get_nprocs() and get_nprocs_conf() use various methods to obtain an
accurate number of processors. Re-introduce __get_nprocs_sched() as
a source of information, and fix the order in which these methods are
used to return the most accurate information. The primary source of
information used in both functions remains unchanged.
This also changes __get_nprocs_sched() error return value from 2 to 0,
but all its users are already prepared to handle that.
Old fallback order:
get_nprocs:
/sys/devices/system/cpu/online -> /proc/stat -> 2
get_nprocs_conf:
/sys/devices/system/cpu/ -> /proc/stat -> 2
New fallback order:
get_nprocs:
/sys/devices/system/cpu/online -> /proc/stat -> sched_getaffinity -> 2
get_nprocs_conf:
/sys/devices/system/cpu/ -> /proc/stat -> sched_getaffinity -> 2
Fixes: 342298278e ("linux: Revert the use of sched_getaffinity on get_nproc")
Closes: BZ #28865
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Pass the actual number of bytes returned by the kernel.
Fixes: 33099d72e4 ("linux: Simplify get_nprocs")
Reviewed-by: Dmitry V. Levin <ldv@altlinux.org>
I used these shell commands:
../glibc/scripts/update-copyrights $PWD/../gnulib/build-aux/update-copyright
(cd ../glibc && git commit -am"[this commit message]")
and then ignored the output, which consisted lines saying "FOO: warning:
copyright statement not found" for each of 7061 files FOO.
I then removed trailing white space from math/tgmath.h,
support/tst-support-open-dev-null-range.c, and
sysdeps/x86_64/multiarch/strlen-vec.S, to work around the following
obscure pre-commit check failure diagnostics from Savannah. I don't
know why I run into these diagnostics whereas others evidently do not.
remote: *** 912-#endif
remote: *** 913:
remote: *** 914-
remote: *** error: lines with trailing whitespace found
...
remote: *** error: sysdeps/unix/sysv/linux/statx_cp.c: trailing lines
The use of sched_getaffinity on get_nproc and
sysconf (_SC_NPROCESSORS_ONLN) done in 903bc7dcc2 (BZ #27645)
breaks the top command in common hypervisor configurations and also
other monitoring tools.
The main issue using sched_getaffinity changed the symbols semantic
from system-wide scope of online CPUs to per-process one (which can
be changed with kernel cpusets or book parameters in VM).
This patch reverts mostly of the 903bc7dcc2, with the
exceptions:
* No more cached values and atomic updates, since they are inherent
racy.
* No /proc/cpuinfo fallback, since /proc/stat is already used and
it would require to revert more arch-specific code.
* The alloca is replace with a static buffer of 1024 bytes.
So the implementation first consult the sysfs, and fallbacks to procfs.
Checked on x86_64-linux-gnu.
Reviewed-by: Florian Weimer <fweimer@redhat.com>
This patch simplifies the memory allocation code and uses the sched
routines instead of reimplement it. This still uses a stack
allocation buffer, so it can be used on malloc initialization code.
Linux currently supports at maximum of 4096 cpus for most architectures:
$ find -iname Kconfig | xargs git grep -A10 -w NR_CPUS | grep -w range
arch/alpha/Kconfig- range 2 32
arch/arc/Kconfig- range 2 4096
arch/arm/Kconfig- range 2 16 if DEBUG_KMAP_LOCAL
arch/arm/Kconfig- range 2 32 if !DEBUG_KMAP_LOCAL
arch/arm64/Kconfig- range 2 4096
arch/csky/Kconfig- range 2 32
arch/hexagon/Kconfig- range 2 6 if SMP
arch/ia64/Kconfig- range 2 4096
arch/mips/Kconfig- range 2 256
arch/openrisc/Kconfig- range 2 32
arch/parisc/Kconfig- range 2 32
arch/riscv/Kconfig- range 2 32
arch/s390/Kconfig- range 2 512
arch/sh/Kconfig- range 2 32
arch/sparc/Kconfig- range 2 32 if SPARC32
arch/sparc/Kconfig- range 2 4096 if SPARC64
arch/um/Kconfig- range 1 1
arch/x86/Kconfig-# [NR_CPUS_RANGE_BEGIN ... NR_CPUS_RANGE_END] range.
arch/x86/Kconfig- range NR_CPUS_RANGE_BEGIN NR_CPUS_RANGE_END
arch/xtensa/Kconfig- range 2 32
With x86 supporting 8192:
arch/x86/Kconfig
976 config NR_CPUS_RANGE_END
977 int
978 depends on X86_64
979 default 8192 if SMP && CPUMASK_OFFSTACK
980 default 512 if SMP && !CPUMASK_OFFSTACK
981 default 1 if !SMP
So using a maximum of 32k cpu should cover all cases (and I would
expect once we start to have many more CPUs that Linux would provide
a more straightforward way to query for such information).
