-)
target/arm/tcg/arm_ldst.h +0 -47
@@ -1,47 +0,0 @@ -/* -* ARM load/store instructions for code (armeb-user support) -* -* Copyright (c) 2012 CodeSourcery, LLC -* -* This library is free software; you can redistribute it and/or -* modify it under the terms of the GNU Lesser General Public -* License as published by the Free Software Foundation; either -* version 2.1 of the License, or (at your option) any later version. -* -* This library is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU -* Lesser General Public License for more details. -* -* You should have received a copy of the GNU Lesser General Public -* License along with this library; if not, see <http://www.gnu.org/licenses/>. -*/ - -#ifndef ARM_LDST_H -#define ARM_LDST_H - -#include "exec/translator.h" -#include "qemu/bswap.h" - -/* Load an instruction and return it in the standard little-endian order */ -static inline uint32_t arm_ldl_code (CPUARMState *env,DisasContextBase *s, -target_ulong addr,bool sctlr_b) -{ -return translator_ldl_swap (env, s, addr,bswap_code (sctlr_b)); -} - -/* Ditto, for a halfword (Thumb) instruction */ -static inline uint16_t arm_lduw_code (CPUARMState *env,DisasContextBase * s, -target_ulong addr,bool sctlr_b) -{ -#ifndef CONFIG_USER_ONLY -/* In big-endian (BE32) mode, adjacent Thumb instructions have been swapped -within each word. Undo that now. */ -if (sctlr_b) { - addr ^=2 ; - } -#endif -return translator_lduw_swap (env, s, addr,bswap_code (sctlr_b)); -} - -#endif
target/arm/tcg/translate.c +21 -1
@@ -25,7 +25,6 @@#define TCG_ADDRESS_BITS 32#include "tcg/tcg-op-mem.h" #include "qemu/log.h" -#include "arm_ldst.h" #include "semihosting/semihost.h" #include "cpregs.h" #include "exec/target_page.h"
@@ -6281,6 +6280,20 @@ static void disas_thumb_insn(DisasContext *s, uint32_t insn) } } +/* Ditto, for a halfword (Thumb) instruction */ +static uint16_t arm_lduw_code (CPUARMState *env,DisasContextBase * s, +target_ulong addr,bool sctlr_b) +{ +#ifndef CONFIG_USER_ONLY +/* In big-endian (BE32) mode, adjacent Thumb instructions have been swapped +within each word. Undo that now. */ +if (sctlr_b) { + addr ^=2 ; + } +#endif +return translator_lduw_swap (env, s, addr,bswap_code (sctlr_b)); +} +static bool insn_crosses_page (CPUARMState *env,DisasContext *s) {/* Return true if the insn at dc->base.pc_next might cross a page boundary.
@@ -6515,6 +6528,13 @@ static void arm_post_translate_insn(DisasContext *dc) } } +/* Load an instruction and return it in the standard little-endian order */ +static uint32_t arm_ldl_code (CPUARMState *env,DisasContextBase *s, +target_ulong addr,bool sctlr_b) +{ +return translator_ldl_swap (env, s, addr,bswap_code (sctlr_b)); +} +static void arm_tr_translate_insn (DisasContextBase *dcbase,CPUState *cpu) {DisasContext *dc =container_of (dcbase, DisasContext, base);
Signature · 1 line
2.47.3
Re: [PATCH v10 17/21] target/arm/tcg/arm_ldst.h: merge in translate.c
Quoted 10 lines · Pierrick Bouvier
Only
translate.cuses those functions, thus move them back to this file.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/arm_ldst.h | 47 -------------------------------------- target/arm/tcg/translate.c | 22 +++++++++++++++++- 2 files changed, 21 insertions(+), 48 deletions(-)
target/arm/tcg/translate.c static void disas_thumb_insn(DisasContext *s, uint32_t insn) · line 6289
Quoted 7 lines
target/arm/tcg/translate.c +7
-0+/* Ditto, for a halfword (Thumb) instruction */ +static uint16_t arm_lduw_code (CPUARMState *env,DisasContextBase * s, +target_ulong addr,bool sctlr_b) +{ +#ifndef CONFIG_USER_ONLY +/* In big-endian (BE32) mode, adjacent Thumb instructions have been swapped +within each word. Undo that now. */
