tools/nolibc: shrink arch support
Subsystems (3)
v2 cover tools/nolibc: shrink arch support
Hi, Willy
This is v2 of the "tools/nolibc: shrink arch support" [1].
This v2 has no core code logic change, but applies some suggestions from Willy and Thomas, one is using post-whitespaces instead of post-tab, another is restructuring the arch support directory and files [2].
Like musl, this v2 creates <ARCH> directory for every arch and splits the old arch-<ARCH>.h to <ARCH>/{crt.h, sys.h} and at the same time, splits the old arch.h to crt_arch.h and sys_arch.h. at last, only need to include crt_arch.h in crt.h and sys_arch.h in sys.h respectively, and no longer need to include arch.h in the other common headers:
crt.h <-- crt_arch.h <-- <ARCH>/crt.h
sys.h <-- sys_arch.h <-- <ARCH>/sys.hIt is based on the 20230705-nolibc-series2 branch of nolibc repo [3]. It should be applied after the v6 __sysret helper series [4] and the v4 min config support series [5].
Here is the test report for all of the supported architectures:
arch/board | result
------------|------------
arm/vexpress-a9 | 142 test(s) passed, 1 skipped, 0 failed.
arm/virt | 142 test(s) passed, 1 skipped, 0 failed.
aarch64/virt | 142 test(s) passed, 1 skipped, 0 failed.
ppc/g3beige | not supported
ppc/ppce500 | not supported
i386/pc | 142 test(s) passed, 1 skipped, 0 failed.
x86_64/pc | 142 test(s) passed, 1 skipped, 0 failed.
mipsel/malta | 142 test(s) passed, 1 skipped, 0 failed.
loongarch64/virt | 142 test(s) passed, 1 skipped, 0 failed.
riscv64/virt | 142 test(s) passed, 1 skipped, 0 failed.
riscv32/virt | 0 test(s) passed, 0 skipped, 0 failed.
s390x/s390-ccw-virtio | 142 test(s) passed, 1 skipped, 0 failed.Changes from v1 --> v2:
* tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h tools/nolibc: split arch.h to crt.h and sys.h
Restruct the arch support directory and files.
Fix up the errors reported by scripts/checkpatch.pl.
* tools/nolibc: sys.h: remove the old sys_stat support
Rebase on the new arch support directory and files.
* tools/nolibc: crt.h: add _start_c
Move #include "compiler.h" in the common crt.h too.
* tools/nolibc: arm/crt.h: shrink _start with _start_c tools/nolibc: aarch64/crt.h: shrink _start with _start_c tools/nolibc: i386/crt.h: shrink _start with _start_c tools/nolibc: x86_64/crt.h: shrink _start with _start_c tools/nolibc: mips/crt.h: shrink _start with _start_c tools/nolibc: loongarch/crt.h: shrink _start with _start_c tools/nolibc: riscv/crt.h: shrink _start with _start_c tools/nolibc: s390/crt.h: shrink _start with _start_c
Rebase on the new arch support directory and files.
Use post-whitespaces instead of post-tab.
Best regards,
Zhangjin
---
[1]: https://lore.kernel.org/lkml/[email protected]/
[2]: https://lore.kernel.org/lkml/[email protected]/
[3]: https://git.kernel.org/pub/scm/linux/kernel/git/wtarreau/nolibc.git
[4]: https://lore.kernel.org/lkml/[email protected]/
[5]: https://lore.kernel.org/lkml/[email protected]/
Zhangjin Wu (12):
tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h tools/nolibc: split arch.h to crt.h and sys.h tools/nolibc: sys.h: remove the old sys_stat support tools/nolibc: crt.h: add _start_c
tools/nolibc: arm/crt.h: shrink _start with _start_c tools/nolibc: aarch64/crt.h: shrink _start with _start_c tools/nolibc: i386/crt.h: shrink _start with _start_c tools/nolibc: x86_64/crt.h: shrink _start with _start_c tools/nolibc: mips/crt.h: shrink _start with _start_c tools/nolibc: loongarch/crt.h: shrink _start with _start_c tools/nolibc: riscv/crt.h: shrink _start with _start_c tools/nolibc: s390/crt.h: shrink _start with _start_c
tools/include/nolibc/Makefile | 36 ++++---
tools/include/nolibc/aarch64/crt.h | 24 +++++
.../nolibc/{arch-aarch64.h => aarch64/sys.h} | 68 +------------
tools/include/nolibc/arch.h | 36 -------
tools/include/nolibc/arm/crt.h | 25 +++++
.../include/nolibc/{arch-arm.h => arm/sys.h} | 96 +------------------
tools/include/nolibc/crt.h | 60 ++++++++++++
tools/include/nolibc/crt_arch.h | 32 +++++++
tools/include/nolibc/i386/crt.h | 33 +++++++
.../nolibc/{arch-i386.h => i386/sys.h} | 77 +--------------
tools/include/nolibc/loongarch/crt.h | 30 ++++++
.../{arch-loongarch.h => loongarch/sys.h} | 64 +------------
tools/include/nolibc/mips/crt.h | 32 +++++++
.../nolibc/{arch-mips.h => mips/sys.h} | 87 +----------------
tools/include/nolibc/nolibc.h | 2 +-
tools/include/nolibc/riscv/crt.h | 28 ++++++
.../nolibc/{arch-riscv.h => riscv/sys.h} | 83 +---------------
tools/include/nolibc/s390/crt.h | 21 ++++
.../nolibc/{arch-s390.h => s390/sys.h} | 74 +-------------
tools/include/nolibc/signal.h | 1 -
tools/include/nolibc/stdio.h | 1 -
tools/include/nolibc/stdlib.h | 2 +-
tools/include/nolibc/sys.h | 65 +++----------
tools/include/nolibc/sys_arch.h | 32 +++++++
tools/include/nolibc/time.h | 1 -
tools/include/nolibc/types.h | 4 +-
tools/include/nolibc/unistd.h | 1 -
tools/include/nolibc/x86_64/crt.h | 33 +++++++
.../nolibc/{arch-x86_64.h => x86_64/sys.h} | 74 +-------------
29 files changed, 421 insertions(+), 701 deletions(-)
create mode 100644 tools/include/nolibc/aarch64/crt.h
rename tools/include/nolibc/{arch-aarch64.h => aarch64/sys.h} (76%)
delete mode 100644 tools/include/nolibc/arch.h
create mode 100644 tools/include/nolibc/arm/crt.h
rename tools/include/nolibc/{arch-arm.h => arm/sys.h} (74%)
create mode 100644 tools/include/nolibc/crt.h
create mode 100644 tools/include/nolibc/crt_arch.h
create mode 100644 tools/include/nolibc/i386/crt.h
rename tools/include/nolibc/{arch-i386.h => i386/sys.h} (73%)
create mode 100644 tools/include/nolibc/loongarch/crt.h
rename tools/include/nolibc/{arch-loongarch.h => loongarch/sys.h} (73%)
create mode 100644 tools/include/nolibc/mips/crt.h
rename tools/include/nolibc/{arch-mips.h => mips/sys.h} (74%)
create mode 100644 tools/include/nolibc/riscv/crt.h
rename tools/include/nolibc/{arch-riscv.h => riscv/sys.h} (70%)
create mode 100644 tools/include/nolibc/s390/crt.h
rename tools/include/nolibc/{arch-s390.h => s390/sys.h} (68%)
create mode 100644 tools/include/nolibc/sys_arch.h
create mode 100644 tools/include/nolibc/x86_64/crt.h
rename tools/include/nolibc/{arch-x86_64.h => x86_64/sys.h} (76%)Signature · 1 line
2.25.1
PATCH v2 1/12 tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
Currently, the architecture specific arch.h has two parts, one is the syscall declarations for sys.h, another is the _start code definition for startup support.
The coming crt.h will provide the startup support with a new common _start_c(), it will replace most of the assembly _start code and shrink the original _start code to be minimal, as a result, _start_c() and the left minimal _start code will work together to provide the startup support, therefore, the left _start code will be only required by crt.h.
So, the syscall declarations part of arch.h can be split to sys_arch.h and the _start code part of arch.h can be split to crt_arch.h and then, they should only be included in sys.h and crt.h respectively.
At the same time, the architecture specific arch-<ARCH>.h should be split to <ARCH>/crt.h and <ARCH>/sys.h.
As a preparation, this creates the architecture specific directory and moves tools/include/nolibc/arch-<ARCH>.h to tools/include/nolibc/<ARCH>/arch.h.
---
tools/include/nolibc/Makefile | 10 +++++-----
.../nolibc/{arch-aarch64.h => aarch64/arch.h} | 6 +++---
tools/include/nolibc/arch.h | 16 ++++++++--------
tools/include/nolibc/{arch-arm.h => arm/arch.h} | 6 +++---
.../include/nolibc/{arch-i386.h => i386/arch.h} | 6 +++---
.../{arch-loongarch.h => loongarch/arch.h} | 6 +++---
.../include/nolibc/{arch-mips.h => mips/arch.h} | 6 +++---
.../nolibc/{arch-riscv.h => riscv/arch.h} | 6 +++---
.../include/nolibc/{arch-s390.h => s390/arch.h} | 6 +++---
.../nolibc/{arch-x86_64.h => x86_64/arch.h} | 6 +++---
10 files changed, 37 insertions(+), 37 deletions(-)
rename tools/include/nolibc/{arch-aarch64.h => aarch64/arch.h} (99%)
rename tools/include/nolibc/{arch-arm.h => arm/arch.h} (99%)
rename tools/include/nolibc/{arch-i386.h => i386/arch.h} (99%)
rename tools/include/nolibc/{arch-loongarch.h => loongarch/arch.h} (99%)
rename tools/include/nolibc/{arch-mips.h => mips/arch.h} (99%)
rename tools/include/nolibc/{arch-riscv.h => riscv/arch.h} (99%)
rename tools/include/nolibc/{arch-s390.h => s390/arch.h} (98%)
rename tools/include/nolibc/{arch-x86_64.h => x86_64/arch.h} (99%)tools/include/nolibc/Makefile +5 -5
@@ -24,7 +24,7 @@ Q=@ endif nolibc_arch := $(patsubst arm64,aarch64,$(ARCH)) -arch_file := arch-$(nolibc_arch).h +arch_file := $(nolibc_arch)/arch.h all_files := \ compiler.h \ ctype.h \
@@ -70,11 +70,11 @@ headers: $(Q)cp $(all_files) $(OUTPUT)sysroot/include/ $(Q)if [ "$(ARCH)" = "x86" ]; then \ sed -e \ - 's,^#ifndef _NOLIBC_ARCH_X86_64_H,#if !defined(_NOLIBC_ARCH_X86_64_H) \&\& defined(__x86_64__),' \ - arch-x86_64.h; \ + 's,^#ifndef _NOLIBC_X86_64_ARCH_H,#if !defined(_NOLIBC_X86_64_ARCH_H) \&\& defined(__x86_64__),' \ + x86_64/arch.h; \ sed -e \ - 's,^#ifndef _NOLIBC_ARCH_I386_H,#if !defined(_NOLIBC_ARCH_I386_H) \&\& !defined(__x86_64__),' \ - arch-i386.h; \ + 's,^#ifndef _NOLIBC_I386_ARCH_H,#if !defined(_NOLIBC_I386_ARCH_H) \&\& !defined(__x86_64__),' \ + i386/arch.h; \ elif [ -e "$(arch_file)" ]; then \ cat $(arch_file); \ else \
tools/include/nolibc/aarch64/arch.h renamed from tools/include/nolibc/arch-aarch64.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_ARCH_AARCH64_H -#define _NOLIBC_ARCH_AARCH64_H +#ifndef _NOLIBC_AARCH64_ARCH_H +#define _NOLIBC_AARCH64_ARCH_H#include "compiler.h"
@@ -201,4 +201,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr ); __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_AARCH64_H */ +#endif /* _NOLIBC_AARCH64_ARCH_H */
tools/include/nolibc/arch.h +8 -8
@@ -16,21 +16,21 @@#define _NOLIBC_ARCH_H#if defined(__x86_64__) -#include "arch-x86_64.h" +#include "x86_64/arch.h" #elif defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) -#include "arch-i386.h" +#include "i386/arch.h" #elif defined(__ARM_EABI__) -#include "arch-arm.h" +#include "arm/arch.h" #elif defined(__aarch64__) -#include "arch-aarch64.h" +#include "aarch64/arch.h" #elif defined(__mips__) && defined(_ABIO32) -#include "arch-mips.h" +#include "mips/arch.h" #elif defined(__riscv) -#include "arch-riscv.h" +#include "riscv/arch.h" #elif defined(__s390x__) -#include "arch-s390.h" +#include "s390/arch.h" #elif defined(__loongarch__) -#include "arch-loongarch.h" +#include "loongarch/arch.h" #endif #endif /* _NOLIBC_ARCH_H */
tools/include/nolibc/arm/arch.h renamed from tools/include/nolibc/arch-arm.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_ARCH_ARM_H -#define _NOLIBC_ARCH_ARM_H +#ifndef _NOLIBC_ARM_ARCH_H +#define _NOLIBC_ARM_ARCH_H#include "compiler.h"
@@ -267,4 +267,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_ARM_H */ +#endif /* _NOLIBC_ARM_ARCH_H */
tools/include/nolibc/i386/arch.h renamed from tools/include/nolibc/arch-i386.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_ARCH_I386_H -#define _NOLIBC_ARCH_I386_H +#ifndef _NOLIBC_I386_ARCH_H +#define _NOLIBC_I386_ARCH_H#include "compiler.h"
@@ -221,4 +221,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_I386_H */ +#endif /* _NOLIBC_I386_ARCH_H */
tools/include/nolibc/loongarch/arch.h renamed from tools/include/nolibc/arch-loongarch.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2023 Loongson Technology Corporation Limited */ -#ifndef _NOLIBC_ARCH_LOONGARCH_H -#define _NOLIBC_ARCH_LOONGARCH_H +#ifndef _NOLIBC_LOONGARCH_ARCH_H +#define _NOLIBC_LOONGARCH_ARCH_H#include "compiler.h"
@@ -197,4 +197,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_LOONGARCH_H */ +#endif /* _NOLIBC_LOONGARCH_ARCH_H */
tools/include/nolibc/mips/arch.h renamed from tools/include/nolibc/arch-mips.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_ARCH_MIPS_H -#define _NOLIBC_ARCH_MIPS_H +#ifndef _NOLIBC_MIPS_ARCH_H +#define _NOLIBC_MIPS_ARCH_H#include "compiler.h"
@@ -250,4 +250,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_MIPS_H */ +#endif /* _NOLIBC_MIPS_ARCH_H */
tools/include/nolibc/riscv/arch.h renamed from tools/include/nolibc/arch-riscv.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_ARCH_RISCV_H -#define _NOLIBC_ARCH_RISCV_H +#ifndef _NOLIBC_RISCV_ARCH_H +#define _NOLIBC_RISCV_ARCH_H#include "compiler.h"
@@ -214,4 +214,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_RISCV_H */ +#endif /* _NOLIBC_RISCV_ARCH_H */
tools/include/nolibc/s390/arch.h renamed from tools/include/nolibc/arch-s390.h +3 -3
@@ -3,8 +3,8 @@* s390 specific definitions for NOLIBC */ -#ifndef _NOLIBC_ARCH_S390_H -#define _NOLIBC_ARCH_S390_H +#ifndef _NOLIBC_S390_ARCH_H +#define _NOLIBC_S390_ARCH_H#include <asm/signal.h> #include <asm/unistd.h>
@@ -234,4 +234,4 @@ pid_t sys_fork(void) }#define sys_fork sys_fork -#endif /* _NOLIBC_ARCH_S390_H */ +#endif /* _NOLIBC_S390_ARCH_H */
tools/include/nolibc/x86_64/arch.h renamed from tools/include/nolibc/arch-x86_64.h +3 -3
@@ -4,8 +4,8 @@* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_ARCH_X86_64_H -#define _NOLIBC_ARCH_X86_64_H +#ifndef _NOLIBC_X86_64_ARCH_H +#define _NOLIBC_X86_64_ARCH_H#include "compiler.h"
@@ -217,4 +217,4 @@ void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_pr __builtin_unreachable(); } -#endif /* _NOLIBC_ARCH_X86_64_H */ +#endif /* _NOLIBC_X86_64_ARCH_H */
Signature · 1 line
2.25.1
PATCH v2 2/12 tools/nolibc: split arch.h to crt.h and sys.h
In reply to Zhangjin WuFor each architecture, arch.h has the syscall declarations and the _start code definition, let's split it to sys.h and crt.h respectively:
- local : <ARCH>/arch.h -> <ARCH>/{sys.h, crt.h}
- global: arch.h -> sys_arch.h, crt_arch.h
And then, we can include the architecture specific file to the common file like this:
- crt.h: #include "crt_arch.h"
- sys.h: #include "sys_arch.h"
At last, the "arch.h" header can be removed from all of the other headers, instead, we only need to include "sys.h" and "crt.h" on demand.
