From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754036AbcCANof (ORCPT ); Tue, 1 Mar 2016 08:44:35 -0500 Received: from outbound-smtp12.blacknight.com ([46.22.139.17]:41654 "EHLO outbound-smtp12.blacknight.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752688AbcCANod (ORCPT ); Tue, 1 Mar 2016 08:44:33 -0500 Date: Tue, 1 Mar 2016 13:44:28 +0000 From: Mel Gorman To: Andrew Morton Cc: Vlastimil Babka , Rik van Riel , Johannes Weiner , Andrea Arcangeli , Linux-MM , LKML Subject: [PATCH] mm: thp: Set THP defrag by default to madvise and add a stall-free defrag option -fix Message-ID: <20160301134428.GH2854@techsingularity.net> References: <1456503359-4910-1-git-send-email-mgorman@techsingularity.net> <56D58E1E.5090708@suse.cz> <20160301133102.GG2854@techsingularity.net> <56D59AC7.1020104@suse.cz> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: <56D59AC7.1020104@suse.cz> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org The following is a fix to the patch mm-thp-set-thp-defrag-by-default-to-madvise-and-add-a-stall-free-defrag-option.patch based on feedback from Vlastimil Babka. It removes an unnecessary VM_BUG_ON for tidyness, clarifies documentation and adds a check forbidding someone writing "defer" to the enable knob for transparent huge pages. The ack from Vlastimil applies for the original patch plus this fix combined. Signed-off-by: Mel Gorman Acked-by: Vlastimil Babka --- Documentation/vm/transhuge.txt | 2 +- mm/huge_memory.c | 7 ++++--- 2 files changed, 5 insertions(+), 4 deletions(-) diff --git a/Documentation/vm/transhuge.txt b/Documentation/vm/transhuge.txt index a19b173cbc57..1943fe051a36 100644 --- a/Documentation/vm/transhuge.txt +++ b/Documentation/vm/transhuge.txt @@ -129,7 +129,7 @@ available in the near future. It's the responsibility of khugepaged to then install the THP pages later. "madvise" will enter direct reclaim like "always" but only for regions -that are have used madvise(). This is the default behaviour. +that are have used madvise(MADV_HUGEPAGE). This is the default behaviour. "never" should be self-explanatory. diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 206f35f06d83..9161b3a83720 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -246,6 +246,8 @@ static ssize_t triple_flag_store(struct kobject *kobj, { if (!memcmp("defer", buf, min(sizeof("defer")-1, count))) { + if (enabled == deferred) + return -EINVAL; clear_bit(enabled, &transparent_hugepage_flags); clear_bit(req_madv, &transparent_hugepage_flags); set_bit(deferred, &transparent_hugepage_flags); @@ -273,10 +275,9 @@ static ssize_t triple_flag_store(struct kobject *kobj, static ssize_t enabled_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf) { - if (test_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags)) { - VM_BUG_ON(test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flags)); + if (test_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags)) return sprintf(buf, "[always] madvise never\n"); - } else if (test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flags)) + else if (test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flags)) return sprintf(buf, "always [madvise] never\n"); else return sprintf(buf, "always madvise [never]\n");