spec: prerelease 0.5 for testing ppoll

Change-Id: I51deb1c1703a986ba0aa4e02da9f53009554dbb7

CMakeLists.txt

@@ -10,7 +10,7 @@ project(mckernel C ASM)
 set(MCKERNEL_VERSION "1.7.0")
 
 # See "Fedora Packaging Guidlines -- Versioning"
-set(MCKERNEL_RELEASE "0.4")
+set(MCKERNEL_RELEASE "0.5")
 
 set(CMAKE_MODULE_PATH ${CMAKE_SOURCE_DIR}/cmake/modules)
 # for rpmbuild

arch/arm64/kernel/include/syscall_list.h

@@ -142,4 +142,3 @@ SYSCALL_HANDLED(1045, signalfd)
 SYSCALL_DELEGATED(1049, stat)
 SYSCALL_DELEGATED(1060, getpgrp)
 SYSCALL_HANDLED(1062, time)
-SYSCALL_DELEGATED(1069, epoll_wait)

executer/include/uprotocol.h

@@ -91,7 +91,10 @@ struct program_image_section {
 
 struct get_cpu_set_arg {
 	int nr_processes;
+	char *req_cpu_list;   // Requested by user-space
+	int req_cpu_list_len; // Lenght of request string
 	int *process_rank;
+	pid_t ppid;
 	void *cpu_set;
 	size_t cpu_set_size;	// Size in bytes
 	int *target_core;

executer/kernel/mcctrl/control.c

@@ -587,13 +587,14 @@ extern int mckernel_cpu_2_linux_cpu(struct mcctrl_usrdata *udp, int cpu_id);
 static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 {
 	struct mcctrl_usrdata *udp = ihk_host_os_get_usrdata(os);
-	struct mcctrl_part_exec *pe;
+	struct mcctrl_part_exec *pe = NULL, *pe_itr;
 	struct get_cpu_set_arg req;
 	struct mcctrl_cpu_topology *cpu_top, *cpu_top_i;
 	struct cache_topology *cache_top;
 	int cpu, cpus_assigned, cpus_to_assign, cpu_prev;
 	int ret = 0;
 	int mcexec_linux_numa;
+	int pe_list_len = 0;
 	cpumask_t *mcexec_cpu_set = NULL;
 	cpumask_t *cpus_used = NULL;
 	cpumask_t *cpus_to_use = NULL;
@@ -613,24 +614,126 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 		return -EINVAL;
 	}
 
-	pe = &udp->part_exec;
-
-	mutex_lock(&pe->lock);
-
 	if (copy_from_user(&req, (void *)arg, sizeof(req))) {
-		printk("%s: error copying user request\n", __FUNCTION__);
+		pr_err("%s: error copying user request\n", __func__);
 		ret = -EINVAL;
-		goto put_and_unlock_out;
+		goto put_out;
 	}
 
