devel

devel@uclibc-ng.org

3 participants
1149 discussions

by Yann Sionneau

This test was changed from using fork to vfork by: https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/commit/?id=822e4896c1072b… This change was wrong because man page says the child of vfork() must ONLY do either: - a call to one of exec(2) - a call to _exit(2) - get killed by a signal BUT the test case does: - a call to kill - and also a call to exit(3) and not _exit(2) This test produces a double free() causing assert in free because this is called several times: https://elixir.bootlin.com/uclibc-ng/latest/source/libc/stdlib/_atexit.c#L2… So, it seems wrong both in theory (according to man page) and in fact in practice (assert in free). Here I propose to get back to testing fork(), which is the original test case from: https://sourceware.org/bugzilla/show_bug.cgi?id=838 And to just "PASS" on no-mmu systems. For the record, here is the assert call stack: FAIL tst-mallocfork got 1 expected 0 ./tst-mallocfork: libc/stdlib/malloc-standard/malloc.c: 149: __do_check_inuse_chunk: Assertion `((((mchunkptr)(((char*)(p))+((p)->size & ~0x1)))->size) & 0x1)' failed. Didn't expect signal from child: got `Aborted' 0 free (mem=0x3a010) at libc/stdlib/malloc-standard/free.c:285 1 0x00000000000220c4 in __exit_handler (status=1) at libc/stdlib/_atexit.c:270 2 0x000000000001afa0 in __GI_exit (rv=1) at libc/stdlib/_atexit.c:301 3 0x0000000000010dcc in main (argc=1, argv=0x7ffffffd98) at ../test-skeleton.c:405 4 0x000000000001cba8 in __uClibc_main (main=0x10988 <main>, argc=1, argv=0x7ffffffd98, app_init=0x100e8, app_fini=0x23918 <_fini>, rtld_fini=0x0 <k1c_start>, stack_end=0x7ffffffd90) at libc/misc/internals/__uClibc_main.c:512 5 0x00000000000eb8b8 in ?? () Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- test/malloc/tst-mallocfork.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-) diff --git a/test/malloc/tst-mallocfork.c b/test/malloc/tst-mallocfork.c index edd9c39..e9e5133 100644 --- a/test/malloc/tst-mallocfork.c +++ b/test/malloc/tst-mallocfork.c @@ -9,10 +9,12 @@ #include <sys/wait.h> #include "../test-skeleton.h" +#ifdef __ARCH_USE_MMU__ + static void sig_handler (int signum) { - pid_t child = vfork (); + pid_t child = fork (); if (child == 0) exit (0); TEMP_FAILURE_RETRY (waitpid (child, NULL, 0)); @@ -35,7 +37,7 @@ do_test (void) } /* Create a child that sends the signal to be caught. */ - pid_t child = vfork (); + pid_t child = fork (); if (child == 0) { if (kill (parent, SIGALRM) == -1) @@ -48,5 +50,16 @@ do_test (void) return 0; } +#else + +static int +do_test (void) +{ + printf("Skipping test on non-mmu host!\n"); + return EXIT_SUCCESS; +} + +#endif + #define TEST_FUNCTION do_test () #include "../test-skeleton.c" -- 1.8.3.1

5 years, 11 months

[PATCH] Fix TLS allocation and loading in !SHARED case

by Yann Sionneau

For ARCH where shared lib are not supported: - SHARED is not set (!SHARED is true) - __ARCH_HAS_NO_LDSO__ is set so code inside #if !defined __ARCH_HAS_NO_LDSO__ && !defined SHARED is compiled-out. But without a call do _dl_aux_init(), _dl_phdr stays NULL and __libc_setup_tls won't be able to allocate memory for the in-executable TLS and also won't be able to load the initimage from ELF TLS segment. This results in segfault when doing things like "errno = 0" like in tst-cancel15.c for instance in uClibc-ng testsuite. Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- libc/misc/internals/__uClibc_main.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/libc/misc/internals/__uClibc_main.c b/libc/misc/internals/__uClibc_main.c index 849bca8..affa0ce 100644 --- a/libc/misc/internals/__uClibc_main.c +++ b/libc/misc/internals/__uClibc_main.c @@ -370,7 +370,7 @@ void __uClibc_main(int (*main)(int, char **, char **), int argc, char **argv, void (*app_init)(void), void (*app_fini)(void), void (*rtld_fini)(void), void *stack_end attribute_unused) { -#if !defined __ARCH_HAS_NO_LDSO__ && !defined SHARED +#ifndef SHARED unsigned long *aux_dat; ElfW(auxv_t) auxvt[AT_EGID + 1]; #endif @@ -396,7 +396,7 @@ void __uClibc_main(int (*main)(int, char **, char **), int argc, __environ = &argv[argc]; } -#if !defined __ARCH_HAS_NO_LDSO__ && !defined SHARED +#ifndef SHARED /* Pull stuff from the ELF header when possible */ memset(auxvt, 0x00, sizeof(auxvt)); aux_dat = (unsigned long*)__environ; -- 1.8.3.1

