Timo Teräs [Fri, 22 Oct 2010 11:58:13 +0000 (13:58 +0200)]
libm/x86: use call instead of jump for wrappers
GCC can emit prologue/epilogue code for the functions in various
different cases:
- frame pointers
- PIC build (to load ebx for indirect calls/jumps)
- forced stack smashing protection
If we used jump in such cases, we'd corrupt the call stack and
crash.
Denys Vlasenko [Thu, 21 Oct 2010 16:19:35 +0000 (18:19 +0200)]
sleep: code shrink
Use less stack by using same "sigset_t set" object for new and saved
signal set, remove redundant clearing of set, and do not save/restore
errno around sigprocmask(SIG_SETMASK) - it never changes it.
While at it, improve comments and make code style consistent
across sleep.c file.
text data bss dec hex filename
- 242 0 0 242 f2 libc/unistd/sleep.o
+ 197 0 0 197 c5 libc/unistd/sleep.o
Natanael Copa [Thu, 14 Oct 2010 06:35:08 +0000 (06:35 +0000)]
getservice: fix handling of long lines
Don't try to be smart by dynamically realloc buffersize as it doesn't
work. Instead, be simple and allocate a buffer big enough.
This fixes a memory leak when calling getserv{ent,byname,byport}
multiple times.
To save memory we reduce number of max aliases. We seldomly will need
more than 1 anyways. While here, fix segfault that happened if there
were too many aliases.
Natanael Copa [Thu, 14 Oct 2010 06:35:04 +0000 (06:35 +0000)]
config parser: do not assume that realloc return same pointer
We need to update the parser->line pointer on realloc and do not
initialize the token array til after the potensial realloc in
bb_get_chunk_with_continuation().
While here, also replace a realloc() with malloc() where pointer always
is NULL.
Protected symbols should not be overridden by
symbols from other modules. Such symbols are exported
i.e. globally visible, but references from whithin
defining modules are satisfied locally.
nptl: Fix libpthread build when UCLIBC_LINUX_SPECIFIC is disabled
NPTL library needs both madvise and statfs symbols, that are guarded
by UCLIBC_LINUX_SPECIFIC option. This fix provides these symbols too
when NPTL is used, indipendently by UCLIBC_LINUX_SPECIFIC choice.
Otherwise libpthread link fails as below:
LD libpthread-0.9.32-git.so
libpthread/nptl/libpthread_so.a(pthread_create.oS): In function `__free_tcb':
pthread_create.c:(.text+0x1184): undefined reference to `madvise'
libpthread/nptl/libpthread_so.a(sem_open.oS): In function `__where_is_shmfs':
sem_open.c:(.text+0x764): undefined reference to `statfs'
collect2: ld returned 1 exit status
make: *** [lib/libpthread.so] Error 1
Carmelo Amoroso [Wed, 30 Jun 2010 06:32:15 +0000 (08:32 +0200)]
libc: Handle cancellation in non multiplexed socket calls
For those archs that provide non multiplexed socket calls
it possible to implement the lib C wrappers without calling the
multi-purpose __socketcall. For a subset of these functions that
are cancellation point, it needs to correctly handle cancellation.
nptl: Remove nptl/sysdeps/generic directory from include list
All archs have now moved their generic sysdep.h header from
sysdeps/generic to sysdeps/common directory, so it is possible
to remove the sysdeps/generic directory from the include list.
sh: update the memcpy adding a new loop with aggressive prefetching
After exploring different prefetch distance-degree combinations
in this new update of the memcpy function, a new loop has been added
for moving many cache lines with an aggressive prefetching schema.
Prefetch has been removed when move few cache line aligned blocks.
As final result, this memcpy gives us the same performances for small
sizes (we already had!) and better numbers for big copies.
In case of SH4-300 CPU Series, benchmarks show a gain of ~20% for sizes
from 4KiB to 256KiB.
In case of the SH4-200, there is a gain of ~40% for sizes bigger than
32KiB.
sh: move data without fetching cache block within the memset
With this patch the movca.l instruction is used within the memset.
The current memset implementation only uses the FPU and there is
an real gain for all the sizes.
Adding the movca.l instruction numbers always are better than the generic code.
There is a big gain for size greater than 64 KiB but number are worst for 4-32KiB
sizes compared with the implementation without movca.l.
libc: Fix cancellation handling in some C functions
According to POSIX.1-2008 standard, the following syscalls shall be
cancellation points : waitid, sleep, fdatasync, ppoll.
