Current allocate_stack logic for create stacks is to first mmap all
the required memory with the desirable memory and then mprotect the
guard area with PROT_NONE if required. Although it works as expected,
it pessimizes the allocation because it requires the kernel to actually
increase commit charge (it counts against the available physical/swap
memory available for the system).
The only issue is to actually check this change since side-effects are
really Linux specific and to actually account them it would require a
kernel specific tests to parse the system wide information. On the kernel
I checked /proc/self/statm does not show any meaningful difference for
vmm and/or rss before and after thread creation. I could only see
really meaningful information checking on system wide /proc/meminfo
between thread creation: MemFree, MemAvailable, and Committed_AS shows
large difference without the patch. I think trying to use these
kind of information on a testcase is fragile.
The BZ#18988 reports shows that the commit pages are easily seen with
mlockall (MCL_FUTURE) (with lock all pages that become mapped in the
process) however a more straighfoward testcase shows that pthread_create
could be faster using this patch:
--
static const int inner_count = 256;
static const int outer_count = 128;
static
void *thread1(void *arg)
{
return NULL;
}
static
void *sleeper(void *arg)
{
pthread_t ts[inner_count];
for (int i = 0; i < inner_count; i++)
pthread_create (&ts[i], &a, thread1, NULL);
for (int i = 0; i < inner_count; i++)
pthread_join (ts[i], NULL);
return NULL;
}
int main(void)
{
pthread_attr_init(&a);
pthread_attr_setguardsize(&a, 1<<20);
pthread_attr_setstacksize(&a, 1134592);
pthread_t ts[outer_count];
for (int i = 0; i < outer_count; i++)
pthread_create(&ts[i], &a, sleeper, NULL);
for (int i = 0; i < outer_count; i++)
pthread_join(ts[i], NULL);
assert(r == 0);
}
return 0;
}
--
On x86_64 (4.4.0-45-generic, gcc 5.4.0) running the small benchtests
I see:
$ time ./test
real 0m3.647s
user 0m0.080s
sys 0m11.836s
While with the patch I see:
$ time ./test
real 0m0.696s
user 0m0.040s
sys 0m1.152s
So I added a pthread_create benchtest (thread_create) which check
the thread creation latency. As for the simple benchtests, I saw
improvements in thread creation on all architectures I tested the
change.
Checked on x86_64-linux-gnu, i686-linux-gnu, aarch64-linux-gnu,
arm-linux-gnueabihf, powerpc64le-linux-gnu, sparc64-linux-gnu,
and sparcv9-linux-gnu.
[BZ #18988]
* benchtests/thread_create-inputs: New file.
* benchtests/thread_create-source.c: Likewise.
* support/xpthread_attr_setguardsize.c: Likewise.
* support/Makefile (libsupport-routines): Add
xpthread_attr_setguardsize object.
* support/xthread.h: Add xpthread_attr_setguardsize prototype.
* benchtests/Makefile (bench-pthread): Add thread_create.
* nptl/allocatestack.c (allocate_stack): Call mmap with PROT_NONE and
then mprotect the required area.
This is intended as a type-safe alternative to obstacks and
hand-written realloc constructs. The implementation avoids
writing function pointers to the heap.
Simplify the Linux accept4 implementation based on the assumption
that it is available in some way. __ASSUME_ACCEPT4_SOCKETCALL was
previously unused, so remove it.
For ia64, the accept4 system call (and socket call) were backported
in kernel version 3.2.18. Reflect this in the installation
instructions.
Before this change, the loop iterating over RRs in the answer
section stopped at the first CNAME record, never printing them.
The CNAME and PTR record contents was extracted from the wrong
buffer (whole packet instead RDATA). This desynced the parsing
after the first CNAME or PTR record.
Also fix the AAAA record parsing by checking their sizes.
The commit documents the ownership rules around 'struct pthread' and
when a thread can read or write to the descriptor. With those ownership
rules in place it becomes obvious that pd->stopped_start should not be
touched in several of the paths during thread startup, particularly so
for detached threads. In the case of detached threads, between the time
the thread is created by the OS kernel and the creating thread checks
pd->stopped_start, the detached thread might have already exited and the
memory for pd unmapped. As a regression test we add a simple test which
exercises this exact case by quickly creating detached threads with
large enough stacks to ensure the thread stack cache is bypassed and the
stacks are unmapped. Before the fix the testcase segfaults, after the
fix it works correctly and completes without issue.
For a detailed discussion see:
https://www.sourceware.org/ml/libc-alpha/2017-01/msg00505.html
The support/tst-support_record_failure-2.out test attempts to run
built code even if run-built-tests = no, so failing with
build-many-glibcs.py for all architectures whose code cannot be run on
the system running the script. This patch disables the test in that
case.
Tested for x86_64 (native), and for aarch64 with build-many-glibcs.py.
* support/Makefile (tests-special): Make definition conditional on
[$(run-built-tests) = yes].
($(objpfx)tst-support_record_failure-2.out): Make rule conditional
on [$(run-built-tests) = yes].
The new functions support_record_failure records a test failure,
but does not terminate the process. The macros TEST_VERIFY
and TEST_VERIFY_EXIT check that a condition is true.
The new test driver in <support/test-driver.c> has feature parity with
the old one. The main difference is that its hooking mechanism is
based on functions and function pointers instead of macros. This
commit also implements a new environment variable, TEST_COREDUMPS,
which disables the code which disables coredumps (that is, it enables
them if the invocation environment has not disabled them).
<test-skeleton.c> defines wrapper functions so that it is possible to
use existing macros with the new-style hook functionality.
This commit changes only a few test cases to the new test driver, to
make sure that it works as expected.
