From: Dan Williams <dan.j.williams@intel.com>
To: Joao Martins <joao.m.martins@oracle.com>
Cc: Linux MM <linux-mm@kvack.org>, Ira Weiny <ira.weiny@intel.com>,
linux-nvdimm <linux-nvdimm@lists.01.org>,
Matthew Wilcox <willy@infradead.org>,
Jason Gunthorpe <jgg@ziepe.ca>, Jane Chu <jane.chu@oracle.com>,
Muchun Song <songmuchun@bytedance.com>,
Mike Kravetz <mike.kravetz@oracle.com>,
Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [PATCH RFC 0/9] mm, sparse-vmemmap: Introduce compound pagemaps
Date: Fri, 19 Feb 2021 17:18:51 -0800 [thread overview]
Message-ID: <CAPcyv4gQQ03-nhBNwLK6KDc953SVD1rOs7HFBo_Mu9LFTkXRgw@mail.gmail.com> (raw)
In-Reply-To: <20201208172901.17384-1-joao.m.martins@oracle.com>
On Tue, Dec 8, 2020 at 9:32 AM Joao Martins <joao.m.martins@oracle.com> wrote:
>
> Hey,
>
> This small series, attempts at minimizing 'struct page' overhead by
> pursuing a similar approach as Muchun Song series "Free some vmemmap
> pages of hugetlb page"[0] but applied to devmap/ZONE_DEVICE.
>
> [0] https://lore.kernel.org/linux-mm/20201130151838.11208-1-songmuchun@bytedance.com/
Clever!
>
> The link above describes it quite nicely, but the idea is to reuse tail
> page vmemmap areas, particular the area which only describes tail pages.
> So a vmemmap page describes 64 struct pages, and the first page for a given
> ZONE_DEVICE vmemmap would contain the head page and 63 tail pages. The second
> vmemmap page would contain only tail pages, and that's what gets reused across
> the rest of the subsection/section. The bigger the page size, the bigger the
> savings (2M hpage -> save 6 vmemmap pages; 1G hpage -> save 4094 vmemmap pages).
>
> In terms of savings, per 1Tb of memory, the struct page cost would go down
> with compound pagemap:
>
> * with 2M pages we lose 4G instead of 16G (0.39% instead of 1.5% of total memory)
> * with 1G pages we lose 8MB instead of 16G (0.0007% instead of 1.5% of total memory)
Nice!
>
> Along the way I've extended it past 'struct page' overhead *trying* to address a
> few performance issues we knew about for pmem, specifically on the
> {pin,get}_user_pages* function family with device-dax vmas which are really
> slow even of the fast variants. THP is great on -fast variants but all except
> hugetlbfs perform rather poorly on non-fast gup.
>
> So to summarize what the series does:
>
> Patches 1-5: Much like Muchun series, we reuse tail page areas across a given
> page size (namely @align was referred by remaining memremap/dax code) and
> enabling of memremap to initialize the ZONE_DEVICE pages as compound pages or a
> given @align order. The main difference though, is that contrary to the hugetlbfs
> series, there's no vmemmap for the area, because we are onlining it. IOW no
> freeing of pages of already initialized vmemmap like the case for hugetlbfs,
> which simplifies the logic (besides not being arch-specific). After these,
> there's quite visible region bootstrap of pmem memmap given that we would
> initialize fewer struct pages depending on the page size.
>
> NVDIMM namespace bootstrap improves from ~750ms to ~190ms/<=1ms on emulated NVDIMMs
> with 2M and 1G respectivally. The net gain in improvement is similarly observed
> in proportion when running on actual NVDIMMs.
I
>
> Patch 6 - 8: Optimize grabbing/release a page refcount changes given that we
> are working with compound pages i.e. we do 1 increment/decrement to the head
> page for a given set of N subpages compared as opposed to N individual writes.
