From: David Rientjes <rientjes@google.com>
To: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>,
Linus Torvalds <torvalds@linux-foundation.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: Linux 4.9-rc6
Date: Mon, 21 Nov 2016 00:34:13 -0800 (PST) [thread overview]
Message-ID: <alpine.DEB.2.10.1611210024210.66057@chino.kir.corp.google.com> (raw)
In-Reply-To: <1479704390.8455.398.camel@edumazet-glaptop3.roam.corp.google.com>
On Sun, 20 Nov 2016, Eric Dumazet wrote:
> Another potential issue with CONFIG_VMAP_STACK is that we make no
> attempt to allocate 4 consecutive pages.
>
> Even if we have plenty of memory, 4 calls to alloc_page() are likely to
> give us 4 pages in completely different locations.
>
> Here I printed the hugepage number of the 4 pages for some stacks :
>
>
> 0xffffc9001a07c000-0xffffc9001a081000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfcac Hfeba Hfec0 Hfc9d N0=4
> 0xffffc9001a084000-0xffffc9001a089000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfc79 Hfc79 Hfc79 Hfc83 N0=4
> 0xffffc9001a08c000-0xffffc9001a091000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfc9b Hfe91 Hfebe Hfca2 N0=4
> 0xffffc9001a094000-0xffffc9001a099000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfcaa Hfcaa Hfca6 Hfebc N0=4
> 0xffffc9001a09c000-0xffffc9001a0a1000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfe9b Hfe90 Hff09 Hfefb N0=4
> 0xffffc9001a0a4000-0xffffc9001a0a9000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfe94 Hfe62 Hfea0 Hfe7b N0=4
> 0xffffc9001a0ac000-0xffffc9001a0b1000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfe78 Hff05 Hff05 Hfc74 N0=4
> 0xffffc9001a0b4000-0xffffc9001a0b9000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfc9b Hfc9b Hfe83 Hf782 N0=4
> 0xffffc9001a0bc000-0xffffc9001a0c1000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfe78 Hfe78 Hfc7f Hfc7f N0=4
> 0xffffc9001a0c4000-0xffffc9001a0c9000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfebe Hfebe Hfe82 Hfe85 N0=4
> 0xffffc9001a0cc000-0xffffc9001a0d1000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfc6b Hfe62 Hfe62 Hfcaa N0=4
> 0xffffc9001a0d4000-0xffffc9001a0d9000 20480 _do_fork+0xe1/0x360 pages=4 vmalloc Hfebd Hfebd Hfc92 Hfc92 N0=4
>
> This is a vmalloc() generic issue that is worth fixing now ?
>
> Note this RFC might conflict with NUMA interleave policy.
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index f2481cb4e6b2..0123e97debb9 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -1602,9 +1602,10 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
> pgprot_t prot, int node)
> {
> struct page **pages;
> - unsigned int nr_pages, array_size, i;
> + unsigned int nr_pages, array_size, i, j;
> const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
> const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
> + const gfp_t multi_alloc_mask = (gfp_mask & ~__GFP_DIRECT_RECLAIM) | __GFP_NORETRY;
>
> nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
> array_size = (nr_pages * sizeof(struct page *));
I think multi_alloc_mask wants to use alloc_mask rather than gfp_mask
before clearing the bit, otherwise the failed high-order allocations with
no chance to reclaim will spew page allocation failure warnings. Using
__GFP_NORETRY here would be a no-op, but it depends on the implementation
so no problems setting it.
> @@ -1624,20 +1625,34 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
> return NULL;
> }
>
> - for (i = 0; i < area->nr_pages; i++) {
> - struct page *page;
> -
> - if (node == NUMA_NO_NODE)
> - page = alloc_page(alloc_mask);
> - else
> - page = alloc_pages_node(node, alloc_mask, 0);
> + for (i = 0; i < area->nr_pages;) {
> + struct page *page = NULL;
> + unsigned int chunk_order = min(ilog2(area->nr_pages - i), MAX_ORDER - 1);
> +
> + while (chunk_order && !page) {
> + if (node == NUMA_NO_NODE)
> + page = alloc_pages(multi_alloc_mask, chunk_order);
> + else
> + page = alloc_pages_node(node, multi_alloc_mask, chunk_order);
> + if (page)
> + split_page(page, chunk_order);
> + else
> + chunk_order--;
> + }
> + if (!page) {
> + if (node == NUMA_NO_NODE)
> + page = alloc_pages(alloc_mask, 0);
> + else
> + page = alloc_pages_node(node, alloc_mask, 0);
> + }
>
> if (unlikely(!page)) {
> /* Successfully allocated i pages, free them in __vunmap() */
> area->nr_pages = i;
> goto fail;
> }
> - area->pages[i] = page;
> + for (j = 0; j < (1 << chunk_order); j++)
> + area->pages[i++] = page++;
> if (gfpflags_allow_blocking(gfp_mask))
> cond_resched();
> }
>
>
>
next prev parent reply other threads:[~2016-11-21 8:34 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-11-20 22:05 Linux 4.9-rc6 Linus Torvalds
2016-11-20 22:27 ` Eric Dumazet
2016-11-20 23:27 ` Linus Torvalds
2016-11-21 1:35 ` Al Viro
2016-11-21 4:59 ` Eric Dumazet
2016-11-21 8:34 ` David Rientjes [this message]
2016-11-21 13:32 ` Eric Dumazet
2016-11-21 13:51 ` Eric Dumazet
2016-11-21 16:49 ` Eric Dumazet
2016-12-04 10:43 ` Thorsten Leemhuis
[not found] ` <CA+55aFzPiZW4FfWbvM-+AFraa0fkUHv4C1Y9SCzHdXEcUSPqdg@mail.gmail.com>
2016-12-04 17:17 ` Eric Dumazet
2016-12-21 15:30 ` Eric Dumazet
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=alpine.DEB.2.10.1611210024210.66057@chino.kir.corp.google.com \
--to=rientjes@google.com \
--cc=eric.dumazet@gmail.com \
--cc=linux-kernel@vger.kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=viro@zeniv.linux.org.uk \
/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).