-	/* First process to enter CPU partitioning */
-	if (pe->nr_processes == -1) {
+	/* User requested CPU mask? */
+	if (req.req_cpu_list && req.req_cpu_list_len) {
+		char *cpu_list = NULL;
+
+		cpu_list = kmalloc(req.req_cpu_list_len, GFP_KERNEL);
+		if (!cpu_list) {
+			printk("%s: error: allocating CPU list\n", __FUNCTION__);
+			ret = -ENOMEM;
+			goto put_out;
+		}
+
+		if (copy_from_user(cpu_list,
+					req.req_cpu_list, req.req_cpu_list_len)) {
+			printk("%s: error copying CPU list request\n", __FUNCTION__);
+			kfree(cpu_list);
+			ret = -EINVAL;
+			goto put_out;
+		}
+
+		cpus_used = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
+		cpus_to_use = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
+		if (!cpus_to_use || !cpus_used) {
+			printk("%s: error: allocating CPU mask\n", __FUNCTION__);
+			ret = -ENOMEM;
+			kfree(cpu_list);
+			goto put_out;
+		}
+		memset(cpus_used, 0, sizeof(cpumask_t));
+		memset(cpus_to_use, 0, sizeof(cpumask_t));
+
+		/* Parse CPU list */
+		if (cpulist_parse(cpu_list, cpus_to_use) < 0) {
+			printk("%s: invalid CPUs requested: %s\n",
+				__FUNCTION__, cpu_list);
+			ret = -EINVAL;
+			kfree(cpu_list);
+			goto put_out;
+		}
+
+		memcpy(cpus_used, cpus_to_use, sizeof(cpumask_t));
+
+		/* Copy mask to user-space */
+		if (copy_to_user(req.cpu_set, cpus_used,
+					(req.cpu_set_size < sizeof(cpumask_t) ?
+					 req.cpu_set_size : sizeof(cpumask_t)))) {
+			printk("%s: error copying mask to user\n", __FUNCTION__);
+			ret = -EINVAL;
+			kfree(cpu_list);
+			goto put_out;
+		}
+
+		/* Copy IKC target core */
+		cpu = cpumask_next(-1, cpus_used);
+		if (copy_to_user(req.target_core, &cpu, sizeof(cpu))) {
+			printk("%s: error copying target core to user\n",
+					__FUNCTION__);
+			ret = -EINVAL;
+			kfree(cpu_list);
+			goto put_out;
+		}
+
+		/* Save in per-process structure */
+		memcpy(&ppd->cpu_set, cpus_used, sizeof(cpumask_t));
+		ppd->ikc_target_cpu = cpu;
+		printk("%s: %s -> target McKernel CPU: %d\n",
+			__func__, cpu_list, cpu);
+
+		ret = 0;
+		kfree(cpu_list);
+		goto put_out;
+	}
+
+	mutex_lock(&udp->part_exec_lock);
+	/* Find part_exec having same node_proxy */
+	list_for_each_entry_reverse(pe_itr, &udp->part_exec_list, chain) {
+		pe_list_len++;
+		if (pe_itr->node_proxy_pid == req.ppid) {
+			pe = pe_itr;
+			break;
+		}
+	}
+
+	if (!pe) {
+		/* First process to enter CPU partitioning */
+		pr_debug("%s: pe_list_len:%d\n", __func__, pe_list_len);
+		if (pe_list_len >= PE_LIST_MAXLEN) {
+			/* delete head entry of pe_list */
+			pe_itr = list_first_entry(&udp->part_exec_list,
+					struct mcctrl_part_exec, chain);
+			list_del(&pe_itr->chain);
+			kfree(pe_itr);
+		}
+
+		pe = kzalloc(sizeof(struct mcctrl_part_exec), GFP_KERNEL);
+		if (!pe) {
+			mutex_unlock(&udp->part_exec_lock);
+			ret = -ENOMEM;
+			goto put_out;
+		}
+		/* Init part_exec */
+		mutex_init(&pe->lock);
+		INIT_LIST_HEAD(&pe->pli_list);
 		pe->nr_processes = req.nr_processes;
 		pe->nr_processes_left = req.nr_processes;
+		pe->nr_processes_joined = 0;
+		pe->node_proxy_pid = req.ppid;
+
+		list_add_tail(&pe->chain, &udp->part_exec_list);
 		dprintk("%s: nr_processes: %d (partitioned exec starts)\n",
-				__FUNCTION__,
-				pe->nr_processes);
+				__func__, pe->nr_processes);
 	}
+	mutex_unlock(&udp->part_exec_lock);
+
+	mutex_lock(&pe->lock);
 
 	if (pe->nr_processes != req.nr_processes) {
 		printk("%s: error: requested number of processes"
@@ -640,7 +743,15 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 		goto put_and_unlock_out;
 	}
 