Further, if generic syscall is not available and stubs are
configured, provide the stub implementation for function.
libubacktrace: Provide uClibc with backtrace functions
A new shared object, libubacktrace.so.0 is added to uClibc
to provide backtrace functions to support application self-debugging.
This set of functions requires to dynamically load libgcc_s.so so they
need to call dlopen/dlsym that are provided by libdl. For this reason
they cannot be included into libc.so.0 but are provided by a new library.
User application that wants to use backtrace needs to be compiled with
-fexceptions option and -rdynamic to get full symbols printed and must be
linked against libubacktrace.so
CC libc/misc/internals/__h_errno_location.os
CC libc/misc/internals/parse_config.os
libc/misc/internals/parse_config.c: In function
'bb_get_chunk_with_continuation':
libc/misc/internals/parse_config.c:77: error: 'PAGE_SIZE' undeclared (first use in this function)
libc/misc/internals/parse_config.c:77: error: (Each undeclared identifier is reported only once
libc/misc/internals/parse_config.c:77: error: for each function it appears in.)
make[2]: *** [libc/misc/internals/parse_config.os] Error 1
Carmelo Amoroso [Mon, 23 Aug 2010 22:14:00 +0000 (00:14 +0200)]
libc: Provide both getpgid and __getpgid symbols
Indeed unistd.h header does export '__getpgid' by default, while
uClibc provide only 'getpgid' implementation. The 'getpgid' symbol
is exported by standard header if __USE_XOPEN_EXTENDED is defined.
This patch alignes uClibc implementation with standard header (matching
with glibc behaviour).
.. and NPTL results in broken resolver in very annoying ways.
Now, it seems that most of the uclibc code does not work well
if res_state is TLS variable. Technically, this is the correct thing
to do since this gives proper per-thread resolving behavior, and
it also makes the config options overridable per thread. This
probably what apps expect as glibc does it too.
But alas, most places use _res to sync up static global variables
which results in breakage. It gets more or less randomly selected
which threads options get applied. Also in case of multiple servers
it looks like the retry logic is shared between all threads, e.g.
two concurrent resolutions can make other resolvers skip nameservers
due to shared "last_ns_num".
And finally the timeout/attempts commit breaks the accumulated stuff
horribly. What happens is:
1. multithreaded application startups, initializes resolver,
resolves things just fine
2. resolv.conf gets changed, application calls res_init
after res_init uclibc will call res_sync on all resolver functions
to refresh globals from the TLS variable _res
3. res_init was called only in one thread, so other thread's
_res contains all zeroes (yes, this is correct app usage:
res_init should be called only from one thread)
4. threads not calling res_init get broken resolver due to timeout
being set to zero
Now, one proper solution would be to:
1. make __open_nameservers return the configuration options
2. pass the config options struct to res_sync_func so it can do
the proper overrides from per-thread _res
3. remove the related globals and use locally config options from
__open_nameservers
But technically, the correct thing (as in glibc does this) is:
- res_init increases global "res_init timestamp"
- use _res (or pointer within there) for resolver retries etc.
get proper per-thread behavior
- resolvers functions call "maybe_init" which reinitialize the
TLS'ed resolver state (e.g. reload resolv.conf) if their res_init
timestamp is out dated
As an immediate emergency kludge fix, the following might do:
resolv: fix options handling for TLS _res
If _res is in TLS (NPTL), it might not get initialized. Assume
zeroes mean default values.
Timo Teräs [Fri, 6 Aug 2010 10:31:35 +0000 (13:31 +0300)]
nptl: fix calling convention for __pthread_mutex_cond_lock
The assembly versions of pthread_cond_wait calls
__pthread_mutex_cond_lock and __pthread_mutex_cond_lock_adjust
using internal calling convention (which differs from default
calling convention at least on x86). Thus these two functions
must be defined with internal_function or the call sequence goes
wrong.
__pthread_mutex_cond_lock resides in
sysdeps/unix/sysv/linux/pthread_mutex_cond_lock.c, but it does
evil macro definitions and includes pthread_mutex_lock.c, so
we need to add some extra kludge to pthread_mutex_lock.c to get
the prototypes correctly.
Timo Teräs [Fri, 6 Aug 2010 08:34:11 +0000 (11:34 +0300)]
config parser: fix memory corruption
fgets will happily write over allocated area limits. Adjusted the
buffer size according to how much is already read.
