From 93581cb142b4841dd94064c29383f77bb90e1cb7 Mon Sep 17 00:00:00 2001 From: "Shiratori, Takehiro" Date: Thu, 5 Mar 2020 15:57:57 +0900 Subject: [PATCH] test: runq_lock and over-scheduling fix. Change-Id: I236ab585403076d716be350c8b51e8d352122f2b Refs: #1400 --- test/issues/1400/arm64/1400.sh | 68 ++++++++ test/issues/1400/arm64/1400_arm64.c | 238 ++++++++++++++++++++++++++++ test/issues/1400/arm64/README | 57 +++++++ test/issues/1400/arm64/ltp_list.txt | 23 +++ 4 files changed, 386 insertions(+) create mode 100755 test/issues/1400/arm64/1400.sh create mode 100644 test/issues/1400/arm64/1400_arm64.c create mode 100644 test/issues/1400/arm64/README create mode 100644 test/issues/1400/arm64/ltp_list.txt diff --git a/test/issues/1400/arm64/1400.sh b/test/issues/1400/arm64/1400.sh new file mode 100755 index 00000000..4b568f97 --- /dev/null +++ b/test/issues/1400/arm64/1400.sh @@ -0,0 +1,68 @@ +#!/bin/sh +# 1400.sh COPYRIGHT FUJITSU LIMITED 2020 + +. $HOME/.mck_test_config + +BOOTPARAM="-c 12-15 -m 1G@4 -O" +USELTP=1 + +. ../../../common.sh + +PWD=`pwd` +LOGFILE="${PWD}/1400_log" +STOPFILE="./1400_stop" +LTPLIST="${PWD}/ltp_list.txt" + +echo "issue-1400 test run." +echo "start-time: `date`" +stime=`date "+%s"` +failed=0 + +while : +do + ${MCEXEC} ./1400_arm64 > ${LOGFILE} 2>&1 + if [ $? -ne 0 ]; then + echo "1400_arm64 failed." + failed=1 + break + fi + + pushd ${LTPBIN} > /dev/null + while read line + do + ${MCEXEC} ./${line} > ${LOGFILE} 2>&1 + if [ $? -ne 0 ]; then + echo "${line} failed." + failed=1 + break + fi + done < ${LTPLIST} + popd > /dev/null + + if [ -e ${STOPFILE} ]; then + rm -f ${STOPFILE} + break + fi + + if [ ${failed} -eq 1 ]; then + break + fi +done + +echo "end-time: `date`" +etime=`date "+%s"` + +run_time=$((${etime} - ${stime})) + +if [ 43200 -le ${run_time} ]; then + if [ ${failed} -eq 0 ]; then + echo "issue-1400 test OK." + rm -f ${LOGFILE} + else + echo "issue-1400 test NG." + fi +else + echo "issue-1400 test NG." +fi + +mcstop diff --git a/test/issues/1400/arm64/1400_arm64.c b/test/issues/1400/arm64/1400_arm64.c new file mode 100644 index 00000000..c305160a --- /dev/null +++ b/test/issues/1400/arm64/1400_arm64.c @@ -0,0 +1,238 @@ +/* 1400_arm64.c COPYRIGHT FUJITSU LIMITED 2020 */ +#define _GNU_SOURCE +#include +#include +#include +#include +#include +#include +#include + +#define POINT_ORDER_NUM 2 + +static int *sync1 = MAP_FAILED; +static int *parent_core = MAP_FAILED; +static int *point_order = MAP_FAILED; +static int *od = MAP_FAILED; + +int main(int argc, char *argv[]) +{ + pid_t pid = -1; + pid_t ret_pid = -1; + int status = 0; + int i = 0; + int result = -1; + int ret = -1; + int failed = 0; + + /* get shared memory */ + sync1 = (int *)mmap(NULL, sizeof(int), PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_ANONYMOUS, -1, 0); + parent_core = (int *)mmap(NULL, sizeof(int), PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_ANONYMOUS, -1, 0); + point_order = (int *)mmap(NULL, sizeof(int) * POINT_ORDER_NUM, + PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_ANONYMOUS, -1, 0); + od = (int *)mmap(NULL, sizeof(int), PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_ANONYMOUS, -1, 0); + + /* mmap check */ + if (sync1 == MAP_FAILED || + parent_core == MAP_FAILED || + point_order == MAP_FAILED || + od == MAP_FAILED) { + printf("mmap() Failed.