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>,
Vlastimil Babka <vbabka@suse.cz>,
David Hildenbrand <david@redhat.com>,
Muchun Song <songmuchun@bytedance.com>,
linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v7 5/7] mm: Make alloc_contig_range handle free hugetlb pages
Date: Tue, 13 Apr 2021 15:40:18 +0200 [thread overview]
Message-ID: <YHWfQjegLi4xekhM@dhcp22.suse.cz> (raw)
In-Reply-To: <20210413104747.12177-6-osalvador@suse.de>
On Tue 13-04-21 12:47:45, Oscar Salvador wrote:
> alloc_contig_range will fail if it ever sees a HugeTLB page within the
> range we are trying to allocate, even when that page is free and can be
> easily reallocated.
> This has proved to be problematic for some users of alloc_contic_range,
> e.g: CMA and virtio-mem, where those would fail the call even when those
> pages lay in ZONE_MOVABLE and are free.
>
> We can do better by trying to replace such page.
>
> Free hugepages are tricky to handle so as to no userspace application
> notices disruption, we need to replace the current free hugepage with
> a new one.
>
> In order to do that, a new function called alloc_and_dissolve_huge_page
> is introduced.
> This function will first try to get a new fresh hugepage, and if it
> succeeds, it will replace the old one in the free hugepage pool.
>
> The free page replacement is done under hugetlb_lock, so no external
> users of hugetlb will notice the change.
> To allocate the new huge page, we use alloc_buddy_huge_page(), so we
> do not have to deal with any counters, and prep_new_huge_page() is not
> called. This is valulable because in case we need to free the new page,
> we only need to call __free_pages().
>
> Once we know that the page to be replaced is a genuine 0-refcounted
> huge page, we remove the old page from the freelist by remove_hugetlb_page().
> Then, we can call __prep_new_huge_page() and __prep_account_new_huge_page()
> for the new huge page to properly initialize it and increment the
> hstate->nr_huge_pages counter (previously decremented by
> remove_hugetlb_page()).
> Once done, the page is enqueued by enqueue_huge_page() and it is ready
> to be used.
>
> There is one tricky case when
> page's refcount is 0 because it is in the process of being released.
> A missing PageHugeFreed bit will tell us that freeing is in flight so
> we retry after dropping the hugetlb_lock. The race window should be
> small and the next retry should make a forward progress.
>
> E.g:
>
> CPU0 CPU1
> free_huge_page() isolate_or_dissolve_huge_page
> PageHuge() == T
> alloc_and_dissolve_huge_page
> alloc_buddy_huge_page()
> spin_lock_irq(hugetlb_lock)
> // PageHuge() && !PageHugeFreed &&
> // !PageCount()
> spin_unlock_irq(hugetlb_lock)
> spin_lock_irq(hugetlb_lock)
> 1) update_and_free_page
> PageHuge() == F
> __free_pages()
> 2) enqueue_huge_page
> SetPageHugeFreed()
> spin_unlock(&hugetlb_lock)
> spin_lock_irq(hugetlb_lock)
> 1) PageHuge() == F (freed by case#1 from CPU0)
> 2) PageHuge() == T
> PageHugeFreed() == T
> - proceed with replacing the page
>
> In the case above we retry as the window race is quite small and we have high
> chances to succeed next time.
>
> With regard to the allocation, we restrict it to the node the page belongs
> to with __GFP_THISNODE, meaning we do not fallback on other node's zones.
>
> Note that gigantic hugetlb pages are fenced off since there is a cyclic
> dependency between them and alloc_contig_range.
>
> Signed-off-by: Oscar Salvador <osalvador@suse.de>
Acked-by: Michal Hocko <mhocko@suse.com>
One minor nit below
[...]
> + /*
> + * Ok, old_page is still a genuine free hugepage. Remove it from
> + * the freelist and decrease the counters. These will be
> + * incremented again when calling __prep_account_new_huge_page()
> + * and enqueue_huge_page() for new_page. The counters will remain
> + * stable since this happens under the lock.