+	if (pe->nr_processes_joined >= pe->nr_processes) {
+		printk("%s: too many processes have joined to the group of %d\n",
+				__func__, req.ppid);
+		ret = -EINVAL;
+		goto put_and_unlock_out;
+	}
+
 	--pe->nr_processes_left;
+	++pe->nr_processes_joined;
 	dprintk("%s: nr_processes: %d, nr_processes_left: %d\n",
 			__FUNCTION__,
 			pe->nr_processes,
@@ -726,8 +837,6 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 				wake_up_interruptible(&pli_next->pli_wq);
 			}
 
-			/* Reset process counter to start state */
-			pe->nr_processes = -1;
 			ret = -ETIMEDOUT;
 			goto put_and_unlock_out;
 		}
@@ -975,16 +1084,8 @@ next_cpu:
 	/* Commit used cores to OS structure */
 	memcpy(&pe->cpus_used, cpus_used, sizeof(*cpus_used));
 
-	/* Reset if last process */
-	if (pe->nr_processes_left == 0) {
-		dprintk("%s: nr_processes: %d (partitioned exec ends)\n",
-				__FUNCTION__,
-				pe->nr_processes);
-		pe->nr_processes = -1;
-		memset(&pe->cpus_used, 0, sizeof(pe->cpus_used));
-	}
-	/* Otherwise wake up next process in list */
-	else {
+	/* If not last process, wake up next process in list */
+	if (pe->nr_processes_left != 0) {
 		++pe->process_rank;
 		pli_next = list_first_entry(&pe->pli_list,
 			struct process_list_item, list);
@@ -997,11 +1098,14 @@ next_cpu:
 	ret = 0;
 
 put_and_unlock_out:
+	mutex_unlock(&pe->lock);
+
+put_out:
+	mcctrl_put_per_proc_data(ppd);
+
 	kfree(cpus_to_use);
 	kfree(cpus_used);
 	kfree(mcexec_cpu_set);
-	mcctrl_put_per_proc_data(ppd);
-	mutex_unlock(&pe->lock);
 
 	return ret;
 }

executer/kernel/mcctrl/ikc.c

@@ -513,6 +513,7 @@ int prepare_ikc_channels(ihk_os_t os)
 
 	init_waitqueue_head(&usrdata->wq_procfs);
 	mutex_init(&usrdata->reserve_lock);
+	mutex_init(&usrdata->part_exec_lock);
 
 	for (i = 0; i < MCCTRL_PER_PROC_DATA_HASH_SIZE; ++i) {
 		INIT_LIST_HEAD(&usrdata->per_proc_data_hash[i]);
@@ -521,10 +522,8 @@ int prepare_ikc_channels(ihk_os_t os)
 
 	INIT_LIST_HEAD(&usrdata->cpu_topology_list);
 	INIT_LIST_HEAD(&usrdata->node_topology_list);
+	INIT_LIST_HEAD(&usrdata->part_exec_list);
 
-	mutex_init(&usrdata->part_exec.lock);
-	INIT_LIST_HEAD(&usrdata->part_exec.pli_list);
-	usrdata->part_exec.nr_processes = -1;
 	INIT_LIST_HEAD(&usrdata->wakeup_descs_list);
 	spin_lock_init(&usrdata->wakeup_descs_lock);
 
@@ -580,6 +579,18 @@ void destroy_ikc_channels(ihk_os_t os)
 
 	kfree(usrdata->channels);
 	kfree(usrdata->ikc2linux);
+
+	mutex_lock(&usrdata->part_exec_lock);
+	while (!list_empty(&usrdata->part_exec_list)) {
+		struct mcctrl_part_exec *pe;
+
+		pe = list_first_entry(&usrdata->part_exec_list,
+				struct mcctrl_part_exec, chain);
+		list_del(&pe->chain);
+		kfree(pe);
+	}
+	mutex_unlock(&usrdata->part_exec_lock);
+
 	kfree(usrdata);
 }
 

executer/kernel/mcctrl/mcctrl.h

@@ -324,13 +324,20 @@ struct process_list_item {
 	wait_queue_head_t pli_wq;
 };
 