Pre-existing and removed in the next patch, but in case you have to respin:
WARNING: Block comments use * on subsequent lines
#91: FILE: target/arm/tcg/translate.c:6289:
+ /* In big-endian (BE32) mode, adjacent Thumb instructions have been swapped
+ within each word. Undo that now. */
Quoted 6 lines
target/arm/tcg/translate.c +6
-0+if (sctlr_b) { + addr ^=2 ; + } +#endif +return translator_lduw_swap (env, s, addr,bswap_code (sctlr_b)); +}
Re: [PATCH v10 17/21] target/arm/tcg/arm_ldst.h: merge in translate.c
In reply to Pierrick BouvierQuoted 8 lines · Pierrick Bouvier
Only
translate.cuses those functions, thus move them back to this file.Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/arm_ldst.h | 47 -------------------------------------- target/arm/tcg/translate.c | 22 +++++++++++++++++- 2 files changed, 21 insertions(+), 48 deletions(-)
PATCH v10 18/21 target/arm/tcg/translate.c: replace translator_ldl_swap with translator_ldl_end
In reply to Pierrick Bouvier--- target/arm/tcg/translate.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-)
target/arm/tcg/translate.c +5 -6
@@ -6284,14 +6284,13 @@ static void disas_thumb_insn(DisasContext *s, uint32_t insn)static uint16_t arm_lduw_code (CPUARMState *env,DisasContextBase * s,target_ulong addr,bool sctlr_b) { -#ifndef CONFIG_USER_ONLY -/* In big-endian (BE32) mode, adjacent Thumb instructions have been swapped -within each word. Undo that now. */ +MemOp end =MO_LE ;if (sctlr_b) { +/* In BE32 mode, adjacent Thumb instructions are swapped. */ addr ^=2 ; + end =MO_BE ; } -#endif -return translator_lduw_swap (env, s, addr,bswap_code (sctlr_b)); +return translator_lduw_end (env, s, addr, end); }static bool insn_crosses_page (CPUARMState *env,DisasContext *s)
@@ -6532,7 +6531,7 @@ static void arm_post_translate_insn(DisasContext *dc)static uint32_t arm_ldl_code (CPUARMState *env,DisasContextBase *s,target_ulong addr,bool sctlr_b) { -return translator_ldl_swap (env, s, addr,bswap_code (sctlr_b)); +return translator_ldl_end (env, s, addr, sctlr_b ?MO_BE :MO_LE ); }static void arm_tr_translate_insn (DisasContextBase *dcbase,CPUState *cpu)
Signature · 1 line
2.47.3
Re: [PATCH v10 18/21] target/arm/tcg/translate.c: replace translator_ldl_swap with translator_ldl_end
Quoted 16 lines · Pierrick Bouvier
Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/tcg/translate.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-)target/arm/tcg/translate.c +0
-2@@ -6284,14 +6284,13 @@ static void disas_thumb_insn(DisasContext *s, uint32_t insn)static uint16_t arm_lduw_code (CPUARMState *env,DisasContextBase * s,target_ulong addr,bool sctlr_b) { -#ifndef CONFIG_USER_ONLY -/* In big-endian (BE32) mode, adjacent Thumb instructions have been swapped
target/arm/tcg/translate.c static void disas_thumb_insn(DisasContext *s, uint32_t insn) · line 6293
Quoted 10 lines
target/arm/tcg/translate.c +4
-3-within each word. Undo that now. */ +MemOp end =MO_LE ;if (sctlr_b) { +/* In BE32 mode, adjacent Thumb instructions are swapped. */ addr ^=2 ; + end =MO_BE ; } -#endif -return translator_lduw_swap (env, s, addr,bswap_code (sctlr_b)); +return translator_lduw_end (env, s, addr, end);
Great, this removes the last translator_lduw_swap() use \o/
And a lot of complexity is removed :~)
PATCH v10 19/21 target/arm/cpu.c: simplify endianness handling in arm_disas_set_info
In reply to Pierrick Bouvier--- target/arm/cpu.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