Also increase the maximum default line length, as 80 is slightly
small. It might be better if bb_get_chunk_with_continuation would
reallocate the line buffer if it was not user given.
Ingo van Lil [Wed, 28 Jul 2010 12:52:08 +0000 (14:52 +0200)]
Immediately try next nameserver on recv() failure
If there is a problem communicating with a nameserver the __dns_lookup()
function will not immediately advance to the next nameserver but instead
continue waiting until the timeout expires. This will cause a 30 second
delay even if no nameserver is configured in resolv.conf and no DNS is
running on localhost.
Mike Frysinger [Tue, 27 Jul 2010 05:48:22 +0000 (01:48 -0400)]
new helper funcs for alloca/malloc with mmu/nommu
The rpc rcmd code has some ugly ifdef mazes to handle mmu/nommu differences
just to select alloca or malloc. Unify those with some helper macros in a
new header, and then convert the rcmd code over to it.
This is all geared towards fixing the getdents helper functions which only
use alloca() atm. Now that we have helper functions, convert the getdents
functions over too.
The 0.9.31 release included a change to malloc-simple to request
uninitialized memory from noMMU kernels. Unfortunately, the corresponding
calloc() code assumed that memory returned by malloc() was already zeroed,
which leads to all kinds of nastiness.
David A Ramos [Tue, 27 Jul 2010 11:10:15 +0000 (13:10 +0200)]
Fix ctime() standard compliance bug
fixes issue2209:
ctime() was updated in 0.9.31 to call localtime_r() instead of
localtime() to
avoid using a static buffer. Unfortunately, this change replaces the
static
buffer (which is zeroed out on initialization) with an uninitialized
local
buffer.
In the common case, this has no effect. However, with a sufficiently
large
time_t value, the value returned differs from that returned by
asctime(localtime(t)), and thus violates the ANSI/ISO standard.
An example input is (on a 64-bit machine):
time_t t = 0x7ffffffffff6c600;
ldso/mips: pltgot should array not address of array to dynamic info.
* This change was introduced in commit 9381d622e2411a35a5fd73a5a573eb269e2dd9c9
which rendered uclibc not booting on mips targets. Restoring it makes
it work again.
Roman I Khimov [Wed, 14 Jul 2010 13:19:29 +0000 (17:19 +0400)]
nptl i686: fix pthread_cond_wait.S compilation
__i686 is a gcc-defined macro, so i686 build failed with:
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S: Assembler messages:
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S:572: Error: junk at end of line, first unrecognized character is `1'
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S:573: Error: junk at end of line, first unrecognized character is `1'
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S:574: Error: junk at end of line, first unrecognized character is `1'
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S:575: Error: junk at end of line, first unrecognized character is `.'
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S:576: Error: junk at end of line, first unrecognized character is `1'
libpthread/nptl/sysdeps/unix/sysv/linux/i386/i686/../i486/pthread_cond_wait.S:579: Error: expected comma after name `' in .size directive
libc/inet/socketcalls.c: In function ‘__libc_recvfrom’:
libc/inet/socketcalls.c:229:1: warning: declaration of ‘_sys_result’
shadows a previous local
libc/inet/socketcalls.c:229:1: warning: shadowed declaration is here
Now automatically picks the correct (arch and subarch specific) impls in
favour of generic impls.
make O=/tmp/objs PREFIX=/my/sysroot -j
works now as expected (both out-of-tree as well as parallel-safe).
__errno_location / __h_errno_location access has to go through the PLT
like malloc/free, so the linuxthread variants gets used instead when
compiling with -pthread.
Timo Teräs [Wed, 30 Jun 2010 11:46:37 +0000 (14:46 +0300)]
more workarounds for GCC PR32219
Commit 2e53dd645d5348f207cec7f8595969dc566c5a55 workarounds GCC
bug when accessing _locale_init and _stdio_init. We need the same
fix for __errno_location and __h_errno_location otherwise we crash
calling null with static and non-threaded builds.
Khem Raj [Wed, 30 Jun 2010 11:10:37 +0000 (04:10 -0700)]
mips/syscall-error: Choose the correct version for setting up errno.
* Current function is a C protype and PSEUDO macro does not
transfer syscall return parameters to correct argument registers
for a C function. This causes problem with syscalls setting wrong
value for errno when they encounter an error.
projects / uclibc-ng.git / log