Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Chuck Lever <chuck.lever@oracle.com>
To:     Mel Gorman <mgorman@techsingularity.net>
CC:     Mel Gorman <mgorman@suse.de>,
        Jesper Dangaard Brouer <brouer@redhat.com>,
        Linux NFS Mailing List <linux-nfs@vger.kernel.org>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>,
        Jakub Kicinski <kuba@kernel.org>
Subject: Re: alloc_pages_bulk()
Thread-Topic: alloc_pages_bulk()
Thread-Index: AQHW/jD1lENfvnr+s0+/RnBes7PyXKpPoPgAgAFz0YCAADINgIAAGAqAgAClRACAAKuMgIAAd5CAgAYCQYCAAEGOAA==
Date:   Mon, 15 Feb 2021 16:00:46 +0000
Message-ID: <345E0497-BFA9-4634-8017-DC9BFF643290@oracle.com>
References: <2A0C36E7-8CB0-486F-A8DB-463CA28C5C5D@oracle.com>
 <EEB0B974-6E63-41A0-9C01-F0DEA39FC4BF@oracle.com>
 <20210209113108.1ca16cfa@carbon> <20210210084155.GA3697@techsingularity.net>
 <20210210124103.56ed1e95@carbon> <20210210130705.GC3629@suse.de>
 <B123FB11-661F-45A6-8235-2982BF3C4B83@oracle.com>
 <20210211091235.GC3697@techsingularity.net>
 <F3CD435E-905F-4262-B4DA-0C721A4235E1@oracle.com>
 <20210215120608.GE3697@techsingularity.net>
In-Reply-To: <20210215120608.GE3697@techsingularity.net>
Accept-Language: en-US
Content-Language: en-US
Content-Type: text/plain; charset="us-ascii"
Content-ID: <F6DD35621BFF3A48A4241680BDEE03F2@namprd10.prod.outlook.com>
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
X-MS-Exchange-CrossTenant-Network-Message-Id: 3392558f-a7c3-4f10-4232-08d8d1cadb89
X-MS-Exchange-CrossTenant-originalarrivaltime: 15 Feb 2021 16:00:46.8953
 (UTC)
X-MS-Exchange-CrossTenant-fromentityheader: Hosted
X-MS-Exchange-CrossTenant-id: 4e2c6054-71cb-48f1-bd6c-3a9705aca71b
X-MS-Exchange-CrossTenant-mailboxtype: HOSTED
X-MS-Exchange-CrossTenant-userprincipalname: DDz1y+ztTornMvhn7wOwo2KCewDbXr74JU3TVSDgP9IsKNg4OjMT+xKQ+OFM7G9H45X5ROSE7rT7uXnMQRFM7A==
X-MS-Exchange-Transport-CrossTenantHeadersStamped: BY5PR10MB3954
X-Proofpoint-Virus-Version: vendor=nai engine=6200 definitions=9895 signatures=668683
X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 malwarescore=0 spamscore=0 mlxscore=0
 phishscore=0 adultscore=0 bulkscore=0 mlxlogscore=999 suspectscore=0
 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2009150000
 definitions=main-2102150125
X-Proofpoint-Virus-Version: vendor=nai engine=6200 definitions=9896 signatures=668683
X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 spamscore=0 priorityscore=1501
 lowpriorityscore=0 bulkscore=0 impostorscore=0 mlxlogscore=999
 adultscore=0 malwarescore=0 phishscore=0 clxscore=1015 mlxscore=0
 suspectscore=0 classifier=spam adjust=0 reason=mlx scancount=1
 engine=8.12.0-2009150000 definitions=main-2102150125
Precedence: bulk
List-ID: <linux-nfs.vger.kernel.org>
X-Mailing-List: linux-nfs@vger.kernel.org


