From: Matthew Wilcox <willy@infradead.org>
To: Michal Hocko <mhocko@suse.com>
Cc: Zhou Guanghui <zhouguanghui1@huawei.com>,
linux-kernel@vger.kernel.org, linux-mm@kvack.org,
akpm@linux-foundation.org, hannes@cmpxchg.org, hughd@google.com,
kirill.shutemov@linux.intel.com, npiggin@gmail.com,
ziy@nvidia.com, wangkefeng.wang@huawei.com, guohanjun@huawei.com,
dingtianhong@huawei.com, chenweilong@huawei.com,
rui.xiang@huawei.com
Subject: Re: [PATCH v2 2/2] mm/memcg: set memcg when split page
Date: Tue, 9 Mar 2021 12:32:55 +0000 [thread overview]
Message-ID: <20210309123255.GI3479805@casper.infradead.org> (raw)
In-Reply-To: <YEc5iI+ZP7dWr2fC@dhcp22.suse.cz>
On Tue, Mar 09, 2021 at 10:02:00AM +0100, Michal Hocko wrote:
> On Mon 08-03-21 21:02:25, Matthew Wilcox wrote:
> > On Thu, Mar 04, 2021 at 07:40:53AM +0000, Zhou Guanghui wrote:
> > > For example, when alloc_pages_exact is used to allocate 1MB continuous
> > > physical memory, 2MB is charged(kmemcg is enabled and __GFP_ACCOUNT is
> > > set). When make_alloc_exact free the unused 1MB and free_pages_exact
> > > free the applied 1MB, actually, only 4KB(one page) is uncharged.
> > @@ -5081,9 +5081,15 @@ void __free_pages(struct page *page, unsigned int order)
> > {
> > if (put_page_testzero(page))
> > free_the_page(page, order);
> > - else if (!PageHead(page))
> > - while (order-- > 0)
> > - free_the_page(page + (1 << order), order);
> > + else if (!PageHead(page)) {
> > + while (order-- > 0) {
> > + struct page *tail = page + (1 << order);
> > +#ifdef CONFIG_MEMCG
> > + tail->memcg_data = page->memcg_data;
> > +#endif
> > + free_the_page(tail, order);
> > + }
> > + }
> > }
> > EXPORT_SYMBOL(__free_pages);
>
> Hmm, I was not aware of this code. This is really a tricky code.
Yes. I only added it recently. I don't see a better way to solve this
problem. We could turn the non-compound page into a compound page at
this point, but I'm not sure that's really less tricky.
> > I wonder if we shouldn't initialise memcg_data on all subsequent pages
> > of non-compound allocations instead? Because I'm not sure this is the
> > only place that needs to be fixed.
>
> That would be safer for sure. Do you mean this as a replacement to the
> original patch?
>
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 913c2b9e5c72..d44dea2b8d22 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -3135,8 +3135,21 @@ int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order)
> if (memcg && !mem_cgroup_is_root(memcg)) {
> ret = __memcg_kmem_charge(memcg, gfp, 1 << order);
> if (!ret) {
> + int nr_pages = 1 << order;
> page->memcg_data = (unsigned long)memcg |
> MEMCG_DATA_KMEM;
> +
> + /*
> + * Compound pages are normally split or freed
> + * via their head pages so memcg_data in in the
> + * head page should be sufficient but there
> + * are exceptions to the rule (see __free_pages).
> + * Non compound pages would need to copy memcg anyway.
> + */
> + for (i = 1; i < nr_pages; i++) {
> + struct page * p = page + i;
> + p->memcg_data = page->memcg_data
> + }
> return 0;
I would condition this loop on if (!(gfp & __GFP_COMP)), but yes, something
along these lines. I might phrase the comment a little differently ...
/*
* Compound pages are treated as a single unit,
* but non-compound pages can be freed individually
* so each page needs to have its memcg set to get
* the accounting right.
*/
next prev parent reply other threads:[~2021-03-09 12:33 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-04 7:40 [PATCH v2 0/2] set memcg when split page Zhou Guanghui
2021-03-04 7:40 ` [PATCH v2 1/2] mm/memcg: rename mem_cgroup_split_huge_fixup to split_page_memcg Zhou Guanghui
2021-03-04 15:50 ` Johannes Weiner
2021-03-04 16:20 ` Zi Yan
2021-03-04 18:54 ` Shakeel Butt
2021-03-05 11:48 ` Michal Hocko
2021-03-08 22:37 ` Singh, Balbir
2021-03-09 8:28 ` Michal Hocko
2021-03-10 21:44 ` Singh, Balbir
2021-03-10 22:00 ` Hugh Dickins
2021-03-10 23:50 ` Singh, Balbir
2021-03-04 7:40 ` [PATCH v2 2/2] mm/memcg: set memcg when split page Zhou Guanghui
2021-03-04 15:52 ` Johannes Weiner
2021-03-04 16:22 ` Zi Yan
2021-03-04 18:55 ` Shakeel Butt
2021-03-05 11:52 ` Michal Hocko
2021-03-05 23:58 ` Andrew Morton
2021-03-08 8:41 ` Michal Hocko
2021-03-08 20:42 ` Andrew Morton
2021-03-08 20:47 ` Matthew Wilcox
2021-03-09 0:10 ` Andrew Morton
2021-03-08 21:02 ` Matthew Wilcox
2021-03-09 9:02 ` Michal Hocko
2021-03-09 12:32 ` Matthew Wilcox [this message]
2021-03-09 13:03 ` Michal Hocko
2021-03-11 8:37 ` Michal Hocko
2021-03-11 15:21 ` Johannes Weiner
2021-03-11 16:23 ` Matthew Wilcox
2021-03-11 16:26 ` Michal Hocko
2021-03-11 20:37 ` Hugh Dickins
2021-03-18 14:05 ` Michal Hocko
2021-03-18 15:02 ` Matthew Wilcox
2021-03-18 15:07 ` Johannes Weiner
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=20210309123255.GI3479805@casper.infradead.org \
--to=willy@infradead.org \
--cc=akpm@linux-foundation.org \
--cc=chenweilong@huawei.com \
--cc=dingtianhong@huawei.com \
--cc=guohanjun@huawei.com \
--cc=hannes@cmpxchg.org \
--cc=hughd@google.com \
--cc=kirill.shutemov@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@suse.com \
--cc=npiggin@gmail.com \
--cc=rui.xiang@huawei.com \
--cc=wangkefeng.wang@huawei.com \
--cc=zhouguanghui1@huawei.com \
--cc=ziy@nvidia.com \
/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).