A test is added to check if sched_getaffinity can successfully return
with large buffers.
Checked on x86_64-linux-gnu and i686-linux-gnu.
Reviewed-by: Florian Weimer <fweimer@redhat.com>
This is an internal function meant to return the number of avaliable
processor where the process can scheduled, different than the
__get_nprocs which returns a the system available online CPU.
The Linux implementation currently only calls __get_nprocs(), which
in tuns calls sched_getaffinity.
Reviewed-by: Florian Weimer <fweimer@redhat.com>
We stopped adding "Contributed by" or similar lines in sources in 2012
in favour of git logs and keeping the Contributors section of the
glibc manual up to date. Removing these lines makes the license
header a bit more consistent across files and also removes the
possibility of error in attribution when license blocks or files are
copied across since the contributed-by lines don't actually reflect
reality in those cases.
Move all "Contributed by" and similar lines (Written by, Test by,
etc.) into a new file CONTRIBUTED-BY to retain record of these
contributions. These contributors are also mentioned in
manual/contrib.texi, so we just maintain this additional record as a
courtesy to the earlier developers.
The following scripts were used to filter a list of files to edit in
place and to clean up the CONTRIBUTED-BY file respectively. These
were not added to the glibc sources because they're not expected to be
of any use in future given that this is a one time task:
https://gist.github.com/siddhesh/b5ecac94eabfd72ed2916d6d8157e7dchttps://gist.github.com/siddhesh/15ea1f5e435ace9774f485030695ee02
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
malloc initialization depends on __get_nprocs, so using
scratch buffers in __get_nprocs may result in infinite recursion.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
There is no much gain in fallback to cpuinfo if sysfs is no present,
usually on restricted environment neither will be present. It also
simplifies the code and make all architecture use the sched_getaffinity
as the sysfs fallback.
Checked on sparc64-linux-gnu.
Both the sysfs and procfs parsing (through GET_NPROCS_PARSER) are
removed in favor the syscall. The initial scratch buffer should
fit to most of the common usage (1024 bytes with maps to 8192 CPUs).
Checked on x86_64-linux-gnu and aarch64-linux-gnu.
I used these shell commands:
../glibc/scripts/update-copyrights $PWD/../gnulib/build-aux/update-copyright
(cd ../glibc && git commit -am"[this commit message]")
and then ignored the output, which consisted lines saying "FOO: warning:
copyright statement not found" for each of 6694 files FOO.
I then removed trailing white space from benchtests/bench-pthread-locks.c
and iconvdata/tst-iconv-big5-hkscs-to-2ucs4.c, to work around this
diagnostic from Savannah:
remote: *** pre-commit check failed ...
remote: *** error: lines with trailing whitespace found
remote: error: hook declined to update refs/heads/master
As for gettimeofday, time will be implemented based on clock_gettime
on all platforms and internal code should use clock_gettime
directly. In addition to removing a layer of indirection, this will
allow us to remove the PLT-bypass gunk for gettimeofday.
The changed code always assumes __clock_gettime (CLOCK_REALTIME)
or __clock_gettime (CLOCK_REALTIME_COARSE) (for Linux case) cannot
fail, using the same rationale for gettimeofday change. And internal
helper was added (time_now).
Checked on x86_64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
powerpc64-linux-gnu, and powerpc-linux-gnu.
Reviewed-by: Lukasz Majewski <lukma@denx.de>
This patch consolidates all the non cancellable read calls to use
the __read_nocancel identifier. For non cancellable targets it will
be just a macro to call the default respective symbol while on Linux
will be a internal one.
Also, since it is used on libcrypto it is also exported in GLIBC_PRIVATE
namespace.
Checked on x86_64-linux-gnu, x86_64-linux-gnu-x32, and i686-linux-gnu.
* sysdeps/generic/not-cancel.h (read_not_cancel): Remove macro.