target/arm/cpu.c +1 -5
@@ -828,15 +828,11 @@ static void arm_disas_set_info(const CPUState *cpu, disassemble_info *info) } info->endian =BFD_ENDIAN_LITTLE ; -if (bswap_code (sctlr_b)) { - info->endian =target_big_endian () ?BFD_ENDIAN_LITTLE :BFD_ENDIAN_BIG ; - } info->flags &= ~INSN_ARM_BE32 ; -#ifndef CONFIG_USER_ONLY if (sctlr_b) { + info->endian |=BFD_ENDIAN_BIG ; info->flags |=INSN_ARM_BE32 ; } -#endif }static void aarch64_cpu_dump_state (CPUState *cs,FILE *f,int flags)
Signature · 1 line
2.47.3
Re: [PATCH v10 19/21] target/arm/cpu.c: simplify endianness handling in arm_disas_set_info
Quoted 6 lines · Pierrick Bouvier
Suggested-by Richard Henderson <[email protected]>Reviewed-by Richard Henderson <[email protected]>Signed-off-by Pierrick Bouvier <[email protected]>--- target/arm/cpu.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
:~)
PATCH v10 20/21 target/arm/tcg/translate.c: remove target_ulong
In reply to Pierrick BouvierWe can replace it with uint32_t, because we know it's limited to 32-bit target.
--- target/arm/tcg/translate.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
target/arm/tcg/translate.c +2 -2
@@ -6282,7 +6282,7 @@ static void disas_thumb_insn(DisasContext *s, uint32_t insn)/* Ditto, for a halfword (Thumb) instruction */ static uint16_t arm_lduw_code (CPUARMState *env,DisasContextBase * s, -target_ulong addr,bool sctlr_b) +uint32_t addr,bool sctlr_b) {MemOp end =MO_LE ;if (sctlr_b) {
@@ -6529,7 +6529,7 @@ static void arm_post_translate_insn(DisasContext *dc)/* Load an instruction and return it in the standard little-endian order */ static uint32_t arm_ldl_code (CPUARMState *env,DisasContextBase *s, -target_ulong addr,bool sctlr_b) +uint32_t addr,bool sctlr_b) {return translator_ldl_end (env, s, addr, sctlr_b ?MO_BE :MO_LE ); }
Signature · 1 line
2.47.3
PATCH v10 21/21 target/arm/tcg/translate.c: make compilation unit common
In reply to Pierrick BouvierWe removed all target specifics, and can finally compile this file only once.
--- target/arm/tcg/meson.build | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
target/arm/tcg/meson.build +5 -3
@@ -20,20 +20,18 @@ mve_d = decodetree.process('mve.decode', extra_args: '--decode=disas_mve') m_nocp_d = decodetree.process('m-nocp.decode', extra_args: '--decode=disas_m_nocp') -gen_a32 = [ +translate32_d = [ decodetree.process('a32.decode', extra_args: '--static-decode=disas_a32'), decodetree.process('a32-uncond.decode', extra_args: '--static-decode=disas_a32_uncond'), decodetree.process('t32.decode', extra_args: '--static-decode=disas_t32'), decodetree.process('t16.decode', extra_args: ['-w', '16', '--static-decode=disas_t16']), ] -arm_ss.add(gen_a32) arm_ss.add(when: 'TARGET_AARCH64', if_true: gen_a64) arm_ss.add(when: 'TARGET_AARCH64', if_false: files('stubs32.c')) arm_ss.add(files( 'cpu32.c', - 'translate.c', 'm_helper.c', 'mve_helper.c', ))
@@ -66,6 +64,7 @@ arm_common_system_ss.add( m_nocp_d, mve_d, neon_d, + translate32_d, vfp_d, files( 'cpregs-at.c',
@@ -77,6 +76,7 @@ arm_common_system_ss.add( 'psci.c', 'tlb_helper.c', 'tlb-insns.c', + 'translate.c', 'translate-m-nocp.c', 'translate-mve.c', 'translate-neon.c',
@@ -88,6 +88,7 @@ arm_user_ss.add( m_nocp_d, mve_d, neon_d, + translate32_d, vfp_d, files( 'debug.c',
@@ -96,6 +97,7 @@ arm_user_ss.add( 'neon_helper.c', 'op_helper.c', 'tlb_helper.c', + 'translate.c', 'translate-m-nocp.c', 'translate-mve.c', 'translate-neon.c',
Signature · 1 line
2.47.3
Quoted 11 lines · Pierrick Bouvier
This series tackles (most of) the rest of target/arm, especially tcg folder.
