| Age | Commit message (Collapse) | Author |
|---|
|
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
The asm code has high maintainance cost.
The cost is extremely high if we take into
consideration time64 implementation which implies
operating with 64bit numbers on 32bit system.
Also glibc uses mainly C in NPTL code.
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
Remove ^L (0x0c) chars from source code.
Signed-off-by: Petr Vorel <petr.vorel@gmail.com>
|
|
- Use TIME64 by default for rv32, usage of 32-bit time
leads to a lot of incompatibilities with linux kernel 6.6.x and later
versions.
- Add some other corrections to use proper system calls on riscv32
platform.
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
Quelches a compiler warning by that memset was implicitly declared.
|
|
The variable is used since 04a676f3c8d2443499f27612f69ee88e12089e61.
|
|
|
|
With time64 enabled we need to pass structure
which consists of two 64bit fields to clock_gettime64()
and clock_nanosleep_time64() syscalls with proper conversion
to regular timespec structure after syscall execution.
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
For BE architectures there is one significant difference
in comparison with time64 support for little-endian
architectures like ARMv7.
The difference is that we strictly need to pass two 64bit
values to system calls because Linux Kernel internally uses
`struct __kernel_timespec` and similar, which consists of two
64bit fields.
For this reason many files have been changed to convert
pointers to timespec-family structures (mixed of 64bit and 32bit values)
to the pointer of the similar but 64bit-only structures
for using as system calls args.
This is general prerequisite for any BE architecture.
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
Inside `if` branches the conditions
`as->ino == bs->ino` and `as->dev == bs->dev`
are always false.
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
This patch introduces *time64 syscalls support for uClibc-ng.
Currently the redirection of syscalls to their *time64
analogs is fully supported for 32bit ARM (ARMv5, ARMv6, ARMv7).
The main changes that take effect when time64 feature is enabled are:
- sizeof(time_t) is 8.
- There is a possibility os setting date beyond year 2038.
- some syscalls are redirected:
clock_adjtime -> clock_adjtime64
clock_getres -> clock_getres_time64
clock_gettime -> clock_gettime64
clock_nanosleep -> clock_nanosleep_time64
clock_settime -> clock_settime64
futex -> futex_time64
mq_timedreceive -> mq_timedreceive_time64
mq_timedsend -> mq_timedsend_time64
ppoll -> ppoll_time64
pselect6 -> pselect6_time64
recvmmsg -> recvmmsg_time64
rt_sigtimedwait -> rt_sigtimedwait_time64
sched_rr_get_interval -> sched_rr_get_interval_time64
semtimedop -> semtimedop_time64
timer_gettime -> timer_gettime64
timer_settime -> timer_settime64
timerfd_gettime -> timerfd_gettime64
timerfd_settime -> timerfd_settime64
utimensat -> utimensat_time64.
- settimeofday uses clock_settime (like in glibc/musl).
- gettimeofday uses clock_gettime (like in glibc/musl).
- nanosleep uses clock_nanosleep (like in glibc/musl).
- There are some fixes in data structures used by libc and kernel
for correct data handling both with and without enabled time64 support.
Signed-off-by: Dmitry Chestnykh <dm.chestnykh@gmail.com>
|
|
The spurious_wakeup_count variable is set but is never actually used for
the semaphore implementation. To avoid a clang warning for this case
remove the unused variable.
|
|
Threads currently have 2-4 MiB stacks by default (depending on the
platform). This is fine on MMU platforms, where this stack space is not
actually allocated until it is used, but tends to waste a large amount
of memory on no-MMU platforms.
This patch adds a PTHREADS_STACK_DEFAULT_SIZE Kconfig option that allows
the user to override the default stack size at build time. This allows
the user to select a reasonable default stack size for the software that
runs on their system, and avoids the need to patch every package to add
calls to pthread_attr_setstacksize().
An alternative to this patch would be to change the hardcoded default
stack size on no-MMU platforms, but it is difficult to choose an
appropriate value because the minimum required stack depends on the
software in use. This would also be a breaking change.
Signed-off-by: Ben Wolsieffer <ben.wolsieffer@hefring.com>
|
|
Fix the [-Warray-parameter=] warning for __sigsetjmp generated by GCC 11 and
later GCC versions:
|
| warning: argument 1 of type 'struct __jmp_buf_tag *' declared as a pointer [-Warray-parameter=]
| extern int __sigsetjmp (struct __jmp_buf_tag *__env, int __savemask) __THROWNL;
| ...
| note: previously declared as an array 'struct __jmp_buf_tag[1]'
| extern int __sigsetjmp (struct __jmp_buf_tag __env[1], int __savemask)
|
Use the same fix as in glibc. The fix is to move the struct __jmp_buf_tag
definition to a separate bits/ header so it can be included in
pthread.h, to allow to use an array (as in setjmp.h) rather than a pointer
in the declaration.