BTW, the following errors reported by scripts/checkpatch.pl are fixed:
ERROR: space required after that ',' (ctx:VxV)
#1327: FILE: tools/include/nolibc/x86_64/crt.h:17:
+void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_protector _start(void)
---
tools/include/nolibc/Makefile | 36 +++++++----
tools/include/nolibc/aarch64/crt.h | 37 +++++++++++
.../include/nolibc/aarch64/{arch.h => sys.h} | 40 ++----------
tools/include/nolibc/arch.h | 36 -----------
tools/include/nolibc/arm/crt.h | 53 +++++++++++++++
tools/include/nolibc/arm/{arch.h => sys.h} | 59 ++---------------
tools/include/nolibc/crt.h | 16 +++++
tools/include/nolibc/crt_arch.h | 32 ++++++++++
tools/include/nolibc/i386/crt.h | 48 ++++++++++++++
tools/include/nolibc/i386/{arch.h => sys.h} | 51 ++-------------
tools/include/nolibc/loongarch/crt.h | 61 ++++++++++++++++++
.../nolibc/loongarch/{arch.h => sys.h} | 64 ++-----------------
tools/include/nolibc/mips/crt.h | 56 ++++++++++++++++
tools/include/nolibc/mips/{arch.h => sys.h} | 59 ++---------------
tools/include/nolibc/nolibc.h | 2 +-
tools/include/nolibc/riscv/crt.h | 57 +++++++++++++++++
tools/include/nolibc/riscv/{arch.h => sys.h} | 60 ++---------------
tools/include/nolibc/s390/crt.h | 45 +++++++++++++
tools/include/nolibc/s390/{arch.h => sys.h} | 49 ++------------
tools/include/nolibc/signal.h | 1 -
tools/include/nolibc/stdio.h | 1 -
tools/include/nolibc/stdlib.h | 2 +-
tools/include/nolibc/sys.h | 2 +-
tools/include/nolibc/sys_arch.h | 32 ++++++++++
tools/include/nolibc/time.h | 1 -
tools/include/nolibc/unistd.h | 1 -
tools/include/nolibc/x86_64/crt.h | 44 +++++++++++++
tools/include/nolibc/x86_64/{arch.h => sys.h} | 47 ++------------
28 files changed, 539 insertions(+), 453 deletions(-)
create mode 100644 tools/include/nolibc/aarch64/crt.h
rename tools/include/nolibc/aarch64/{arch.h => sys.h} (82%)
delete mode 100644 tools/include/nolibc/arch.h
create mode 100644 tools/include/nolibc/arm/crt.h
rename tools/include/nolibc/arm/{arch.h => sys.h} (81%)
create mode 100644 tools/include/nolibc/crt.h
create mode 100644 tools/include/nolibc/crt_arch.h
create mode 100644 tools/include/nolibc/i386/crt.h
rename tools/include/nolibc/i386/{arch.h => sys.h} (79%)
create mode 100644 tools/include/nolibc/loongarch/crt.h
rename tools/include/nolibc/loongarch/{arch.h => sys.h} (73%)
create mode 100644 tools/include/nolibc/mips/crt.h
rename tools/include/nolibc/mips/{arch.h => sys.h} (79%)
create mode 100644 tools/include/nolibc/riscv/crt.h
rename tools/include/nolibc/riscv/{arch.h => sys.h} (78%)
create mode 100644 tools/include/nolibc/s390/crt.h
rename tools/include/nolibc/s390/{arch.h => sys.h} (78%)
create mode 100644 tools/include/nolibc/sys_arch.h
create mode 100644 tools/include/nolibc/x86_64/crt.h
rename tools/include/nolibc/x86_64/{arch.h => sys.h} (81%)tools/include/nolibc/Makefile +22 -14
@@ -24,9 +24,13 @@ Q=@ endif nolibc_arch := $(patsubst arm64,aarch64,$(ARCH)) -arch_file := $(nolibc_arch)/arch.h +arch_files := \ + $(nolibc_arch)/crt.h \ + $(nolibc_arch)/sys.h + all_files := \ compiler.h \ + crt.h \ ctype.h \ errno.h \ nolibc.h \
@@ -68,19 +72,23 @@ headers: $(Q)mkdir -p $(OUTPUT)sysroot $(Q)mkdir -p $(OUTPUT)sysroot/include $(Q)cp $(all_files) $(OUTPUT)sysroot/include/ - $(Q)if [ "$(ARCH)" = "x86" ]; then \ - sed -e \ - 's,^#ifndef _NOLIBC_X86_64_ARCH_H,#if !defined(_NOLIBC_X86_64_ARCH_H) \&\& defined(__x86_64__),' \ - x86_64/arch.h; \ - sed -e \ - 's,^#ifndef _NOLIBC_I386_ARCH_H,#if !defined(_NOLIBC_I386_ARCH_H) \&\& !defined(__x86_64__),' \ - i386/arch.h; \ - elif [ -e "$(arch_file)" ]; then \ - cat $(arch_file); \ - else \ - echo "Fatal: architecture $(ARCH) not yet supported by nolibc." >&2; \ - exit 1; \ - fi > $(OUTPUT)sysroot/include/arch.h + $(Q)for arch_file in $(arch_files); do \ + base_file=$$(basename $$arch_file); \ + arch_header=$$(echo $$base_file | sed -e 's/\.h$$/_arch.h/g'); \ + if [ "$(ARCH)" = "x86" ]; then \ + sed -e \ + 's,^#ifndef _NOLIBC_X86_64_\([^_]*\)_H,#if !defined(_NOLIBC_X86_64_\1_H) \&\& defined(__x86_64__),' \ + x86_64/$$base_file; \ + sed -e \ + 's,^#ifndef _NOLIBC_I386_\([^_]*\)_H,#if !defined(_NOLIBC_I386_\1_H) \&\& !defined(__x86_64__),' \ + i386/$$base_file; \ + elif [ -e "$$arch_file" ]; then \ + cat $$arch_file; \ + else \ + echo "Fatal: architecture $(ARCH) not yet supported by nolibc." >&2; \ + exit 1; \ + fi > $(OUTPUT)sysroot/include/$$arch_header; \ + done headers_standalone: headers $(Q)$(MAKE) -C $(srctree) headers
tools/include/nolibc/aarch64/crt.h new file +37 -0
@@ -0,0 +1,37 @@ +/* SPDX-License-Identifier: LGPL-2.1 OR MIT */ +/* +* AARCH64 specific startup code for NOLIBC +* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> +*/ + +#ifndef _NOLIBC_AARCH64_CRT_H +#define _NOLIBC_AARCH64_CRT_H + +/* startup code */ +void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_protector _start(void) +{ + __asm__volatile ( +#ifdef _NOLIBC_STACKPROTECTOR + "bl __stack_chk_init\n" /* initialize stack protector */ +#endif +"ldr x0, [sp]\n" /* argc (x0) was in the stack */ +"add x1, sp, 8\n" /* argv (x1) = sp */ +"lsl x2, x0, 3\n" /* envp (x2) = 8*argc ... */ +"add x2, x2, 8\n" /* + 8 (skip null) */ +"add x2, x2, x1\n" /* + argv */ +"adrp x3, environ\n" /* x3 = &environ (high bits) */ +"str x2, [x3, #:lo12:environ]\n" /* store envp into environ */ +"mov x4, x2\n" /* search for auxv (follows NULL after last env) */ +"0:\n" +"ldr x5, [x4], 8\n" /* x5 = *x4; x4 += 8 */ +"cbnz x5, 0b\n" /* and stop at NULL after last env */ +"adrp x3, _auxv\n" /* x3 = &_auxv (high bits) */ +"str x4, [x3, #:lo12:_auxv]\n" /* store x4 into _auxv */ +"and sp, x1, -16\n" /* sp must be 16-byte aligned in the callee */ +"bl main\n" /* main() returns the status code, we'll exit with it. */ +"mov x8, 93\n" /* NR_exit == 93 */ +"svc #0\n" + ); +__builtin_unreachable (); +} +#endif /* _NOLIBC_AARCH64_CRT_H */
tools/include/nolibc/aarch64/sys.h renamed from tools/include/nolibc/aarch64/arch.h +4 -36
@@ -1,13 +1,11 @@/* SPDX-License-Identifier: LGPL-2.1 OR MIT */ /* -* AARCH64 specific definitions for NOLIBC +* AARCH64 specific syscall declarations for NOLIBC * Copyright (C) 2017-2022 Willy Tarreau <[email protected]> */ -#ifndef _NOLIBC_AARCH64_ARCH_H -#define _NOLIBC_AARCH64_ARCH_H - -#include "compiler.h" +#ifndef _NOLIBC_AARCH64_SYS_H +#define _NOLIBC_AARCH64_SYS_H/* The struct returned by the newfstatat() syscall. Differs slightly from the * x86_64's stat one by field ordering, so be careful.
@@ -171,34 +169,4 @@ struct sys_stat_struct { _arg1; \ }) -char **environ __attribute__((weak)); -const unsigned long *_auxv __attribute__((weak)); - -/* startup code */ -void __attribute__((weak,noreturn,optimize("omit-frame-pointer"))) __no_stack_protector _start(void) -{ - __asm__volatile ( -#ifdef _NOLIBC_STACKPROTECTOR - "bl __stack_chk_init\n" /* initialize stack protector */ -#endif -"ldr x0, [sp]\n" /* argc (x0) was in the stack */ -"add x1, sp, 8\n" /* argv (x1) = sp */ -"lsl x2, x0, 3\n" /* envp (x2) = 8*argc ... */ -"add x2, x2, 8\n" /* + 8 (skip null) */ -"add x2, x2, x1\n" /* + argv */ -"adrp x3, environ\n" /* x3 = &environ (high bits) */ -"str x2, [x3, #:lo12:environ]\n" /* store envp into environ */ -"mov x4, x2\n" /* search for auxv (follows NULL after last env) */ -"0:\n" -"ldr x5, [x4], 8\n" /* x5 = *x4; x4 += 8 */ -"cbnz x5, 0b\n" /* and stop at NULL after last env */ -"adrp x3, _auxv\n" /* x3 = &_auxv (high bits) */ -"str x4, [x3, #:lo12:_auxv]\n" /* store x4 into _auxv */ -"and sp, x1, -16\n" /* sp must be 16-byte aligned in the callee */ -"bl main\n" /* main() returns the status code, we'll exit with it. */ -"mov x8, 93\n" /* NR_exit == 93 */ -"svc #0\n" - ); -__builtin_unreachable (); -} -#endif /* _NOLIBC_AARCH64_ARCH_H */ +#endif /* _NOLIBC_AARCH64_SYS_H */
tools/include/nolibc/arch.h deleted +0 -36
@@ -1,36 +0,0 @@ -/* SPDX-License-Identifier: LGPL-2.1 OR MIT */ -/* -* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> -*/ - -/* Below comes the architecture-specific code. For each architecture, we have -* the syscall declarations and the _start code definition. This is the only -* global part. On all architectures the kernel puts everything in the stack -* before jumping to _start just above us, without any return address (_start -* is not a function but an entry point). So at the stack pointer we find argc. -* Then argv[] begins, and ends at the first NULL. Then we have envp which -* starts and ends with a NULL as well. So envp=argv+argc+1. -*/ - -#ifndef _NOLIBC_ARCH_H -#define _NOLIBC_ARCH_H - -#if defined(__x86_64__) -#include "x86_64/arch.h" -#elif defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) -#include "i386/arch.h" -#elif defined(__ARM_EABI__) -#include "arm/arch.h" -#elif defined(__aarch64__) -#include "aarch64/arch.h" -#elif defined(__mips__) && defined(_ABIO32) -#include "mips/arch.h" -#elif defined(__riscv) -#include "riscv/arch.h" -#elif defined(__s390x__) -#include "s390/arch.h" -#elif defined(__loongarch__) -#include "loongarch/arch.h" -#endif - -#endif /* _NOLIBC_ARCH_H */
tools/include/nolibc/arm/crt.h new file +53 -0
53 lines not on this page · raw message
tools/include/nolibc/arm/sys.h renamed from tools/include/nolibc/arm/arch.h +5 -54
76 lines not on this page · raw message
tools/include/nolibc/crt.h new file +16 -0
@@ -0,0 +1,16 @@ +/* SPDX-License-Identifier: LGPL-2.1 OR MIT */ +/* +* C Run Time support for NOLIBC +* Copyright (C) 2023 Zhangjin Wu <[email protected]> +*/ + +#ifndef _NOLIBC_CRT_H +#define _NOLIBC_CRT_H + +#include "compiler.h" +#include "crt_arch.h" + +char **environ __attribute__((weak)); +const unsigned long *_auxv __attribute__((weak)); + +#endif /* _NOLIBC_CRT_H */
tools/include/nolibc/crt_arch.h new file +32 -0
@@ -0,0 +1,32 @@ +/* SPDX-License-Identifier: LGPL-2.1 OR MIT */ +/* +* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> +*/ + +/* Below comes the architecture-specific code. For each architecture, we have +* the syscall declarations and the _start code definition. This is the global +* part for _start code definition. +*/ + +#ifndef _NOLIBC_CRT_ARCH_H +#define _NOLIBC_CRT_ARCH_H + +#if defined(__x86_64__) +#include "x86_64/crt.h" +#elif defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) +#include "i386/crt.h" +#elif defined(__ARM_EABI__) +#include "arm/crt.h" +#elif defined(__aarch64__) +#include "aarch64/crt.h" +#elif defined(__mips__) && defined(_ABIO32) +#include "mips/crt.h" +#elif defined(__riscv) +#include "riscv/crt.h" +#elif defined(__s390x__) +#include "s390/crt.h" +#elif defined(__loongarch__) +#include "loongarch/crt.h" +#endif + +#endif /* _NOLIBC_CRT_ARCH_H */
tools/include/nolibc/i386/crt.h new file +48 -0
48 lines not on this page · raw message
tools/include/nolibc/i386/sys.h renamed from tools/include/nolibc/i386/arch.h +4 -47
62 lines not on this page · raw message
tools/include/nolibc/loongarch/crt.h new file +61 -0
61 lines not on this page · raw message
tools/include/nolibc/loongarch/sys.h renamed from tools/include/nolibc/loongarch/arch.h +4 -60
75 lines not on this page · raw message
tools/include/nolibc/mips/crt.h new file +56 -0
56 lines not on this page · raw message
tools/include/nolibc/mips/sys.h renamed from tools/include/nolibc/mips/arch.h +4 -55
