From: Richard Cheng <icheng@nvidia.com>
To: "Gregory Price (Meta)" <gourry@gourry.net>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org,
kernel-team@meta.com, linux-doc@vger.kernel.org,
driver-core@lists.linux.dev, linux-serial@vger.kernel.org,
linux-pm@vger.kernel.org, damon@lists.linux.dev,
linux-cxl@vger.kernel.org, akpm@linux-foundation.org,
david@kernel.org, liam@infradead.org, vbabka@kernel.org,
rppt@kernel.org, surenb@google.com, mhocko@suse.com,
corbet@lwn.net, skhan@linuxfoundation.org,
rdunlap@infradead.org, gregkh@linuxfoundation.org,
rafael@kernel.org, dakr@kernel.org, johan@kernel.org,
jirislaby@kernel.org, kasong@tencent.com, qi.zheng@linux.dev,
shakeel.butt@linux.dev, baohua@kernel.org,
axelrasmussen@google.com, yuanchu@google.com,
weixugc@google.com, baoquan.he@linux.dev,
baolin.wang@linux.alibaba.com, yury.norov@gmail.com,
linux@rasmusvillemoes.dk, pavel@kernel.org, lenb@kernel.org,
hannes@cmpxchg.org, brendan.jackman@linux.dev, ziy@nvidia.com,
sj@kernel.org, muchun.song@linux.dev, osalvador@suse.de,
nico.pache@linux.dev, ryan.roberts@arm.com, dev.jain@arm.com,
lance.yang@linux.dev, usama.arif@linux.dev, kas@kernel.org,
xu.xin@linux.dev, chengming.zhou@linux.dev,
joshua.hahnjy@gmail.com, rakie.kim@sk.com,
ying.huang@linux.alibaba.com, matthew.brost@intel.com,
byungchul@sk.com, apopple@nvidia.com, rientjes@google.com
Subject: Re: [PATCH 13/22] mm/hugetlb: restrict pool growth to common memory
Date: Wed, 18 Nov 2026 15:32:10 +0800 [thread overview]
Message-ID: <av1PyBvUVJcy_Psd@MWDK4CY14F> (raw)
In-Reply-To: <20261114070959.3490008-14-gourry@gourry.net>
On Sat, Nov 14, 2026 at 02:09:50AM +0800, Gregory Price (Meta) wrote:
> Fresh HugeTLB pool folios come from ordinary allocations, which only
> N_MEMORY_COMMON nodes serve.
>
> Restrict HugeTLB allocations and pool-growth to N_MEMORY_COMMON nodes.
>
> Keep the original allowed mask for pool shrinking and demotion so
> existing pool folios can still be found on any online-memory node.
>
> The result of this change is that hugepage allocations will fail on
> nodes without N_MEMORY_COMMON. This restriction can be loosened later
> when/if it grows a user.
>
> All memory is common at this point, so behavior is unchanged.
