September 2023 - devel

[PATCH 1/4] fstatat64: define it as a wrapper of statx if the kernel does not support fstatat64 syscall
by yann＠sionneau.net 31 Oct '23

31 Oct '23

From: Yann Sionneau <ysionneau(a)kalray.eu> Define fstatat64 as a wrapper of statx if the kernel does not support fstatat64 syscall This is the case for non-legacy architectures that don't define __ARCH_WANT_NEW_STAT in their linux arch/xxx/include/asm/unistd.h Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- libc/sysdeps/linux/common/fstatat64.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+) diff --git a/libc/sysdeps/linux/common/fstatat64.c b/libc/sysdeps/linux/common/fstatat64.c index 711521a6a..836ed4114 100644 --- a/libc/sysdeps/linux/common/fstatat64.c +++ b/libc/sysdeps/linux/common/fstatat64.c @@ -43,5 +43,28 @@ int fstatat64(int fd, const char *file, struct stat64 *buf, int flag) } libc_hidden_def(fstatat64) #else + +#if defined(__NR_statx) && defined(__UCLIBC_HAVE_STATX__) +# include <sys/stat.h> +# include <statx_cp.h> +# include <fcntl.h> // for AT_NO_AUTOMOUNT + +int fstatat64(int fd, const char *file, struct stat64 *buf, int flag) +{ + struct statx tmp; + + int r = INLINE_SYSCALL(statx, 5, fd, file, AT_NO_AUTOMOUNT | flag, + STATX_BASIC_STATS, &tmp); + + if (r != 0) + return r; + + __cp_stat_statx ((struct stat *)buf, &tmp); + + return 0; +} +libc_hidden_def(fstatat64) +#endif + /* should add emulation with fstat64() and /proc/self/fd/ ... */ #endif -- 2.42.0

3 11

[PATCH v2] fork: generate stub on no-MMU systems
by Ben Wolsieffer 18 Sep '23

18 Sep '23

From: Ben Wolsieffer <Ben.Wolsieffer(a)hefring.com> fork() can be implemented using either the fork or clone syscalls on MMU systems. Therefore the stub is only generated if neither __NR_fork nor __NR_clone are defined. The stub code manually undefines __NR_fork on no-MMU systems in an attempt to enable the stub, but this doesn't work because __NR_clone is still defined. It is not appropriate to undefine __NR_clone because clone is available on no-MMU, it is just not capable of implementing fork. This patch directly enables the fork stub if __ARCH_USE_MMU__ is not defined. This eliminates the need to undefine __NR_fork, so this code is removed Signed-off-by: Ben Wolsieffer <ben.wolsieffer(a)hefring.com> --- libc/sysdeps/linux/common/stubs.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/libc/sysdeps/linux/common/stubs.c b/libc/sysdeps/linux/common/stubs.c index c17e509b9..634dcde0d 100644 --- a/libc/sysdeps/linux/common/stubs.c +++ b/libc/sysdeps/linux/common/stubs.c @@ -34,10 +34,6 @@ __attribute_used__ static int ret_enosys_stub(void) link_warning(stub, #stub ": this function is not implemented") \ strong_alias(ret_enosys_stub, stub) -#ifndef __ARCH_USE_MMU__ -# undef __NR_fork -#endif - #ifdef __arm__ # define __NR_fadvise64_64 __NR_arm_fadvise64_64 # define __NR_fadvise64 __NR_arm_fadvise64_64 @@ -120,7 +116,7 @@ make_stub(fgetxattr) make_stub(flistxattr) #endif -#if !defined __NR_fork && !defined __NR_clone +#if !defined __ARCH_USE_MMU__ || (!defined __NR_fork && !defined __NR_clone) make_stub(fork) #endif -- 2.42.0

3 3

[PATCH] kvx: fix asm syntax
by yann＠sionneau.net 16 Sep '23

16 Sep '23

From: Paul Iannetta <piannetta(a)kalray.eu> Previously kvx assembler considered all separators (",", "?", "=", "[]") to be the same, this is not the case anymore hence we need to fix all the misformed assembly. Signed-off-by: Paul Iannetta <piannetta(a)kalray.eu> Acked-by: Yann Sionneau <ysionneau(a)kalray.eu> Tested-by: Yann Sionneau <ysionneau(a)kalray.eu> Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- libc/string/kvx/memcpy.S | 4 ++-- libc/sysdeps/linux/kvx/clone.S | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/libc/string/kvx/memcpy.S b/libc/string/kvx/memcpy.S index 290e705b4..70e8db910 100644 --- a/libc/string/kvx/memcpy.S +++ b/libc/string/kvx/memcpy.S @@ -53,7 +53,7 @@ ENTRY(memcpy) ;; cb.deqz $r7? .Lstreaming_loop_end ;; - loopdo $r7? .Lstreaming_loop_end + loopdo $r7, .Lstreaming_loop_end ;; sq 0[$r0] = $r32r33 addd $r2 = $r2, -256 @@ -162,7 +162,7 @@ ENTRY(memcpy) ;; cb.deqz $r7? .Lloop_32_end ;; - loopdo $r7? .Lloop_32_end + loopdo $r7, .Lloop_32_end ;; lo $r32r33r34r35 = 0[$r1] addd $r1 = $r1, 32 diff --git a/libc/sysdeps/linux/kvx/clone.S b/libc/sysdeps/linux/kvx/clone.S index 71ea80a01..71553a4a9 100644 --- a/libc/sysdeps/linux/kvx/clone.S +++ b/libc/sysdeps/linux/kvx/clone.S @@ -59,10 +59,10 @@ ENTRY (__clone) scall SYS_ify(clone) ;; /* If 0, then we are the child */ - cb.deqz $r0, L(child_start) + cb.deqz $r0? L(child_start) ;; /* Else we are the parent, and we need to check for errors */ - cb.dltz $r0, L(clone_error) + cb.dltz $r0? L(clone_error) ;; /* No error ! Yeepa ! */ ret -- 2.42.0

