From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:55075) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1gITFT-0003SK-MB for qemu-devel@nongnu.org; Fri, 02 Nov 2018 02:48:05 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1gITFR-0002s0-OJ for qemu-devel@nongnu.org; Fri, 02 Nov 2018 02:48:03 -0400 Date: Fri, 2 Nov 2018 14:47:47 +0800 From: Fam Zheng Message-ID: <20181102064747.GA21032@magic> References: <20181101103807.25862-1-lifeng1519@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <20181101103807.25862-1-lifeng1519@gmail.com> Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH] block/nvme: optimize the performance of nvme driver based on vfio-pci List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Li Feng Cc: fengli@smartx.com, Kevin Wolf , Max Reitz , "open list:NVMe Block Driver" , "open list:All patches CC here" On Thu, 11/01 18:38, Li Feng wrote: > When the IO size is larger than 2 pages, we move the the pointer one by > one in the pagelist, this is inefficient. >=20 > This is a simple benchmark result: >=20 > Before: > $ qemu-io -c 'write 0 1G' nvme://0000:00:04.0/1 >=20 > wrote 1073741824/1073741824 bytes at offset 0 > 1 GiB, 1 ops; 0:00:02.41 (424.504 MiB/sec and 0.4146 ops/sec) >=20 > $ qemu-io -c 'read 0 1G' nvme://0000:00:04.0/1 >=20 > read 1073741824/1073741824 bytes at offset 0 > 1 GiB, 1 ops; 0:00:02.03 (503.055 MiB/sec and 0.4913 ops/sec) >=20 > After: > $ qemu-io -c 'write 0 1G' nvme://0000:00:04.0/1 >=20 > wrote 1073741824/1073741824 bytes at offset 0 > 1 GiB, 1 ops; 0:00:02.17 (471.517 MiB/sec and 0.4605 ops/sec) >=20 > $ qemu-io -c 'read 0 1G' nvme://0000:00:04.0/1 = = = = 1 =E2=86=B5 >=20 > read 1073741824/1073741824 bytes at offset 0 > 1 GiB, 1 ops; 0:00:01.94 (526.770 MiB/sec and 0.5144 ops/sec) >=20 > Signed-off-by: Li Feng > --- > block/nvme.c | 16 ++++++---------- > 1 file changed, 6 insertions(+), 10 deletions(-) >=20 > diff --git a/block/nvme.c b/block/nvme.c > index 29294038fc..982097b5b1 100644 > --- a/block/nvme.c > +++ b/block/nvme.c > @@ -837,7 +837,7 @@ try_map: > } > =20 > for (j =3D 0; j < qiov->iov[i].iov_len / s->page_size; j++) { > - pagelist[entries++] =3D iova + j * s->page_size; > + pagelist[entries++] =3D cpu_to_le64(iova + j * s->page_siz= e); > } > trace_nvme_cmd_map_qiov_iov(s, i, qiov->iov[i].iov_base, > qiov->iov[i].iov_len / s->page_siz= e); > @@ -850,20 +850,16 @@ try_map: > case 0: > abort(); > case 1: > - cmd->prp1 =3D cpu_to_le64(pagelist[0]); > + cmd->prp1 =3D pagelist[0]; > cmd->prp2 =3D 0; > break; > case 2: > - cmd->prp1 =3D cpu_to_le64(pagelist[0]); > - cmd->prp2 =3D cpu_to_le64(pagelist[1]);; > + cmd->prp1 =3D pagelist[0]; > + cmd->prp2 =3D pagelist[1]; > break; > default: > - cmd->prp1 =3D cpu_to_le64(pagelist[0]); > - cmd->prp2 =3D cpu_to_le64(req->prp_list_iova); > - for (i =3D 0; i < entries - 1; ++i) { > - pagelist[i] =3D cpu_to_le64(pagelist[i + 1]); > - } > - pagelist[entries - 1] =3D 0; > + cmd->prp1 =3D pagelist[0]; > + cmd->prp2 =3D cpu_to_le64(req->prp_list_iova + sizeof(uint64_t= )); > break; > } > trace_nvme_cmd_map_qiov(s, cmd, req, qiov, entries); > --=20 > 2.11.0 >=20 Nice! Thanks. I've queued the patch. Fam