Signed-off-by: Pavel Kozlov <pavel.kozlov@synopsys.com>
|
|
Setting signal handler in the kernel and then updating sighandler[sig]
results in a crash if a signal which handler is being changed from
SIG_DFL to a non-default was pending. Improve the race a little and
update the sighandler[sig] before the sigaction syscall. It doesn't
eliminate the race entirely, but fixes this particular failing case.
E.g. this fixes the 100% reproducible segfault in the busybox hush shell
built with FEATURE_EDITING_WINCH on ssh client's terminal window resize,
but in that case there's one more even bigger issue: busybox calls
sigaction with both old and new signal pointers pointing to the same
structure instance, as a result act->sa_handler after the sigaction
syscall is not what the user requested, but the previous handler.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Changed linux thread's stack allocation mmap to use new MAP_FIXED_NOREPLACE flag on kernels >4.17.
For older kernels, a check is added to see if requested address matches the address received.
If the addresses don't match, an error is returned and thread creation is aborted.
Signed-off-by: linted <linted@users.noreply.github.com>
|
|
When uclibc is built with static PIE support the _dl_load_base variable
shared between the libc-tls.c and reloc_static_pie.c creates the
dependency that requires linking reloc_static_pie.o into static
position-dependent executables resulting in the following build errors:
gcc -static test.c -o test
...ld:
...usr/lib/libc.a(reloc_static_pie.os):(.text+0x0):
undefined reference to `_DYNAMIC'
Move _dl_load_base definition to libc-tls.c to resolve this dependency
and fix static PDE build.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Static PIE ELFs may be loaded on noMMU linux platforms with FDPIC
support, but they don't have adjustable brk, and thus cannot allocate
memory for the TLS. Use mmap instead of sbrk to allocate initial TLS
memory when building with static PIE support for noMMU.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
on i386, x86_64, and arm.
This patch adds the generation of rcrt1.o which is used by gcc when compiling with the --static-pie flag.
rcrt1.o differs from crt1.o and Scrt1.o in that it the executable has a dynamic section but no relocations have been performed prior to _start being called.
crt1.o assumes there to be no dynamic relocations, and Scrt1.o has all relocations performed prior to execution by lsdo.
The new reloc_static_pie function handles parsing the dynamic section, and performing the relocations in a architecture agnostic method.
It also sets _dl_load_base which is used when initalizing TLS to ensure loading from the proper location.
This allows for easier porting of static-pie support to additional architectures as only modifications to crt1.S to find the load address are required.
Signed-off-by: linted <linted@users.noreply.github.com>
|
|
Thomas has repored failure building ARM 32-bit systems for ARMv8 cores
CC libpthread/linuxthreads/mutex.os
/tmp/ccn8SFKU.s: Assembler messages:
/tmp/ccn8SFKU.s:162: Error: swp{b} use is obsoleted for ARMv8 and later
/tmp/ccn8SFKU.s:186: Error: swp{b} use is obsoleted for ARMv8 and later
/tmp/ccn8SFKU.s:203: Error: swp{b} use is obsoleted for ARMv8 and later
/tmp/ccn8SFKU.s:224: Error: swp{b} use is obsoleted for ARMv8 and later
make[1]: *** [Makerules:369: libpthread/linuxthreads/mutex.os] Error 1
This is due to libpthread/linuxthreads/sysdeps/arm/pt-machine.h which
uses the swp instruction that is not allowed on ARMv8.
All __ARM_ARCH >= 7 support ldrex/strex instructions, so unlock
testandset() varaint for them.