70 lines not on this page · raw message
tools/include/nolibc/nolibc.h +1 -1
@@ -93,8 +93,8 @@#define _NOLIBC_H#include "std.h" -#include "arch.h" #include "types.h" +#include "crt.h" #include "sys.h" #include "ctype.h" #include "signal.h"
tools/include/nolibc/riscv/crt.h new file +57 -0
57 lines not on this page · raw message
tools/include/nolibc/riscv/sys.h renamed from tools/include/nolibc/riscv/arch.h +4 -56
77 lines not on this page · raw message
tools/include/nolibc/s390/crt.h new file +45 -0
45 lines not on this page · raw message
tools/include/nolibc/s390/sys.h renamed from tools/include/nolibc/s390/arch.h +4 -45
68 lines not on this page · raw message
tools/include/nolibc/signal.h +0 -1
@@ -8,7 +8,6 @@#define _NOLIBC_SIGNAL_H#include "std.h" -#include "arch.h" #include "types.h" #include "sys.h"
tools/include/nolibc/stdio.h +0 -1
@@ -10,7 +10,6 @@#include <stdarg.h> #include "std.h" -#include "arch.h" #include "errno.h" #include "types.h" #include "sys.h"
tools/include/nolibc/stdlib.h +1 -1
@@ -8,8 +8,8 @@#define _NOLIBC_STDLIB_H#include "std.h" -#include "arch.h" #include "types.h" +#include "crt.h" #include "sys.h" #include "string.h" #include <linux/auxvec.h>
tools/include/nolibc/sys.h +1 -1
@@ -23,7 +23,7 @@#include <linux/stat.h> /* for statx() */ #include <linux/prctl.h> -#include "arch.h" +#include "sys_arch.h" #include "errno.h" #include "types.h"
tools/include/nolibc/sys_arch.h new file +32 -0
@@ -0,0 +1,32 @@ +/* SPDX-License-Identifier: LGPL-2.1 OR MIT */ +/* +* Copyright (C) 2017-2022 Willy Tarreau <[email protected]> +*/ + +/* Below comes the architecture-specific code. For each architecture, we have +* the syscall declarations and the _start code definition. This is the global +* part for syscall declarations. +*/ + +#ifndef _NOLIBC_SYS_ARCH_H +#define _NOLIBC_SYS_ARCH_H + +#if defined(__x86_64__) +#include "x86_64/sys.h" +#elif defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) +#include "i386/sys.h" +#elif defined(__ARM_EABI__) +#include "arm/sys.h" +#elif defined(__aarch64__) +#include "aarch64/sys.h" +#elif defined(__mips__) && defined(_ABIO32) +#include "mips/sys.h" +#elif defined(__riscv) +#include "riscv/sys.h" +#elif defined(__s390x__) +#include "s390/sys.h" +#elif defined(__loongarch__) +#include "loongarch/sys.h" +#endif + +#endif /* _NOLIBC_SYS_ARCH_H */
tools/include/nolibc/time.h +0 -1
@@ -8,7 +8,6 @@#define _NOLIBC_TIME_H#include "std.h" -#include "arch.h" #include "types.h" #include "sys.h"
tools/include/nolibc/unistd.h +0 -1
@@ -8,7 +8,6 @@#define _NOLIBC_UNISTD_H#include "std.h" -#include "arch.h" #include "types.h" #include "sys.h"
tools/include/nolibc/x86_64/crt.h new file +44 -0
44 lines not on this page · raw message
tools/include/nolibc/x86_64/sys.h renamed from tools/include/nolibc/x86_64/arch.h +4 -43
58 lines not on this page · raw message
Signature · 1 line
2.25.1
PATCH v2 3/12 tools/nolibc: sys.h: remove the old sys_stat support
In reply to Zhangjin Wu__NR_statx has been added from v4.10:
commit a528d35e8bfc ("statx: Add a system call to make enhanced file info available")
It has been supported by all of the platforms since at least from v4.20 and glibc 2.28.
Let's remove the old arch related and dependent sys_stat support completely.
This is friendly to the future new architecture porting.
--- tools/include/nolibc/aarch64/sys.h | 28 ------------- tools/include/nolibc/arm/sys.h | 37 ------------------ tools/include/nolibc/i386/sys.h | 26 ------------ tools/include/nolibc/mips/sys.h | 28 ------------- tools/include/nolibc/riscv/sys.h | 23 ----------- tools/include/nolibc/s390/sys.h | 25 ------------ tools/include/nolibc/sys.h | 63 ++++++------------------------ tools/include/nolibc/types.h | 4 +- tools/include/nolibc/x86_64/sys.h | 27 ------------- 9 files changed, 13 insertions(+), 248 deletions(-)
tools/include/nolibc/aarch64/sys.h +0 -28
@@ -7,34 +7,6 @@#ifndef _NOLIBC_AARCH64_SYS_H#define _NOLIBC_AARCH64_SYS_H -/* The struct returned by the newfstatat() syscall. Differs slightly from the -* x86_64's stat one by field ordering, so be careful. -*/ -struct sys_stat_struct { -unsigned long st_dev; -unsigned long st_ino; -unsigned int st_mode; -unsigned int st_nlink; -unsigned int st_uid; -unsigned int st_gid; - -unsigned long st_rdev; -unsigned long __pad1; -long st_size; -int st_blksize; -int __pad2; - -long st_blocks; -long st_atime; -unsigned long st_atime_nsec; -long st_mtime; - -unsigned long st_mtime_nsec; -long st_ctime; -unsigned long st_ctime_nsec; -unsigned int __unused[2 ]; -}; -/* Syscalls for AARCH64 : * - registers are 64-bit * - stack is 16-byte aligned
tools/include/nolibc/arm/sys.h +0 -37
@@ -7,43 +7,6 @@#ifndef _NOLIBC_ARM_SYS_H#define _NOLIBC_ARM_SYS_H -/* The struct returned by the stat() syscall, 32-bit only, the syscall returns -* exactly 56 bytes (stops before the unused array). In big endian, the format -* differs as devices are returned as short only. -*/ -struct sys_stat_struct { -#if defined(__ARMEB__) -unsigned short st_dev; -unsigned short __pad1; -#else -unsigned long st_dev; -#endif -unsigned long st_ino; -unsigned short st_mode; -unsigned short st_nlink; -unsigned short st_uid; -unsigned short st_gid; - -#if defined(__ARMEB__) -unsigned short st_rdev; -unsigned short __pad2; -#else -unsigned long st_rdev; -#endif -unsigned long st_size; -unsigned long st_blksize; -unsigned long st_blocks; - -unsigned long st_atime; -unsigned long st_atime_nsec; -unsigned long st_mtime; -unsigned long st_mtime_nsec; - -unsigned long st_ctime; -unsigned long st_ctime_nsec; -unsigned long __unused[2 ]; -}; -/* Syscalls for ARM in ARM or Thumb modes : * - registers are 32-bit * - stack is 8-byte aligned
tools/include/nolibc/i386/sys.h +0 -26
@@ -7,32 +7,6 @@#ifndef _NOLIBC_I386_SYS_H#define _NOLIBC_I386_SYS_H -/* The struct returned by the stat() syscall, 32-bit only, the syscall returns -* exactly 56 bytes (stops before the unused array). -*/ -struct sys_stat_struct { -unsigned long st_dev; -unsigned long st_ino; -unsigned short st_mode; -unsigned short st_nlink; -unsigned short st_uid; -unsigned short st_gid; - -unsigned long st_rdev; -unsigned long st_size; -unsigned long st_blksize; -unsigned long st_blocks; - -unsigned long st_atime; -unsigned long st_atime_nsec; -unsigned long st_mtime; -unsigned long st_mtime_nsec; - -unsigned long st_ctime; -unsigned long st_ctime_nsec; -unsigned long __unused[2 ]; -}; -/* Syscalls for i386 : * - mostly similar to x86_64 * - registers are 32-bit
tools/include/nolibc/mips/sys.h +0 -28
@@ -7,34 +7,6 @@#ifndef _NOLIBC_MIPS_SYS_H#define _NOLIBC_MIPS_SYS_H -/* The struct returned by the stat() syscall. 88 bytes are returned by the -* syscall. -*/ -struct sys_stat_struct { -unsigned int st_dev; -long st_pad1[3 ]; -unsigned long st_ino; -unsigned int st_mode; -unsigned int st_nlink; -unsigned int st_uid; -unsigned int st_gid; -unsigned int st_rdev; -long st_pad2[2 ]; -long st_size; -long st_pad3; - -long st_atime; -long st_atime_nsec; -long st_mtime; -long st_mtime_nsec; - -long st_ctime; -long st_ctime_nsec; -long st_blksize; -long st_blocks; -long st_pad4[14 ]; -}; -/* Syscalls for MIPS ABI O32 : * - WARNING! there's always a delayed slot! * - WARNING again, the syntax is different, registers take a '$' and numbers
tools/include/nolibc/riscv/sys.h +0 -23
@@ -7,29 +7,6 @@#ifndef _NOLIBC_RISCV_SYS_H#define _NOLIBC_RISCV_SYS_H -struct sys_stat_struct { -unsigned long st_dev;/* Device. */ -unsigned long st_ino;/* File serial number. */ -unsigned int st_mode;/* File mode. */ -unsigned int st_nlink;/* Link count. */ -unsigned int st_uid;/* User ID of the file's owner. */ -unsigned int st_gid;/* Group ID of the file's group. */ -unsigned long st_rdev;/* Device number, if device. */ -unsigned long __pad1; -long st_size;/* Size of file, in bytes. */ -int st_blksize;/* Optimal block size for I/O. */ -int __pad2; -long st_blocks;/* Number 512-byte blocks allocated. */ -long st_atime;/* Time of last access. */ -unsigned long st_atime_nsec; -long st_mtime;/* Time of last modification. */ -unsigned long st_mtime_nsec; -long st_ctime;/* Time of last status change. */ -unsigned long st_ctime_nsec; -unsigned int __unused4; -unsigned int __unused5; -}; -/* Syscalls for RISCV : * - stack is 16-byte aligned * - syscall number is passed in a7
tools/include/nolibc/s390/sys.h +0 -25
@@ -8,31 +8,6 @@#include <asm/signal.h> #include <asm/unistd.h> -/* The struct returned by the stat() syscall, equivalent to stat64(). The -* syscall returns 116 bytes and stops in the middle of __unused. -*/ - -struct sys_stat_struct { -unsigned long st_dev; -unsigned long st_ino; -unsigned long st_nlink; -unsigned int st_mode; -unsigned int st_uid; -unsigned int st_gid; -unsigned int __pad1; -unsigned long st_rdev; -unsigned long st_size; -unsigned long st_atime; -unsigned long st_atime_nsec; -unsigned long st_mtime; -unsigned long st_mtime_nsec; -unsigned long st_ctime; -unsigned long st_ctime_nsec; -unsigned long st_blksize; -long st_blocks; -unsigned long __unused[3 ]; -}; -/* Syscalls for s390: * - registers are 64-bit * - syscall number is passed in r1
tools/include/nolibc/sys.h +11 -52
97 lines not on this page · raw message
tools/include/nolibc/types.h +2 -2
@@ -15,8 +15,8 @@/* Only the generic macros and types may be defined here. The arch-specific -* ones such as the O_RDONLY and related macros used by fcntl() and open(), or -* the layout of sys_stat_struct must not be defined here. +* ones such as the O_RDONLY and related macros used by fcntl() and open() +* must not be defined here. */ /* stat flags (WARNING, octal here). We need to check for an existing
tools/include/nolibc/x86_64/sys.h +0 -27
@@ -7,33 +7,6 @@#ifndef _NOLIBC_X86_64_SYS_H#define _NOLIBC_X86_64_SYS_H -/* The struct returned by the stat() syscall, equivalent to stat64(). The -* syscall returns 116 bytes and stops in the middle of __unused. -*/ -struct sys_stat_struct { -unsigned long st_dev; -unsigned long st_ino; -unsigned long st_nlink; -unsigned int st_mode; -unsigned int st_uid; - -unsigned int st_gid; -unsigned int __pad0; -unsigned long st_rdev; -long st_size; -long st_blksize; - -long st_blocks; -unsigned long st_atime; -unsigned long st_atime_nsec; -unsigned long st_mtime; - -unsigned long st_mtime_nsec; -unsigned long st_ctime; -unsigned long st_ctime_nsec; -long __unused[3 ]; -}; -/* Syscalls for x86_64 : * - registers are 64-bit * - syscall number is passed in rax
Signature · 1 line
2.25.1
PATCH v2 4/12 tools/nolibc: crt.h: add _start_c
In reply to Zhangjin WuAs the environ and _auxv support added for nolibc, the assembly _start function becomes more and more complex and therefore makes the porting of nolibc to new architectures harder and harder.
To simplify portability, this c version of _start_c() is added to do most of the assembly start operations in C, which reduces the complexity a lot and will eventually simplify the porting of nolibc to the new architectures.
The new _start_c() only requires a stack pointer argument, it will find argv, envp and _auxv for us, and then call main(), finally, it exit() with main's return status. With this new _start_c(), the future new architectures only require to add very few assembly instructions.