>
> Signed-off-by: Gregory Price (Meta) <gourry@gourry.net>
> ---
> mm/hugetlb.c | 44 ++++++++++++++++++++------------------------
> 1 file changed, 20 insertions(+), 24 deletions(-)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index cb2de24affdc4..cc1468f68901b 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -3452,13 +3452,13 @@ static void __init hugetlb_hstate_alloc_pages_onenode(struct hstate *h, int nid)
> gfp_t gfp_mask = htlb_alloc_mask(h) | __GFP_THISNODE;
>
> folio = only_alloc_fresh_hugetlb_folio(h, gfp_mask, nid,
> - &node_states[N_MEMORY], NULL);
> + &node_states[N_MEMORY_COMMON], NULL);
> if (!folio && !list_empty(&folio_list) &&
> hugetlb_vmemmap_optimizable(h)) {
> prep_and_add_allocated_folios(h, &folio_list);
> INIT_LIST_HEAD(&folio_list);
> folio = only_alloc_fresh_hugetlb_folio(h, gfp_mask, nid,
> - &node_states[N_MEMORY], NULL);
> + &node_states[N_MEMORY_COMMON], NULL);
> }
> if (!folio)
> break;
> @@ -3528,7 +3528,7 @@ static void __init hugetlb_pages_alloc_boot_node(unsigned long start, unsigned l
> prep_and_add_allocated_folios(h, &folio_list);
> INIT_LIST_HEAD(&folio_list);
> }
> - folio = alloc_pool_huge_folio(h, &node_states[N_MEMORY],
> + folio = alloc_pool_huge_folio(h, &node_states[N_MEMORY_COMMON],
> &node_alloc_noretry, &next_node);
> if (!folio)
> break;
> @@ -3672,13 +3672,12 @@ static void __init hugetlb_init_hstates(void)
>
> for_each_hstate(h) {
> /*
> - * Always reset to first_memory_node here, even if
> - * next_nid_to_alloc was set before - we can't
> - * reference hugetlb_bootmem_nodes after init, and
> - * first_memory_node is right for all further allocations.
> + * Always reset the allocation cursor here, even if it was set
> + * before: hugetlb_bootmem_nodes cannot be referenced after init,
> + * and further pool allocations use common memory.
> */
> - h->next_nid_to_alloc = first_memory_node;
> - h->next_nid_to_free = first_memory_node;
> + h->next_nid_to_alloc = first_node_state(N_MEMORY_COMMON);
> + h->next_nid_to_free = first_node_state(N_MEMORY_COMMON);
>
> /* oversize hugepages were init'ed in early boot */
> if (!hstate_is_gigantic(h))
> @@ -3802,23 +3801,21 @@ static int set_max_huge_pages(struct hstate *h, unsigned long count, int nid,
> unsigned long allocated;
> struct folio *folio;
> LIST_HEAD(page_list);
> - NODEMASK_ALLOC(nodemask_t, node_alloc_noretry, GFP_KERNEL);
> + NODEMASK_ALLOC(struct nodemask_scratch, scratch, GFP_KERNEL);
>
> - /*
> - * Bit mask controlling how hard we retry per-node allocations.
> - * If we can not allocate the bit mask, do not attempt to allocate
> - * the requested huge pages.
> - */
> - if (node_alloc_noretry)
> - nodes_clear(*node_alloc_noretry);
> - else
> + /* Scratch masks for allocation eligibility and per-node retries. */
> + if (!scratch)
> return -ENOMEM;
> + nodes_clear(scratch->mask2);
>
> /*
> * resize_lock mutex prevents concurrent adjustments to number of
> * pages in hstate via the proc/sysfs interfaces.
> */
> mutex_lock(&h->resize_lock);
> +
> + /* Fresh pool pages come from the ordinary allocation domain. */
> + nodes_and(scratch->mask1, *nodes_allowed, node_states[N_MEMORY_COMMON]);
For a global request the caller passes &node_states[N_MEMORY], so before this patch
the grow loop read the live mask on every alloc_pool_huge_folio().
Now the mask is sampled once, before the loop.
The loop drops hugetlb_lock and can run for a long time. If a node gets its first
memoroy in that window, the old code would start using it. The new code won't, until
the next write to nr_hugepages.
So the pool can end up uneven, or smaller than requested if the other nodes are full.
I think the nodes_and() should be placed inside the loop to keep the old behavior ?
Best regards,
Richard Cheng.
> flush_free_hpage_work(h);
> spin_lock_irq(&hugetlb_lock);
>
> @@ -3855,7 +3852,7 @@ static int set_max_huge_pages(struct hstate *h, unsigned long count, int nid,
> if (count > persistent_huge_pages(h)) {
> spin_unlock_irq(&hugetlb_lock);
> mutex_unlock(&h->resize_lock);
> - NODEMASK_FREE(node_alloc_noretry);
> + NODEMASK_FREE(scratch);
> return -EINVAL;
> }
> /* Fall through to decrease pool */
> @@ -3873,7 +3870,7 @@ static int set_max_huge_pages(struct hstate *h, unsigned long count, int nid,
> * within all the constraints specified by the sysctls.