Reported-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>
|
|
O0 build result in the following codegen
00000000 <ldrex>:
0: b480 push {r7}
2: b085 sub sp, #20
4: af00 add r7, sp, #0
6: 6078 str r0, [r7, #4]
8: 687b ldr r3, [r7, #4]
a: e853 3f00 ldrex r3, [r3]
e: 60fb str r3, [r7, #12]
10: 68fb ldr r3, [r7, #12]
12: 4618 mov r0, r3
14: 3714 adds r7, #20
16: 46bd mov sp, r7
18: f85d 7b04 ldr.w r7, [sp], #4
1c: 4770 bx lr
0000001e <strex>:
1e: b480 push {r7}
20: b085 sub sp, #20
22: af00 add r7, sp, #0
24: 6078 str r0, [r7, #4]
26: 6039 str r1, [r7, #0]
28: 687b ldr r3, [r7, #4]
2a: 683a ldr r2, [r7, #0]
2c: e842 3300 strex r3, r3, [r2]
30: 60fb str r3, [r7, #12]
32: 68fb ldr r3, [r7, #12]
34: 4618 mov r0, r3
36: 3714 adds r7, #20
38: 46bd mov sp, r7
3a: f85d 7b04 ldr.w r7, [sp], #4
3e: 4770 bx lr
00000040 <testandset>:
40: b590 push {r4, r7, lr}
42: b083 sub sp, #12
44: af00 add r7, sp, #0
46: 6078 str r0, [r7, #4]
48: 6878 ldr r0, [r7, #4]
4a: f7ff fffe bl 0 <ldrex>
4e: 4603 mov r3, r0
50: 461c mov r4, r3
52: 6879 ldr r1, [r7, #4]
54: 2001 movs r0, #1
56: f7ff fffe bl 1e <strex>
5a: 4603 mov r3, r0
5c: 2b00 cmp r3, #0
5e: d1f3 bne.n 48 <testandset+0x8>
60: 4623 mov r3, r4
62: 4618 mov r0, r3
64: 370c adds r7, #12
66: 46bd mov sp, r7
68: bd90 pop {r4, r7, pc}
ARM ARM suggests that LoadExcl/StoreExcl loops are guaranteed to make
forward progress only if, for any LoadExcl/StoreExcl loop within a
single thread of execution, the software meets all of the following
conditions:
1 Between the Load-Exclusive and the Store-Exclusive, there are no
explicit memory accesses, preloads, direct or indirect System
register writes, address translation instructions, cache or TLB
maintenance instructions, exception generating instructions,
exception returns, or indirect branches.
...
Obviously condition is not met for O0 builds.
O2 build (which is highly likely the most common setting) able to do
the right thing resulting in
00000000 <ldrex>:
0: e850 0f00 ldrex r0, [r0]
4: 4770 bx lr
6: bf00 nop
00000008 <strex>:
8: e841 0000 strex r0, r0, [r1]
c: 4770 bx lr
e: bf00 nop
00000010 <testandset>:
10: 2101 movs r1, #1
12: 4603 mov r3, r0
14: e853 0f00 ldrex r0, [r3]
18: e843 1200 strex r2, r1, [r3]
1c: 2a00 cmp r2, #0
1e: d1f9 bne.n 14 <testandset+0x4>
20: 4770 bx lr
22: bf00 nop
Rather than depending on level of optimisation implement whole
ldrex/strex loop in inline assembly.
Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>
|
|
Signed-off-by: Yann Sionneau <ysionneau@kalray.eu>
|
|
TLS)
Create timer thread with sufficiant stack size (take into account
allocated space for thread-local-storage), for this backport glibc
commit 'Create internal threads with sufficient stack size' ([1],
[2]) introducing __pthread_get_minstack() and use it in
__start_helper_thread().
Fixes timer_create() in case of linking with library using large
TLS area (e.g openblas, see [3]).
[1] https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=2c1094bd700e63a8d7f547b3f5495bedb55c0a08
[2] https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=630f4cc3aa019ede55976ea561f1a7af2f068639
[3] http://lists.busybox.net/pipermail/buildroot/2021-April/308281.html
Signed-off-by: Peter Seiderer <ps.report@gmx.net>
|
|
This commit adds support for Kalray VLIW family (kvx)
Kalray kv3 core is embedded in Kalray Coolidge SoC. This core which is the
third of the KV family has the following features:
32/64 bits execution mode
6-issue VLIW architecture
64 x 64bits general purpose registers
SIMD instructions
little-endian
In order to build a usable toolchain, build scripts are provided at the
following address: https://github.com/kalray/build-scripts.
Kalray uses FOSS which is available at https://github.com/kalray
This includes Linux kernel, uClibc-ng, gcc, binutils, etc.
Signed-off-by: Clément Léger <cleger@kalray.eu>
Signed-off-by: Guillaume Thouvenin <gthouvenin@kalray.eu>
Signed-off-by: Laurent Thevenoux <lthevenoux@kalray.eu>
Signed-off-by: Marc Poulhies <mpoulhies@kalray.eu>
Signed-off-by: Marius Gligor <mgligor@kalray.eu>
Signed-off-by: Yann Sionneau <ysionneau@kalray.eu>
|
|
Fixes this:
In file included from libpthread/nptl/pthread_create.c:48:0:
libpthread/nptl/allocatestack.c: In function 'allocate_stack':
libpthread/nptl/allocatestack.c:602:6: warning: label 'mprot_error' defined but not used [-Wunused-label]
mprot_error:
^~~~~~~~~~~
Signed-off-by: Yann Sionneau <ysionneau@kalray.eu>
|
|
This reverts commit 5b58a1ebd89a4f05778441814e81817c82193fa3.
