From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from cn.fujitsu.com ([59.151.112.132]:1385 "EHLO heian.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1752774AbcKVJ1n (ORCPT ); Tue, 22 Nov 2016 04:27:43 -0500 Subject: Re: [PATCH] btrfs: limit the number of asynchronous delalloc pages to reasonable value To: , , References: <1478597458-29287-1-git-send-email-wangxg.fnst@cn.fujitsu.com> <20161121163941.GJ12522@twin.jikos.cz> From: Wang Xiaoguang Message-ID: <58340DCA.7050107@cn.fujitsu.com> Date: Tue, 22 Nov 2016 17:20:10 +0800 MIME-Version: 1.0 In-Reply-To: <20161121163941.GJ12522@twin.jikos.cz> Content-Type: text/plain; charset="windows-1252"; format=flowed Sender: linux-btrfs-owner@vger.kernel.org List-ID: hello, On 11/22/2016 12:39 AM, David Sterba wrote: > On Tue, Nov 08, 2016 at 05:30:58PM +0800, Wang Xiaoguang wrote: >> The current limit of number of asynchronous delalloc pages is (10 * SZ_1M). >> For 4K page, the total ram bytes would be 40G, very big value, I think in >> most cases, this limit will not work, here I set limit of the number of >> asynchronous delalloc pages to SZ_1M(4GB ram bytes). >> >> Signed-off-by: Wang Xiaoguang >> --- >> fs/btrfs/inode.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c >> index 8e3a5a2..3a910f6 100644 >> --- a/fs/btrfs/inode.c >> +++ b/fs/btrfs/inode.c >> @@ -1158,7 +1158,7 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page, >> struct btrfs_root *root = BTRFS_I(inode)->root; >> unsigned long nr_pages; >> u64 cur_end; >> - int limit = 10 * SZ_1M; >> + int limit = SZ_1M; > This looks like mismatch in units, the limit seems to be "10MiB" but > it's used to compare async_delalloc_pages, which is in pages. I'm not > sure what's the expected value for limit, as 10MiB is too small. There are such codes in cow_file_range_async() nr_pages = (cur_end - start + PAGE_SIZE) >> PAGE_SHIFT; atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); So here the real limit is 10 * SZ_1M * 4096, 40GB memory, I think this value is too big, so modify it to 4GB. Regards, Xiaoguang Wang > >