\n"); + goto out; + } + + for (i = 0; i < POINT_ORDER_NUM; i++) { + point_order[i] = 0; + } + *od = 0; + *sync1 = 0; + + /* create child process */ + pid = fork(); + + switch (pid) { + case -1: + /* error */ + printf("fork() Failed.\n"); + goto out; + + case 0: { + /* child */ + /* before migrate, get cpunum */ + int old_mycore = sched_getcpu(); + + printf("[child:%d] running core %d\n", getpid(), old_mycore); + + /* sync parent */ + *sync1 = 1; + + /* wait until migrated */ + while (sched_getcpu() == old_mycore) { + __asm__ __volatile__("yield" ::: "memory"); + } + point_order[0] = ++(*od); + + _exit(0); + break; + } + + default: { + /* parent */ + cpu_set_t cpuset; + + /* sync child */ + while (*sync1 != 1) { + __asm__ __volatile__("yield" ::: "memory"); + } + + /* parent corenum get */ + *parent_core = sched_getcpu(); + + /* child process to migrate parent core */ + printf("[parent:%d] running core %d\n", getpid(), *parent_core); + printf("[parent] child process(pid=%d) " + "migrate/bind to core %d\n", + pid, *parent_core); + + CPU_ZERO(&cpuset); + CPU_SET(*parent_core, &cpuset); + + result = sched_setaffinity(pid, sizeof(cpuset), &cpuset); + if (result == -1) { + printf("errno = %d\n", errno); + printf("child migrate/bind " + "sched_setaffinity failed.\n"); + } + + /* parent core bind */ + printf("[parent] parent process bind to core %d\n", + *parent_core); + result = sched_setaffinity(0, sizeof(cpuset), &cpuset); + if (result == -1) { + printf("errno = %d\n", errno); + printf("parent bind sched_setaffinity failed.\n"); + } + + /* sched_setaffinity interval */ + usleep(10000); + + /* sync child, switch to child process */ + printf("[parent] send sched_yield.\n"); + + result = 0; + + result = sched_yield(); + + point_order[1] = ++(*od); + + break; + } + } + + if (result == -1) { + printf("sched_yield failed.\n"); + } + + /* child process status check. */ + ret_pid = wait(&status); + if (ret_pid == pid) { + if (WIFEXITED(status)) { + if (WEXITSTATUS(status)) { + printf("TP failed, child migrate fail.\n"); + } + else { + goto wait_ok; + } + } + else { + printf("TP failed, child is not exited.\n"); + } + + if (WIFSIGNALED(status)) { + printf("TP failed, child signaled by %d.\n", + WTERMSIG(status)); + + if (WCOREDUMP(status)) { + printf("coredumped.\n"); + } + } + else { + printf("TP failed, child is not signaled.\n"); + } + + + if (WIFSTOPPED(status)) { + printf("TP failed, child is stopped by signal %d.\n", + WSTOPSIG(status)); + } + else { + printf("TP failed, child is not stopped.\n"); + } + + if (WIFCONTINUED(status)) { + printf("TP failed, child is continued.