Only unblock an asynchronous signal if it is used

[babsingh]

Currently, all supported asynchronous signals are unblocked during
startup. A signal is unblocked even when a signal handler is not
installed for it. This conflicts with an OMR consumer, who wants to
block a set of signals. The OMR consumer blocks a set of signals but the
OMR signal library unblocks those signals at startup. OMR signal library
should only unblock a signal if a signal handler is registered for it.
In other words, a signal should not be unblocked if it is not used i.e.
a signal handler is not registered for it.

Related:

1. Customer Issue (IntelliJ IDE doesn't receive signals; 28th Oct. 2018)
2. OMR Fix (Handle signals blocked by a parent process; 23rd Nov. 2018)
3. Add support for new signals (SIGUSR1, SIGUSR2, ...; 15th April 2019)

Signed-off-by: Babneet Singh sbabneet@ca.ibm.com

[babsingh]

Performance Results

Compile the micro-benchmark

gcc -pthread -O3 benchmark_pthread_sigmask.c -o benchmark_pthread_sigmask

Run the micro-benchmark

/usr/bin/time -v ./benchmark_pthread_sigmask <ITERATIONS>

Elapsed time

Number of ITERATIONS	Elapsed time (m:ss)
1	0:00.00
10	0:00.00
100	0:00.00
1000	0:00.01
1000	0:00.14

Currently, pthread_sigmask is invoked once at startup for all the supported signals. With this fix, pthread_sigmask will be invoked everytime a signal handler is installed for a signal. Generally, a master handler is only installed once for a signal. Before this fix, pthread_sigmask is invoked once. After this fix, pthread_sigmask will be invoked ~32 times (== number of signals).

Each iteration of the micro-benchmark invokes pthread_sigmask 32 times. The above results for the elapsed time suggest that this fix will have negligible impact on performance.

Micro-benchmark (benchmark_pthread_sigmask.c):

#include <stdio.h>
#include <signal.h>
#include <stdlib.h>

static int
unblockSignal(int signal)
{
	int rc = 0;
	sigset_t signalSet;

	rc = sigemptyset(&signalSet);
	if (0 != rc) {
		goto exit;
	}

	rc = sigaddset(&signalSet, signal);
	if (0 != rc) {
		goto exit;
	}

	/* Unblock the signal. */
	rc = pthread_sigmask(SIG_UNBLOCK, &signalSet, NULL);
exit:
	return rc;
}

int main(int argc,char* argv[]) 
{
	int i = 0;
	int j = 0;
	int rc = 0;
	int iterations = 0;
	int signals[] = {SIGHUP, SIGINT, SIGQUIT, SIGILL, SIGTRAP, SIGABRT, SIGBUS, SIGFPE, SIGKILL, SIGUSR1, SIGSEGV, SIGUSR2, SIGPIPE, SIGALRM, SIGTERM, SIGSTKFLT, SIGCHLD, SIGCONT, SIGSTOP, SIGTSTP, SIGTTIN, SIGTTOU, SIGURG, SIGXCPU, SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH, SIGIO, SIGPWR, SIGSYS};
	int numSignals = (int)sizeof(signals)/sizeof(signals[0]);
	if (argc !=2) {
		printf("Error: Number of iterations not provided.\n");
		exit(-1);
	}
	
	iterations = atoi(argv[1]);
	if (iterations <= 0) {
		printf("Error: Number of iterations is not a valid positive integer\n");
		exit(-1);
	}

	for (i = 0; i < iterations; i++) {
		for (j = 0; j < numSignals; j++) {
			rc = unblockSignal(signals[j]);
			if (0 != rc) {
				printf("Error: Unable to unblock signal=%d j=%d\n", signals[j], j);
				exit(-1);
			}
		}
	}

	return rc;
}

[babsingh]

Initial testing looks good. Awaiting for test results from customer's nightly build. Code is ready-to-be-reviewed. But it should not be merged until test results from customer's nightly build are received.

[babsingh]

The failures have been resolved in the customer's nightly build. So, this pull request can be merged after the code-review.

[rwy7]

@genie-omr build all

[rwy7]

@genie-omr build all

[Leonardo2718]

FYI: I'm manually stopping some tests for this PR that look like they're hanging.

[babsingh]

@rwy0717 Unrelated errors were happening in the CI build jobs. I have rebased my branch. Can you please retry the builds?

[rwy7]

@genie-omr build all

[rwy7]

@genie-omr build all

[youngar]

@genie-omr build x32linux
Relaunching build due to machine issues