From patchwork Thu Jul 5 10:52:30 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Peter Lieven X-Patchwork-Id: 939843 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (mailfrom) smtp.mailfrom=nongnu.org (client-ip=2001:4830:134:3::11; helo=lists.gnu.org; envelope-from=qemu-devel-bounces+incoming=patchwork.ozlabs.org@nongnu.org; receiver=) Authentication-Results: ozlabs.org; dmarc=none (p=none dis=none) header.from=kamp.de Received: from lists.gnu.org (lists.gnu.org [IPv6:2001:4830:134:3::11]) (using TLSv1 with cipher AES256-SHA (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 41Lvpg5qC2z9s29 for ; Thu, 5 Jul 2018 20:53:14 +1000 (AEST) Received: from localhost ([::1]:51784 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1fb1st-0006aB-6W for incoming@patchwork.ozlabs.org; Thu, 05 Jul 2018 06:53:11 -0400 Received: from eggs.gnu.org ([2001:4830:134:3::10]:35625) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1fb1sW-0006a1-KO for qemu-devel@nongnu.org; Thu, 05 Jul 2018 06:52:50 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1fb1sT-0003pW-FT for qemu-devel@nongnu.org; Thu, 05 Jul 2018 06:52:48 -0400 Received: from mx-v6.kamp.de ([2a02:248:0:51::16]:58514 helo=mx01.kamp.de) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1fb1sT-0003dd-6s for qemu-devel@nongnu.org; Thu, 05 Jul 2018 06:52:45 -0400 Received: (qmail 28628 invoked by uid 89); 5 Jul 2018 10:52:37 -0000 Received: from [195.62.97.192] by client-16-kamp (envelope-from , uid 89) with qmail-scanner-2010/03/19-MF (clamdscan: 0.100.0/24724. avast: 1.2.2/17010300. spamassassin: 3.4.1. Clear:RC:1(195.62.97.192):. Processed in 0.11218 secs); 05 Jul 2018 10:52:37 -0000 Received: from kerio.kamp.de ([195.62.97.192]) by mx01.kamp.de with ESMTPS (DHE-RSA-AES256-SHA encrypted); 5 Jul 2018 10:52:36 -0000 X-GL_Whitelist: yes X-Footer: a2FtcC5kZQ== Received: from submission.kamp.de ([195.62.97.28]) by kerio.kamp.de with ESMTPS (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256 bits)) for qemu-devel@nongnu.org; Thu, 5 Jul 2018 12:52:35 +0200 Received: (qmail 1867 invoked from network); 5 Jul 2018 10:52:36 -0000 Received: from lieven-vm.kamp-intra.net (HELO lieven-vm-neu) (relay@kamp.de@::ffff:172.21.12.69) by submission.kamp.de with ESMTPS (DHE-RSA-AES256-GCM-SHA384 encrypted) ESMTPA; 5 Jul 2018 10:52:36 -0000 Received: by lieven-vm-neu (Postfix, from userid 1060) id 0D61E2025C; Thu, 5 Jul 2018 12:52:36 +0200 (CEST) From: Peter Lieven To: qemu-devel@nongnu.org, qemu-block@nongnu.org Date: Thu, 5 Jul 2018 12:52:30 +0200 Message-Id: <1530787950-25306-1-git-send-email-pl@kamp.de> X-Mailer: git-send-email 1.9.1 X-detected-operating-system: by eggs.gnu.org: GNU/Linux 2.2.x-3.x [generic] X-Received-From: 2a02:248:0:51::16 Subject: [Qemu-devel] [PATCH V3] qemu-img: align result of is_allocated_sectors X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: kwolf@redhat.com, Peter Lieven , mreitz@redhat.com Errors-To: qemu-devel-bounces+incoming=patchwork.ozlabs.org@nongnu.org Sender: "Qemu-devel" We currently don't enforce that the sparse segments we detect during convert are aligned. This leads to unnecessary and costly read-modify-write cycles either internally in Qemu or in the background on the storage device as nearly all modern filesystems or hardware have a 4k alignment internally. The number of RMW cycles when converting an example image [1] to a raw device that has 4k sector size is about 4600 4k read requests to perform a total of about 15000 write requests. With this path the additional 4600 read requests are eliminated. [1] https://cloud-images.ubuntu.com/releases/16.04/release/ubuntu-16.04-server-cloudimg-amd64-disk1.vmdk Signed-off-by: Peter Lieven --- V2->V3: - ensure that s.alignment is a power of 2 - correctly handle n < alignment in is_allocated_sectors if sector_num % alignment > 0. V1->V2: - take the current sector offset into account [Max] - try to figure out the target alignment [Max] qemu-img.c | 46 ++++++++++++++++++++++++++++++++++++---------- 1 file changed, 36 insertions(+), 10 deletions(-) diff --git a/qemu-img.c b/qemu-img.c index e1a506f..db91b9e 100644 --- a/qemu-img.c +++ b/qemu-img.c @@ -1105,8 +1105,11 @@ static int64_t find_nonzero(const uint8_t *buf, int64_t n) * * 'pnum' is set to the number of sectors (including and immediately following * the first one) that are known to be in the same allocated/unallocated state. + * The function will try to align 'pnum' to the number of sectors specified + * in 'alignment' to avoid unnecassary RMW cycles on modern hardware. */ -static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum) +static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum, + int64_t sector_num, int alignment) { bool is_zero; int i; @@ -1115,14 +1118,26 @@ static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum) *pnum = 0; return 0; } - is_zero = buffer_is_zero(buf, 512); - for(i = 1; i < n; i++) { - buf += 512; - if (is_zero != buffer_is_zero(buf, 512)) { + + if (n % alignment) { + alignment = 1; + } + + if (sector_num % alignment) { + n = ROUND_UP(sector_num, alignment) - sector_num; + alignment = 1; + } + + n /= alignment; + + is_zero = buffer_is_zero(buf, BDRV_SECTOR_SIZE * alignment); + for (i = 1; i < n; i++) { + buf += BDRV_SECTOR_SIZE * alignment; + if (is_zero != buffer_is_zero(buf, BDRV_SECTOR_SIZE * alignment)) { break; } } - *pnum = i; + *pnum = i * alignment; return !is_zero; } @@ -1132,7 +1147,7 @@ static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum) * breaking up write requests for only small sparse areas. */ static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum, - int min) + int min, int64_t sector_num, int alignment) { int ret; int num_checked, num_used; @@ -1141,7 +1156,7 @@ static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum, min = n; } - ret = is_allocated_sectors(buf, n, pnum); + ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment); if (!ret) { return ret; } @@ -1149,13 +1164,15 @@ static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum, num_used = *pnum; buf += BDRV_SECTOR_SIZE * *pnum; n -= *pnum; + sector_num += *pnum; num_checked = num_used; while (n > 0) { - ret = is_allocated_sectors(buf, n, pnum); + ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment); buf += BDRV_SECTOR_SIZE * *pnum; n -= *pnum; + sector_num += *pnum; num_checked += *pnum; if (ret) { num_used = num_checked; @@ -1560,6 +1577,7 @@ typedef struct ImgConvertState { bool wr_in_order; bool copy_range; int min_sparse; + int alignment; size_t cluster_sectors; size_t buf_sectors; long num_coroutines; @@ -1724,7 +1742,8 @@ static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num, * zeroed. */ if (!s->min_sparse || (!s->compressed && - is_allocated_sectors_min(buf, n, &n, s->min_sparse)) || + is_allocated_sectors_min(buf, n, &n, s->min_sparse, + sector_num, s->alignment)) || (s->compressed && !buffer_is_zero(buf, n * BDRV_SECTOR_SIZE))) { @@ -2373,6 +2392,13 @@ static int img_convert(int argc, char **argv) out_bs->bl.pdiscard_alignment >> BDRV_SECTOR_BITS))); + /* try to align the write requests to the destination to avoid unnecessary + * RMW cycles. */ + s.alignment = MAX(pow2floor(s.min_sparse), + DIV_ROUND_UP(out_bs->bl.request_alignment, + BDRV_SECTOR_SIZE)); + assert(is_power_of_2(s.alignment)); + if (skip_create) { int64_t output_sectors = blk_nb_sectors(s.target); if (output_sectors < 0) {