mcexec: --disable-sched-yield: avoid kernel/user switch

configure

@@ -634,6 +634,7 @@ KERNDIR
 KMODDIR
 ETCDIR
 SBINDIR
+LIBDIR
 BINDIR
 TARGET
 UNAME_R
@@ -4039,6 +4040,9 @@ case $WITH_TARGET in
 	if test "X$BINDIR" = X; then
 		BINDIR="$prefix/bin"
 	fi
+	if test "X$LIBDIR" = X; then
+		LIBDIR="$prefix/lib"
+	fi
 	if test "X$SBINDIR" = X; then
 		SBINDIR="$prefix/sbin"
 	fi
@@ -4506,6 +4510,15 @@ else
 $as_echo "$as_me: mcoverlayfs is disabled" >&6;}
 fi
 
+if test "x$LIBDIR" != "x" ; then
+
+cat >>confdefs.h <<_ACEOF
+#define MCKERNEL_LIBDIR "$LIBDIR"
+_ACEOF
+
+fi
+
+
 
 
 

configure.ac

@@ -162,6 +162,9 @@ case $WITH_TARGET in
 	if test "X$BINDIR" = X; then
 		BINDIR="$prefix/bin"
 	fi
+	if test "X$LIBDIR" = X; then
+		LIBDIR="$prefix/lib"
+	fi
 	if test "X$SBINDIR" = X; then
 		SBINDIR="$prefix/sbin"
 	fi
@@ -295,6 +298,10 @@ else
 	AC_MSG_NOTICE([mcoverlayfs is disabled])
 fi
 
+if test "x$LIBDIR" != "x" ; then
+	AC_DEFINE_UNQUOTED(MCKERNEL_LIBDIR,"$LIBDIR",[McKernel specific libraries])
+fi
+
 AC_SUBST(CC)
 AC_SUBST(XCC)
 AC_SUBST(ARCH)
@@ -302,6 +309,7 @@ AC_SUBST(KDIR)
 AC_SUBST(UNAME_R)
 AC_SUBST(TARGET)
 AC_SUBST(BINDIR)
+AC_SUBST(LIBDIR)
 AC_SUBST(SBINDIR)
 AC_SUBST(ETCDIR)
 AC_SUBST(KMODDIR)

executer/config.h.in

@@ -72,6 +72,9 @@
 /* Define to address of kernel symbol zap_page_range, or 0 if exported */
 #undef MCCTRL_KSYM_zap_page_range
 
+/* McKernel specific libraries */
+#undef MCKERNEL_LIBDIR
+
 /* Define to the address where bug reports for this package should be sent. */
 #undef PACKAGE_BUGREPORT
 

executer/user/Makefile.in

@@ -1,9 +1,10 @@
 CC=@CC@
 BINDIR=@BINDIR@
+LIBDIR=@LIBDIR@
 KDIR ?= @KDIR@
 CFLAGS=-Wall -O -I.
 VPATH=@abs_srcdir@
-TARGET=mcexec
+TARGET=mcexec libsched_yield
 @uncomment_if_ENABLE_MEMDUMP@TARGET+=eclair
 LIBS=@LIBS@
 IHKDIR ?= $(VPATH)/../../../ihk/linux/include/
@@ -16,6 +17,9 @@ mcexec: mcexec.c
 eclair: eclair.c
 	$(CC) $(CFLAGS) -I${IHKDIR} -o $@ $^ $(LIBS)
 
+libsched_yield: libsched_yield.c
+	$(CC) -shared -fPIC -Wl,-soname,sched_yield.so.1 -o libsched_yield.so.1.0.0 $^ -lc -ldl
+
 clean:
 	$(RM) $(TARGET) *.o
 
