From patchwork Thu Jan 15 07:42:19 2009 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Hisashi Hifumi X-Patchwork-Id: 18599 Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from vger.kernel.org (vger.kernel.org [209.132.176.167]) by ozlabs.org (Postfix) with ESMTP id 048AADE10D for ; Thu, 15 Jan 2009 18:45:14 +1100 (EST) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1762072AbZAOHpM (ORCPT ); Thu, 15 Jan 2009 02:45:12 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1762086AbZAOHpM (ORCPT ); Thu, 15 Jan 2009 02:45:12 -0500 Received: from serv2.oss.ntt.co.jp ([222.151.198.100]:56526 "EHLO serv2.oss.ntt.co.jp" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1761068AbZAOHpJ (ORCPT ); Thu, 15 Jan 2009 02:45:09 -0500 Received: from serv2.oss.ntt.co.jp (localhost [127.0.0.1]) by serv2.oss.ntt.co.jp (Postfix) with ESMTP id C69052480EC; Thu, 15 Jan 2009 16:45:07 +0900 (JST) Received: from serv1.oss.ntt.co.jp (serv1.localdomain [172.19.0.2]) by serv2.oss.ntt.co.jp (Postfix) with ESMTP id A46FC2480E0; Thu, 15 Jan 2009 16:45:07 +0900 (JST) Received: from hifumi-PC.oss.ntt.co.jp (unknown [172.17.1.27]) by serv1.oss.ntt.co.jp (Postfix) with ESMTP id CACB8104167; Thu, 15 Jan 2009 16:45:04 +0900 (JST) Message-Id: <6.0.0.20.2.20090115163853.07056ed0@172.19.0.2> X-Sender: hhifumi@172.19.0.2 X-Mailer: QUALCOMM Windows Eudora Version 6J-Jr3 Date: Thu, 15 Jan 2009 16:42:19 +0900 To: akpm@linux-foundation.org, linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org From: Hisashi Hifumi Subject: [RESEND] [PATCH] lseek: change i_mutex usage. Mime-Version: 1.0 Sender: linux-ext4-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-ext4@vger.kernel.org I changed i_mutex usage on generic_file_llseek. This function is inside i_mutex, but I think there is room for optimization in some cases. When SEEK_END is specified from caller, in this case we should handle inode->i_size so i_mutex is needed. But in other cases such as SEEK_CUR or SEEK_SET, i_mutex is not needed because just changing file->f_pos value without touching i_size. I did some test to measure i_mutex contention. This test do: 1. make an 128MB file. 2. fork 100 processes. repeat 10000000 times lseeking randomly on each process to this file. 3, gauge seconds between start and end of this test. The result was: -2.6.29-rc1 # time ./lseek_test 315 sec real 5m15.407s user 1m19.128s sys 5m38.884s -2.6.29-rc1-patched # time ./lseek_test 13 sec real 0m13.039s user 1m14.730s sys 2m9.633s Hardware environment: CPU　2.4GHz(Quad Core) *4 Memory 64GB This improvement is derived from just removal of lseek's i_mutex contention. There is i_mutex contention not only around lseek, but also fsync or write. So, I think we also can mitigate i_mutex contention between fsync and lseek. Thanks. Signed-off-by: Hisashi Hifumi Following is the test program "lseek_test.c". #include #include #include #include #include #include #include #include #include #include #define NUM 100 #define LEN 4096 #define LOOP 32*1024 int num; void catch_SIGCHLD(int signo) { pid_t child_pid = 0; do { int child_ret; child_pid = waitpid(-1, &child_ret, WNOHANG); if (child_pid > 0) num++; } while (child_pid > 0); } main() { int i, pid; char buf[LEN]; unsigned long offset, filesize; time_t t1, t2; struct sigaction act; memset(buf, 0, LEN); memset(&act, 0, sizeof(act)); act.sa_handler = catch_SIGCHLD; act.sa_flags = SA_NOCLDSTOP|SA_RESTART; sigemptyset(&act.sa_mask); sigaction(SIGCHLD, &act, NULL); filesize = LEN * LOOP; int fd = open("targetfile1",O_RDWR|O_CREAT); /* create a 128MB file */ for(i = 0; i < LOOP; i++) write(fd, buf, LEN); fsync(fd); close(fd); time(&t1); for (i = 0; i < NUM; i++) { pid = fork(); if(pid == 0){ /* child */ int fd = open("targetfile1", O_RDWR); int j; for (j = 0; j < 10000000; j++) { offset = (random() % filesize); lseek(fd, offset, SEEK_SET); } close(fd); exit(0); } } while(num < NUM) sleep(1); time(&t2); printf("%d sec\n",t2-t1); } --- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html diff -Nrup linux-2.6.29-rc1.org/fs/read_write.c linux-2.6.29-rc1.lseek/fs/read_write.c --- linux-2.6.29-rc1.org/fs/read_write.c 2009-01-15 15:37:35.000000000 +0900 +++ linux-2.6.29-rc1.lseek/fs/read_write.c 2009-01-15 16:12:17.000000000 +0900 @@ -89,9 +89,12 @@ loff_t generic_file_llseek(struct file * { loff_t rval; - mutex_lock(&file->f_dentry->d_inode->i_mutex); - rval = generic_file_llseek_unlocked(file, offset, origin); - mutex_unlock(&file->f_dentry->d_inode->i_mutex); + if (origin == SEEK_END) { + mutex_lock(&file->f_dentry->d_inode->i_mutex); + rval = generic_file_llseek_unlocked(file, offset, origin); + mutex_unlock(&file->f_dentry->d_inode->i_mutex); + } else + rval = generic_file_llseek_unlocked(file, offset, origin); return rval; }