(__read_nocancel): New macro.
* sysdeps/unix/sysv/linux/Versions (libc) [GLIBC_PRIVATE]: Add
__read_nocancel.
* sysdeps/unix/sysv/linux/not-cancel.h (__read_nocancel): Remove
macro.
(__read_nocancel): New prototype.
* sysdeps/unix/sysv/linux/read.c (__read_nocancel): New function.
* catgets/open_catalog.c (__open_catalog): Replace read_not_cancel
with __read_nocancel.
* intl/loadmsgcat.c (read): Likewise.
* libio/fileops.c (_IO_file_read): Likewise.
* locale/loadlocale.c (_nl_load_locale): Likewise.
* login/utmp_file.c (getutent_r_file): Likewise.
(internal_getut_r): Likewise.
(getutline_r_file): Likewise.
* sysdeps/unix/sysv/linux/fips-private.h (fips_enable_p): Likewise.
* sysdeps/unix/sysv/linux/gethostid.c (gethostid): Likewise.
* sysdeps/unix/sysv/linux/getloadavg.c (getloadavg): Likewise.
* sysdeps/unix/sysv/linux/getlogin_r.c (__getlogin_r_loginuid):
Likewise.
* sysdeps/unix/sysv/linux/getsysstats.c (next_line): Likewise.
* sysdeps/unix/sysv/linux/i386/smp.h (is_smp_system): Likewise.
* sysdeps/unix/sysv/linux/ia64/has_cpuclock.c (has_cpuclock):
Likewise.
* sysdeps/unix/sysv/linux/libc_fatal.c (backtrace_and_maps):
Likewise.
* sysdeps/unix/sysv/linux/malloc-sysdep.h (check_may_shrink_heap):
Likewise.
* sysdeps/unix/sysv/linux/pthread_getname.c (pthread_getname_np):
Likewise.
* sysdeps/unix/sysv/linux/sysconf.c (__sysconf): Likewise.
Profiling git's test suite, Linus noted [1] that a disproportionately
large amount of time was spent reading /proc/meminfo. This is done by
the glibc functions get_phys_pages and get_avphys_pages, but they only
need the MemTotal and MemFree fields, respectively. That same
information can be obtained with a single syscall, sysinfo, instead of
six: open, fstat, mmap, read, close, munmap. While sysinfo also
provides more than necessary, it does a lot less work than what the
kernel needs to do to provide the entire /proc/meminfo. Both strace -T
and in-app microbenchmarks shows that the sysinfo() approach is
roughly an order of magnitude faster.
sysinfo() is much older than what glibc currently requires, so I don't
think there's any reason to keep the old parsing code. Moreover, this
makes get_[av]phys_pages work even in the absence of /proc.
Linus noted that something as simple as 'bash -c "echo"' would trigger
the reading of /proc/meminfo, but gdb says that many more applications
than just bash are affected:
Starting program: /bin/bash "-c" "echo"
Breakpoint 1, __get_phys_pages () at ../sysdeps/unix/sysv/linux/getsysstats.c:283
283 ../sysdeps/unix/sysv/linux/getsysstats.c: No such file or directory.
(gdb) bt
So it seems that any application that uses qsort on a moderately sized
array will incur this cost (once), which is obviously proportionately
more expensive for lots of short-lived processes (such as the git test
suite).
[1] http://thread.gmane.org/gmane.linux.kernel/2019285
Signed-off-by: Rasmus Villemoes <rv@rasmusvillemoes.dk>
* sysdeps/unix/sysv/linux/getsysstats.c (__get_phys_pages):
Use sysinfo system call instead of parsing /proc/meminfo.
* sysdeps/unix/sysv/linux/getsysstats.c (__get_avphys_pages):
Likewise.
__get_nprocs is called from malloc code, but calls fgets_unlocked,
which is not an ISO C or POSIX function. This patch fixes it to call
a new __fgets_unlocked name instead.
Note: there are various other uses of fgets_unlocked in glibc's
libraries, and I haven't yet investigated which others might also be
problematic (called directly or indirectly from standard functions)
and so need to change to use __fgets_unlocked.
Tested for x86_64 (testsuite, and that disassembly of installed shared
libraries is unchanged by the patch).
[BZ #17582]
* libio/iofgets.c [weak_alias && !_IO_MTSAFE_IO]
(__fgets_unlocked): Add alias of _IO_fgets. Use libc_hidden_def.