> {get,pin}_user_pages_fast() for zone_device with compound pagemap consequently
> improves considerably, and unpin_user_pages() improves as well when passed a
> set of consecutive pages:
>
> before after
> (get_user_pages_fast 1G;2M page size) ~75k us -> ~3.2k ; ~5.2k us
> (pin_user_pages_fast 1G;2M page size) ~125k us -> ~3.4k ; ~5.5k us
Compelling!
>
> The RDMA patch (patch 8/9) is to demonstrate the improvement for an existing
> user. For unpin_user_pages() we have an additional test to demonstrate the
> improvement. The test performs MR reg/unreg continuously and measuring its
> rate for a given period. So essentially ib_mem_get and ib_mem_release being
> stress tested which at the end of day means: pin_user_pages_longterm() and
> unpin_user_pages() for a scatterlist:
>
> Before:
> 159 rounds in 5.027 sec: 31617.923 usec / round (device-dax)
> 466 rounds in 5.009 sec: 10748.456 usec / round (hugetlbfs)
>
> After:
> 305 rounds in 5.010 sec: 16426.047 usec / round (device-dax)
> 1073 rounds in 5.004 sec: 4663.622 usec / round (hugetlbfs)
Why does hugetlbfs get faster for a ZONE_DEVICE change? Might answer
that question myself when I get to patch 8.
>
> Patch 9: Improves {pin,get}_user_pages() and its longterm counterpart. It
> is very experimental, and I imported most of follow_hugetlb_page(), except
> that we do the same trick as gup-fast. In doing the patch I feel this batching
> should live in follow_page_mask() and having that being changed to return a set
> of pages/something-else when walking over PMD/PUDs for THP / devmap pages. This
> patch then brings the previous test of mr reg/unreg (above) on parity
> between device-dax and hugetlbfs.
>
> Some of the patches are a little fresh/WIP (specially patch 3 and 9) and we are
> still running tests. Hence the RFC, asking for comments and general direction
> of the work before continuing.
Will go look at the code, but I don't see anything scary conceptually
here. The fact that pfn_to_page() does not need to change is among the
most compelling features of this approach.
next prev parent reply other threads:[~2021-02-20 1:19 UTC|newest]
Thread overview: 80+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-12-08 17:28 [PATCH RFC 0/9] mm, sparse-vmemmap: Introduce compound pagemaps Joao Martins
2020-12-08 17:28 ` [PATCH RFC 1/9] memremap: add ZONE_DEVICE support for compound pages Joao Martins
2020-12-09 5:59 ` John Hubbard
2020-12-09 6:33 ` Matthew Wilcox
2020-12-09 13:12 ` Joao Martins
2021-02-20 1:43 ` Dan Williams
2021-02-22 11:24 ` Joao Martins
2021-02-22 20:37 ` Dan Williams
2021-02-23 15:46 ` Joao Martins
2021-02-23 16:50 ` Dan Williams
2021-02-23 17:18 ` Joao Martins
2021-02-23 18:18 ` Dan Williams
2021-03-10 18:12 ` Joao Martins
2021-03-12 5:54 ` Dan Williams
2021-02-20 1:24 ` Dan Williams
2021-02-22 11:09 ` Joao Martins
2020-12-08 17:28 ` [PATCH RFC 2/9] sparse-vmemmap: Consolidate arguments in vmemmap section populate Joao Martins
2020-12-09 6:16 ` John Hubbard
2020-12-09 13:51 ` Joao Martins
2021-02-20 1:49 ` Dan Williams