> */
> while (h->surplus_huge_pages && count > persistent_huge_pages(h)) {
> - if (!adjust_pool_surplus(h, nodes_allowed, -1))
> + if (!adjust_pool_surplus(h, &scratch->mask1, -1))
> break;
> }
>
> @@ -3889,9 +3886,8 @@ static int set_max_huge_pages(struct hstate *h, unsigned long count, int nid,
> /* yield cpu to avoid soft lockup */
> cond_resched();
>
> - folio = alloc_pool_huge_folio(h, nodes_allowed,
> - node_alloc_noretry,
> - &h->next_nid_to_alloc);
> + folio = alloc_pool_huge_folio(h, &scratch->mask1, &scratch->mask2,
> + &h->next_nid_to_alloc);
> if (!folio) {
> prep_and_add_allocated_folios(h, &page_list);
> spin_lock_irq(&hugetlb_lock);
> @@ -3978,7 +3974,7 @@ static int set_max_huge_pages(struct hstate *h, unsigned long count, int nid,
> spin_unlock_irq(&hugetlb_lock);
> mutex_unlock(&h->resize_lock);
>
> - NODEMASK_FREE(node_alloc_noretry);
> + NODEMASK_FREE(scratch);
>
> return 0;
> }
> --
> 2.53.1-Meta
>
>
next prev parent reply other threads:[~2026-11-18 7:32 UTC|newest]
Thread overview: 52+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-14 7:09 [PATCH 00/22] mm: introduce N_MEMORY_COMMON Gregory Price (Meta)
2026-11-14 7:09 ` [PATCH 01/22] mm: add node-state-filtered zone and node iterators Gregory Price (Meta)
2026-11-14 7:19 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 02/22] mm: publish and clear node memory states via helpers Gregory Price (Meta)
2026-11-14 7:20 ` sashiko-bot
2026-11-18 7:53 ` Richard Cheng
2026-11-14 7:09 ` [PATCH 03/22] mm: introduce the common-memory node state Gregory Price (Meta)
2026-11-14 7:17 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 04/22] mm/page_alloc: build zonelists from common memory Gregory Price (Meta)
2026-11-14 7:16 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 05/22] mm/mempolicy: keep interleave and tmpfs mount policies on " Gregory Price (Meta)
2026-11-14 7:18 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 06/22] mm: base hash distribution on common memory nodes Gregory Price (Meta)
2026-11-14 7:15 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 07/22] mm/memory_hotplug: migrate offlined pages to common memory Gregory Price (Meta)
2026-11-14 7:16 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 08/22] mm/memory-tiers: restrict demotion topology " Gregory Price (Meta)
2026-11-14 7:16 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 09/22] mm/ksm: restrict merging " Gregory Price (Meta)
2026-11-14 7:20 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 10/22] mm/khugepaged: restrict collapse " Gregory Price (Meta)
2026-11-14 7:21 ` sashiko-bot
2026-11-14 7:21 ` sashiko-bot
2026-11-14 7:35 ` Gregory Price (Meta)
2026-11-14 7:09 ` [PATCH 11/22] mm/mempolicy: don't NUMA-hint folios outside " Gregory Price (Meta)
2026-11-14 7:16 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 12/22] mm/damon: restrict DAMON-native folio operations to " Gregory Price (Meta)
2026-11-14 7:15 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 13/22] mm/hugetlb: restrict pool growth " Gregory Price (Meta)
2026-11-14 7:18 ` sashiko-bot
2026-11-18 7:32 ` Richard Cheng [this message]