We extract tcg memory operation in a new header, with a new type representing virtual addresses.
Initially, I went down the way to have a dynamic definition of TCGv, which required much more boilerplate and runtime checks. After discussing with Richard, I realized it was not needed, and that could simply split 32 and 64 bit properly in different files instead. This approach will not work with other architectures, requiring something else, but it's enough for target/arm.
Quoted 4 lines
Files left are:
-target/arm/tcg/mve_helper.c
-target/arm/tcg/m_helper.c
-target/arm/tcg/cpu32.c
I remember having an issue with cpu32.c when it registers TYPE_CPU QOM objects, "max" overlaps with cpu64.c one.
As a kludge we could rename arm_max_initfn -> aarch64_max_tcg_initfn and move it to target/arm/tcg/stubs32.c =)
Quoted 129 lines · Pierrick Bouvier
This series tackles (most of) the rest of target/arm, especially tcg folder.
We extract tcg memory operation in a new header, with a new type representing virtual addresses.
Initially, I went down the way to have a dynamic definition of TCGv, which required much more boilerplate and runtime checks. After discussing with Richard, I realized it was not needed, and that could simply split 32 and 64 bit properly in different files instead. This approach will not work with other architectures, requiring something else, but it's enough for target/arm.
Files left are:
-target/arm/tcg/mve_helper.c
-target/arm/tcg/m_helper.c
-target/arm/tcg/cpu32.cv10
---- extract patch for endianness handling in
arm_disas_set_info
- added some commentsv9
--- introduce
aarch64_translate_code*before* adding parameter totranslator_loop
- useTCG_TYPE_VAas address type fortranslator_loop
- remove inline marker on functions merged fromarm_ldst.htotranslate.c
- refactor arm_ldl_code/arm_lduw_codev8
--- make
translate.ccompilation unit common.v7
--- add parameter to
translator_loopto indicate address type for current tb.
- remove TCGv fromtarget/arm/tcg/translate.cv6
--- meson.build: extract decode files in separate variables and explicitly include them in user and system code. The duplication is now minimized: one instance for all system binaries, and one per user binary.
v5
--- reimplemented what v0 did with new approach defining new tcg-op-mem header
- reapply all straightforward patches from v0
- onlytranslate.cis left, but first want to validate that changes here are ok for maintainers before doing more.v0
--See original thread:
https://lore.kernel.org/qemu-devel/[email protected]/#tNote about v1 -> v4
-------------------- After v0 which was trying to do too many changes, v1 to v4 focused on smaller things, and all patches have been merged already.