+#define PE_LIST_MAXLEN 5
+
 struct mcctrl_part_exec {
 	struct mutex lock;	
 	int nr_processes;
+	/* number of processes to let in / out the synchronization point */
 	int nr_processes_left;
+	/* number of processes which have joined the partition */
+	int nr_processes_joined;
 	int process_rank;
+	pid_t node_proxy_pid;
 	cpumask_t cpus_used;
 	struct list_head pli_list;
+	struct list_head chain;
 };
 
 #define CPU_LONGS (((NR_CPUS) + (BITS_PER_LONG) - 1) / (BITS_PER_LONG))
@@ -353,6 +360,7 @@ struct mcctrl_usrdata {
 	int	job_pos;
 	int	mcctrl_dma_abort;
 	struct mutex reserve_lock;
+	struct mutex part_exec_lock;
 	unsigned long	last_thread_exec;
 	wait_queue_head_t wq_procfs;
 	struct list_head per_proc_data_hash[MCCTRL_PER_PROC_DATA_HASH_SIZE];
@@ -368,7 +376,7 @@ struct mcctrl_usrdata {
 	nodemask_t numa_online;
 	struct list_head cpu_topology_list;
 	struct list_head node_topology_list;
-	struct mcctrl_part_exec part_exec;
+	struct list_head part_exec_list;
 	int perf_event_num;
 };
 

executer/user/mcexec.c

@@ -2211,6 +2211,64 @@ int main(int argc, char **argv)
 
 	pthread_spin_init(&overlay_fd_lock, 0);
 
+	/* XXX: Fugaku: Fujitsu process placement fix */
+	if (getenv("FLIB_AFFINITY_ON_PROCESS")) {
+		char *cpu_s;
+		int flib_size;
+		char *flib_aff_orig, *flib_aff;
+		int cpu, off = 0;
+
+		flib_aff_orig = strdup(getenv("FLIB_AFFINITY_ON_PROCESS"));
+		if (!flib_aff_orig) {
+			fprintf(stderr, "error: dupping FLIB_AFFINITY_ON_PROCESS\n");
+			exit(EXIT_FAILURE);
+		}
+
+		flib_size = strlen(flib_aff_orig) * 2;
+
+		flib_aff = malloc(flib_size);
+		if (!flib_aff) {
+			fprintf(stderr, "error: allocating memory for "
+					"FLIB_AFFINITY_ON_PROCESS\n");
+			exit(EXIT_FAILURE);
+		}
+		memset(flib_aff, 0, flib_size);
+
+		cpu_s = strtok(flib_aff_orig, ",");
+		while (cpu_s) {
+			int ret;
+
+			/* "Shift" left by 12 CPUs */
+			cpu = atoi(cpu_s) - 12;
+
+			/* Prepend "," */
+			if (off > 0) {
+				ret = snprintf(flib_aff + off, flib_size - off, "%s", ",");
+				if (ret < 0) {
+					fprintf(stderr, "error: constructing "
+							"FLIB_AFFINITY_ON_PROCESS\n");
+					exit(EXIT_FAILURE);
+				}
+
+				off += ret;
+			}
+
+			ret = snprintf(flib_aff + off, flib_size - off, "%d", cpu);
+			if (ret < 0) {
+				fprintf(stderr, "error: constructing "
+						"FLIB_AFFINITY_ON_PROCESS\n");
+				exit(EXIT_FAILURE);
+			}
+
+			off += ret;
+			cpu_s = strtok(NULL, ",");
+		}
+
+		__dprintf("FLIB_AFFINITY_ON_PROCESS: %s -> %s\n",
+				getenv("FLIB_AFFINITY_ON_PROCESS"), flib_aff);
+		setenv("FLIB_AFFINITY_ON_PROCESS", flib_aff, 1);
+	}
+
 	ld_preload_init();
 
 #ifdef ADD_ENVS_OPTION
@@ -2483,9 +2541,12 @@ int main(int argc, char **argv)
 
 		CPU_ZERO(&mcexec_cpu_set);
 