--- tools/include/nolibc/crt.h | 44 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)
tools/include/nolibc/crt.h +44 -0
@@ -13,4 +13,48 @@ char **environ __attribute__((weak)); const unsigned long *_auxv __attribute__((weak)); +int main (int argc,char *argv[],char **envp); +static void exit (int ); + +void _start_c (long *sp) +{ +int argc, i; +char **argv; +char **envp; + +/* +* sp : argc <-- argument count, required by main() +* argv: argv[0] <-- argument vector, required by main() +* argv[1] +* ... +* argv[argc-1] +* null +* envp: envp[0] <-- environment variables, required by main() and getenv() +* envp[1] +* ... +* null +* _auxv: auxv[0] <-- auxiliary vector, required by getauxval() +* auxv[1] +* ... +* null +*/ + +/* assign argc and argv */ + argc = sp[0 ]; + argv = (void *)(sp +1 ); + +/* find envp */ + envp = argv + argc +1 ; + environ = envp; + +/* find auxv */ + i =0 ; +while (envp[i]) + i++; + _auxv = (void *)(envp + i +1 ); + +/* go to application */ +exit (main (argc, argv, envp)); +} +#endif /* _NOLIBC_CRT_H */
Signature · 1 line
2.25.1
PATCH v2 5/12 tools/nolibc: arm/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/arm/crt.h | 38 +++++----------------------------- 1 file changed, 5 insertions(+), 33 deletions(-)
tools/include/nolibc/arm/crt.h +5 -33
@@ -12,40 +12,12 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_ { __asm__volatile ( #ifdef _NOLIBC_STACKPROTECTOR - "bl __stack_chk_init\n" /* initialize stack protector */ + "bl __stack_chk_init\n" /* initialize stack protector */ #endif -"pop {%r0}\n" /* argc was in the stack */ -"mov %r1, %sp\n" /* argv = sp */ - -"add %r2, %r0, $1\n" /* envp = (argc + 1) ... */ -"lsl %r2, %r2, $2\n" /* * 4 ... */ -"add %r2, %r2, %r1\n" /* + argv */ -"ldr %r3, 1f\n" /* r3 = &environ (see below) */ -"str %r2, [r3]\n" /* store envp into environ */ - -"mov r4, r2\n" /* search for auxv (follows NULL after last env) */ -"0:\n" -"mov r5, r4\n" /* r5 = r4 */ -"add r4, r4, #4\n" /* r4 += 4 */ -"ldr r5,[r5]\n" /* r5 = *r5 = *(r4-4) */ -"cmp r5, #0\n" /* and stop at NULL after last env */ -"bne 0b\n" -"ldr %r3, 2f\n" /* r3 = &_auxv (low bits) */ -"str r4, [r3]\n" /* store r4 into _auxv */ - -"mov %r3, $8\n" /* AAPCS : sp must be 8-byte aligned in the */ -"neg %r3, %r3\n" /* callee, and bl doesn't push (lr=pc) */ -"and %r3, %r3, %r1\n" /* so we do sp = r1(=sp) & r3(=-8); */ -"mov %sp, %r3\n" - -"bl main\n" /* main() returns the status code, we'll exit with it. */ -"movs r7, $1\n" /* NR_exit == 1 */ -"svc $0x00\n" -".align 2\n" /* below are the pointers to a few variables */ -"1:\n" -".word environ\n" -"2:\n" -".word _auxv\n" +"mov %r0, sp\n" /* save stack pointer to %r0, as arg1 of _start_c */ +"and ip, %r0, #-8\n" /* sp must be 8-byte aligned in the callee */ +"mov sp, ip\n" +"bl _start_c\n" /* transfer to c runtime */ );__builtin_unreachable (); }
Signature · 1 line
2.25.1
PATCH v2 6/12 tools/nolibc: aarch64/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/aarch64/crt.h | 23 +++++------------------ 1 file changed, 5 insertions(+), 18 deletions(-)
tools/include/nolibc/aarch64/crt.h +5 -18
@@ -12,26 +12,13 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_ { __asm__volatile ( #ifdef _NOLIBC_STACKPROTECTOR - "bl __stack_chk_init\n" /* initialize stack protector */ + "bl __stack_chk_init\n" /* initialize stack protector */ #endif -"ldr x0, [sp]\n" /* argc (x0) was in the stack */ -"add x1, sp, 8\n" /* argv (x1) = sp */ -"lsl x2, x0, 3\n" /* envp (x2) = 8*argc ... */ -"add x2, x2, 8\n" /* + 8 (skip null) */ -"add x2, x2, x1\n" /* + argv */ -"adrp x3, environ\n" /* x3 = &environ (high bits) */ -"str x2, [x3, #:lo12:environ]\n" /* store envp into environ */ -"mov x4, x2\n" /* search for auxv (follows NULL after last env) */ -"0:\n" -"ldr x5, [x4], 8\n" /* x5 = *x4; x4 += 8 */ -"cbnz x5, 0b\n" /* and stop at NULL after last env */ -"adrp x3, _auxv\n" /* x3 = &_auxv (high bits) */ -"str x4, [x3, #:lo12:_auxv]\n" /* store x4 into _auxv */ -"and sp, x1, -16\n" /* sp must be 16-byte aligned in the callee */ -"bl main\n" /* main() returns the status code, we'll exit with it. */ -"mov x8, 93\n" /* NR_exit == 93 */ -"svc #0\n" +"mov x0, sp\n" /* save stack pointer to x0, as arg1 of _start_c */ +"and sp, x0, -16\n" /* sp must be 16-byte aligned in the callee */ +"bl _start_c\n" /* transfer to c runtime */ );__builtin_unreachable (); + }#endif /* _NOLIBC_AARCH64_CRT_H */
Signature · 1 line
2.25.1
PATCH v2 7/12 tools/nolibc: i386/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/i386/crt.h | 29 +++++++---------------------- 1 file changed, 7 insertions(+), 22 deletions(-)
tools/include/nolibc/i386/crt.h +7 -22
@@ -18,29 +18,14 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_ { __asm__volatile ( #ifdef _NOLIBC_STACKPROTECTOR - "call __stack_chk_init\n" /* initialize stack protector */ + "call __stack_chk_init\n" /* initialize stack protector */ #endif -"pop %eax\n" /* argc (first arg, %eax) */ -"mov %esp, %ebx\n" /* argv[] (second arg, %ebx) */ -"lea 4(%ebx,%eax,4),%ecx\n" /* then a NULL then envp (third arg, %ecx) */ -"mov %ecx, environ\n" /* save environ */ -"xor %ebp, %ebp\n" /* zero the stack frame */ -"mov %ecx, %edx\n" /* search for auxv (follows NULL after last env) */ -"0:\n" -"add $4, %edx\n" /* search for auxv using edx, it follows the */ -"cmp -4(%edx), %ebp\n" /* ... NULL after last env (ebp is zero here) */ -"jnz 0b\n" -"mov %edx, _auxv\n" /* save it into _auxv */ -"and $-16, %esp\n" /* x86 ABI : esp must be 16-byte aligned before */ -"sub $4, %esp\n" /* the call instruction (args are aligned) */ -"push %ecx\n" /* push all registers on the stack so that we */ -"push %ebx\n" /* support both regparm and plain stack modes */ -"push %eax\n" -"call main\n" /* main() returns the status code in %eax */ -"mov %eax, %ebx\n" /* retrieve exit code (32-bit int) */ -"movl $1, %eax\n" /* NR_exit == 1 */ -"int $0x80\n" /* exit now */ -"hlt\n" /* ensure it does not */ +"xor %ebp, %ebp\n" /* zero the stack frame */ +"mov %esp, %eax\n" /* save stack pointer to %eax, as arg1 of _start_c */ +"and $-16, %esp\n" /* last pushed argument must be 16-byte aligned */ +"push %eax\n" /* push arg1 on stack to support plain stack modes too */ +"call _start_c\n" /* transfer to c runtime */ +"hlt\n" /* ensure it does not return */ );__builtin_unreachable (); }
Signature · 1 line
2.25.1
PATCH v2 8/12 tools/nolibc: x86_64/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/x86_64/crt.h | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-)
tools/include/nolibc/x86_64/crt.h +7 -18
@@ -18,27 +18,16 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_ { __asm__volatile ( #ifdef _NOLIBC_STACKPROTECTOR - "call __stack_chk_init\n" /* initialize stack protector */ + "call __stack_chk_init\n" /* initialize stack protector */ #endif -"pop %rdi\n" /* argc (first arg, %rdi) */ -"mov %rsp, %rsi\n" /* argv[] (second arg, %rsi) */ -"lea 8(%rsi,%rdi,8),%rdx\n" /* then a NULL then envp (third arg, %rdx) */ -"mov %rdx, environ\n" /* save environ */ -"xor %ebp, %ebp\n" /* zero the stack frame */ -"mov %rdx, %rax\n" /* search for auxv (follows NULL after last env) */ -"0:\n" -"add $8, %rax\n" /* search for auxv using rax, it follows the */ -"cmp -8(%rax), %rbp\n" /* ... NULL after last env (rbp is zero here) */ -"jnz 0b\n" -"mov %rax, _auxv\n" /* save it into _auxv */ -"and $-16, %rsp\n" /* x86 ABI : esp must be 16-byte aligned before call */ -"call main\n" /* main() returns the status code, we'll exit with it. */ -"mov %eax, %edi\n" /* retrieve exit code (32 bit) */ -"mov $60, %eax\n" /* NR_exit == 60 */ -"syscall\n" /* really exit */ -"hlt\n" /* ensure it does not return */ +"xor %ebp, %ebp\n" /* zero the stack frame */ +"mov %rsp, %rdi\n" /* save stack pointer to %rdi, as arg1 of _start_c */ +"and $-16, %rsp\n" /* %rsp must be 16-byte aligned before call */ +"call _start_c\n" /* transfer to c runtime */ +"hlt\n" /* ensure it does not return */ );__builtin_unreachable (); + }#endif /* _NOLIBC_X86_64_CRT_H */
Signature · 1 line
2.25.1
PATCH v2 9/12 tools/nolibc: mips/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
Also clean up the instructions in delay slots.
--- tools/include/nolibc/mips/crt.h | 42 +++++++-------------------------- 1 file changed, 9 insertions(+), 33 deletions(-)
tools/include/nolibc/mips/crt.h +9 -33
@@ -11,43 +11,19 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_protector __start(void) { __asm__volatile ( -/*".set nomips16\n"*/ ".set push\n" - ".set noreorder\n" + ".set noreorder\n" ".option pic0\n" #ifdef _NOLIBC_STACKPROTECTOR - "jal __stack_chk_init\n" /* initialize stack protector */ - "nop\n" /* delayed slot */ + "jal __stack_chk_init\n" /* initialize stack protector */ + " nop\n" /* delayed slot */ #endif -/*".ent __start\n"*/ -/*"__start:\n"*/ -"lw $a0,($sp)\n" /* argc was in the stack */ -"addiu $a1, $sp, 4\n" /* argv = sp + 4 */ -"sll $a2, $a0, 2\n" /* a2 = argc * 4 */ -"add $a2, $a2, $a1\n" /* envp = argv + 4*argc ... */ -"addiu $a2, $a2, 4\n" /* ... + 4 */ -"lui $a3, %hi(environ)\n" /* load environ into a3 (hi) */ -"addiu $a3, %lo(environ)\n" /* load environ into a3 (lo) */ -"sw $a2,($a3)\n" /* store envp(a2) into environ */ - -"move $t0, $a2\n" /* iterate t0 over envp, look for NULL */ -"0:" /* do { */ -"lw $a3, ($t0)\n" /* a3=*(t0); */ -"bne $a3, $0, 0b\n" /* } while (a3); */ -"addiu $t0, $t0, 4\n" /* delayed slot: t0+=4; */ -"lui $a3, %hi(_auxv)\n" /* load _auxv into a3 (hi) */ -"addiu $a3, %lo(_auxv)\n" /* load _auxv into a3 (lo) */ -"sw $t0, ($a3)\n" /* store t0 into _auxv */ - -"li $t0, -8\n" -"and $sp, $sp, $t0\n" /* sp must be 8-byte aligned */ -"addiu $sp,$sp,-16\n" /* the callee expects to save a0..a3 there! */ -"jal main\n" /* main() returns the status code, we'll exit with it. */ -"nop\n" /* delayed slot */ -"move $a0, $v0\n" /* retrieve 32-bit exit code from v0 */ -"li $v0, 4001\n" /* NR_exit == 4001 */ -"syscall\n" -/*".end __start\n"*/ +"move $a0, $sp\n" /* save stack pointer to $a0, as arg1 of _start_c */ +"li $t0, -8\n" +"and $sp, $sp, $t0\n" /* $sp must be 8-byte aligned */ +"addiu $sp, $sp, -16\n" /* the callee expects to save a0..a3 there */ +"jal _start_c\n" /* transfer to c runtime */ +" nop\n" /* delayed slot */ ".set pop\n" );__builtin_unreachable ();
Signature · 1 line
2.25.1
PATCH v2 10/12 tools/nolibc: loongarch/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/loongarch/crt.h | 39 +++------------------------- 1 file changed, 4 insertions(+), 35 deletions(-)
tools/include/nolibc/loongarch/crt.h +4 -35
@@ -8,22 +8,8 @@#define _NOLIBC_LOONGARCH_CRT_H#if __loongarch_grlen ==32 -#define LONGLOG "2" -#define SZREG "4" -#define REG_L "ld.w" -#define LONG_S "st.w" -#define LONG_ADD "add.w" -#define LONG_ADDI "addi.w" -#define LONG_SLL "slli.w"#define LONG_BSTRINS "bstrins.w"#else /* __loongarch_grlen == 64 */ -#define LONGLOG "3" -#define SZREG "8" -#define REG_L "ld.d" -#define LONG_S "st.d" -#define LONG_ADD "add.d" -#define LONG_ADDI "addi.d" -#define LONG_SLL "slli.d"#define LONG_BSTRINS "bstrins.d"#endif
@@ -32,28 +18,11 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_ { __asm__volatile ( #ifdef _NOLIBC_STACKPROTECTOR - "bl __stack_chk_init\n" /* initialize stack protector */ + "bl __stack_chk_init\n" /* initialize stack protector */ #endif -REG_L " $a0, $sp, 0\n" /* argc (a0) was in the stack */ -LONG_ADDI " $a1, $sp, " SZREG "\n" /* argv (a1) = sp + SZREG */ -LONG_SLL " $a2, $a0, " LONGLOG "\n" /* envp (a2) = SZREG*argc ... */ -LONG_ADDI " $a2, $a2, " SZREG "\n" /* + SZREG (skip null) */ -LONG_ADD " $a2, $a2, $a1\n" /* + argv */ - -"move $a3, $a2\n" /* iterate a3 over envp to find auxv (after NULL) */ -"0:\n" /* do { */ -REG_L " $a4, $a3, 0\n" /* a4 = *a3; */ -LONG_ADDI " $a3, $a3, " SZREG "\n" /* a3 += sizeof(void*); */ -"bne $a4, $zero, 0b\n" /* } while (a4); */ -"la.pcrel $a4, _auxv\n" /* a4 = &_auxv */ -LONG_S " $a3, $a4, 0\n" /* store a3 into _auxv */ - -"la.pcrel $a3, environ\n" /* a3 = &environ */ -LONG_S " $a2, $a3, 0\n" /* store envp(a2) into environ */ -LONG_BSTRINS " $sp, $zero, 3, 0\n" /* sp must be 16-byte aligned */ -"bl main\n" /* main() returns the status code, we'll exit with it. */ -"li.w $a7, 93\n" /* NR_exit == 93 */ -"syscall 0\n" +"move $a0, $sp\n" /* save stack pointer to $a0, as arg1 of _start_c */ +LONG_BSTRINS " $sp, $zero, 3, 0\n" /* $sp must be 16-byte aligned */ +"bl _start_c\n" /* transfer to c runtime */ );__builtin_unreachable (); }
Signature · 1 line
2.25.1
PATCH v2 11/12 tools/nolibc: riscv/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/riscv/crt.h | 39 ++++---------------------------- 1 file changed, 5 insertions(+), 34 deletions(-)
tools/include/nolibc/riscv/crt.h +5 -34
@@ -7,49 +7,20 @@#ifndef _NOLIBC_RISCV_CRT_H#define _NOLIBC_RISCV_CRT_H -#if __riscv_xlen ==64 -#define PTRLOG "3" -#define SZREG "8" -#define REG_L "ld" -#define REG_S "sd" -#elif __riscv_xlen ==32 -#define PTRLOG "2" -#define SZREG "4" -#define REG_L "lw" -#define REG_S "sw" -#endif -/* startup code */ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_protector _start(void) { __asm__volatile ( ".option push\n" ".option norelax\n" - "lla gp, __global_pointer$\n" + "lla gp, __global_pointer$\n" ".option pop\n" #ifdef _NOLIBC_STACKPROTECTOR - "call __stack_chk_init\n" /* initialize stack protector */ + "call __stack_chk_init\n" /* initialize stack protector */ #endif -REG_L " a0, 0(sp)\n" /* argc (a0) was in the stack */ -"add a1, sp, " SZREG "\n" /* argv (a1) = sp */ -"slli a2, a0, " PTRLOG "\n" /* envp (a2) = SZREG*argc ... */ -"add a2, a2, " SZREG "\n" /* + SZREG (skip null) */ -"add a2,a2,a1\n" /* + argv */ - -"add a3, a2, zero\n" /* iterate a3 over envp to find auxv (after NULL) */ -"0:\n" /* do { */ -REG_L " a4, 0(a3)\n" /* a4 = *a3; */ -"add a3, a3, " SZREG "\n" /* a3 += sizeof(void*); */ -"bne a4, zero, 0b\n" /* } while (a4); */ -"lui a4, %hi(_auxv)\n" /* a4 = &_auxv (high bits) */ -REG_S " a3, %lo(_auxv)(a4)\n" /* store a3 into _auxv */ - -"lui a3, %hi(environ)\n" /* a3 = &environ (high bits) */ -REG_S " a2,%lo(environ)(a3)\n" /* store envp(a2) into environ */ -"andi sp,a1,-16\n" /* sp must be 16-byte aligned */ -"call main\n" /* main() returns the status code, we'll exit with it. */ -"li a7, 93\n" /* NR_exit == 93 */ -"ecall\n" +"mv a0, sp\n" /* save stack pointer to a0, as arg1 of _start_c */ +"andi sp, a0, -16\n" /* sp must be 16-byte aligned */ +"call _start_c\n" /* transfer to c runtime */ );__builtin_unreachable (); }
Signature · 1 line
2.25.1
PATCH v2 12/12 tools/nolibc: s390/crt.h: shrink _start with _start_c
In reply to Zhangjin Wumove most of the _start operations to _start_c().