5 years, 11 months

[PATCH] Fix _dl_deallocate_tls in !SHARED case

by Yann Sionneau

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-stan… Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- libpthread/nptl/sysdeps/generic/dl-tls.c | 5 +++++ libpthread/nptl/sysdeps/generic/libc-tls.c | 5 ++++- 2 files changed, 9 insertions(+), 1 deletion(-) diff --git a/libpthread/nptl/sysdeps/generic/dl-tls.c b/libpthread/nptl/sysdeps/generic/dl-tls.c index 989e587a2..7d25e4706 100644 --- a/libpthread/nptl/sysdeps/generic/dl-tls.c +++ b/libpthread/nptl/sysdeps/generic/dl-tls.c @@ -48,6 +48,9 @@ /* Value used for dtv entries for which the allocation is delayed. */ # define TLS_DTV_UNALLOCATED ((void *) -1l) +#ifndef SHARED +extern dtv_t static_dtv; +#endif /* Out-of-memory handler. */ # ifdef SHARED @@ -584,6 +587,8 @@ _dl_deallocate_tls (void *tcb, bool dealloc_tcb) /* The array starts with dtv[-1]. */ #ifdef SHARED if (dtv != GL(dl_initial_dtv)) +#else + if ((dtv - 1) != &static_dtv) #endif free (dtv - 1); diff --git a/libpthread/nptl/sysdeps/generic/libc-tls.c b/libpthread/nptl/sysdeps/generic/libc-tls.c index a6df4cdc4..d746c9a38 100644 --- a/libpthread/nptl/sysdeps/generic/libc-tls.c +++ b/libpthread/nptl/sysdeps/generic/libc-tls.c @@ -42,7 +42,10 @@ extern size_t _dl_phnum; extern int __tdata_start; #endif -static dtv_t static_dtv[2 + TLS_SLOTINFO_SURPLUS]; +#ifdef SHARED +static +#endif +dtv_t static_dtv[2 + TLS_SLOTINFO_SURPLUS]; static struct -- 2.21.0

5 years, 11 months

[PATCH] Fix _dl_deallocate_tls in !SHARED case

by Yann Sionneau

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: 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-stan… Signed-off-by: Yann Sionneau <ysionneau(a)kalray.eu> --- libpthread/nptl/sysdeps/generic/dl-tls.c | 5 +++++ libpthread/nptl/sysdeps/generic/libc-tls.c | 5 ++++- 2 files changed, 9 insertions(+), 1 deletion(-) diff --git a/libpthread/nptl/sysdeps/generic/dl-tls.c b/libpthread/nptl/sysdeps/generic/dl-tls.c index 989e587a2..7d25e4706 100644 --- a/libpthread/nptl/sysdeps/generic/dl-tls.c +++ b/libpthread/nptl/sysdeps/generic/dl-tls.c @@ -48,6 +48,9 @@ /* Value used for dtv entries for which the allocation is delayed. */ # define TLS_DTV_UNALLOCATED ((void *) -1l) +#ifndef SHARED +extern dtv_t static_dtv; +#endif /* Out-of-memory handler. */ # ifdef SHARED @@ -584,6 +587,8 @@ _dl_deallocate_tls (void *tcb, bool dealloc_tcb) /* The array starts with dtv[-1]. */ #ifdef SHARED if (dtv != GL(dl_initial_dtv)) +#else + if ((dtv - 1) != &static_dtv) #endif free (dtv - 1); diff --git a/libpthread/nptl/sysdeps/generic/libc-tls.c b/libpthread/nptl/sysdeps/generic/libc-tls.c index a6df4cdc4..d746c9a38 100644 --- a/libpthread/nptl/sysdeps/generic/libc-tls.c +++ b/libpthread/nptl/sysdeps/generic/libc-tls.c @@ -42,7 +42,10 @@ extern size_t _dl_phnum; extern int __tdata_start; #endif -static dtv_t static_dtv[2 + TLS_SLOTINFO_SURPLUS]; +#ifdef SHARED +static +#endif +dtv_t static_dtv[2 + TLS_SLOTINFO_SURPLUS]; static struct -- 2.21.0

5 years, 11 months

__dl_iterate_phdr includes main application binary in list

by Bogdan Harjoc

After building uclibc-1.0.31 with AddressSanitizer enabled (gcc7.3 and 8.2), asan complains at runtime that it's not the first library in the so-list. With a breakpoint on AsanCheckDynamicRTPrereqs which calls __dl_iterate_phdr, I see the first module that uClibc returns has info->libname = "/home/user/myapp", while on glibc-2.27, libname is an empty string. ASAN depends on seeing this empty string. Should __dl_iterate_phdr return an empty string here, or is there an alternative to get asan-enabled uClibc toolchains ? Thanks.