+		cpu_set_arg.req_cpu_list = NULL;
+		cpu_set_arg.req_cpu_list_len = 0;
 		cpu_set_arg.cpu_set = (void *)&desc->cpu_set;
 		cpu_set_arg.cpu_set_size = sizeof(desc->cpu_set);
 		cpu_set_arg.nr_processes = nr_processes;
+		cpu_set_arg.ppid = getppid();
 		cpu_set_arg.target_core = &target_core;
 		cpu_set_arg.process_rank = &process_rank;
 		cpu_set_arg.mcexec_linux_numa = &mcexec_linux_numa;
@@ -2493,6 +2554,16 @@ int main(int argc, char **argv)
 		cpu_set_arg.mcexec_cpu_set_size = sizeof(mcexec_cpu_set);
 		cpu_set_arg.ikc_mapped = &ikc_mapped;
 
+		/* Fugaku specific: Fujitsu CPU binding */
+		if (getenv("FLIB_AFFINITY_ON_PROCESS")) {
+			cpu_set_arg.req_cpu_list =
+				getenv("FLIB_AFFINITY_ON_PROCESS");
+			cpu_set_arg.req_cpu_list_len =
+				strlen(cpu_set_arg.req_cpu_list) + 1;
+			__dprintf("%s: requesting CPUs: %s\n",
+				__func__, cpu_set_arg.req_cpu_list);
+		}
+
 		if (ioctl(fd, MCEXEC_UP_GET_CPUSET, (void *)&cpu_set_arg) != 0) {
 			perror("getting CPU set for partitioned execution");
 			close(fd);
@@ -2501,6 +2572,12 @@ int main(int argc, char **argv)
 
 		desc->cpu = target_core;
 		desc->process_rank = process_rank;
+		/* Fugaku specific: Fujitsu node-local rank */
+		if (getenv("PLE_RANK_ON_NODE")) {
+			desc->process_rank = atoi(getenv("PLE_RANK_ON_NODE"));
+			__dprintf("%s: rank: %d, target CPU: %d\n",
+				__func__, desc->process_rank, desc->cpu);
+		}
 
 		/* Bind to CPU cores where the LWK process' IKC target maps to */
 		if (ikc_mapped && !no_bind_ikc_map) {

kernel/syscall.c

@@ -242,11 +242,6 @@ long do_syscall(struct syscall_request *req, int cpu)
 			unsigned long flags;
 			DECLARE_WAITQ_ENTRY(scd_wq_entry, cpu_local_var(current));
 
-			if (req->number == __NR_epoll_wait ||
-				req->number == __NR_epoll_pwait ||
-				req->number == __NR_ppoll)
-				goto schedule;
-
 			cpu_pause();
 
 			/* Spin if not preemptable */
@@ -275,7 +270,6 @@ long do_syscall(struct syscall_request *req, int cpu)
 				continue;
 			}
 
-schedule:
 			flags = cpu_disable_interrupt_save();
 
 			/* Try to sleep until notified */
@@ -10320,11 +10314,7 @@ long syscall(int num, ihk_mc_user_context_t *ctx)
 	}
 #endif // PROFILE_ENABLE
 
-	/* Do not deschedule when returning from an event (e.g., MPI) */
-	if (!(num == __NR_epoll_wait ||
-				num == __NR_epoll_pwait ||
-				num == __NR_ppoll) &&
-			smp_load_acquire(&v->flags) & CPU_FLAG_NEED_RESCHED) {
+	if (smp_load_acquire(&v->flags) & CPU_FLAG_NEED_RESCHED) {
 		check_need_resched();
 	}
 