\n"); + } + else { + printf("TP failed, child is not continued.\n"); + } + + for (i = 0; i < POINT_ORDER_NUM; i++) { + printf("point_order[%d] = %d\n", i, point_order[i]); + } + goto out; + } + else { + printf("TP failed, child process wait() fail.\n"); + + for (i = 0; i < POINT_ORDER_NUM; i++) { + printf("point_order[%d] = %d\n", i, point_order[i]); + } + goto out; + } + +wait_ok: + for (i = 0; i < POINT_ORDER_NUM; i++) { + printf("point_order[%d] = %d\n", i, point_order[i]); + + if (point_order[i] == 0) { + failed = 1; + } + } + + if (failed != 0) { + printf("TP failed, parent or child process is not running.\n"); + goto out; + } + + if (result != -1) { + if (point_order[0] < point_order[1]) { + ret = 0; + } + else { + printf("TP failed, out of order.\n"); + } + } + +out: + /* unmap semaphore memory */ + if (od != MAP_FAILED) { + munmap(od, sizeof(int)); + } + + if (point_order != MAP_FAILED) { + munmap(point_order, sizeof(int) * POINT_ORDER_NUM); + } + + if (parent_core != MAP_FAILED) { + munmap(parent_core, sizeof(int)); + } + + if (sync1 != MAP_FAILED) { + munmap(sync1, sizeof(int)); + } + return ret; +} diff --git a/test/issues/1400/arm64/README b/test/issues/1400/arm64/README new file mode 100644 index 00000000..7c86bfe7 --- /dev/null +++ b/test/issues/1400/arm64/README @@ -0,0 +1,57 @@ +/* README COPYRIGHT FUJITSU LIMITED 2020 */ + +ostest-sched_yield.000: 親プロセスが子プロセスと自信を同一CPUにバインドしてsched_yield()しても子プロセスがスケジュールされない +https://postpeta.pccluster.org/redmine/issues/1400 + +テストセットREADME + +(0) 事前準備 + 1. $HOME/.mck_test_configを用意する + +(1) テスト実行方法 + 1. make test + +(2) テスト終了方法 + 1. touch ./1400_stop + + テスト項目全項目を1セットとして行うため、終了までにタイムラグがあることがある。 + +(3) テスト項目詳細 + + 1400_arm64.c + 親プロセスが子プロセスと自身を同一CPUにバインドしてsched_yield()した場合、 + 子プロセス、親プロセスの順序で実行されることを確認する。 + + 本修正が既存機能へのレベルダウンとなっていないか確認するため、 + 下記のLTP項目についても実施対象とする。 + - sched_yield01 + - signal01,02,03,04,05 + - rt_sigaction01,02,03 + - rt_sigprocmask01,02 + - rt_sigsuspend01 + - rt_tgsigqueueinfo01 + - futex_wait01,02,03,04 + - futex_wake01,03 + - futex_wait_bitset01,02 + - execveat02 + + 現象はタイミングによって発生有無が変化するため、テストは無限ループによる + 長時間実行を行う。 + 12時間以上の実行継続、テスト成功を目安とし、合否判定とする。 + + なお、本テストの実行は別障害のため、NGとなることがある。 + 別障害は https://postpeta.pccluster.org/redmine/issues/1454 で管理。 + + + また、本修正により明らかになった上記とは別の障害により以下のLTP項目が +　FAILになる。 + - msgrcv05 + - msgsnd05 + - semctl01 + - semop05 + + こちらの別障害については、 + https://postpeta.pccluster.org/redmine/issues/1455 + にて管理する。 + +以上。 diff --git a/test/issues/1400/arm64/ltp_list.txt b/test/issues/1400/arm64/ltp_list.txt new file mode 100644 index 00000000..7bdbe7aa --- /dev/null +++ b/test/issues/1400/arm64/ltp_list.txt @@ -0,0 +1,23 @@ +sched_yield01 +signal01 +signal02 +signal03 +signal04 +signal05 +rt_sigaction01 +rt_sigaction02 +rt_sigaction03 +rt_sigprocmask01 +rt_sigprocmask02 +rt_sigsuspend01 +rt_tgsigqueueinfo01 +futex_wait01 +futex_wait02 +futex_wait03 +futex_wait04 +futex_wake01 +futex_wake03 +futex_wait_bitset01 +futex_wait_bitset02 +execveat02 +sigsuspend01