5 years, 11 months

Re: [uclibc-ng-devel] [LTP] [PATCH] cpuset: disable for UCLIBC

by Vineet Gupta

On 3/14/19 3:01 AM, Petr Vorel wrote: > Hi Vineet, > >> libcpuset.c relies on FTS which is not enabled by default on uClibc. >> Instead of needing to reconfigure, rebuild uClibc lets disable this test >> for uClibc > Which uClibc version are you using? > Both uClibc and uclibc-ng has it [1] since 2011, version v0.9.32, in commit > a4aa01c12 ("Added fts support for traversing UNIX file hierarchies.") Right I'm using bleeding edge as well, and do have the header in sources, its just not configured/enabled by default hence not available in installation. > It's actually MUSL, what is missing it [2]. Although there is some > implementation [3], we should skip it for musl. But the clearest way is really > to check presence of <fts.h>, that's fix it for all libc. Spot on, that is indeed the best/ideal way of doing it. Do you know if such a mechanism already exists in LTP. If you point me to an example I can respin the patch accordingly. Thx, -Vineet

5 years, 12 months

[PATCH] cpuset: disable for UCLIBC

by Vineet Gupta

libcpuset.c relies on FTS which is not enabled by default on uClibc. Instead of needing to reconfigure, rebuild uClibc lets disable this test for uClibc Signed-off-by: Vineet Gupta <vgupta(a)synopsys.com> --- testcases/kernel/controllers/Makefile | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/testcases/kernel/controllers/Makefile b/testcases/kernel/controllers/Makefile index 404073947904..9f4abe28bfcf 100644 --- a/testcases/kernel/controllers/Makefile +++ b/testcases/kernel/controllers/Makefile @@ -29,6 +29,11 @@ LIBDIR := libcontrollers LIB := $(LIBDIR)/libcontrollers.a FILTER_OUT_DIRS := $(LIBDIR) +# controllers/cpuset/cpuset_lib/libcpuset.c uses fts +ifeq ($(UCLIBC),1) +FILTER_OUT_DIRS += cpuset +endif + $(LIBDIR): mkdir -p "$@" -- 2.7.4

6 years

Problems with circular dependencies when trying to dynamically load a GStreamer plugin on Raspberry Pi / Buildroot

by avnerus

Hi! I've been having troubles loading a gstreamer plugin for Raspberry Pi (libgstrpicamsrc.so) on Buildroot with Uclibc-ng 1.0.30. It seems that some circualr dependencies are preventing the plugin from loading (specifically libmmal_core.so and libmmal_util.so). I noticed that GStreamer loads its plugins using RTLD_NOW, but shouldn't it still be able to handle circular dependencies? The plugin does load correctly on glibc. I'm attaching the output of the load using LD_DEBUG. There seem to be a lot of messages regarding circular dependencies. There is also a deeper analysis of the issue and how I eventually worked around it in here: https://wiki.avner.us/doku.php?id=embedded-web-interaction:uclibc-ng-proble… Would appreciate any insights, Thank you! /Avner

6 years

[PATCH] Fix NULL pointer dereference in open_memstream()

by Eugene Yudin

Hi, Current version of uClibc-ng has issue in open_memstream() function. If the cookie variable is NULL (due malloc() fail) then null pointer is dereferenced after if block. The attached patch fixes this issue. -- Best regards, Eugene

6 years

uClibc-ng 1.0.31 allowed ARM/FDPIC

by Alexey Neyman

Hi, I've noticed that uClibc-ng now allows the FDPIC ELF for ARM: one of the crosstool-NG samples that configures a toolchain for ARM with no MMU now defaults to FDPIC rather than flat format. The problem is, though, it then fails - because GCC complains that: armv6-nommu-linux-uclibcgnueabi-gcc: error: unrecognized command line option '-mfdpic'; did you mean '-fpic'? I'll fix the default in crosstool-NG, but what intrigues me is that the -mfdpic option isn't listed in GCC manual as one of the supported options for ARM (it is only listed in FRV's options). How is this configuration built? Regards, Alexey.

6 years

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

devel