> On Feb 15, 2021, at 7:06 AM, Mel Gorman <mgorman@techsingularity.net> wro=
te:
>=20
> On Thu, Feb 11, 2021 at 04:20:31PM +0000, Chuck Lever wrote:
>>> On Feb 11, 2021, at 4:12 AM, Mel Gorman <mgorman@techsingularity.net> w=
rote:
>>>=20
>>> <SNIP>
>>>=20
>>> Parameters to __rmqueue_pcplist are garbage as the parameter order chan=
ged.
>>> I'm surprised it didn't blow up in a spectacular fashion. Again, this
>>> hasn't been near any testing and passing a list with high orders to
>>> free_pages_bulk() will corrupt lists too. Mostly it's a curiousity to s=
ee
>>> if there is justification for reworking the allocator to fundamentally
>>> deal in batches and then feed batches to pcp lists and the bulk allocat=
or
>>> while leaving the normal GFP API as single page "batches". While that
>>> would be ideal, it's relatively high risk for regressions. There is sti=
ll
>>> some scope for adding a basic bulk allocator before considering a major
>>> refactoring effort.
>>>=20
>>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>>> index f8353ea7b977..8f3fe7de2cf7 100644
>>> --- a/mm/page_alloc.c
>>> +++ b/mm/page_alloc.c
>>> @@ -5892,7 +5892,7 @@ __alloc_pages_bulk_nodemask(gfp_t gfp_mask, unsig=
ned int order,
>>> 	pcp_list =3D &pcp->lists[migratetype];
>>>=20
>>> 	while (nr_pages) {
>>> -		page =3D __rmqueue_pcplist(zone, gfp_mask, migratetype,
>>> +		page =3D __rmqueue_pcplist(zone, migratetype, alloc_flags,
>>> 								pcp, pcp_list);
>>> 		if (!page)
>>> 			break;
>>=20
>> The NFS server is considerably more stable now. Thank you!
>>=20
>=20
> Thanks for testing!
>=20
>> I confirmed that my patch is requesting and getting multiple pages.
>> The new NFSD code and the API seem to be working as expected.
>>=20
>> The results are stunning. Each svc_alloc_arg() call here allocates
>> 65 pages to satisfy a 256KB NFS READ request.
>>=20
>> Before:
>>=20
>>            nfsd-972   [000]   584.513817: funcgraph_entry:      + 35.385=
 us  |  svc_alloc_arg();
>>            nfsd-979   [002]   584.513870: funcgraph_entry:      + 29.051=
 us  |  svc_alloc_arg();
>>            nfsd-980   [001]   584.513951: funcgraph_entry:      + 29.178=
 us  |  svc_alloc_arg();
>>            nfsd-983   [000]   584.514014: funcgraph_entry:      + 29.211=
 us  |  svc_alloc_arg();
>>            nfsd-976   [002]   584.514059: funcgraph_entry:      + 29.315=
 us  |  svc_alloc_arg();
>>            nfsd-974   [001]   584.514127: funcgraph_entry:      + 29.237=
 us  |  svc_alloc_arg();
>>=20
>> After:
>>=20
>>            nfsd-977   [002]    87.049425: funcgraph_entry:        4.293 =
us   |  svc_alloc_arg();
>>            nfsd-981   [000]    87.049478: funcgraph_entry:        4.059 =
us   |  svc_alloc_arg();
>>            nfsd-988   [001]    87.049549: funcgraph_entry:        4.474 =
us   |  svc_alloc_arg();
>>            nfsd-983   [003]    87.049612: funcgraph_entry:        3.819 =
us   |  svc_alloc_arg();
>>            nfsd-976   [000]    87.049619: funcgraph_entry:        3.869 =
us   |  svc_alloc_arg();
>>            nfsd-980   [002]    87.049738: funcgraph_entry:        4.124 =
us   |  svc_alloc_arg();
>>            nfsd-975   [000]    87.049769: funcgraph_entry:        3.734 =
us   |  svc_alloc_arg();
>>=20
>=20
> Uhhhh, that is much better than I expected given how lame the
> implementation is.

My experience with function tracing is the entry and exit
timestamping adds significant overhead. I'd bet the actual
timing improvement is still good but much less.


> Sure -- it works, but it has more overhead than it
> should with the downside that reducing it requires fairly deep surgery. I=
t
> may be enough to tidy this up to handle order-0 pages only to start with
> and see how far it gets. That's a fairly trivial modification.

I'd like to see an "order-0 only" implementation go in soon.
The improvement is palpable and there are several worthy
consumers on deck.


>> There appears to be little cost change for single-page allocations
>> using the bulk allocator (nr_pages=3D1):
>>=20
>> Before:
>>=20
>>            nfsd-985   [003]   572.324517: funcgraph_entry:        0.332 =
us   |  svc_alloc_arg();
>>            nfsd-986   [001]   572.324531: funcgraph_entry:        0.311 =
us   |  svc_alloc_arg();
>>            nfsd-985   [003]   572.324701: funcgraph_entry:        0.311 =
us   |  svc_alloc_arg();
>>            nfsd-986   [001]   572.324727: funcgraph_entry:        0.424 =
us   |  svc_alloc_arg();
>>            nfsd-985   [003]   572.324760: funcgraph_entry:        0.332 =
us   |  svc_alloc_arg();
>>            nfsd-986   [001]   572.324786: funcgraph_entry:        0.390 =
us   |  svc_alloc_arg();
>>=20
>> After:
>>=20
>>            nfsd-989   [002]    75.043226: funcgraph_entry:        0.322 =
us   |  svc_alloc_arg();
>>            nfsd-988   [001]    75.043436: funcgraph_entry:        0.368 =
us   |  svc_alloc_arg();
>>            nfsd-989   [002]    75.043464: funcgraph_entry:        0.424 =
us   |  svc_alloc_arg();
>>            nfsd-988   [001]    75.043490: funcgraph_entry:        0.317 =
us   |  svc_alloc_arg();
>>            nfsd-989   [002]    75.043517: funcgraph_entry:        0.425 =
us   |  svc_alloc_arg();
>>            nfsd-988   [001]    75.050025: funcgraph_entry:        0.407 =
us   |  svc_alloc_arg();
>>=20
>=20
> That is not too surprising given that there would be some additional
> overhead to manage a list of 1 page. I would hope that users of the bulk
> allocator are not routinely calling it with nr_pages =3D=3D 1.

The NFSD implementation I did uses alloc_pages_bulk() to fill however
many pages are needed. Often that's just one page.

Sometimes it's zero pages. alloc_pages_bulk() does not behave very
well, so NFSD avoids calling it in that case.

I can post the patch for review. I cleaned it up recently but haven't
had a chance to test the clean-ups, so it might not work in its
current state.

--
Chuck Lever