--- tools/include/nolibc/s390/crt.h | 32 ++++---------------------------- 1 file changed, 4 insertions(+), 28 deletions(-)
tools/include/nolibc/s390/crt.h +4 -28
@@ -10,34 +10,10 @@ void __attribute__((weak, noreturn, optimize("omit-frame-pointer"))) __no_stack_protector _start(void) { __asm__volatile ( -"lg %r2,0(%r15)\n" /* argument count */ -"la %r3,8(%r15)\n" /* argument pointers */ - -"xgr %r0,%r0\n" /* r0 will be our NULL value */ -/* search for envp */ -"lgr %r4,%r3\n" /* start at argv */ -"0:\n" -"clg %r0,0(%r4)\n" /* entry zero? */ -"la %r4,8(%r4)\n" /* advance pointer */ -"jnz 0b\n" /* no -> test next pointer */ -/* yes -> r4 now contains start of envp */ -"larl %r1,environ\n" -"stg %r4,0(%r1)\n" - -/* search for auxv */ -"lgr %r5,%r4\n" /* start at envp */ -"1:\n" -"clg %r0,0(%r5)\n" /* entry zero? */ -"la %r5,8(%r5)\n" /* advance pointer */ -"jnz 1b\n" /* no -> test next pointer */ -"larl %r1,_auxv\n" /* yes -> store value in _auxv */ -"stg %r5,0(%r1)\n" - -"aghi %r15,-160\n" /* allocate new stackframe */ -"xc 0(8,%r15),0(%r15)\n" /* clear backchain */ -"brasl %r14,main\n" /* ret value of main is arg to exit */ -"lghi %r1,1\n" /* __NR_exit */ -"svc 0\n" +"lgr %r2, %r15\n" /* save stack pointer to %r2, as arg1 of _start_c */ +"aghi %r15, -160\n" /* allocate new stackframe */ +"xc 0(8,%r15), 0(%r15)\n" /* clear backchain */ +"brasl %r14, _start_c\n" /* transfer to c runtime */ );__builtin_unreachable (); }
Signature · 1 line
2.25.1
Willy Tarreau
maintainerRe: [PATCH v2 01/12] tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
In reply to Zhangjin WuQuoted 14 lines · Zhangjin Wu
Currently, the architecture specific
arch.hhas two parts, one is the syscall declarations forsys.h, another is the_startcode definition for startup support.The coming
crt.hwill provide the startup support with a new common_start_c(), it will replace most of the assembly_startcode and shrink the original_startcode to be minimal, as a result,_start_c()and the left minimal_startcode will work together to provide the startup support, therefore, the left_startcode will be only required bycrt.h.So, the syscall declarations part of
arch.hcan be split tosys_arch.hand the_startcode part ofarch.hcan be split tocrt_arch.hand then, they should only be included insys.handcrt.hrespectively.
Quoted 6 lines
At the same time, the architecture specific arch-<ARCH>.h should be split to <ARCH>/crt.h and <ARCH>/sys.h.
As a preparation, this creates the architecture specific directory and moves tools/include/nolibc/arch-<ARCH>.h to tools/include/nolibc/<ARCH>/arch.h.
I'm sorry but I still don't understand what it *provides*. I'm reading it as "we *can* do this so let's do it". But what is the specific purpose of adding this extra directory structure ? It's really unclear to me and worries me that it'll only result in complicating maintenance by adding even more files, thus even more "include" lines and cross dependencies.
Zhangjin, very often in your series, the justification for a change is missing, instead it's only explaining what is being changed, and I must confess that it makes it particularly difficult to figure the benefits. I'm only seeing this as an opportunity for a change ("can be split"). I could have missed something of course, but I can't figure what problem it is trying to solve.
As a general advice, I tend to remind people that when sending a patch series, they should consider they're trying to sell it, so they must emphasize the benefits of accepting the series for the maintainer(s).
You very likely have a good reason for doing this but I can't see it here so I'm just seeing a change that will possibly add some extra cost (if at least because file locations change again) and nothing more. When you try to reorganize things, it's often much more efficient to try to discuss it before proposing patches, because reorg patches are generally unreadable and take time for you to create and for others to review. Instead, just explaining what you think you can improve is faster for everyone, and others can chime in and propose alternate approaches (something which is very hard to do with a patch series).
Thanks!
Willy
Re: [PATCH v2 04/12] tools/nolibc: crt.h: add _start_c
In reply to Zhangjin WuQuoted 9 lines · Zhangjin Wu
As the environ and
_auxvsupport added for nolibc, the assembly_startfunction becomes more and more complex and therefore makes the porting of nolibc to new architectures harder and harder.To simplify portability, this c version of
_start_c()is added to do most of the assembly start operations in C, which reduces the complexity a lot and will eventually simplify the porting of nolibc to the new architectures.
Quoted 4 lines
The new
_start_c()only requires a stack pointer argument, it will find argv, envp and_auxvfor us, and then callmain(), finally, itexit()with main's return status. With this new_start_c(), the future new architectures only require to add very few assembly instructions.
I like it!
A quick test indicates that the initialization of the stackprotectors could also be moved into the C function.
It also seems like a good opportunity to add some tests for argv/environment variable passing.
And as general note to the full series I think that splitting the arch files is not necessary and confusing.
tools/include/nolibc/crt.h line 13
Quoted 11 lines
Signed-off-by Zhangjin Wu <[email protected]>--- tools/include/nolibc/crt.h | 44 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)tools/include/nolibc/crt.h +0
-0@@ -13,4 +13,48 @@ char **environ __attribute__((weak));
The old code seems to avoid putting "environ" into the global symbol namespace. Could this declaration be moved into the function like in getenv()?
tools/include/nolibc/crt.h line 16
Quoted 3 lines
tools/include/nolibc/crt.h +1
-0const unsigned long *_auxv __attribute__((weak)); +int main (int argc,char *argv[],char **envp);
This will lead to conflicting declarations if the users use a different signature. I'm not (yet?) sure how to work around this.
Also how is the case handled where main() returns "void"? I'm not sure how this is currently handled or if the compiler takes core of returning 0 in this case.
Quoted 47 lines
tools/include/nolibc/crt.h +43
-0+static void exit (int ); + +void _start_c (long *sp) +{ +int argc, i; +char **argv; +char **envp; + +/* +* sp : argc <-- argument count, required by main() +* argv: argv[0] <-- argument vector, required by main() +* argv[1] +* ... +* argv[argc-1] +* null +* envp: envp[0] <-- environment variables, required by main() and getenv() +* envp[1] +* ... +* null +* _auxv: auxv[0] <-- auxiliary vector, required by getauxval() +* auxv[1] +* ... +* null +*/ + +/* assign argc and argv */ + argc = sp[0 ]; + argv = (void *)(sp +1 ); + +/* find envp */ + envp = argv + argc +1 ; + environ = envp; + +/* find auxv */ + i =0 ; +while (envp[i]) + i++; + _auxv = (void *)(envp + i +1 ); + +/* go to application */ +exit (main (argc, argv, envp)); +} +#endif /* _NOLIBC_CRT_H */ Signature · 1 line
2.25.1
Re: [PATCH v2 04/12] tools/nolibc: crt.h: add _start_c
Quoted 18 lines · Thomas Weißschuh
On 2023-07-08 23:29:58+0800, Zhangjin Wu wrote:
[..]
Quoted 9 lines
--- tools/include/nolibc/crt.h | 44 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)tools/include/nolibc/crt.h +0
-0@@ -13,4 +13,48 @@[..]
Quoted 2 lines
const unsigned long
*_auxv__attribute__((weak));
Quoted 8 lines
Quoted 1 line
+int
main(int argc, char *argv[], char **envp);This will lead to conflicting declarations if the users use a different signature. I'm not (yet?) sure how to work around this.
Also how is the case handled where
main()returns "void"? I'm not sure how this is currently handled or if the compiler takes core of returning 0 in this case.
I looked into this some more.
The diff below allows it to accept different signatures for main(). (Maybe you can improve the naming)
Implicit return values seem to be handled by the compiler automatically. In C89 mode we get garbage values, in C2X/C23 we get zero. As per the respective C specs.
tools/include/nolibc/crt.h +2 -2
@@ -13,7 +13,6 @@ char **environ __attribute__((weak)); const unsigned long *_auxv __attribute__((weak)); -int main (int argc,char *argv[],char **envp);static void exit (int );void _start_c (long *sp)
@@ -21,6 +20,7 @@ void _start_c(long *sp)int argc, i;char **argv;char **envp; +int _nolibc_main_alias (int ,char **,char **) __asm__("main" );/* * sp : argc <-- argument count, required by main()
@@ -54,7 +54,7 @@ void _start_c(long *sp) _auxv = (void *)(envp + i +1 );/* go to application */ - exit(main(argc, argv, envp)); + exit(_nolibc_main_alias(argc, argv, envp)); }#endif /* _NOLIBC_CRT_H */
Re: [PATCH v2 01/12] tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
In reply to Willy TarreauHi, Willy
Quoted 23 lines · Zhangjin Wu
Quoted 14 lines
Currently, the architecture specific
arch.hhas two parts, one is the syscall declarations forsys.h, another is the_startcode definition for startup support.The coming
crt.hwill provide the startup support with a new common_start_c(), it will replace most of the assembly_startcode and shrink the original_startcode to be minimal, as a result,_start_c()and the left minimal_startcode will work together to provide the startup support, therefore, the left_startcode will be only required bycrt.h.So, the syscall declarations part of
arch.hcan be split tosys_arch.hand the_startcode part ofarch.hcan be split tocrt_arch.hand then, they should only be included insys.handcrt.hrespectively.Quoted 6 lines
At the same time, the architecture specific arch-<ARCH>.h should be split to <ARCH>/crt.h and <ARCH>/sys.h.
As a preparation, this creates the architecture specific directory and moves tools/include/nolibc/arch-<ARCH>.h to tools/include/nolibc/<ARCH>/arch.h.
Quoted 6 lines
I'm sorry but I still don't understand what it *provides*. I'm reading it as "we *can* do this so let's do it". But what is the specific purpose of adding this extra directory structure ? It's really unclear to me and worries me that it'll only result in complicating maintenance by adding even more files, thus even more "include" lines and cross dependencies.
Willy, I was assuming you had a look at the discussion between Thomas and me, so, I didn't add the link to our discussion, it is more about the 'clarity' of code "include" [1].
I have proposed the idea in the discussion but got no response yet, so, sent this revision for more discussion, obviously, it is better to discuss more there and get more feedback from Thomas and you.
The v0 included "crt.h" before "arch.h", Thomas suggested me include "crt.h" in arch_<ARCH>.h, just like the "compiler.h" did. His suggestion did inspire me to think about how to treat the relationship among crt.h, sys.h and arch.h.
The idea behind is we have many directions to divide nolibc to different parts/modules:
- one is arch specific (arch.h) and non-arch specific (the others)
This method is used by us currently, It is very good to put all of the arch specific parts together to simplify (in the files to be added/maintained) the porting of a new architecture.
But to be honest, It also confuse the modularity a little, for example, like sys.h, crt.h should be a core function/feature of nolibc, arch.h is not so. arch.h only provides the necessary minimal assembly "pieces".
both sys.h and crt.h are not a sub modules of arch.h (although they have minimal arch specific code), so, like sys.h, crt.h should be included in the top-level headers, not in arch.h, reversely, the minimal arch specific should be included in crt.h. To do so and to avoid include the non-crt part, the split of arch.h is required, and therefore, the <ARCH>/ is created to put the divided <ARCH>/sys.h and <ARCH>/crt.h, otherwise, there will be many sys-<ARCH>.h and crt-<ARCH>.h in the top-level directory of nolibc.
- another is the parallel functions/features (like crt.h, sys.h, stack protector ...)
This is used by musl and glibc, before sending this proposal, I have taken a look at both of them, musl is simpler and clearer, we apply the similar method:
musl:
crt/crt1.c
#include "crt_arch.h" /* arch/<ARCH>/crt_arch.h */ void _start_c(long *p)
{
int argc = p[0];
char **argv = (void *)(p+1);
...
}
src/internal/syscall.h:
##include "syscall_arch.h" /* arch/<ARCH>/syscall_arch.h */
...glibc: (it is more complicated than musl)
csu/libc-start.c, sysdeps/<ARCH>/start.S
sysdeps/unix/sysv/linux/sysdep.h, sysdeps/unix/sysv/linux/<ARCH>/sysdep.h,
With this method, the "crt_arch.h + crt.h" together provide the C RunTime (startup code, stack protector, environ and _auxv currently) function, the "sys_arch.h + sys.h" together provide the syscall definitions. The arch specific parts are hidden behind, and only require to include the crt_arch.h in crt.h and sys_arch.h in sys.h, no need to include the whole arch.h for all.