test/issues/929/C929.sh

@@ -0,0 +1,145 @@
+#/bin/sh
+
+USELTP=0
+USEOSTEST=0
+
+. ../../common.sh
+
+issue="929"
+tid=01
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+TEST_CMD="mpirun -f ./hostfile -ppn 5 ${MCEXEC} -n 5 ./test_prog.sh"
+echo ${TEST_CMD}
+${TEST_CMD} &> ${tname}.txt
+mpi_ret=$?
+
+cat ./${tname}.txt
+started_num=`grep 'test_prog is started' ./${tname}.txt | wc -l`
+
+if [ ${mpi_ret} -eq 0 -a ${started_num} -eq 5 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+TEST_CMD="mpirun -f ./hostfile -ppn 5 ${MCEXEC} -n 3 ./test_prog.sh"
+echo ${TEST_CMD}
+${TEST_CMD} &> ${tname}.txt
+mpi_ret=$?
+
+cat ./${tname}.txt
+started_num=`grep 'test_prog is started' ./${tname}.txt | wc -l`
+
+if [ ${mpi_ret} -ne 0 -a ${started_num} -eq 3 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+TEST_CMD="mpirun -f ./hostfile -ppn 3 ${MCEXEC} -n 5 ./test_prog.sh"
+echo ${TEST_CMD}
+${TEST_CMD} &> ${tname}.txt
+mpi_ret=$?
+
+cat ./${tname}.txt
+started_num=`grep 'test_prog is started' ./${tname}.txt | wc -l`
+
+if [ ${mpi_ret} -ne 0 -a ${started_num} -eq 0 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+TEST_CMD="mpirun -f ./hostfile -ppn 6 ${MCEXEC} -n 3 ./test_prog.sh"
+echo ${TEST_CMD}
+${TEST_CMD} &> ${tname}.txt
+mpi_ret=$?
+
+cat ./${tname}.txt
+started_num=`grep 'test_prog is started' ./${tname}.txt | wc -l`
+
+if [ ${mpi_ret} -ne 0 -a ${started_num} -eq 3 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+TEST_CMD="mpirun -f ./hostfile -ppn 250 ${MCEXEC} -n 250 ./test_prog.sh"
+echo ${TEST_CMD}
+${TEST_CMD} &> ${tname}.txt
+mpi_ret=$?
+
+head -n 10 ./${tname}.txt
+echo "..."
+started_num=`grep 'test_prog is started' ./${tname}.txt | wc -l`
+
+if [ ${mpi_ret} -ne 0 -a ${started_num} -eq 0 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+ng=0
+TEST_CMD="mpirun -f ./hostfile -ppn 5 ${MCEXEC} -n 5 ./test_prog.sh"
+echo "** reboot mcrernel for check pe_list_len"
+mcreboot
+echo "** enable debug message in mcexec_get_cpuset"
+sudo sh -c "echo -n 'func mcexec_get_cpuset +p' > /sys/kernel/debug/dynamic_debug/control"
+echo ${TEST_CMD}
+for i in `seq 1 20`
+do
+	${TEST_CMD} &> ${tname}.txt
+	mpi_ret=$?
+	started_num=`grep 'test_prog is started' ./${tname}.txt | wc -l`
+	if [ ${mpi_ret} -eq 0 -a ${started_num} -eq 5 ]; then
+		echo "[OK] exec: $i"
+	else
+		echo "[NG] exec: $i"
+		let ng++
+	fi
+done
+echo "** check pe_list_len"
+dmesg --notime | grep "mcexec_get_cpuset: pe_list" | tail -n 20 | cut -f 2-3 -d ':' > ./pe_list_len.txt
+cat ./pe_list_len.txt | while read line
+do
+	len=`echo ${line} | cut -f 2 -d ':'`
+	if [ ${len} -ge 0 -a ${len} -le 5 ]; then
+		echo "[OK] ${line}"
+	else
+		echo "[NG] ${line}"
+		let ng++
+	fi
+done
+echo "** disable debug message in mcexec_get_cpuset"
+sudo sh -c "echo -n 'func mcexec_get_cpuset -p' > /sys/kernel/debug/dynamic_debug/control"
+
+if [ ${ng} -eq 0 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+