https://patchew.org/QEMU/[email protected]/Pierrick Bouvier (21):
include/tcg/tcg-op: extract memory operations totcg-op-mem.htarget/arm/translate.h: removetcg-op.hincludetarget/arm/tcg/translate.h: removetcg-op-gvec.hincludetarget/arm/tcg/translate.h: removeTARGET_AARCH64target/arm/tcg/translate-vfp.c: make compilation unit commontarget/arm/tcg/translate-neon.c: make compilation unit commontarget/arm/tcg/translate-mve.c: make compilation unit commontarget/arm/tcg/translate-m-nocp.c: make compilation unit commontarget/arm/tcg/op_helper.c: make compilation unit commontarget/arm/tcg/gengvec.c: make compilation unit commontarget/arm/tcg/translate.c: removeMO_TEusagetarget/arm/tcg/translate.c: replacetarget_ulongwithuint32_ttarget/arm/tcg/translate.c: extractaarch64_translate_code()tcg/translator: add parameter totranslator_loopfor current addr typetarget/arm/tcg/translate.c: replace TCGv withTCGv_vatarget/arm/tcg/translate-a64.c: usetranslator_ldl_endinstead of
arm_ldl_code
target/arm/tcg/arm_ldst.h: merge intranslate.ctarget/arm/tcg/translate.c: replacetranslator_ldl_swapwith
translator_ldl_end
target/arm/cpu.c: simplify endianness handling inarm_disas_set_infotarget/arm/tcg/translate.c: removetarget_ulongtarget/arm/tcg/translate.c: make compilation unit commoninclude/exec/translator.h | 4 +- include/tcg/tcg-op-mem.h | 126 +++++++++++++++++++++++++++++++ include/tcg/tcg-op.h | 100 +----------------------- target/arm/internals.h | 2 + target/arm/tcg/arm_ldst.h | 47 ------------ target/arm/tcg/translate.h | 14 +--- accel/tcg/translate-all.c | 1 - accel/tcg/translator.c | 4 +- target/alpha/translate.c | 3 +- target/arm/cpu.c | 6 +- target/arm/tcg/op_helper.c | 1 - target/arm/tcg/stubs32.c | 24 ++++++ target/arm/tcg/translate-a64.c | 17 ++++- target/arm/tcg/translate-sme.c | 1 + target/arm/tcg/translate-sve.c | 1 + target/arm/tcg/translate.c | 73 +++++++++++------- target/avr/translate.c | 3 +- target/hexagon/translate.c | 3 +- target/hppa/translate.c | 3 +- target/i386/tcg/translate.c | 3 +- target/loongarch/tcg/translate.c | 3 +- target/m68k/translate.c | 3 +- target/microblaze/translate.c | 3 +- target/mips/tcg/translate.c | 3 +- target/or1k/translate.c | 3 +- target/ppc/translate.c | 3 +- target/riscv/translate.c | 3 +- target/rx/translate.c | 3 +- target/s390x/tcg/translate.c | 3 +- target/sh4/translate.c | 3 +- target/sparc/translate.c | 3 +- target/tricore/translate.c | 3 +- target/xtensa/translate.c | 3 +- target/arm/tcg/meson.build | 57 ++++++++++---- 34 files changed, 304 insertions(+), 228 deletions(-)
Quoted 3 lines
create mode 100644
include/tcg/tcg-op-mem.h
create mode 100644target/arm/tcg/stubs32.c
v11 sent:
https://lore.kernel.org/qemu-devel/[email protected]/
Regards,
Pierrick
PATCH v10 1/1 linux-user/arm: Inline bswap_code()
In reply to Pierrick Bouvierbswap_code() is now used once in get_user_code_u32(), inline it and simplify removing dead system code.