As a summary, the core obvious reason here is, to this crt.h itself, it is ok for us to include crt.h in arch.h in code side, but reversely, I do prefer to include arch.h (and therefore the crt_arch.h) in crt.h, crt.h is the core function should be exported, arch.h is not, it only provide some low-level helpers for crt.h. If we treat sys.h as a core function and crt.h as a arch specific thing, it does confuse a little. This reorg may also help the similar future functions who require arch specific support, but of course, it does require to add/maintain more files for a new architecture, but it also allow to develop/debug at a smaller fineness.
In current stage, include crt.h in arch.h is not that unacceptable, but if reorg is a better direction, why not do it currently, because we do have two functions (crt.h and sys.h) in <ARCH>/, if only one, it is not urgent ;-)
Is this explanation better than before? welcome to discuss more ;-)
Like musl, if required, another top-level arch/ may be required to put all of the <ARCH>/ directories together to clean up the top-level nolibc directory.
Quoted 7 lines
Zhangjin, very often in your series, the justification for a change is missing, instead it's only explaining what is being changed, and I must confess that it makes it particularly difficult to figure the benefits. I'm only seeing this as an opportunity for a change ("can be split"). I could have missed something of course, but I can't figure what problem it is trying to solve.
Willy, thanks very much for pointing out this and so sorry, "commit message should tell why but not how" is in my mind, but sometimes, it may be lost especially when the change list are 'huge' (must improve).
In reality, It is a little difficult to just explain it at the right dimension for this change, so I have wrotten several versions of the commit message for this change locally (and also for the other changes too), at last, I choose the one currently used.
As explained in another reply, it is really hard to write a just ok commit message for every change, sometimes, the justification is 'obvious' to some develoers who have the background information or who have dicussed together, sometimes, it is not easy to just write precisely about the core justification, to improve this, here is the list I plan to do, hope it help the others too:
- discuss more before send new patches, especially for 'new' or 'big' change
- reword carefully about every change before sending patches (show benefits to let maintainer 'buy' (merge) them)
- send the patches not frequently, keep the mind conscious, not like a "flood"
Quoted 5 lines
As a general advice, I tend to remind people that when sending a patch series, they should consider they're trying to sell it, so they must emphasize the benefits of accepting the series for the maintainer(s).
Yeah, why maintainer want to 'buy' (merge) is the right direction.
Quoted 7 lines
You very likely have a good reason for doing this but I can't see it here so I'm just seeing a change that will possibly add some extra cost (if at least because file locations change again) and nothing more. When you try to reorganize things, it's often much more efficient to try to discuss it before proposing patches, because reorg patches are generally unreadable and take time for you to create and for others to review.
Yes, reorg is really hard to do and also hard to review.
Quoted 5 lines
Instead, just explaining what you
think you can improve is faster for everyone, and others can chime in and propose alternate approaches (something which is very hard to do with a patch series).
Great suggestion, a ping on the discussion [1] would be better than just reorg it and send the patches. Thanks a lot.
Best regards,
Zhangjin
---
[1]: https://lore.kernel.org/lkml/[email protected]/
Quoted 2 lines
Thanks!
Willy
Re: [PATCH v2 04/12] tools/nolibc: crt.h: add _start_c
In reply to Thomas WeißschuhHi, Thomas
Quoted 16 lines · Zhangjin Wu
Quoted 9 lines
As the environ and
_auxvsupport added for nolibc, the assembly_startfunction becomes more and more complex and therefore makes the porting of nolibc to new architectures harder and harder.To simplify portability, this c version of
_start_c()is added to do most of the assembly start operations in C, which reduces the complexity a lot and will eventually simplify the porting of nolibc to the new architectures.Quoted 4 lines
The new
_start_c()only requires a stack pointer argument, it will find argv, envp and_auxvfor us, and then callmain(), finally, itexit()with main's return status. With this new_start_c(), the future new architectures only require to add very few assembly instructions.
Quoted 5 lines
I like it!
A quick test indicates that the initialization of the stackprotectors could also be moved into the C function.
Cool, do you mean directly call __stack_chk_init() at the beginning of _start_c()?
Quoted 2 lines
It also seems like a good opportunity to add some tests for argv/environment variable passing.
Yes, and even further, we can do more on auxv, just like musl does in src/env/__libc_start_main.c, not that urgent currently:
libc.auxv = auxv = (void *)(envp+i+1);
...
__hwcap = aux[AT_HWCAP];
if (aux[AT_SYSINFO]) __sysinfo = aux[AT_SYSINFO];
...
libc.page_size = aux[AT_PAGESZ];
if (!pn) pn = (void*)aux[AT_EXECFN];
if (!pn) pn = "";
__progname = __progname_full = pn;
for (i=0; pn[i]; i++) if (pn[i]=='/') __progname = pn+i+1;
__init_tls(aux);
__init_ssp((void *)aux[AT_RANDOM]);
if (aux[AT_UID]==aux[AT_EUID] && aux[AT_GID]==aux[AT_EGID]
&& !aux[AT_SECURE]) return;
...
libc.secure = 1;Quoted 4 lines
And as general note to the full series I think that splitting the arch files is not necessary and confusing.
Ok, welcome to discuss more in this thread:
https://lore.kernel.org/lkml/[email protected]/
and let's choose a better method as possible as we can, Just replied Willy to explain more.
Quoted 5 lines
Quoted 5 lines
Signed-off-by Zhangjin Wu <[email protected]>--- tools/include/nolibc/crt.h | 44 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)
Quoted 11 lines
tools/include/nolibc/crt.h line 13
Quoted 6 lines
tools/include/nolibc/crt.h +0
-0@@ -13,4 +13,48 @@ char **environ __attribute__((weak));The old code seems to avoid putting "environ" into the global symbol namespace. Could this declaration be moved into the function like in
getenv()?
ok, do you mean just move it to stdlib.h like this? I moved _auxv (used by getauxv()) to stdlib.h too:
tools/nolibc: move environ and _auxv from crt.h to stdlib.h
Move the definitions of environ and _auxv from crt.h to stdlib.h, where
the place who uses those definitions.
- getenv uses environ
- getauxv uses _auxcv
Signed-off-by: Zhangjin Wu <[email protected]>
diff --git a/tools/include/nolibc/crt.h b/tools/include/nolibc/crt.h
index b269294e9664..d2f84cbe73d0 100644
--- a/tools/include/nolibc/crt.h
+++ b/tools/include/nolibc/crt.h
@@ -10,14 +10,13 @@
#include "compiler.h"
#include "crt_arch.h"
-char **environ __attribute__((weak));
-const unsigned long *_auxv __attribute__((weak));
-
int main(int argc, char *argv[], char **envp);
static void exit(int);
void _start_c(long *sp)
{
+ extern char **environ;
+ extern const unsigned long *_auxv;
int argc, i;
char **argv;
char **envp;
diff --git a/tools/include/nolibc/stdlib.h b/tools/include/nolibc/stdlib.h
index 2f9b4b3c6d26..5eadadc2d0f5 100644
--- a/tools/include/nolibc/stdlib.h
+++ b/tools/include/nolibc/stdlib.h
@@ -14,6 +14,9 @@
#include "string.h"
#include <linux/auxvec.h>
+char **environ __attribute__((weak));
+const unsigned long *_auxv __attribute__((weak));
+
struct nolibc_heap {
size_t len;
char user_p[] __attribute__((__aligned__));Quoted 7 lines
tools/include/nolibc/crt.h line 16
Quoted 3 lines
tools/include/nolibc/crt.h +1
-0const unsigned long *_auxv __attribute__((weak)); +int main (int argc,char *argv[],char **envp);This will lead to conflicting declarations if the users use a different signature. I'm not (yet?) sure how to work around this.
Ah yes, I forgot this critical case, people may use something like:
int main(void)
int main(int argc, char *argv[])Just applied the method suggested by you in anothe reply [1]:
diff --git a/tools/include/nolibc/crt.h b/tools/include/nolibc/crt.h
index d2f84cbe73d0..8fe38ef8c5b2 100644
--- a/tools/include/nolibc/crt.h
+++ b/tools/include/nolibc/crt.h
@@ -10,7 +10,7 @@
#include "compiler.h"
#include "crt_arch.h"
-int main(int argc, char *argv[], char **envp);
+typedef int (_nolibc_main_fn)(int, char **, char **);
static void exit(int);
void _start_c(long *sp)
@@ -20,6 +20,7 @@ void _start_c(long *sp)
int argc, i;
char **argv;
char **envp;
+ _nolibc_main_fn _nolibc_main __asm__ ("main");
/*
* sp : argc <-- argument count, required by main()
@@ -53,7 +54,7 @@ void _start_c(long *sp)
_auxv = (void *)(envp + i + 1);
/* go to application */
- exit(main(argc, argv, envp));
+ exit(_nolibc_main(argc, argv, envp));
}
#endif /* _NOLIBC_CRT_H */It works as expected, thanks very much!
Quoted 1 line
Also how is the case handled where
main()returns "void"?
Do you mean the main() without an explicit "return 0"?
int main(int argc, char *argv[], char **envp)
{
printf("Hello\n");
}Tested c89, c99, c2x, c23, gnu11, the same result as yours:
std | return value
----------|------------
c89 | 6 /* return value of printf() */
c99 | 0 /* set by compiler with: xorl %eax, %eax */
c2x | 0
gnu11 | 0Quoted 2 lines
I'm not sure how this is currently handled or if the compiler takes care of returning 0 in this case.
void main(void)
{
printf("Hello\n");
}Tested c89, c99, c2x, c23, gnu11, the same result as yours, the return value is 6, the return value is the return status of the last called functon.
Thanks,
Zhangjin
---
[1]: https://lore.kernel.org/lkml/[email protected]/
Re: [PATCH v2 04/12] tools/nolibc: crt.h: add _start_c
Quoted 15 lines · Zhangjin Wu
Quoted 15 lines · Zhangjin Wu
Quoted 13 lines
As the environ and
_auxvsupport added for nolibc, the assembly_startfunction becomes more and more complex and therefore makes the porting of nolibc to new architectures harder and harder.To simplify portability, this c version of
_start_c()is added to do most of the assembly start operations in C, which reduces the complexity a lot and will eventually simplify the porting of nolibc to the new architectures.The new
_start_c()only requires a stack pointer argument, it will find argv, envp and_auxvfor us, and then callmain(), finally, itexit()with main's return status. With this new_start_c(), the future new architectures only require to add very few assembly instructions.
Quoted 8 lines
Quoted 5 lines
I like it!
A quick test indicates that the initialization of the stackprotectors could also be moved into the C function.
Cool, do you mean directly call
__stack_chk_init()at the beginning of_start_c()?
Yes, exactly.
Quoted 5 lines
Quoted 2 lines
It also seems like a good opportunity to add some tests for argv/environment variable passing.
Yes, and even further, we can do more on auxv, just like musl does in
src/env/__libc_start_main.c, not that urgent currently:
With tests I mean nolibc-test.c to make sure we don't introduce any regressions.
Only some tiny testcases to validate that argv and environ are picked up correctly by the startup code on all arches.
Quoted 27 lines
libc.auxv = auxv = (void *)(envp+i+1); ... __hwcap = aux[AT_HWCAP]; if (aux[AT_SYSINFO]) __sysinfo = aux[AT_SYSINFO]; ... libc.page_size = aux[AT_PAGESZ]; if (!pn) pn = (void*)aux[AT_EXECFN]; if (!pn) pn = ""; __progname = __progname_full = pn; for (i=0; pn[i]; i++) if (pn[i]=='/') __progname = pn+i+1; __init_tls(aux); __init_ssp((void *)aux[AT_RANDOM]); if (aux[AT_UID]==aux[AT_EUID] && aux[AT_GID]==aux[AT_EGID] && !aux[AT_SECURE]) return; ... libc.secure = 1;Quoted 4 lines
And as general note to the full series I think that splitting the arch files is not necessary and confusing.
Quoted 6 lines
Ok, welcome to discuss more in this thread:
https://lore.kernel.org/lkml/[email protected]/and let's choose a better method as possible as we can, Just replied Willy to explain more.
Will do.
Quoted 17 lines
Quoted 5 lines
Quoted 5 lines
Signed-off-by Zhangjin Wu <[email protected]>--- tools/include/nolibc/crt.h | 44 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)Quoted 11 lines
tools/include/nolibc/crt.h line 13
Quoted 6 lines
tools/include/nolibc/crt.h +0
-0@@ -13,4 +13,48 @@ char **environ __attribute__((weak));The old code seems to avoid putting "environ" into the global symbol namespace. Could this declaration be moved into the function like in
getenv()?
Quoted 2 lines
ok, do you mean just move it to
stdlib.hlike this? I moved_auxv(used bygetauxv()) tostdlib.htoo:
Nevermind, I got confused by the in-function declaration of "extern char **environ" inside "getenv()".
Actually this in-function declaration doesn't do anything and can be dropped.
Quoted 55 lines
tools/nolibc: move environ and _auxv from crt.h to stdlib.h Move the definitions of environ and _auxv from crt.h to stdlib.h, where the place who uses those definitions. - getenv uses environ - getauxv uses _auxcv Signed-off-by: Zhangjin Wu <[email protected]> diff --git a/tools/include/nolibc/crt.h b/tools/include/nolibc/crt.h index b269294e9664..d2f84cbe73d0 100644 --- a/tools/include/nolibc/crt.h +++ b/tools/include/nolibc/crt.h @@ -10,14 +10,13 @@ #include "compiler.h" #include "crt_arch.h" -char **environ __attribute__((weak)); -const unsigned long *_auxv __attribute__((weak)); - int main(int argc, char *argv[], char **envp); static void exit(int); void _start_c(long *sp) { + extern char **environ; + extern const unsigned long *_auxv; int argc, i; char **argv; char **envp; diff --git a/tools/include/nolibc/stdlib.h b/tools/include/nolibc/stdlib.h index 2f9b4b3c6d26..5eadadc2d0f5 100644 --- a/tools/include/nolibc/stdlib.h +++ b/tools/include/nolibc/stdlib.h @@ -14,6 +14,9 @@ #include "string.h" #include <linux/auxvec.h> +char **environ __attribute__((weak)); +const unsigned long *_auxv __attribute__((weak)); + struct nolibc_heap { size_t len; char user_p[] __attribute__((__aligned__));Quoted 7 lines
tools/include/nolibc/crt.h line 16
Quoted 3 lines
tools/include/nolibc/crt.h +1
-0const unsigned long *_auxv __attribute__((weak)); +int main (int argc,char *argv[],char **envp);This will lead to conflicting declarations if the users use a different signature. I'm not (yet?) sure how to work around this.
Quoted 6 lines
Ah yes, I forgot this critical case, people may use something like:
int main(void) int main(int argc, char *argv[])[..]
I thought about this general problem and it turns out that there is nothing that any libc can do to distinguish these special cases. So it has to be handled in the compiler and we do not have to care.
Thomas
Re: [PATCH v2 01/12] tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
In reply to Zhangjin WuQuoted 15 lines · Zhangjin Wu
Quoted 1 line · Zhangjin Wu
[..]