2026-11-14 7:09 ` [PATCH 14/22] mm/page_alloc: count common memory nodes for kswapd recovery Gregory Price (Meta)
2026-11-14 7:18 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 15/22] mm/oom_kill: compare OOM constraints against common memory Gregory Price (Meta)
2026-11-14 7:19 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 16/22] tty/sysrq: use common memory for the manual OOM zonelist Gregory Price (Meta)
2026-11-14 7:19 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 17/22] mm/vmscan: restrict reclaim to common memory Gregory Price (Meta)
2026-11-14 7:22 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 18/22] mm/compaction: restrict compaction " Gregory Price (Meta)
2026-11-14 7:19 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 19/22] mm/migrate: restrict move_pages() destinations " Gregory Price (Meta)
2026-11-14 7:20 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 20/22] PM: hibernate: size the image from " Gregory Price (Meta)
2026-11-14 7:34 ` sashiko-bot
2026-11-14 7:34 ` sashiko-bot
2026-11-14 7:50 ` Gregory Price (Meta)
2026-11-14 7:09 ` [PATCH 21/22] mm/khugepaged: base the recommended min_free_kbytes on " Gregory Price (Meta)
2026-11-14 7:20 ` sashiko-bot
2026-11-14 7:09 ` [PATCH 22/22] nodemask: remove the now-unused first_memory_node/next_memory_node Gregory Price (Meta)
2026-11-14 7:19 ` sashiko-bot
2026-11-18 7:01 ` [PATCH 00/22] mm: introduce N_MEMORY_COMMON Richard Cheng
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=av1PyBvUVJcy_Psd@MWDK4CY14F \
--to=icheng@nvidia.com \
--cc=akpm@linux-foundation.org \
--cc=apopple@nvidia.com \
--cc=axelrasmussen@google.com \
--cc=baohua@kernel.org \
--cc=baolin.wang@linux.alibaba.com \
--cc=baoquan.he@linux.dev \
--cc=brendan.jackman@linux.dev \
--cc=byungchul@sk.com \
--cc=chengming.zhou@linux.dev \
--cc=corbet@lwn.net \
--cc=dakr@kernel.org \
--cc=damon@lists.linux.dev \
--cc=david@kernel.org \
--cc=dev.jain@arm.com \
--cc=driver-core@lists.linux.dev \
--cc=gourry@gourry.net \
--cc=gregkh@linuxfoundation.org \
--cc=hannes@cmpxchg.org \
--cc=jirislaby@kernel.org \
--cc=johan@kernel.org \
--cc=joshua.hahnjy@gmail.com \
--cc=kas@kernel.org \
--cc=kasong@tencent.com \
--cc=kernel-team@meta.com \
--cc=lance.yang@linux.dev \
--cc=lenb@kernel.org \
--cc=liam@infradead.org \
--cc=linux-cxl@vger.kernel.org \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-pm@vger.kernel.org \
--cc=linux-serial@vger.kernel.org \
--cc=linux@rasmusvillemoes.dk \
--cc=matthew.brost@intel.com \
--cc=mhocko@suse.com \
--cc=muchun.song@linux.dev \
--cc=nico.pache@linux.dev \
--cc=osalvador@suse.de \
--cc=pavel@kernel.org \
--cc=qi.zheng@linux.dev \
--cc=rafael@kernel.org \
--cc=rakie.kim@sk.com \
--cc=rdunlap@infradead.org \
--cc=rientjes@google.com \
--cc=rppt@kernel.org \
--cc=ryan.roberts@arm.com \
--cc=shakeel.butt@linux.dev \
--cc=sj@kernel.org \
--cc=skhan@linuxfoundation.org \
--cc=surenb@google.com \
--cc=usama.arif@linux.dev \
--cc=vbabka@kernel.org \
--cc=weixugc@google.com \
--cc=xu.xin@linux.dev \
--cc=ying.huang@linux.alibaba.com \
--cc=yuanchu@google.com \
--cc=yury.norov@gmail.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.