netdev - Re: [RFC PATCH 2/4] page_pool: basic implementation of page

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 9 Jan 2017 21:58:26 +0000
From:   Mel Gorman <mgorman@...hsingularity.net>
To:     Jesper Dangaard Brouer <brouer@...hat.com>
Cc:     Vlastimil Babka <vbabka@...e.cz>, linux-mm@...ck.org,
        Alexander Duyck <alexander.duyck@...il.com>,
        willemdebruijn.kernel@...il.com, netdev@...r.kernel.org,
        john.fastabend@...il.com, Saeed Mahameed <saeedm@...lanox.com>,
        bjorn.topel@...el.com,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Tariq Toukan <tariqt@...lanox.com>
Subject: Re: [RFC PATCH 2/4] page_pool: basic implementation of page_pool

On Mon, Jan 09, 2017 at 09:45:24PM +0100, Jesper Dangaard Brouer wrote:
> > I see. I guess if all page pool pages were order>0 compound pages, you
> > could hook this to the existing compound_dtor functionality instead.
> 
> The page_pool will support order>0 pages, but it is the order-0 case
> that is optimized for.
> 

The bulk allocator is currently not suitable for high-order pages. It would
take more work to do that but is not necessarily even a good idea. FWIW,
the high-order per-cpu page allocator posted some weeks ago would be the
basis. I didn't push that series as the benefit to SLUB was too marginal
given the complexity.

> > Well typically the VMA mapped pages are those on the LRU list (anonymous
> > or file). But I don't suppose you will want memory reclaim to free your
> > pages, so seems lru field should be reusable for you.
> 
> Thanks for the info.
> 
> So, LRU-list area could be reusable, but I does not align so well with
> the bulking API Mel just introduced/proposed, but still doable.
> 

That's a relatively minor implementation detail. I needed something to
hang the pages onto for returning. Using a list and page->lru is a standard
approach but it does not mandate that the caller preserve page->lru or that
it's related to the LRU. The caller simply needs to put the pages back onto
a list if it's bulk freeing or call __free_pages() directly for each page.
If any in-kernel user uses __free_pages() then the free_pages_bulk()
API can be dropped entirely.

I'm not intending to merge the bulk allocator due to a lack of in-kernel
users and an inability to test in-kernel users.  It was simply designed to
illustrate how to call the core of the page allocator in a way that avoids
the really expensive checks. If required, the pages could be returned on
a caller-allocated array or something exotic like using one page to store
pointers to the rest. Either of those alternatives are harder to use. A
caller-allocated array must be sure the nr_pages parameter is correct and
the exotic approach would require careful use by the caller. Using page->lru
was more straight-forward when the requirements of the callers was unknown.

It opens the question of what to do with that series. I was going to wait
for feedback but my intent was to try merge patches 1-3 if there were no
objections and preferably with your reviewed-by or ack. I would then hand
patch 4 over to you for addition to a series that added in-kernel callers to
alloc_pages_bulk() be that the generic pool recycle or modifying drivers.
You are then free to modify the API to suit your needs without having to
figure out the best way of calling the page allocator.

Any thoughts?

-- 
Mel Gorman
SUSE Labs