@@ -24,5 +28,7 @@ clean:
 install:
 	mkdir -p -m 755 $(BINDIR)
 	install -m 755 mcexec $(BINDIR)
+	mkdir -p -m 755 $(LIBDIR)
+	install -m 755 libsched_yield.so.1.0.0 $(LIBDIR)
 	@uncomment_if_ENABLE_MEMDUMP@install -m 755 eclair $(BINDIR)
 

executer/user/libsched_yield.c

@@ -0,0 +1,27 @@
+#define _GNU_SOURCE
+#include <dlfcn.h>
+#include <sys/time.h>
+#include <sched.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+
+#undef sched_yield
+
+typedef int (*int_void_fn)(void);
+
+static int_void_fn orig_sched_yield = 0;
+
+int sched_yield(void)
+{
+#if 0
+	if (!orig_sched_yield) {
+		orig_sched_yield = (int_void_fn)dlsym(RTLD_NEXT, "sched_yield");
+	}
+
+	printf("sched_yield() called\n");
+#endif
+
+	return 0;
+}

executer/user/mcexec.c

@@ -152,6 +152,7 @@ static int fd;
 static char *exec_path = NULL;
 static char *altroot;
 static const char rlimit_stack_envname[] = "MCKERNEL_RLIMIT_STACK";
+static const char ld_preload_envname[] = "MCKERNEL_LD_PRELOAD";
 static int ischild;
 static int enable_vdso = 1;
 static int mpol_no_heap = 0;
@@ -160,6 +161,7 @@ static int mpol_no_bss = 0;
 static int no_bind_ikc_map = 0;
 static unsigned long mpol_threshold = 0;
 static int profile = 0;
+static int disable_sched_yield = 0;
 
 /* Partitioned execution (e.g., for MPI) */
 static int nr_processes = 0;
@@ -1321,6 +1323,12 @@ static struct option mcexec_options[] = {
 		.flag =		NULL,
 		.val =		'm',
 	},
+	{
+		.name =		"disable-sched-yield",
+		.has_arg =	no_argument,
+		.flag =		&disable_sched_yield,
+		.val =		1,
+	},
 	/* end */
 	{ NULL, 0, NULL, 0, },
 };
@@ -1361,10 +1369,6 @@ int main(int argc, char **argv)
 		altroot = "/usr/linux-k1om-4.7/linux-k1om";
 	}
 
-	/* Collect environment variables */
-	envs_len = flatten_strings(-1, NULL, environ, &envs);
-	envs = envs;
-
 	rlim_stack.rlim_cur = MCEXEC_DEF_CUR_STACK_SIZE;
 	rlim_stack.rlim_max = MCEXEC_DEF_MAX_STACK_SIZE;
 
@@ -1436,6 +1440,29 @@ int main(int argc, char **argv)
 		return 1;
 	}
 
+	if (disable_sched_yield) {
+		char sched_yield_lib_path[PATH_MAX];
+		sprintf(sched_yield_lib_path, "%s/libsched_yield.so.1.0.0",
+			MCKERNEL_LIBDIR);
+		__dprintf("%s: %s\n", __FUNCTION__, sched_yield_lib_path);
+		if (setenv("LD_PRELOAD", sched_yield_lib_path, 1) < 0) {
+			printf("%s: warning: failed to set LD_PRELOAD for sched_yield\n",
+					__FUNCTION__);
+		}
+	}
+	/* Set LD_PRELOAD to McKernel specific value */
+	else if (getenv(ld_preload_envname)) {
+		if (setenv("LD_PRELOAD", getenv(ld_preload_envname), 1) < 0) {
+			printf("%s: warning: failed to set LD_PRELOAD environment variable\n",
+					__FUNCTION__);
+		}
+		unsetenv(ld_preload_envname);
+	}
+
+	/* Collect environment variables */
+	envs_len = flatten_strings(-1, NULL, environ, &envs);
+	envs = envs;
+
 #ifdef ENABLE_MCOVERLAYFS
 	__dprintf("mcoverlay enable\n");
 	char mcos_procdir[PATH_MAX];