Quoted 10 lines
Quoted 3 lines
As a preparation, this creates the architecture specific directory and moves tools/include/nolibc/arch-<ARCH>.h to tools/include/nolibc/<ARCH>/arch.h.
I'm sorry but I still don't understand what it *provides*. I'm reading it as "we *can* do this so let's do it". But what is the specific purpose of adding this extra directory structure ? It's really unclear to me and worries me that it'll only result in complicating maintenance by adding even more files, thus even more "include" lines and cross dependencies.
Quoted 7 lines
Willy, I was assuming you had a look at the discussion between Thomas and me, so, I didn't add the link to our discussion, it is more about the 'clarity' of code "include" [1].
I have proposed the idea in the discussion but got no response yet, so, sent this revision for more discussion, obviously, it is better to discuss more there and get more feedback from Thomas and you.
To be honest I got overwhelmed at some point and instead of figuring out to which series' I already responded and which not I only responded to those where I had time to do so immediately.
Keeping the amount of in-flight serieses lower would help this.
Quoted 30 lines
The v0 included "
crt.h" before "arch.h", Thomas suggested me include "crt.h" in arch_<ARCH>.h, just like the "compiler.h" did. His suggestion did inspire me to think about how to treat the relationship amongcrt.h,sys.handarch.h.The idea behind is we have many directions to divide nolibc to different parts/modules:
- one is arch specific (
arch.h) and non-arch specific (the others)This method is used by us currently, It is very good to put all of the arch specific parts together to simplify (in the files to be added/maintained) the porting of a new architecture.
But to be honest, It also confuse the modularity a little, for example, like
sys.h,crt.hshould be a core function/feature of nolibc,arch.his not so.arch.honly provides the necessary minimal assembly "pieces".both
sys.handcrt.hare not a sub modules ofarch.h(although they have minimal arch specific code), so, likesys.h,crt.hshould be included in the top-level headers, not inarch.h, reversely, the minimal arch specific should be included incrt.h. To do so and to avoid include the non-crt part, the split ofarch.his required, and therefore, the <ARCH>/ is created to put the divided <ARCH>/sys.h and <ARCH>/crt.h, otherwise, there will be many sys-<ARCH>.h and crt-<ARCH>.h in the top-level directory of nolibc.- another is the parallel functions/features (like
crt.h,sys.h, stack protector ...)
Quoted 7 lines
This is used by musl and glibc, before sending this proposal, I have taken a look at both of them, musl is simpler and clearer, we apply the similar method:
musl: crt/crt1.c #include "crt_arch.h" /* arch/<ARCH>/crt_arch.h */
In musl crt_arch.h seems to be used in different ways. So it makes sense to split it from syscall_arch.h. In nolibc there is no such distinction. And everything will end up in a global namespace anyways.
Quoted 37 lines
void _start_c(long *p) { int argc = p[0]; char **argv = (void *)(p+1); ... } src/internal/syscall.h: ##include "syscall_arch.h" /* arch/<ARCH>/syscall_arch.h */ ...glibc: (it is more complicated than musl)
csu/libc-start.c, sysdeps/<ARCH>/start.S
sysdeps/unix/sysv/linux/sysdep.h, sysdeps/unix/sysv/linux/<ARCH>/sysdep.h,With this method, the "
crt_arch.h+crt.h" together provide the C RunTime (startup code, stack protector, environ and_auxvcurrently) function, the "sys_arch.h+sys.h" together provide the syscall definitions. The arch specific parts are hidden behind, and only require to include thecrt_arch.hincrt.handsys_arch.hinsys.h, no need to include the wholearch.hfor all.As a summary, the core obvious reason here is, to this
crt.hitself, it is ok for us to includecrt.hinarch.hin code side, but reversely, I do prefer to includearch.h(and therefore thecrt_arch.h) incrt.h,crt.his the core function should be exported,arch.his not, it only provide some low-level helpers forcrt.h. If we treatsys.has a core function andcrt.has a arch specific thing, it does confuse a little. This reorg may also help the similar future functions who require arch specific support, but of course, it does require to add/maintain more files for a new architecture, but it also allow to develop/debug at a smaller fineness.
Quoted 1 line
In current stage, include
crt.hinarch.his not that unacceptable, but
Why would it be more unacceptable in the future?
Quoted 5 lines
if reorg is a better direction, why not do it currently, because we do have two functions (
crt.handsys.h) in <ARCH>/, if only one, it is not urgent ;-)Is this explanation better than before? welcome to discuss more ;-)
Personally I'm not convinced :-)
The arch-specific code in nolibc in mainline is currentl ~200 lines per arch. With this series in general it will be even less.
If at some point there are many more architectures it may make sense to introduce an arch/ directory then.
Quoted 3 lines
Like musl, if required, another top-level
arch/may be required to put all of the <ARCH>/ directories together to clean up the top-level nolibc directory.
At the moment in mainline there are 26 files in nolibc. That does not seem excessive, in fact it looks to be less than most other kernel directories.
Quoted 23 lines
Quoted 6 lines
Zhangjin, very often in your series, the justification for a change is missing, instead it's only explaining what is being changed, and I must confess that it makes it particularly difficult to figure the benefits. I'm only seeing this as an opportunity for a change ("can be split"). I could have missed something of course, but I can't figure what problem it is trying to solve.
Willy, thanks very much for pointing out this and so sorry, "commit message should tell why but not how" is in my mind, but sometimes, it may be lost especially when the change list are 'huge' (must improve).
In reality, It is a little difficult to just explain it at the right dimension for this change, so I have wrotten several versions of the commit message for this change locally (and also for the other changes too), at last, I choose the one currently used.
As explained in another reply, it is really hard to write a just ok commit message for every change, sometimes, the justification is 'obvious' to some develoers who have the background information or who have dicussed together, sometimes, it is not easy to just write precisely about the core justification, to improve this, here is the list I plan to do, hope it help the others too:
Quoted 2 lines
- discuss more before send new patches, especially for 'new' or 'big' change
This sounds like the correct thing to do for reorganization patches. It should be very easy to describe and very annoying to actually do.
Quoted 7 lines
- reword carefully about every change before sending patches (show benefits to let maintainer 'buy' (merge) them)
- send the patches not frequently, keep the mind conscious, not like a "flood"
[..]
Re: [PATCH v2 04/12] tools/nolibc: crt.h: add _start_c
In reply to Thomas WeißschuhHi, Thomas
Quoted 2 lines · Zhangjin Wu
Quoted 1 line
Quoted 1 line · Zhangjin Wu
[...]
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
It also seems like a good opportunity to add some tests for argv/environment variable passing.
Quoted 8 lines
Quoted 2 lines
Yes, and even further, we can do more on auxv, just like musl does in
src/env/__libc_start_main.c, not that urgent currently:With tests I mean
nolibc-test.cto make sure we don't introduce any regressions.
Only some tiny testcases to validate that argv and environ are picked up correctly by the startup code on all arches.
Thomas, seems we already have some testcases for argv, environ and auxv currently:
run_syscall: chmod_argv0 <-- argv[0]
chdir_root <-- chdir(getenv("PWD")) getpagesize <-- getauxval(AT_PAGESZ)
run_stdlib : getenv_TERM <-- getenv
Quoted 1 line
Quoted 1 line
[...]
Quoted 28 lines
Quoted 8 lines
Quoted 1 line
Ok, welcome to discuss more in this thread:
https://lore.kernel.org/lkml/[email protected]/and let's choose a better method as possible as we can, Just replied Willy to explain more.
Will do.
Quoted 17 lines
Quoted 16 lines
tools/include/nolibc/crt.h line 13
Quoted 11 lines
Signed-off-by Zhangjin Wu <[email protected]>--- tools/include/nolibc/crt.h | 44 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)tools/include/nolibc/crt.h +0
-0@@ -13,4 +13,48 @@ char **environ __attribute__((weak));The old code seems to avoid putting "environ" into the global symbol namespace. Could this declaration be moved into the function like in
getenv()?
Quoted 8 lines
Quoted 2 lines
ok, do you mean just move it to
stdlib.hlike this? I moved_auxv(used bygetauxv()) tostdlib.htoo:Nevermind, I got confused by the in-function declaration of "extern char **environ" inside "
getenv()".
Actually this in-function declaration doesn't do anything and can be dropped.
Yes. for nolibc application is in one-file style, let's remove it.
Quoted 1 line
Quoted 1 line
[...]
Quoted 7 lines
Quoted 7 lines
Quoted 4 lines
This will lead to conflicting declarations if the users use a different signature. I'm not (yet?) sure how to work around this.
Ah yes, I forgot this critical case, people may use something like:
Quoted 8 lines
Quoted 4 lines
int main(void) int main(int argc, char *argv[])[..]
I thought about this general problem and it turns out that there is nothing that any libc can do to distinguish these special cases. So it has to be handled in the compiler and we do not have to care.
Ok.
Thanks,
Zhangjin
Quoted 2 lines
Thomas
Willy Tarreau
maintainerRe: [PATCH v2 01/12] tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
In reply to Thomas WeißschuhHi Thomas, Zhangjin,
Quoted 24 lines · Thomas Weißschuh
Quoted 15 lines · Zhangjin Wu
Quoted 1 line · Zhangjin Wu
[..]
Quoted 10 lines
Quoted 3 lines
As a preparation, this creates the architecture specific directory and moves tools/include/nolibc/arch-<ARCH>.h to tools/include/nolibc/<ARCH>/arch.h.
I'm sorry but I still don't understand what it *provides*. I'm reading it as "we *can* do this so let's do it". But what is the specific purpose of adding this extra directory structure ? It's really unclear to me and worries me that it'll only result in complicating maintenance by adding even more files, thus even more "include" lines and cross dependencies.
Quoted 7 lines
Willy, I was assuming you had a look at the discussion between Thomas and me, so, I didn't add the link to our discussion, it is more about the 'clarity' of code "include" [1].
I have proposed the idea in the discussion but got no response yet, so, sent this revision for more discussion, obviously, it is better to discuss more there and get more feedback from Thomas and you.
Quoted 5 lines
To be honest I got overwhelmed at some point and instead of figuring out to which series' I already responded and which not I only responded to those where I had time to do so immediately.
Keeping the amount of in-flight serieses lower would help this.
+1 on this. First it's difficult for me to assign contiguous time on the subject so I can't grasp all series at once, and I'm terribly bad at context-switching, which takes even more time and induces confusion. Less topics at once, more focused with less reviews of reorganizations will definitely help.
Quoted 7 lines
Quoted 7 lines
The v0 included "
crt.h" before "arch.h", Thomas suggested me include "crt.h" in arch_<ARCH>.h, just like the "compiler.h" did. His suggestion did inspire me to think about how to treat the relationship amongcrt.h,sys.handarch.h.The idea behind is we have many directions to divide nolibc to different parts/modules:
Again above I'm seeing an opportunity but no explanation of why this is needed. Thomas already mentioned that you're speaking about just trying to factor out a few tens of lines. I'm not seeing *why* we need to re-split everything yet again.
Quoted 10 lines
Quoted 10 lines
- one is arch specific (
arch.h) and non-arch specific (the others)This method is used by us currently, It is very good to put all of the arch specific parts together to simplify (in the files to be added/maintained) the porting of a new architecture.
But to be honest, It also confuse the modularity a little, for example, like
sys.h,crt.hshould be a core function/feature of nolibc,arch.his not so.arch.honly provides the necessary minimal assembly "pieces".
But that's precisely the principle: keep arch-specific stuff as minimal as possible, keep most of the rest generic but easily overloadable if needed as we know that archs are not all 1:1 equivalent.
Quoted 3 lines
Quoted 3 lines
both
sys.handcrt.hare not a sub modules ofarch.h(although they have minimal arch specific code), so, likesys.h,crt.hshould be included in the top-level headers, not inarch.h,
Why ? Keep in mind that these are only include files, to in the end, *all* of them are included. The ordering is the only thing that really matters.
Quoted 6 lines
Quoted 6 lines
reversely, the
minimal arch specific should be included incrt.h. To do so and to avoid include the non-crt part, the split ofarch.his required, and therefore, the <ARCH>/ is created to put the divided <ARCH>/sys.h and <ARCH>/crt.h, otherwise, there will be many sys-<ARCH>.h and crt-<ARCH>.h in the top-level directory of nolibc.
Then doesn't it prove that you don't need that crt-<ARCH>.h and that instead it should just be in arch-<ARCH> like the rest of the same arch ?
Quoted 6 lines
Quoted 6 lines
- another is the parallel functions/features (like
crt.h,sys.h, stack protector ...)This is used by musl and glibc, before sending this proposal, I have taken a look at both of them, musl is simpler and clearer, we apply the similar method:
Quoted 7 lines
Quoted 3 lines
musl: crt/crt1.c #include "crt_arch.h" /* arch/<ARCH>/crt_arch.h */In musl
crt_arch.hseems to be used in different ways. So it makes sense to split it fromsyscall_arch.h. In nolibc there is no such distinction. And everything will end up in a global namespace anyways.
Exactly. Musl is musl and nolibc is nolibc. Musl is a regular libc in that it provides a .so that is built from many .c files. As such it's desirable to split along certain edges. nolibc contains no single C file. It's only meant to be included as-is in the user's C file. This changes a lot of things, even in terms of splitting. Also keep in mind that musl is a general-purpose libc, and that some distros are entirely built on it. nolibc doesn't have such goal nor expectation, the first user was a preinit code I wrote long ago, and the second one is rcutorture which contains a while() loop around gettimeofday() IIRC. We must not just blindly imitate other components' choices because they work, when we're dealing with different constraints. If ours are acceptable, no need to complicate everything.
Quoted 6 lines
Quoted 6 lines
With this method, the "
crt_arch.h+crt.h" together provide the C RunTime (startup code, stack protector, environ and_auxvcurrently) function, the "sys_arch.h+sys.h" together provide the syscall definitions. The arch specific parts are hidden behind, and only require to include thecrt_arch.hincrt.handsys_arch.hinsys.h, no need to include the wholearch.hfor all.
Everything is included all the time. *everything*. The more files we create, the more "#ifdef FOO_H" gets evaluated, and the more maintenance burden it adds.
Quoted 15 lines
Quoted 11 lines
As a summary, the core obvious reason here is, to this
crt.hitself, it is ok for us to includecrt.hinarch.hin code side, but reversely, I do prefer to includearch.h(and therefore thecrt_arch.h) incrt.h,crt.his the core function should be exported,arch.his not, it only provide some low-level helpers forcrt.h. If we treatsys.has a core function andcrt.has a arch specific thing, it does confuse a little. This reorg may also help the similar future functions who require arch specific support, but of course, it does require to add/maintain more files for a new architecture, but it also allow to develop/debug at a smaller fineness.Quoted 1 line
In current stage, include
crt.hinarch.his not that unacceptable, butWhy would it be more unacceptable in the future?
Quoted 7 lines
Quoted 5 lines
if reorg is a better direction, why not do it currently, because we do have two functions (
crt.handsys.h) in <ARCH>/, if only one, it is not urgent ;-)Is this explanation better than before? welcome to discuss more ;-)
Personally I'm not convinced :-)
Neither am I.
Quoted 4 lines
The arch-specific code in nolibc in mainline is currentl ~200 lines per arch. With this series in general it will be even less.