test/issues/929/Makefile

@@ -0,0 +1,11 @@
+CFLAGS=-g
+LDFLAGS=
+
+TARGET=
+
+all: $(TARGET)
+
+test: all
+	./C929.sh
+clean:
+	rm -f $(TARGET) *.o *.txt

test/issues/929/README

@@ -0,0 +1,36 @@
+【Issue#929 動作確認】
+□ テスト内容
+1. mpirunで指定する-ppnと、mcexecで指定する-n の指定状況ごとに
+  想定どおりの動作となることを確認
+C929T01:
+  -ppn == -n の場合に、プログラムが実行され、mpirunが成功する
+
+C929T02:
+  -ppn > -n の場合に、プログラムの一部が実行され、mpirunが失敗する
+
+C929T03:
+  -ppn < -n の場合に、プログラムが実行されず、mpirunが失敗する
+
+C929T04:
+  -ppn が -n の整数倍である場合に、プログラムの一部が実行され、mpirunが失敗する
+
+C929T05:
+  -ppn と -n がMcKernelに割り当てたCPU数よりも大きい場合に、
+  プログラムが実行されず、mpirunが失敗する
+
+C929T06:
+  -ppn == -n での正常実行を20回連続で行った場合に、
+  プログラムが実行され、mpirunが成功する
+  また、mcctrlで管理しているpart_exec_list の要素数が5を超えない
+
+□ 実行手順
+$ make test
+
+McKernelのインストール先や、OSTEST, LTPの配置場所は、
+$HOME/.mck_test_config を参照している
+.mck_test_config は、McKernelをビルドした際に生成されるmck_test_config.sample ファイルを
+$HOMEにコピーし、適宜編集する
+
+□ 実行結果
+x86_64_result.log aarch64_result.log 参照。
+すべての項目をPASSしていることを確認。

test/issues/929/aarch64_result.log

@@ -0,0 +1,99 @@
+*** C929T01 start *******************************
+mpirun -f ./hostfile -ppn 5 /home/satoken/ihk+mckernel/bin/mcexec -n 5 ./test_prog.sh
+test_prog is started.
+test_prog is started.
+test_prog is started.
+test_prog is started.
+test_prog is started.
+*** C929T01 PASSED ******************************
+
+*** C929T02 start *******************************
+mpirun -f ./hostfile -ppn 5 /home/satoken/ihk+mckernel/bin/mcexec -n 3 ./test_prog.sh
+getting CPU set for partitioned execution: Invalid argument
+getting CPU set for partitioned execution: Invalid argument
+test_prog is started.
+test_prog is started.
+test_prog is started.
+*** C929T02 PASSED ******************************
+
+*** C929T03 start *******************************
+mpirun -f ./hostfile -ppn 3 /home/satoken/ihk+mckernel/bin/mcexec -n 5 ./test_prog.sh
+getting CPU set for partitioned execution: Connection timed out
+getting CPU set for partitioned execution: Connection timed out
+getting CPU set for partitioned execution: Connection timed out
+*** C929T03 PASSED ******************************
+
+*** C929T04 start *******************************
+mpirun -f ./hostfile -ppn 6 /home/satoken/ihk+mckernel/bin/mcexec -n 3 ./test_prog.sh
+getting CPU set for partitioned execution: Invalid argument
+getting CPU set for partitioned execution: Invalid argument
+getting CPU set for partitioned execution: Invalid argument
+test_prog is started.
+test_prog is started.
+test_prog is started.
+*** C929T04 PASSED ******************************
+
+*** C929T05 start *******************************
+mpirun -f ./hostfile -ppn 250 /home/satoken/ihk+mckernel/bin/mcexec -n 250 ./test_prog.sh
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+...
+*** C929T05 PASSED ******************************
+
+*** C929T06 start *******************************
+** reboot mcrernel for check pe_list_len
+mcreboot.sh -c 1-6,29-34 -m 50G@0,50G@1 -r 1-6:0+29-34:28 -O ... done
+** enable debug message in mcexec_get_cpuset
+mpirun -f ./hostfile -ppn 5 /home/satoken/ihk+mckernel/bin/mcexec -n 5 ./test_prog.sh
+[OK] exec: 1
+[OK] exec: 2
+[OK] exec: 3
+[OK] exec: 4
+[OK] exec: 5
+[OK] exec: 6
+[OK] exec: 7
+[OK] exec: 8
+[OK] exec: 9
+[OK] exec: 10
+[OK] exec: 11
+[OK] exec: 12
+[OK] exec: 13
+[OK] exec: 14
+[OK] exec: 15
+[OK] exec: 16
+[OK] exec: 17
+[OK] exec: 18
+[OK] exec: 19
+[OK] exec: 20
+** check pe_list_len
+[OK] pe_list_len:0
+[OK] pe_list_len:1
+[OK] pe_list_len:2
+[OK] pe_list_len:3
+[OK] pe_list_len:4
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+[OK] pe_list_len:5
+** disable debug message in mcexec_get_cpuset
+*** C929T06 PASSED ******************************
+