2021-02-22 11:26 ` Joao Martins
2020-12-08 17:28 ` [PATCH RFC 3/9] sparse-vmemmap: Reuse vmemmap areas for a given mhp_params::align Joao Martins
2020-12-08 17:38 ` Joao Martins
2020-12-08 17:28 ` [PATCH RFC 3/9] sparse-vmemmap: Reuse vmemmap areas for a given page size Joao Martins
2021-02-20 3:34 ` Dan Williams
2021-02-22 11:42 ` Joao Martins
2021-02-22 22:40 ` Dan Williams
2021-02-23 15:46 ` Joao Martins
2020-12-08 17:28 ` [PATCH RFC 4/9] mm/page_alloc: Reuse tail struct pages for compound pagemaps Joao Martins
2021-02-20 6:17 ` Dan Williams
2021-02-22 12:01 ` Joao Martins
2020-12-08 17:28 ` [PATCH RFC 5/9] device-dax: Compound pagemap support Joao Martins
2020-12-08 17:28 ` [PATCH RFC 6/9] mm/gup: Grab head page refcount once for group of subpages Joao Martins
2020-12-08 19:49 ` Jason Gunthorpe
2020-12-09 11:05 ` Joao Martins
2020-12-09 15:15 ` Jason Gunthorpe
2020-12-09 16:02 ` Joao Martins
2020-12-09 16:24 ` Jason Gunthorpe
2020-12-09 17:27 ` Joao Martins
2020-12-09 18:14 ` Matthew Wilcox
2020-12-09 19:08 ` Jason Gunthorpe
2020-12-10 15:43 ` Joao Martins
2020-12-09 4:40 ` John Hubbard
2020-12-09 13:44 ` Joao Martins
2020-12-08 17:28 ` [PATCH RFC 7/9] mm/gup: Decrement head page " Joao Martins
2020-12-08 19:34 ` Jason Gunthorpe
2020-12-09 5:06 ` John Hubbard
2020-12-09 13:43 ` Jason Gunthorpe
2020-12-09 12:17 ` Joao Martins
2020-12-17 19:05 ` Joao Martins
2020-12-17 20:05 ` Jason Gunthorpe
2020-12-17 22:34 ` Joao Martins
2020-12-18 14:25 ` Jason Gunthorpe
2020-12-19 2:06 ` John Hubbard
2020-12-19 13:10 ` Joao Martins
2020-12-08 17:29 ` [PATCH RFC 8/9] RDMA/umem: batch page unpin in __ib_mem_release() Joao Martins
2020-12-08 19:29 ` Jason Gunthorpe
2020-12-09 10:59 ` Joao Martins
2020-12-19 13:15 ` Joao Martins
2020-12-09 5:18 ` John Hubbard
2020-12-08 17:29 ` [PATCH RFC 9/9] mm: Add follow_devmap_page() for devdax vmas Joao Martins
2020-12-08 19:57 ` Jason Gunthorpe
2020-12-09 8:05 ` Christoph Hellwig
2020-12-09 11:19 ` Joao Martins
2020-12-09 5:23 ` John Hubbard
2020-12-09 9:38 ` [PATCH RFC 0/9] mm, sparse-vmemmap: Introduce compound pagemaps David Hildenbrand
2020-12-09 9:52 ` [External] " Muchun Song
2021-02-20 1:18 ` Dan Williams [this message]
2021-02-22 11:06 ` Joao Martins
2021-02-22 14:32 ` Joao Martins
2021-02-23 16:28 ` Joao Martins
2021-02-23 16:44 ` Dan Williams
2021-02-23 17:15 ` Joao Martins
2021-02-23 18:15 ` Dan Williams
2021-02-23 18:54 ` Jason Gunthorpe
2021-02-23 22:48 ` Dan Williams
2021-02-23 23:07 ` Jason Gunthorpe
2021-02-24 0:14 ` Dan Williams
2021-02-24 1:00 ` Jason Gunthorpe
2021-02-24 1:32 ` Dan Williams
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAPcyv4gQQ03-nhBNwLK6KDc953SVD1rOs7HFBo_Mu9LFTkXRgw@mail.gmail.com \
--to=dan.j.williams@intel.com \
--cc=akpm@linux-foundation.org \
--cc=ira.weiny@intel.com \
--cc=jane.chu@oracle.com \
--cc=jgg@ziepe.ca \
--cc=joao.m.martins@oracle.com \
--cc=linux-mm@kvack.org \
--cc=linux-nvdimm@lists.01.org \
--cc=mike.kravetz@oracle.com \
--cc=songmuchun@bytedance.com \
--cc=willy@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).