From: Michal Hocko <mhocko@suse.com>
To: Oscar Salvador <osalvador@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>,
Mike Kravetz <mike.kravetz@oracle.com>,
David Hildenbrand <david@redhat.com>,
Muchun Song <songmuchun@bytedance.com>,
linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/2] mm: Make alloc_contig_range handle free hugetlb pages
Date: Wed, 17 Feb 2021 16:00:11 +0100 [thread overview]
Message-ID: <YC0ve4PP+VTrEEtw@dhcp22.suse.cz> (raw)
In-Reply-To: <20210217100816.28860-2-osalvador@suse.de>
On Wed 17-02-21 11:08:15, Oscar Salvador wrote:
[...]
> +static bool alloc_and_dissolve_huge_page(struct hstate *h, struct page *page)
> +{
> + gfp_t gfp_mask = htlb_alloc_mask(h);
> + nodemask_t *nmask = &node_states[N_MEMORY];
> + struct page *new_page;
> + bool ret = false;
> + int nid;
> +
> + spin_lock(&hugetlb_lock);
> + /*
> + * Check one more time to make race-window smaller.
> + */
> + if (!PageHuge(page)) {
> + /*
> + * Dissolved from under our feet.
> + */
> + spin_unlock(&hugetlb_lock);
> + return true;
> + }
Is this really necessary? dissolve_free_huge_page will take care of this
and the race windown you are covering is really tiny.
> +
> + nid = page_to_nid(page);
> + spin_unlock(&hugetlb_lock);
> +
> + /*
> + * Before dissolving the page, we need to allocate a new one,
> + * so the pool remains stable.
> + */
> + new_page = alloc_fresh_huge_page(h, gfp_mask, nid, nmask, NULL);
wrt. fallback to other zones, I haven't realized that the primary
usecase is a form of memory offlining (from virt-mem). I am not yet sure
what the proper behavior is in that case but if breaking hugetlb pools,
similar to the normal hotplug operation, is viable then this needs a
special mode. We do not want a random alloc_contig_range user to do the
same. So for starter I would go with __GFP_THISNODE here.
> + if (new_page) {
> + /*
> + * Ok, we got a new free hugepage to replace this one. Try to
> + * dissolve the old page.
> + */
> + if (!dissolve_free_huge_page(page)) {
> + ret = true;
> + } else if (dissolve_free_huge_page(new_page)) {
> + /*
> + * Seems the old page could not be dissolved, so try to
> + * dissolve the freshly allocated page. If that fails
> + * too, let us count the new page as a surplus. Doing so
> + * allows the pool to be re-balanced when pages are freed
> + * instead of enqueued again.
> + */
> + spin_lock(&hugetlb_lock);
> + h->surplus_huge_pages++;
> + h->surplus_huge_pages_node[nid]++;
> + spin_unlock(&hugetlb_lock);
> + }
> + /*
> + * Free it into the hugepage allocator
> + */
> + put_page(new_page);
> + }
> +
> + return ret;
> +}
> +
> +bool isolate_or_dissolve_huge_page(struct page *page)
> +{
> + struct hstate *h = NULL;
> + struct page *head;
> + bool ret = false;
> +
> + spin_lock(&hugetlb_lock);
> + if (PageHuge(page)) {
> + head = compound_head(page);
> + h = page_hstate(head);
> + }
> + spin_unlock(&hugetlb_lock);
> +
> + if (!h)
> + /*
> + * The page might have been dissolved from under our feet.
> + * If that is the case, return success as if we dissolved it
> + * ourselves.
> + */
> + return true;
nit I would put the comment above the conditin for both cases. It reads
more easily that way. At least without { }.
> +
> + if (hstate_is_gigantic(h))
> + /*
> + * Fence off gigantic pages as there is a cyclic dependency
> + * between alloc_contig_range and them.
> + */
> + return ret;
> +
> + if(!page_count(head) && alloc_and_dissolve_huge_page(h, head))
> + ret = true;
> +
> + return ret;
> +}
> +
> struct page *alloc_huge_page(struct vm_area_struct *vma,
> unsigned long addr, int avoid_reserve)
> {
Other than that I haven't noticed any surprises.
--
Michal Hocko
SUSE Labs
next prev parent reply other threads:[~2021-02-17 15:02 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-02-17 10:08 [PATCH 0/2] Make alloc_contig_range handle Hugetlb pages Oscar Salvador
2021-02-17 10:08 ` [PATCH 1/2] mm: Make alloc_contig_range handle free hugetlb pages Oscar Salvador
2021-02-17 13:30 ` Michal Hocko
2021-02-17 13:36 ` David Hildenbrand
2021-02-17 13:50 ` Michal Hocko
2021-02-17 13:53 ` David Hildenbrand
2021-02-17 13:59 ` Michal Hocko
2021-02-17 14:08 ` David Hildenbrand
2021-02-17 14:14 ` Michal Hocko
2021-02-17 14:23 ` Oscar Salvador
2021-02-17 13:42 ` Oscar Salvador
2021-02-17 15:00 ` Michal Hocko [this message]
2021-02-18 10:09 ` Oscar Salvador
2021-02-18 12:52 ` Michal Hocko
2021-02-18 13:32 ` Oscar Salvador
2021-02-18 13:59 ` Michal Hocko
2021-02-18 16:53 ` Oscar Salvador
2021-02-19 9:05 ` Oscar Salvador
2021-02-19 9:56 ` Michal Hocko
2021-02-19 10:14 ` Oscar Salvador
2021-02-19 20:00 ` Mike Kravetz
2021-02-19 10:40 ` Oscar Salvador
2021-02-19 10:55 ` Michal Hocko
2021-02-19 11:17 ` Oscar Salvador
2021-02-19 11:24 ` Michal Hocko
2021-02-17 10:08 ` [PATCH 2/2] mm: Make alloc_contig_range handle in-use " Oscar Salvador
2021-02-17 13:36 ` Michal Hocko
2021-02-17 13:46 ` Oscar Salvador
2021-02-17 13:54 ` Michal Hocko
2021-02-17 15:06 ` Michal Hocko
2021-02-17 15:27 ` Oscar Salvador
2021-02-17 15:33 ` Michal Hocko
2021-02-18 6:01 ` Oscar Salvador
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=YC0ve4PP+VTrEEtw@dhcp22.suse.cz \
--to=mhocko@suse.com \
--cc=akpm@linux-foundation.org \
--cc=david@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mike.kravetz@oracle.com \
--cc=osalvador@suse.de \
--cc=songmuchun@bytedance.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).