* libio/iofgets_u.c (fgets_unlocked): Rename to __fgets_unlocked
and define as weak alias of __fgets_unlocked. Use
libc_hidden_weak.
(__fgets_unlocked): Use libc_hidden_def.
* include/stdio.h (__fgets_unlocked): Declare. Use
libc_hidden_proto.
* sysdeps/unix/sysv/linux/getsysstats.c (phys_pages_info): Use
__fgets_unlocked instead of fgets_unlocked.
* sysdeps/unix/sysv/linux/alpha/getsysstats.c
(GET_NPROCS_CONF_PARSER): Likewise.
* sysdeps/unix/sysv/linux/sparc/getsysstats.c
(GET_NPROCS_CONF_PARSER): Likewise.
This patch removes conditionals on __ASSUME_O_CLOEXEC, and on
O_CLOEXEC being defined, in sysdeps/unix/sysv/linux/, now that
O_CLOEXEC support can be unconditionally assumed.
The patch is conservative in what it changes and further followup
cleanups may be possible. It may be possible to remove dl-opendir.c,
but the patch does not do so, just removing a redundant undefine and
redefine of __ASSUME_O_CLOEXEC. Also, __ASSUME_O_CLOEXEC is defined
unconditionally for Hurd as well as Linux. Thus, if we decide that
O_CLOEXEC support is a required feature of any glibc port, we could
remove __ASSUME_O_CLOEXEC and all conditionals on it throughout glibc,
rather than just cleaning up sysdeps/unix/sysv/linux/.
Tested x86_64 that the disassembly of installed shared libraries is
unchanged by this patch.
* sysdeps/unix/sysv/linux/dl-opendir.c (__ASSUME_O_CLOEXEC): Do
not undefine and redefine.
* sysdeps/unix/sysv/linux/getsysstats.c (__get_nprocs)
[O_CLOEXEC]: Make code unconditional.
(__get_nprocs) [!O_CLOEXEC]: Remove conditional code.
* sysdeps/unix/sysv/linux/shm_open.c: Do not include
<kernel-features.h>.
[O_CLOEXEC && !__ASSUME_O_CLOEXEC] (have_o_cloexec): Remove
conditional variable definition.
(shm_open) [O_CLOEXEC]: Make code unconditional.
(shm_open) [!O_CLOEXEC || !__ASSUME_O_CLOEXEC]: Remove conditional
code.
The implementation of __get_nprocs uses a stactic variable to cache
the value of the current number of processors. The caching breaks when
'time (NULL) == 0':
$ cat nproc.c
#include <stdio.h>
#include <time.h>
#include <sys/time.h>
int main(int argc, char *argv[])
{
time_t t;
struct timeval tv = {0, 0};
printf("settimeofday({0, 0}, NULL) = %d\n", settimeofday(&tv, NULL));
t = time(NULL);
printf("Time: %d, CPUs: %d\n", (unsigned int)t, get_nprocs());
return 0;
}
$ gcc -O3 nproc.c
$ ./a.out
settimeofday({0, 0}, NULL) = -1
Time: 1401311578, CPUs: 4
$ sudo ./a.out
settimeofday({0, 0}, NULL) = 0
Time: 0, CPUs: 0
The problem is with the condition used to check whether a cached
value should be returned or not:
static int cached_result;
static time_t timestamp;
time_t now = time (NULL);
time_t prev = timestamp;
atomic_read_barrier ();
if (now == prev)
return cached_result;
This patch fixes the problem by ensuring that 'cached_result' has
been set at least once before returning it.
are always at least 4 bytes in the returned line.
2009-04-15 Jakub Jelinek <jakub@redhat.com>
* sysdeps/unix/sysv/linux/getsysstats.c (__get_nprocs): Check
__libc_use_alloca (8192), if the stack is too small use 512 bytes
instead of 8K. Stop searching in /proc/stat after hitting first
line not starting with cpu.
(next_line): Truncate too long
lines at buffer size * 3/4 instead of pretending there were line
breaks inside of large lines.
(GET_NPROCS_PARSER): Change parameters and use next_line.
(__get_nprocs): Rewrite to not use stdio routines.
* sysdeps/unix/sysv/linux/sparc/getsysstats.c (GET_NPROCS_PARSER):
Change parameters and use next_line.