> + */
> + remove_hugetlb_page(h, old_page, false);
> +
> + /*
> + * Call __prep_new_huge_page() to construct the hugetlb page, and
> + * enqueue it then to place it in the freelists. After this,
> + * counters are back on track. Free hugepages have a refcount of 0,
> + * so we need to decrease new_page's count as well.
> + */
> + __prep_new_huge_page(new_page);
> + __prep_account_new_huge_page(h, nid);
I think it would help to put something like the following into the
comment above this really strange construct.
/*
* new_page needs to be initialized with the standard
* hugetlb state. This is normally done by
* prep_new_huge_page but that takes hugetlb_lock which
* is already held so we need to open code it here.
* Reference count trick is needed because allocator
* gives us referenced page but the pool requires pages
* with 0 refcount.
*/
> + page_ref_dec(new_page);
> + enqueue_huge_page(h, new_page);
> +
> + /*
> + * Pages have been replaced, we can safely free the old one.
> + */
> + spin_unlock_irq(&hugetlb_lock);
> + update_and_free_page(h, old_page);
--
Michal Hocko
SUSE Labs
next prev parent reply other threads:[~2021-04-13 13:40 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-13 10:47 [PATCH v7 0/7] Make alloc_contig_range handle Hugetlb pages Oscar Salvador
2021-04-13 10:47 ` [PATCH v7 1/7] mm,page_alloc: Bail out earlier on -ENOMEM in alloc_contig_migrate_range Oscar Salvador
2021-04-13 17:00 ` Mike Kravetz
2021-04-13 10:47 ` [PATCH v7 2/7] mm,compaction: Let isolate_migratepages_{range,block} return error codes Oscar Salvador
2021-04-13 17:52 ` Mike Kravetz
2021-04-14 11:54 ` David Hildenbrand
2021-04-15 8:42 ` Oscar Salvador
2021-04-13 10:47 ` [PATCH v7 3/7] mm,hugetlb: Clear HPageFreed outside of the lock Oscar Salvador
2021-04-13 13:23 ` Michal Hocko
2021-04-13 21:19 ` Mike Kravetz
2021-04-14 6:04 ` Michal Hocko
2021-04-14 7:41 ` Oscar Salvador
2021-04-14 8:28 ` Michal Hocko
2021-04-14 10:01 ` Oscar Salvador
2021-04-14 10:03 ` Oscar Salvador
2021-04-14 10:32 ` Michal Hocko
2021-04-14 10:49 ` Oscar Salvador
2021-04-14 11:09 ` Michal Hocko
2021-04-14 12:02 ` David Hildenbrand
2021-04-14 16:45 ` Mike Kravetz
2021-04-13 10:47 ` [PATCH v7 4/7] mm,hugetlb: Split prep_new_huge_page functionality Oscar Salvador
2021-04-13 13:24 ` Michal Hocko
2021-04-13 13:26 ` Michal Hocko
2021-04-13 21:33 ` Mike Kravetz
2021-04-14 4:59 ` Oscar Salvador
2021-04-14 12:15 ` David Hildenbrand
2021-04-14 17:03 ` Mike Kravetz
2021-04-13 10:47 ` [PATCH v7 5/7] mm: Make alloc_contig_range handle free hugetlb pages Oscar Salvador
2021-04-13 13:40 ` Michal Hocko [this message]
2021-04-15 8:29 ` Oscar Salvador
2021-04-13 22:29 ` Mike Kravetz
2021-04-14 4:54 ` Oscar Salvador
2021-04-14 12:26 ` David Hildenbrand
2021-04-15 8:28 ` Oscar Salvador
2021-04-13 10:47 ` [PATCH v7 6/7] mm: Make alloc_contig_range handle in-use " Oscar Salvador
2021-04-13 22:48 ` Mike Kravetz
2021-04-14 4:52 ` Oscar Salvador
2021-04-14 17:07 ` Mike Kravetz
2021-04-13 10:47 ` [PATCH v7 7/7] mm,page_alloc: Drop unnecessary checks from pfn_range_valid_contig Oscar Salvador
2021-04-13 22:53 ` Mike Kravetz
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=YHWfQjegLi4xekhM@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 \
--cc=vbabka@suse.cz \
/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).