This breaks all static builds earlier to gcc 10 :(
Bad testing on my side.
|
|
Starting with GCC-10 multiple definitions of global variables by will be
rejected.
https://gcc.gnu.org/gcc-10/porting_to.html
This fixes multiple definitions of _dl_pagesize and _dl_tls_static_size
while attempting static linking.
Of course this only occurs when compiling something that requires these
symbols.
First patch submission so hopefully all done correctly.
thanks,
Lance Fredrickson
From e0686f7c03ce8e51ccffefeb6365e50311e6dd10 Mon Sep 17 00:00:00 2001
From: lancethepants <lancethepants@gmail.com>
Date: Wed, 15 Jul 2020 13:09:26 -0600
Subject: [PATCH] Starting with GCC-10 multiple definitions of global variables
by will be rejected. This fixes multiple definitions of _dl_pagesize and
_dl_tls_static_size while attempting static linking.
|
|
Add XCHAL definitions for S32C1I and EXCLUSIVE options to
xtensa-config.h, include it in places that implement atomic operations
and add implementations with exclusive access option opcodes.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Replace "a" constraints with "+m" to avoid forcing atomic variable
address into a register and let the compiler use non-zero offset in
load/store opcodes.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Rename various spare fields in structs to include a namespace
This should avoid accidental clashes with uses of the __unused symbol
in upstream projects. eg currently it causes a compile error in dhcpcd 8.x
due to their re-use of the __unused symbol as a macro
This follows the style of glibc which does something equivalent
|
|
|
|
Signed-off-by: Yann Sionneau <ysionneau@kalray.eu>
|
|
basically from or1k port of uClibc-ng, with fixes for structures in
pthreadtypes.h from 64 bit architectures.
18 testsuite failures counted.
|
|
The return type of syscall() is long so __syscall_error, which is jumped
to by syscall handlers to stash an error number into errno, must return
long too otherwhise it returs 4294967295L instead of -1L. For example,
syscall for x86_64 is defined in libc/sysdeps/linux/x86_64/syscall.S as
syscall:
movq %rdi, %rax /* Syscall number -> rax. */
movq %rsi, %rdi /* shift arg1 - arg5. */
movq %rdx, %rsi
movq %rcx, %rdx
movq %r8, %r10
movq %r9, %r8
movq 8(%rsp),%r9 /* arg6 is on the stack. */
syscall /* Do the system call. */
cmpq $-4095, %rax /* Check %rax for error. */
jae __syscall_error /* Branch forward if it failed. */
ret /* Return to caller. */
In libc/sysdeps/linux/x86_64/__syscall_error.c, __syscall_error is
defined as
int __syscall_error(void) attribute_hidden;
int __syscall_error(void)
{
register int err_no __asm__ ("%rcx");
__asm__ ("mov %rax, %rcx\n\t"
"neg %rcx");
__set_errno(err_no);
return -1;
}
So __syscall_error returns -1 as a 32-bit int in a 64-bit register, %rax
(0x00000000ffffffff, whose decimal value is decimal 4294967295) and a
test like this always returns false:
if (syscall(number, ...) == -1)
foo();
Fix the error by making __syscall_error return a long, like syscall().
The problem can be circumvented by the caller by coercing the returned
value to int before comparing it to -1:
if ((int) syscall(number, ...) == -1)
foo();
The same problem probably occurs on other 64-bit systems but so far only
x86_64 was tested, so this change must be considered experimental.
Signed-off-by: Carlos Santos <unixmania@gmail.com>
|
|
This patch seems needed in builds where
- SHARED is not defined (no shared lib support)
- and USE_TLS is set
Without this patch, static_dtv is free'ed.