If at some point there are many more architectures it may make sense to introduce anarch/directory then.
And even then I'm not convinced because the number of archs will remain low anyway.
Quoted 7 lines
Quoted 3 lines
Like musl, if required, another top-level
arch/may be required to put all of the <ARCH>/ directories together to clean up the top-level nolibc directory.At the moment in mainline there are 26 files in nolibc. That does not seem excessive, in fact it looks to be less than most other kernel directories.
Indeed :-) Note that it started with a single one!
Quoted 4 lines
Quoted 4 lines
As explained in another reply, it is really hard to write a just ok commit message for every change, sometimes, the justification is 'obvious' to some develoers who have the background information or who have dicussed together, sometimes,
Sometimes yes, but most of the series come with propositions to improve something. The commits making the major changes (and the cover letter) should justify why this is a desirable change, what it implies not to have it and what it may imply to have it, what possible alternatives were considered and dropped sometimes (e.g. when hesitating between two approaches), etc.
Thanks,
Willy
Re: [PATCH v2 01/12] tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
Hi, Willy
Quoted 31 lines
Hi Thomas, Zhangjin,
Quoted 28 lines · Thomas Wei�schuh
Quoted 15 lines · Zhangjin Wu
Quoted 1 line · Zhangjin Wu
[..]
Quoted 10 lines
Quoted 3 lines
As a preparation, this creates the architecture specific directory and moves tools/include/nolibc/arch-<ARCH>.h to tools/include/nolibc/<ARCH>/arch.h.
I'm sorry but I still don't understand what it *provides*. I'm reading it as "we *can* do this so let's do it". But what is the specific purpose of adding this extra directory structure ? It's really unclear to me and worries me that it'll only result in complicating maintenance by adding even more files, thus even more "include" lines and cross dependencies.
Quoted 7 lines
Willy, I was assuming you had a look at the discussion between Thomas and me, so, I didn't add the link to our discussion, it is more about the 'clarity' of code "include" [1].
I have proposed the idea in the discussion but got no response yet, so, sent this revision for more discussion, obviously, it is better to discuss more there and get more feedback from Thomas and you.
To be honest I got overwhelmed at some point and instead of figuring out to which series' I already responded and which not I only responded to those where I had time to do so immediately.
Quoted 7 lines
Quoted 1 line
Keeping the amount of in-flight serieses lower would help this.
+1 on this. First it's difficult for me to assign contiguous time on the subject so I can't grasp all series at once, and I'm terribly bad at context-switching, which takes even more time and induces confusion. Less topics at once, more focused with less reviews of reorganizations will definitely help.
me too, let's go back to the original rv32 support asap ;-)
Quoted 20 lines
Quoted 8 lines
Quoted 8 lines
The v0 included "
crt.h" before "arch.h", Thomas suggested me include "crt.h" in arch_<ARCH>.h, just like the "compiler.h" did. His suggestion did inspire me to think about how to treat the relationship amongcrt.h,sys.handarch.h.The idea behind is we have many directions to divide nolibc to different parts/modules:
Again above I'm seeing an opportunity but no explanation of why this is needed. Thomas already mentioned that you're speaking about just trying to factor out a few tens of lines. I'm not seeing *why* we need to re-split everything yet again.
Quoted 6 lines
Quoted 6 lines
- one is arch specific (
arch.h) and non-arch specific (the others)This method is used by us currently, It is very good to put all of the arch specific parts together to simplify (in the files to be added/maintained) the porting of a new architecture.
Quoted 9 lines
Quoted 4 lines
Quoted 4 lines
But to be honest, It also confuse the modularity a little, for example, like
sys.h,crt.hshould be a core function/feature of nolibc,arch.his not so.arch.honly provides the necessary minimal assembly "pieces".But that's precisely the principle: keep arch-specific stuff as minimal as possible, keep most of the rest generic but easily overloadable if needed as we know that archs are not all 1:1 equivalent.
This patchset applies the same principle but mixes some other requirements ;-)
Agree with "archs are not all 1:1 equivalent".
Quoted 8 lines
Quoted 3 lines
Quoted 3 lines
both
sys.handcrt.hare not a sub modules ofarch.h(although they have minimal arch specific code), so, likesys.h,crt.hshould be included in the top-level headers, not inarch.h,Why ? Keep in mind that these are only include files, to in the end, *all* of them are included. The ordering is the only thing that really matters.
Yeah, I know this. The only thing confused me is the relationship among crt.h, sys.h and arch.h, the include position should better reflects their relationship, currently, we have mixed two "divide" methods together, one is arch and non-arch, another is the parallel functions/features.
so, the only left question is where should we include crt.h in? Firstly, I put it like sys.h (In my mind, it should be), If you two both agree to put it in arch-<ARCH.h>, I will renew this series with it, this is definitely a lighter way than the reorg method, I don't persist ;-)
Quoted 24 lines
Quoted 6 lines
Quoted 6 lines
reversely, the
minimal arch specific should be included incrt.h. To do so and to avoid include the non-crt part, the split ofarch.his required, and therefore, the <ARCH>/ is created to put the divided <ARCH>/sys.h and <ARCH>/crt.h, otherwise, there will be many sys-<ARCH>.h and crt-<ARCH>.h in the top-level directory of nolibc.Then doesn't it prove that you don't need that crt-<ARCH>.h and that instead it should just be in arch-<ARCH> like the rest of the same arch ?
Quoted 6 lines
Quoted 6 lines
- another is the parallel functions/features (like
crt.h,sys.h, stack protector ...)This is used by musl and glibc, before sending this proposal, I have taken a look at both of them, musl is simpler and clearer, we apply the similar method:
Quoted 7 lines
Quoted 3 lines
musl: crt/crt1.c #include "crt_arch.h" /* arch/<ARCH>/crt_arch.h */In musl
crt_arch.hseems to be used in different ways. So it makes sense to split it fromsyscall_arch.h. In nolibc there is no such distinction. And everything will end up in a global namespace anyways.
Quoted 13 lines
Exactly. Musl is musl and nolibc is nolibc. Musl is a regular libc in that it provides a .so that is built from many .c files. As such it's desirable to split along certain edges. nolibc contains no single C file. It's only meant to be included as-is in the user's C file. This changes a lot of things, even in terms of splitting. Also keep in mind that musl is a general-purpose libc, and that some distros are entirely built on it. nolibc doesn't have such goal nor expectation, the first user was a preinit code I wrote long ago, and the second one is rcutorture which contains a
while()loop aroundgettimeofday()IIRC. We must not just blindly imitate other components' choices because they work, when we're dealing with different constraints. If ours are acceptable, no need to complicate everything.
Willy, I know the difference between musl and nolibc, the musl code referenced here is only used to clear my confusion about "the relationship among crt.h, sys.h and arch.h" I mentiond above.
BTW, I do think nolibc have more using scenes, I like it very much and have used it in my own "Linux Lab" open source project [1] to let it work as the minimal rootfs to speed up kernel features learning and development, I do like the 'kernel-only deployments' feature behind nolibc [2], although there is something like "Unikernel Linux" [3], but that differs from a normal Linux system and is more complicated ;-)
I'm even imaging using it with a pure-header shell and a pure-header gui to let them further work together as a tiny rtos ;-)
[1]: https://github.com/tinyclub/linux-lab
[2]: https://lwn.net/Articles/920158/
[3]: https://github.com/unikernelLinux/ukl
Quoted 11 lines
Quoted 6 lines
Quoted 6 lines
With this method, the "
crt_arch.h+crt.h" together provide the C RunTime (startup code, stack protector, environ and_auxvcurrently) function, the "sys_arch.h+sys.h" together provide the syscall definitions. The arch specific parts are hidden behind, and only require to include thecrt_arch.hincrt.handsys_arch.hinsys.h, no need to include the wholearch.hfor all.Everything is included all the time. *everything*. The more files we create, the more "#ifdef
FOO_H" gets evaluated, and the more maintenance burden it adds.
Agree with less "#ifdef"s.
Quoted 25 lines
Quoted 15 lines
Quoted 11 lines
As a summary, the core obvious reason here is, to this
crt.hitself, it is ok for us to includecrt.hinarch.hin code side, but reversely, I do prefer to includearch.h(and therefore thecrt_arch.h) incrt.h,crt.his the core function should be exported,arch.his not, it only provide some low-level helpers forcrt.h. If we treatsys.has a core function andcrt.has a arch specific thing, it does confuse a little. This reorg may also help the similar future functions who require arch specific support, but of course, it does require to add/maintain more files for a new architecture, but it also allow to develop/debug at a smaller fineness.Quoted 1 line
In current stage, include
crt.hinarch.his not that unacceptable, butWhy would it be more unacceptable in the future?
Quoted 7 lines
Quoted 5 lines
if reorg is a better direction, why not do it currently, because we do have two functions (
crt.handsys.h) in <ARCH>/, if only one, it is not urgent ;-)Is this explanation better than before? welcome to discuss more ;-)
Personally I'm not convinced :-)
Neither am I.
Quoted 8 lines
Quoted 4 lines
The arch-specific code in nolibc in mainline is currentl ~200 lines per arch. With this series in general it will be even less.
If at some point there are many more architectures it may make sense to introduce anarch/directory then.And even then I'm not convinced because the number of archs will remain low anyway.
We have 8 now, the maximum may be 'ls -d arch/*/ | wc -l', it is 22 currently ;-)
Quoted 9 lines
Quoted 7 lines
Quoted 3 lines
Like musl, if required, another top-level
arch/may be required to put all of the <ARCH>/ directories together to clean up the top-level nolibc directory.At the moment in mainline there are 26 files in nolibc. That does not seem excessive, in fact it looks to be less than most other kernel directories.
Indeed :-) Note that it started with a single one!
Yeah, I learned the history, but I do think we will have more, as the requirements become more and more ;-)
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
As explained in another reply, it is really hard to write a just ok commit message for every change, sometimes, the justification is 'obvious' to some develoers who have the background information or who have dicussed together, sometimes,
Quoted 6 lines
Sometimes yes, but most of the series come with propositions to improve something. The commits making the major changes (and the cover letter) should justify why this is a desirable change, what it implies not to have it and what it may imply to have it, what possible alternatives were considered and dropped sometimes (e.g. when hesitating between two approaches), etc.
Ok.
Thanks,
Zhangjin
Quoted 3 lines
Thanks,
Willy
Willy Tarreau
maintainerRe: [PATCH v2 01/12] tools/nolibc: rename arch-<ARCH>.h to <ARCH>/arch.h
Hi Zhangjin,
Quoted 15 lines · Zhangjin Wu
Quoted 8 lines
Quoted 3 lines
Quoted 3 lines
both
sys.handcrt.hare not a sub modules ofarch.h(although they have minimal arch specific code), so, likesys.h,crt.hshould be included in the top-level headers, not inarch.h,Why ? Keep in mind that these are only include files, to in the end, *all* of them are included. The ordering is the only thing that really matters.
Yeah, I know this. The only thing confused me is the relationship among
crt.h,sys.handarch.h, the include position should better reflects their relationship, currently, we have mixed two "divide" methods together, one is arch and non-arch, another is the parallel functions/features.
Quoted 4 lines
so, the only left question is where should we include
crt.hin? Firstly, I put it likesys.h(In my mind, it should be), If you two both agree to put it in arch-<ARCH.h>, I will renew this series with it, this is definitely a lighter way than the reorg method, I don't persist ;-)
Then let's start this way. And if later we find a better organization *and* a good reason to change it, we can do it. Please also keep in mind that constantly renaming/moving inflicts extra pain to the stable team when backporting fixes if any. That's another reason for not moving code around often.
Quoted 18 lines
Quoted 4 lines
Quoted 3 lines
In musl
crt_arch.hseems to be used in different ways. So it makes sense to split it fromsyscall_arch.h. In nolibc there is no such distinction. And everything will end up in a global namespace anyways.Quoted 13 lines
Exactly. Musl is musl and nolibc is nolibc. Musl is a regular libc in that it provides a .so that is built from many .c files. As such it's desirable to split along certain edges. nolibc contains no single C file. It's only meant to be included as-is in the user's C file. This changes a lot of things, even in terms of splitting. Also keep in mind that musl is a general-purpose libc, and that some distros are entirely built on it. nolibc doesn't have such goal nor expectation, the first user was a preinit code I wrote long ago, and the second one is rcutorture which contains a
while()loop aroundgettimeofday()IIRC. We must not just blindly imitate other components' choices because they work, when we're dealing with different constraints. If ours are acceptable, no need to complicate everything.
Quoted 3 lines
Willy, I know the difference between musl and nolibc, the musl code referenced here is only used to clear my confusion about "the relationship among
crt.h,sys.handarch.h" I mentiond above.
OK, but even so, our includes are "private" and do not necessarily need to be organized like other projects. That doesn't mean that good ideas from other ones should not help us decide, of course, I just mean that it's normal to see differences.
Quoted 9 lines
BTW, I do think nolibc have more using scenes, I like it very much and have used it in my own "Linux Lab" open source project [1] to let it work as the minimal rootfs to speed up kernel features learning and development, I do like the 'kernel-only deployments' feature behind nolibc [2], although there is something like "Unikernel Linux" [3], but that differs from a normal Linux system and is more complicated ;-)
I'm even imaging using it with a pure-header shell and a pure-header gui to let them further work together as a tiny rtos ;-)
I'm not denying the possibilities for other use cases. I still have somewhere a small 1MB kernel image+rootfs that is sufficient to SSH into remotely or download and flash a firmware image over tftp/http/serial, so I see pretty well how this can be useful, and suspect this could eventually happen again. But doing incremental changes that don't seem to go in a particular direction is not much helpful and complicates participation from everyone.
However of course, if someone comes and say "I'd like to use it in this or that environment but for this I need all that", it's easier to follow and try to steer in a direction that adapts smoothly to all users' needs. Another point, please also remember that we moved it into the kernel to help those who need to adjust their tests easily contribute the missing calls (like you did for a number of things). But I tend to think that as long as it's in the kernel, the activity should remain related to the kernel usage. If you have bigger plans for a much larger project, the kernel's process will slow you down and you'll experience some rejection for breaking certain principles, so in this case such work would better be done outside (after re-updating the original repo that I left rot for a while).
Quoted 11 lines
Quoted 8 lines
Quoted 4 lines
The arch-specific code in nolibc in mainline is currentl ~200 lines per arch. With this series in general it will be even less.
If at some point there are many more architectures it may make sense to introduce anarch/directory then.And even then I'm not convinced because the number of archs will remain low anyway.
We have 8 now, the maximum may be 'ls -d
arch/*/| wc -l', it is 22 currently ;-)
Yes but I'm not convinced that all of them will be ported there, for various reasons ranging from lack of interest or lack of use cases, to lack of maintainer time.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Like musl, if required, another top-level
arch/may be required to put all of the <ARCH>/ directories together to clean up the top-level nolibc directory.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
At the moment in mainline there are 26 files in nolibc. That does not seem excessive, in fact it looks to be less than most other kernel directories.
Indeed :-) Note that it started with a single one!
Yeah, I learned the history, but I do think we will have more, as the requirements become more and more ;-)
Sure, but it grew fast thanks to being easy to understand and locate stuff. This should be seen as a feature I guess.
Cheers,
Willy