2 1

[PATCH 1/5] kvx: add support for kv3-2 (Coolidge v2 SoC)
by yann＠sionneau.net 16 Sep '23

16 Sep '23

From: Yann Sionneau <ysionneau(a)kalray.eu> The only difference, with regard to libc, is the compile flag: -march= Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- Rules.mak | 3 ++- extra/Configs/Config.kvx | 18 ++++++++++++++++++ 2 files changed, 20 insertions(+), 1 deletion(-) diff --git a/Rules.mak b/Rules.mak index 8984edb38..c3313b257 100644 --- a/Rules.mak +++ b/Rules.mak @@ -459,7 +459,8 @@ ifeq ($(TARGET_ARCH),csky) endif ifeq ($(TARGET_ARCH),kvx) - CPU_CFLAGS-$(CONFIG_KVX) += -march=kvx + CPU_CFLAGS-y += -march=$(call qstrip,$(TARGET_MARCH)) + CPU_LDFLAGS-y += -march=$(call qstrip,$(TARGET_MARCH)) endif ifeq ($(TARGET_ARCH),m68k) diff --git a/extra/Configs/Config.kvx b/extra/Configs/Config.kvx index 398ffceaa..04df53c38 100644 --- a/extra/Configs/Config.kvx +++ b/extra/Configs/Config.kvx @@ -7,6 +7,24 @@ config TARGET_ARCH string default "kvx" +choice + prompt "Target architecture variant" + help + Select CPU variant to use + +config CONFIG_KVX_COOLIDGE_V1 + bool "Coolidge V1" + +config CONFIG_KVX_COOLIDGE_V2 + bool "Coolidge V2" + +endchoice + +config TARGET_MARCH + string + default "kv3-1" if CONFIG_KVX_COOLIDGE_V1 + default "kv3-2" if CONFIG_KVX_COOLIDGE_V2 + config FORCE_OPTIONS_FOR_ARCH bool default y -- 2.42.0

2 5

[PATCH] fork: fix stub on no-MMU systems
by Ben Wolsieffer 14 Sep '23

14 Sep '23

From: Ben Wolsieffer <Ben.Wolsieffer(a)hefring.com> The stub code undefs __NR_fork on no-MMU systems, but it also needs to undef __NR_clone since fork can now be implemented using clone, and clone is not available either on no-MMU systems. --- libc/sysdeps/linux/common/stubs.c | 1 + 1 file changed, 1 insertion(+) diff --git a/libc/sysdeps/linux/common/stubs.c b/libc/sysdeps/linux/common/stubs.c index c17e509b9..11849e3ba 100644 --- a/libc/sysdeps/linux/common/stubs.c +++ b/libc/sysdeps/linux/common/stubs.c @@ -36,6 +36,7 @@ __attribute_used__ static int ret_enosys_stub(void) #ifndef __ARCH_USE_MMU__ # undef __NR_fork +# undef __NR_clone #endif #ifdef __arm__ -- 2.42.0

2 2

[PATCH] daemon.c: make _fork_parent static inline again
by Max Filippov 07 Sep '23

07 Sep '23

The commit cf649082c7d4 ("remove forced gcc optimization") removed -O3 optimization specified in the code for the function _fork_parent, but at the same time it removed the 'static inline' part of the function definition. That change seems unintended and _fork_parent is not a part of the libc interface. Make it static inline again. Signed-off-by: Max Filippov <jcmvbkbc(a)gmail.com> --- libc/unistd/daemon.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libc/unistd/daemon.c b/libc/unistd/daemon.c index 53fd902bfd72..599c3c8e6cd4 100644 --- a/libc/unistd/daemon.c +++ b/libc/unistd/daemon.c @@ -63,7 +63,7 @@ /* use clone() to get fork() like behavior here -- we just want to disassociate * from the controlling terminal */ -pid_t _fork_parent(void) +static inline pid_t _fork_parent(void) { INTERNAL_SYSCALL_DECL(err); register long ret = INTERNAL_SYSCALL(clone, err, 2, CLONE_VM, 0); -- 2.30.2

2 1

non-canonical reference to canonical protected function `__pthread_key_create'
by Chris Packham 05 Sep '23

05 Sep '23

Hi, I'm seeing a failure building ISL using uclibc-ng and GCC 13.2 [ALL ] /home/ctng/x-tools/x86_64-multilib-linux-uclibc/lib/gcc/x86_64-multilib-linux-uclibc/13.2.0/../../../../x86_64-multilib-linux-uclibc/bin/ld.bfd: isl_test2.o: non-canonical reference to canonical protected function `__pthread_key_create' in /home/ctng/x-tools/x86_64-multilib-linux-uclibc/x86_64-multilib-linux-uclibc/sysroot/lib64/libc.so.1 [ALL ] /home/ctng/x-tools/x86_64-multilib-linux-uclibc/lib/gcc/x86_64-multilib-linux-uclibc/13.2.0/../../../../x86_64-multilib-linux-uclibc/bin/ld.bfd: failed to set dynamic section sizes: bad value I don't really understand all the ins and outs but it looks like some libstdc++ APIs used by ISLs test code boils down to a call to __pthread_key_create which is marked protected so ld complains. I can successfully build the same code with glibc (haven't tried musl). So I'm reaching out to uclibc-ng rather than binutils or GCC. Any thoughts on what might be the problem and what the solution might be?

1 0