See the following backtrace:
0 __do_check_chunk (p=0x52638 <fork_handler_pool+2296>) at libc/stdlib/malloc-standard/malloc.c:80
1 0x0000000000017fa0 in __do_check_inuse_chunk (p=0x52638 <fork_handler_pool+2296>) at libc/stdlib/malloc-standard/malloc.c:143
2 0x0000000000017354 in free (mem=0x52648 <static_dtv>) at libc/stdlib/malloc-standard/free.c:293
3 0x000000000002d5b0 in _dl_deallocate_tls (tcb=0x58690, dealloc_tcb=false) at libpthread/nptl/sysdeps/generic/dl-tls.c:588
4 0x0000000000021c0c in __deallocate_stack (pd=0x58000) at libpthread/nptl/allocatestack.c:717
5 0x0000000000024408 in __free_tcb (pd=0x58000) at libpthread/nptl/pthread_create.c:217
6 0x00000000000200ac in pthread_join (threadid=360448, thread_return=0x0 <k1c_start>) at libpthread/nptl/pthread_join.c:109
7 0x0000000000010354 in tf (a=0x58000) at tst-basic3.c:42
8 0x00000000000247c8 in start_thread (arg=0x4000200960) at libpthread/nptl/pthread_create.c:285
9 0x0000000000026560 in ?? ()
This backtrace is obtained while debugging tst-basic3 from the uclibc-ng nptl testsuite.
It aborts because of the assert in malloc:
https://elixir.bootlin.com/uclibc-ng/v1.0.31/source/libc/stdlib/malloc-standard/malloc.c#L80
Signed-off-by: Yann Sionneau <ysionneau@kalray.eu>
|
|
I've got several patches to fix ltp/openmp/uclibc-ng-test testcase fail on
c-sky.
- fix a ltp testcase.
- fix the problem that pthread creat will fail when libomp is linked before
libc, the variable pagesize is not init.
- fix tst-cancel4 and tst-cancel16. tst-cancelx4 and tst-cancelx16 still fail
with this patch applied, cleanup handler is not called for open/creat/fcntl,
seems some thing wrong with unwind, I haven't check the rootcause yet.
|
|
Replace sbrk with mmap since this commit disables sbrk area
for FDPIC MMU-less platform:
fs/binfmt_elf_fdpic.c: fix brk area overlap with stack on NOMMU
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/fs/binfmt_elf_fdpic.c?id=4ac313111018cb44ecc250445de5ccb93026a980
* libpthread/nptl/sysdeps/generic/libc-tls.c (__libc_setup_tls):
Handle __FDPIC__.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
busybox init checks it has pid 1, but getpid() returns another value
when building busybox statically. This is because the corresponding
area is not cleared when allocated (it is allocated with
MAP_UNINITIALIZED, whose behavior depends on the Linux kernel's
CONFIG_MMAP_ALLOW_UNINITIALIZED).
This patch fixes the problem by explicitly clearing the memory area.
* libpthread/nptl/sysdeps/generic/libc-tls.c (__libc_tls_setup):
Clear tlsblock.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
* libpthread/nptl/Makefile.in (libpthread-routines-): Remove
pthread_mutex_getprioceiling.c and pthread_mutex_setprioceiling.c.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
Allow both tmpfs and ramfs for shm devices.
* libpthread/nptl/linux_fsinfo.h (SHMFS_SUPER_MAGIC_WITH_MMU): Define.
(SHMFS_SUPER_MAGIC_WITHOUT_MMU): Define.
* libpthread/nptl/sem_open.c (__where_is_shmfs): Add support for
SHMFS_SUPER_MAGIC_WITHOUT_MMU.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
phdr->p_vaddr of TLS segment is not a valid value for FDPIC so we can
either translate phdr->p_vaddr using loadmap (not easy here) or use a
new linker script defined symbol, whih this patch does.
* libpthread/nptl/sysdeps/generic/libc-tls.c (__tdata_start): Declare.
(__libc_setup_tls): Support __FDPIC__.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
start_thread() uses it, but it is not supported on MMU-less systems.
* libpthread/nptl/pthread_create.c (start_thread): Call madvise
only if __ARCH_USE_MMU__ is defined.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
They do not work on MMU-less systems.
For atfork, implement a hidden function that always returns EPERM.
* libpthread/nptl/sysdeps/unix/sysv/linux/fork.c: Disable if
__ARCH_USE_MMU__ is not defined.
* libpthread/nptl/sysdeps/unix/sysv/linux/libc_pthread_init.c
(__libc_pthread_init): Handle __ARCH_USE_MMU__.
* libpthread/nptl/sysdeps/unix/sysv/linux/register-atfork.c: Likewise.
* libpthread/nptl/sysdeps/unix/sysv/linux/unregister-atfork.c: Likewise.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
|
Since mprotect does not work on MMU-less systems, disable it if
__ARCH_USE_MMU__ is not defined.
* libpthread/nptl/allocatestack.c (change_stack_perm): Call
mprotect only if __ARCH_USE_MMU__ is defined.
(allocate_stack): Likewise.
Signed-off-by: Mickaël Guêné <mickael.guene@st.com>
Signed-off-by: Christophe Lyon <christophe.lyon@st.com>
|