test/issues/929/hostfile

@@ -0,0 +1 @@
+localhost

test/issues/929/test_prog.sh

@@ -0,0 +1,3 @@
+#!/bin/sh
+
+echo "test_prog is started."

test/issues/929/x86_64_result.log

@@ -0,0 +1,74 @@
+*** C929T01 start *******************************
+mpirun -f ./hostfile -ppn 5 /home/satoken/ihk+mckernel/bin/mcexec -n 5 ./test_prog.sh
+test_prog is started.
+test_prog is started.
+test_prog is started.
+test_prog is started.
+test_prog is started.
+*** C929T01 PASSED ******************************
+
+*** C929T02 start *******************************
+mpirun -f ./hostfile -ppn 5 /home/satoken/ihk+mckernel/bin/mcexec -n 3 ./test_prog.sh
+getting CPU set for partitioned execution: Invalid argument
+getting CPU set for partitioned execution: Invalid argument
+test_prog is started.
+test_prog is started.
+test_prog is started.
+*** C929T02 PASSED ******************************
+
+*** C929T03 start *******************************
+mpirun -f ./hostfile -ppn 3 /home/satoken/ihk+mckernel/bin/mcexec -n 5 ./test_prog.sh
+getting CPU set for partitioned execution: Connection timed out
+getting CPU set for partitioned execution: Connection timed out
+getting CPU set for partitioned execution: Connection timed out
+*** C929T03 PASSED ******************************
+
+*** C929T04 start *******************************
+mpirun -f ./hostfile -ppn 6 /home/satoken/ihk+mckernel/bin/mcexec -n 3 ./test_prog.sh
+getting CPU set for partitioned execution: Invalid argument
+getting CPU set for partitioned execution: Invalid argument
+getting CPU set for partitioned execution: Invalid argument
+test_prog is started.
+test_prog is started.
+test_prog is started.
+*** C929T04 PASSED ******************************
+
+*** C929T05 start *******************************
+mpirun -f ./hostfile -ppn 250 /home/satoken/ihk+mckernel/bin/mcexec -n 250 ./test_prog.sh
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+error: nr_processes can't exceed nr. of CPUs
+...
+*** C929T05 PASSED ******************************
+
+*** C929T06 start *******************************
+mpirun -f ./hostfile -ppn 5 /home/satoken/ihk+mckernel/bin/mcexec -n 5 ./test_prog.sh
+[OK] exec: 1
+[OK] exec: 2
+[OK] exec: 3
+[OK] exec: 4
+[OK] exec: 5
+[OK] exec: 6
+[OK] exec: 7
+[OK] exec: 8
+[OK] exec: 9
+[OK] exec: 10
+[OK] exec: 11
+[OK] exec: 12
+[OK] exec: 13
+[OK] exec: 14
+[OK] exec: 15
+[OK] exec: 16
+[OK] exec: 17
+[OK] exec: 18
+[OK] exec: 19
+[OK] exec: 20
+*** C929T06 PASSED ******************************
+