Memory stores cause minor page faults after fork() but not after vfork() on Linux

README.md

$ make test
./test-fork-exec-memset
vfork-exec-memset: using vfork

 Performance counter stats for './vfork-exec-memset':

           244,243      page-faults:u                                                         
           244,243      minor-faults:u                                                        
                 0      major-faults:u                                                        

       0.803667298 seconds time elapsed

       0.200547000 seconds user
       0.595543000 seconds sys



 Performance counter stats for './fork-exec-memset':

           488,411      page-faults:u                                                         
           488,411      minor-faults:u                                                        
                 0      major-faults:u                                                        

       1.322967201 seconds time elapsed

       0.288548000 seconds user
       1.004638000 seconds sys

fork-exec-memset.c

// Memory stores cause minor page faults after fork() but not after vfork() on Linux
// See discussion in https://twitter.com/offlinemark/status/1643348064835612672

#include <stdlib.h>
#include <string.h>
#include <libgen.h>
#include <unistd.h>
#include <stdio.h>
#include <errno.h>
#include <sys/wait.h>

enum { gigabyte = 1000000000 };

char *program_name;

enum { FORK_EXEC_USE_VFORK = 1 };
int fork_exec(char * const program, int flags)
{
	int r, wstatus;
	pid_t pid;
	if (flags & FORK_EXEC_USE_VFORK) {
		pid = vfork();
	} else {
		pid = fork();
	}
	char * const argv[] = { NULL };
	if (pid == 0) {
		r = execvp(program, argv);
		if (r != 0) {
			fprintf(stderr, "execvp: %s\n", strerror(errno));
		}
	} else {
		int options = 0;
		r = waitpid(pid, &wstatus, options);
		if (r == -1) {
			fprintf(stderr, "waitpid: %s\n", strerror(errno));
			return r;
		}
		return wstatus;
	}
	return -1;
}

int main(int argc, char **argv)
{
	program_name = basename(argv[0]);

	volatile char *buf = malloc(gigabyte);
	memset(buf, 0x55, gigabyte);
	// fprintf(stderr, "program_name: %s\n", program_name);
	if (strstr(program_name, "vfork-")) {
		fprintf(stderr, "%s: using vfork\n", program_name);
		fork_exec("true", FORK_EXEC_USE_VFORK);
	} else {
		fork_exec("true", 0);
	}
	memset(buf, 0xaa, gigabyte);
	return 0;
}

GNUmakefile

CFLAGS := -Wall -Og
PROGRAMS := fork-exec-memset vfork-exec-memset

.PHONY: all
all: $(PROGRAMS)

.PHONY: test
test: all
	./test-fork-exec-memset

vfork-exec-memset: fork-exec-memset
	ln -sf $< $@

.PHONY: clean
clean:
	rm -f $(PROGRAMS)

test-fork-exec-memset

#!/bin/bash

for i in vfork-exec-memset fork-exec-memset ; do
	perf stat -e page-faults,minor-faults,major-faults -- ./$i
done

I had to compile with -Og instead of -Os to keep the second memset() call from being optimized out by GCC.