--- target/arm/cpu.h | 16 ---------------- linux-user/arm/cpu_loop.c | 10 ++++++++++ 2 files changed, 10 insertions(+), 16 deletions(-)
target/arm/cpu.h +0 -16
@@ -2568,22 +2568,6 @@ static inline int sme_vq(CPUARMState *env)return EX_TBFLAG_A64 (env->hflags,SVL ) +1 ; } -static inline bool bswap_code (bool sctlr_b) -{ -#ifdef CONFIG_USER_ONLY -/* BE8 (SCTLR.B = 0, TARGET_BIG_ENDIAN = 1) is mixed endian. -* The invalid combination SCTLR.B=1/CPSR.E=1/TARGET_BIG_ENDIAN=0 -* would also end up as a mixed-endian mode with BE code, LE data. -*/ -return TARGET_BIG_ENDIAN ^ sctlr_b; -#else -/* All code access in ARM is little endian, and there are no loaders -* doing swaps that need to be reversed -*/ -return 0 ; -#endif -} -enum {QEMU_PSCI_CONDUIT_DISABLED =0 ,QEMU_PSCI_CONDUIT_SMC =1 ,
linux-user/arm/cpu_loop.c +10 -0
@@ -29,6 +29,16 @@#include "user/page-protection.h" #include "target/arm/syndrome.h" +static inline bool bswap_code (bool sctlr_b) +{ +/* +* BE8 (SCTLR.B = 0, TARGET_BIG_ENDIAN = 1) is mixed endian. +* The invalid combination SCTLR.B=1/CPSR.E=1/TARGET_BIG_ENDIAN=0 +* would also end up as a mixed-endian mode with BE code, LE data. +*/ +return TARGET_BIG_ENDIAN ^ sctlr_b; +} +#define get_user_code_u32 (x, gaddr, env) \ ({ abi_long __r = get_user_u32((x), (gaddr)); \ if (!__r && bswap_code(arm_sctlr_b(env))) { \
Signature · 1 line
2.53.0
Re: [PATCH v10 19.5/21] linux-user/arm: Inline bswap_code()
Quoted 47 lines · Philippe Mathieu-Daudé
bswap_code()is now used once inget_user_code_u32(), inline it and simplify removing dead system code.Signed-off-by Philippe Mathieu-Daudé <[email protected]>--- target/arm/cpu.h | 16 ---------------- linux-user/arm/cpu_loop.c | 10 ++++++++++ 2 files changed, 10 insertions(+), 16 deletions(-)target/arm/cpu.h +0
-16@@ -2568,22 +2568,6 @@ static inline int sme_vq(CPUARMState *env)return EX_TBFLAG_A64 (env->hflags,SVL ) +1 ; } -static inline bool bswap_code (bool sctlr_b) -{ -#ifdef CONFIG_USER_ONLY -/* BE8 (SCTLR.B = 0, TARGET_BIG_ENDIAN = 1) is mixed endian. -* The invalid combination SCTLR.B=1/CPSR.E=1/TARGET_BIG_ENDIAN=0 -* would also end up as a mixed-endian mode with BE code, LE data. -*/ -return TARGET_BIG_ENDIAN ^ sctlr_b; -#else -/* All code access in ARM is little endian, and there are no loaders -* doing swaps that need to be reversed -*/ -return 0 ; -#endif -} -enum {QEMU_PSCI_CONDUIT_DISABLED =0 ,QEMU_PSCI_CONDUIT_SMC =1 ,linux-user/arm/cpu_loop.c +3
-0@@ -29,6 +29,16 @@#include "user/page-protection.h" #include "target/arm/syndrome.h" +static inline bool bswap_code (bool sctlr_b) +{ +/*
linux-user/arm/cpu_loop.c line 44
Quoted 10 lines
linux-user/arm/cpu_loop.c +7
-0+* BE8 (SCTLR.B = 0, TARGET_BIG_ENDIAN = 1) is mixed endian. +* The invalid combination SCTLR.B=1/CPSR.E=1/TARGET_BIG_ENDIAN=0 +* would also end up as a mixed-endian mode with BE code, LE data. +*/ +return TARGET_BIG_ENDIAN ^ sctlr_b; +} +#define get_user_code_u32 (x, gaddr, env) \ ({ abi_long __r = get_user_u32((x), (gaddr)); \ if (!__r && bswap_code(arm_sctlr_b(env))) { \
I don't think so, Richard wants to remove bswap_code completely and get_user_code_u32 is last usage of it. I'm waiting for his patch to integrate it in series.